--- /dev/null
+# Set the default behavior, in case people don't have core.autocrlf set.
+* text=auto
+
+*.sh text eol=lf
+sshd_config text eol=lf
+
+# Explicitly declare text files you want to always be normalized and converted
+# to native line endings on checkout.
+*.c text
+*.h text
+*.cpp text
+*.hpp text
+*.py text
+
+# Declare files that will always have CRLF line endings on checkout.
+*.sln text eol=crlf
+
+# Denote all files that are truly binary and should not be modified.
+*.png binary
+*.jpg binary
--- /dev/null
+# ignore build directories and any files in build directories
+build/
+
+# ignore any qt-creator user files for cmake projects. these are files
+# specific to a user and specific to qt-creator
+CMakeLists.txt.user
+
+
--- /dev/null
+SOFTWARE TOOLS LICENSE AGREEMENT
+
+DO NOT DOWNLOAD, INSTALL, ACCESS, COPY, OR USE ANY PORTION OF THE MATERIALS (DEFINED BELOW) UNTIL YOU HAVE READ AND ACCEPTED
+THE TERMS AND CONDITIONS OF THIS AGREEMENT. BY INSTALLING, COPYING, ACCESSING, OR USING THE MATERIALS, YOU AGREE TO BE LEGALLY
+BOUND BY THE TERMS AND CONDITIONS OF THIS AGREEMENT. If You do not agree to be bound by, or the entity for whose benefit You act has not
+authorized You to accept, these terms and conditions, do not install, access, copy, or use the Software and destroy all copies of the Software in Your
+possession.
+
+This DEVELOPMENT TOOLS LICENSE AGREEMENT (this "Agreement") is entered into between Intel Corporation, a Delaware corporation ("Intel") and You.
+"You" refers to you or your employer or other entity for whose benefit you act, as applicable. If you are agreeing to the terms and conditions of this
+Agreement on behalf of a company or other legal entity, you represent and warrant that you have the legal authority to bind that legal entity to the
+Agreement, in which case, "You" or "Your" shall be in reference to such entity. Intel and You are referred to herein individually as a "Party" or, together, as
+the "Parties".
+
+The Parties, in consideration of the mutual covenants contained in this Agreement, and for other good and valuable consideration, the receipt and
+sufficiency of which they acknowledge, and intending to be legally bound, agree as follows:
+
+1. DEFINITIONS. The following definitions are used throughout this Agreement:
+"Affiliate" means any entity controlling, controlled by or under common control with a Party hereto, where "control" means the direct or indirect ownership
+of more than fifty percent (50%) of such entity"s capital or equivalent voting rights. An entity will be deemed an "Affiliate" only as long as such control exists
+during the term of this Agreement.
+
+"Contractor" means a third party consultant or subcontractor who requires access to or use of the Materials to perform work on Your behalf or at Your
+behest.
+
+"Development Tools" means the development, evaluation, production, or test tool software, and associated documentation or other collateral, identified
+in the "development_tools.txt" text files, if any, included in the Materials.
+
+"Derivatives" means derivative works as defined in 17 U.S.C " 101 et seq.
+
+"Intel-based Device" means a device designed, manufactured, or configured by You or Your Affiliates to include or operate Intel hardware, software, or
+services.
+
+"Materials" means the software, documentation, the software product serial number and license key codes (if applicable), Development Tools,
+Redistributables, and other materials or collateral, including any updates and upgrades thereto, in source code or object code form where applicable, that
+are provided or otherwise made available by Intel to You under this Agreement. "Materials" do not include Open Source Software or any computer
+programming code that is subject to an agreement, obligation or license (whether or not accompanying the Materials) intended to supersede this
+Agreement.
+
+"Redistributables" means the software, documentation, or other collateral identified in the "redist.txt" text files, if any, included in the Materials.
+
+2. LIMITED LICENSE.
+(A) Subject to the terms and conditions of this Agreement, Intel grants You and Your Affiliates, a limited, nonexclusive, nontransferable, revocable,
+worldwide, fully paid-up license during the term of this Agreement, without the right to sublicense, unless expressly stated otherwise, to:
+(1) internally reproduce and install a reasonable number of copies of the Materials for Your internal use solely for the purposes of designing,
+developing, manufacturing and testing Intel-based Devices;
+(2) internally reproduce the source code of the Development Tools, if provided to You by Intel, and to internally create and reproduce Derivatives of
+the Development Tools, and to internally reproduce the binary code of the Development Tools, or any Derivatives created by You, in each case solely
+for the purpose of designing, developing, manufacturing and testing the Intel-based Device, solely as necessary for the integration of any Intel software
+and the output generated by the Development Tools, with and into Intel-based Devices;
+(3) create Derivatives of the Redistributables, or any portions thereof, provided to You by Intel in source code form solely for the purposes of designing,
+developing, debugging, modifying, distributing and testing software containing significantly more functionality and features than the Redistributables
+in the form provided to You by Intel;
+(4) distribute (or otherwise make available) on a royalty-free basis, subject to any other terms and conditions which may appear in the Redistributables
+text files, the Redistributables, including any Derivatives of the Redistributables pursuant to Section 2(A)(3), or any portions thereof, only as integrated
+or embedded in software (and not on a stand-alone basis) solely for use on an Intel-based Device; and
+(5) have the tasks set forth in Section 2(A)(1) and (2) above performed by a Contractor on the conditions that You enter into a written confidentiality
+agreement with any such Contractor, subject to Section 7 (Confidentiality), and You remain fully liable to Intel for the actions and inactions of Your
+Contractors.
+(B) You will be liable for Your Affiliate"s breach of these terms. In addition, You acknowledge that Your Affiliates are beneficiaries of the licenses granted by
+Intel under Section 2.
+(C) Intel hereby grants You the right to sub-license (without rights to further sublicense) the Development Tools, including any accompanying
+documentation, to Your manufacturing partners, in the code format provided to You by Intel, solely for designing, developing, manufacturing and testing
+the Intel-based Devices solely as necessary for the integration of any Intel software and the output generated by the Development Tools, with and into
+Intel-based Devices. The sublicense is subject to a written sublicensing agreement that contains confidentiality obligations and license restrictions that are
+no less protective of Intel than those provided in this Agreement. You will be fully responsible and liable towards Intel for Your sub-licensees" compliance
+with all such confidentiality obligations and license restrictions. You may grant Your manufacturing partners the right to further distribute Redistributables
+solely as integrated or embedded in software for Your Intel-based Devices.
+
+3. LICENSE RESTRICTIONS. All right, title and interest in and to the Materials and associated documentation are and will remain the exclusive property of
+Intel and its suppliers. Unless expressly permitted under the Agreement, You will not, and will not allow any third party to (i) use, copy, distribute, sell or
+offer to sell the Materials or associated documentation; (ii) modify, adapt, enhance, disassemble, decompile, reverse engineer, change or create derivative
+works from the Materials except and only to the extent as specifically required by mandatory applicable laws or any applicable third party license terms
+accompanying the Materials; (iii) use or make the Materials available for the use or benefit of third parties; or (iv) use the Materials on Your products other
+than those that include the Intel product(s), platform(s), or software identified in the Materials; or (v) publish or provide any Materials benchmark or
+comparison test results.
+If You received the Materials solely for evaluation purposes, You have no distribution rights to the Materials or any portion thereof.
+
+Distribution of the Redistributables is also subject to the following conditions: You shall: (i) be solely responsible to Your customers and end users for any
+update or support obligation or other liability which may arise from the distribution, (ii) not make any statement that Your software is "certified", or that its
+performance is guaranteed, by Intel, (iii) not use Intel's name or trademarks to promote Your software without prior written permission, (iv) use a license
+agreement that contains provisions that are at least as restrictive as this Agreement and which prohibits disassembly and reverse engineering of the
+Materials provided in object code form, and (v) indemnify, hold harmless, and defend Intel, Intel"s Affiliates, and its licensors from and against any claims
+or lawsuits, including attorney's fees, that arise or result from Your Derivatives or Your distribution of Your software.
+
+The consideration under this Agreement is only for the licenses Intel expressly grants above. Any other rights including, but not limited to, additional patent
+rights, will require an additional license and additional consideration. Nothing in this Agreement requires or will be treated to require Intel to grant any
+additional license. You acknowledge that an essential basis of the bargain in this Agreement is that Intel grants You no licenses or other rights including,
+but not limited to, patent, copyright, trade secret, trademark, trade name, service mark or other intellectual property licenses or rights with respect to the
+Materials and associated documentation, by implication, estoppel or otherwise, except for the licenses expressly granted above. You acknowledge there
+are significant uses of the Materials in their original, unmodified and uncombined form. The consideration for the licenses in this Agreement reflects Intel"s
+continuing right to assert patent claims against any modifications or derivative works (including, without limitation, error corrections and bug fixes) of, or
+combinations with, the Materials that You, Your Affiliates or third parties make that infringe any Intel patent claim.
+
+4. LICENSE TO FEEDBACK. This Agreement does not obligate You to provide Intel with materials, information, comments, suggestions, Your Derivatives or
+other communication regarding the features, functions, performance or use of the Materials ("Feedback"). If any software included in the Materials is
+provided or otherwise made available by Intel in source code form, to the extent You provide Intel with Feedback in a tangible form, You grant to Intel and
+its affiliates a non-exclusive, perpetual, sublicenseable, irrevocable, worldwide, royalty-free, fully paid-up and transferable license, to and under all of Your
+intellectual property rights, whether perfected or not, to publicly perform, publicly display, reproduce, use, make, have made, sell, offer for sale, distribute,
+import, create derivative works of and otherwise exploit any comments, suggestions, descriptions, ideas, Your Derivatives or other feedback regarding the
+Materials provided by You or on Your behalf.
+
+5. OPEN SOURCE STATEMENT. The Materials may include Open Source Software (OSS) licensed pursuant to OSS license agreement(s) identified in the
+OSS comments in the applicable source code file(s) and/or file header(s) provided with or otherwise associated with the Materials. Neither You nor any
+Original Equipment Manufacturer (OEM), Original Device Manufacturer (ODM), customer, or distributor may subject any proprietary portion of the Materials
+to any OSS license obligations including, without limitation, combining or distributing the Materials with OSS in a manner that subjects Intel, the Materials
+or any portion thereof to any OSS license obligation. Nothing in this Agreement limits any rights under, or grants rights that supersede, the terms of any
+applicable OSS license.
+
+6. THIRD PARTY SOFTWARE. Certain third party software provided with or within the Materials may only be used (a) upon securing a license directly from
+the owner of the software or (b) in combination with hardware components purchased from such third party and (c) subject to further license limitations
+by the software owner. A listing of any such third party limitations is in one or more text files accompanying the Materials. You acknowledge Intel is not
+providing You with a license to such third party software and further that it is Your responsibility to obtain appropriate licenses from such third parties
+directly.
+
+7. CONFIDENTIALITY. The terms and conditions of this Agreement, exchanged confidential information, as well as the Materials are subject to the terms
+and conditions of the Non-Disclosure Agreement(s) or Intel Pre-Release Loan Agreement(s) (referred to herein collectively or individually as "NDA") entered
+into by and in force between Intel and You, and in any case no less confidentiality protection than You apply to Your information of similar sensitivity. If
+You would like to have a Contractor perform work on Your behalf that requires any access to or use of Materials You must obtain a written confidentiality
+agreement from the Contractor which contains terms and conditions with respect to access to or use of Materials no less restrictive than those set forth in
+this Agreement, excluding any distribution rights and use for any other purpose, and You will remain fully liable to Intel for the actions and inactions of
+those Contractors. You may not use Intel's name in any publications, advertisements, or other announcements without Intel's prior written consent.
+
+8. NO OBLIGATION; NO AGENCY. Intel may make changes to the Software, or items referenced therein, at any time without notice. Intel is not obligated to
+support, update, provide training for, or develop any further version of the Software or to grant any license thereto. No agency, franchise, partnership, joint-
+venture, or employee-employer relationship is intended or created by this Agreement.
+
+9. EXCLUSION OF WARRANTIES. THE MATERIALS ARE PROVIDED "AS IS" WITHOUT ANY EXPRESS OR IMPLIED WARRANTY OF ANY KIND INCLUDING
+WARRANTIES OF MERCHANTABILITY, NONINFRINGEMENT, OR FITNESS FOR A PARTICULAR PURPOSE. Intel does not warrant or assume responsibility
+for the accuracy or completeness of any information, text, graphics, links or other items within the Materials.
+
+10. LIMITATION OF LIABILITY. IN NO EVENT WILL INTEL OR ITS AFFILIATES, LICENSORS OR SUPPLIERS (INCLUDING THEIR RESPECTIVE DIRECTORS,
+OFFICERS, EMPLOYEES, AND AGENTS) BE LIABLE FOR ANY DAMAGES WHATSOEVER (INCLUDING, WITHOUT LIMITATION, LOST PROFITS, BUSINESS
+INTERRUPTION, OR LOST DATA) ARISING OUT OF OR IN RELATION TO THIS AGREEMENT, INCLUDING THE USE OF OR INABILITY TO USE THE MATERIALS,
+EVEN IF INTEL HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGES. SOME JURISDICTIONS PROHIBIT EXCLUSION OR LIMITATION OF LIABILITY
+FOR IMPLIED WARRANTIES OR CONSEQUENTIAL OR INCIDENTAL DAMAGES, SO THE ABOVE LIMITATION MAY IN PART NOT APPLY TO YOU. YOU MAY
+ALSO HAVE OTHER LEGAL RIGHTS THAT VARY FROM JURISDICTION TO JURISDICTION. THE MATERIALS LICENSED HEREUNDER ARE NOT DESIGNED
+OR INTENDED FOR USE IN ANY MEDICAL, LIFE SAVING OR LIFE SUSTAINING SYSTEMS, TRANSPORTATION SYSTEMS, NUCLEAR SYSTEMS, OR FOR ANY
+OTHER MISSION CRITICAL APPLICATION IN WHICH THE FAILURE OF THE DEVELOPMENT TOOLS COULD LEAD TO PERSONAL INJURY OR DEATH. YOU
+WILL INDEMNIFY AND HOLD INTEL AND ITS AFFILIATES, LICENSORS AND SUPPLIERS (INCLUDING THEIR RESPECTIVE DIRECTORS, OFFICERS,
+EMPLOYEES, AND AGENTS) HARMLESS AGAINST ALL CLAIMS, LIABILITIES, LOSSES, COSTS, DAMAGES, AND EXPENSES (INCLUDING REASONABLE
+ATTORNEY FEES), ARISING OUT OF, DIRECTLY OR INDIRECTLY, THE DISTRIBUTION OF THE MATERIALS AND ANY CLAIM OF PRODUCT LIABILITY,
+PERSONAL INJURY OR DEATH ASSOCIATED WITH ANY UNINTENDED USE, EVEN IF SUCH CLAIM ALLEGES THAT INTEL OR AN INTEL AFFILIATE, LICENSOR
+OR SUPPLIER WAS NEGLIGENT REGARDING THE DESIGN OR MANUFACTURE OF THE MATERIALS. THE LIMITED REMEDIES, WARRANTY DISCLAIMER AND
+LIMITED LIABILITY ARE FUNDAMENTAL ELEMENTS OF THE BASIS OF THE BARGAIN BETWEEN INTEL AND YOU AND INTEL WOULD NOT BE ABLE TO
+PROVIDE THE MATERIALS WITHOUT SUCH LIMITATIONS.
+
+11. TERMINATION AND SURVIVAL. Intel may terminate this Agreement for any reason with thirty (30) days" notice and immediately if You or someone
+acting on Your behalf or at Your behest violates any of its terms or conditions. Upon termination You will immediately destroy and ensure the destruction
+of the Materials (including providing certification of such destruction or return back to Intel). Upon termination of this Agreement, all licenses granted to
+You hereunder terminate immediately. All Sections of this Agreement, except Section 2, will survive termination. In the event of termination of this
+Agreement, the license grant to any Redistributables, including Your Derivatives of the Redistributables, distributed by You prior to the effective date of
+such termination and in accordance with the terms and conditions of this Agreement shall survive any such termination of this Agreement.
+
+12. GOVERNING LAW AND JURISDICTION. This Agreement and any dispute arising out of or relating to it will be governed by the laws of the U.S.A. and
+Delaware, without regard to conflict of laws principles. The Parties exclude the application of the United Nations Convention on Contracts for the
+International Sale of Goods (1980). The state and federal courts sitting in Delaware, U.S.A. will have exclusive jurisdiction over any dispute arising out of or
+relating to this Agreement. The Parties consent to personal jurisdiction and venue in those courts. A Party that obtains a judgment against the other Party
+in the courts identified in this section may enforce that judgment in any court that has jurisdiction over the Parties.
+
+13. EXPORT REGULATIONS/EXPORT CONTROL. You agree that neither You nor Your subsidiaries or Affiliates will export/re-export the Materials, directly
+or indirectly, to any country for which the U.S. Department of Commerce or any other agency or department of the U.S. Government or the foreign
+government from where it is shipping requires an export license, or other governmental approval, without first obtaining any such required license or
+approval. In the event the Materials are exported from the U.S.A. or re-exported from a foreign destination by You, Your subsidiaries, or Your Affiliates, You
+will ensure that the distribution and export/re-export or import of the Materials complies with all laws, regulations, orders, or other restrictions of the U.S.
+Export Administration Regulations and the appropriate foreign government.
+
+14. GOVERNMENT RESTRICTED RIGHTS. The Materials are a commercial item (as defined in 48 C.F.R. 2.101) consisting of commercial computer software
+and commercial computer software documentation (as those terms are used in 48 C.F.R. 12.212). Consistent with 48 C.F.R. 12.212 and 48 C.F.R 227.72021
+through 227.7202-4, You will not provide the Materials to the U.S. Government. Contractor or Manufacturer is Intel Corporation, 2200 Mission College
+Blvd., Santa Clara, CA 95054.
+
+15. TRADEMARKS. Third party trademarks, trade names, product names and logos (the "Trademarks") contained in or used by the Materials are the
+trademarks or registered trademarks of their respective owners, and the use of such Trademarks shall inure to the benefit of the trademark owner. The
+reference to such Trademarks (if any) by Intel in any of the Materials does not constitute: (i) an affiliation by Intel and its licensors with such company, or (ii)
+an endorsement or approval of such company of Intel and its licensors and its products or services.
+
+16. ASSIGNMENT. You may not delegate, assign or transfer this Agreement, the license(s) granted or any of Your rights or duties hereunder, expressly, by
+implication, by operation of law, or otherwise and any attempt to do so, without Intel"s express prior written consent, will be null and void. Intel may assign,
+delegate and transfer this Agreement, and its rights and obligations hereunder, in its sole discretion.
+
+17. ENTIRE AGREEMENT; SEVERABILITY. The terms and conditions of this Agreement and any NDA with Intel constitute the entire agreement between the
+Parties with respect to the subject matter hereof, and merge and supersede all prior or contemporaneous agreements, understandings, negotiations and
+discussions. Neither Party will be bound by any terms, conditions, definitions, warranties, understandings, or representations with respect to the subject
+matter hereof other than as expressly provided herein. In the event any provision of this Agreement is unenforceable or invalid under any applicable law
+or applicable court decision, such unenforceability or invalidity will not render this Agreement unenforceable or invalid as a whole, instead such provision
+will be changed and interpreted so as to best accomplish the objectives of such provision within legal limits.
+
+18. WAIVER. The failure of a Party to require performance by the other Party of any provision hereof will not affect the full right to require such performance
+at any time thereafter; nor will waiver by a Party of a breach of any provision hereof constitute a waiver of the provision itself.
+
+19. PRIVACY. YOUR PRIVACY RIGHTS ARE SET FORTH IN INTEL'S PRIVACY NOTICE, WHICH FORMS A PART OF THIS AGREEMENT. PLEASE REVIEW THE
+PRIVACY NOTICE AT HTTP://WWW.INTEL.COM/PRIVACY TO LEARN HOW INTEL COLLECTS, USES AND SHARES INFORMATION ABOUT YOU.
--- /dev/null
+
+ifneq ($(findstring movidius, $(PYTHONPATH)), movidius)
+ export PYTHONPATH:=/opt/movidius/caffe/python:/opt/movidius/mvnc/python:$(PYTHONPATH)
+endif
+
+
+.PHONY: help
+help:
+ @echo "\nmake help starting."
+ @echo "Make targets are:"
+ @echo " make help - shows this message"
+ @echo " make install - Installs the ncsdk."
+ @echo " make examples - makes the ncsdk examples."
+ @echo " make uninstall - uninstalls the ncsdk."
+ @echo " make clean - removes targets and intermediate files."
+
+.PHONY: all
+all: install examples
+
+.PHONY: opencv
+opencv:
+ ./install-opencv.sh
+
+.PHONY: prereqs
+prereqs:
+ @sed -i 's/\r//' ncsdk.conf
+ @if [ -e ncsdk_redirector.txt ] ; \
+ then \
+ @sed -i 's/\r//' ncsdk_redirector.txt ; \
+ fi
+
+ @sed -i 's/\r//' install.sh
+ @sed -i 's/\r//' uninstall.sh
+ @sed -i 's/\r//' README.md
+ @chmod +x install.sh
+ @chmod +x uninstall.sh
+ @chmod +x install-opencv.sh
+
+.PHONY: install
+install: prereqs
+ @echo "\nmake install starting."
+ ./install.sh
+
+.PHONY: uninstall
+uninstall: prereqs
+ @echo "\nmake uninstall starting."
+ ./uninstall.sh
+
+.PHONY: examples
+examples: prereqs opencv
+ @echo "\nmake examples starting."
+ (cd examples; make)
+
+.PHONY: runexamples
+runexamples: prereqs opencv
+ @echo "\nmake examples starting."
+ (cd examples; make run)
+
+.PHONY: clean
+clean:
+ @echo "\nmake clean starting."
+ (cd examples; make clean)
+
+
-# MvNC_SDK-
-Toolkit/API binaries for the Neural Compute Stick Program
+# Movidius™ Neural Compute Software Development Kit
+This SDK is provided for users of the [Movidius™ Neural Compute Stick (NCS)](https://developer.movidius.com/). It provides software tools, an API, and examples which enable developers to create software that takes advantage of the hardware the accelerated neural network capability provided by the NCS.
+
+
+# Installation
+The provided Makefile helps with installation. Clone this repository and then run the following command to install the SDK.
+
+```
+make install
+```
+
+# Examples
+Also included in the SDK are examples. After cloning and running 'make install' run the following command to install examples.
+```
+make examples
+```
+
+# Documentation
+The complete Neural Compute SDK documentation can be viewed at [https://movidius.github.io/ncsdk/](https://movidius.github.io/ncsdk/)
+
--- /dev/null
+# Caffe Support
+
+## Introduction
+[Caffe](http://caffe.berkeleyvision.org/) is a deep learning framework developed by Berkeley AI Research ([BAIR](http://bair.berkley.edu)) and by community contributors. The setup script currently downloads BVLC Caffe and installs it in a system location. Other versions of Caffe are not supported at this time. For more information please visit http://caffe.berkeleyvision.org/
+
+Default Caffe Installation Location: /opt/movidius/caffe<br>
+Checkout Berkley Vision's Web Image Classification [demo](http://demo.caffe.berkeleyvision.org/)
+
+## Caffe Zoo
+Berkley Vision hosts a Caffe Model Zoo for researchers and engineers to contribute Caffe models for various tasks. Please visit the [Berkley Caffe Zoo](http://caffe.berkeleyvision.org/model_zoo.html) page to learn more about the caffe zoo and how to create your own Caffe Zoo model and contribute.
+
+Caffe Zoo has several models contributed including a model network that can classify images for Age and Gender. This network trained by [Gil Levi](https://gist.github.com/GilLevi) and Tal Hassner is at this [Gender Net Caffe Zoo Model on GitHub](https://gist.github.com/GilLevi/c9e99062283c719c03de)
+
+Caffe models consists of two files that are used for compiling the caffe model using the [Neural Compute Compiler](tools/compile.md)
+* Caffe Network Description (.prototxt): Text file that describes the topology and layers of the network.
+* Caffe Weights (.caffemodel): Contains the weights for each layer that are obtained after training a model.
+
+## Neural Compute Caffe Layer Support
+The following layers are supported in Caffe by the Neural Compute SDK. The Neural Compute Stick does not support training, so some layers that are only required for training are not supported.
+
+### Activation/Neuron
+* bias
+* elu
+* prelu
+* relu
+* scale
+* sigmoid
+* tanh
+
+### Common
+* inner_product
+
+### Normalization
+* batch_norm
+* lrn
+
+### Utility
+* concat
+* eltwise
+* flatten
+* parameter
+* reshape
+* slice
+* softmax
+
+### Vision
+* conv
+ * Regular Convolution - 1x1s1, 3x3s1, 5x5s1, 7x7s1, 7x7s2, 7x7s4
+ * Group Convolution - <1024 groups total
+* deconv
+* pooling
+
+# Known Issues
+### Caffe Input Layer
+
+Limitation: Batch Size which is the first dimension must always be 1
+
+Limitation: The number of inputs must be 1
+
+Limitation: We don't support this "input_param" format for the input layer:
+
+```
+1 name: "GoogleNet"
+2 layer {
+3 name: "data"
+4 type: "Input"
+5 top: "data"
+6 input_param { shape: { dim: *10* dim: 3 dim: 224 dim: 224 } }
+7 }
+```
+
+We only support this "input_shape" format for the input layer:
+
+```
+name: "GoogleNet"
+input: "data"
+ input_shape
+{ dim:1 dim:3 dim:224 dim:224 }
+```
+
+### Input Name
+Input name should be always called "data"
+This works
+```
+name: "GoogleNet"
+input: "data"
+ input_shape
+{ dim:1 dim:3 dim:224 dim:224 }
+```
+This does not
+```
+name: "GoogleNet"
+input: "data_x"
+ input_shape
+{ dim:1 dim:3 dim:224 dim:224 }
+```
+
+### Non-Square Convolutions
+Limitation: We don't support non-square convolutions such as 1x20
+```
+input: "data"
+ input_shape
+{ dim:1 dim:1 dim:1 dim:64 }
+layer {
+ name: "conv1_x"
+ type: "Convolution"
+ bottom: "data"
+ top: "conv1_x"
+ convolution_param {
+ num_output: 3
+ kernel_h: 1
+ kernel_w: 20
+ stride: 1
+ }
+```
+
+### Crop Layer
+Limitation: Crop layer cannot take reference size layer from input:"data"
+
+```
+layer {
+ name: "score"
+ type: "Crop"
+ bottom: "upscore"
+ bottom: "data"
+ top: "score"
+ crop_param {
+ axis: 2
+ offset: 18
+ }
+}
+```
+
+### Size Limitations
+Compiled Movidius™ "graph" file < 320MB
+Intermediate layer buffer size < 100MB
+```
+[Error 35] Setup Error: Not enough resources on Myriad to process this network
+```
+
+Scratch Memory size < 112KB
+
+```
+[Error 25] Myriad Error: "Matmul scratch memory [112640] lower than required [165392]"
+```
+
+## Caffe Networks
+The following networks are validated and known to work on the Movidius™ Neural Compute SDK.
+- GoogleNet V1
+- SqueezeNet V1.1
+- LeNet
+- CaffeNet
+- VGG (Sousmith VGG_A)
+- AlexNet
+
-# Steven should not be seeing this, but he is ...
+# Introduction
+The Movidius™ Neural Compute SDK and Movidius™ Neural Compute Stick (NCS) enables rapid prototyping, validation and deployment of Deep Neural Networks (DNNs.)
+
+The NCS is used in two primary scenarios:
+- Profiling, tuning, and compiling a DNN on a development computer (host system) with the tools provided in the Movidius™ Neural Compute SDK. In this scenario the host system is typically a desktop or laptop machine running Ubuntu 16.04 Desktop (x86, 64 bit), but you can use any supported platform for these steps.
+
+- Prototyping a user application on a development computer (host system) which accesses the hardware of the NCS to accelerate DNN inferences via the API provided with the Movidius™ Neural Compute SDK. In this scenario the host system can be a developer workstation or any developer system that runs an operating system compatible with the API.
+
+The following diagram shows the typical workflow for development with the NCS:
+
+
+The training phase does not utilize the NCS hardware or SDK, while the subsequent phases of “profiling, tuning and compiling” and “prototyping” do require the NCS hardware and the accompanying Movidius™ Neural Compute SDK
+
+The SDK contains a set of software tools to compile, profile, and check validity of your DNN as well as an API for both the C and Python programming languages. The API is provided to allow users to create software which offloads the neural network computation onto the Movidius™ Neural Compute Stick.
+
+Here is more information on the [architecture](ncs1arch.md) of the Neural Compute Stick.
+
+<a name="Frameworks"></a>
+# Frameworks
+Neural Compute SDK currently supports two Deep Learning frameworks.
+1. [Caffe](Caffe.md) : Caffe is a deep learning framework from Berkeley Vision Labs.
+2. [TensorFlow™](TensorFlow.md): TensorFlow™ is a deep learning framework from Google.
+
+[See how to use networks from these supported framework with NCS.](configure_network.md)
+
+
+<a name="InstallAndExamples"></a>
+# Installation and Examples
+The following commands install NCSDK and run examples. Detailed instructions for [installation and Configuration](install.md)
+
+```
+git clone http://github.com/Movidius/ncsdk && cd ncsdk && make install && make examples
+
+```
+<a name="NcSdkTools"></a>
+# Movidius™ Neural Compute SDK Tools
+The SDK comes with a set of tools to assist in development and deployment of applications that utilize hardware accelerated Deep Neural Networks via the Movidius™ Neural Compute Stick (NCS). Each tool and its usage is described on this page below
+
+* [mvNCCompile](tools/compile.md): Converts Caffe/TF network and weights to Movidius™ Internal compiled format
+
+* [mvNCProfile](tools/profile.md): Provides layer by layer statistics to evaluate the performance of Caffe/TF networks on the NCS
+
+* [mvNCCheck](tools/check.md): Compares the results from an inference by running the network on the NCS and Caffe/TF
+
+<a name="NcApi"></a>
+# Neural Compute API
+Applications for inferencing with Neural Compute SDK can be developed either in C/C++ or Python. The API provides a software interface to Open/Close Neural Compute Sticks, load graphs into the NCS and run inferences on the stick.
+
+* [C API](c_api/readme.md)
+* [Python API](py_api/readme.md)
+
+<a name="UserForum"></a>
+# Movidius™ Neural Compute User Forum
+
+There is an active user forum in which users of the Neural Compute Stick discuss ideas and issues they have with regard to the NCS. The link to the NCS User Forum is:
+
+[https://ncsforum.movidius.com](https://ncsforum.movidius.com)
+
+The forum is a good place to go if you need help troubleshooting an issue. You may find other people that have figured out the issue or get ideas for how to fix it. The forum is also monitored by Movidius™ engineers which provide solutions as well.
+
+<a name="Examples"></a>
+# Examples
+
+There are several examples including the following at the github
+* Caffe
+ * GoogLeNet
+ * AlexNet
+ * SqueezeNet
+* TensorFlow™
+ * Inception V1
+ * Inception V3
+* Apps
+ * hello_ncs_py
+ * hello_ncs_cpp
+ * multistick_cpp
+
+The examples demonstrate compiling, profiling and running inferences using the network on the Movidius™ Neural Compute Stick.
+Each example contains a Makefile. Running 'make help' in the example's base directory will give possible make targets.
+
+```
+
+git clone http://github.com/Movidius/ncsdk # Already done during installation
+(cd ncsdk/examples && make) # run all examples
+(cd ncsdk/examples/caffe/GoogLeNet && make) # Run just one example
+
+```
+
+
+[Release Notes](release_notes.md)
--- /dev/null
+# NC SDK Documentation Table of Contents
+
+- [Introduction](readme.md#Introduction)
+ - [NCS Architecture](ncs1arch.md)
+- [Frameworks](readme.md#Frameworks)
+ - [Caffe Support](Caffe.md)
+ - [TensorFlow Support](TensorFlow.md)
+ - [Configure Networks for NCS](configure_network.md)
+- [Installation and examples](readme.md#InstallAndExamples)
+ - [Detailed Installation Instructions](install.md)
+ - [Virtual Machine Configuration](VirtualMachineConfig.md)
+ - [Installation manifiest](manifest.md)
+- [NC SDK Tools](readme.md#NcSdkTools)
+ - [mvNCCompile](tools/compile.md)
+ - [mvNCProfile](tools/profile.md)
+ - [mvNCCheck](tools/check.md)
+- [Neural Compute API](readme.md#NcApi)
+ - [Python API](py_api/readme.md)
+ - [C API](c_api/readme.md)
+- [Neural Compute User Forum](readme.md#UserForum)
+- [Examples](readme.md#Examples)
+
+
--- /dev/null
+# TensorFlow™ Support
+
+# Introduction
+[TensorFlow™](https://www.tensorflow.org/) is a deep learning framework pionered by Google. The NCSDK introduced TensorFlow™ support with the 1.09.xx NCSDK release. Validation has been done with TensorFlow™ r1.3. The TensorFlow™ website describes it as "TensorFlow™ is an open source software library for numerical computation using data flow graphs. Nodes in the graph represent mathematical operations, while the graph edges represent the multidimensional data arrays (tensors) communicated between them."
+
+Default Installation Location: /opt/movidius/tensorflow
+
+# TensorFlow™ Model Zoo
+TensorFlow™ has a model GitHub repo at https://github.com/tensorflow/models similar to the Caffe Zoo for Caffe. The TensorFlow™ models GitHub repository contains several models which are maintained by the respective autors unlike Caffe which is not a single GitHub repo.
+
+# Save Session with graph and checkpoint information
+
+```python
+import numpy as np
+import tensorflow as tf
+
+from tensorflow.contrib.slim.nets import inception
+
+slim = tf.contrib.slim
+
+def run(name, image_size, num_classes):
+ with tf.Graph().as_default():
+ image = tf.placeholder("float", [1, image_size, image_size, 3], name="input")
+ with slim.arg_scope(inception.inception_v1_arg_scope()):
+ logits, _ = inception.inception_v1(image, num_classes, is_training=False, spatial_squeeze=False)
+ probabilities = tf.nn.softmax(logits)
+ init_fn = slim.assign_from_checkpoint_fn('inception_v1.ckpt', slim.get_model_variables('InceptionV1'))
+
+ with tf.Session() as sess:
+ init_fn(sess)
+ saver = tf.train.Saver(tf.global_variables())
+ saver.save(sess, "output/"+name)
+
+run('inception-v1', 224, 1001)
+```
+# Compile for TensorFlow™
+
+```
+mvNCCompile output/inception-v1.meta -in=input -on=InceptionV1/Logits/Predictions/Reshape_1 -s12
+
+```
+
+# Neural Compute TensorFlow™ Layer Support
+
+# TensorFlow™ Networks Supported
+* AlexNet
+* Inception V1
+* Inception V3
+* Inception V4
+* Inception ResNet V1 (FaceNet)
+* Inception ResNet V2
+* MobileNet
+
+
--- /dev/null
+# Virtual Machine Configurations
+
+The following configuration has been tested with the 1.09 SDK release
+
+## General Requirements
+- Virtualbox 5.1.28 (later releases should be fine but not tested)
+- Guest Extensions installed
+- You will need to select usb3.0 and create two filters:
+ - USB2 filter with vendor ID 03e7 and product ID 2150
+ - USB3 filter with vendor ID 040e and product ID f63b
+- Host OS (these have been tested, other may work):
+ - OSX Yosemite 10.10.5
+ - Windows 10 Enterprise
+ - Ubuntu 16.04 Desktop
+- Guest OS:
+ - Ubuntu 16.04 Desktop
+
+## Installation Order
+- Install Ubuntu 16.04 on VM
+- Install updates (apt-get update, apt-get upgrade)
+- Install guest extensions (virtualbox menu devices / Insert guest additions CD image)
+- Setup USB filters
+- Install NCSDK with 'make install' ([Installation Instructions](install.md))
+- Insert NCS device to USB port
+- Install examples with 'make examples' if doesn’t work re-insert key and try again
+
+## Notes
+- During operation applications will need 2s delay between close and re-openign NCS device
+- VM RAM needs to be 2GB or caffe compile will likely fail
-theme: jekyll-theme-architect
\ No newline at end of file
+title: Movidius™ Neural Compute SDK
+description: Home of the ncsdk used to develop software for the Movidius™ Neural Compute Stick
+theme: jekyll-theme-modernist
--- /dev/null
+<!doctype html>
+<html lang="{{ site.lang | default: "en-US" }}">
+ <head>
+ <meta charset="utf-8">
+ <meta http-equiv="X-UA-Compatible" content="chrome=1">
+
+ <link href='https://fonts.googleapis.com/css?family=Lato:300italic,700italic,300,700' rel='stylesheet' type='text/css'>
+ <link rel="stylesheet" href="{{ '/assets/css/style.css?v=' | append: site.github.build_revision | relative_url }}">
+ <script src="{{ '/assets/js/scale.fix.js' | relative_url }}"></script>
+ <meta name="viewport" content="width=device-width, initial-scale=1, user-scalable=no">
+
+ <!--[if lt IE 9]>
+ <script src="//html5shiv.googlecode.com/svn/trunk/html5.js"></script>
+ <![endif]-->
+ </head>
+ <body>
+ <div class="wrapper">
+ <header {% unless site.description or site.github.project_tagline %} class="without-description" {% endunless %}>
+ <h1 style="color:#FFFFFF;text-shadow:none">{{ site.title | default: site.github.repository_name }}</h1>
+ {% if site.description or site.github.project_tagline %}
+ <p style="color:#FFFFFF;">{{ site.description | default: site.github.project_tagline }}</p>
+ {% endif %}
+ <p class="view"><a href="{{ site.github.repository_url }}">View the Project on GitHub <small>{{ github_name }}</small></a></p>
+ <ul>
+ {% if site.show_downloads %}
+ <li><a href="{{ site.github.zip_url }}">Download <strong>ZIP File</strong></a></li>
+ <li><a href="{{ site.github.tar_url }}">Download <strong>TAR Ball</strong></a></li>
+ {% endif %}
+ <li> <a href={{ "index.html#Introduction" | absolute_url }}><small>Docs Home</small></a> </li>
+ <li> <a href={{ "TOC.html" | absolute_url }}><small>Table of Contents</small></a> </li>
+ <li> <a href={{ "index.html#NcSdkTools" | absolute_url }}><small>SDK Tools</small></a> </li>
+ <li> <a href={{ "index.html#NcApi" | absolute_url }}><small>Neural Compute API</small></a>
+ <li> <a href="{{ site.github.repository_url }}">View On <strong>GitHub</strong></a></li>
+ </ul>
+ </header>
+
+ <section>
+
+ {{ content }}
+
+ </section>
+ </div>
+ <footer>
+ {% if site.github.is_project_page %}
+ <p>Project maintained by <a href="{{ site.github.owner_url }}">{{ site.github.owner_name }}</a></p>
+ {% endif %}
+
+ </footer>
+ <!--[if !IE]><script>fixScale(document);</script><![endif]-->
+
+ {% if site.google_analytics %}
+ <script type="text/javascript">
+ (function(i,s,o,g,r,a,m){i['GoogleAnalyticsObject']=r;i[r]=i[r]||function(){
+ (i[r].q=i[r].q||[]).push(arguments)},i[r].l=1*new Date();a=s.createElement(o),
+ m=s.getElementsByTagName(o)[0];a.async=1;a.src=g;m.parentNode.insertBefore(a,m)
+ })(window,document,'script','//www.google-analytics.com/analytics.js','ga');
+ ga('create', '{{ site.google_analytics }}', 'auto');
+ ga('send', 'pageview');
+ </script>
+ {% endif %}
+ </body>
+</html>
--- /dev/null
+---
+---
+
+@import "{{ site.theme }}";
+
+header
+{
+ background:#0066D7;
+ color:#00FF00
+}
+
+header ul
+{
+ background:#2a2e67;
+}
+
+ul a
+{
+ color:#0066d7;
+}
+
+a
+{
+ color:#0066d7;
+}
+
+p
+{
+ color:#61778B;
+ font-weight: bold;
+}
+
+.wrapper
+{
+ width: 900px;
+ background:#FAFAFA;
+}
--- /dev/null
+# mvncAllocateGraph()
+
+Type|Function
+------------ | -------------
+Header|mvnc.h
+Library| libmvnc.so
+Return|[mvncStatus](mvncStatus.md)
+Version|1.0
+See also|[mvncOpenDevice](mvncOpenDevice.md), [mvncDeallocateGraph](mvncDeallocateGraph.md)
+
+## Overview
+This function allocates a graph on the specified device and creates a handle to the graph which can be passed to other API functions such as mvncLoadTensor() and mvncGetResult(). When the caller is done with the graph the mvncDeallocateGraph() function should be called to free the resources associated with the graph.
+
+## Prototype
+
+```C
+mvncStatus mvncAllocateGraph(void *deviceHandle, void **graphHandle, const void *graphFile, unsigned int graphFileLength);
+```
+## Parameters
+
+Name|Type|Description
+----|----|-----------
+deviceHandle|void \*|The deviceHandle pointer to the opaque device datatype (which was created via mvncOpenDevice()) on which the graph should be allocated.
+graphHandle|void\*\*|Address of a pointer that will be set to point to an opaque graph datatype. Upon successful return this graphHandle can be passed to other API funtions.
+graphFile|const void\* | Pointer to a buffer that contains the contents of a graph file. The graph file is a compiled neural network file that gets created by the MvNCCompile SDK tool.
+graphFileLength|unsigned int|The number of bytes allocated for the buffer that graphFile points to.
+
+## Return
+This function returns an appropriate value from the [mvncStatus](mvncStatus.md) enumeration.
+
+## Known Issues
+- Be sure to call mvncDeallocateGraph() for the graphHandle returned from this function.
+
+## Example
+```C
+
+// graph file name
+#define GRAPH_FILE_NAME "graphfile"
+
+int main(int argc, char** argv)
+{
+ void* deviceHandle;
+
+ //
+ // Assume NCS device opened here and deviceHandle is valid now
+ //
+
+ // Now read in a graph file so graphFileBuf will point to the
+ // bytes of the file in a memory buffer and graphFileLen will be
+ // set to the number of bytes in the graph file and memory buffer.
+ unsigned int graphFileLen;
+ void* graphFileBuf = LoadFile(GRAPH_FILE_NAME, &graphFileLen);
+
+ // allocate the graph
+ void* graphHandle;
+ retCode = mvncAllocateGraph(deviceHandle, &graphHandle, graphFileBuf, graphFileLen);
+ if (retCode != MVNC_OK)
+ { // error allocating graph
+ printf("Could not allocate graph for file: %s\n", GRAPH_FILE_NAME);
+ }
+ else
+ { // successfully allocated graph. Now graphHandle is ready to go.
+ // use graphHandle for other API calls and call mvncDeallocateGraph
+ // when done with it.
+ printf("Successfully allocated graph for %s\n", GRAPH_FILE_NAME);
+
+ retCode = mvncDeallocateGraph(graphHandle);
+ graphHandle = NULL;
+ }
+
+ free(graphFileBuf);
+ retCode = mvncCloseDevice(deviceHandle);
+ deviceHandle = NULL;
+}
+
+```
--- /dev/null
+# mvncCloseDevice()
+
+Type|Function
+------------ | -------------
+Header|mvnc.h
+Library| libmvnc.so
+Return|[mvncStatus](mvncStatus.md)
+Version|1.0
+See also|[mvncOpenDevice](mvncOpenDevice.md), [mvncGetDeviceOption](mvncGetDeviceOption.md), [mvncSetDeviceOption](mvncSetDeviceOption.md)
+
+## Overview
+This function is used to cease communication and reset the device.
+
+## Prototype
+
+```C
+mvncStatus mvncCloseDevice(void *deviceHandle);
+```
+## Parameters
+
+Name|Type|Description
+----|----|-----------
+deviceHandle|void*|Pointer to the opaque NCS Device structure that was allocated and returned from the mvncOpenDevice function.
+
+## Return
+This function returns an appropriate value from the [mvncStatus](mvncStatus.md) enumeration.
+
+## Known Issues
+
+## Example
+```C
+#include <stdio.h>
+#include <stdlib.h>
+
+extern "C"
+{
+#include <mvnc.h>
+}
+// somewhat arbitrary buffer size for the device name
+#define NAME_SIZE 100
+int main(int argc, char** argv)
+{
+ mvncStatus retCode;
+ void *deviceHandle;
+ char devName[NAME_SIZE];
+ retCode = mvncGetDeviceName(0, devName, NAME_SIZE);
+ if (retCode != MVNC_OK)
+ { // failed to get device name, maybe none plugged in.
+ printf("No NCS devices found\n");
+ exit(-1);
+ }
+
+ // Try to open the NCS device via the device name
+ retCode = mvncOpenDevice(devName, &deviceHandle);
+ if (retCode != MVNC_OK)
+ { // failed to open the device.
+ printf("Could not open NCS device\n");
+ exit(-1);
+ }
+
+ // deviceHandle is ready to use now.
+ // Pass it to other NC API calls as needed and close it when finished.
+ printf("Successfully opened NCS device!\n");
+
+ // Close the device previously opened by mvncOpenDevice()
+ retCode = mvncCloseDevice(deviceHandle);
+}
+```
--- /dev/null
+# mvncDeallocateGraph()
+
+Type|Function
+------------ | -------------
+Header|mvnc.h
+Library| libmvnc.so
+Return|[mvncStatus](mvncStatus.md)
+Version|1.0
+See also|[mvncOpenDevice](mvncOpenDevice.md), [mvncAllocateGraph](mvncAllocateGraph.md)
+
+## Overview
+This function deallocates a graph that was previously allocated with mvncAllocateGraph(). After successful return from this function the passed graphHandle will be invalid and should not be used.
+
+## Prototype
+
+```C
+mvncStatus mvncDeallocateGraph(void *graphHandle);
+```
+## Parameters
+
+Name|Type|Description
+----|----|-----------
+graphHandle|void\*\*|Pointer to opaque graph data type that was initialized with the mvncAllocateGraph() function.
+
+## Return
+This function returns an appropriate value from the [mvncStatus](mvncStatus.md) enumeration.
+
+## Known Issues
+Using a deallocated graph handle can lead to hard to find bugs. To prevent this it's good practice to set the handle to NULL (or nullptr for C++ 11) after deallocating as shown in this code snippet.
+```C++
+mvncDeallocateGraph(graphHandle);
+graphHandle = NULL;
+```
+
+## Example
+```C
+// graph file name
+#define GRAPH_FILE_NAME "graphfile"
+
+int main(int argc, char** argv)
+{
+ void* deviceHandle;
+
+ .
+ .
+ .
+
+ //
+ // Assume NCS device opened here and deviceHandle is valid now
+ // and the graph file is in graphFileBuf and length in bytes
+ // is in graphFileLen
+ //
+
+ // allocate the graph
+ void* graphHandle;
+ retCode = mvncAllocateGraph(deviceHandle, &graphHandle, graphFileBuf, graphFileLen);
+ if (retCode != MVNC_OK)
+ { // error allocating graph
+ printf("Could not allocate graph for file: %s\n", GRAPH_FILE_NAME);
+ }
+ else
+ { // successfully allocated graph. Now graphHandle is ready to go.
+ // use graphHandle for other API calls and call mvncDeallocateGraph
+ // when done with it.
+ printf("Successfully allocated graph for %s\n", GRAPH_FILE_NAME);
+
+ retCode = mvncDeallocateGraph(graphHandle);
+ graphHandle = NULL;
+ }
+.
+.
+.
+
+}
+```
--- /dev/null
+# mvncDeviceOptions enumeration
+
+Info | Value/s
+------------ | -------------
+Header|mvnc.h
+Version|1.0
+See also|[mvncGetDeviceOption](mvncGetDeviceOption.md), [mvncSetDeviceOption](mvncSetDeviceOption.md), [mvncOpenDevice](mvncOpenDevice.md)
+
+## Overview
+
+This enumeration is used to specify an option on the NCS device that can be written or read via mvncGetDeviceOption() and mvncSetDeviceOption(). The table below provides details on the meaning of each of the values in the enumeration.
+
+constant | Option Type | Possible Values | get/set | Description
+-------- | ------------| --------------- | ------- | -----------
+MVNC_THERMAL_THROTTLING_LEVEL|int|1, or 2|get|Returns 1 if lower guard temperature threshold of chip sensor is reached. This indicates short throttling time is in action between inferences to protect the device. Returns 2 if upper guard temperature of chip sensor is reached. This indicates long throttling time is in action between inferences to protect the device.
--- /dev/null
+# mvncGetDeviceName()
+
+Type|Function
+------------ | -------------
+Header|mvnc.h
+Library| libmvnc.so
+Return|[mvncStatus](mvncStatus.md)
+Version|1.0
+See also|[mvncOpenDevice](mvncOpenDevice.md)
+
+## Overview
+This function is used to get the name of a particular NCS device. Typical usage is to call the function repeatedly starting with index = 0 and incrementing the index each time until an error is returned. These successive calls will give you the names of all the devices in the system.
+
+## Prototype
+
+```C
+mvncStatus mvncGetDeviceName(int index, char* name, unsigned int nameSize);
+```
+## Parameters
+
+Name|Type|Description
+----|----|-----------
+index|int|index of the device for which the name should be retrieved.
+name|char\*|pointer to a character buffer into which the name will be copied. This buffer should be allocated by the caller.
+nameSize|unsigned int| the number of characters allocated to the buffer pointed to by the name parameter.
+
+## Return
+This function returns an appropriate value from the [mvncStatus](mvncStatus.md) enumeration.
+
+## Known Issues
+
+## Example
+The following example shows how to get the name of all NCS devices attached to the system. mvncGetDeviceName is called repeatedly until it returns MVNC_DEVICE_NOT_FOUND.
+
+```C++
+#include <stdio.h>
+extern "C"
+{
+#include <mvnc.h>
+}
+// somewhat arbitrary buffer size for the device name
+#define NAME_SIZE 100
+int main(int argc, char** argv)
+{
+ mvncStatus retCode;
+ int deviceCount = 0;
+ char devName[NAME_SIZE];
+ while ((retCode = mvncGetDeviceName(deviceCount, devName, NAME_SIZE)) != MVNC_DEVICE_NOT_FOUND)
+ {
+ printf("Found NCS device named: \"%s\"\n", devName);
+ deviceCount++;
+ }
+ printf("Total number of NCS devices found: %d\n", deviceCount);
+}
+```
+Output from the example code above with two NCS devices in the system.
+
+```
+Found NCS device named: "2.3"
+Found NCS device named: "1"
+Total number of NCS devices found: 2
+```
--- /dev/null
+# mvncGetDeviceOption()
+
+Type|Function
+------------ | -------------
+Header|mvnc.h
+Library| libmvnc.so
+Return|[mvncStatus](mvncStatus.md)
+Version|1.0
+See also|[mvncOpenDevice](mvncOpenDevice.md), [mvncDeviceOptions](mvncDeviceOptions.md), [mvncSetDeviceOption](mvncSetDeviceOption.md)
+
+## Overview
+This function gets the current value of an option for an NCS device. The available options and their data types can be found in the [DeviceOptions](mvncDeviceOptions.md) enumeration documentation.
+
+## Prototype
+
+```C
+mvncStatus mvncGetDeviceOption(void *deviceHandle, int option, void *data, unsigned int *datalength);
+```
+## Parameters
+
+Name|Type|Description
+----|----|-----------
+deviceHandle|void\*|Pointer to opaque device data type that was initialized with the mvncOpenDevice() function. This specifies which NCS device's option will be retrieved.
+option|int|A value from the DeviceOptions enumeration that specifies which option will be retrieved.
+data|void\*|Pointer to a buffer where the value of the option will be copied. The type of data this points to will depend on the option that is specified. Check mvncDeviceOptions for the data types that each option requires.
+dataLength|unsigned int \*| Pointer to an unsigned int which must point to the size, in bytes, of the buffer allocated to the data parameter when called. Upon successfull return it will be set to the number of bytes copied to the data buffer.
+
+## Return
+This function returns an appropriate value from the [mvncStatus](mvncStatus.md) enumeration.
+
+## Known Issues
+
+## Example
+```C
+TODO
+```
--- /dev/null
+
+# mvncGetGlobalOption()
+
+Type|Function
+------------ | -------------
+Header|mvnc.h
+Library| libmvnc.so
+Return|[mvncStatus](mvncStatus.md)
+Version|1.09
+See also|[mvncGlobalOptions](mvncGlobalOptions.md), [mvncSetGlobalOption](mvncSetGlobalOption.md)
+
+## Overview
+This function gets the current value of an option that is global to the SDK. The available options and their data types can be found in the [mvncGlobalOptions](mvncGlobalOptions.md) enumeration documentation.
+
+## Prototype
+
+```C
+mvncStatus mvncGetGlobalOption(int option, void *data, unsigned int *datalength);
+```
+## Parameters
+
+Name|Type|Description
+----|----|-----------
+option|int|A value from the GlobalOptions enumeration that specifies which option will be retrieved.
+data|void\*|Pointer to a buffer where the value of the option will be copied. The type of data this points to will depend on the option that is specified. Check mvncGlobalOptions for the data types that each option requires.
+dataLength|unsigned int \*| Pointer to an unsigned int which must point to the size, in bytes, of the buffer allocated to the data parameter when called. Upon successfull return it will be set to the number of bytes copied to the data buffer.
+
+## Return
+This function returns an appropriate value from the [mvncStatus](mvncStatus.md) enumeration.
+
+## Known Issues
+
+## Example
+```C
+int len;
+int loggingLevel;
+status = mvncGetGlobalOption(MVNC_LOGLEVEL, &loggingLevel, &len);
+// loggingLevel has the option value now.
+```
--- /dev/null
+# mvncGetGraphOption()
+
+Type|Function
+------------ | -------------
+Header|mvnc.h
+Library| libmvnc.so
+Return|[mvncStatus](mvncStatus.md)
+Version|1.0
+See also|[mvncAllocateGraph](mvncAllocateGraph.md), [mvncGraphOptions](mvncGraphOptions.md), [mvncSetGraphOption](mvncSetGraphOption.md)
+
+## Overview
+This function gets the current value of an option for a graph. The available options can be found in the GraphOptions enumeration.
+
+## Prototype
+
+```C
+mvncStatus mvncGetGraphOption(void *graphHandle, int option, void **data, unsigned int *datalength);
+```
+## Parameters
+
+Name|Type|Description
+----|----|-----------
+graphHandle|void\*|Pointer to opaque graph data type that was initialized with the mvncAllocateGraph() function which represents the neural network. This specifies which graph's option value will be retrieved.
+option|int|A value from the GraphOptions enumeration that specifies which option will be retrieved.
+data|void\*|Pointer to a buffer where the value of the option will be copied. The type of data this points to will depend on the option that is specified. Check mvncGraphOptions for the data types that each option requires.
+dataLength|unsigned int \*| Pointer to an unsigned int which must point to the size, in bytes, of the buffer allocated to the data parameter when called. Upon successfull return it will be set to the number of bytes copied to the data buffer.
+
+## Return
+This function returns an appropriate value from the [mvncStatus](mvncStatus.md) enumeration.
+
+## Known Issues
+
+## Example
+```C++
+.
+.
+.
+ // open device to get device handle,
+ // allocate the graph to get graph handle
+
+ // set the graph option for blocking calls
+ int dontBlockValue;
+ unsigned int sizeOfValue;
+ retCode = mvncGetGraphOption(graphHandle, MVNC_DONTBLOCK, (void**)(&dontBlockValue), &sizeOfValue);
+ if (retCode == MVNC_OK)
+ {
+ printf("Successfully got graph option, value is: %d\n", dontBlockValue);
+ }
+ else
+ {
+ printf("Could not get graph option\n");
+ printf("Error returned from mvncGetGraphOption: %d\n", retCode);
+ }
+
+ // use graph, deallocate graph, close device etc.
+
+ .
+ .
+ .
+
+```
--- /dev/null
+# mvncGetResult()
+
+Type|Function
+------------ | -------------
+Header|mvnc.h
+Library| libmvnc.so
+Return|[mvncStatus](mvncStatus.md)
+Version|1.0
+See also|[mvncOpenDevice](mvncOpenDevice.md), [mvncAllocateGraph](mvncAllocateGraph.md), [mvncLoadTensor](mvncLoadTensor.md)
+
+## Overview
+This function retrieves the result of an inference that was initiated via LoadTensor() on the specified graph.
+
+## Prototype
+
+```C
+mvncStatus mvncGetResult(void *graphHandle, void **outputData, unsigned int *outputDataLength, void **userParam);
+```
+## Parameters
+
+Name|Type|Description
+----|----|-----------
+graphHandle|void\*|Pointer to opaque graph data type that was initialized with the mvncAllocateGraph() function that represents the neural network for which an inference was initiated.
+outputData|void\*\*|Address of the pointer that will be set to a buffer of 16 bit floats which contain the result of the inference. The buffer will contain one 16 bit float for each network category, the values of which are the results of the output node. Typically these values are the probabilities that an image belongs to the category.
+outputDataLength|unsigned int\*|Pointer to an unsigned int that will be set to the number of bytes in the outputData buffer.
+userParam|void \*\*| Address of a pointer that will be set to the user parameter for this inference. This corresponds to the userParam that was passed to the LoadTensor() function which initiated the inference.
+
+## Return
+This function returns an appropriate value from the [mvncStatus](mvncStatus.md) enumeration.
+
+## Known Issues
+
+## Example
+```C++
+.
+.
+.
+
+// use a 16 bit unsigned type to represent half precision floats since C++ has no
+// built in support for 16 but floats.
+typedef unsigned short half;
+
+int main(int argc, char** argv)
+{
+.
+.
+.
+ //
+ // Open NCS device and set deviceHandle to the valid handle
+ //
+
+ //
+ // Read graph from disk and call mvncAllocateGraph to set graphHandle appropriately
+ //
+
+ //
+ // Load an image.png from disk and preprocess it to match network
+ // requirements so that imageBufFp16 list to 16 bit floats
+ //
+
+ // start the inference with call to mvncLoadTensor()
+ retCode = mvncLoadTensor(graphHandle, imageBufFp16, lenBufFp16, NULL);
+ if (retCode == MVNC_OK)
+ { // the inference has been started, now call mvncGetResult() for the
+ // inference result
+ printf("Successfully loaded the tensor for image %s\n", "image.png");
+
+ void* resultData16;
+ void* userParam;
+ unsigned int lenResultData;
+ retCode = mvncGetResult(graphHandle, &resultData16, &lenResultData, &userParam);
+ if (retCode == MVNC_OK)
+ { // Successfully got the result. The inference result is in the buffer pointed to by resultData
+ printf("Successfully got the inference result for image %s\n", IMAGE_FILE_NAME);
+ printf("resultData is %d bytes which is %d 16-bit floats.\n", lenResultData, lenResultData/(int)sizeof(half));
+
+ // convert half precision floats to full floats
+ int numResults = lenResultData / sizeof(half);
+ float* resultData32;
+ resultData32 = (float*)malloc(numResults * sizeof(*resultData32));
+ fp16tofloat(resultData32, (unsigned char*)resultData16, numResults);
+
+ float maxResult = 0.0;
+ int maxIndex = -1;
+ for (int index = 0; index < numResults; index++)
+ {
+ printf("Category %d is: %f\n", index, resultData32[index]);
+ if (resultData32[index] > maxResult)
+ {
+ maxResult = resultData32[index];
+ maxIndex = index;
+ }
+ }
+ printf("index of top result is: %d - probability of top result is: %f\n", maxIndex, resultData32[maxIndex]);
+ free(resultData32);
+ }
+ }
+
+ //
+ // call mvncDeallocateGraph to free the resources tied to graphHandle
+ // close the device with mvncCloseDevice()
+ //
+}
+
+```
--- /dev/null
+# mvncGlobalOptions enumeration
+
+Info | Value/s
+------------ | -------------
+Header|mvnc.h
+Version|1.09
+See also|[mvncGetGlobalOption](mvncGetGlobalOption.md), [mvncSetGlobalOption](mvncSetGlobalOption.md)
+
+## Overview
+
+This enumeration is used to specify a global option that can be written or read via mvncGetDeviceOption() and mvncSetDeviceOption(). The table below provides details on the meaning of each of the values in the enumeration.
+
+constant | Option Type | Possible Values | get/set | Description
+-------- | ------------| --------------- | ------- | -----------
+MVNC_LOGLEVEL | int | 0, 1, 2 |get, set|The logging level for application Value meanings are: 0 = log nothing (default), 1 = log errors only, 2 = log all, verbose logging.
--- /dev/null
+# mvncGraphOptions enumeration
+
+Info | Value/s
+------------ | -------------
+Header|mvnc.h
+Version|1.0
+See also|[mvncGetGraphOption](mvncGetGraphOption.md), [mvncSetGraphOption](mvncSetGraphOption.md), [mvncAllocateGraph](mvncAllocateGraph.md)
+
+## Overview
+
+
+This enumeration is used to specify an option on the a graph that can be written or read via mvncGetGraphOption() and mvncSetGraphOption(). The table below provides details on the meaning of each of the values in the enumeration.
+
+constant | Option Type | Possible Values | get/set | Description
+-------- | ------------| --------------- | ------- | -----------
+MVNC_DONT_BLOCK| int |0, 1|get, set|0: Calls to mvncLoadTensor and mvncGetResult block, 1: calls to those functions don't block.
+MVNC_TIME_TAKEN| float\* | any | get |Time taken for the last inference returned by mvncGetResult.
+MVNC_DEBUG_INFO| char\* | any | get | A string that provides more details when the result of a function call was MVNC_MYRIADERROR.
--- /dev/null
+# mvncLoadTensor()
+
+Type|Function
+------------ | -------------
+Header|mvnc.h
+Library| libmvnc.so
+Return|[mvncStatus](mvncStatus.md)
+Version|1.0
+See also|[mvncOpenDevice](mvncOpenDevice.md), [mvncAllocateGraph](mvncAllocateGraph.md), [mvncGetResult](mvncGetResult.md)
+
+## Overview
+This function initiates an inference on the specified graph via the associated NCS device. After calling this function use the mvncGetResult() function to retrieve the inference result.
+
+## Prototype
+
+```C
+mvncStatus mvncLoadTensor(void *graphHandle, const void *inputTensor, unsigned int inputTensorLength, void *userParam);
+```
+## Parameters
+
+Name|Type|Description
+----|----|-----------
+graphHandle|void\*|Pointer to opaque graph data type that was initialized with the mvncAllocateGraph() function that represents the neural network for which an inference will be initiated.
+inputTensor|const void\*|Pointer to tensor data buffer which contains 16 bit half precision floats (per IEEE 754 half precision binary floating-point format: binary16). The values in the buffer are dependent on the neural network (graph) but are typically representations of each color channel of each pixel of an image.
+inputTensorLength|unsigned int|The length, in bytes, of the buffer pointed to by the inputTensor parameter.
+userParam|void \*| Pointer to user data that will be returned along with the inference result from the GetResult() function.
+
+## Return
+This function returns an appropriate value from the [mvncStatus](mvncStatus.md) enumeration.
+
+## Known Issues
+
+## Example
+```C+
+.
+.
+.
+
+// use a 16 bit unsigned type to represent half precision floats since C++ has no
+// built in support for 16 but floats.
+typedef unsigned short half;
+
+// GoogleNet image dimensions and network mean values for each channel. This information is specific
+// for each network, and usually available from network creators.
+const int networkDim = 224;
+float networkMean[] = {0.40787054*255.0, 0.45752458*255.0, 0.48109378*255.0};
+
+int main(int argc, char** argv)
+{
+.
+.
+.
+ //
+ // Open NCS device and set deviceHandle to the valid handle
+ //
+
+ //
+ // Read graph from disk and call mvncAllocateGraph to set graphHandle appropriately
+ //
+
+ //
+ // Load an image from disk
+ // LoadImage will read image from disk, convert channels to floats
+ // subtract network mean for each value in each channel. Then convert
+ // floats to half precision floats.
+ // return pointer to the buffer of half precision floats
+ half* imageBufFp16 = LoadImage("image.png", networkDim, networkMean);
+
+ // calculate the length of the buffer that contains the half precision floats.
+ // 3 channels * width * height * sizeof a 16bit float
+ unsigned int lenBufFp16 = 3*networkDim*networkDim*sizeof(*imageBufFp16);
+
+ // start the inference with mvncLoadTensor()
+ retCode = mvncLoadTensor(graphHandle, imageBufFp16, lenBufFp16, NULL);
+ if (retCode == MVNC_OK)
+ { // the inference has been started, now call mvncGetResult() for the
+ // inference result
+ printf("Successfully loaded the tensor for image %s\n", "image.png");
+
+ // here mvncGetResult() can be called to get the result of the inference
+ // that was started with mvncLoadTensor() above
+ }
+
+ //
+ // call mvncDeallocateGraph to free the resources tied to graphHandle
+ // close the device with mvncCloseDevice()
+ //
+}
+
+```
--- /dev/null
+# mvncOpenDevice()
+
+Type|Function
+------------ | -------------
+Header|mvnc.h
+Library| libmvnc.so
+Return|[mvncStatus](mvncStatus.md)
+Version|1.0
+See also|[mvncCloseDevice](mvncCloseDevice.md), [mvncGetDeviceName](mvncGetDeviceName.md) [mvncGetDeviceOption](mvncGetDeviceOption.md), [mvncSetDeviceOption](mvncSetDeviceOption.md)
+
+## Overview
+This function is used to initialize the NCS device and return a device handle that can be passed to other API functions.
+
+## Prototype
+
+```C
+mvncStatus mvncOpenDevice(const char *name, void **deviceHandle);
+```
+
+## Parameters
+
+Name|Type|Description
+----|----|------------
+name|const char\*|Pointer to a constant array of chars that contains the name of the device to open. This value is obtained from mvncGetDeviceName.
+deviceHandle|void \*\*|Address of a pointer that will be set to point to an opaque structure representing an NCS device.
+
+## Return
+This function returns an appropriate value from the [mvncStatus](mvncStatus.md) enumeration.
+
+## Known Issues
+
+## Example
+In the example below the code gets the name of the first device and then calls mvncOpenDevice to open the device and set the deviceHandle variable for use to other API calls that expect a device handle for an open device.
+```C++
+#include <stdio.h>
+#include <stdlib.h>
+
+extern "C"
+{
+#include <mvnc.h>
+}
+// somewhat arbitrary buffer size for the device name
+#define NAME_SIZE 100
+int main(int argc, char** argv)
+{
+ mvncStatus retCode;
+ void *deviceHandle;
+ char devName[NAME_SIZE];
+ retCode = mvncGetDeviceName(0, devName, NAME_SIZE);
+ if (retCode != MVNC_OK)
+ { // failed to get device name, maybe none plugged in.
+ printf("No NCS devices found\n");
+ exit(-1);
+ }
+
+ // Try to open the NCS device via the device name
+ retCode = mvncOpenDevice(devName, &deviceHandle);
+ if (retCode != MVNC_OK)
+ { // failed to open the device.
+ printf("Could not open NCS device\n");
+ exit(-1);
+ }
+
+ // deviceHandle is ready to use now.
+ // Pass it to other NC API calls as needed and close it when finished.
+ printf("Successfully opened NCS device!\n");
+
+ retCode = mvncCloseDevice(deviceHandle);
+}
+
+```
--- /dev/null
+# mvncSetDeviceOption()
+
+Type|Function
+------------ | -------------
+Header|mvnc.h
+Library| libmvnc.so
+Return|[mvncStatus](mvncStatus.md)
+Version|1.0
+See also|[mvncOpenDevice](mvncOpenDevice.md), [mvncDeviceOptions](mvncDeviceOptions.md), [mvncGetDeviceOption](mvncGetDeviceOption.md)
+
+## Overview
+This function sets an option for a specific NCS device. The available options can be found in the [DeviceOptions](mvncDeviceOptions.md) enumeration.
+
+## Prototype
+
+```C
+mvncStatus mvncSetDeviceOption(void *deviceHandle, int option, const void *data, unsigned int datalength);
+```
+## Parameters
+
+Name|Type|Description
+----|----|-----------
+deviceHandle|void\*|Pointer to opaque device data type that was initialized with the mvncOpenDevice() function. This specifies which device's option will be set.
+option|int|A value from the DeviceOptions enumeration that specifies which option will be set.
+data|const void\*|Pointer to the data for the new value for the option. The type of data this points to depends on the option that is being set. Check mvncDeviceOptions for the data types that each option requires.
+dataLength|unsigned int| An unsigned int that contains the length, in bytes, of the buffer that the data parameter points to.
+
+## Return
+This function returns an appropriate value from the [mvncStatus](mvncStatus.md) enumeration.
+
+## Known Issues
+
+## Example
+```C
+TBD
+```
--- /dev/null
+# mvncSetGlobalOption()
+
+Type|Function
+------------ | -------------
+Header|mvnc.h
+Library| libmvnc.so
+Return|[mvncStatus](mvncStatus.md)
+Version|1.09
+See also|[mvncGlobalOptions](mvncGlobalOptions.md), [mvncGetGlobalOption](mvncGetGlobalOption.md)
+
+## Overview
+This function sets an option that is global for an application using the SDK. The available options can be found in the [mvncGlobalOptions](mvncGlobalOptions.md) enumeration.
+
+## Prototype
+
+```C
+mvncStatus mvncSetGlobalOption(int option, const void *data, unsigned int datalength);
+```
+## Parameters
+
+Name|Type|Description
+----|----|-----------
+option|int|A value from the mvncGlobalOptions enumeration that specifies which option will be set.
+data|const void\*|Pointer to the data for the new value for the option. The type of data this points to depends on the option that is being set. Check mvncGlobalOptions for the data types that each option requires.
+dataLength|unsigned int| An unsigned int that contains the length, in bytes, of the buffer that the data parameter points to.
+
+## Return
+This function returns an appropriate value from the [mvncStatus](mvncStatus.md) enumeration.
+
+## Known Issues
+
+## Example
+```C
+int logLevel = 3;
+retCode = mvncSetGlobalOption(MVNC_LOGLEVEL, &logLevel, sizeof(int));
+
+```
--- /dev/null
+# mvncSetGraphOption()
+
+Type|Function
+------------ | -------------
+Header|mvnc.h
+Library| libmvnc.so
+Return|[mvncStatus](mvncStatus.md)
+Version|1.0
+See also|[mvncAllocateGraph](mvncAllocateGraph.md), [mvncGraphOptions](mvncGraphOptions.md), [mvncGetGraphOption](mvncGetGraphOption.md)
+
+## Overview
+This function sets an option of a graph. The available options can be found in the GraphOptions enumeration.
+
+## Prototype
+
+```C
+mvncStatus mvncSetGraphOption(void *graphHandle, int option, const void *data, unsigned int datalength);
+```
+## Parameters
+
+Name|Type|Description
+----|----|-----------
+graphHandle|void\*|Pointer to opaque graph data type that was initialized with the mvncAllocateGraph() function that represents the neural network. This specifies which graph's option will be set.
+option|int|A value from the GraphOptions enumeration that specifies which option will be set.
+data|const void\*|Pointer to the data for the new value for the option. The type of data this points to depends on the option that is being set. Check mvncGraphOptions for the data types that each option requires.
+dataLength|unsigned int| An unsigned int that contains the length, in bytes, of the buffer that the data parameter points to.
+
+## Return
+This function returns an appropriate value from the [mvncStatus](mvncStatus.md) enumeration.
+
+## Known Issues
+
+## Example
+```C++
+.
+.
+.
+ // open NCS device to initialize deviceHandle.
+ // read compiled graph file into graphFileBuf and put length of it in graphFileLen
+
+ // allocate the graph
+ void* graphHandle;
+ retCode = mvncAllocateGraph(deviceHandle, &graphHandle, graphFileBuf, graphFileLen);
+ if (retCode != MVNC_OK)
+ { // error allocating graph
+ printf("Could not allocate graph for file: %s\n", GRAPH_FILE_NAME);
+ }
+ else
+ { // successfully allocated graph. Now graphHandle is ready to go.
+ // use graphHandle for other API calls and call mvncDeallocateGraph
+ // when done with it.
+ printf("Successfully allocated graph for %s\n", GRAPH_FILE_NAME);
+
+ // set the graph option for blocking calls
+ int dontBlockValue = 0;
+ retCode = mvncSetGraphOption(graphHandle, MVNC_DONTBLOCK, &dontBlockValue, sizeof(int));
+ if (retCode == MVNC_OK)
+ {
+ printf("Successfully set graph option\n");
+ }
+ else
+ {
+ printf("Could not set graph option\n");
+ printf("Error returned from mvncSetGraphOption: %d\n", retCode);
+ }
+
+ // use graphHandle here with the option set above.
+ // then deallocate the graph and close the device
+ }
+.
+.
+.
+```
--- /dev/null
+# mvncStatus enumeration
+
+Info | Value/s
+------------ | -------------
+Header |mvnc.h
+Version |1.0
+See also |
+
+## Overview
+A value from this enumeration is returned from most of the C API functions. The table below provides details on the meaning of each of the values in the enumeration.
+
+constant | description
+-------- | -----------
+MVNC_OK | The function call worked as expected.
+MVNC_BUSY | The device is busy, retry later.
+MVNC_ERROR | An unexpected error was encountered during the function call.
+MVNC_OUT_OF_MEMORY | The host is out of memory.
+MVNC_DEVICE_NOT_FOUND | There is no device at the given index or name.
+MVNC_INVALID_PARAMETERS | At least one of the given parameters is invalid in the context of the function call.
+MVNC_TIMEOUT | Timeout in the communication with the device.
+MVNC_MVCMD_NOT_FOUND | The file named MvNCAPI.mvcmd should be installed in the mvnc directory. This message may mean that the installation failed.
+MVNC_NO_DATA | No data to return.
+MVNC_GONE | The graph or device has been closed during the operation.
+MVNC_UNSUPPORTED_GRAPH_FILE | The graph file may have been created with an incompatible prior version of the Toolkit. Try to recompile the graph file with the version of the Toolkit that corresponds to the API version.
+MVNC_MYRIAD_ERROR | An error has been reported by Movidius™ VPU. Use MVNC_DEBUGINFO to get more information.
--- /dev/null
+# Movidius™ Neural Compute SDK C API
+
+The SDK comes with a C Language API that enables developers to create applications in C or C++ which utilize hardware-accelerated Deep Neural Networks via the Movidius™ Neural Compute Stick (NCS.) The C API is provided as a header file (mvnc.h) and an associated library file (libmvnc.so) both of which are placed on the development computer when the SDK is installed. Details of the C API are provided below and within the documentation linked from here.
+
+## Enumerations
+- [mvncStatus](mvncStatus.md): Contains possible return values for API functions.
+- [mvncDeviceOptions](mvncDeviceOptions.md): Contains all possible options to get/set for an NCS device, and their data types.
+- [mvncGraphOptions](mvncGraphOptions.md): Contains all possible options to get/set for a graph, and their data types.
+- [mvncGlobalOptions](mvncGlobalOptions.md): Contains all possible global options to get/set and their data types.
+
+## Functions
+- [mvncGetDeviceName](mvncGetDeviceName.md): Retrieves the name of an NCS device that can be opened.
+- [mvncOpenDevice](mvncOpenDevice.md): Opens an NCS device for use by the application.
+- [mvncAllocateGraph](mvncAllocateGraph.md): Allocates a graph for a specific NCS device in preparation for computing inferences.
+- [mvncDeallocateGraph](mvncDeallocateGraph.md): Deallocates and frees resouces associated with a graph.
+- [mvncLoadTensor](mvncLoadTensor.md): Initiates an inference by providing input to the neural network.
+- [mvncGetResult](mvncGetResult.md): Retrieves the result of an inference that was previously initiated.
+- [mvncSetGraphOption](mvncSetGraphOption.md): Sets an option for a graph.
+- [mvncGetGraphOption](mvncGetGraphOption.md): Retrieves an the current value of an option for a graph.
+- [mvncSetDeviceOption](mvncSetDeviceOption.md): Sets an option for an NCS device.
+- [mvncGetDeviceOption](mvncGetDeviceOption.md): Retrieves the current value of an option for an NCS device.
+- [mvncSetGlobalOption](mvncSetGlobalOption.md): Sets a global option for an application
+- [mvncGetGlobalOption](mvncGetGlobalOption.md): Retrieves the current value of a global option for an application.
+- [mvncCloseDevice](mvncCloseDevice.md): Closes a previously opened NCS device.
+
--- /dev/null
+# Error Codes
+
+### Error 1: "Setup Error: Caffe Import Error."
+### Error 2: "Setup Error: Using a version of Python that is unsupported."
+### Error 3: "Setup Error: Caffe Syntax Error: ..."
+### Error 4: "Toolkit Error: Stage Type Not Supported: " + str(extra)
+### Error 5: "Toolkit Error: Stage Details Not Supported: " + str(extra)
+### Error 6: "Setup Error: no Myriad Executable detected."
+### Error 7: "Toolkit Error: USB Failure. Code: ..."
+### Error 8: "Argument Error: Network description cannot be found."
+### Error 9: "Argument Error: Network weight cannot be found."
+### Error 10: "Toolkit Error: No such Mode ..."
+### Error 11: "Argument Error: Expected ID not provided."
+### Error 12: "Argument Error: Image cannot be found."
+### Error 13: "Toolkit Error: Provided OutputNode/InputNode name does not exist or does not match with one contained in caffemodel file Provided: ..."
+### Error 14: "Toolkit Error: Data Type Not Supported: " + str(extra)
+### Error 15: "Toolkit Error: Parser Not Supported: ..."
+### Error 16: "Toolkit Error: Internal Error: Input Stage is not first layer."
+### Error 17: "Toolkit Error: Internal Error: Could not build graph. Missing link: ..."
+### Error 18: "Toolkit Error: Internal Error: Invalid Conversion Optimization. From: ..."
+### Error 19: "Toolkit Error: Setup Error: Not all required arguments were passed / Erroneous arguments."
+### Error 20: "Toolkit Error: Input Layer must be in an input_shape construct."
+### Error 21: "Argument Error: Validation metric not supported "
+### Error 22: "Toolkit Error: Unable to tell what parser is required. Consider overriding with --parser argument."
+### Error 23: "Toolkit Error: Mismatch between input layer of network and given input file."
+### Error 24: "Toolkit Error: >1 image inference not supported."
+### Error 26: "Toolkit Error: No moviUsbBoot Executable detected."
+### Error 25: "Myriad Error: ..."
+### Error 27: "Setup Error: Too Many / Too Few Shave Processors Selected."
+### Error 28: "Caffe Error: MemoryError. Potential Cause: Available RAM not sufficient for Network to be loaded into Caffe"
+### Error 29: "Setup Error: Tuple Syntax Incorrect, should be in form x,y,z "
+### Error 30: "Toolkit Error: Filetype not supported as a input."
+### Error 31: "USB Error: Problem Transferring data."
+### Error 32: "Setup Error: Problem parsing configuration File."
+### Error 33: "Toolkit Error: No Temperature Read from device."
+### Error 34: "Setup Error: Values for input contain placeholder. Pass an absolute value."
+### Error 35: "Setup Error: Not enough resources on Myriad to process this network."
--- /dev/null
+# Configuring your network for NCS
+This guide will help you get all of the configuration information correct when creating your network for the Movidius Neural Compute Stick. All of these parameters are critical, if you don't get them right, your network won't give you the accuracy that was achieved by the team that trained the model. The configuration parameters include:
+* mean subtraction
+* scale
+* color channel configuration
+* class prediction
+* input image size
+
+**let's go through these one at a time**
+
+## Mean Subtraction
+mean substraction on the input data to a CNN is a common technique. The mean is calculated on the data set. For example on Imagenet the mean is calculated on a per channel basis to be:
+```
+104, 117, 123
+these numbers are in BGR orientation
+```
+### Caffe Specific Examples
+this mean calculation can be calculated with a tool that comes with caffe:
+[compute_image_mean.cpp](https://github.com/BVLC/caffe/blob/master/tools/compute_image_mean.cpp),
+and Caffe provides a script to do it as well:
+[make_imagenet_mean.sh](https://github.com/BVLC/caffe/blob/master/examples/imagenet/make_imagenet_mean.sh)
+
+---
+this will create an output file often called mean_binary.proto. You can see an example of this in the training prototxt file for AlexNet
+
+[train_val.prototxt](https://github.com/BVLC/caffe/blob/master/models/bvlc_alexnet/train_val.prototxt)
+```
+ transform_param {
+ mirror: true
+ crop_size: 227
+ mean_file: "data/ilsvrc12/imagenet_mean.binaryproto"
+ }
+```
+---
+in the GoogLeNet prototxt file they have just put the values directly:
+
+[train_val.prototxt](https://github.com/BVLC/caffe/blob/master/models/bvlc_googlenet/train_val.prototxt)
+```
+
+ transform_param {
+ mirror: true
+ crop_size: 224
+ mean_value: 104
+ mean_value: 117
+ mean_value: 123
+ }
+```
+---
+some models don't use mean subtraction, see below for LeNet as an example. There is no mean in the transform_param, but there is a scale which we'll get to later
+
+[lenet_train_test.prototxt](https://github.com/BVLC/caffe/blob/master/examples/mnist/lenet_train_test.prototxt)
+```
+ transform_param {
+ scale: 0.00390625
+ }
+ ```
+### TensorFlow specific examples
+TensorFlow documentation of mean is not as straight forward as Caffe. The TensorFlow Slim models for image classification are a great place to get high quality pre-trained models:
+
+[slim models](https://github.com/tensorflow/models/tree/master/research/slim#pre-trained-models)
+
+I could find this the following file the mean (and scale) for both Inception V3 and MobileNet V1
+
+[retrain script](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/examples/image_retraining/retrain.py#L872)
+```
+input_mean = 128
+```
+in the case of the InceptionV3 model there is not a per color channel mean, the same mean is used for all channels. This mean should apply to all of the Inception and MobileNet models, but other models there might be different.
+for example, the VGG16 model just had the weights converted from Caffe. If we look at the link to the VGG16 for Caffe page, we see the means are done like the other Caffe models:
+```
+https://gist.github.com/ksimonyan/211839e770f7b538e2d8#description
+
+the following BGR values should be subtracted: [103.939, 116.779, 123.68]
+```
+
+## Scale
+typical 8 bit per pixel per channel images will have a scale of 0-255. Many CNN networks use the native scale, but some don't. As was seen in a snippet of the Caffe prototxt file, the **transform_param** would show whether there was a scale. In the example of LeNet for Caffe, you can see it has a scale pameter of **0.00390625**
+
+[lenet_train_test.prototxt](https://github.com/BVLC/caffe/blob/master/examples/mnist/lenet_train_test.prototxt)
+```
+ transform_param {
+ scale: 0.00390625
+ }
+ ```
+ this may seem like a strange number, but it is actually just 1/256. So the input 8 bit image is being scaled down to an image from 0-1 instead of 0-255
+
+ ---
+Back to the example of TensorFlow for Inception V3. Below the **input_mean** the **input_std** is also listed. All this is is a scaling factor. You divide 255/128 and it's about 2. So in this case, the scale is two, but the mean subtraction is 128. So in the end the scale is actually -1 to 1
+
+[retrain script](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/examples/image_retraining/retrain.py#L872)
+```
+input_mean = 128
+input_std = 128
+```
+## Color Channel configuration
+different models may be trained with different color channel orientations (either RGB or BGR). Typically Caffe models seem to be trained with BGR whereas the Slim TensorFlow models (at least Inception and MobileNet) are trained in RGB.
+
+Once you figure out the color channel orientation for your model, you will need to know the way the image is loaded. For example opencv will open images in BGR but skimiage will open the image in RGB.
+```
+skimage.io.imread will open the image in RGB
+cv2.imread will open the image in BGR
+Caffe trained models will probably be BGR
+TensorFlow trained models will probably be in RGB
+```
+## Categories
+for models that are trained on the Imagenet database, some have 1000 output classes, and some have 1001 output classes. The extra output class is a background class. The list below has the list of the 1000 classes not including the background.
+[synset_words.txt](https://github.com/HoldenCaulfieldRye/caffe/blob/master/data/ilsvrc12/synset_words.txt)
+Most Caffe trained models seem to follow the 1000 class convention, and TensorFlow trained models follow the 1001 class convention. So for the TensorFlow models, an offset needs to be added. You can see this as documented in the TensorFlow github [here](https://github.com/tensorflow/models/tree/master/research/slim#the-resnet-and-vgg-models-have-1000-classes-but-the-imagenet-dataset-has-1001)
+
+# Putting it all together
+now with all of these factors, let's go through two examples
+
+## Caffe Example
+let's use the Berkeley Caffe GoogLeNet model as an example. the basic model parameters are:
+```
+Scale: 0-255 (before mean subtraction)
+Mean: based on mean_binary.proto file
+Color channel configuration: BGR
+output categories: 1000
+input size: 224x224
+labels_offset=0
+```
+code snippet:
+```
+#load the label files
+labels_offset=0 # no background class offset
+labels_file='./synset_words.txt'
+labels=numpy.loadtxt(labels_file,str,delimiter='\t')
+#Load blob
+with open('./googlenet.blob', mode='rb') as f:
+ blob = f.read()
+graph = device.AllocateGraph(blob)
+graph.SetGraphOption(mvnc.GraphOption.ITERATIONS, 1)
+iterations = graph.GetGraphOption(mvnc.GraphOption.ITERATIONS)
+img = cv2.imread('./dog.jpg') # using OpenCV for reading the image, it will be in BGR
+img=cv2.resize(img,(224,224)) # resize to 224x224
+img-=[104,117,124] # subtract mean
+#run, get the result and print results per the synset_words.txt
+graph.LoadTensor(img.astype(numpy.float16), 'user object')
+output, userobj = graph.GetResult()
+order = output.argsort()[::-1][:6]
+print('\n------- predictions --------')
+for i in range(0,5):
+ print ('prediction ' + str(i) + ' is ' + labels[order[i]-labels_offset])
+```
+
+## TensorFlow example
+let's use the TensorFlow Slim Inception V3
+```
+Scale: -1 to 1 (after mean subtraction)
+Mean: 128
+Color channel configuration: RBG
+output categories: 1001
+input size: 299x299
+labels_offset=1
+```
+code snippet:
+```
+#load the label files
+labels_offset=1 # background class offset of 1
+labels_file='./synset_words.txt'
+labels=numpy.loadtxt(labels_file,str,delimiter='\t')
+#Load blob
+with open('./inceptionv3.blob', mode='rb') as f:
+ blob = f.read()
+graph = device.AllocateGraph(blob)
+graph.SetGraphOption(mvnc.GraphOption.ITERATIONS, 1)
+iterations = graph.GetGraphOption(mvnc.GraphOption.ITERATIONS)
+#import the image and do the proper scaling
+img = cv2.imread('./dog.jpg').astype(numpy.float32) # using OpenCV for reading the image, it will be in BGR
+img=cv2.resize(img,(299,299)) # resize to 299x299
+img=cv2.cvtColor(img,cv2.COLOR_BGR2RGB) # need to convert to RBG
+img-=[128,128,128] # subtract mean
+img /=128. # scale the image
+#run, get the result and print results per the synset_words.txt
+graph.LoadTensor(img.astype(numpy.float16), 'user object')
+output, userobj = graph.GetResult()
+order = output.argsort()[::-1][:6]
+print('\n------- predictions --------')
+for i in range(0,5):
+ print ('prediction ' + str(i) + ' is ' + labels[order[i]-labels_offset])
+```
+
+feedback or comments? let me know darren.s.crews@intel.com
--- /dev/null
+# Movidius™ Neural Compute User Forum
+
+There is an active user forum in which users of the Neural Compute Stick discuss ideas and issues they have with regard to the NCS. The link to the NCS User Forum is:
+
+https://ncsforum.movidius.com
+
+The forum is a good place to go if you need help troubleshooting an issue. You may find other people that have figured out the issue or get ideas for how to fix it. The forum is also monitored by Movidius™ engineers which provide solutions as well.
+
--- /dev/null
+# Installation and Configuration
+This page provides installation and configuration information needed to use the NCS and the examples provided in this repository. To use the NCS you will need to have the Movidius™ Neural Compute SDK installed on your development computer. The SDK installation provides an option to install the examples in this repostitory. If you've already installed the SDK on your development computer you may have selected the option to also install these examples. If you have not already installed the SDK you should follow the instructions in the Example Installation with SDK section in this page, and when prompted select the option to install the examples.
+
+## Prerequisites
+To build and run the examples in this repository you will need to have the following.
+- Movidius™ Neural Compute Stick (NCS)
+- Movidius™ Neural Compute SDK
+- Development Computer with Supported OS
+ - x86-64 with Ubuntu (64 bit) 16.04 Desktop
+ - Raspberry Pi 3 with Raspian Stretch (starting with SDK 1.09.xx)
+ - See [Upgrade Raspian Jessie to Stretch](https://linuxconfig.org/how-to-upgrade-debian-8-jessie-to-debian-9-stretch)
+ - Virtual Machine per the [supported VM configuration](VirtualMachineConfig.md)
+- Internet Connection.
+- USB Camera (optional)
+
+## Connecting the NCS to a development computer
+The NCS connects to the development computer over a USB 2.0 High Speed interface. Plug the NCS directly to a USB port on your development computer or into a powered USB hub that is plugged into your development computer.
+
+
+
+## Installation SDK and examples
+To install the SDK along with the examples in this repository use the following command on your development computer. This is the typical installation. If you haven't already installed the SDK on your development computer you should use this command to install.
+```
+git clone http://github.com/Movidius/ncsdk && cd ncsdk && make install && make examples
+```
+
+## Installation of examples without SDK
+To install only the examples and not the SDK on your development computer use the following command to clone the repository and then make appropriate examples for your development computer. If you already have the SDK installed and only need the examples on your machine you should use this command to install.
+```
+git clone http://github.com/Movidius/ncsdk && cd ncsdk && make examples
+```
+
+## Building Individual Examples
+Whether installing with the SDK or without it, both methods above will install and build the examples that are appropriate for your development system including prerequisite software. Each example comes with its own Makefile that will install only that specific example and any prerequisites that it requires. To install and build any individual example run the 'make' command from within that example's base directory. For example to build the GoogLeNet examples type the following command.
+
+```
+cd examples/Caffe/GoogLeNet && make
+
+```
+
+The Makefile for each example also has a 'help' target which will display all possible targets. To see all possible targets for any example use the following command from within the examples top directory.
+```
+make help
+
+```
+
+## Uninstallation
+To uninstall the SDK type the following command.
+```
+make uninstall
+
+```
+
+
+## Installation Manifest
+For the list of files that 'make install' will modify on your system (outside of the repository) see the [installation manifest](manifest.md).
--- /dev/null
+
+# NC SDK installation manifest
+These are the files and directories installed outside of repository directory as part of the NC SDK.
+
+- /usr/local/include/mvnc.h
+- /usr/local/include/mvnc_deprecated.h
+- /usr/local/lib/libmvnc.so
+- /usr/local/lib/libmvnc.so.0
+- /usr/local/lib/mvnc/libmvnc.so
+- /usr/local/lib/mvnc/MvNCAPI.mvcmd
+- /usr/local/bin/mvNCCheck
+- /usr/local/bin/mvNCProfile
+- /usr/local/bin/mvNCCompile
+- /etc/udev/rules.d/97-usbboot.rules
+- /opt/movidius/*
--- /dev/null
+# Introduction
+
+The following explains how the Neural Compute SDK works on compiling and executing a given Caffe or TensorFlow™ Neural Network on the Neural Compute Stick.
+
+# Architecture Details
+The following diagram shows the inner workings of the Neural Compute Stick. The Neural Compute Stick primarily contains Movidius™ Myriad 2 VPU (Vision Processing Unit), and some power delivery voltage regulators. The Myriad 2 VPU includes 4 Gbit of LPDDR3 DRAM and its architecture includes specific imaging and vision accelerators and an array of 12 VLIW vector processors called SHAVE processors, used to accelerate neural networks by running parts of the neural networks in parallel for achieving the highest performance. The Neural Compute Stick is connected to an Application Processor (AP) such as a Raspberry Pi or Up Squared board using the USB interface on the Myriad 2 VPU. The USB3 interface can be used both in Super Speed (5Gbps) or High Speed (480Mbps) modes.
+
+The CPU in the Myriad 2 VPU is a SPARC microprocessor core that runs custom firmware. When the Neural Compute Stick is first plugged in there is no firmware loaded onto it. The Myriad 2 VPU boots from the internal ROM and connects to the host computer(application processor) as a USB2 device.
+
+Applications executing on the host computer (AP) communicate to the Myriad SOC using the Neural Compute API. When the API initializes and opens a device, the firmware from the Neural Compute SDK is loaded onto the Neural Compute Stick. At this time, the Neural Compute Stick resets and now shows up to the host computer as a USB2 or USB3 device depending on the host type. It is now ready to accept the neural network graph files and commands to execute inferences on the graph files.
+
+
+
+
+
+A graph file is loaded into the DRAM attached to the Myriad-2 VPU via the API. The Leon processor coordinates receiving the graph file and images for inference via the USB connection. It also parses the graph file and schedules kernels to the SHAVE neural compute accelerator engines. In addition, the Leon processor also takes care of monitoring die temperature and throttling processing on high temperature alerts. Statistics and the output of the neural network are sent back to the host computer via the USB connection and they are received by a host application via the API.
+
+In addition to the API, the SDK provides the tools mvNCCompile, mvNCCheck, and mvNCProfile that run on the host computer during application and neural network development. The checker and profiler tools run an inference on the Neural Compute Stick to validate against Caffe/TensorFlow™ and generate per layer statistics respectively.
--- /dev/null
+# Device.AllocateGraph()
+
+|Info | Value |
+|----------|---------------|
+|Package | mvnc |
+|Module | mvncapi |
+|Version | 1.0 |
+|See also|Graph<br>Graph.DeallocateGraph<br>|
+
+## Overview
+This function is used to create an instance of a Graph that represents a neural network which can be used to infer results via methods Graph.LoadTensor() and Graph.GetResult().
+
+## Syntax
+
+```python
+dev.AllocateGraph(graphPath)
+```
+
+## Parameters
+
+|Parameter | Description |
+|---------------|---------------|
+|graphPath | A string that is the path to the graph file. The graph file must have been created with the NC SDK graph compiler.|
+
+## Return
+Returns an instance of a Graph object that is ready to use.
+
+## Known Issues
+After the Graph that is created is no longer needed Graph.DeallocateGraph() must be called to free the graph resources.
+
+## Example
+```python
+
+from mvnc import mvncapi as ncs
+# Enumerate Devices
+device_List = ncs.Enumerate()
+
+# Initialize and open the first device
+device = ncs.Device(device_List[0])
+device.OpenDevice()
+
+# Allocate a graph on the device by specifying the path to a graph file
+graph = device.AllocateGraph("../networks/myNetwork/graph")
+
+# Use graph here
+
+# Deallocate the graph to free resources
+graph.DeallocateGraph()
+
+#close device
+```
--- /dev/null
+# Device.CloseDevice()
+
+|Info | Value |
+|----------|---------------|
+|Package | mvnc |
+|Module | mvncapi |
+|Version | 1.0 |
+|See also|Device.\_\_init\_\_()<br>Device.OpenDevice()|
+
+## Overview
+This function is used to cease communication and reset the device.
+
+## Syntax
+```python
+dev.CloseDevice()
+```
+
+## Parameters
+None.
+
+## Return
+None.
+
+## Known Issues
+
+## Example
+```Python
+# Open and close the first NCS device
+import mvnc.mvncapi as ncs
+
+deviceNames = ncs.EnumerateDevices()
+firstDevice = ncs.Device(deviceNames[0])
+
+# Open the device
+firstDevice.OpenDevice()
+
+# Use device here
+
+# Close the device
+firstDevice.CloseDevice()
+```
--- /dev/null
+# Device.GetDeviceOption()
+
+|Info | Value |
+|----------|---------------|
+|Package | mvnc |
+|Module | mvncapi |
+|Version | 1.0 |
+|See also |Device.SetDeviceOption<br>DeviceOption|
+
+## Overview
+This function is used to get an option for the device. The options can be found in the DeviceOption enumeration table.
+
+## Syntax
+```python
+GetDeviceOption(option)
+```
+
+## Parameters
+
+|Parameter | Description |
+|---------------|---------------|
+| option | Member of the DeviceOption enumeration that specifies which option to get.|
+
+## Return
+The value for the specified device option. The type of the returned value depends on the option specified. See the DeviceOption enumeration for the type that will be returned for each option.
+
+## Known Issues
+
+## Example
+```python
+import mvnc.mvncapi as ncs
+
+# Initialize and open device
+
+# Get the device option THERMAL_THROTTLING_LEVEL
+optionValue = device.GetDeviceOption(mvnc.DeviceOption.THERMAL_THROTTLING_LEVEL)
+
+```
--- /dev/null
+# Device.OpenDevice()
+
+|Info | Value |
+|----------|---------------|
+|Package | mvnc |
+|Module | mvncapi |
+|Version | 1.0 |
+|See also|Device.\_\_init\_\_()<br>Device.CloseDevice()|
+
+## Overview
+This function is used to initialize the device.
+
+## Syntax
+```python
+OpenDevice()
+```
+
+## Parameters
+None.
+
+## Return
+None.
+
+## Known Issues
+
+## Example
+```python
+#############################################
+# Open first device only
+import mvnc.mvncapi as ncs
+deviceNames = ncs.EnumerateDevices()
+firstDevice = ncs.Device(deviceNames[0])
+firstDevice.OpenDevice()
+# use device
+firstDevice.CloseDevice()
+#############################################
+
+
+#############################################
+# Open all devices
+import mvnc.mvncapi as ncs
+devices = ncs.EnumerateDevices()
+devlist = list()
+for devnum in range(len(devices)):
+ devlist.append(ncs.Device(devices[devnum]))
+ devlist[devnum].OpenDevice()
+#
+# Use devices in devlist
+#
+for dev in devList:
+ dev.CloseDevice()
+#############################################
+
+```
--- /dev/null
+# Device.SetDeviceOption()
+
+|Info | Value |
+|----------|---------------|
+|Package | mvnc |
+|Module | mvncapi |
+|Version | 1.0 |
+|See also|Device.GetDeviceOption()<br>DeviceOption|
+
+## Overview
+This function is used to set an option for the device. The options can be found in the DeviceOption enumeration class.
+
+## Syntax
+```python
+SetDeviceOption(option, value)
+```
+
+## Parameters
+
+|Parameter | Description |
+|-----------|---------------|
+|option | Member of the DeviceOption enumeration class that specifies which device option to set |
+|value | The new value to which the device option will be set. The type to pass for this parameter depends on which option is being set. See the DeviceOption enumeration class for the types that correspond to each option.|
+
+
+## Return
+None.
+
+## Known Issues
+
+## Example
+```python
+
+
+```
--- /dev/null
+# Device.\_\_init\_\_()
+
+|Info | Value |
+|----------|---------------|
+|Package | mvnc |
+|Module | mvncapi |
+|Version | 1.0 |
+|See also||
+
+## Overview
+This function is used to initialize a device object.
+
+## Syntax
+
+```python
+mvnc.Device("device name here")
+```
+
+## Parameters
+
+|Parameter | Description |
+|---------------|---------------|
+|deviceName | The name of the device to initialize. This must come from calling mvncapi module function EnumerateDevices()|
+
+## Return
+None.
+
+## Known Issues
+
+## Example
+```python
+import mvnc.mvncapi as ncs
+
+# Enumerate devices
+deviceNames = ncs.EnumerateDevices()
+
+# create and init a Device instance.
+ncsDevice = ncs.Device(deviceNames[0])
+
+# Open device, use device, close device
+
+```
--- /dev/null
+# Device class
+
+The Device class represents the NCS device. Typically one instance of this class is created for each physical NCS device that is plugged into the system so multiple instances may exist if you have multiple devices attached ot your system.
+
+# Usage
+To use the Device class you must create and initialize it by name. The valid names to use can be determined by calling the mvncapi module function EnumerateDevices(). Once you have successfully created an instance of this class the typical usage is to call OpenDevice(), AllocateGraph(), use the graph, CloseDevice().
+
+# Device methods
+## [\_\_init\_\_](Device.__init__.md)
+## [OpenDevice](Device.OpenDevice.md)
+## [CloseDevice](Device.CloseDevice.md)
+## [SetDeviceOption](Device.SetDeviceOption.md)
+## [GetDeviceOption](Device.GetDeviceOption.md)
+## [AllocateGraph](Device.AllocateGraph.md)
--- /dev/null
+# DeviceOption enumeration class
+
+|Info | Value |
+|----------|---------------|
+|Package | mvnc |
+|Module | mvncapi |
+|Version | 1.0 |
+|See also |Device.SetDeviceOption()<br>Device.GetDeviceOption()|
+
+
+
+## Overview
+The DeviceOption class is an enumeration class that defines the options that are passed to and received from the SetDeviceOption and the GetDeviceOption functions.
+
+
+Enumerator Values|Description
+------------ | -------------
+THERMAL_THROTTLING_LEVEL |Returns 1 if lower guard temperature threshold of chip sensor is reached. This indicates short throttling time is in action between inferences to protect the device. Returns 2 if upper guard temperature of chip sensor is reached. This indicates long throttling time is in action between inferences to protect the device.
--- /dev/null
+# EnumerateDevices()
+
+|Info | Value |
+|----------|---------------|
+|Package | mvnc |
+|Module | mvncapi |
+|Version | 1.0 |
+|See also | Device |
+
+## Overview
+This function is used to get a list of the names of the devices present in the system. Each of the returned names can be used to create an instance of the Device class.
+
+## Syntax
+
+```python
+deviceNames = EnumerateDevices()
+```
+
+## Parameters
+None.
+
+## Return
+An array of device names each of which can be used to create a new instance of the Device class.
+
+## Known Issues
+
+## Example
+```Python
+import mvnc.mvncapi as ncs
+deviceNames = ncs.EnumerateDevices()
+if len(deviceNames) == 0:
+ print("Error - No devices detected.")
+ quit()
+
+# open first NCS device found
+device = ncs.Device(deviceNames[0])
+
+# Allocate graph / otherwise use device as needed
+
+# Deallocate graph if allocated
+
+device.CloseDevice()
+```
--- /dev/null
+# GetGlobalOption
+
+|Info | Value |
+|----------|---------------|
+|Package | mvnc |
+|Module | mvncapi |
+|Version | 1.0 |
+|See also | GlobalOption<br>SetGlobalOption()|
+
+## Overview
+This function is used to get a global option. The available options can be found in the GlobalOption enumeration section.
+
+## Syntax
+
+```python
+value = GetGlobalOption(option)
+```
+
+## Parameters
+
+|Parameter | Description |
+|---------------|---------------|
+|option |Member of the GlobalOption enumeration that specifies which option to get.|
+
+## Return
+The value for the specified option. The type of the returned value depends on the option specified. See the GlobalOption enumeration for the type that will be returned for each option.
+
+## Known Issues
+
+## Example
+```Python
+from mvnc import mvncapi as mvnc
+
+# Get current logging level Global Option
+logLevel = mvnc.GetGlobalOption(mvnc.GlobalOption.LOGLEVEL)
+
+print("The current global logging level is: ", logLevel)
+```
--- /dev/null
+# GlobalOption enumeration class
+
+|Info | Value |
+|----------|---------------|
+|Package | mvnc |
+|Module | mvncapi |
+|Version | 1.0 |
+|See also | SetGlobalOption()<br>GetGlobalOption()|
+
+
+
+## Overview
+The GlobalOption class is an enumeration class that defines the options that are passed to and received from the SetGlobalOption and the GetGlobalOption functions.
+
+
+Enumerator Values|Description
+------------ | -------------
+LOGLEVEL |0=Nothing is printed. <br>1=Print errors only. <br>2=Verbose.
+
+
--- /dev/null
+# Graph.DeallocateGraph()
+
+|Info | Value |
+|----------|---------------|
+|Package | mvnc |
+|Module | mvncapi |
+|Version | 1.0 |
+|See also|Device.AllocateGraph()|
+
+## Overview
+This function is used to deallocate a graph that was allocated for an NCS device with the Device.AllocateGraph() method. This should be called for every graph that is created to free resources associated with the graph.
+
+## Syntax
+
+```python
+graph.DeallocateGraph()
+```
+## Parameters
+None.
+
+## Return
+None.
+
+## Known Issues
+The Graph class can only be created via the Device.AllocateGraph() function.
+
+## Example
+```python
+
+import mvnc.mvncapi as ncs
+deviceNames = ncs.EnumerateDevices()
+if len(deviceNames) == 0:
+ print("Error - No devices detected.")
+ quit()
+
+# open first NCS device found
+device = mvnc.Device(devices[0])
+
+# Allocate the graph
+device.AllocateGraph("my_graph")
+
+# Use device here
+
+graph.DeallocateGraph()
+
+device.CloseDevice()
+```
--- /dev/null
+# Graph.GetGraphOption()
+
+|Info | Value |
+|----------|---------------|
+|Package | mvnc |
+|Module | mvncapi |
+|Version | 1.0 |
+|See also|Graph.SetGraphOption()<br>GraphOption|
+
+## Overview
+This function is used to get a graph option. The available options can be found in GraphOption enumeration class.
+
+## Syntax
+```python
+value = graph.GetGraphOption(option)
+```
+
+## Parameters
+
+|Parameter | Description |
+|---------------|---------------|
+|option | A value from the GraphOption enumeration to specify which option's value should be returned. |
+
+## Return
+The value for the specified GraphOption is returned. The type of the returned value depends on the option specified. See the GraphOption class for the value types for each option.
+
+## Known Issues
+
+## Example
+```python
+import mvnc.mvncapi as ncs
+deviceNames = ncs.EnumerateDevices()
+ncsDevice = ncs.Device(deviceNames[0])
+ncsDevice.OpenDevice()
+graph = ncsDevice.AllocateGraph("../networks/myNetwork/graph")
+
+# Get the graph option
+optionValue = graph.GetGraphOption(mvnc.GraphOption.DONTBLOCK)
+
+# Use device here
+
+graph.DeallocateGraph()
+ncsDevice.CloseDevice()
+
+
+
+```
--- /dev/null
+# Graph.GetResult()
+
+|Info | Value |
+|----------|---------------|
+|Package | mvnc |
+|Module | mvncapi |
+|Version | 1.0 |
+|See also |Graph.LoadTensor()|
+
+## Overview
+This function retrieves the result of an inference that was initiated via Graph.LoadTensor() on the specified graph.
+
+## Syntax
+```python
+inferenceResult, userObj = graph.GetResult()
+```
+
+## Parameters
+None.
+
+## Return
+The inference result and user object that was specified in the Graph.LoadTensor() call that initiated this inference.
+
+## Known Issues
+
+## Example
+```python
+
+# Open NCS device and allocate a graph
+
+graph.LoadTensor(img.astype(numpy.float16), "user object for this inference")
+inferenceResult, userObj = graph.GetResult()
+
+# Deallocate graph
+# Close device
+
+```
--- /dev/null
+# Graph.LoadTensor()
+
+|Info | Value |
+|----------|---------------|
+|Package | mvnc |
+|Module | mvncapi |
+|Version | 1.0 |
+|See also |Graph.GetResult()|
+
+## Overview
+This function initiates an inference on the specified graph via the associated NCS device. After calling this function use the Graph.GetResult() function to retrieve the inference result.
+
+## Syntax
+
+```python
+graph.LoadTensor(inputTensor, userObject)
+```
+## Parameters
+
+|Parameter | Description |
+|---------------|---------------|
+|inputTensor | Input data on which an inderence will be run. The data must be passed in a NumPy ndarray of half precision floats (float 16) | |
+|userObject | A user-defined parameter that is returned by the GetResult function along with the inference result for this tensor.|
+
+## Return
+Returns True if the function works, False if not. When the graph is in non blocking mode (GraphOption.DONTBLOCK) this function will return False if the device is busy.
+
+## Known Issues
+
+## Example
+```python
+
+# Enumerate Device
+# Open Device, # Allocate Graph, # Set Graph Option
+# Read an image, resize the image and adjust for mean if necessary to match the network expected size
+
+if (graph.LoadTensor(img.astype(numpy.float16), 'user object')):
+ print("LoadTensor success")
+ output, userobj = graph.GetResult()
+
+# Deallocate the graph and Close the device
+```
--- /dev/null
+# Graph.SetGraphOption()
+
+|Info | Value |
+|----------|---------------|
+|Package | mvnc |
+|Module | mvncapi |
+|Version | 1.0 |
+|See also|Graph.GetGraphOption()<br>GraphOption|
+
+## Overview
+This function is used to set a graph option. The available options can be found in the GraphOption enumeration class.
+
+## Syntax
+```python
+graph.SetGraphOption(option, value)
+```
+
+## Parameters
+
+Parameter | Description
+---------------|---------------
+option | Member of the GraphOption enumeration specifying which option's value will be set.
+value | The new value to which the specified graph option will be set. See the GraphOption enumeration class for the type of value for each option.
+
+
+## Return
+None.
+
+## Known Issues
+
+## Example
+```python
+import mvnc.mvncapi as ncs
+deviceNames = ncs.EnumerateDevices()
+ncsDevice = ncs.Device(deviceNames[0])
+ncsDevice.OpenDevice()
+graph = ncsDevice.AllocateGraph("../networks/myNetwork/graph")
+
+graph.SetGraphOption(mvnc.GraphOption.DONTBLOCK, 1)
+
+# Use device here
+
+graph.DeallocateGraph()
+ncsDevice.CloseDevice()
+```
--- /dev/null
+# Graph class
+
+The Graph class is a container for a neural network graph file that is associated with a particular NCS device.
+
+# Usage
+To use the Graph class you must create a graph handle by calling AllocateGraph() from the Device class. The location of the graph file will be passed to AllocateGraph() and it will return an instance of the Graph class. Once you have successfully created an instance of this class, the typical usage is to optionally get/set graph options, then call LoadTensor() and GetResult(), to perform inferencing with the graph that was allocated, and finally call DeallocateGraph() when the neural network is no longer needed.
+
+# Graph methods
+## [DeallocateGraph](Graph.DeallocateGraph.md)
+## [SetGraphOption](Graph.SetGraphOption.md)
+## [GetGraphOption](Graph.GetGraphOption.md)
+## [LoadTensor](Graph.LoadTensor.md)
+## [GetResult](Graph.GetResult.md)
--- /dev/null
+# GraphOption enumeration class
+
+|Info | Value |
+|----------|---------------|
+|Package | mvnc |
+|Module | mvncapi |
+|Version | 1.0 |
+|See also | Graph.SetGraphOption()<br>Graph.GetGraphOption()|
+
+
+
+## Overview
+The GraphOption class is an enumeration class that defines the options that are passed to and received from the SetDeviceOption and the GetDeviceOption functions.
+
+
+enum| option type | possible values|get/set|Description
+--- | ----------- | -------------- |-------|-----------
+DONTBLOCK |integer |0 or 1|get/set|0: LoadTensor and GetResult Block<br>1: LoadTensor returns BUSY instead of blocking. GetResult will return NODATA instead of blocking.
+TIMETAKEN |string |any|get|Return a NumPy float array [numpy.array()] of inference times per layer in float data type.
+DEBUGINFO | string |any|get|Return a string with the error text as returned by the device.
--- /dev/null
+# SetGlobalOption
+
+|Info | Value |
+|----------|---------------|
+|Package | mvnc |
+|Module | mvncapi |
+|Version | 1.0 |
+|See also | GlobalOption<br>GetGlobalOption()|
+
+## Overview
+This function is used to set a global option. The available Global options and possible values can be found in the documentation for the GlobalOption enumeration.
+
+## Syntax
+
+```python
+SetGlobalOption(option, value)
+```
+
+## Paramerers
+
+|Parameter | Description |
+|---------------|---------------|
+|option|Member of the GlobalOptions enumeration which specifies which option to set.|
+|value |The new value to which the option will be set. See the GlobalOption enumeration class for the type of value for each option.|
+
+## Known Issues
+
+## Example
+```Python
+import mvnc.mvncapi as ncs
+
+# set the global logging level to verbose
+ncs.SetGlobalOption(ncs.GlobalOption.LOGLEVEL, 2)
+```
--- /dev/null
+# Status enumeration class
+
+|Info | Value |
+|----------|---------------|
+|Package | mvnc |
+|Module | mvncapi |
+|Version | 1.0 |
+|See also | |
+
+
+## Overview
+The Status class is an enumeration class that defines status codes returned from calls to the API functions. If the underlying API returns a non-zero status, an exception is usually raised with the corresponding status. The possible status codes are shown below.
+
+
+Enumeration Values|Description
+------------ | -------------
+MVNC_OK |The function call worked as expected.
+MVNC_BUSY |The device is busy, retry later.
+MVNC_ERROR |An unexpected error was encountered during the function call.
+MVNC_OUT_OF_MEMORY |The host is out of memory.
+MVNC_DEVICE_NOT_FOUND |There is no device at the given index or name.
+MVNC_INVALID_PARAMETERS |At least one of the given parameters is invalid in the context of the function call.
+MVNC_TIMEOUT |Timeout in the communication with the device
+MVNC_MVCMD_NOT_FOUND |The file named MvNCAPI.mvcmd is installed in the mvnc directory. This message means that the file has been moved or installer failed.
+MVNC_NO_DATA |No data to return.
+MVNC_GONE |The graph or device has been closed during the operation.
+MVNC_UNSUPPORTED_GRAPH_FILE |The graph file is corrupt or may have been created with an incompatible prior version of the NCS toolkit. Try to recompile the graph file with the version of the Toolkit that corresponds to the API version.
+MVNC_MYRIAD_ERROR |An error has been reported by the Movidius™ VPU. Use MVNC_DEBUGINFO.
+
--- /dev/null
+# Movidius™ Neural Compute SDK Python API
+
+The SDK comes with a Python3 Language API that enables applications that utilize hardware accelerated Deep Neural Networks via the Movidius™ Neural Compute Stick (NCS.) The API is provided as a single python script (mvncapi.py) which is placed on the development computer when the SDK is installed. Details of the Python API are provided below and within the documents linked from here.
+
+## Class Diagram
+The Python class diagram follows. In this diagram it is shown that the entire python api is in the mvnc Python package. Within the mvnc package there is one Python module which is mvncapi. The mvncapi module is defined by the contents of the mvncapi.py file.
+
+
+
+In the sections that follow, the details of the elements within the mvncapi module are provided. This includes module level enumerations, functions that are global to the module, and the classes defined in the module.
+
+## Enumerations
+### [GlobalOption](GlobalOption.md)
+### [DeviceOption](DeviceOption.md)
+### [GraphOption](GraphOption.md)
+### [Status](Status.md)
+
+## Global Functions
+### [EnumerateDevices](EnumerateDevices.md)
+### [SetGlobalOption](SetGlobalOption.md)
+### [GetGlobalOption](GetGlobalOption.md)
+
+
+## Classes
+### [Device](Device.md)
+### [Graph](Graph.md)
+
+
--- /dev/null
+============================================================
+# Movidius Neural Compute SDK Release Notes
+# V1.09.00 2017-10-09
+============================================================
+
+## SDK Notes:
+SDK has been refactored and contains many new features and structural changes. It is recommended you read the documentation to familiarize with the new features and contents. A partial list of new features:
+
+ 1. New, unified, faster installer and uninstaller.
+ 2. Now supports complete SDK installation on Raspberry Pi
+ 3. System installation of tools and API libraries.
+ 4. API support for Python 2.7.
+ 5. Source code included for API, for porting to other architectures or Linux distributions
+ 6. Tools support for Pi.
+ 7. Tensorflow R1.3 support for tools (only on Ubuntu 16.04 LTS currently).
+ 8. More network support, see documentation for details!
+ 9. Support for SDK on Ubuntu 16.04 LTS as guest OS, and Win10, OSX, and Ubuntu 16.04 as host OS. See docs/VirtualMachineConfig.
+
+## API Notes:
+ 1. API supported on both python 2.7 and python 3.5.
+ 2. Some APIs deprecated, will emit the "deprecated" warning if used. Users expected to move to using new APIs for these functions.
+
+## Network Notes:
+Support for the following networks has been tested.
+
+### Caffe
+ 1. GoogleNet V1
+ 2. SqueezeNet V1.1
+ 3. LeNet
+ 4. CaffeNet
+ 5. VGG (Sousmith VGG_A)
+ 6. Alexnet
+ 7. TinyYolo v1
+
+
+### Tensorflow r1.3
+ 1. inception-v1
+ 2. inception-v3
+ 3. inception-v4
+ 4. Inception ResNet v1 (FaceNet)
+ 5. Inception ResNet v2
+ 6. Mobilenet_V1_1.0_224
+
+## Bug Fixes:
+ 1. USB protocol bug fixes, for expanded compatibility with hubs and hosts. Fix for devices with maxpacket of 64.
+ 2. When a graph execution fails, the result for a previous execution is erroneously returned.
+
+## Errata:
+ 1. Python 2.7 is fully supported for making your own applications, but only the helloworld_py example runs as-is in both python 2.7 and 3.5 due to dependencies on modules
+ 2. SDK tools for tensorflow on Rasbpian Stretch are not supported for this release, due to lack of an integrated tensorflow installer for Rasbpian in the SDK. TF examples are provided with pre-compiled graph files to allow them to run on Pi, however the compile, profile, and check functions will not be available on Pi, and 'make examples' will generate failures for the tensorflow examples on Rasbpian.
+ 3. mobilenet is not optimized. Support for this network should be considered a preview.
+ 4. Restriction on fully connected layers -- input dimension must be a multiple of 16.
+ 5. If working behind proxy, proper proxy settings must be applied for the installer to succeed.
+ 6. Although improved, the Installer is known to take a long time on Raspberry Pi. Date/time must be correct for SDK installation to succeed on Raspbian.
+ 7. Default system virtual memory swap file size is too small to compile AlexNet on Raspberry Pi.
+ 8. Raspbian users will need to upgrade to Raspbian Stretch for this release.
--- /dev/null
+# mvNCCheck
+
+Type|Function
+------------ | -------------
+Library|Command Line Tools
+Output|
+Revision|1.08
+See also| [mvNCCompile](compile.md), [mvNCProfile](profile.md), [TensorFlow™ Info](../TensorFlowUsage.md)
+
+## Overview
+This commandline tool compiles the provided network, runs the network on the connected Neural Compute Stick (NCS) and creates a text/HTML profiling output. The profiling data contains layer by layer stats about the performance of the input network. This is very helpful in determining how much time is spent on each layer and is helpful in determining changes to the network to improve the total inference time for a network on the Neural Compute Stick.
+
+The weights file is not required when profiling a network to determine bottlenecks.
+
+## Syntax
+
+### Caffe
+```bash
+mvNCCheck network.prototxt [-w weights_file] [-s Max Number of Shaves] [-in Input Node Name] [-on Output Node Name] [-is Input-Width Input-Height] [-o Output Graph Filename] [-i image filename] [-id Top-1 Validation ID] [-S scale factor] [-M Mean Subtraction Number or npy filename] [-cs Color Sequence]
+```
+
+### TensorFlow™
+```bash
+mvNCCheck network.meta [-s Max Number of Shaves] [-in Input Node Name] [-on Output Node Name] [-is Input-Width Input-Height] [-o Output Graph Filename] [-i image filename] [-id Top-1 Validation ID] [-S scale factor] [-M Mean Subtraction Number or npy filename] [-cs Color Sequence]
+```
+
+Argument|Description
+------------ | -------------
+network.prototxt(caffe)<br>network.meta(TensorFlow™)|Name of the network file.
+[-w weights_file]|Weights filename from training (Only applies to Caffe, not to be used with TensorFlow™.) If omitted zero weights will be used.
+[-s Max # of Shaves]|Default: 1<br><br>Selects the maximum number of SHAVEs (1,2,4,8 or 12.) to use for network layers.<br><br>Note: The NCS runtime code may use less than the MAX SHAVE value for some layers where measurements have typically shown no inference performance degradation (and consequently a power benefit) of using fewer SHAVEs.
+[-in Input Node Name]|By default the network is processed from the input tensor. This option allows a user to select an alternative start point in the network.<br><br>This enables partial network processing. When used together with the -on option a user can isolate one or more layers in a network for analysis.
+[-on Output Node Name]|By default the network is processed through to the output tensor. This option allows a user to select an alternative end point in the network.<br><br>This enables partial network processing. When used together with the -in option a user can isolate one or more layers in a network for analysis.
+[-is Input-Width Input-Height]|Input size is typically described as a part of the network. For networks that do not have dimension constraints on the input tensor, this option can be used to set the desired input dimensions.<br><br>Only two dimensions are defined because the batch size is always 1 and the number of color planes is assumed to be 3.
+[-o Output Graph Filename]|Default: "graph"<br><br>Output graph container filename. If not provided, “graph” will be used.
+[-i image filename]|Image to use as input to validation run.<br>If not set, a randomly generated image will be used
+[-id Top-1 Validation ID]|Expected id for Top-1 validation
+[-S scale factor]|Scale each value of the input by this amount.<br>E.g. if the network expects input values in the range 0-255, put 255 here (1 is default, as the range 0-1 is the default).
+[-M Mean Subtraction Number or npy filename]|Subtract this from the input (applied after scale). E.g. If the network expects a mean file to be subtracted from the input image, put it here.
+[-cs Color Sequence]|Color Sequence of Input channels<br> 2,1,0: BGR (Default) <br> 0,1,2 : RGB
+
+## Known Issues
+
+## Example
+### Caffe
+```bash
+
+mvNCCheck deploy.prototxt -w bvlc_googlenet.caffemodel -s 12 -in input -on prob -is 224 224 -o GoogLeNet.graph -cs 2,1,0
+
+```
+### TensorFlow™
+```bash
+mvNCCheck inception_v1.meta -s 12 -in=input -on=InceptionV1/Logits/Predictions/Reshape_1 -is 224 224 -o InceptionV1.graph -cs 0,1,2
+
+```
--- /dev/null
+# mvNCCompile
+
+Type|Function
+------------ | -------------
+Library|Command Line Tools
+Output| Compiled 'graph' file
+Revision|1.08
+See also| [mvNCProfile](profile.md), [mvNCCheck](check.md), [TensorFlow™ Info](../TensorFlow.md)
+
+## Overview
+This commandline tool compiles and converts the network file and weights file described in Caffe or TensorFlow™ into Movidius™ internal Graphfile format. The graph file is loaded into the Neural Compute Stick during runtime using the NCSDK API. The graph file then can be executed by sending an image to the NCS for inferencing.
+
+## Syntax
+
+### Caffe
+```bash
+mvNCCompile network.prototxt [-w weights_file] [-s Max Number of Shaves] [-in Input Node Name] [-on Output Node Name] [-is Input-Width Input-Height] [-o Output Graph Filename]
+```
+### TensorFlow™
+```bash
+mvNCCompile network.meta [-s Max Number of Shaves] [-in Input Node Name] [-on Output Node Name] [-is Input-Width Input-Height] [-o Output Graph Filename]
+```
+
+Argument|Description
+------------ | -------------
+network.prototxt(Caffe)<br>network.meta(TensorFlow™)|Name of the network file.
+[-w weights_file]|Weights filename from training (only applies to Caffe, not to be used with TensorFlow™.) If omitted zero weights will be used.
+[-s Max # of Shaves]|Default: 1<br><br>Selects the maximum number of SHAVEs (1, 2, 4, 8 or 12.) to use for network layers.<br><br>Note: The NCS runtime code may use less than the MAX SHAVE value for some layers where measurements have typically shown no inference performance degradation (and consequently a power benefit) of using fewer SHAVEs.
+[-in Input Node Name]|By default the network is processed from the input tensor. This option allows a user to select an alternative start point in the network.<br><br>This enables partial network processing. When used together with the -on option a user can isolate one or more layers in a network for analysis.
+[-on Output Node Name]|By default the network is processed through to the output tensor. This option allows a user to select an alternative end point in the network.<br><br>This enables partial network processing. When used together with the -in option a user can isolate one or more layers in a network for analysis. Note: beware that the parser stops at the first instance of the output node name (e.g., a Relu following a Conv will not be processed if it shares the same name).
+[-is Input-Width Input-Height]|Input size is typically described as a part of the network. For networks that do not have dimension constraints on the input tensor, this option can be used to set the desired input dimensions.<br><br>Only two dimensions are defined because the batch size is always 1 and the number of color planes is assumed to be 3.
+[-o Output Graph Filename]|Default: "graph"<br><br>Output graph container filename. If not provided, “graph” will be used.
+
+## Known Issues
+
+## Example
+### Caffe
+```bash
+
+mvNCCompile deploy.prototxt -w bvlc_googlenet.caffemodel -s 12 -in input -on prob -is 224 224 -o GoogLeNet.graph
+
+```
+### TensorFlow™
+```bash
+
+mvNCCompile inception-v1.meta -s 12 -in=input -on=InceptionV1/Logits/Predictions/Reshape_1 -is 224 224 -o InceptionV1.graph
+
+```
+
--- /dev/null
+# mvNCProfile
+
+Type|Function
+------------ | -------------
+Library|Command Line Tools
+Output|Profile Data in HTML/graphic/Text Format
+Revision|1.08
+See also| [mvNCCompile](compile.md), [mvNCCheck](check.md), [TensorFlow™ info](../TensorFlow.md)
+
+## Overview
+This commandline tool compiles the provided network, runs the network on the connected Neural Compute Stick (NCS) and creates a text/HTML profiling output. The profiling data contains layer by layer stats about the performance of the input network. This is very helpful in determining how much time is spent on each layer and is helpful in determining changes to the network to improve the total inference time for a network on the Neural Compute Stick.
+
+The weights file is not required when profiling a network to determine bottlenecks.
+
+## Syntax
+
+### Caffe
+```bash
+mvNCProfile network.prototxt [-w weights_file] [-s Max Number of Shaves] [-in Input Node Name] [-on Output Node Name] [-is Input-Width Input-Height] [-o Output Graph Filename]
+```
+### TensorFlow™
+```bash
+mvNCProfile network.meta [-s Max Number of Shaves] [-in Input Node Name] [-on Output Node Name] [-is Input-Width Input-Height] [-o Output Graph Filename]
+```
+
+Argument|Description
+------------ | -------------
+network.prototxt(caffe)<br>network.meta(TensorFlow™)|Name of the network file.
+[-w weights_file]|Weights filename from training (Only applies to Caffe, not to be used with TensorFlow™.) If omitted zero weights will be used.
+[-s Max # of Shaves]|Default: 1<br><br>Selects the maximum number of SHAVEs (1,2,4,8 or 12.) to use for network layers.<br><br>Note: The NCS runtime code may use less than the MAX SHAVE value for some layers where measurements have typically shown no inference performance degradation (and consequently a power benefit) of using fewer SHAVEs.
+[-in Input Node Name]|By default the network is processed from the input tensor. This option allows a user to select an alternative start point in the network.<br><br>This enables partial network processing. When used together with the -on option a user can isolate one or more layers in a network for analysis.
+[-on Output Node Name]|By default the network is processed through to the output tensor. This option allows a user to select an alternative end point in the network.<br><br>This enables partial network processing. When used together with the -in option a user can isolate one or more layers in a network for analysis.
+[-is Input-Width Input-Height]|Input size is typically described as a part of the network. For networks that do not have dimension constraints on the input tensor, this option can be used to set the desired input dimensions.<br><br>Only two dimensions are defined because the batch size is always 1 and the number of color planes is assumed to be 3.
+[-o Output Graph Filename]|Default: "graph"<br><br>Output graph container filename. If not provided, “graph” will be used.
+
+
+## Known Issues
+- The -o option is yet to be implemented.
+- the -w option can be omitted if the network's .caffemodel file has the same base filename as the .prototxt file it will be used.
+
+## Example
+### Caffe
+```bash
+
+mvNCProfile deploy.prototxt -w bvlc_googlenet.caffemodel -s 12 -in input -on prob -is 224 224 -o GoogLeNet.graph
+
+```
+### TensorFlow™
+```bash
+
+mvNCProfile inception_v1.meta -s 12 -in=input -on=InceptionV1/Logits/Predictions/Reshape_1 -is 224 224 -o InceptionV1.graph
+
+```
+
+# Profile Output for GoogLeNet
+
+## Text Format
+The console output from mvNCProfile is shown below.
+```
+Detailed Per Layer Profile
+Layer Name MFLOPs Bandwidth MB/s time(ms)
+========================================================================================
+0 conv1/7x7_s2 236.028 2505.00 5.63
+1 pool1/3x3_s2 1.806 1441.66 1.06
+2 pool1/norm1 0.000 712.67 0.54
+3 conv2/3x3_reduce 25.690 404.11 0.97
+4 conv2/3x3 693.633 316.67 11.55
+5 conv2/norm2 0.000 797.05 1.44
+6 pool2/3x3_s2 1.355 1495.52 0.77
+7 inception_3a/1x1 19.268 462.47 0.67
+8 inception_3a/3x3_reduce 28.901 399.64 0.81
+9 inception_3a/3x3 173.408 333.13 4.52
+10 inception_3a/5x5_reduce 4.817 793.78 0.37
+11 inception_3a/5x5 20.070 849.91 0.73
+12 inception_3a/pool 1.355 686.68 0.42
+13 inception_3a/pool_proj 9.634 558.60 0.54
+14 inception_3b/1x1 51.380 470.46 0.95
+15 inception_3b/3x3_reduce 51.380 472.93 0.94
+16 inception_3b/3x3 346.817 268.78 7.99
+17 inception_3b/5x5_reduce 12.845 1098.70 0.36
+18 inception_3b/5x5 120.422 580.92 2.32
+19 inception_3b/pool 1.806 695.31 0.55
+20 inception_3b/pool_proj 25.690 683.06 0.61
+21 pool3/3x3_s2 0.847 1305.34 0.55
+22 inception_4a/1x1 36.127 374.89 0.95
+23 inception_4a/3x3_reduce 18.063 574.14 0.47
+24 inception_4a/3x3 70.447 320.50 2.09
+25 inception_4a/5x5_reduce 3.011 1034.04 0.19
+26 inception_4a/5x5 7.526 616.84 0.31
+27 inception_4a/pool 0.847 630.87 0.28
+28 inception_4a/pool_proj 12.042 661.36 0.36
+29 inception_4b/1x1 32.113 294.21 1.18
+30 inception_4b/3x3_reduce 22.479 377.09 0.80
+31 inception_4b/3x3 88.510 313.94 2.58
+32 inception_4b/5x5_reduce 4.817 838.52 0.26
+33 inception_4b/5x5 15.053 384.82 0.78
+34 inception_4b/pool 0.903 612.12 0.31
+35 inception_4b/pool_proj 12.845 552.44 0.46
+36 inception_4c/1x1 25.690 486.52 0.65
+37 inception_4c/3x3_reduce 25.690 488.53 0.65
+38 inception_4c/3x3 115.606 308.59 3.23
+39 inception_4c/5x5_reduce 4.817 835.81 0.26
+40 inception_4c/5x5 15.053 387.14 0.78
+41 inception_4c/pool 0.903 614.42 0.31
+42 inception_4c/pool_proj 12.845 550.52 0.46
+43 inception_4d/1x1 22.479 393.44 0.77
+44 inception_4d/3x3_reduce 28.901 388.96 0.85
+45 inception_4d/3x3 146.313 428.44 2.80
+46 inception_4d/5x5_reduce 6.423 725.47 0.31
+47 inception_4d/5x5 20.070 474.31 0.84
+48 inception_4d/pool 0.903 657.23 0.29
+49 inception_4d/pool_proj 12.845 583.48 0.44
+50 inception_4e/1x1 52.986 309.60 1.47
+51 inception_4e/3x3_reduce 33.116 279.09 1.28
+52 inception_4e/3x3 180.634 307.91 4.62
+53 inception_4e/5x5_reduce 6.623 594.87 0.39
+54 inception_4e/5x5 40.141 416.06 1.20
+55 inception_4e/pool 0.931 636.86 0.31
+56 inception_4e/pool_proj 26.493 477.56 0.68
+57 pool4/3x3_s2 0.367 1303.53 0.24
+58 inception_5a/1x1 20.873 631.79 0.77
+59 inception_5a/3x3_reduce 13.046 657.84 0.50
+60 inception_5a/3x3 45.158 615.42 1.66
+61 inception_5a/5x5_reduce 2.609 468.53 0.27
+62 inception_5a/5x5 10.035 554.62 0.50
+63 inception_5a/pool 0.367 540.50 0.14
+64 inception_5a/pool_proj 10.437 593.71 0.47
+65 inception_5b/1x1 31.310 667.18 1.03
+66 inception_5b/3x3_reduce 15.655 688.70 0.56
+67 inception_5b/3x3 65.028 799.92 1.79
+68 inception_5b/5x5_reduce 3.914 459.85 0.33
+69 inception_5b/5x5 15.053 563.79 0.73
+70 inception_5b/pool 0.367 533.47 0.15
+71 inception_5b/pool_proj 10.437 592.62 0.47
+72 pool5/7x7_s1 0.100 481.97 0.20
+73 loss3/classifier 0.002 2519.16 0.78
+74 prob 0.003 10.62 0.18
+----------------------------------------------------------------------------------------
+ Total inference time 88.66
+----------------------------------------------------------------------------------------
+```
+
+## Graphical Format
+The mvNCProfile also creates the output_report.html and output.gv.svg files which contain a graphcial representation of the profile information as shown below.
+
+
+
--- /dev/null
+
+ifneq ($(findstring movidius, $(PYTHONPATH)), movidius)
+ export PYTHONPATH:=/opt/movidius/caffe/python:/opt/movidius/mvnc/python:$(PYTHONPATH)
+endif
+
+TOPTARGETS := all clean check profile compile run
+
+SUBDIRS := $(wildcard */.)
+
+$(TOPTARGETS): $(SUBDIRS)
+$(SUBDIRS):
+ $(MAKE) -C $@ $(MAKECMDGOALS)
+
+.PHONY: $(TOPTARGETS) $(SUBDIRS)
+
+.PHONY: help
+help:
+ @echo "Possible Make targets"
+ @echo " make help - shows this message"
+ @echo " make all - Makes all targets"
+ @echo " make clean - Removes all temp files from all directories"
+ @echo " make check - Runs check on all caffe/tensorflow models"
+ @echo " make profile - Runs profile on all caffe/tensorflow models"
+ @echo " make compile - Runs compile on all caffe/tensorflow models"
+ @echo " make run - Runs all caffe/tensorflow/apps"
--- /dev/null
+# Introduction
+A Makefile is provided to compile profile and check all the caffe and tensorflow networks on the Neural Compute Stick using the Neural Compute SDK. The Makefile also builds some example applications that demonstrate the usage of the compiled networks. Each network includes python and/or C++ example code right along with the network and the apps directory contains some other example programs as well.
+
+## make all
+Executing 'make all' builds all the following networks and applications
+
+* Caffe
+ * GoogLeNet
+ * AlexNet
+ * SqueezeNet
+
+* TensorFlow
+ * Inception V1
+ * Inception V3
+
+* Applications
+ * hello_ncs_py
+ * hello_ncs_cpp
+ * multistick_cpp
+
+## make compile
+ Compiles all the networks
+## make profile
+ Runs profile on all the networks
+## make check
+ Runs 'make check' on all the networks
+## make run
+ Runs the python example in all the networks
--- /dev/null
+
+ifneq ($(findstring movidius, $(PYTHONPATH)), movidius)
+ export PYTHONPATH:=/opt/movidius/caffe/python:/opt/movidius/mvnc/python:$(PYTHONPATH)
+endif
+
+TOPTARGETS := all clean
+
+SUBDIRS := $(wildcard */.)
+
+$(TOPTARGETS): $(SUBDIRS)
+$(SUBDIRS):
+ $(MAKE) -C $@ $(MAKECMDGOALS)
+
+.PHONY: $(TOPTARGETS) $(SUBDIRS)
+
+.PHONY: help
+help:
+ @echo "Possible make targets: ";
+ @echo " make help - shows this message";
+ @echo " make all - does make all in each sub directory";
+ @echo " make clean - does make clean in each subdirectory";
+
+check profile compile run:
--- /dev/null
+#Application examples
--- /dev/null
+
+
+
+.PHONY: all
+all: hello_ncs_cpp
+
+
+.PHONY: hello_ncs_cpp
+hello_ncs_cpp:
+ @echo "\nmaking hello_ncs_cpp"
+ g++ cpp/hello_ncs.cpp -o cpp/hello_ncs_cpp -lmvnc
+ @echo "Created cpp/hello_ncs_cpp executable"
+
+.PHONY: run
+run: hello_ncs_cpp
+ @echo "\nmaking run"
+ cd cpp; ./hello_ncs_cpp; cd ..
+
+.PHONY: help
+help:
+ @echo "possible make targets: ";
+ @echo " make help - shows this message";
+ @echo " make all - makes the following: cpp, run_cpp";
+ @echo " make hello_ncs_cpp - builds the hello_ncs_cpp executable example";
+ @echo " make run - runs the hello_ncs_cpp executable example program";
+ @echo " make clean - removes all created content and temporary files";
+
+clean: clean
+ @echo "\nmaking clean";
+ rm -f cpp/hello_ncs_cpp;
--- /dev/null
+// Copyright 2017 Intel Corporation.
+// The source code, information and material ("Material") contained herein is
+// owned by Intel Corporation or its suppliers or licensors, and title to such
+// Material remains with Intel Corporation or its suppliers or licensors.
+// The Material contains proprietary information of Intel or its suppliers and
+// licensors. The Material is protected by worldwide copyright laws and treaty
+// provisions.
+// No part of the Material may be used, copied, reproduced, modified, published,
+// uploaded, posted, transmitted, distributed or disclosed in any way without
+// Intel's prior express written permission. No license under any patent,
+// copyright or other intellectual property rights in the Material is granted to
+// or conferred upon you, either expressly, by implication, inducement, estoppel
+// or otherwise.
+// Any license under such intellectual property rights must be express and
+// approved by Intel in writing.
+
+
+
+#include <stdio.h>
+#include <stdlib.h>
+
+#include <mvnc.h>
+
+// somewhat arbitrary buffer size for the device name
+#define NAME_SIZE 100
+
+
+int main(int argc, char** argv)
+{
+ mvncStatus retCode;
+ void *deviceHandle;
+ char devName[NAME_SIZE];
+ retCode = mvncGetDeviceName(0, devName, NAME_SIZE);
+ if (retCode != MVNC_OK)
+ { // failed to get device name, maybe none plugged in.
+ printf("Error - No NCS devices found.\n");
+ printf(" mvncStatus value: %d\n", retCode);
+ exit(-1);
+ }
+
+ // Try to open the NCS device via the device name
+ retCode = mvncOpenDevice(devName, &deviceHandle);
+ if (retCode != MVNC_OK)
+ { // failed to open the device.
+ printf("Error - Could not open NCS device.\n");
+ printf(" mvncStatus value: %d\n", retCode);
+ exit(-1);
+ }
+
+ // deviceHandle is ready to use now.
+ // Pass it to other NC API calls as needed and close it when finished.
+ printf("Hello NCS! Device opened normally.\n");
+
+ retCode = mvncCloseDevice(deviceHandle);
+ deviceHandle = NULL;
+ if (retCode != MVNC_OK)
+ {
+ printf("Error - Could not close NCS device.\n");
+ printf(" mvncStatus value: %d\n", retCode);
+ exit(-1);
+ }
+
+ printf("Goodbye NCS! Device Closed normally.\n");
+ printf("NCS device working.\n");
+}
--- /dev/null
+# hello_ncs_cpp: A Movidius Neural Compute Stick Hello World example for C++
+
+This directory contains a C++ example that shows the Movidius NC SDK software is installed and is correctly configured on your system such that an application can access the stick via the NC SDK's API.
+
+## Prerequisites
+
+This code example requires that the following components are available:
+1. Movidius Neural Compute Stick
+2. Movidius Neural Compute SDK
+
+
+## Building the example
+To run the example code do the following :
+1. Open a terminal and change directory to the hello_ncs_cpp base directory
+2. Type the following command in the terminal: make hello_ncs_cpp
+
+
+## Running the Example
+To run the example code do the following :
+1. Open a terminal and change directory to the hello_ncs_cpp base directory
+2. Type the following command in the terminal: make run
+
+When the application runs normally and is able to connect to the NCS device the output will be similar to this:
+
+~~~
+Hello NCS! Device opened normally.
+Goodbye NCS! Device closed normally.
+NCS device working.
+~~~
+
+
+
--- /dev/null
+
+
+
+.PHONY: all
+all:
+ @echo "nothing to make, use 'make run' to run."
+
+.PHONY: run
+run:
+ @echo "\nmaking run";
+ python3 hello_ncs.py;
+
+.PHONY: help
+help:
+ @echo "possible make targets: ";
+ @echo " make help - shows this message";
+ @echo " make all - makes the following: run";
+ @echo " make run - runs the hello_ncs_cpp executable example program";
+ @echo " make clean - removes all created content and temporary files";
+
+.PHONY: clean
+clean:
+ @echo "clean done.";
--- /dev/null
+#!/usr/bin/python3
+#
+# Copyright 2017 Intel Corporation.
+# The source code, information and material ("Material") contained herein is
+# owned by Intel Corporation or its suppliers or licensors, and title to such
+# Material remains with Intel Corporation or its suppliers or licensors.
+# The Material contains proprietary information of Intel or its suppliers and
+# licensors. The Material is protected by worldwide copyright laws and treaty
+# provisions.
+# No part of the Material may be used, copied, reproduced, modified, published,
+# uploaded, posted, transmitted, distributed or disclosed in any way without
+# Intel's prior express written permission. No license under any patent,
+# copyright or other intellectual property rights in the Material is granted to
+# or conferred upon you, either expressly, by implication, inducement, estoppel
+# or otherwise.
+# Any license under such intellectual property rights must be express and
+# approved by Intel in writing.
+
+# Python script to open and close a single NCS device
+
+import mvnc.mvncapi as fx
+
+# main entry point for the program
+if __name__=="__main__":
+
+ # set the logging level for the NC API
+ fx.SetGlobalOption(fx.GlobalOption.LOG_LEVEL, 0)
+
+ # get a list of names for all the devices plugged into the system
+ ncs_names = fx.EnumerateDevices()
+ if (len(ncs_names) < 1):
+ print("Error - no NCS devices detected, verify an NCS device is connected.")
+ quit()
+
+
+ # get the first NCS device by its name. For this program we will always open the first NCS device.
+ dev = fx.Device(ncs_names[0])
+
+
+ # try to open the device. this will throw an exception if someone else has it open already
+ try:
+ dev.OpenDevice()
+ except:
+ print("Error - Could not open NCS device.")
+ quit()
+
+
+ print("Hello NCS! Device opened normally.")
+
+
+ try:
+ dev.CloseDevice()
+ except:
+ print("Error - could not close NCS device.")
+ quit()
+
+ print("Goodbye NCS! Device closed normally.")
+ print("NCS device working.")
+
--- /dev/null
+# hello_ncs_cpp: A Movidius Neural Compute Stick Hello World example for Python
+
+This directory contains a python3 example that shows the Movidius NC SDK software is installed and is correctly configured on your system such that an application can access the stick via the NC SDK's API.
+
+## Prerequisites
+
+This code example requires that the following components are available:
+1. Movidius Neural Compute Stick
+2. Movidius Neural Compute SDK
+3. Python3
+
+
+## Running the Example
+To run the example code do the following :
+1. Open a terminal and change directory to the hello_ncs_cpp base directory
+2. Type the following command in the terminal: make run
+
+When the application runs normally and is able to connect to the NCS device the output will be similar to this:
+
+~~~
+Hello NCS! Device opened normally.
+Goodbye NCS! Device closed normally.
+NCS device working.
+~~~
+
+
+
--- /dev/null
+
+
+
+.PHONY: all
+all: multistick_cpp
+
+.PHONY: googlenet
+googlenet:
+ @echo "\nmaking googlenet"
+ (cd ../../caffe/GoogLeNet; make compile; cd ../../apps/multistick_cpp; cp ../../caffe/GoogLeNet/graph ./googlenet.graph;)
+
+.PHONY: squeezenet
+squeezenet:
+ @echo "\nmaking squeezenet"
+ (cd ../../caffe/SqueezeNet; make compile; cd ../../apps/multistick_cpp; cp ../../caffe/SqueezeNet/graph ./squeezenet.graph;)
+
+.PHONY: multistick_cpp
+multistick_cpp: googlenet squeezenet
+ @echo "\nmaking multistick_cpp"
+ cp googlenet.graph cpp/googlenet.graph;
+ cp squeezenet.graph cpp/squeezenet.graph;
+ g++ cpp/multistick.cpp cpp/fp16.c -o cpp/multistick_cpp -lmvnc
+ @echo "Created cpp/multistick_cpp executable"
+
+.PHONY: run
+run: multistick_cpp
+ @echo "\nmaking run";
+ cd cpp; ./multistick_cpp; cd ..;
+
+.PHONY: help
+help:
+ @echo "possible make targets: ";
+ @echo " make help - shows this message";
+ @echo " make all - makes the following: multistick_cpp, run";
+ @echo " make multistick_cpp - builds the multistick_cpp executable example";
+ @echo " make run - runs the multistick_cpp executable example program";
+ @echo " make clean - removes all created content and temporary files";
+
+clean: clean
+ @echo "\nmaking clean";
+ rm -f cpp/multistick_cpp;
+ rm -f cpp/googlenet.graph;
+ rm -f cpp/squeezenet.graph;
+ rm -f squeezenet.graph;
+ rm -f googlenet.graph;
+
--- /dev/null
+#include "fp16.h"
+
+// Copied from Numpy
+
+static unsigned half2float(unsigned short h)
+{
+ unsigned short h_exp, h_sig;
+ unsigned f_sgn, f_exp, f_sig;
+
+ h_exp = (h&0x7c00u);
+ f_sgn = ((unsigned)h&0x8000u) << 16;
+ switch (h_exp) {
+ case 0x0000u: /* 0 or subnormal */
+ h_sig = (h&0x03ffu);
+ /* Signed zero */
+ if (h_sig == 0) {
+ return f_sgn;
+ }
+ /* Subnormal */
+ h_sig <<= 1;
+ while ((h_sig&0x0400u) == 0) {
+ h_sig <<= 1;
+ h_exp++;
+ }
+ f_exp = ((unsigned)(127 - 15 - h_exp)) << 23;
+ f_sig = ((unsigned)(h_sig&0x03ffu)) << 13;
+ return f_sgn + f_exp + f_sig;
+ case 0x7c00u: /* inf or NaN */
+ /* All-ones exponent and a copy of the significand */
+ return f_sgn + 0x7f800000u + (((unsigned)(h&0x03ffu)) << 13);
+ default: /* normalized */
+ /* Just need to adjust the exponent and shift */
+ return f_sgn + (((unsigned)(h&0x7fffu) + 0x1c000u) << 13);
+ }
+}
+
+unsigned short float2half(unsigned f)
+{
+ unsigned f_exp, f_sig;
+ unsigned short h_sgn, h_exp, h_sig;
+
+ h_sgn = (unsigned short) ((f&0x80000000u) >> 16);
+ f_exp = (f&0x7f800000u);
+
+ /* Exponent overflow/NaN converts to signed inf/NaN */
+ if (f_exp >= 0x47800000u) {
+ if (f_exp == 0x7f800000u) {
+ /* Inf or NaN */
+ f_sig = (f&0x007fffffu);
+ if (f_sig != 0) {
+ /* NaN - propagate the flag in the significand... */
+ unsigned short ret = (unsigned short) (0x7c00u + (f_sig >> 13));
+ /* ...but make sure it stays a NaN */
+ if (ret == 0x7c00u) {
+ ret++;
+ }
+ return h_sgn + ret;
+ } else {
+ /* signed inf */
+ return (unsigned short) (h_sgn + 0x7c00u);
+ }
+ } else {
+ /* overflow to signed inf */
+#if NPY_HALF_GENERATE_OVERFLOW
+ npy_set_floatstatus_overflow();
+#endif
+ return (unsigned short) (h_sgn + 0x7c00u);
+ }
+ }
+
+ /* Exponent underflow converts to a subnormal half or signed zero */
+ if (f_exp <= 0x38000000u) {
+ /*
+ * Signed zeros, subnormal floats, and floats with small
+ * exponents all convert to signed zero halfs.
+ */
+ if (f_exp < 0x33000000u) {
+#if NPY_HALF_GENERATE_UNDERFLOW
+ /* If f != 0, it underflowed to 0 */
+ if ((f&0x7fffffff) != 0) {
+ npy_set_floatstatus_underflow();
+ }
+#endif
+ return h_sgn;
+ }
+ /* Make the subnormal significand */
+ f_exp >>= 23;
+ f_sig = (0x00800000u + (f&0x007fffffu));
+#if NPY_HALF_GENERATE_UNDERFLOW
+ /* If it's not exactly represented, it underflowed */
+ if ((f_sig&(((unsigned)1 << (126 - f_exp)) - 1)) != 0) {
+ npy_set_floatstatus_underflow();
+ }
+#endif
+ f_sig >>= (113 - f_exp);
+ /* Handle rounding by adding 1 to the bit beyond half precision */
+#if NPY_HALF_ROUND_TIES_TO_EVEN
+ /*
+ * If the last bit in the half significand is 0 (already even), and
+ * the remaining bit pattern is 1000...0, then we do not add one
+ * to the bit after the half significand. In all other cases, we do.
+ */
+ if ((f_sig&0x00003fffu) != 0x00001000u) {
+ f_sig += 0x00001000u;
+ }
+#else
+ f_sig += 0x00001000u;
+#endif
+ h_sig = (unsigned short) (f_sig >> 13);
+ /*
+ * If the rounding causes a bit to spill into h_exp, it will
+ * increment h_exp from zero to one and h_sig will be zero.
+ * This is the correct result.
+ */
+ return (unsigned short) (h_sgn + h_sig);
+ }
+
+ /* Regular case with no overflow or underflow */
+ h_exp = (unsigned short) ((f_exp - 0x38000000u) >> 13);
+ /* Handle rounding by adding 1 to the bit beyond half precision */
+ f_sig = (f&0x007fffffu);
+#if NPY_HALF_ROUND_TIES_TO_EVEN
+ /*
+ * If the last bit in the half significand is 0 (already even), and
+ * the remaining bit pattern is 1000...0, then we do not add one
+ * to the bit after the half significand. In all other cases, we do.
+ */
+ if ((f_sig&0x00003fffu) != 0x00001000u) {
+ f_sig += 0x00001000u;
+ }
+#else
+ f_sig += 0x00001000u;
+#endif
+ h_sig = (unsigned short) (f_sig >> 13);
+ /*
+ * If the rounding causes a bit to spill into h_exp, it will
+ * increment h_exp by one and h_sig will be zero. This is the
+ * correct result. h_exp may increment to 15, at greatest, in
+ * which case the result overflows to a signed inf.
+ */
+#if NPY_HALF_GENERATE_OVERFLOW
+ h_sig += h_exp;
+ if (h_sig == 0x7c00u) {
+ npy_set_floatstatus_overflow();
+ }
+ return h_sgn + h_sig;
+#else
+ return h_sgn + h_exp + h_sig;
+#endif
+}
+
+void floattofp16(unsigned char *dst, float *src, unsigned nelem)
+{
+ unsigned i;
+ unsigned short *_dst = (unsigned short *)dst;
+ unsigned *_src = (unsigned *)src;
+
+ for(i = 0; i < nelem; i++)
+ _dst[i] = float2half(_src[i]);
+}
+
+void fp16tofloat(float *dst, unsigned char *src, unsigned nelem)
+{
+ unsigned i;
+ unsigned *_dst = (unsigned *)dst;
+ unsigned short *_src = (unsigned short *)src;
+
+ for(i = 0; i < nelem; i++)
+ _dst[i] = half2float(_src[i]);
+}
+
--- /dev/null
+// Copied from Numpy
+
+
+static unsigned half2float(unsigned short h);
+unsigned short float2half(unsigned f);
+void floattofp16(unsigned char *dst, float *src, unsigned nelem);
+void fp16tofloat(float *dst, unsigned char *src, unsigned nelem);
+
+
--- /dev/null
+// Copyright 2017 Intel Corporation.
+// The source code, information and material ("Material") contained herein is
+// owned by Intel Corporation or its suppliers or licensors, and title to such
+// Material remains with Intel Corporation or its suppliers or licensors.
+// The Material contains proprietary information of Intel or its suppliers and
+// licensors. The Material is protected by worldwide copyright laws and treaty
+// provisions.
+// No part of the Material may be used, copied, reproduced, modified, published,
+// uploaded, posted, transmitted, distributed or disclosed in any way without
+// Intel's prior express written permission. No license under any patent,
+// copyright or other intellectual property rights in the Material is granted to
+// or conferred upon you, either expressly, by implication, inducement, estoppel
+// or otherwise.
+// Any license under such intellectual property rights must be express and
+// approved by Intel in writing.
+
+#include <stdio.h>
+#include <stdlib.h>
+
+#define STB_IMAGE_IMPLEMENTATION
+#include "stb_image.h"
+
+#define STB_IMAGE_RESIZE_IMPLEMENTATION
+#include "stb_image_resize.h"
+
+#include "fp16.h"
+#include <mvnc.h>
+
+
+// somewhat arbitrary buffer size for the device name
+#define NAME_SIZE 100
+
+// from current director to examples base director
+// #define APP_BASE_DIR "../"
+
+#define EXAMPLES_BASE_DIR "../../../"
+
+// graph file names - assume the graph file is in the current directory.
+#define GOOGLENET_GRAPH_FILE_NAME "googlenet.graph"
+#define SQUEEZENET_GRAPH_FILE_NAME "squeezenet.graph"
+
+// image file name - assume we are running in this directory: ncsdk/examples/caffe/GoogLeNet/cpp
+#define GOOGLENET_IMAGE_FILE_NAME EXAMPLES_BASE_DIR "data/images/nps_electric_guitar.png"
+#define SQUEEZENET_IMAGE_FILE_NAME EXAMPLES_BASE_DIR "data/images/nps_baseball.png"
+
+
+// 16 bits. will use this to store half precision floats since C++ has no
+// built in support for it.
+typedef unsigned short half;
+
+// GoogleNet image dimensions, network mean values for each channel in BGR order.
+const int networkDimGoogleNet = 224;
+const int networkDimSqueezeNet = 227;
+float networkMeanGoogleNet[] = {0.40787054*255.0, 0.45752458*255.0, 0.48109378*255.0};
+float networkMeanSqueezeNet[] = {0.40787054*255.0, 0.45752458*255.0, 0.48109378*255.0};
+
+// Prototypes
+void *LoadGraphFile(const char *path, unsigned int *length);
+half *LoadImage(const char *path, int reqsize, float *mean);
+// end prototypes
+
+// Reads a graph file from the file system and copies it to a buffer
+// that is allocated internally via malloc.
+// Param path is a pointer to a null terminate string that must be set to the path to the
+// graph file on disk before calling
+// Param length must must point to an integer that will get set to the number of bytes
+// allocated for the buffer
+// Returns pointer to the buffer allcoated.
+// Note: The caller must free the buffer returned.
+void *LoadGraphFile(const char *path, unsigned int *length)
+{
+ FILE *fp;
+ char *buf;
+
+ fp = fopen(path, "rb");
+ if(fp == NULL)
+ return 0;
+ fseek(fp, 0, SEEK_END);
+ *length = ftell(fp);
+ rewind(fp);
+ if(!(buf = (char*) malloc(*length)))
+ {
+ fclose(fp);
+ return 0;
+ }
+ if(fread(buf, 1, *length, fp) != *length)
+ {
+ fclose(fp);
+ free(buf);
+ return 0;
+ }
+ fclose(fp);
+ return buf;
+}
+
+// Reads an image file from disk (8 bit per channel RGB .jpg or .png or other formats
+// supported by stbi_load.) Resizes it, subtracts the mean from each channel, and then
+// converts to an array of half precision floats that is suitable to pass to mvncLoadTensor.
+// The returned array will contain 3 floats for each pixel in the image the first float
+// for a pixel is it's the Blue channel value the next is Green and then Red. The array
+// contains the pixel values in row major order.
+// Param path is a pointer to a null terminated string that must be set to the path of the
+// to read before calling.
+// Param reqsize must be set to the width and height that the image will be resized to.
+// Its assumed width and height are the same size.
+// Param mean must be set to point to an array of 3 floating point numbers. The three
+// numbers are the mean values for the blue, green, and red channels in that order.
+// each B, G, and R value from the image will have this value subtracted from it.
+// Returns a pointer to a buffer that is allocated internally via malloc. this buffer contains
+// the 16 bit float values that can be passed to mvncLoadTensor(). The returned buffer
+// will contain reqSize*reqSize*3 half floats.
+half *LoadImage(const char *path, int reqSize, float *mean)
+{
+ int width, height, cp, i;
+ unsigned char *img, *imgresized;
+ float *imgfp32;
+ half *imgfp16;
+
+ img = stbi_load(path, &width, &height, &cp, 3);
+ if(!img)
+ {
+ printf("Error - the image file %s could not be loaded\n", path);
+ return NULL;
+ }
+ imgresized = (unsigned char*) malloc(3*reqSize*reqSize);
+ if(!imgresized)
+ {
+ free(img);
+ perror("malloc");
+ return NULL;
+ }
+ stbir_resize_uint8(img, width, height, 0, imgresized, reqSize, reqSize, 0, 3);
+ free(img);
+ imgfp32 = (float*) malloc(sizeof(*imgfp32) * reqSize * reqSize * 3);
+ if(!imgfp32)
+ {
+ free(imgresized);
+ perror("malloc");
+ return NULL;
+ }
+ for(i = 0; i < reqSize * reqSize * 3; i++)
+ imgfp32[i] = imgresized[i];
+ free(imgresized);
+ imgfp16 = (half*) malloc(sizeof(*imgfp16) * reqSize * reqSize * 3);
+ if(!imgfp16)
+ {
+ free(imgfp32);
+ perror("malloc");
+ return NULL;
+ }
+ for(i = 0; i < reqSize*reqSize; i++)
+ {
+ float blue, green, red;
+ blue = imgfp32[3*i+2];
+ green = imgfp32[3*i+1];
+ red = imgfp32[3*i+0];
+
+ imgfp32[3*i+0] = blue-mean[0];
+ imgfp32[3*i+1] = green-mean[1];
+ imgfp32[3*i+2] = red-mean[2];
+
+ // uncomment to see what values are getting passed to mvncLoadTensor() before conversion to half float
+ //printf("Blue: %f, Grean: %f, Red: %f \n", imgfp32[3*i+0], imgfp32[3*i+1], imgfp32[3*i+2]);
+ }
+ floattofp16((unsigned char *)imgfp16, imgfp32, 3*reqSize*reqSize);
+ free(imgfp32);
+ return imgfp16;
+}
+
+
+// Opens one NCS device.
+// Param deviceIndex is the zero-based index of the device to open
+// Param deviceHandle is the address of a device handle that will be set
+// if opening is successful
+// Returns true if works or false if doesn't.
+bool OpenOneNCS(int deviceIndex, void** deviceHandle)
+{
+ mvncStatus retCode;
+ char devName[NAME_SIZE];
+ retCode = mvncGetDeviceName(deviceIndex, devName, NAME_SIZE);
+ if (retCode != MVNC_OK)
+ { // failed to get this device's name, maybe none plugged in.
+ printf("Error - NCS device at index %d not found\n", deviceIndex);
+ return false;
+ }
+
+ // Try to open the NCS device via the device name
+ retCode = mvncOpenDevice(devName, deviceHandle);
+ if (retCode != MVNC_OK)
+ { // failed to open the device.
+ printf("Error - Could not open NCS device at index %d\n", deviceIndex);
+ return false;
+ }
+
+ // deviceHandle is ready to use now.
+ // Pass it to other NC API calls as needed and close it when finished.
+ printf("Successfully opened NCS device at index %d!\n", deviceIndex);
+ return true;
+}
+
+
+// Loads a compiled network graph onto the NCS device.
+// Param deviceHandle is the open device handle for the device that will allocate the graph
+// Param graphFilename is the name of the compiled network graph file to load on the NCS
+// Param graphHandle is the address of the graph handle that will be created internally.
+// the caller must call mvncDeallocateGraph when done with the handle.
+// Returns true if works or false if doesn't.
+bool LoadGraphToNCS(void* deviceHandle, const char* graphFilename, void** graphHandle)
+{
+ mvncStatus retCode;
+
+ // Read in a graph file
+ unsigned int graphFileLen;
+ void* graphFileBuf = LoadGraphFile(graphFilename, &graphFileLen);
+
+ // allocate the graph
+ retCode = mvncAllocateGraph(deviceHandle, graphHandle, graphFileBuf, graphFileLen);
+ free(graphFileBuf);
+ if (retCode != MVNC_OK)
+ { // error allocating graph
+ printf("Could not allocate graph for file: %s\n", graphFilename);
+ printf("Error from mvncAllocateGraph is: %d\n", retCode);
+ return false;
+ }
+
+ // successfully allocated graph. Now graphHandle is ready to go.
+ // use graphHandle for other API calls and call mvncDeallocateGraph
+ // when done with it.
+ printf("Successfully allocated graph for %s\n", graphFilename);
+
+ return true;
+}
+
+
+// Runs an inference and outputs result to console
+// Param graphHandle is the graphHandle from mvncAllocateGraph for the network that
+// will be used for the inference
+// Param imageFileName is the name of the image file that will be used as input for
+// the neural network for the inference
+// Param networkDim is the height and width (assumed to be the same) for images that the
+// network expects. The image will be resized to this prior to inference.
+// Param networkMean is pointer to array of 3 floats that are the mean values for the network
+// for each color channel, blue, green, and red in that order.
+// Returns tru if works or false if doesn't
+bool DoInferenceOnImageFile(void* graphHandle, const char* imageFileName, int networkDim, float* networkMean)
+{
+ mvncStatus retCode;
+
+ // LoadImage will read image from disk, convert channels to floats
+ // subtract network mean for each value in each channel. Then, convert
+ // floats to half precision floats and return pointer to the buffer
+ // of half precision floats (Fp16s)
+ half* imageBufFp16 = LoadImage(imageFileName, networkDim, networkMean);
+
+ // calculate the length of the buffer that contains the half precision floats.
+ // 3 channels * width * height * sizeof a 16bit float
+ unsigned int lenBufFp16 = 3*networkDim*networkDim*sizeof(*imageBufFp16);
+
+ // start the inference with mvncLoadTensor()
+ retCode = mvncLoadTensor(graphHandle, imageBufFp16, lenBufFp16, NULL);
+ free(imageBufFp16);
+ if (retCode != MVNC_OK)
+ { // error loading tensor
+ printf("Error - Could not load tensor\n");
+ printf(" mvncStatus from mvncLoadTensor is: %d\n", retCode);
+ return false;
+ }
+
+ // the inference has been started, now call mvncGetResult() for the
+ // inference result
+ printf("Successfully loaded the tensor for image %s\n", imageFileName);
+
+ void* resultData16;
+ void* userParam;
+ unsigned int lenResultData;
+ retCode = mvncGetResult(graphHandle, &resultData16, &lenResultData, &userParam);
+ if (retCode != MVNC_OK)
+ {
+ printf("Error - Could not get result for image %s\n", imageFileName);
+ printf(" mvncStatus from mvncGetResult is: %d\n", retCode);
+ return false;
+ }
+
+ // Successfully got the result. The inference result is in the buffer pointed to by resultData
+ printf("Successfully got the inference result for image %s\n", imageFileName);
+ //printf("resultData is %d bytes which is %d 16-bit floats.\n", lenResultData, lenResultData/(int)sizeof(half));
+
+ // convert half precision floats to full floats
+ int numResults = lenResultData / sizeof(half);
+ float* resultData32;
+ resultData32 = (float*)malloc(numResults * sizeof(*resultData32));
+ fp16tofloat(resultData32, (unsigned char*)resultData16, numResults);
+
+ float maxResult = 0.0;
+ int maxIndex = -1;
+ for (int index = 0; index < numResults; index++)
+ {
+ // printf("Category %d is: %f\n", index, resultData32[index]);
+ if (resultData32[index] > maxResult)
+ {
+ maxResult = resultData32[index];
+ maxIndex = index;
+ }
+ }
+ printf("Index of top result is: %d\n", maxIndex);
+ printf("Probability of top result is: %f\n", resultData32[maxIndex]);
+}
+
+// Main entry point for the program
+int main(int argc, char** argv)
+{
+ mvncStatus retCode;
+ void* devHandle1;
+ void* devHandle2;
+ void* graphHandleGoogleNet;
+ void* graphHandleSqueezeNet;
+
+ if (!OpenOneNCS(0, &devHandle1))
+ { // couldn't open first NCS device
+ exit(-1);
+ }
+ if (!OpenOneNCS(1, &devHandle2))
+ { // couldn't open second NCS device
+ exit(-1);
+ }
+
+ void* deviceHandle = devHandle1;
+
+ if (!LoadGraphToNCS(devHandle1, GOOGLENET_GRAPH_FILE_NAME, &graphHandleGoogleNet))
+ {
+ mvncCloseDevice(devHandle1);
+ mvncCloseDevice(devHandle2);
+ exit(-2);
+ }
+ if (!LoadGraphToNCS(devHandle2, SQUEEZENET_GRAPH_FILE_NAME, &graphHandleSqueezeNet))
+ {
+ mvncDeallocateGraph(graphHandleGoogleNet);
+ graphHandleGoogleNet = NULL;
+ mvncCloseDevice(devHandle1);
+ mvncCloseDevice(devHandle2);
+ exit(-2);
+ }
+
+ printf("\n--- NCS 1 inference ---\n");
+ DoInferenceOnImageFile(graphHandleGoogleNet, GOOGLENET_IMAGE_FILE_NAME, networkDimGoogleNet, networkMeanGoogleNet);
+ printf("-----------------------\n");
+
+ printf("\n--- NCS 2 inference ---\n");
+ DoInferenceOnImageFile(graphHandleSqueezeNet, SQUEEZENET_IMAGE_FILE_NAME, networkDimSqueezeNet, networkMeanSqueezeNet);
+ printf("-----------------------\n");
+
+ retCode = mvncDeallocateGraph(graphHandleSqueezeNet);
+ graphHandleSqueezeNet = NULL;
+ retCode = mvncDeallocateGraph(graphHandleGoogleNet);
+ graphHandleGoogleNet = NULL;
+
+ retCode = mvncCloseDevice(devHandle1);
+ devHandle1 = NULL;
+
+ retCode = mvncCloseDevice(devHandle2);
+ devHandle2 = NULL;
+}
--- /dev/null
+/* stb_image - v2.15 - public domain image loader - http://nothings.org/stb_image.h
+ no warranty implied; use at your own risk
+
+ Do this:
+ #define STB_IMAGE_IMPLEMENTATION
+ before you include this file in *one* C or C++ file to create the implementation.
+
+ // i.e. it should look like this:
+ #include ...
+ #include ...
+ #include ...
+ #define STB_IMAGE_IMPLEMENTATION
+ #include "stb_image.h"
+
+ You can #define STBI_ASSERT(x) before the #include to avoid using assert.h.
+ And #define STBI_MALLOC, STBI_REALLOC, and STBI_FREE to avoid using malloc,realloc,free
+
+
+ QUICK NOTES:
+ Primarily of interest to game developers and other people who can
+ avoid problematic images and only need the trivial interface
+
+ JPEG baseline & progressive (12 bpc/arithmetic not supported, same as stock IJG lib)
+ PNG 1/2/4/8/16-bit-per-channel
+
+ TGA (not sure what subset, if a subset)
+ BMP non-1bpp, non-RLE
+ PSD (composited view only, no extra channels, 8/16 bit-per-channel)
+
+ GIF (*comp always reports as 4-channel)
+ HDR (radiance rgbE format)
+ PIC (Softimage PIC)
+ PNM (PPM and PGM binary only)
+
+ Animated GIF still needs a proper API, but here's one way to do it:
+ http://gist.github.com/urraka/685d9a6340b26b830d49
+
+ - decode from memory or through FILE (define STBI_NO_STDIO to remove code)
+ - decode from arbitrary I/O callbacks
+ - SIMD acceleration on x86/x64 (SSE2) and ARM (NEON)
+
+ Full documentation under "DOCUMENTATION" below.
+
+
+LICENSE
+
+ See end of file for license information.
+
+RECENT REVISION HISTORY:
+
+ 2.15 (2017-03-18) fix png-1,2,4; all Imagenet JPGs; no runtime SSE detection on GCC
+ 2.14 (2017-03-03) remove deprecated STBI_JPEG_OLD; fixes for Imagenet JPGs
+ 2.13 (2016-12-04) experimental 16-bit API, only for PNG so far; fixes
+ 2.12 (2016-04-02) fix typo in 2.11 PSD fix that caused crashes
+ 2.11 (2016-04-02) 16-bit PNGS; enable SSE2 in non-gcc x64
+ RGB-format JPEG; remove white matting in PSD;
+ allocate large structures on the stack;
+ correct channel count for PNG & BMP
+ 2.10 (2016-01-22) avoid warning introduced in 2.09
+ 2.09 (2016-01-16) 16-bit TGA; comments in PNM files; STBI_REALLOC_SIZED
+ 2.08 (2015-09-13) fix to 2.07 cleanup, reading RGB PSD as RGBA
+ 2.07 (2015-09-13) partial animated GIF support
+ limited 16-bit PSD support
+ minor bugs, code cleanup, and compiler warnings
+
+ See end of file for full revision history.
+
+
+ ============================ Contributors =========================
+
+ Image formats Extensions, features
+ Sean Barrett (jpeg, png, bmp) Jetro Lauha (stbi_info)
+ Nicolas Schulz (hdr, psd) Martin "SpartanJ" Golini (stbi_info)
+ Jonathan Dummer (tga) James "moose2000" Brown (iPhone PNG)
+ Jean-Marc Lienher (gif) Ben "Disch" Wenger (io callbacks)
+ Tom Seddon (pic) Omar Cornut (1/2/4-bit PNG)
+ Thatcher Ulrich (psd) Nicolas Guillemot (vertical flip)
+ Ken Miller (pgm, ppm) Richard Mitton (16-bit PSD)
+ github:urraka (animated gif) Junggon Kim (PNM comments)
+ Daniel Gibson (16-bit TGA)
+ socks-the-fox (16-bit PNG)
+ Jeremy Sawicki (handle all ImageNet JPGs)
+ Optimizations & bugfixes
+ Fabian "ryg" Giesen
+ Arseny Kapoulkine
+
+ Bug & warning fixes
+ Marc LeBlanc David Woo Guillaume George Martins Mozeiko
+ Christpher Lloyd Jerry Jansson Joseph Thomson Phil Jordan
+ Dave Moore Roy Eltham Hayaki Saito Nathan Reed
+ Won Chun Luke Graham Johan Duparc Nick Verigakis
+ the Horde3D community Thomas Ruf Ronny Chevalier Baldur Karlsson
+ Janez Zemva John Bartholomew Michal Cichon github:rlyeh
+ Jonathan Blow Ken Hamada Tero Hanninen github:romigrou
+ Laurent Gomila Cort Stratton Sergio Gonzalez github:svdijk
+ Aruelien Pocheville Thibault Reuille Cass Everitt github:snagar
+ Ryamond Barbiero Paul Du Bois Engin Manap github:Zelex
+ Michaelangel007@github Philipp Wiesemann Dale Weiler github:grim210
+ Oriol Ferrer Mesia Josh Tobin Matthew Gregan github:sammyhw
+ Blazej Dariusz Roszkowski Gregory Mullen github:phprus
+
+*/
+
+#ifndef STBI_INCLUDE_STB_IMAGE_H
+#define STBI_INCLUDE_STB_IMAGE_H
+
+// DOCUMENTATION
+//
+// Limitations:
+// - no 16-bit-per-channel PNG
+// - no 12-bit-per-channel JPEG
+// - no JPEGs with arithmetic coding
+// - no 1-bit BMP
+// - GIF always returns *comp=4
+//
+// Basic usage (see HDR discussion below for HDR usage):
+// int x,y,n;
+// unsigned char *data = stbi_load(filename, &x, &y, &n, 0);
+// // ... process data if not NULL ...
+// // ... x = width, y = height, n = # 8-bit components per pixel ...
+// // ... replace '0' with '1'..'4' to force that many components per pixel
+// // ... but 'n' will always be the number that it would have been if you said 0
+// stbi_image_free(data)
+//
+// Standard parameters:
+// int *x -- outputs image width in pixels
+// int *y -- outputs image height in pixels
+// int *channels_in_file -- outputs # of image components in image file
+// int desired_channels -- if non-zero, # of image components requested in result
+//
+// The return value from an image loader is an 'unsigned char *' which points
+// to the pixel data, or NULL on an allocation failure or if the image is
+// corrupt or invalid. The pixel data consists of *y scanlines of *x pixels,
+// with each pixel consisting of N interleaved 8-bit components; the first
+// pixel pointed to is top-left-most in the image. There is no padding between
+// image scanlines or between pixels, regardless of format. The number of
+// components N is 'req_comp' if req_comp is non-zero, or *comp otherwise.
+// If req_comp is non-zero, *comp has the number of components that _would_
+// have been output otherwise. E.g. if you set req_comp to 4, you will always
+// get RGBA output, but you can check *comp to see if it's trivially opaque
+// because e.g. there were only 3 channels in the source image.
+//
+// An output image with N components has the following components interleaved
+// in this order in each pixel:
+//
+// N=#comp components
+// 1 grey
+// 2 grey, alpha
+// 3 red, green, blue
+// 4 red, green, blue, alpha
+//
+// If image loading fails for any reason, the return value will be NULL,
+// and *x, *y, *comp will be unchanged. The function stbi_failure_reason()
+// can be queried for an extremely brief, end-user unfriendly explanation
+// of why the load failed. Define STBI_NO_FAILURE_STRINGS to avoid
+// compiling these strings at all, and STBI_FAILURE_USERMSG to get slightly
+// more user-friendly ones.
+//
+// Paletted PNG, BMP, GIF, and PIC images are automatically depalettized.
+//
+// ===========================================================================
+//
+// Philosophy
+//
+// stb libraries are designed with the following priorities:
+//
+// 1. easy to use
+// 2. easy to maintain
+// 3. good performance
+//
+// Sometimes I let "good performance" creep up in priority over "easy to maintain",
+// and for best performance I may provide less-easy-to-use APIs that give higher
+// performance, in addition to the easy to use ones. Nevertheless, it's important
+// to keep in mind that from the standpoint of you, a client of this library,
+// all you care about is #1 and #3, and stb libraries DO NOT emphasize #3 above all.
+//
+// Some secondary priorities arise directly from the first two, some of which
+// make more explicit reasons why performance can't be emphasized.
+//
+// - Portable ("ease of use")
+// - Small source code footprint ("easy to maintain")
+// - No dependencies ("ease of use")
+//
+// ===========================================================================
+//
+// I/O callbacks
+//
+// I/O callbacks allow you to read from arbitrary sources, like packaged
+// files or some other source. Data read from callbacks are processed
+// through a small internal buffer (currently 128 bytes) to try to reduce
+// overhead.
+//
+// The three functions you must define are "read" (reads some bytes of data),
+// "skip" (skips some bytes of data), "eof" (reports if the stream is at the end).
+//
+// ===========================================================================
+//
+// SIMD support
+//
+// The JPEG decoder will try to automatically use SIMD kernels on x86 when
+// supported by the compiler. For ARM Neon support, you must explicitly
+// request it.
+//
+// (The old do-it-yourself SIMD API is no longer supported in the current
+// code.)
+//
+// On x86, SSE2 will automatically be used when available based on a run-time
+// test; if not, the generic C versions are used as a fall-back. On ARM targets,
+// the typical path is to have separate builds for NEON and non-NEON devices
+// (at least this is true for iOS and Android). Therefore, the NEON support is
+// toggled by a build flag: define STBI_NEON to get NEON loops.
+//
+// If for some reason you do not want to use any of SIMD code, or if
+// you have issues compiling it, you can disable it entirely by
+// defining STBI_NO_SIMD.
+//
+// ===========================================================================
+//
+// HDR image support (disable by defining STBI_NO_HDR)
+//
+// stb_image now supports loading HDR images in general, and currently
+// the Radiance .HDR file format, although the support is provided
+// generically. You can still load any file through the existing interface;
+// if you attempt to load an HDR file, it will be automatically remapped to
+// LDR, assuming gamma 2.2 and an arbitrary scale factor defaulting to 1;
+// both of these constants can be reconfigured through this interface:
+//
+// stbi_hdr_to_ldr_gamma(2.2f);
+// stbi_hdr_to_ldr_scale(1.0f);
+//
+// (note, do not use _inverse_ constants; stbi_image will invert them
+// appropriately).
+//
+// Additionally, there is a new, parallel interface for loading files as
+// (linear) floats to preserve the full dynamic range:
+//
+// float *data = stbi_loadf(filename, &x, &y, &n, 0);
+//
+// If you load LDR images through this interface, those images will
+// be promoted to floating point values, run through the inverse of
+// constants corresponding to the above:
+//
+// stbi_ldr_to_hdr_scale(1.0f);
+// stbi_ldr_to_hdr_gamma(2.2f);
+//
+// Finally, given a filename (or an open file or memory block--see header
+// file for details) containing image data, you can query for the "most
+// appropriate" interface to use (that is, whether the image is HDR or
+// not), using:
+//
+// stbi_is_hdr(char *filename);
+//
+// ===========================================================================
+//
+// iPhone PNG support:
+//
+// By default we convert iphone-formatted PNGs back to RGB, even though
+// they are internally encoded differently. You can disable this conversion
+// by by calling stbi_convert_iphone_png_to_rgb(0), in which case
+// you will always just get the native iphone "format" through (which
+// is BGR stored in RGB).
+//
+// Call stbi_set_unpremultiply_on_load(1) as well to force a divide per
+// pixel to remove any premultiplied alpha *only* if the image file explicitly
+// says there's premultiplied data (currently only happens in iPhone images,
+// and only if iPhone convert-to-rgb processing is on).
+//
+// ===========================================================================
+//
+// ADDITIONAL CONFIGURATION
+//
+// - You can suppress implementation of any of the decoders to reduce
+// your code footprint by #defining one or more of the following
+// symbols before creating the implementation.
+//
+// STBI_NO_JPEG
+// STBI_NO_PNG
+// STBI_NO_BMP
+// STBI_NO_PSD
+// STBI_NO_TGA
+// STBI_NO_GIF
+// STBI_NO_HDR
+// STBI_NO_PIC
+// STBI_NO_PNM (.ppm and .pgm)
+//
+// - You can request *only* certain decoders and suppress all other ones
+// (this will be more forward-compatible, as addition of new decoders
+// doesn't require you to disable them explicitly):
+//
+// STBI_ONLY_JPEG
+// STBI_ONLY_PNG
+// STBI_ONLY_BMP
+// STBI_ONLY_PSD
+// STBI_ONLY_TGA
+// STBI_ONLY_GIF
+// STBI_ONLY_HDR
+// STBI_ONLY_PIC
+// STBI_ONLY_PNM (.ppm and .pgm)
+//
+// - If you use STBI_NO_PNG (or _ONLY_ without PNG), and you still
+// want the zlib decoder to be available, #define STBI_SUPPORT_ZLIB
+//
+
+
+#ifndef STBI_NO_STDIO
+#include <stdio.h>
+#endif // STBI_NO_STDIO
+
+#define STBI_VERSION 1
+
+enum
+{
+ STBI_default = 0, // only used for req_comp
+
+ STBI_grey = 1,
+ STBI_grey_alpha = 2,
+ STBI_rgb = 3,
+ STBI_rgb_alpha = 4
+};
+
+typedef unsigned char stbi_uc;
+typedef unsigned short stbi_us;
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifdef STB_IMAGE_STATIC
+#define STBIDEF static
+#else
+#define STBIDEF extern
+#endif
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// PRIMARY API - works on images of any type
+//
+
+//
+// load image by filename, open file, or memory buffer
+//
+
+typedef struct
+{
+ int (*read) (void *user,char *data,int size); // fill 'data' with 'size' bytes. return number of bytes actually read
+ void (*skip) (void *user,int n); // skip the next 'n' bytes, or 'unget' the last -n bytes if negative
+ int (*eof) (void *user); // returns nonzero if we are at end of file/data
+} stbi_io_callbacks;
+
+////////////////////////////////////
+//
+// 8-bits-per-channel interface
+//
+
+STBIDEF stbi_uc *stbi_load (char const *filename, int *x, int *y, int *channels_in_file, int desired_channels);
+STBIDEF stbi_uc *stbi_load_from_memory (stbi_uc const *buffer, int len , int *x, int *y, int *channels_in_file, int desired_channels);
+STBIDEF stbi_uc *stbi_load_from_callbacks(stbi_io_callbacks const *clbk , void *user, int *x, int *y, int *channels_in_file, int desired_channels);
+
+#ifndef STBI_NO_STDIO
+STBIDEF stbi_uc *stbi_load_from_file (FILE *f, int *x, int *y, int *channels_in_file, int desired_channels);
+// for stbi_load_from_file, file pointer is left pointing immediately after image
+#endif
+
+////////////////////////////////////
+//
+// 16-bits-per-channel interface
+//
+
+STBIDEF stbi_us *stbi_load_16(char const *filename, int *x, int *y, int *channels_in_file, int desired_channels);
+#ifndef STBI_NO_STDIO
+STBIDEF stbi_us *stbi_load_from_file_16(FILE *f, int *x, int *y, int *channels_in_file, int desired_channels);
+#endif
+// @TODO the other variants
+
+////////////////////////////////////
+//
+// float-per-channel interface
+//
+#ifndef STBI_NO_LINEAR
+ STBIDEF float *stbi_loadf (char const *filename, int *x, int *y, int *channels_in_file, int desired_channels);
+ STBIDEF float *stbi_loadf_from_memory (stbi_uc const *buffer, int len, int *x, int *y, int *channels_in_file, int desired_channels);
+ STBIDEF float *stbi_loadf_from_callbacks (stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *channels_in_file, int desired_channels);
+
+ #ifndef STBI_NO_STDIO
+ STBIDEF float *stbi_loadf_from_file (FILE *f, int *x, int *y, int *channels_in_file, int desired_channels);
+ #endif
+#endif
+
+#ifndef STBI_NO_HDR
+ STBIDEF void stbi_hdr_to_ldr_gamma(float gamma);
+ STBIDEF void stbi_hdr_to_ldr_scale(float scale);
+#endif // STBI_NO_HDR
+
+#ifndef STBI_NO_LINEAR
+ STBIDEF void stbi_ldr_to_hdr_gamma(float gamma);
+ STBIDEF void stbi_ldr_to_hdr_scale(float scale);
+#endif // STBI_NO_LINEAR
+
+// stbi_is_hdr is always defined, but always returns false if STBI_NO_HDR
+STBIDEF int stbi_is_hdr_from_callbacks(stbi_io_callbacks const *clbk, void *user);
+STBIDEF int stbi_is_hdr_from_memory(stbi_uc const *buffer, int len);
+#ifndef STBI_NO_STDIO
+STBIDEF int stbi_is_hdr (char const *filename);
+STBIDEF int stbi_is_hdr_from_file(FILE *f);
+#endif // STBI_NO_STDIO
+
+
+// get a VERY brief reason for failure
+// NOT THREADSAFE
+STBIDEF const char *stbi_failure_reason (void);
+
+// free the loaded image -- this is just free()
+STBIDEF void stbi_image_free (void *retval_from_stbi_load);
+
+// get image dimensions & components without fully decoding
+STBIDEF int stbi_info_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp);
+STBIDEF int stbi_info_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp);
+
+#ifndef STBI_NO_STDIO
+STBIDEF int stbi_info (char const *filename, int *x, int *y, int *comp);
+STBIDEF int stbi_info_from_file (FILE *f, int *x, int *y, int *comp);
+
+#endif
+
+
+
+// for image formats that explicitly notate that they have premultiplied alpha,
+// we just return the colors as stored in the file. set this flag to force
+// unpremultiplication. results are undefined if the unpremultiply overflow.
+STBIDEF void stbi_set_unpremultiply_on_load(int flag_true_if_should_unpremultiply);
+
+// indicate whether we should process iphone images back to canonical format,
+// or just pass them through "as-is"
+STBIDEF void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert);
+
+// flip the image vertically, so the first pixel in the output array is the bottom left
+STBIDEF void stbi_set_flip_vertically_on_load(int flag_true_if_should_flip);
+
+// ZLIB client - used by PNG, available for other purposes
+
+STBIDEF char *stbi_zlib_decode_malloc_guesssize(const char *buffer, int len, int initial_size, int *outlen);
+STBIDEF char *stbi_zlib_decode_malloc_guesssize_headerflag(const char *buffer, int len, int initial_size, int *outlen, int parse_header);
+STBIDEF char *stbi_zlib_decode_malloc(const char *buffer, int len, int *outlen);
+STBIDEF int stbi_zlib_decode_buffer(char *obuffer, int olen, const char *ibuffer, int ilen);
+
+STBIDEF char *stbi_zlib_decode_noheader_malloc(const char *buffer, int len, int *outlen);
+STBIDEF int stbi_zlib_decode_noheader_buffer(char *obuffer, int olen, const char *ibuffer, int ilen);
+
+
+#ifdef __cplusplus
+}
+#endif
+
+//
+//
+//// end header file /////////////////////////////////////////////////////
+#endif // STBI_INCLUDE_STB_IMAGE_H
+
+#ifdef STB_IMAGE_IMPLEMENTATION
+
+#if defined(STBI_ONLY_JPEG) || defined(STBI_ONLY_PNG) || defined(STBI_ONLY_BMP) \
+ || defined(STBI_ONLY_TGA) || defined(STBI_ONLY_GIF) || defined(STBI_ONLY_PSD) \
+ || defined(STBI_ONLY_HDR) || defined(STBI_ONLY_PIC) || defined(STBI_ONLY_PNM) \
+ || defined(STBI_ONLY_ZLIB)
+ #ifndef STBI_ONLY_JPEG
+ #define STBI_NO_JPEG
+ #endif
+ #ifndef STBI_ONLY_PNG
+ #define STBI_NO_PNG
+ #endif
+ #ifndef STBI_ONLY_BMP
+ #define STBI_NO_BMP
+ #endif
+ #ifndef STBI_ONLY_PSD
+ #define STBI_NO_PSD
+ #endif
+ #ifndef STBI_ONLY_TGA
+ #define STBI_NO_TGA
+ #endif
+ #ifndef STBI_ONLY_GIF
+ #define STBI_NO_GIF
+ #endif
+ #ifndef STBI_ONLY_HDR
+ #define STBI_NO_HDR
+ #endif
+ #ifndef STBI_ONLY_PIC
+ #define STBI_NO_PIC
+ #endif
+ #ifndef STBI_ONLY_PNM
+ #define STBI_NO_PNM
+ #endif
+#endif
+
+#if defined(STBI_NO_PNG) && !defined(STBI_SUPPORT_ZLIB) && !defined(STBI_NO_ZLIB)
+#define STBI_NO_ZLIB
+#endif
+
+
+#include <stdarg.h>
+#include <stddef.h> // ptrdiff_t on osx
+#include <stdlib.h>
+#include <string.h>
+#include <limits.h>
+
+#if !defined(STBI_NO_LINEAR) || !defined(STBI_NO_HDR)
+#include <math.h> // ldexp
+#endif
+
+#ifndef STBI_NO_STDIO
+#include <stdio.h>
+#endif
+
+#ifndef STBI_ASSERT
+#include <assert.h>
+#define STBI_ASSERT(x) assert(x)
+#endif
+
+
+#ifndef _MSC_VER
+ #ifdef __cplusplus
+ #define stbi_inline inline
+ #else
+ #define stbi_inline
+ #endif
+#else
+ #define stbi_inline __forceinline
+#endif
+
+
+#ifdef _MSC_VER
+typedef unsigned short stbi__uint16;
+typedef signed short stbi__int16;
+typedef unsigned int stbi__uint32;
+typedef signed int stbi__int32;
+#else
+#include <stdint.h>
+typedef uint16_t stbi__uint16;
+typedef int16_t stbi__int16;
+typedef uint32_t stbi__uint32;
+typedef int32_t stbi__int32;
+#endif
+
+// should produce compiler error if size is wrong
+typedef unsigned char validate_uint32[sizeof(stbi__uint32)==4 ? 1 : -1];
+
+#ifdef _MSC_VER
+#define STBI_NOTUSED(v) (void)(v)
+#else
+#define STBI_NOTUSED(v) (void)sizeof(v)
+#endif
+
+#ifdef _MSC_VER
+#define STBI_HAS_LROTL
+#endif
+
+#ifdef STBI_HAS_LROTL
+ #define stbi_lrot(x,y) _lrotl(x,y)
+#else
+ #define stbi_lrot(x,y) (((x) << (y)) | ((x) >> (32 - (y))))
+#endif
+
+#if defined(STBI_MALLOC) && defined(STBI_FREE) && (defined(STBI_REALLOC) || defined(STBI_REALLOC_SIZED))
+// ok
+#elif !defined(STBI_MALLOC) && !defined(STBI_FREE) && !defined(STBI_REALLOC) && !defined(STBI_REALLOC_SIZED)
+// ok
+#else
+#error "Must define all or none of STBI_MALLOC, STBI_FREE, and STBI_REALLOC (or STBI_REALLOC_SIZED)."
+#endif
+
+#ifndef STBI_MALLOC
+#define STBI_MALLOC(sz) malloc(sz)
+#define STBI_REALLOC(p,newsz) realloc(p,newsz)
+#define STBI_FREE(p) free(p)
+#endif
+
+#ifndef STBI_REALLOC_SIZED
+#define STBI_REALLOC_SIZED(p,oldsz,newsz) STBI_REALLOC(p,newsz)
+#endif
+
+// x86/x64 detection
+#if defined(__x86_64__) || defined(_M_X64)
+#define STBI__X64_TARGET
+#elif defined(__i386) || defined(_M_IX86)
+#define STBI__X86_TARGET
+#endif
+
+#if defined(__GNUC__) && defined(STBI__X86_TARGET) && !defined(__SSE2__) && !defined(STBI_NO_SIMD)
+// gcc doesn't support sse2 intrinsics unless you compile with -msse2,
+// which in turn means it gets to use SSE2 everywhere. This is unfortunate,
+// but previous attempts to provide the SSE2 functions with runtime
+// detection caused numerous issues. The way architecture extensions are
+// exposed in GCC/Clang is, sadly, not really suited for one-file libs.
+// New behavior: if compiled with -msse2, we use SSE2 without any
+// detection; if not, we don't use it at all.
+#define STBI_NO_SIMD
+#endif
+
+#if defined(__MINGW32__) && defined(STBI__X86_TARGET) && !defined(STBI_MINGW_ENABLE_SSE2) && !defined(STBI_NO_SIMD)
+// Note that __MINGW32__ doesn't actually mean 32-bit, so we have to avoid STBI__X64_TARGET
+//
+// 32-bit MinGW wants ESP to be 16-byte aligned, but this is not in the
+// Windows ABI and VC++ as well as Windows DLLs don't maintain that invariant.
+// As a result, enabling SSE2 on 32-bit MinGW is dangerous when not
+// simultaneously enabling "-mstackrealign".
+//
+// See https://github.com/nothings/stb/issues/81 for more information.
+//
+// So default to no SSE2 on 32-bit MinGW. If you've read this far and added
+// -mstackrealign to your build settings, feel free to #define STBI_MINGW_ENABLE_SSE2.
+#define STBI_NO_SIMD
+#endif
+
+#if !defined(STBI_NO_SIMD) && (defined(STBI__X86_TARGET) || defined(STBI__X64_TARGET))
+#define STBI_SSE2
+#include <emmintrin.h>
+
+#ifdef _MSC_VER
+
+#if _MSC_VER >= 1400 // not VC6
+#include <intrin.h> // __cpuid
+static int stbi__cpuid3(void)
+{
+ int info[4];
+ __cpuid(info,1);
+ return info[3];
+}
+#else
+static int stbi__cpuid3(void)
+{
+ int res;
+ __asm {
+ mov eax,1
+ cpuid
+ mov res,edx
+ }
+ return res;
+}
+#endif
+
+#define STBI_SIMD_ALIGN(type, name) __declspec(align(16)) type name
+
+static int stbi__sse2_available()
+{
+ int info3 = stbi__cpuid3();
+ return ((info3 >> 26) & 1) != 0;
+}
+#else // assume GCC-style if not VC++
+#define STBI_SIMD_ALIGN(type, name) type name __attribute__((aligned(16)))
+
+static int stbi__sse2_available()
+{
+ // If we're even attempting to compile this on GCC/Clang, that means
+ // -msse2 is on, which means the compiler is allowed to use SSE2
+ // instructions at will, and so are we.
+ return 1;
+}
+#endif
+#endif
+
+// ARM NEON
+#if defined(STBI_NO_SIMD) && defined(STBI_NEON)
+#undef STBI_NEON
+#endif
+
+#ifdef STBI_NEON
+#include <arm_neon.h>
+// assume GCC or Clang on ARM targets
+#define STBI_SIMD_ALIGN(type, name) type name __attribute__((aligned(16)))
+#endif
+
+#ifndef STBI_SIMD_ALIGN
+#define STBI_SIMD_ALIGN(type, name) type name
+#endif
+
+///////////////////////////////////////////////
+//
+// stbi__context struct and start_xxx functions
+
+// stbi__context structure is our basic context used by all images, so it
+// contains all the IO context, plus some basic image information
+typedef struct
+{
+ stbi__uint32 img_x, img_y;
+ int img_n, img_out_n;
+
+ stbi_io_callbacks io;
+ void *io_user_data;
+
+ int read_from_callbacks;
+ int buflen;
+ stbi_uc buffer_start[128];
+
+ stbi_uc *img_buffer, *img_buffer_end;
+ stbi_uc *img_buffer_original, *img_buffer_original_end;
+} stbi__context;
+
+
+static void stbi__refill_buffer(stbi__context *s);
+
+// initialize a memory-decode context
+static void stbi__start_mem(stbi__context *s, stbi_uc const *buffer, int len)
+{
+ s->io.read = NULL;
+ s->read_from_callbacks = 0;
+ s->img_buffer = s->img_buffer_original = (stbi_uc *) buffer;
+ s->img_buffer_end = s->img_buffer_original_end = (stbi_uc *) buffer+len;
+}
+
+// initialize a callback-based context
+static void stbi__start_callbacks(stbi__context *s, stbi_io_callbacks *c, void *user)
+{
+ s->io = *c;
+ s->io_user_data = user;
+ s->buflen = sizeof(s->buffer_start);
+ s->read_from_callbacks = 1;
+ s->img_buffer_original = s->buffer_start;
+ stbi__refill_buffer(s);
+ s->img_buffer_original_end = s->img_buffer_end;
+}
+
+#ifndef STBI_NO_STDIO
+
+static int stbi__stdio_read(void *user, char *data, int size)
+{
+ return (int) fread(data,1,size,(FILE*) user);
+}
+
+static void stbi__stdio_skip(void *user, int n)
+{
+ fseek((FILE*) user, n, SEEK_CUR);
+}
+
+static int stbi__stdio_eof(void *user)
+{
+ return feof((FILE*) user);
+}
+
+static stbi_io_callbacks stbi__stdio_callbacks =
+{
+ stbi__stdio_read,
+ stbi__stdio_skip,
+ stbi__stdio_eof,
+};
+
+static void stbi__start_file(stbi__context *s, FILE *f)
+{
+ stbi__start_callbacks(s, &stbi__stdio_callbacks, (void *) f);
+}
+
+//static void stop_file(stbi__context *s) { }
+
+#endif // !STBI_NO_STDIO
+
+static void stbi__rewind(stbi__context *s)
+{
+ // conceptually rewind SHOULD rewind to the beginning of the stream,
+ // but we just rewind to the beginning of the initial buffer, because
+ // we only use it after doing 'test', which only ever looks at at most 92 bytes
+ s->img_buffer = s->img_buffer_original;
+ s->img_buffer_end = s->img_buffer_original_end;
+}
+
+enum
+{
+ STBI_ORDER_RGB,
+ STBI_ORDER_BGR
+};
+
+typedef struct
+{
+ int bits_per_channel;
+ int num_channels;
+ int channel_order;
+} stbi__result_info;
+
+#ifndef STBI_NO_JPEG
+static int stbi__jpeg_test(stbi__context *s);
+static void *stbi__jpeg_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__jpeg_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_PNG
+static int stbi__png_test(stbi__context *s);
+static void *stbi__png_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__png_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_BMP
+static int stbi__bmp_test(stbi__context *s);
+static void *stbi__bmp_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__bmp_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_TGA
+static int stbi__tga_test(stbi__context *s);
+static void *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__tga_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_PSD
+static int stbi__psd_test(stbi__context *s);
+static void *stbi__psd_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri, int bpc);
+static int stbi__psd_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_HDR
+static int stbi__hdr_test(stbi__context *s);
+static float *stbi__hdr_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__hdr_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_PIC
+static int stbi__pic_test(stbi__context *s);
+static void *stbi__pic_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__pic_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_GIF
+static int stbi__gif_test(stbi__context *s);
+static void *stbi__gif_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__gif_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_PNM
+static int stbi__pnm_test(stbi__context *s);
+static void *stbi__pnm_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__pnm_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+// this is not threadsafe
+static const char *stbi__g_failure_reason;
+
+STBIDEF const char *stbi_failure_reason(void)
+{
+ return stbi__g_failure_reason;
+}
+
+static int stbi__err(const char *str)
+{
+ stbi__g_failure_reason = str;
+ return 0;
+}
+
+static void *stbi__malloc(size_t size)
+{
+ return STBI_MALLOC(size);
+}
+
+// stb_image uses ints pervasively, including for offset calculations.
+// therefore the largest decoded image size we can support with the
+// current code, even on 64-bit targets, is INT_MAX. this is not a
+// significant limitation for the intended use case.
+//
+// we do, however, need to make sure our size calculations don't
+// overflow. hence a few helper functions for size calculations that
+// multiply integers together, making sure that they're non-negative
+// and no overflow occurs.
+
+// return 1 if the sum is valid, 0 on overflow.
+// negative terms are considered invalid.
+static int stbi__addsizes_valid(int a, int b)
+{
+ if (b < 0) return 0;
+ // now 0 <= b <= INT_MAX, hence also
+ // 0 <= INT_MAX - b <= INTMAX.
+ // And "a + b <= INT_MAX" (which might overflow) is the
+ // same as a <= INT_MAX - b (no overflow)
+ return a <= INT_MAX - b;
+}
+
+// returns 1 if the product is valid, 0 on overflow.
+// negative factors are considered invalid.
+static int stbi__mul2sizes_valid(int a, int b)
+{
+ if (a < 0 || b < 0) return 0;
+ if (b == 0) return 1; // mul-by-0 is always safe
+ // portable way to check for no overflows in a*b
+ return a <= INT_MAX/b;
+}
+
+// returns 1 if "a*b + add" has no negative terms/factors and doesn't overflow
+static int stbi__mad2sizes_valid(int a, int b, int add)
+{
+ return stbi__mul2sizes_valid(a, b) && stbi__addsizes_valid(a*b, add);
+}
+
+// returns 1 if "a*b*c + add" has no negative terms/factors and doesn't overflow
+static int stbi__mad3sizes_valid(int a, int b, int c, int add)
+{
+ return stbi__mul2sizes_valid(a, b) && stbi__mul2sizes_valid(a*b, c) &&
+ stbi__addsizes_valid(a*b*c, add);
+}
+
+// returns 1 if "a*b*c*d + add" has no negative terms/factors and doesn't overflow
+static int stbi__mad4sizes_valid(int a, int b, int c, int d, int add)
+{
+ return stbi__mul2sizes_valid(a, b) && stbi__mul2sizes_valid(a*b, c) &&
+ stbi__mul2sizes_valid(a*b*c, d) && stbi__addsizes_valid(a*b*c*d, add);
+}
+
+// mallocs with size overflow checking
+static void *stbi__malloc_mad2(int a, int b, int add)
+{
+ if (!stbi__mad2sizes_valid(a, b, add)) return NULL;
+ return stbi__malloc(a*b + add);
+}
+
+static void *stbi__malloc_mad3(int a, int b, int c, int add)
+{
+ if (!stbi__mad3sizes_valid(a, b, c, add)) return NULL;
+ return stbi__malloc(a*b*c + add);
+}
+
+static void *stbi__malloc_mad4(int a, int b, int c, int d, int add)
+{
+ if (!stbi__mad4sizes_valid(a, b, c, d, add)) return NULL;
+ return stbi__malloc(a*b*c*d + add);
+}
+
+// stbi__err - error
+// stbi__errpf - error returning pointer to float
+// stbi__errpuc - error returning pointer to unsigned char
+
+#ifdef STBI_NO_FAILURE_STRINGS
+ #define stbi__err(x,y) 0
+#elif defined(STBI_FAILURE_USERMSG)
+ #define stbi__err(x,y) stbi__err(y)
+#else
+ #define stbi__err(x,y) stbi__err(x)
+#endif
+
+#define stbi__errpf(x,y) ((float *)(size_t) (stbi__err(x,y)?NULL:NULL))
+#define stbi__errpuc(x,y) ((unsigned char *)(size_t) (stbi__err(x,y)?NULL:NULL))
+
+STBIDEF void stbi_image_free(void *retval_from_stbi_load)
+{
+ STBI_FREE(retval_from_stbi_load);
+}
+
+#ifndef STBI_NO_LINEAR
+static float *stbi__ldr_to_hdr(stbi_uc *data, int x, int y, int comp);
+#endif
+
+#ifndef STBI_NO_HDR
+static stbi_uc *stbi__hdr_to_ldr(float *data, int x, int y, int comp);
+#endif
+
+static int stbi__vertically_flip_on_load = 0;
+
+STBIDEF void stbi_set_flip_vertically_on_load(int flag_true_if_should_flip)
+{
+ stbi__vertically_flip_on_load = flag_true_if_should_flip;
+}
+
+static void *stbi__load_main(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri, int bpc)
+{
+ memset(ri, 0, sizeof(*ri)); // make sure it's initialized if we add new fields
+ ri->bits_per_channel = 8; // default is 8 so most paths don't have to be changed
+ ri->channel_order = STBI_ORDER_RGB; // all current input & output are this, but this is here so we can add BGR order
+ ri->num_channels = 0;
+
+ #ifndef STBI_NO_JPEG
+ if (stbi__jpeg_test(s)) return stbi__jpeg_load(s,x,y,comp,req_comp, ri);
+ #endif
+ #ifndef STBI_NO_PNG
+ if (stbi__png_test(s)) return stbi__png_load(s,x,y,comp,req_comp, ri);
+ #endif
+ #ifndef STBI_NO_BMP
+ if (stbi__bmp_test(s)) return stbi__bmp_load(s,x,y,comp,req_comp, ri);
+ #endif
+ #ifndef STBI_NO_GIF
+ if (stbi__gif_test(s)) return stbi__gif_load(s,x,y,comp,req_comp, ri);
+ #endif
+ #ifndef STBI_NO_PSD
+ if (stbi__psd_test(s)) return stbi__psd_load(s,x,y,comp,req_comp, ri, bpc);
+ #endif
+ #ifndef STBI_NO_PIC
+ if (stbi__pic_test(s)) return stbi__pic_load(s,x,y,comp,req_comp, ri);
+ #endif
+ #ifndef STBI_NO_PNM
+ if (stbi__pnm_test(s)) return stbi__pnm_load(s,x,y,comp,req_comp, ri);
+ #endif
+
+ #ifndef STBI_NO_HDR
+ if (stbi__hdr_test(s)) {
+ float *hdr = stbi__hdr_load(s, x,y,comp,req_comp, ri);
+ return stbi__hdr_to_ldr(hdr, *x, *y, req_comp ? req_comp : *comp);
+ }
+ #endif
+
+ #ifndef STBI_NO_TGA
+ // test tga last because it's a crappy test!
+ if (stbi__tga_test(s))
+ return stbi__tga_load(s,x,y,comp,req_comp, ri);
+ #endif
+
+ return stbi__errpuc("unknown image type", "Image not of any known type, or corrupt");
+}
+
+static stbi_uc *stbi__convert_16_to_8(stbi__uint16 *orig, int w, int h, int channels)
+{
+ int i;
+ int img_len = w * h * channels;
+ stbi_uc *reduced;
+
+ reduced = (stbi_uc *) stbi__malloc(img_len);
+ if (reduced == NULL) return stbi__errpuc("outofmem", "Out of memory");
+
+ for (i = 0; i < img_len; ++i)
+ reduced[i] = (stbi_uc)((orig[i] >> 8) & 0xFF); // top half of each byte is sufficient approx of 16->8 bit scaling
+
+ STBI_FREE(orig);
+ return reduced;
+}
+
+static stbi__uint16 *stbi__convert_8_to_16(stbi_uc *orig, int w, int h, int channels)
+{
+ int i;
+ int img_len = w * h * channels;
+ stbi__uint16 *enlarged;
+
+ enlarged = (stbi__uint16 *) stbi__malloc(img_len*2);
+ if (enlarged == NULL) return (stbi__uint16 *) stbi__errpuc("outofmem", "Out of memory");
+
+ for (i = 0; i < img_len; ++i)
+ enlarged[i] = (stbi__uint16)((orig[i] << 8) + orig[i]); // replicate to high and low byte, maps 0->0, 255->0xffff
+
+ STBI_FREE(orig);
+ return enlarged;
+}
+
+static unsigned char *stbi__load_and_postprocess_8bit(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__result_info ri;
+ void *result = stbi__load_main(s, x, y, comp, req_comp, &ri, 8);
+
+ if (result == NULL)
+ return NULL;
+
+ if (ri.bits_per_channel != 8) {
+ STBI_ASSERT(ri.bits_per_channel == 16);
+ result = stbi__convert_16_to_8((stbi__uint16 *) result, *x, *y, req_comp == 0 ? *comp : req_comp);
+ ri.bits_per_channel = 8;
+ }
+
+ // @TODO: move stbi__convert_format to here
+
+ if (stbi__vertically_flip_on_load) {
+ int w = *x, h = *y;
+ int channels = req_comp ? req_comp : *comp;
+ int row,col,z;
+ stbi_uc *image = (stbi_uc *) result;
+
+ // @OPTIMIZE: use a bigger temp buffer and memcpy multiple pixels at once
+ for (row = 0; row < (h>>1); row++) {
+ for (col = 0; col < w; col++) {
+ for (z = 0; z < channels; z++) {
+ stbi_uc temp = image[(row * w + col) * channels + z];
+ image[(row * w + col) * channels + z] = image[((h - row - 1) * w + col) * channels + z];
+ image[((h - row - 1) * w + col) * channels + z] = temp;
+ }
+ }
+ }
+ }
+
+ return (unsigned char *) result;
+}
+
+static stbi__uint16 *stbi__load_and_postprocess_16bit(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__result_info ri;
+ void *result = stbi__load_main(s, x, y, comp, req_comp, &ri, 16);
+
+ if (result == NULL)
+ return NULL;
+
+ if (ri.bits_per_channel != 16) {
+ STBI_ASSERT(ri.bits_per_channel == 8);
+ result = stbi__convert_8_to_16((stbi_uc *) result, *x, *y, req_comp == 0 ? *comp : req_comp);
+ ri.bits_per_channel = 16;
+ }
+
+ // @TODO: move stbi__convert_format16 to here
+ // @TODO: special case RGB-to-Y (and RGBA-to-YA) for 8-bit-to-16-bit case to keep more precision
+
+ if (stbi__vertically_flip_on_load) {
+ int w = *x, h = *y;
+ int channels = req_comp ? req_comp : *comp;
+ int row,col,z;
+ stbi__uint16 *image = (stbi__uint16 *) result;
+
+ // @OPTIMIZE: use a bigger temp buffer and memcpy multiple pixels at once
+ for (row = 0; row < (h>>1); row++) {
+ for (col = 0; col < w; col++) {
+ for (z = 0; z < channels; z++) {
+ stbi__uint16 temp = image[(row * w + col) * channels + z];
+ image[(row * w + col) * channels + z] = image[((h - row - 1) * w + col) * channels + z];
+ image[((h - row - 1) * w + col) * channels + z] = temp;
+ }
+ }
+ }
+ }
+
+ return (stbi__uint16 *) result;
+}
+
+#ifndef STBI_NO_HDR
+static void stbi__float_postprocess(float *result, int *x, int *y, int *comp, int req_comp)
+{
+ if (stbi__vertically_flip_on_load && result != NULL) {
+ int w = *x, h = *y;
+ int depth = req_comp ? req_comp : *comp;
+ int row,col,z;
+ float temp;
+
+ // @OPTIMIZE: use a bigger temp buffer and memcpy multiple pixels at once
+ for (row = 0; row < (h>>1); row++) {
+ for (col = 0; col < w; col++) {
+ for (z = 0; z < depth; z++) {
+ temp = result[(row * w + col) * depth + z];
+ result[(row * w + col) * depth + z] = result[((h - row - 1) * w + col) * depth + z];
+ result[((h - row - 1) * w + col) * depth + z] = temp;
+ }
+ }
+ }
+ }
+}
+#endif
+
+#ifndef STBI_NO_STDIO
+
+static FILE *stbi__fopen(char const *filename, char const *mode)
+{
+ FILE *f;
+#if defined(_MSC_VER) && _MSC_VER >= 1400
+ if (0 != fopen_s(&f, filename, mode))
+ f=0;
+#else
+ f = fopen(filename, mode);
+#endif
+ return f;
+}
+
+
+STBIDEF stbi_uc *stbi_load(char const *filename, int *x, int *y, int *comp, int req_comp)
+{
+ FILE *f = stbi__fopen(filename, "rb");
+ unsigned char *result;
+ if (!f) return stbi__errpuc("can't fopen", "Unable to open file");
+ result = stbi_load_from_file(f,x,y,comp,req_comp);
+ fclose(f);
+ return result;
+}
+
+STBIDEF stbi_uc *stbi_load_from_file(FILE *f, int *x, int *y, int *comp, int req_comp)
+{
+ unsigned char *result;
+ stbi__context s;
+ stbi__start_file(&s,f);
+ result = stbi__load_and_postprocess_8bit(&s,x,y,comp,req_comp);
+ if (result) {
+ // need to 'unget' all the characters in the IO buffer
+ fseek(f, - (int) (s.img_buffer_end - s.img_buffer), SEEK_CUR);
+ }
+ return result;
+}
+
+STBIDEF stbi__uint16 *stbi_load_from_file_16(FILE *f, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__uint16 *result;
+ stbi__context s;
+ stbi__start_file(&s,f);
+ result = stbi__load_and_postprocess_16bit(&s,x,y,comp,req_comp);
+ if (result) {
+ // need to 'unget' all the characters in the IO buffer
+ fseek(f, - (int) (s.img_buffer_end - s.img_buffer), SEEK_CUR);
+ }
+ return result;
+}
+
+STBIDEF stbi_us *stbi_load_16(char const *filename, int *x, int *y, int *comp, int req_comp)
+{
+ FILE *f = stbi__fopen(filename, "rb");
+ stbi__uint16 *result;
+ if (!f) return (stbi_us *) stbi__errpuc("can't fopen", "Unable to open file");
+ result = stbi_load_from_file_16(f,x,y,comp,req_comp);
+ fclose(f);
+ return result;
+}
+
+
+#endif //!STBI_NO_STDIO
+
+STBIDEF stbi_uc *stbi_load_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__context s;
+ stbi__start_mem(&s,buffer,len);
+ return stbi__load_and_postprocess_8bit(&s,x,y,comp,req_comp);
+}
+
+STBIDEF stbi_uc *stbi_load_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__context s;
+ stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user);
+ return stbi__load_and_postprocess_8bit(&s,x,y,comp,req_comp);
+}
+
+#ifndef STBI_NO_LINEAR
+static float *stbi__loadf_main(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+{
+ unsigned char *data;
+ #ifndef STBI_NO_HDR
+ if (stbi__hdr_test(s)) {
+ stbi__result_info ri;
+ float *hdr_data = stbi__hdr_load(s,x,y,comp,req_comp, &ri);
+ if (hdr_data)
+ stbi__float_postprocess(hdr_data,x,y,comp,req_comp);
+ return hdr_data;
+ }
+ #endif
+ data = stbi__load_and_postprocess_8bit(s, x, y, comp, req_comp);
+ if (data)
+ return stbi__ldr_to_hdr(data, *x, *y, req_comp ? req_comp : *comp);
+ return stbi__errpf("unknown image type", "Image not of any known type, or corrupt");
+}
+
+STBIDEF float *stbi_loadf_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__context s;
+ stbi__start_mem(&s,buffer,len);
+ return stbi__loadf_main(&s,x,y,comp,req_comp);
+}
+
+STBIDEF float *stbi_loadf_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__context s;
+ stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user);
+ return stbi__loadf_main(&s,x,y,comp,req_comp);
+}
+
+#ifndef STBI_NO_STDIO
+STBIDEF float *stbi_loadf(char const *filename, int *x, int *y, int *comp, int req_comp)
+{
+ float *result;
+ FILE *f = stbi__fopen(filename, "rb");
+ if (!f) return stbi__errpf("can't fopen", "Unable to open file");
+ result = stbi_loadf_from_file(f,x,y,comp,req_comp);
+ fclose(f);
+ return result;
+}
+
+STBIDEF float *stbi_loadf_from_file(FILE *f, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__context s;
+ stbi__start_file(&s,f);
+ return stbi__loadf_main(&s,x,y,comp,req_comp);
+}
+#endif // !STBI_NO_STDIO
+
+#endif // !STBI_NO_LINEAR
+
+// these is-hdr-or-not is defined independent of whether STBI_NO_LINEAR is
+// defined, for API simplicity; if STBI_NO_LINEAR is defined, it always
+// reports false!
+
+STBIDEF int stbi_is_hdr_from_memory(stbi_uc const *buffer, int len)
+{
+ #ifndef STBI_NO_HDR
+ stbi__context s;
+ stbi__start_mem(&s,buffer,len);
+ return stbi__hdr_test(&s);
+ #else
+ STBI_NOTUSED(buffer);
+ STBI_NOTUSED(len);
+ return 0;
+ #endif
+}
+
+#ifndef STBI_NO_STDIO
+STBIDEF int stbi_is_hdr (char const *filename)
+{
+ FILE *f = stbi__fopen(filename, "rb");
+ int result=0;
+ if (f) {
+ result = stbi_is_hdr_from_file(f);
+ fclose(f);
+ }
+ return result;
+}
+
+STBIDEF int stbi_is_hdr_from_file(FILE *f)
+{
+ #ifndef STBI_NO_HDR
+ stbi__context s;
+ stbi__start_file(&s,f);
+ return stbi__hdr_test(&s);
+ #else
+ STBI_NOTUSED(f);
+ return 0;
+ #endif
+}
+#endif // !STBI_NO_STDIO
+
+STBIDEF int stbi_is_hdr_from_callbacks(stbi_io_callbacks const *clbk, void *user)
+{
+ #ifndef STBI_NO_HDR
+ stbi__context s;
+ stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user);
+ return stbi__hdr_test(&s);
+ #else
+ STBI_NOTUSED(clbk);
+ STBI_NOTUSED(user);
+ return 0;
+ #endif
+}
+
+#ifndef STBI_NO_LINEAR
+static float stbi__l2h_gamma=2.2f, stbi__l2h_scale=1.0f;
+
+STBIDEF void stbi_ldr_to_hdr_gamma(float gamma) { stbi__l2h_gamma = gamma; }
+STBIDEF void stbi_ldr_to_hdr_scale(float scale) { stbi__l2h_scale = scale; }
+#endif
+
+static float stbi__h2l_gamma_i=1.0f/2.2f, stbi__h2l_scale_i=1.0f;
+
+STBIDEF void stbi_hdr_to_ldr_gamma(float gamma) { stbi__h2l_gamma_i = 1/gamma; }
+STBIDEF void stbi_hdr_to_ldr_scale(float scale) { stbi__h2l_scale_i = 1/scale; }
+
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// Common code used by all image loaders
+//
+
+enum
+{
+ STBI__SCAN_load=0,
+ STBI__SCAN_type,
+ STBI__SCAN_header
+};
+
+static void stbi__refill_buffer(stbi__context *s)
+{
+ int n = (s->io.read)(s->io_user_data,(char*)s->buffer_start,s->buflen);
+ if (n == 0) {
+ // at end of file, treat same as if from memory, but need to handle case
+ // where s->img_buffer isn't pointing to safe memory, e.g. 0-byte file
+ s->read_from_callbacks = 0;
+ s->img_buffer = s->buffer_start;
+ s->img_buffer_end = s->buffer_start+1;
+ *s->img_buffer = 0;
+ } else {
+ s->img_buffer = s->buffer_start;
+ s->img_buffer_end = s->buffer_start + n;
+ }
+}
+
+stbi_inline static stbi_uc stbi__get8(stbi__context *s)
+{
+ if (s->img_buffer < s->img_buffer_end)
+ return *s->img_buffer++;
+ if (s->read_from_callbacks) {
+ stbi__refill_buffer(s);
+ return *s->img_buffer++;
+ }
+ return 0;
+}
+
+stbi_inline static int stbi__at_eof(stbi__context *s)
+{
+ if (s->io.read) {
+ if (!(s->io.eof)(s->io_user_data)) return 0;
+ // if feof() is true, check if buffer = end
+ // special case: we've only got the special 0 character at the end
+ if (s->read_from_callbacks == 0) return 1;
+ }
+
+ return s->img_buffer >= s->img_buffer_end;
+}
+
+static void stbi__skip(stbi__context *s, int n)
+{
+ if (n < 0) {
+ s->img_buffer = s->img_buffer_end;
+ return;
+ }
+ if (s->io.read) {
+ int blen = (int) (s->img_buffer_end - s->img_buffer);
+ if (blen < n) {
+ s->img_buffer = s->img_buffer_end;
+ (s->io.skip)(s->io_user_data, n - blen);
+ return;
+ }
+ }
+ s->img_buffer += n;
+}
+
+static int stbi__getn(stbi__context *s, stbi_uc *buffer, int n)
+{
+ if (s->io.read) {
+ int blen = (int) (s->img_buffer_end - s->img_buffer);
+ if (blen < n) {
+ int res, count;
+
+ memcpy(buffer, s->img_buffer, blen);
+
+ count = (s->io.read)(s->io_user_data, (char*) buffer + blen, n - blen);
+ res = (count == (n-blen));
+ s->img_buffer = s->img_buffer_end;
+ return res;
+ }
+ }
+
+ if (s->img_buffer+n <= s->img_buffer_end) {
+ memcpy(buffer, s->img_buffer, n);
+ s->img_buffer += n;
+ return 1;
+ } else
+ return 0;
+}
+
+static int stbi__get16be(stbi__context *s)
+{
+ int z = stbi__get8(s);
+ return (z << 8) + stbi__get8(s);
+}
+
+static stbi__uint32 stbi__get32be(stbi__context *s)
+{
+ stbi__uint32 z = stbi__get16be(s);
+ return (z << 16) + stbi__get16be(s);
+}
+
+#if defined(STBI_NO_BMP) && defined(STBI_NO_TGA) && defined(STBI_NO_GIF)
+// nothing
+#else
+static int stbi__get16le(stbi__context *s)
+{
+ int z = stbi__get8(s);
+ return z + (stbi__get8(s) << 8);
+}
+#endif
+
+#ifndef STBI_NO_BMP
+static stbi__uint32 stbi__get32le(stbi__context *s)
+{
+ stbi__uint32 z = stbi__get16le(s);
+ return z + (stbi__get16le(s) << 16);
+}
+#endif
+
+#define STBI__BYTECAST(x) ((stbi_uc) ((x) & 255)) // truncate int to byte without warnings
+
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// generic converter from built-in img_n to req_comp
+// individual types do this automatically as much as possible (e.g. jpeg
+// does all cases internally since it needs to colorspace convert anyway,
+// and it never has alpha, so very few cases ). png can automatically
+// interleave an alpha=255 channel, but falls back to this for other cases
+//
+// assume data buffer is malloced, so malloc a new one and free that one
+// only failure mode is malloc failing
+
+static stbi_uc stbi__compute_y(int r, int g, int b)
+{
+ return (stbi_uc) (((r*77) + (g*150) + (29*b)) >> 8);
+}
+
+static unsigned char *stbi__convert_format(unsigned char *data, int img_n, int req_comp, unsigned int x, unsigned int y)
+{
+ int i,j;
+ unsigned char *good;
+
+ if (req_comp == img_n) return data;
+ STBI_ASSERT(req_comp >= 1 && req_comp <= 4);
+
+ good = (unsigned char *) stbi__malloc_mad3(req_comp, x, y, 0);
+ if (good == NULL) {
+ STBI_FREE(data);
+ return stbi__errpuc("outofmem", "Out of memory");
+ }
+
+ for (j=0; j < (int) y; ++j) {
+ unsigned char *src = data + j * x * img_n ;
+ unsigned char *dest = good + j * x * req_comp;
+
+ #define STBI__COMBO(a,b) ((a)*8+(b))
+ #define STBI__CASE(a,b) case STBI__COMBO(a,b): for(i=x-1; i >= 0; --i, src += a, dest += b)
+ // convert source image with img_n components to one with req_comp components;
+ // avoid switch per pixel, so use switch per scanline and massive macros
+ switch (STBI__COMBO(img_n, req_comp)) {
+ STBI__CASE(1,2) { dest[0]=src[0], dest[1]=255; } break;
+ STBI__CASE(1,3) { dest[0]=dest[1]=dest[2]=src[0]; } break;
+ STBI__CASE(1,4) { dest[0]=dest[1]=dest[2]=src[0], dest[3]=255; } break;
+ STBI__CASE(2,1) { dest[0]=src[0]; } break;
+ STBI__CASE(2,3) { dest[0]=dest[1]=dest[2]=src[0]; } break;
+ STBI__CASE(2,4) { dest[0]=dest[1]=dest[2]=src[0], dest[3]=src[1]; } break;
+ STBI__CASE(3,4) { dest[0]=src[0],dest[1]=src[1],dest[2]=src[2],dest[3]=255; } break;
+ STBI__CASE(3,1) { dest[0]=stbi__compute_y(src[0],src[1],src[2]); } break;
+ STBI__CASE(3,2) { dest[0]=stbi__compute_y(src[0],src[1],src[2]), dest[1] = 255; } break;
+ STBI__CASE(4,1) { dest[0]=stbi__compute_y(src[0],src[1],src[2]); } break;
+ STBI__CASE(4,2) { dest[0]=stbi__compute_y(src[0],src[1],src[2]), dest[1] = src[3]; } break;
+ STBI__CASE(4,3) { dest[0]=src[0],dest[1]=src[1],dest[2]=src[2]; } break;
+ default: STBI_ASSERT(0);
+ }
+ #undef STBI__CASE
+ }
+
+ STBI_FREE(data);
+ return good;
+}
+
+static stbi__uint16 stbi__compute_y_16(int r, int g, int b)
+{
+ return (stbi__uint16) (((r*77) + (g*150) + (29*b)) >> 8);
+}
+
+static stbi__uint16 *stbi__convert_format16(stbi__uint16 *data, int img_n, int req_comp, unsigned int x, unsigned int y)
+{
+ int i,j;
+ stbi__uint16 *good;
+
+ if (req_comp == img_n) return data;
+ STBI_ASSERT(req_comp >= 1 && req_comp <= 4);
+
+ good = (stbi__uint16 *) stbi__malloc(req_comp * x * y * 2);
+ if (good == NULL) {
+ STBI_FREE(data);
+ return (stbi__uint16 *) stbi__errpuc("outofmem", "Out of memory");
+ }
+
+ for (j=0; j < (int) y; ++j) {
+ stbi__uint16 *src = data + j * x * img_n ;
+ stbi__uint16 *dest = good + j * x * req_comp;
+
+ #define STBI__COMBO(a,b) ((a)*8+(b))
+ #define STBI__CASE(a,b) case STBI__COMBO(a,b): for(i=x-1; i >= 0; --i, src += a, dest += b)
+ // convert source image with img_n components to one with req_comp components;
+ // avoid switch per pixel, so use switch per scanline and massive macros
+ switch (STBI__COMBO(img_n, req_comp)) {
+ STBI__CASE(1,2) { dest[0]=src[0], dest[1]=0xffff; } break;
+ STBI__CASE(1,3) { dest[0]=dest[1]=dest[2]=src[0]; } break;
+ STBI__CASE(1,4) { dest[0]=dest[1]=dest[2]=src[0], dest[3]=0xffff; } break;
+ STBI__CASE(2,1) { dest[0]=src[0]; } break;
+ STBI__CASE(2,3) { dest[0]=dest[1]=dest[2]=src[0]; } break;
+ STBI__CASE(2,4) { dest[0]=dest[1]=dest[2]=src[0], dest[3]=src[1]; } break;
+ STBI__CASE(3,4) { dest[0]=src[0],dest[1]=src[1],dest[2]=src[2],dest[3]=0xffff; } break;
+ STBI__CASE(3,1) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]); } break;
+ STBI__CASE(3,2) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]), dest[1] = 0xffff; } break;
+ STBI__CASE(4,1) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]); } break;
+ STBI__CASE(4,2) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]), dest[1] = src[3]; } break;
+ STBI__CASE(4,3) { dest[0]=src[0],dest[1]=src[1],dest[2]=src[2]; } break;
+ default: STBI_ASSERT(0);
+ }
+ #undef STBI__CASE
+ }
+
+ STBI_FREE(data);
+ return good;
+}
+
+#ifndef STBI_NO_LINEAR
+static float *stbi__ldr_to_hdr(stbi_uc *data, int x, int y, int comp)
+{
+ int i,k,n;
+ float *output;
+ if (!data) return NULL;
+ output = (float *) stbi__malloc_mad4(x, y, comp, sizeof(float), 0);
+ if (output == NULL) { STBI_FREE(data); return stbi__errpf("outofmem", "Out of memory"); }
+ // compute number of non-alpha components
+ if (comp & 1) n = comp; else n = comp-1;
+ for (i=0; i < x*y; ++i) {
+ for (k=0; k < n; ++k) {
+ output[i*comp + k] = (float) (pow(data[i*comp+k]/255.0f, stbi__l2h_gamma) * stbi__l2h_scale);
+ }
+ if (k < comp) output[i*comp + k] = data[i*comp+k]/255.0f;
+ }
+ STBI_FREE(data);
+ return output;
+}
+#endif
+
+#ifndef STBI_NO_HDR
+#define stbi__float2int(x) ((int) (x))
+static stbi_uc *stbi__hdr_to_ldr(float *data, int x, int y, int comp)
+{
+ int i,k,n;
+ stbi_uc *output;
+ if (!data) return NULL;
+ output = (stbi_uc *) stbi__malloc_mad3(x, y, comp, 0);
+ if (output == NULL) { STBI_FREE(data); return stbi__errpuc("outofmem", "Out of memory"); }
+ // compute number of non-alpha components
+ if (comp & 1) n = comp; else n = comp-1;
+ for (i=0; i < x*y; ++i) {
+ for (k=0; k < n; ++k) {
+ float z = (float) pow(data[i*comp+k]*stbi__h2l_scale_i, stbi__h2l_gamma_i) * 255 + 0.5f;
+ if (z < 0) z = 0;
+ if (z > 255) z = 255;
+ output[i*comp + k] = (stbi_uc) stbi__float2int(z);
+ }
+ if (k < comp) {
+ float z = data[i*comp+k] * 255 + 0.5f;
+ if (z < 0) z = 0;
+ if (z > 255) z = 255;
+ output[i*comp + k] = (stbi_uc) stbi__float2int(z);
+ }
+ }
+ STBI_FREE(data);
+ return output;
+}
+#endif
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// "baseline" JPEG/JFIF decoder
+//
+// simple implementation
+// - doesn't support delayed output of y-dimension
+// - simple interface (only one output format: 8-bit interleaved RGB)
+// - doesn't try to recover corrupt jpegs
+// - doesn't allow partial loading, loading multiple at once
+// - still fast on x86 (copying globals into locals doesn't help x86)
+// - allocates lots of intermediate memory (full size of all components)
+// - non-interleaved case requires this anyway
+// - allows good upsampling (see next)
+// high-quality
+// - upsampled channels are bilinearly interpolated, even across blocks
+// - quality integer IDCT derived from IJG's 'slow'
+// performance
+// - fast huffman; reasonable integer IDCT
+// - some SIMD kernels for common paths on targets with SSE2/NEON
+// - uses a lot of intermediate memory, could cache poorly
+
+#ifndef STBI_NO_JPEG
+
+// huffman decoding acceleration
+#define FAST_BITS 9 // larger handles more cases; smaller stomps less cache
+
+typedef struct
+{
+ stbi_uc fast[1 << FAST_BITS];
+ // weirdly, repacking this into AoS is a 10% speed loss, instead of a win
+ stbi__uint16 code[256];
+ stbi_uc values[256];
+ stbi_uc size[257];
+ unsigned int maxcode[18];
+ int delta[17]; // old 'firstsymbol' - old 'firstcode'
+} stbi__huffman;
+
+typedef struct
+{
+ stbi__context *s;
+ stbi__huffman huff_dc[4];
+ stbi__huffman huff_ac[4];
+ stbi__uint16 dequant[4][64];
+ stbi__int16 fast_ac[4][1 << FAST_BITS];
+
+// sizes for components, interleaved MCUs
+ int img_h_max, img_v_max;
+ int img_mcu_x, img_mcu_y;
+ int img_mcu_w, img_mcu_h;
+
+// definition of jpeg image component
+ struct
+ {
+ int id;
+ int h,v;
+ int tq;
+ int hd,ha;
+ int dc_pred;
+
+ int x,y,w2,h2;
+ stbi_uc *data;
+ void *raw_data, *raw_coeff;
+ stbi_uc *linebuf;
+ short *coeff; // progressive only
+ int coeff_w, coeff_h; // number of 8x8 coefficient blocks
+ } img_comp[4];
+
+ stbi__uint32 code_buffer; // jpeg entropy-coded buffer
+ int code_bits; // number of valid bits
+ unsigned char marker; // marker seen while filling entropy buffer
+ int nomore; // flag if we saw a marker so must stop
+
+ int progressive;
+ int spec_start;
+ int spec_end;
+ int succ_high;
+ int succ_low;
+ int eob_run;
+ int jfif;
+ int app14_color_transform; // Adobe APP14 tag
+ int rgb;
+
+ int scan_n, order[4];
+ int restart_interval, todo;
+
+// kernels
+ void (*idct_block_kernel)(stbi_uc *out, int out_stride, short data[64]);
+ void (*YCbCr_to_RGB_kernel)(stbi_uc *out, const stbi_uc *y, const stbi_uc *pcb, const stbi_uc *pcr, int count, int step);
+ stbi_uc *(*resample_row_hv_2_kernel)(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs);
+} stbi__jpeg;
+
+static int stbi__build_huffman(stbi__huffman *h, int *count)
+{
+ int i,j,k=0,code;
+ // build size list for each symbol (from JPEG spec)
+ for (i=0; i < 16; ++i)
+ for (j=0; j < count[i]; ++j)
+ h->size[k++] = (stbi_uc) (i+1);
+ h->size[k] = 0;
+
+ // compute actual symbols (from jpeg spec)
+ code = 0;
+ k = 0;
+ for(j=1; j <= 16; ++j) {
+ // compute delta to add to code to compute symbol id
+ h->delta[j] = k - code;
+ if (h->size[k] == j) {
+ while (h->size[k] == j)
+ h->code[k++] = (stbi__uint16) (code++);
+ if (code-1 >= (1 << j)) return stbi__err("bad code lengths","Corrupt JPEG");
+ }
+ // compute largest code + 1 for this size, preshifted as needed later
+ h->maxcode[j] = code << (16-j);
+ code <<= 1;
+ }
+ h->maxcode[j] = 0xffffffff;
+
+ // build non-spec acceleration table; 255 is flag for not-accelerated
+ memset(h->fast, 255, 1 << FAST_BITS);
+ for (i=0; i < k; ++i) {
+ int s = h->size[i];
+ if (s <= FAST_BITS) {
+ int c = h->code[i] << (FAST_BITS-s);
+ int m = 1 << (FAST_BITS-s);
+ for (j=0; j < m; ++j) {
+ h->fast[c+j] = (stbi_uc) i;
+ }
+ }
+ }
+ return 1;
+}
+
+// build a table that decodes both magnitude and value of small ACs in
+// one go.
+static void stbi__build_fast_ac(stbi__int16 *fast_ac, stbi__huffman *h)
+{
+ int i;
+ for (i=0; i < (1 << FAST_BITS); ++i) {
+ stbi_uc fast = h->fast[i];
+ fast_ac[i] = 0;
+ if (fast < 255) {
+ int rs = h->values[fast];
+ int run = (rs >> 4) & 15;
+ int magbits = rs & 15;
+ int len = h->size[fast];
+
+ if (magbits && len + magbits <= FAST_BITS) {
+ // magnitude code followed by receive_extend code
+ int k = ((i << len) & ((1 << FAST_BITS) - 1)) >> (FAST_BITS - magbits);
+ int m = 1 << (magbits - 1);
+ if (k < m) k += (~0U << magbits) + 1;
+ // if the result is small enough, we can fit it in fast_ac table
+ if (k >= -128 && k <= 127)
+ fast_ac[i] = (stbi__int16) ((k << 8) + (run << 4) + (len + magbits));
+ }
+ }
+ }
+}
+
+static void stbi__grow_buffer_unsafe(stbi__jpeg *j)
+{
+ do {
+ int b = j->nomore ? 0 : stbi__get8(j->s);
+ if (b == 0xff) {
+ int c = stbi__get8(j->s);
+ while (c == 0xff) c = stbi__get8(j->s); // consume fill bytes
+ if (c != 0) {
+ j->marker = (unsigned char) c;
+ j->nomore = 1;
+ return;
+ }
+ }
+ j->code_buffer |= b << (24 - j->code_bits);
+ j->code_bits += 8;
+ } while (j->code_bits <= 24);
+}
+
+// (1 << n) - 1
+static stbi__uint32 stbi__bmask[17]={0,1,3,7,15,31,63,127,255,511,1023,2047,4095,8191,16383,32767,65535};
+
+// decode a jpeg huffman value from the bitstream
+stbi_inline static int stbi__jpeg_huff_decode(stbi__jpeg *j, stbi__huffman *h)
+{
+ unsigned int temp;
+ int c,k;
+
+ if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
+
+ // look at the top FAST_BITS and determine what symbol ID it is,
+ // if the code is <= FAST_BITS
+ c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1);
+ k = h->fast[c];
+ if (k < 255) {
+ int s = h->size[k];
+ if (s > j->code_bits)
+ return -1;
+ j->code_buffer <<= s;
+ j->code_bits -= s;
+ return h->values[k];
+ }
+
+ // naive test is to shift the code_buffer down so k bits are
+ // valid, then test against maxcode. To speed this up, we've
+ // preshifted maxcode left so that it has (16-k) 0s at the
+ // end; in other words, regardless of the number of bits, it
+ // wants to be compared against something shifted to have 16;
+ // that way we don't need to shift inside the loop.
+ temp = j->code_buffer >> 16;
+ for (k=FAST_BITS+1 ; ; ++k)
+ if (temp < h->maxcode[k])
+ break;
+ if (k == 17) {
+ // error! code not found
+ j->code_bits -= 16;
+ return -1;
+ }
+
+ if (k > j->code_bits)
+ return -1;
+
+ // convert the huffman code to the symbol id
+ c = ((j->code_buffer >> (32 - k)) & stbi__bmask[k]) + h->delta[k];
+ STBI_ASSERT((((j->code_buffer) >> (32 - h->size[c])) & stbi__bmask[h->size[c]]) == h->code[c]);
+
+ // convert the id to a symbol
+ j->code_bits -= k;
+ j->code_buffer <<= k;
+ return h->values[c];
+}
+
+// bias[n] = (-1<<n) + 1
+static int const stbi__jbias[16] = {0,-1,-3,-7,-15,-31,-63,-127,-255,-511,-1023,-2047,-4095,-8191,-16383,-32767};
+
+// combined JPEG 'receive' and JPEG 'extend', since baseline
+// always extends everything it receives.
+stbi_inline static int stbi__extend_receive(stbi__jpeg *j, int n)
+{
+ unsigned int k;
+ int sgn;
+ if (j->code_bits < n) stbi__grow_buffer_unsafe(j);
+
+ sgn = (stbi__int32)j->code_buffer >> 31; // sign bit is always in MSB
+ k = stbi_lrot(j->code_buffer, n);
+ STBI_ASSERT(n >= 0 && n < (int) (sizeof(stbi__bmask)/sizeof(*stbi__bmask)));
+ j->code_buffer = k & ~stbi__bmask[n];
+ k &= stbi__bmask[n];
+ j->code_bits -= n;
+ return k + (stbi__jbias[n] & ~sgn);
+}
+
+// get some unsigned bits
+stbi_inline static int stbi__jpeg_get_bits(stbi__jpeg *j, int n)
+{
+ unsigned int k;
+ if (j->code_bits < n) stbi__grow_buffer_unsafe(j);
+ k = stbi_lrot(j->code_buffer, n);
+ j->code_buffer = k & ~stbi__bmask[n];
+ k &= stbi__bmask[n];
+ j->code_bits -= n;
+ return k;
+}
+
+stbi_inline static int stbi__jpeg_get_bit(stbi__jpeg *j)
+{
+ unsigned int k;
+ if (j->code_bits < 1) stbi__grow_buffer_unsafe(j);
+ k = j->code_buffer;
+ j->code_buffer <<= 1;
+ --j->code_bits;
+ return k & 0x80000000;
+}
+
+// given a value that's at position X in the zigzag stream,
+// where does it appear in the 8x8 matrix coded as row-major?
+static stbi_uc stbi__jpeg_dezigzag[64+15] =
+{
+ 0, 1, 8, 16, 9, 2, 3, 10,
+ 17, 24, 32, 25, 18, 11, 4, 5,
+ 12, 19, 26, 33, 40, 48, 41, 34,
+ 27, 20, 13, 6, 7, 14, 21, 28,
+ 35, 42, 49, 56, 57, 50, 43, 36,
+ 29, 22, 15, 23, 30, 37, 44, 51,
+ 58, 59, 52, 45, 38, 31, 39, 46,
+ 53, 60, 61, 54, 47, 55, 62, 63,
+ // let corrupt input sample past end
+ 63, 63, 63, 63, 63, 63, 63, 63,
+ 63, 63, 63, 63, 63, 63, 63
+};
+
+// decode one 64-entry block--
+static int stbi__jpeg_decode_block(stbi__jpeg *j, short data[64], stbi__huffman *hdc, stbi__huffman *hac, stbi__int16 *fac, int b, stbi__uint16 *dequant)
+{
+ int diff,dc,k;
+ int t;
+
+ if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
+ t = stbi__jpeg_huff_decode(j, hdc);
+ if (t < 0) return stbi__err("bad huffman code","Corrupt JPEG");
+
+ // 0 all the ac values now so we can do it 32-bits at a time
+ memset(data,0,64*sizeof(data[0]));
+
+ diff = t ? stbi__extend_receive(j, t) : 0;
+ dc = j->img_comp[b].dc_pred + diff;
+ j->img_comp[b].dc_pred = dc;
+ data[0] = (short) (dc * dequant[0]);
+
+ // decode AC components, see JPEG spec
+ k = 1;
+ do {
+ unsigned int zig;
+ int c,r,s;
+ if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
+ c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1);
+ r = fac[c];
+ if (r) { // fast-AC path
+ k += (r >> 4) & 15; // run
+ s = r & 15; // combined length
+ j->code_buffer <<= s;
+ j->code_bits -= s;
+ // decode into unzigzag'd location
+ zig = stbi__jpeg_dezigzag[k++];
+ data[zig] = (short) ((r >> 8) * dequant[zig]);
+ } else {
+ int rs = stbi__jpeg_huff_decode(j, hac);
+ if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG");
+ s = rs & 15;
+ r = rs >> 4;
+ if (s == 0) {
+ if (rs != 0xf0) break; // end block
+ k += 16;
+ } else {
+ k += r;
+ // decode into unzigzag'd location
+ zig = stbi__jpeg_dezigzag[k++];
+ data[zig] = (short) (stbi__extend_receive(j,s) * dequant[zig]);
+ }
+ }
+ } while (k < 64);
+ return 1;
+}
+
+static int stbi__jpeg_decode_block_prog_dc(stbi__jpeg *j, short data[64], stbi__huffman *hdc, int b)
+{
+ int diff,dc;
+ int t;
+ if (j->spec_end != 0) return stbi__err("can't merge dc and ac", "Corrupt JPEG");
+
+ if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
+
+ if (j->succ_high == 0) {
+ // first scan for DC coefficient, must be first
+ memset(data,0,64*sizeof(data[0])); // 0 all the ac values now
+ t = stbi__jpeg_huff_decode(j, hdc);
+ diff = t ? stbi__extend_receive(j, t) : 0;
+
+ dc = j->img_comp[b].dc_pred + diff;
+ j->img_comp[b].dc_pred = dc;
+ data[0] = (short) (dc << j->succ_low);
+ } else {
+ // refinement scan for DC coefficient
+ if (stbi__jpeg_get_bit(j))
+ data[0] += (short) (1 << j->succ_low);
+ }
+ return 1;
+}
+
+// @OPTIMIZE: store non-zigzagged during the decode passes,
+// and only de-zigzag when dequantizing
+static int stbi__jpeg_decode_block_prog_ac(stbi__jpeg *j, short data[64], stbi__huffman *hac, stbi__int16 *fac)
+{
+ int k;
+ if (j->spec_start == 0) return stbi__err("can't merge dc and ac", "Corrupt JPEG");
+
+ if (j->succ_high == 0) {
+ int shift = j->succ_low;
+
+ if (j->eob_run) {
+ --j->eob_run;
+ return 1;
+ }
+
+ k = j->spec_start;
+ do {
+ unsigned int zig;
+ int c,r,s;
+ if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
+ c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1);
+ r = fac[c];
+ if (r) { // fast-AC path
+ k += (r >> 4) & 15; // run
+ s = r & 15; // combined length
+ j->code_buffer <<= s;
+ j->code_bits -= s;
+ zig = stbi__jpeg_dezigzag[k++];
+ data[zig] = (short) ((r >> 8) << shift);
+ } else {
+ int rs = stbi__jpeg_huff_decode(j, hac);
+ if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG");
+ s = rs & 15;
+ r = rs >> 4;
+ if (s == 0) {
+ if (r < 15) {
+ j->eob_run = (1 << r);
+ if (r)
+ j->eob_run += stbi__jpeg_get_bits(j, r);
+ --j->eob_run;
+ break;
+ }
+ k += 16;
+ } else {
+ k += r;
+ zig = stbi__jpeg_dezigzag[k++];
+ data[zig] = (short) (stbi__extend_receive(j,s) << shift);
+ }
+ }
+ } while (k <= j->spec_end);
+ } else {
+ // refinement scan for these AC coefficients
+
+ short bit = (short) (1 << j->succ_low);
+
+ if (j->eob_run) {
+ --j->eob_run;
+ for (k = j->spec_start; k <= j->spec_end; ++k) {
+ short *p = &data[stbi__jpeg_dezigzag[k]];
+ if (*p != 0)
+ if (stbi__jpeg_get_bit(j))
+ if ((*p & bit)==0) {
+ if (*p > 0)
+ *p += bit;
+ else
+ *p -= bit;
+ }
+ }
+ } else {
+ k = j->spec_start;
+ do {
+ int r,s;
+ int rs = stbi__jpeg_huff_decode(j, hac); // @OPTIMIZE see if we can use the fast path here, advance-by-r is so slow, eh
+ if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG");
+ s = rs & 15;
+ r = rs >> 4;
+ if (s == 0) {
+ if (r < 15) {
+ j->eob_run = (1 << r) - 1;
+ if (r)
+ j->eob_run += stbi__jpeg_get_bits(j, r);
+ r = 64; // force end of block
+ } else {
+ // r=15 s=0 should write 16 0s, so we just do
+ // a run of 15 0s and then write s (which is 0),
+ // so we don't have to do anything special here
+ }
+ } else {
+ if (s != 1) return stbi__err("bad huffman code", "Corrupt JPEG");
+ // sign bit
+ if (stbi__jpeg_get_bit(j))
+ s = bit;
+ else
+ s = -bit;
+ }
+
+ // advance by r
+ while (k <= j->spec_end) {
+ short *p = &data[stbi__jpeg_dezigzag[k++]];
+ if (*p != 0) {
+ if (stbi__jpeg_get_bit(j))
+ if ((*p & bit)==0) {
+ if (*p > 0)
+ *p += bit;
+ else
+ *p -= bit;
+ }
+ } else {
+ if (r == 0) {
+ *p = (short) s;
+ break;
+ }
+ --r;
+ }
+ }
+ } while (k <= j->spec_end);
+ }
+ }
+ return 1;
+}
+
+// take a -128..127 value and stbi__clamp it and convert to 0..255
+stbi_inline static stbi_uc stbi__clamp(int x)
+{
+ // trick to use a single test to catch both cases
+ if ((unsigned int) x > 255) {
+ if (x < 0) return 0;
+ if (x > 255) return 255;
+ }
+ return (stbi_uc) x;
+}
+
+#define stbi__f2f(x) ((int) (((x) * 4096 + 0.5)))
+#define stbi__fsh(x) ((x) << 12)
+
+// derived from jidctint -- DCT_ISLOW
+#define STBI__IDCT_1D(s0,s1,s2,s3,s4,s5,s6,s7) \
+ int t0,t1,t2,t3,p1,p2,p3,p4,p5,x0,x1,x2,x3; \
+ p2 = s2; \
+ p3 = s6; \
+ p1 = (p2+p3) * stbi__f2f(0.5411961f); \
+ t2 = p1 + p3*stbi__f2f(-1.847759065f); \
+ t3 = p1 + p2*stbi__f2f( 0.765366865f); \
+ p2 = s0; \
+ p3 = s4; \
+ t0 = stbi__fsh(p2+p3); \
+ t1 = stbi__fsh(p2-p3); \
+ x0 = t0+t3; \
+ x3 = t0-t3; \
+ x1 = t1+t2; \
+ x2 = t1-t2; \
+ t0 = s7; \
+ t1 = s5; \
+ t2 = s3; \
+ t3 = s1; \
+ p3 = t0+t2; \
+ p4 = t1+t3; \
+ p1 = t0+t3; \
+ p2 = t1+t2; \
+ p5 = (p3+p4)*stbi__f2f( 1.175875602f); \
+ t0 = t0*stbi__f2f( 0.298631336f); \
+ t1 = t1*stbi__f2f( 2.053119869f); \
+ t2 = t2*stbi__f2f( 3.072711026f); \
+ t3 = t3*stbi__f2f( 1.501321110f); \
+ p1 = p5 + p1*stbi__f2f(-0.899976223f); \
+ p2 = p5 + p2*stbi__f2f(-2.562915447f); \
+ p3 = p3*stbi__f2f(-1.961570560f); \
+ p4 = p4*stbi__f2f(-0.390180644f); \
+ t3 += p1+p4; \
+ t2 += p2+p3; \
+ t1 += p2+p4; \
+ t0 += p1+p3;
+
+static void stbi__idct_block(stbi_uc *out, int out_stride, short data[64])
+{
+ int i,val[64],*v=val;
+ stbi_uc *o;
+ short *d = data;
+
+ // columns
+ for (i=0; i < 8; ++i,++d, ++v) {
+ // if all zeroes, shortcut -- this avoids dequantizing 0s and IDCTing
+ if (d[ 8]==0 && d[16]==0 && d[24]==0 && d[32]==0
+ && d[40]==0 && d[48]==0 && d[56]==0) {
+ // no shortcut 0 seconds
+ // (1|2|3|4|5|6|7)==0 0 seconds
+ // all separate -0.047 seconds
+ // 1 && 2|3 && 4|5 && 6|7: -0.047 seconds
+ int dcterm = d[0] << 2;
+ v[0] = v[8] = v[16] = v[24] = v[32] = v[40] = v[48] = v[56] = dcterm;
+ } else {
+ STBI__IDCT_1D(d[ 0],d[ 8],d[16],d[24],d[32],d[40],d[48],d[56])
+ // constants scaled things up by 1<<12; let's bring them back
+ // down, but keep 2 extra bits of precision
+ x0 += 512; x1 += 512; x2 += 512; x3 += 512;
+ v[ 0] = (x0+t3) >> 10;
+ v[56] = (x0-t3) >> 10;
+ v[ 8] = (x1+t2) >> 10;
+ v[48] = (x1-t2) >> 10;
+ v[16] = (x2+t1) >> 10;
+ v[40] = (x2-t1) >> 10;
+ v[24] = (x3+t0) >> 10;
+ v[32] = (x3-t0) >> 10;
+ }
+ }
+
+ for (i=0, v=val, o=out; i < 8; ++i,v+=8,o+=out_stride) {
+ // no fast case since the first 1D IDCT spread components out
+ STBI__IDCT_1D(v[0],v[1],v[2],v[3],v[4],v[5],v[6],v[7])
+ // constants scaled things up by 1<<12, plus we had 1<<2 from first
+ // loop, plus horizontal and vertical each scale by sqrt(8) so together
+ // we've got an extra 1<<3, so 1<<17 total we need to remove.
+ // so we want to round that, which means adding 0.5 * 1<<17,
+ // aka 65536. Also, we'll end up with -128 to 127 that we want
+ // to encode as 0..255 by adding 128, so we'll add that before the shift
+ x0 += 65536 + (128<<17);
+ x1 += 65536 + (128<<17);
+ x2 += 65536 + (128<<17);
+ x3 += 65536 + (128<<17);
+ // tried computing the shifts into temps, or'ing the temps to see
+ // if any were out of range, but that was slower
+ o[0] = stbi__clamp((x0+t3) >> 17);
+ o[7] = stbi__clamp((x0-t3) >> 17);
+ o[1] = stbi__clamp((x1+t2) >> 17);
+ o[6] = stbi__clamp((x1-t2) >> 17);
+ o[2] = stbi__clamp((x2+t1) >> 17);
+ o[5] = stbi__clamp((x2-t1) >> 17);
+ o[3] = stbi__clamp((x3+t0) >> 17);
+ o[4] = stbi__clamp((x3-t0) >> 17);
+ }
+}
+
+#ifdef STBI_SSE2
+// sse2 integer IDCT. not the fastest possible implementation but it
+// produces bit-identical results to the generic C version so it's
+// fully "transparent".
+static void stbi__idct_simd(stbi_uc *out, int out_stride, short data[64])
+{
+ // This is constructed to match our regular (generic) integer IDCT exactly.
+ __m128i row0, row1, row2, row3, row4, row5, row6, row7;
+ __m128i tmp;
+
+ // dot product constant: even elems=x, odd elems=y
+ #define dct_const(x,y) _mm_setr_epi16((x),(y),(x),(y),(x),(y),(x),(y))
+
+ // out(0) = c0[even]*x + c0[odd]*y (c0, x, y 16-bit, out 32-bit)
+ // out(1) = c1[even]*x + c1[odd]*y
+ #define dct_rot(out0,out1, x,y,c0,c1) \
+ __m128i c0##lo = _mm_unpacklo_epi16((x),(y)); \
+ __m128i c0##hi = _mm_unpackhi_epi16((x),(y)); \
+ __m128i out0##_l = _mm_madd_epi16(c0##lo, c0); \
+ __m128i out0##_h = _mm_madd_epi16(c0##hi, c0); \
+ __m128i out1##_l = _mm_madd_epi16(c0##lo, c1); \
+ __m128i out1##_h = _mm_madd_epi16(c0##hi, c1)
+
+ // out = in << 12 (in 16-bit, out 32-bit)
+ #define dct_widen(out, in) \
+ __m128i out##_l = _mm_srai_epi32(_mm_unpacklo_epi16(_mm_setzero_si128(), (in)), 4); \
+ __m128i out##_h = _mm_srai_epi32(_mm_unpackhi_epi16(_mm_setzero_si128(), (in)), 4)
+
+ // wide add
+ #define dct_wadd(out, a, b) \
+ __m128i out##_l = _mm_add_epi32(a##_l, b##_l); \
+ __m128i out##_h = _mm_add_epi32(a##_h, b##_h)
+
+ // wide sub
+ #define dct_wsub(out, a, b) \
+ __m128i out##_l = _mm_sub_epi32(a##_l, b##_l); \
+ __m128i out##_h = _mm_sub_epi32(a##_h, b##_h)
+
+ // butterfly a/b, add bias, then shift by "s" and pack
+ #define dct_bfly32o(out0, out1, a,b,bias,s) \
+ { \
+ __m128i abiased_l = _mm_add_epi32(a##_l, bias); \
+ __m128i abiased_h = _mm_add_epi32(a##_h, bias); \
+ dct_wadd(sum, abiased, b); \
+ dct_wsub(dif, abiased, b); \
+ out0 = _mm_packs_epi32(_mm_srai_epi32(sum_l, s), _mm_srai_epi32(sum_h, s)); \
+ out1 = _mm_packs_epi32(_mm_srai_epi32(dif_l, s), _mm_srai_epi32(dif_h, s)); \
+ }
+
+ // 8-bit interleave step (for transposes)
+ #define dct_interleave8(a, b) \
+ tmp = a; \
+ a = _mm_unpacklo_epi8(a, b); \
+ b = _mm_unpackhi_epi8(tmp, b)
+
+ // 16-bit interleave step (for transposes)
+ #define dct_interleave16(a, b) \
+ tmp = a; \
+ a = _mm_unpacklo_epi16(a, b); \
+ b = _mm_unpackhi_epi16(tmp, b)
+
+ #define dct_pass(bias,shift) \
+ { \
+ /* even part */ \
+ dct_rot(t2e,t3e, row2,row6, rot0_0,rot0_1); \
+ __m128i sum04 = _mm_add_epi16(row0, row4); \
+ __m128i dif04 = _mm_sub_epi16(row0, row4); \
+ dct_widen(t0e, sum04); \
+ dct_widen(t1e, dif04); \
+ dct_wadd(x0, t0e, t3e); \
+ dct_wsub(x3, t0e, t3e); \
+ dct_wadd(x1, t1e, t2e); \
+ dct_wsub(x2, t1e, t2e); \
+ /* odd part */ \
+ dct_rot(y0o,y2o, row7,row3, rot2_0,rot2_1); \
+ dct_rot(y1o,y3o, row5,row1, rot3_0,rot3_1); \
+ __m128i sum17 = _mm_add_epi16(row1, row7); \
+ __m128i sum35 = _mm_add_epi16(row3, row5); \
+ dct_rot(y4o,y5o, sum17,sum35, rot1_0,rot1_1); \
+ dct_wadd(x4, y0o, y4o); \
+ dct_wadd(x5, y1o, y5o); \
+ dct_wadd(x6, y2o, y5o); \
+ dct_wadd(x7, y3o, y4o); \
+ dct_bfly32o(row0,row7, x0,x7,bias,shift); \
+ dct_bfly32o(row1,row6, x1,x6,bias,shift); \
+ dct_bfly32o(row2,row5, x2,x5,bias,shift); \
+ dct_bfly32o(row3,row4, x3,x4,bias,shift); \
+ }
+
+ __m128i rot0_0 = dct_const(stbi__f2f(0.5411961f), stbi__f2f(0.5411961f) + stbi__f2f(-1.847759065f));
+ __m128i rot0_1 = dct_const(stbi__f2f(0.5411961f) + stbi__f2f( 0.765366865f), stbi__f2f(0.5411961f));
+ __m128i rot1_0 = dct_const(stbi__f2f(1.175875602f) + stbi__f2f(-0.899976223f), stbi__f2f(1.175875602f));
+ __m128i rot1_1 = dct_const(stbi__f2f(1.175875602f), stbi__f2f(1.175875602f) + stbi__f2f(-2.562915447f));
+ __m128i rot2_0 = dct_const(stbi__f2f(-1.961570560f) + stbi__f2f( 0.298631336f), stbi__f2f(-1.961570560f));
+ __m128i rot2_1 = dct_const(stbi__f2f(-1.961570560f), stbi__f2f(-1.961570560f) + stbi__f2f( 3.072711026f));
+ __m128i rot3_0 = dct_const(stbi__f2f(-0.390180644f) + stbi__f2f( 2.053119869f), stbi__f2f(-0.390180644f));
+ __m128i rot3_1 = dct_const(stbi__f2f(-0.390180644f), stbi__f2f(-0.390180644f) + stbi__f2f( 1.501321110f));
+
+ // rounding biases in column/row passes, see stbi__idct_block for explanation.
+ __m128i bias_0 = _mm_set1_epi32(512);
+ __m128i bias_1 = _mm_set1_epi32(65536 + (128<<17));
+
+ // load
+ row0 = _mm_load_si128((const __m128i *) (data + 0*8));
+ row1 = _mm_load_si128((const __m128i *) (data + 1*8));
+ row2 = _mm_load_si128((const __m128i *) (data + 2*8));
+ row3 = _mm_load_si128((const __m128i *) (data + 3*8));
+ row4 = _mm_load_si128((const __m128i *) (data + 4*8));
+ row5 = _mm_load_si128((const __m128i *) (data + 5*8));
+ row6 = _mm_load_si128((const __m128i *) (data + 6*8));
+ row7 = _mm_load_si128((const __m128i *) (data + 7*8));
+
+ // column pass
+ dct_pass(bias_0, 10);
+
+ {
+ // 16bit 8x8 transpose pass 1
+ dct_interleave16(row0, row4);
+ dct_interleave16(row1, row5);
+ dct_interleave16(row2, row6);
+ dct_interleave16(row3, row7);
+
+ // transpose pass 2
+ dct_interleave16(row0, row2);
+ dct_interleave16(row1, row3);
+ dct_interleave16(row4, row6);
+ dct_interleave16(row5, row7);
+
+ // transpose pass 3
+ dct_interleave16(row0, row1);
+ dct_interleave16(row2, row3);
+ dct_interleave16(row4, row5);
+ dct_interleave16(row6, row7);
+ }
+
+ // row pass
+ dct_pass(bias_1, 17);
+
+ {
+ // pack
+ __m128i p0 = _mm_packus_epi16(row0, row1); // a0a1a2a3...a7b0b1b2b3...b7
+ __m128i p1 = _mm_packus_epi16(row2, row3);
+ __m128i p2 = _mm_packus_epi16(row4, row5);
+ __m128i p3 = _mm_packus_epi16(row6, row7);
+
+ // 8bit 8x8 transpose pass 1
+ dct_interleave8(p0, p2); // a0e0a1e1...
+ dct_interleave8(p1, p3); // c0g0c1g1...
+
+ // transpose pass 2
+ dct_interleave8(p0, p1); // a0c0e0g0...
+ dct_interleave8(p2, p3); // b0d0f0h0...
+
+ // transpose pass 3
+ dct_interleave8(p0, p2); // a0b0c0d0...
+ dct_interleave8(p1, p3); // a4b4c4d4...
+
+ // store
+ _mm_storel_epi64((__m128i *) out, p0); out += out_stride;
+ _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p0, 0x4e)); out += out_stride;
+ _mm_storel_epi64((__m128i *) out, p2); out += out_stride;
+ _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p2, 0x4e)); out += out_stride;
+ _mm_storel_epi64((__m128i *) out, p1); out += out_stride;
+ _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p1, 0x4e)); out += out_stride;
+ _mm_storel_epi64((__m128i *) out, p3); out += out_stride;
+ _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p3, 0x4e));
+ }
+
+#undef dct_const
+#undef dct_rot
+#undef dct_widen
+#undef dct_wadd
+#undef dct_wsub
+#undef dct_bfly32o
+#undef dct_interleave8
+#undef dct_interleave16
+#undef dct_pass
+}
+
+#endif // STBI_SSE2
+
+#ifdef STBI_NEON
+
+// NEON integer IDCT. should produce bit-identical
+// results to the generic C version.
+static void stbi__idct_simd(stbi_uc *out, int out_stride, short data[64])
+{
+ int16x8_t row0, row1, row2, row3, row4, row5, row6, row7;
+
+ int16x4_t rot0_0 = vdup_n_s16(stbi__f2f(0.5411961f));
+ int16x4_t rot0_1 = vdup_n_s16(stbi__f2f(-1.847759065f));
+ int16x4_t rot0_2 = vdup_n_s16(stbi__f2f( 0.765366865f));
+ int16x4_t rot1_0 = vdup_n_s16(stbi__f2f( 1.175875602f));
+ int16x4_t rot1_1 = vdup_n_s16(stbi__f2f(-0.899976223f));
+ int16x4_t rot1_2 = vdup_n_s16(stbi__f2f(-2.562915447f));
+ int16x4_t rot2_0 = vdup_n_s16(stbi__f2f(-1.961570560f));
+ int16x4_t rot2_1 = vdup_n_s16(stbi__f2f(-0.390180644f));
+ int16x4_t rot3_0 = vdup_n_s16(stbi__f2f( 0.298631336f));
+ int16x4_t rot3_1 = vdup_n_s16(stbi__f2f( 2.053119869f));
+ int16x4_t rot3_2 = vdup_n_s16(stbi__f2f( 3.072711026f));
+ int16x4_t rot3_3 = vdup_n_s16(stbi__f2f( 1.501321110f));
+
+#define dct_long_mul(out, inq, coeff) \
+ int32x4_t out##_l = vmull_s16(vget_low_s16(inq), coeff); \
+ int32x4_t out##_h = vmull_s16(vget_high_s16(inq), coeff)
+
+#define dct_long_mac(out, acc, inq, coeff) \
+ int32x4_t out##_l = vmlal_s16(acc##_l, vget_low_s16(inq), coeff); \
+ int32x4_t out##_h = vmlal_s16(acc##_h, vget_high_s16(inq), coeff)
+
+#define dct_widen(out, inq) \
+ int32x4_t out##_l = vshll_n_s16(vget_low_s16(inq), 12); \
+ int32x4_t out##_h = vshll_n_s16(vget_high_s16(inq), 12)
+
+// wide add
+#define dct_wadd(out, a, b) \
+ int32x4_t out##_l = vaddq_s32(a##_l, b##_l); \
+ int32x4_t out##_h = vaddq_s32(a##_h, b##_h)
+
+// wide sub
+#define dct_wsub(out, a, b) \
+ int32x4_t out##_l = vsubq_s32(a##_l, b##_l); \
+ int32x4_t out##_h = vsubq_s32(a##_h, b##_h)
+
+// butterfly a/b, then shift using "shiftop" by "s" and pack
+#define dct_bfly32o(out0,out1, a,b,shiftop,s) \
+ { \
+ dct_wadd(sum, a, b); \
+ dct_wsub(dif, a, b); \
+ out0 = vcombine_s16(shiftop(sum_l, s), shiftop(sum_h, s)); \
+ out1 = vcombine_s16(shiftop(dif_l, s), shiftop(dif_h, s)); \
+ }
+
+#define dct_pass(shiftop, shift) \
+ { \
+ /* even part */ \
+ int16x8_t sum26 = vaddq_s16(row2, row6); \
+ dct_long_mul(p1e, sum26, rot0_0); \
+ dct_long_mac(t2e, p1e, row6, rot0_1); \
+ dct_long_mac(t3e, p1e, row2, rot0_2); \
+ int16x8_t sum04 = vaddq_s16(row0, row4); \
+ int16x8_t dif04 = vsubq_s16(row0, row4); \
+ dct_widen(t0e, sum04); \
+ dct_widen(t1e, dif04); \
+ dct_wadd(x0, t0e, t3e); \
+ dct_wsub(x3, t0e, t3e); \
+ dct_wadd(x1, t1e, t2e); \
+ dct_wsub(x2, t1e, t2e); \
+ /* odd part */ \
+ int16x8_t sum15 = vaddq_s16(row1, row5); \
+ int16x8_t sum17 = vaddq_s16(row1, row7); \
+ int16x8_t sum35 = vaddq_s16(row3, row5); \
+ int16x8_t sum37 = vaddq_s16(row3, row7); \
+ int16x8_t sumodd = vaddq_s16(sum17, sum35); \
+ dct_long_mul(p5o, sumodd, rot1_0); \
+ dct_long_mac(p1o, p5o, sum17, rot1_1); \
+ dct_long_mac(p2o, p5o, sum35, rot1_2); \
+ dct_long_mul(p3o, sum37, rot2_0); \
+ dct_long_mul(p4o, sum15, rot2_1); \
+ dct_wadd(sump13o, p1o, p3o); \
+ dct_wadd(sump24o, p2o, p4o); \
+ dct_wadd(sump23o, p2o, p3o); \
+ dct_wadd(sump14o, p1o, p4o); \
+ dct_long_mac(x4, sump13o, row7, rot3_0); \
+ dct_long_mac(x5, sump24o, row5, rot3_1); \
+ dct_long_mac(x6, sump23o, row3, rot3_2); \
+ dct_long_mac(x7, sump14o, row1, rot3_3); \
+ dct_bfly32o(row0,row7, x0,x7,shiftop,shift); \
+ dct_bfly32o(row1,row6, x1,x6,shiftop,shift); \
+ dct_bfly32o(row2,row5, x2,x5,shiftop,shift); \
+ dct_bfly32o(row3,row4, x3,x4,shiftop,shift); \
+ }
+
+ // load
+ row0 = vld1q_s16(data + 0*8);
+ row1 = vld1q_s16(data + 1*8);
+ row2 = vld1q_s16(data + 2*8);
+ row3 = vld1q_s16(data + 3*8);
+ row4 = vld1q_s16(data + 4*8);
+ row5 = vld1q_s16(data + 5*8);
+ row6 = vld1q_s16(data + 6*8);
+ row7 = vld1q_s16(data + 7*8);
+
+ // add DC bias
+ row0 = vaddq_s16(row0, vsetq_lane_s16(1024, vdupq_n_s16(0), 0));
+
+ // column pass
+ dct_pass(vrshrn_n_s32, 10);
+
+ // 16bit 8x8 transpose
+ {
+// these three map to a single VTRN.16, VTRN.32, and VSWP, respectively.
+// whether compilers actually get this is another story, sadly.
+#define dct_trn16(x, y) { int16x8x2_t t = vtrnq_s16(x, y); x = t.val[0]; y = t.val[1]; }
+#define dct_trn32(x, y) { int32x4x2_t t = vtrnq_s32(vreinterpretq_s32_s16(x), vreinterpretq_s32_s16(y)); x = vreinterpretq_s16_s32(t.val[0]); y = vreinterpretq_s16_s32(t.val[1]); }
+#define dct_trn64(x, y) { int16x8_t x0 = x; int16x8_t y0 = y; x = vcombine_s16(vget_low_s16(x0), vget_low_s16(y0)); y = vcombine_s16(vget_high_s16(x0), vget_high_s16(y0)); }
+
+ // pass 1
+ dct_trn16(row0, row1); // a0b0a2b2a4b4a6b6
+ dct_trn16(row2, row3);
+ dct_trn16(row4, row5);
+ dct_trn16(row6, row7);
+
+ // pass 2
+ dct_trn32(row0, row2); // a0b0c0d0a4b4c4d4
+ dct_trn32(row1, row3);
+ dct_trn32(row4, row6);
+ dct_trn32(row5, row7);
+
+ // pass 3
+ dct_trn64(row0, row4); // a0b0c0d0e0f0g0h0
+ dct_trn64(row1, row5);
+ dct_trn64(row2, row6);
+ dct_trn64(row3, row7);
+
+#undef dct_trn16
+#undef dct_trn32
+#undef dct_trn64
+ }
+
+ // row pass
+ // vrshrn_n_s32 only supports shifts up to 16, we need
+ // 17. so do a non-rounding shift of 16 first then follow
+ // up with a rounding shift by 1.
+ dct_pass(vshrn_n_s32, 16);
+
+ {
+ // pack and round
+ uint8x8_t p0 = vqrshrun_n_s16(row0, 1);
+ uint8x8_t p1 = vqrshrun_n_s16(row1, 1);
+ uint8x8_t p2 = vqrshrun_n_s16(row2, 1);
+ uint8x8_t p3 = vqrshrun_n_s16(row3, 1);
+ uint8x8_t p4 = vqrshrun_n_s16(row4, 1);
+ uint8x8_t p5 = vqrshrun_n_s16(row5, 1);
+ uint8x8_t p6 = vqrshrun_n_s16(row6, 1);
+ uint8x8_t p7 = vqrshrun_n_s16(row7, 1);
+
+ // again, these can translate into one instruction, but often don't.
+#define dct_trn8_8(x, y) { uint8x8x2_t t = vtrn_u8(x, y); x = t.val[0]; y = t.val[1]; }
+#define dct_trn8_16(x, y) { uint16x4x2_t t = vtrn_u16(vreinterpret_u16_u8(x), vreinterpret_u16_u8(y)); x = vreinterpret_u8_u16(t.val[0]); y = vreinterpret_u8_u16(t.val[1]); }
+#define dct_trn8_32(x, y) { uint32x2x2_t t = vtrn_u32(vreinterpret_u32_u8(x), vreinterpret_u32_u8(y)); x = vreinterpret_u8_u32(t.val[0]); y = vreinterpret_u8_u32(t.val[1]); }
+
+ // sadly can't use interleaved stores here since we only write
+ // 8 bytes to each scan line!
+
+ // 8x8 8-bit transpose pass 1
+ dct_trn8_8(p0, p1);
+ dct_trn8_8(p2, p3);
+ dct_trn8_8(p4, p5);
+ dct_trn8_8(p6, p7);
+
+ // pass 2
+ dct_trn8_16(p0, p2);
+ dct_trn8_16(p1, p3);
+ dct_trn8_16(p4, p6);
+ dct_trn8_16(p5, p7);
+
+ // pass 3
+ dct_trn8_32(p0, p4);
+ dct_trn8_32(p1, p5);
+ dct_trn8_32(p2, p6);
+ dct_trn8_32(p3, p7);
+
+ // store
+ vst1_u8(out, p0); out += out_stride;
+ vst1_u8(out, p1); out += out_stride;
+ vst1_u8(out, p2); out += out_stride;
+ vst1_u8(out, p3); out += out_stride;
+ vst1_u8(out, p4); out += out_stride;
+ vst1_u8(out, p5); out += out_stride;
+ vst1_u8(out, p6); out += out_stride;
+ vst1_u8(out, p7);
+
+#undef dct_trn8_8
+#undef dct_trn8_16
+#undef dct_trn8_32
+ }
+
+#undef dct_long_mul
+#undef dct_long_mac
+#undef dct_widen
+#undef dct_wadd
+#undef dct_wsub
+#undef dct_bfly32o
+#undef dct_pass
+}
+
+#endif // STBI_NEON
+
+#define STBI__MARKER_none 0xff
+// if there's a pending marker from the entropy stream, return that
+// otherwise, fetch from the stream and get a marker. if there's no
+// marker, return 0xff, which is never a valid marker value
+static stbi_uc stbi__get_marker(stbi__jpeg *j)
+{
+ stbi_uc x;
+ if (j->marker != STBI__MARKER_none) { x = j->marker; j->marker = STBI__MARKER_none; return x; }
+ x = stbi__get8(j->s);
+ if (x != 0xff) return STBI__MARKER_none;
+ while (x == 0xff)
+ x = stbi__get8(j->s); // consume repeated 0xff fill bytes
+ return x;
+}
+
+// in each scan, we'll have scan_n components, and the order
+// of the components is specified by order[]
+#define STBI__RESTART(x) ((x) >= 0xd0 && (x) <= 0xd7)
+
+// after a restart interval, stbi__jpeg_reset the entropy decoder and
+// the dc prediction
+static void stbi__jpeg_reset(stbi__jpeg *j)
+{
+ j->code_bits = 0;
+ j->code_buffer = 0;
+ j->nomore = 0;
+ j->img_comp[0].dc_pred = j->img_comp[1].dc_pred = j->img_comp[2].dc_pred = j->img_comp[3].dc_pred = 0;
+ j->marker = STBI__MARKER_none;
+ j->todo = j->restart_interval ? j->restart_interval : 0x7fffffff;
+ j->eob_run = 0;
+ // no more than 1<<31 MCUs if no restart_interal? that's plenty safe,
+ // since we don't even allow 1<<30 pixels
+}
+
+static int stbi__parse_entropy_coded_data(stbi__jpeg *z)
+{
+ stbi__jpeg_reset(z);
+ if (!z->progressive) {
+ if (z->scan_n == 1) {
+ int i,j;
+ STBI_SIMD_ALIGN(short, data[64]);
+ int n = z->order[0];
+ // non-interleaved data, we just need to process one block at a time,
+ // in trivial scanline order
+ // number of blocks to do just depends on how many actual "pixels" this
+ // component has, independent of interleaved MCU blocking and such
+ int w = (z->img_comp[n].x+7) >> 3;
+ int h = (z->img_comp[n].y+7) >> 3;
+ for (j=0; j < h; ++j) {
+ for (i=0; i < w; ++i) {
+ int ha = z->img_comp[n].ha;
+ if (!stbi__jpeg_decode_block(z, data, z->huff_dc+z->img_comp[n].hd, z->huff_ac+ha, z->fast_ac[ha], n, z->dequant[z->img_comp[n].tq])) return 0;
+ z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*j*8+i*8, z->img_comp[n].w2, data);
+ // every data block is an MCU, so countdown the restart interval
+ if (--z->todo <= 0) {
+ if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
+ // if it's NOT a restart, then just bail, so we get corrupt data
+ // rather than no data
+ if (!STBI__RESTART(z->marker)) return 1;
+ stbi__jpeg_reset(z);
+ }
+ }
+ }
+ return 1;
+ } else { // interleaved
+ int i,j,k,x,y;
+ STBI_SIMD_ALIGN(short, data[64]);
+ for (j=0; j < z->img_mcu_y; ++j) {
+ for (i=0; i < z->img_mcu_x; ++i) {
+ // scan an interleaved mcu... process scan_n components in order
+ for (k=0; k < z->scan_n; ++k) {
+ int n = z->order[k];
+ // scan out an mcu's worth of this component; that's just determined
+ // by the basic H and V specified for the component
+ for (y=0; y < z->img_comp[n].v; ++y) {
+ for (x=0; x < z->img_comp[n].h; ++x) {
+ int x2 = (i*z->img_comp[n].h + x)*8;
+ int y2 = (j*z->img_comp[n].v + y)*8;
+ int ha = z->img_comp[n].ha;
+ if (!stbi__jpeg_decode_block(z, data, z->huff_dc+z->img_comp[n].hd, z->huff_ac+ha, z->fast_ac[ha], n, z->dequant[z->img_comp[n].tq])) return 0;
+ z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*y2+x2, z->img_comp[n].w2, data);
+ }
+ }
+ }
+ // after all interleaved components, that's an interleaved MCU,
+ // so now count down the restart interval
+ if (--z->todo <= 0) {
+ if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
+ if (!STBI__RESTART(z->marker)) return 1;
+ stbi__jpeg_reset(z);
+ }
+ }
+ }
+ return 1;
+ }
+ } else {
+ if (z->scan_n == 1) {
+ int i,j;
+ int n = z->order[0];
+ // non-interleaved data, we just need to process one block at a time,
+ // in trivial scanline order
+ // number of blocks to do just depends on how many actual "pixels" this
+ // component has, independent of interleaved MCU blocking and such
+ int w = (z->img_comp[n].x+7) >> 3;
+ int h = (z->img_comp[n].y+7) >> 3;
+ for (j=0; j < h; ++j) {
+ for (i=0; i < w; ++i) {
+ short *data = z->img_comp[n].coeff + 64 * (i + j * z->img_comp[n].coeff_w);
+ if (z->spec_start == 0) {
+ if (!stbi__jpeg_decode_block_prog_dc(z, data, &z->huff_dc[z->img_comp[n].hd], n))
+ return 0;
+ } else {
+ int ha = z->img_comp[n].ha;
+ if (!stbi__jpeg_decode_block_prog_ac(z, data, &z->huff_ac[ha], z->fast_ac[ha]))
+ return 0;
+ }
+ // every data block is an MCU, so countdown the restart interval
+ if (--z->todo <= 0) {
+ if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
+ if (!STBI__RESTART(z->marker)) return 1;
+ stbi__jpeg_reset(z);
+ }
+ }
+ }
+ return 1;
+ } else { // interleaved
+ int i,j,k,x,y;
+ for (j=0; j < z->img_mcu_y; ++j) {
+ for (i=0; i < z->img_mcu_x; ++i) {
+ // scan an interleaved mcu... process scan_n components in order
+ for (k=0; k < z->scan_n; ++k) {
+ int n = z->order[k];
+ // scan out an mcu's worth of this component; that's just determined
+ // by the basic H and V specified for the component
+ for (y=0; y < z->img_comp[n].v; ++y) {
+ for (x=0; x < z->img_comp[n].h; ++x) {
+ int x2 = (i*z->img_comp[n].h + x);
+ int y2 = (j*z->img_comp[n].v + y);
+ short *data = z->img_comp[n].coeff + 64 * (x2 + y2 * z->img_comp[n].coeff_w);
+ if (!stbi__jpeg_decode_block_prog_dc(z, data, &z->huff_dc[z->img_comp[n].hd], n))
+ return 0;
+ }
+ }
+ }
+ // after all interleaved components, that's an interleaved MCU,
+ // so now count down the restart interval
+ if (--z->todo <= 0) {
+ if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
+ if (!STBI__RESTART(z->marker)) return 1;
+ stbi__jpeg_reset(z);
+ }
+ }
+ }
+ return 1;
+ }
+ }
+}
+
+static void stbi__jpeg_dequantize(short *data, stbi__uint16 *dequant)
+{
+ int i;
+ for (i=0; i < 64; ++i)
+ data[i] *= dequant[i];
+}
+
+static void stbi__jpeg_finish(stbi__jpeg *z)
+{
+ if (z->progressive) {
+ // dequantize and idct the data
+ int i,j,n;
+ for (n=0; n < z->s->img_n; ++n) {
+ int w = (z->img_comp[n].x+7) >> 3;
+ int h = (z->img_comp[n].y+7) >> 3;
+ for (j=0; j < h; ++j) {
+ for (i=0; i < w; ++i) {
+ short *data = z->img_comp[n].coeff + 64 * (i + j * z->img_comp[n].coeff_w);
+ stbi__jpeg_dequantize(data, z->dequant[z->img_comp[n].tq]);
+ z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*j*8+i*8, z->img_comp[n].w2, data);
+ }
+ }
+ }
+ }
+}
+
+static int stbi__process_marker(stbi__jpeg *z, int m)
+{
+ int L;
+ switch (m) {
+ case STBI__MARKER_none: // no marker found
+ return stbi__err("expected marker","Corrupt JPEG");
+
+ case 0xDD: // DRI - specify restart interval
+ if (stbi__get16be(z->s) != 4) return stbi__err("bad DRI len","Corrupt JPEG");
+ z->restart_interval = stbi__get16be(z->s);
+ return 1;
+
+ case 0xDB: // DQT - define quantization table
+ L = stbi__get16be(z->s)-2;
+ while (L > 0) {
+ int q = stbi__get8(z->s);
+ int p = q >> 4, sixteen = (p != 0);
+ int t = q & 15,i;
+ if (p != 0 && p != 1) return stbi__err("bad DQT type","Corrupt JPEG");
+ if (t > 3) return stbi__err("bad DQT table","Corrupt JPEG");
+
+ for (i=0; i < 64; ++i)
+ z->dequant[t][stbi__jpeg_dezigzag[i]] = sixteen ? stbi__get16be(z->s) : stbi__get8(z->s);
+ L -= (sixteen ? 129 : 65);
+ }
+ return L==0;
+
+ case 0xC4: // DHT - define huffman table
+ L = stbi__get16be(z->s)-2;
+ while (L > 0) {
+ stbi_uc *v;
+ int sizes[16],i,n=0;
+ int q = stbi__get8(z->s);
+ int tc = q >> 4;
+ int th = q & 15;
+ if (tc > 1 || th > 3) return stbi__err("bad DHT header","Corrupt JPEG");
+ for (i=0; i < 16; ++i) {
+ sizes[i] = stbi__get8(z->s);
+ n += sizes[i];
+ }
+ L -= 17;
+ if (tc == 0) {
+ if (!stbi__build_huffman(z->huff_dc+th, sizes)) return 0;
+ v = z->huff_dc[th].values;
+ } else {
+ if (!stbi__build_huffman(z->huff_ac+th, sizes)) return 0;
+ v = z->huff_ac[th].values;
+ }
+ for (i=0; i < n; ++i)
+ v[i] = stbi__get8(z->s);
+ if (tc != 0)
+ stbi__build_fast_ac(z->fast_ac[th], z->huff_ac + th);
+ L -= n;
+ }
+ return L==0;
+ }
+
+ // check for comment block or APP blocks
+ if ((m >= 0xE0 && m <= 0xEF) || m == 0xFE) {
+ L = stbi__get16be(z->s);
+ if (L < 2) {
+ if (m == 0xFE)
+ return stbi__err("bad COM len","Corrupt JPEG");
+ else
+ return stbi__err("bad APP len","Corrupt JPEG");
+ }
+ L -= 2;
+
+ if (m == 0xE0 && L >= 5) { // JFIF APP0 segment
+ static const unsigned char tag[5] = {'J','F','I','F','\0'};
+ int ok = 1;
+ int i;
+ for (i=0; i < 5; ++i)
+ if (stbi__get8(z->s) != tag[i])
+ ok = 0;
+ L -= 5;
+ if (ok)
+ z->jfif = 1;
+ } else if (m == 0xEE && L >= 12) { // Adobe APP14 segment
+ static const unsigned char tag[6] = {'A','d','o','b','e','\0'};
+ int ok = 1;
+ int i;
+ for (i=0; i < 6; ++i)
+ if (stbi__get8(z->s) != tag[i])
+ ok = 0;
+ L -= 6;
+ if (ok) {
+ stbi__get8(z->s); // version
+ stbi__get16be(z->s); // flags0
+ stbi__get16be(z->s); // flags1
+ z->app14_color_transform = stbi__get8(z->s); // color transform
+ L -= 6;
+ }
+ }
+
+ stbi__skip(z->s, L);
+ return 1;
+ }
+
+ return stbi__err("unknown marker","Corrupt JPEG");
+}
+
+// after we see SOS
+static int stbi__process_scan_header(stbi__jpeg *z)
+{
+ int i;
+ int Ls = stbi__get16be(z->s);
+ z->scan_n = stbi__get8(z->s);
+ if (z->scan_n < 1 || z->scan_n > 4 || z->scan_n > (int) z->s->img_n) return stbi__err("bad SOS component count","Corrupt JPEG");
+ if (Ls != 6+2*z->scan_n) return stbi__err("bad SOS len","Corrupt JPEG");
+ for (i=0; i < z->scan_n; ++i) {
+ int id = stbi__get8(z->s), which;
+ int q = stbi__get8(z->s);
+ for (which = 0; which < z->s->img_n; ++which)
+ if (z->img_comp[which].id == id)
+ break;
+ if (which == z->s->img_n) return 0; // no match
+ z->img_comp[which].hd = q >> 4; if (z->img_comp[which].hd > 3) return stbi__err("bad DC huff","Corrupt JPEG");
+ z->img_comp[which].ha = q & 15; if (z->img_comp[which].ha > 3) return stbi__err("bad AC huff","Corrupt JPEG");
+ z->order[i] = which;
+ }
+
+ {
+ int aa;
+ z->spec_start = stbi__get8(z->s);
+ z->spec_end = stbi__get8(z->s); // should be 63, but might be 0
+ aa = stbi__get8(z->s);
+ z->succ_high = (aa >> 4);
+ z->succ_low = (aa & 15);
+ if (z->progressive) {
+ if (z->spec_start > 63 || z->spec_end > 63 || z->spec_start > z->spec_end || z->succ_high > 13 || z->succ_low > 13)
+ return stbi__err("bad SOS", "Corrupt JPEG");
+ } else {
+ if (z->spec_start != 0) return stbi__err("bad SOS","Corrupt JPEG");
+ if (z->succ_high != 0 || z->succ_low != 0) return stbi__err("bad SOS","Corrupt JPEG");
+ z->spec_end = 63;
+ }
+ }
+
+ return 1;
+}
+
+static int stbi__free_jpeg_components(stbi__jpeg *z, int ncomp, int why)
+{
+ int i;
+ for (i=0; i < ncomp; ++i) {
+ if (z->img_comp[i].raw_data) {
+ STBI_FREE(z->img_comp[i].raw_data);
+ z->img_comp[i].raw_data = NULL;
+ z->img_comp[i].data = NULL;
+ }
+ if (z->img_comp[i].raw_coeff) {
+ STBI_FREE(z->img_comp[i].raw_coeff);
+ z->img_comp[i].raw_coeff = 0;
+ z->img_comp[i].coeff = 0;
+ }
+ if (z->img_comp[i].linebuf) {
+ STBI_FREE(z->img_comp[i].linebuf);
+ z->img_comp[i].linebuf = NULL;
+ }
+ }
+ return why;
+}
+
+static int stbi__process_frame_header(stbi__jpeg *z, int scan)
+{
+ stbi__context *s = z->s;
+ int Lf,p,i,q, h_max=1,v_max=1,c;
+ Lf = stbi__get16be(s); if (Lf < 11) return stbi__err("bad SOF len","Corrupt JPEG"); // JPEG
+ p = stbi__get8(s); if (p != 8) return stbi__err("only 8-bit","JPEG format not supported: 8-bit only"); // JPEG baseline
+ s->img_y = stbi__get16be(s); if (s->img_y == 0) return stbi__err("no header height", "JPEG format not supported: delayed height"); // Legal, but we don't handle it--but neither does IJG
+ s->img_x = stbi__get16be(s); if (s->img_x == 0) return stbi__err("0 width","Corrupt JPEG"); // JPEG requires
+ c = stbi__get8(s);
+ if (c != 3 && c != 1 && c != 4) return stbi__err("bad component count","Corrupt JPEG");
+ s->img_n = c;
+ for (i=0; i < c; ++i) {
+ z->img_comp[i].data = NULL;
+ z->img_comp[i].linebuf = NULL;
+ }
+
+ if (Lf != 8+3*s->img_n) return stbi__err("bad SOF len","Corrupt JPEG");
+
+ z->rgb = 0;
+ for (i=0; i < s->img_n; ++i) {
+ static unsigned char rgb[3] = { 'R', 'G', 'B' };
+ z->img_comp[i].id = stbi__get8(s);
+ if (s->img_n == 3 && z->img_comp[i].id == rgb[i])
+ ++z->rgb;
+ q = stbi__get8(s);
+ z->img_comp[i].h = (q >> 4); if (!z->img_comp[i].h || z->img_comp[i].h > 4) return stbi__err("bad H","Corrupt JPEG");
+ z->img_comp[i].v = q & 15; if (!z->img_comp[i].v || z->img_comp[i].v > 4) return stbi__err("bad V","Corrupt JPEG");
+ z->img_comp[i].tq = stbi__get8(s); if (z->img_comp[i].tq > 3) return stbi__err("bad TQ","Corrupt JPEG");
+ }
+
+ if (scan != STBI__SCAN_load) return 1;
+
+ if (!stbi__mad3sizes_valid(s->img_x, s->img_y, s->img_n, 0)) return stbi__err("too large", "Image too large to decode");
+
+ for (i=0; i < s->img_n; ++i) {
+ if (z->img_comp[i].h > h_max) h_max = z->img_comp[i].h;
+ if (z->img_comp[i].v > v_max) v_max = z->img_comp[i].v;
+ }
+
+ // compute interleaved mcu info
+ z->img_h_max = h_max;
+ z->img_v_max = v_max;
+ z->img_mcu_w = h_max * 8;
+ z->img_mcu_h = v_max * 8;
+ // these sizes can't be more than 17 bits
+ z->img_mcu_x = (s->img_x + z->img_mcu_w-1) / z->img_mcu_w;
+ z->img_mcu_y = (s->img_y + z->img_mcu_h-1) / z->img_mcu_h;
+
+ for (i=0; i < s->img_n; ++i) {
+ // number of effective pixels (e.g. for non-interleaved MCU)
+ z->img_comp[i].x = (s->img_x * z->img_comp[i].h + h_max-1) / h_max;
+ z->img_comp[i].y = (s->img_y * z->img_comp[i].v + v_max-1) / v_max;
+ // to simplify generation, we'll allocate enough memory to decode
+ // the bogus oversized data from using interleaved MCUs and their
+ // big blocks (e.g. a 16x16 iMCU on an image of width 33); we won't
+ // discard the extra data until colorspace conversion
+ //
+ // img_mcu_x, img_mcu_y: <=17 bits; comp[i].h and .v are <=4 (checked earlier)
+ // so these muls can't overflow with 32-bit ints (which we require)
+ z->img_comp[i].w2 = z->img_mcu_x * z->img_comp[i].h * 8;
+ z->img_comp[i].h2 = z->img_mcu_y * z->img_comp[i].v * 8;
+ z->img_comp[i].coeff = 0;
+ z->img_comp[i].raw_coeff = 0;
+ z->img_comp[i].linebuf = NULL;
+ z->img_comp[i].raw_data = stbi__malloc_mad2(z->img_comp[i].w2, z->img_comp[i].h2, 15);
+ if (z->img_comp[i].raw_data == NULL)
+ return stbi__free_jpeg_components(z, i+1, stbi__err("outofmem", "Out of memory"));
+ // align blocks for idct using mmx/sse
+ z->img_comp[i].data = (stbi_uc*) (((size_t) z->img_comp[i].raw_data + 15) & ~15);
+ if (z->progressive) {
+ // w2, h2 are multiples of 8 (see above)
+ z->img_comp[i].coeff_w = z->img_comp[i].w2 / 8;
+ z->img_comp[i].coeff_h = z->img_comp[i].h2 / 8;
+ z->img_comp[i].raw_coeff = stbi__malloc_mad3(z->img_comp[i].w2, z->img_comp[i].h2, sizeof(short), 15);
+ if (z->img_comp[i].raw_coeff == NULL)
+ return stbi__free_jpeg_components(z, i+1, stbi__err("outofmem", "Out of memory"));
+ z->img_comp[i].coeff = (short*) (((size_t) z->img_comp[i].raw_coeff + 15) & ~15);
+ }
+ }
+
+ return 1;
+}
+
+// use comparisons since in some cases we handle more than one case (e.g. SOF)
+#define stbi__DNL(x) ((x) == 0xdc)
+#define stbi__SOI(x) ((x) == 0xd8)
+#define stbi__EOI(x) ((x) == 0xd9)
+#define stbi__SOF(x) ((x) == 0xc0 || (x) == 0xc1 || (x) == 0xc2)
+#define stbi__SOS(x) ((x) == 0xda)
+
+#define stbi__SOF_progressive(x) ((x) == 0xc2)
+
+static int stbi__decode_jpeg_header(stbi__jpeg *z, int scan)
+{
+ int m;
+ z->jfif = 0;
+ z->app14_color_transform = -1; // valid values are 0,1,2
+ z->marker = STBI__MARKER_none; // initialize cached marker to empty
+ m = stbi__get_marker(z);
+ if (!stbi__SOI(m)) return stbi__err("no SOI","Corrupt JPEG");
+ if (scan == STBI__SCAN_type) return 1;
+ m = stbi__get_marker(z);
+ while (!stbi__SOF(m)) {
+ if (!stbi__process_marker(z,m)) return 0;
+ m = stbi__get_marker(z);
+ while (m == STBI__MARKER_none) {
+ // some files have extra padding after their blocks, so ok, we'll scan
+ if (stbi__at_eof(z->s)) return stbi__err("no SOF", "Corrupt JPEG");
+ m = stbi__get_marker(z);
+ }
+ }
+ z->progressive = stbi__SOF_progressive(m);
+ if (!stbi__process_frame_header(z, scan)) return 0;
+ return 1;
+}
+
+// decode image to YCbCr format
+static int stbi__decode_jpeg_image(stbi__jpeg *j)
+{
+ int m;
+ for (m = 0; m < 4; m++) {
+ j->img_comp[m].raw_data = NULL;
+ j->img_comp[m].raw_coeff = NULL;
+ }
+ j->restart_interval = 0;
+ if (!stbi__decode_jpeg_header(j, STBI__SCAN_load)) return 0;
+ m = stbi__get_marker(j);
+ while (!stbi__EOI(m)) {
+ if (stbi__SOS(m)) {
+ if (!stbi__process_scan_header(j)) return 0;
+ if (!stbi__parse_entropy_coded_data(j)) return 0;
+ if (j->marker == STBI__MARKER_none ) {
+ // handle 0s at the end of image data from IP Kamera 9060
+ while (!stbi__at_eof(j->s)) {
+ int x = stbi__get8(j->s);
+ if (x == 255) {
+ j->marker = stbi__get8(j->s);
+ break;
+ }
+ }
+ // if we reach eof without hitting a marker, stbi__get_marker() below will fail and we'll eventually return 0
+ }
+ } else if (stbi__DNL(m)) {
+ int Ld = stbi__get16be(j->s);
+ stbi__uint32 NL = stbi__get16be(j->s);
+ if (Ld != 4) stbi__err("bad DNL len", "Corrupt JPEG");
+ if (NL != j->s->img_y) stbi__err("bad DNL height", "Corrupt JPEG");
+ } else {
+ if (!stbi__process_marker(j, m)) return 0;
+ }
+ m = stbi__get_marker(j);
+ }
+ if (j->progressive)
+ stbi__jpeg_finish(j);
+ return 1;
+}
+
+// static jfif-centered resampling (across block boundaries)
+
+typedef stbi_uc *(*resample_row_func)(stbi_uc *out, stbi_uc *in0, stbi_uc *in1,
+ int w, int hs);
+
+#define stbi__div4(x) ((stbi_uc) ((x) >> 2))
+
+static stbi_uc *resample_row_1(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+{
+ STBI_NOTUSED(out);
+ STBI_NOTUSED(in_far);
+ STBI_NOTUSED(w);
+ STBI_NOTUSED(hs);
+ return in_near;
+}
+
+static stbi_uc* stbi__resample_row_v_2(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+{
+ // need to generate two samples vertically for every one in input
+ int i;
+ STBI_NOTUSED(hs);
+ for (i=0; i < w; ++i)
+ out[i] = stbi__div4(3*in_near[i] + in_far[i] + 2);
+ return out;
+}
+
+static stbi_uc* stbi__resample_row_h_2(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+{
+ // need to generate two samples horizontally for every one in input
+ int i;
+ stbi_uc *input = in_near;
+
+ if (w == 1) {
+ // if only one sample, can't do any interpolation
+ out[0] = out[1] = input[0];
+ return out;
+ }
+
+ out[0] = input[0];
+ out[1] = stbi__div4(input[0]*3 + input[1] + 2);
+ for (i=1; i < w-1; ++i) {
+ int n = 3*input[i]+2;
+ out[i*2+0] = stbi__div4(n+input[i-1]);
+ out[i*2+1] = stbi__div4(n+input[i+1]);
+ }
+ out[i*2+0] = stbi__div4(input[w-2]*3 + input[w-1] + 2);
+ out[i*2+1] = input[w-1];
+
+ STBI_NOTUSED(in_far);
+ STBI_NOTUSED(hs);
+
+ return out;
+}
+
+#define stbi__div16(x) ((stbi_uc) ((x) >> 4))
+
+static stbi_uc *stbi__resample_row_hv_2(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+{
+ // need to generate 2x2 samples for every one in input
+ int i,t0,t1;
+ if (w == 1) {
+ out[0] = out[1] = stbi__div4(3*in_near[0] + in_far[0] + 2);
+ return out;
+ }
+
+ t1 = 3*in_near[0] + in_far[0];
+ out[0] = stbi__div4(t1+2);
+ for (i=1; i < w; ++i) {
+ t0 = t1;
+ t1 = 3*in_near[i]+in_far[i];
+ out[i*2-1] = stbi__div16(3*t0 + t1 + 8);
+ out[i*2 ] = stbi__div16(3*t1 + t0 + 8);
+ }
+ out[w*2-1] = stbi__div4(t1+2);
+
+ STBI_NOTUSED(hs);
+
+ return out;
+}
+
+#if defined(STBI_SSE2) || defined(STBI_NEON)
+static stbi_uc *stbi__resample_row_hv_2_simd(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+{
+ // need to generate 2x2 samples for every one in input
+ int i=0,t0,t1;
+
+ if (w == 1) {
+ out[0] = out[1] = stbi__div4(3*in_near[0] + in_far[0] + 2);
+ return out;
+ }
+
+ t1 = 3*in_near[0] + in_far[0];
+ // process groups of 8 pixels for as long as we can.
+ // note we can't handle the last pixel in a row in this loop
+ // because we need to handle the filter boundary conditions.
+ for (; i < ((w-1) & ~7); i += 8) {
+#if defined(STBI_SSE2)
+ // load and perform the vertical filtering pass
+ // this uses 3*x + y = 4*x + (y - x)
+ __m128i zero = _mm_setzero_si128();
+ __m128i farb = _mm_loadl_epi64((__m128i *) (in_far + i));
+ __m128i nearb = _mm_loadl_epi64((__m128i *) (in_near + i));
+ __m128i farw = _mm_unpacklo_epi8(farb, zero);
+ __m128i nearw = _mm_unpacklo_epi8(nearb, zero);
+ __m128i diff = _mm_sub_epi16(farw, nearw);
+ __m128i nears = _mm_slli_epi16(nearw, 2);
+ __m128i curr = _mm_add_epi16(nears, diff); // current row
+
+ // horizontal filter works the same based on shifted vers of current
+ // row. "prev" is current row shifted right by 1 pixel; we need to
+ // insert the previous pixel value (from t1).
+ // "next" is current row shifted left by 1 pixel, with first pixel
+ // of next block of 8 pixels added in.
+ __m128i prv0 = _mm_slli_si128(curr, 2);
+ __m128i nxt0 = _mm_srli_si128(curr, 2);
+ __m128i prev = _mm_insert_epi16(prv0, t1, 0);
+ __m128i next = _mm_insert_epi16(nxt0, 3*in_near[i+8] + in_far[i+8], 7);
+
+ // horizontal filter, polyphase implementation since it's convenient:
+ // even pixels = 3*cur + prev = cur*4 + (prev - cur)
+ // odd pixels = 3*cur + next = cur*4 + (next - cur)
+ // note the shared term.
+ __m128i bias = _mm_set1_epi16(8);
+ __m128i curs = _mm_slli_epi16(curr, 2);
+ __m128i prvd = _mm_sub_epi16(prev, curr);
+ __m128i nxtd = _mm_sub_epi16(next, curr);
+ __m128i curb = _mm_add_epi16(curs, bias);
+ __m128i even = _mm_add_epi16(prvd, curb);
+ __m128i odd = _mm_add_epi16(nxtd, curb);
+
+ // interleave even and odd pixels, then undo scaling.
+ __m128i int0 = _mm_unpacklo_epi16(even, odd);
+ __m128i int1 = _mm_unpackhi_epi16(even, odd);
+ __m128i de0 = _mm_srli_epi16(int0, 4);
+ __m128i de1 = _mm_srli_epi16(int1, 4);
+
+ // pack and write output
+ __m128i outv = _mm_packus_epi16(de0, de1);
+ _mm_storeu_si128((__m128i *) (out + i*2), outv);
+#elif defined(STBI_NEON)
+ // load and perform the vertical filtering pass
+ // this uses 3*x + y = 4*x + (y - x)
+ uint8x8_t farb = vld1_u8(in_far + i);
+ uint8x8_t nearb = vld1_u8(in_near + i);
+ int16x8_t diff = vreinterpretq_s16_u16(vsubl_u8(farb, nearb));
+ int16x8_t nears = vreinterpretq_s16_u16(vshll_n_u8(nearb, 2));
+ int16x8_t curr = vaddq_s16(nears, diff); // current row
+
+ // horizontal filter works the same based on shifted vers of current
+ // row. "prev" is current row shifted right by 1 pixel; we need to
+ // insert the previous pixel value (from t1).
+ // "next" is current row shifted left by 1 pixel, with first pixel
+ // of next block of 8 pixels added in.
+ int16x8_t prv0 = vextq_s16(curr, curr, 7);
+ int16x8_t nxt0 = vextq_s16(curr, curr, 1);
+ int16x8_t prev = vsetq_lane_s16(t1, prv0, 0);
+ int16x8_t next = vsetq_lane_s16(3*in_near[i+8] + in_far[i+8], nxt0, 7);
+
+ // horizontal filter, polyphase implementation since it's convenient:
+ // even pixels = 3*cur + prev = cur*4 + (prev - cur)
+ // odd pixels = 3*cur + next = cur*4 + (next - cur)
+ // note the shared term.
+ int16x8_t curs = vshlq_n_s16(curr, 2);
+ int16x8_t prvd = vsubq_s16(prev, curr);
+ int16x8_t nxtd = vsubq_s16(next, curr);
+ int16x8_t even = vaddq_s16(curs, prvd);
+ int16x8_t odd = vaddq_s16(curs, nxtd);
+
+ // undo scaling and round, then store with even/odd phases interleaved
+ uint8x8x2_t o;
+ o.val[0] = vqrshrun_n_s16(even, 4);
+ o.val[1] = vqrshrun_n_s16(odd, 4);
+ vst2_u8(out + i*2, o);
+#endif
+
+ // "previous" value for next iter
+ t1 = 3*in_near[i+7] + in_far[i+7];
+ }
+
+ t0 = t1;
+ t1 = 3*in_near[i] + in_far[i];
+ out[i*2] = stbi__div16(3*t1 + t0 + 8);
+
+ for (++i; i < w; ++i) {
+ t0 = t1;
+ t1 = 3*in_near[i]+in_far[i];
+ out[i*2-1] = stbi__div16(3*t0 + t1 + 8);
+ out[i*2 ] = stbi__div16(3*t1 + t0 + 8);
+ }
+ out[w*2-1] = stbi__div4(t1+2);
+
+ STBI_NOTUSED(hs);
+
+ return out;
+}
+#endif
+
+static stbi_uc *stbi__resample_row_generic(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+{
+ // resample with nearest-neighbor
+ int i,j;
+ STBI_NOTUSED(in_far);
+ for (i=0; i < w; ++i)
+ for (j=0; j < hs; ++j)
+ out[i*hs+j] = in_near[i];
+ return out;
+}
+
+// this is a reduced-precision calculation of YCbCr-to-RGB introduced
+// to make sure the code produces the same results in both SIMD and scalar
+#define stbi__float2fixed(x) (((int) ((x) * 4096.0f + 0.5f)) << 8)
+static void stbi__YCbCr_to_RGB_row(stbi_uc *out, const stbi_uc *y, const stbi_uc *pcb, const stbi_uc *pcr, int count, int step)
+{
+ int i;
+ for (i=0; i < count; ++i) {
+ int y_fixed = (y[i] << 20) + (1<<19); // rounding
+ int r,g,b;
+ int cr = pcr[i] - 128;
+ int cb = pcb[i] - 128;
+ r = y_fixed + cr* stbi__float2fixed(1.40200f);
+ g = y_fixed + (cr*-stbi__float2fixed(0.71414f)) + ((cb*-stbi__float2fixed(0.34414f)) & 0xffff0000);
+ b = y_fixed + cb* stbi__float2fixed(1.77200f);
+ r >>= 20;
+ g >>= 20;
+ b >>= 20;
+ if ((unsigned) r > 255) { if (r < 0) r = 0; else r = 255; }
+ if ((unsigned) g > 255) { if (g < 0) g = 0; else g = 255; }
+ if ((unsigned) b > 255) { if (b < 0) b = 0; else b = 255; }
+ out[0] = (stbi_uc)r;
+ out[1] = (stbi_uc)g;
+ out[2] = (stbi_uc)b;
+ out[3] = 255;
+ out += step;
+ }
+}
+
+#if defined(STBI_SSE2) || defined(STBI_NEON)
+static void stbi__YCbCr_to_RGB_simd(stbi_uc *out, stbi_uc const *y, stbi_uc const *pcb, stbi_uc const *pcr, int count, int step)
+{
+ int i = 0;
+
+#ifdef STBI_SSE2
+ // step == 3 is pretty ugly on the final interleave, and i'm not convinced
+ // it's useful in practice (you wouldn't use it for textures, for example).
+ // so just accelerate step == 4 case.
+ if (step == 4) {
+ // this is a fairly straightforward implementation and not super-optimized.
+ __m128i signflip = _mm_set1_epi8(-0x80);
+ __m128i cr_const0 = _mm_set1_epi16( (short) ( 1.40200f*4096.0f+0.5f));
+ __m128i cr_const1 = _mm_set1_epi16( - (short) ( 0.71414f*4096.0f+0.5f));
+ __m128i cb_const0 = _mm_set1_epi16( - (short) ( 0.34414f*4096.0f+0.5f));
+ __m128i cb_const1 = _mm_set1_epi16( (short) ( 1.77200f*4096.0f+0.5f));
+ __m128i y_bias = _mm_set1_epi8((char) (unsigned char) 128);
+ __m128i xw = _mm_set1_epi16(255); // alpha channel
+
+ for (; i+7 < count; i += 8) {
+ // load
+ __m128i y_bytes = _mm_loadl_epi64((__m128i *) (y+i));
+ __m128i cr_bytes = _mm_loadl_epi64((__m128i *) (pcr+i));
+ __m128i cb_bytes = _mm_loadl_epi64((__m128i *) (pcb+i));
+ __m128i cr_biased = _mm_xor_si128(cr_bytes, signflip); // -128
+ __m128i cb_biased = _mm_xor_si128(cb_bytes, signflip); // -128
+
+ // unpack to short (and left-shift cr, cb by 8)
+ __m128i yw = _mm_unpacklo_epi8(y_bias, y_bytes);
+ __m128i crw = _mm_unpacklo_epi8(_mm_setzero_si128(), cr_biased);
+ __m128i cbw = _mm_unpacklo_epi8(_mm_setzero_si128(), cb_biased);
+
+ // color transform
+ __m128i yws = _mm_srli_epi16(yw, 4);
+ __m128i cr0 = _mm_mulhi_epi16(cr_const0, crw);
+ __m128i cb0 = _mm_mulhi_epi16(cb_const0, cbw);
+ __m128i cb1 = _mm_mulhi_epi16(cbw, cb_const1);
+ __m128i cr1 = _mm_mulhi_epi16(crw, cr_const1);
+ __m128i rws = _mm_add_epi16(cr0, yws);
+ __m128i gwt = _mm_add_epi16(cb0, yws);
+ __m128i bws = _mm_add_epi16(yws, cb1);
+ __m128i gws = _mm_add_epi16(gwt, cr1);
+
+ // descale
+ __m128i rw = _mm_srai_epi16(rws, 4);
+ __m128i bw = _mm_srai_epi16(bws, 4);
+ __m128i gw = _mm_srai_epi16(gws, 4);
+
+ // back to byte, set up for transpose
+ __m128i brb = _mm_packus_epi16(rw, bw);
+ __m128i gxb = _mm_packus_epi16(gw, xw);
+
+ // transpose to interleave channels
+ __m128i t0 = _mm_unpacklo_epi8(brb, gxb);
+ __m128i t1 = _mm_unpackhi_epi8(brb, gxb);
+ __m128i o0 = _mm_unpacklo_epi16(t0, t1);
+ __m128i o1 = _mm_unpackhi_epi16(t0, t1);
+
+ // store
+ _mm_storeu_si128((__m128i *) (out + 0), o0);
+ _mm_storeu_si128((__m128i *) (out + 16), o1);
+ out += 32;
+ }
+ }
+#endif
+
+#ifdef STBI_NEON
+ // in this version, step=3 support would be easy to add. but is there demand?
+ if (step == 4) {
+ // this is a fairly straightforward implementation and not super-optimized.
+ uint8x8_t signflip = vdup_n_u8(0x80);
+ int16x8_t cr_const0 = vdupq_n_s16( (short) ( 1.40200f*4096.0f+0.5f));
+ int16x8_t cr_const1 = vdupq_n_s16( - (short) ( 0.71414f*4096.0f+0.5f));
+ int16x8_t cb_const0 = vdupq_n_s16( - (short) ( 0.34414f*4096.0f+0.5f));
+ int16x8_t cb_const1 = vdupq_n_s16( (short) ( 1.77200f*4096.0f+0.5f));
+
+ for (; i+7 < count; i += 8) {
+ // load
+ uint8x8_t y_bytes = vld1_u8(y + i);
+ uint8x8_t cr_bytes = vld1_u8(pcr + i);
+ uint8x8_t cb_bytes = vld1_u8(pcb + i);
+ int8x8_t cr_biased = vreinterpret_s8_u8(vsub_u8(cr_bytes, signflip));
+ int8x8_t cb_biased = vreinterpret_s8_u8(vsub_u8(cb_bytes, signflip));
+
+ // expand to s16
+ int16x8_t yws = vreinterpretq_s16_u16(vshll_n_u8(y_bytes, 4));
+ int16x8_t crw = vshll_n_s8(cr_biased, 7);
+ int16x8_t cbw = vshll_n_s8(cb_biased, 7);
+
+ // color transform
+ int16x8_t cr0 = vqdmulhq_s16(crw, cr_const0);
+ int16x8_t cb0 = vqdmulhq_s16(cbw, cb_const0);
+ int16x8_t cr1 = vqdmulhq_s16(crw, cr_const1);
+ int16x8_t cb1 = vqdmulhq_s16(cbw, cb_const1);
+ int16x8_t rws = vaddq_s16(yws, cr0);
+ int16x8_t gws = vaddq_s16(vaddq_s16(yws, cb0), cr1);
+ int16x8_t bws = vaddq_s16(yws, cb1);
+
+ // undo scaling, round, convert to byte
+ uint8x8x4_t o;
+ o.val[0] = vqrshrun_n_s16(rws, 4);
+ o.val[1] = vqrshrun_n_s16(gws, 4);
+ o.val[2] = vqrshrun_n_s16(bws, 4);
+ o.val[3] = vdup_n_u8(255);
+
+ // store, interleaving r/g/b/a
+ vst4_u8(out, o);
+ out += 8*4;
+ }
+ }
+#endif
+
+ for (; i < count; ++i) {
+ int y_fixed = (y[i] << 20) + (1<<19); // rounding
+ int r,g,b;
+ int cr = pcr[i] - 128;
+ int cb = pcb[i] - 128;
+ r = y_fixed + cr* stbi__float2fixed(1.40200f);
+ g = y_fixed + cr*-stbi__float2fixed(0.71414f) + ((cb*-stbi__float2fixed(0.34414f)) & 0xffff0000);
+ b = y_fixed + cb* stbi__float2fixed(1.77200f);
+ r >>= 20;
+ g >>= 20;
+ b >>= 20;
+ if ((unsigned) r > 255) { if (r < 0) r = 0; else r = 255; }
+ if ((unsigned) g > 255) { if (g < 0) g = 0; else g = 255; }
+ if ((unsigned) b > 255) { if (b < 0) b = 0; else b = 255; }
+ out[0] = (stbi_uc)r;
+ out[1] = (stbi_uc)g;
+ out[2] = (stbi_uc)b;
+ out[3] = 255;
+ out += step;
+ }
+}
+#endif
+
+// set up the kernels
+static void stbi__setup_jpeg(stbi__jpeg *j)
+{
+ j->idct_block_kernel = stbi__idct_block;
+ j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_row;
+ j->resample_row_hv_2_kernel = stbi__resample_row_hv_2;
+
+#ifdef STBI_SSE2
+ if (stbi__sse2_available()) {
+ j->idct_block_kernel = stbi__idct_simd;
+ j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_simd;
+ j->resample_row_hv_2_kernel = stbi__resample_row_hv_2_simd;
+ }
+#endif
+
+#ifdef STBI_NEON
+ j->idct_block_kernel = stbi__idct_simd;
+ j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_simd;
+ j->resample_row_hv_2_kernel = stbi__resample_row_hv_2_simd;
+#endif
+}
+
+// clean up the temporary component buffers
+static void stbi__cleanup_jpeg(stbi__jpeg *j)
+{
+ stbi__free_jpeg_components(j, j->s->img_n, 0);
+}
+
+typedef struct
+{
+ resample_row_func resample;
+ stbi_uc *line0,*line1;
+ int hs,vs; // expansion factor in each axis
+ int w_lores; // horizontal pixels pre-expansion
+ int ystep; // how far through vertical expansion we are
+ int ypos; // which pre-expansion row we're on
+} stbi__resample;
+
+// fast 0..255 * 0..255 => 0..255 rounded multiplication
+static stbi_uc stbi__blinn_8x8(stbi_uc x, stbi_uc y)
+{
+ unsigned int t = x*y + 128;
+ return (stbi_uc) ((t + (t >>8)) >> 8);
+}
+
+static stbi_uc *load_jpeg_image(stbi__jpeg *z, int *out_x, int *out_y, int *comp, int req_comp)
+{
+ int n, decode_n, is_rgb;
+ z->s->img_n = 0; // make stbi__cleanup_jpeg safe
+
+ // validate req_comp
+ if (req_comp < 0 || req_comp > 4) return stbi__errpuc("bad req_comp", "Internal error");
+
+ // load a jpeg image from whichever source, but leave in YCbCr format
+ if (!stbi__decode_jpeg_image(z)) { stbi__cleanup_jpeg(z); return NULL; }
+
+ // determine actual number of components to generate
+ n = req_comp ? req_comp : z->s->img_n >= 3 ? 3 : 1;
+
+ is_rgb = z->s->img_n == 3 && (z->rgb == 3 || (z->app14_color_transform == 0 && !z->jfif));
+
+ if (z->s->img_n == 3 && n < 3 && !is_rgb)
+ decode_n = 1;
+ else
+ decode_n = z->s->img_n;
+
+ // resample and color-convert
+ {
+ int k;
+ unsigned int i,j;
+ stbi_uc *output;
+ stbi_uc *coutput[4];
+
+ stbi__resample res_comp[4];
+
+ for (k=0; k < decode_n; ++k) {
+ stbi__resample *r = &res_comp[k];
+
+ // allocate line buffer big enough for upsampling off the edges
+ // with upsample factor of 4
+ z->img_comp[k].linebuf = (stbi_uc *) stbi__malloc(z->s->img_x + 3);
+ if (!z->img_comp[k].linebuf) { stbi__cleanup_jpeg(z); return stbi__errpuc("outofmem", "Out of memory"); }
+
+ r->hs = z->img_h_max / z->img_comp[k].h;
+ r->vs = z->img_v_max / z->img_comp[k].v;
+ r->ystep = r->vs >> 1;
+ r->w_lores = (z->s->img_x + r->hs-1) / r->hs;
+ r->ypos = 0;
+ r->line0 = r->line1 = z->img_comp[k].data;
+
+ if (r->hs == 1 && r->vs == 1) r->resample = resample_row_1;
+ else if (r->hs == 1 && r->vs == 2) r->resample = stbi__resample_row_v_2;
+ else if (r->hs == 2 && r->vs == 1) r->resample = stbi__resample_row_h_2;
+ else if (r->hs == 2 && r->vs == 2) r->resample = z->resample_row_hv_2_kernel;
+ else r->resample = stbi__resample_row_generic;
+ }
+
+ // can't error after this so, this is safe
+ output = (stbi_uc *) stbi__malloc_mad3(n, z->s->img_x, z->s->img_y, 1);
+ if (!output) { stbi__cleanup_jpeg(z); return stbi__errpuc("outofmem", "Out of memory"); }
+
+ // now go ahead and resample
+ for (j=0; j < z->s->img_y; ++j) {
+ stbi_uc *out = output + n * z->s->img_x * j;
+ for (k=0; k < decode_n; ++k) {
+ stbi__resample *r = &res_comp[k];
+ int y_bot = r->ystep >= (r->vs >> 1);
+ coutput[k] = r->resample(z->img_comp[k].linebuf,
+ y_bot ? r->line1 : r->line0,
+ y_bot ? r->line0 : r->line1,
+ r->w_lores, r->hs);
+ if (++r->ystep >= r->vs) {
+ r->ystep = 0;
+ r->line0 = r->line1;
+ if (++r->ypos < z->img_comp[k].y)
+ r->line1 += z->img_comp[k].w2;
+ }
+ }
+ if (n >= 3) {
+ stbi_uc *y = coutput[0];
+ if (z->s->img_n == 3) {
+ if (is_rgb) {
+ for (i=0; i < z->s->img_x; ++i) {
+ out[0] = y[i];
+ out[1] = coutput[1][i];
+ out[2] = coutput[2][i];
+ out[3] = 255;
+ out += n;
+ }
+ } else {
+ z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n);
+ }
+ } else if (z->s->img_n == 4) {
+ if (z->app14_color_transform == 0) { // CMYK
+ for (i=0; i < z->s->img_x; ++i) {
+ stbi_uc k = coutput[3][i];
+ out[0] = stbi__blinn_8x8(coutput[0][i], k);
+ out[1] = stbi__blinn_8x8(coutput[1][i], k);
+ out[2] = stbi__blinn_8x8(coutput[2][i], k);
+ out[3] = 255;
+ out += n;
+ }
+ } else if (z->app14_color_transform == 2) { // YCCK
+ z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n);
+ for (i=0; i < z->s->img_x; ++i) {
+ stbi_uc k = coutput[3][i];
+ out[0] = stbi__blinn_8x8(255 - out[0], k);
+ out[1] = stbi__blinn_8x8(255 - out[1], k);
+ out[2] = stbi__blinn_8x8(255 - out[2], k);
+ out += n;
+ }
+ } else { // YCbCr + alpha? Ignore the fourth channel for now
+ z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n);
+ }
+ } else
+ for (i=0; i < z->s->img_x; ++i) {
+ out[0] = out[1] = out[2] = y[i];
+ out[3] = 255; // not used if n==3
+ out += n;
+ }
+ } else {
+ if (is_rgb) {
+ if (n == 1)
+ for (i=0; i < z->s->img_x; ++i)
+ *out++ = stbi__compute_y(coutput[0][i], coutput[1][i], coutput[2][i]);
+ else {
+ for (i=0; i < z->s->img_x; ++i, out += 2) {
+ out[0] = stbi__compute_y(coutput[0][i], coutput[1][i], coutput[2][i]);
+ out[1] = 255;
+ }
+ }
+ } else if (z->s->img_n == 4 && z->app14_color_transform == 0) {
+ for (i=0; i < z->s->img_x; ++i) {
+ stbi_uc k = coutput[3][i];
+ stbi_uc r = stbi__blinn_8x8(coutput[0][i], k);
+ stbi_uc g = stbi__blinn_8x8(coutput[1][i], k);
+ stbi_uc b = stbi__blinn_8x8(coutput[2][i], k);
+ out[0] = stbi__compute_y(r, g, b);
+ out[1] = 255;
+ out += n;
+ }
+ } else if (z->s->img_n == 4 && z->app14_color_transform == 2) {
+ for (i=0; i < z->s->img_x; ++i) {
+ out[0] = stbi__blinn_8x8(255 - coutput[0][i], coutput[3][i]);
+ out[1] = 255;
+ out += n;
+ }
+ } else {
+ stbi_uc *y = coutput[0];
+ if (n == 1)
+ for (i=0; i < z->s->img_x; ++i) out[i] = y[i];
+ else
+ for (i=0; i < z->s->img_x; ++i) *out++ = y[i], *out++ = 255;
+ }
+ }
+ }
+ stbi__cleanup_jpeg(z);
+ *out_x = z->s->img_x;
+ *out_y = z->s->img_y;
+ if (comp) *comp = z->s->img_n >= 3 ? 3 : 1; // report original components, not output
+ return output;
+ }
+}
+
+static void *stbi__jpeg_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+ unsigned char* result;
+ stbi__jpeg* j = (stbi__jpeg*) stbi__malloc(sizeof(stbi__jpeg));
+ STBI_NOTUSED(ri);
+ j->s = s;
+ stbi__setup_jpeg(j);
+ result = load_jpeg_image(j, x,y,comp,req_comp);
+ STBI_FREE(j);
+ return result;
+}
+
+static int stbi__jpeg_test(stbi__context *s)
+{
+ int r;
+ stbi__jpeg* j = (stbi__jpeg*)stbi__malloc(sizeof(stbi__jpeg));
+ j->s = s;
+ stbi__setup_jpeg(j);
+ r = stbi__decode_jpeg_header(j, STBI__SCAN_type);
+ stbi__rewind(s);
+ STBI_FREE(j);
+ return r;
+}
+
+static int stbi__jpeg_info_raw(stbi__jpeg *j, int *x, int *y, int *comp)
+{
+ if (!stbi__decode_jpeg_header(j, STBI__SCAN_header)) {
+ stbi__rewind( j->s );
+ return 0;
+ }
+ if (x) *x = j->s->img_x;
+ if (y) *y = j->s->img_y;
+ if (comp) *comp = j->s->img_n >= 3 ? 3 : 1;
+ return 1;
+}
+
+static int stbi__jpeg_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ int result;
+ stbi__jpeg* j = (stbi__jpeg*) (stbi__malloc(sizeof(stbi__jpeg)));
+ j->s = s;
+ result = stbi__jpeg_info_raw(j, x, y, comp);
+ STBI_FREE(j);
+ return result;
+}
+#endif
+
+// public domain zlib decode v0.2 Sean Barrett 2006-11-18
+// simple implementation
+// - all input must be provided in an upfront buffer
+// - all output is written to a single output buffer (can malloc/realloc)
+// performance
+// - fast huffman
+
+#ifndef STBI_NO_ZLIB
+
+// fast-way is faster to check than jpeg huffman, but slow way is slower
+#define STBI__ZFAST_BITS 9 // accelerate all cases in default tables
+#define STBI__ZFAST_MASK ((1 << STBI__ZFAST_BITS) - 1)
+
+// zlib-style huffman encoding
+// (jpegs packs from left, zlib from right, so can't share code)
+typedef struct
+{
+ stbi__uint16 fast[1 << STBI__ZFAST_BITS];
+ stbi__uint16 firstcode[16];
+ int maxcode[17];
+ stbi__uint16 firstsymbol[16];
+ stbi_uc size[288];
+ stbi__uint16 value[288];
+} stbi__zhuffman;
+
+stbi_inline static int stbi__bitreverse16(int n)
+{
+ n = ((n & 0xAAAA) >> 1) | ((n & 0x5555) << 1);
+ n = ((n & 0xCCCC) >> 2) | ((n & 0x3333) << 2);
+ n = ((n & 0xF0F0) >> 4) | ((n & 0x0F0F) << 4);
+ n = ((n & 0xFF00) >> 8) | ((n & 0x00FF) << 8);
+ return n;
+}
+
+stbi_inline static int stbi__bit_reverse(int v, int bits)
+{
+ STBI_ASSERT(bits <= 16);
+ // to bit reverse n bits, reverse 16 and shift
+ // e.g. 11 bits, bit reverse and shift away 5
+ return stbi__bitreverse16(v) >> (16-bits);
+}
+
+static int stbi__zbuild_huffman(stbi__zhuffman *z, const stbi_uc *sizelist, int num)
+{
+ int i,k=0;
+ int code, next_code[16], sizes[17];
+
+ // DEFLATE spec for generating codes
+ memset(sizes, 0, sizeof(sizes));
+ memset(z->fast, 0, sizeof(z->fast));
+ for (i=0; i < num; ++i)
+ ++sizes[sizelist[i]];
+ sizes[0] = 0;
+ for (i=1; i < 16; ++i)
+ if (sizes[i] > (1 << i))
+ return stbi__err("bad sizes", "Corrupt PNG");
+ code = 0;
+ for (i=1; i < 16; ++i) {
+ next_code[i] = code;
+ z->firstcode[i] = (stbi__uint16) code;
+ z->firstsymbol[i] = (stbi__uint16) k;
+ code = (code + sizes[i]);
+ if (sizes[i])
+ if (code-1 >= (1 << i)) return stbi__err("bad codelengths","Corrupt PNG");
+ z->maxcode[i] = code << (16-i); // preshift for inner loop
+ code <<= 1;
+ k += sizes[i];
+ }
+ z->maxcode[16] = 0x10000; // sentinel
+ for (i=0; i < num; ++i) {
+ int s = sizelist[i];
+ if (s) {
+ int c = next_code[s] - z->firstcode[s] + z->firstsymbol[s];
+ stbi__uint16 fastv = (stbi__uint16) ((s << 9) | i);
+ z->size [c] = (stbi_uc ) s;
+ z->value[c] = (stbi__uint16) i;
+ if (s <= STBI__ZFAST_BITS) {
+ int j = stbi__bit_reverse(next_code[s],s);
+ while (j < (1 << STBI__ZFAST_BITS)) {
+ z->fast[j] = fastv;
+ j += (1 << s);
+ }
+ }
+ ++next_code[s];
+ }
+ }
+ return 1;
+}
+
+// zlib-from-memory implementation for PNG reading
+// because PNG allows splitting the zlib stream arbitrarily,
+// and it's annoying structurally to have PNG call ZLIB call PNG,
+// we require PNG read all the IDATs and combine them into a single
+// memory buffer
+
+typedef struct
+{
+ stbi_uc *zbuffer, *zbuffer_end;
+ int num_bits;
+ stbi__uint32 code_buffer;
+
+ char *zout;
+ char *zout_start;
+ char *zout_end;
+ int z_expandable;
+
+ stbi__zhuffman z_length, z_distance;
+} stbi__zbuf;
+
+stbi_inline static stbi_uc stbi__zget8(stbi__zbuf *z)
+{
+ if (z->zbuffer >= z->zbuffer_end) return 0;
+ return *z->zbuffer++;
+}
+
+static void stbi__fill_bits(stbi__zbuf *z)
+{
+ do {
+ STBI_ASSERT(z->code_buffer < (1U << z->num_bits));
+ z->code_buffer |= (unsigned int) stbi__zget8(z) << z->num_bits;
+ z->num_bits += 8;
+ } while (z->num_bits <= 24);
+}
+
+stbi_inline static unsigned int stbi__zreceive(stbi__zbuf *z, int n)
+{
+ unsigned int k;
+ if (z->num_bits < n) stbi__fill_bits(z);
+ k = z->code_buffer & ((1 << n) - 1);
+ z->code_buffer >>= n;
+ z->num_bits -= n;
+ return k;
+}
+
+static int stbi__zhuffman_decode_slowpath(stbi__zbuf *a, stbi__zhuffman *z)
+{
+ int b,s,k;
+ // not resolved by fast table, so compute it the slow way
+ // use jpeg approach, which requires MSbits at top
+ k = stbi__bit_reverse(a->code_buffer, 16);
+ for (s=STBI__ZFAST_BITS+1; ; ++s)
+ if (k < z->maxcode[s])
+ break;
+ if (s == 16) return -1; // invalid code!
+ // code size is s, so:
+ b = (k >> (16-s)) - z->firstcode[s] + z->firstsymbol[s];
+ STBI_ASSERT(z->size[b] == s);
+ a->code_buffer >>= s;
+ a->num_bits -= s;
+ return z->value[b];
+}
+
+stbi_inline static int stbi__zhuffman_decode(stbi__zbuf *a, stbi__zhuffman *z)
+{
+ int b,s;
+ if (a->num_bits < 16) stbi__fill_bits(a);
+ b = z->fast[a->code_buffer & STBI__ZFAST_MASK];
+ if (b) {
+ s = b >> 9;
+ a->code_buffer >>= s;
+ a->num_bits -= s;
+ return b & 511;
+ }
+ return stbi__zhuffman_decode_slowpath(a, z);
+}
+
+static int stbi__zexpand(stbi__zbuf *z, char *zout, int n) // need to make room for n bytes
+{
+ char *q;
+ int cur, limit, old_limit;
+ z->zout = zout;
+ if (!z->z_expandable) return stbi__err("output buffer limit","Corrupt PNG");
+ cur = (int) (z->zout - z->zout_start);
+ limit = old_limit = (int) (z->zout_end - z->zout_start);
+ while (cur + n > limit)
+ limit *= 2;
+ q = (char *) STBI_REALLOC_SIZED(z->zout_start, old_limit, limit);
+ STBI_NOTUSED(old_limit);
+ if (q == NULL) return stbi__err("outofmem", "Out of memory");
+ z->zout_start = q;
+ z->zout = q + cur;
+ z->zout_end = q + limit;
+ return 1;
+}
+
+static int stbi__zlength_base[31] = {
+ 3,4,5,6,7,8,9,10,11,13,
+ 15,17,19,23,27,31,35,43,51,59,
+ 67,83,99,115,131,163,195,227,258,0,0 };
+
+static int stbi__zlength_extra[31]=
+{ 0,0,0,0,0,0,0,0,1,1,1,1,2,2,2,2,3,3,3,3,4,4,4,4,5,5,5,5,0,0,0 };
+
+static int stbi__zdist_base[32] = { 1,2,3,4,5,7,9,13,17,25,33,49,65,97,129,193,
+257,385,513,769,1025,1537,2049,3073,4097,6145,8193,12289,16385,24577,0,0};
+
+static int stbi__zdist_extra[32] =
+{ 0,0,0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,8,8,9,9,10,10,11,11,12,12,13,13};
+
+static int stbi__parse_huffman_block(stbi__zbuf *a)
+{
+ char *zout = a->zout;
+ for(;;) {
+ int z = stbi__zhuffman_decode(a, &a->z_length);
+ if (z < 256) {
+ if (z < 0) return stbi__err("bad huffman code","Corrupt PNG"); // error in huffman codes
+ if (zout >= a->zout_end) {
+ if (!stbi__zexpand(a, zout, 1)) return 0;
+ zout = a->zout;
+ }
+ *zout++ = (char) z;
+ } else {
+ stbi_uc *p;
+ int len,dist;
+ if (z == 256) {
+ a->zout = zout;
+ return 1;
+ }
+ z -= 257;
+ len = stbi__zlength_base[z];
+ if (stbi__zlength_extra[z]) len += stbi__zreceive(a, stbi__zlength_extra[z]);
+ z = stbi__zhuffman_decode(a, &a->z_distance);
+ if (z < 0) return stbi__err("bad huffman code","Corrupt PNG");
+ dist = stbi__zdist_base[z];
+ if (stbi__zdist_extra[z]) dist += stbi__zreceive(a, stbi__zdist_extra[z]);
+ if (zout - a->zout_start < dist) return stbi__err("bad dist","Corrupt PNG");
+ if (zout + len > a->zout_end) {
+ if (!stbi__zexpand(a, zout, len)) return 0;
+ zout = a->zout;
+ }
+ p = (stbi_uc *) (zout - dist);
+ if (dist == 1) { // run of one byte; common in images.
+ stbi_uc v = *p;
+ if (len) { do *zout++ = v; while (--len); }
+ } else {
+ if (len) { do *zout++ = *p++; while (--len); }
+ }
+ }
+ }
+}
+
+static int stbi__compute_huffman_codes(stbi__zbuf *a)
+{
+ static stbi_uc length_dezigzag[19] = { 16,17,18,0,8,7,9,6,10,5,11,4,12,3,13,2,14,1,15 };
+ stbi__zhuffman z_codelength;
+ stbi_uc lencodes[286+32+137];//padding for maximum single op
+ stbi_uc codelength_sizes[19];
+ int i,n;
+
+ int hlit = stbi__zreceive(a,5) + 257;
+ int hdist = stbi__zreceive(a,5) + 1;
+ int hclen = stbi__zreceive(a,4) + 4;
+ int ntot = hlit + hdist;
+
+ memset(codelength_sizes, 0, sizeof(codelength_sizes));
+ for (i=0; i < hclen; ++i) {
+ int s = stbi__zreceive(a,3);
+ codelength_sizes[length_dezigzag[i]] = (stbi_uc) s;
+ }
+ if (!stbi__zbuild_huffman(&z_codelength, codelength_sizes, 19)) return 0;
+
+ n = 0;
+ while (n < ntot) {
+ int c = stbi__zhuffman_decode(a, &z_codelength);
+ if (c < 0 || c >= 19) return stbi__err("bad codelengths", "Corrupt PNG");
+ if (c < 16)
+ lencodes[n++] = (stbi_uc) c;
+ else {
+ stbi_uc fill = 0;
+ if (c == 16) {
+ c = stbi__zreceive(a,2)+3;
+ if (n == 0) return stbi__err("bad codelengths", "Corrupt PNG");
+ fill = lencodes[n-1];
+ } else if (c == 17)
+ c = stbi__zreceive(a,3)+3;
+ else {
+ STBI_ASSERT(c == 18);
+ c = stbi__zreceive(a,7)+11;
+ }
+ if (ntot - n < c) return stbi__err("bad codelengths", "Corrupt PNG");
+ memset(lencodes+n, fill, c);
+ n += c;
+ }
+ }
+ if (n != ntot) return stbi__err("bad codelengths","Corrupt PNG");
+ if (!stbi__zbuild_huffman(&a->z_length, lencodes, hlit)) return 0;
+ if (!stbi__zbuild_huffman(&a->z_distance, lencodes+hlit, hdist)) return 0;
+ return 1;
+}
+
+static int stbi__parse_uncompressed_block(stbi__zbuf *a)
+{
+ stbi_uc header[4];
+ int len,nlen,k;
+ if (a->num_bits & 7)
+ stbi__zreceive(a, a->num_bits & 7); // discard
+ // drain the bit-packed data into header
+ k = 0;
+ while (a->num_bits > 0) {
+ header[k++] = (stbi_uc) (a->code_buffer & 255); // suppress MSVC run-time check
+ a->code_buffer >>= 8;
+ a->num_bits -= 8;
+ }
+ STBI_ASSERT(a->num_bits == 0);
+ // now fill header the normal way
+ while (k < 4)
+ header[k++] = stbi__zget8(a);
+ len = header[1] * 256 + header[0];
+ nlen = header[3] * 256 + header[2];
+ if (nlen != (len ^ 0xffff)) return stbi__err("zlib corrupt","Corrupt PNG");
+ if (a->zbuffer + len > a->zbuffer_end) return stbi__err("read past buffer","Corrupt PNG");
+ if (a->zout + len > a->zout_end)
+ if (!stbi__zexpand(a, a->zout, len)) return 0;
+ memcpy(a->zout, a->zbuffer, len);
+ a->zbuffer += len;
+ a->zout += len;
+ return 1;
+}
+
+static int stbi__parse_zlib_header(stbi__zbuf *a)
+{
+ int cmf = stbi__zget8(a);
+ int cm = cmf & 15;
+ /* int cinfo = cmf >> 4; */
+ int flg = stbi__zget8(a);
+ if ((cmf*256+flg) % 31 != 0) return stbi__err("bad zlib header","Corrupt PNG"); // zlib spec
+ if (flg & 32) return stbi__err("no preset dict","Corrupt PNG"); // preset dictionary not allowed in png
+ if (cm != 8) return stbi__err("bad compression","Corrupt PNG"); // DEFLATE required for png
+ // window = 1 << (8 + cinfo)... but who cares, we fully buffer output
+ return 1;
+}
+
+static const stbi_uc stbi__zdefault_length[288] =
+{
+ 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
+ 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
+ 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
+ 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
+ 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,
+ 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,
+ 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,
+ 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,
+ 7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7, 7,7,7,7,7,7,7,7,8,8,8,8,8,8,8,8
+};
+static const stbi_uc stbi__zdefault_distance[32] =
+{
+ 5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5
+};
+/*
+Init algorithm:
+{
+ int i; // use <= to match clearly with spec
+ for (i=0; i <= 143; ++i) stbi__zdefault_length[i] = 8;
+ for ( ; i <= 255; ++i) stbi__zdefault_length[i] = 9;
+ for ( ; i <= 279; ++i) stbi__zdefault_length[i] = 7;
+ for ( ; i <= 287; ++i) stbi__zdefault_length[i] = 8;
+
+ for (i=0; i <= 31; ++i) stbi__zdefault_distance[i] = 5;
+}
+*/
+
+static int stbi__parse_zlib(stbi__zbuf *a, int parse_header)
+{
+ int final, type;
+ if (parse_header)
+ if (!stbi__parse_zlib_header(a)) return 0;
+ a->num_bits = 0;
+ a->code_buffer = 0;
+ do {
+ final = stbi__zreceive(a,1);
+ type = stbi__zreceive(a,2);
+ if (type == 0) {
+ if (!stbi__parse_uncompressed_block(a)) return 0;
+ } else if (type == 3) {
+ return 0;
+ } else {
+ if (type == 1) {
+ // use fixed code lengths
+ if (!stbi__zbuild_huffman(&a->z_length , stbi__zdefault_length , 288)) return 0;
+ if (!stbi__zbuild_huffman(&a->z_distance, stbi__zdefault_distance, 32)) return 0;
+ } else {
+ if (!stbi__compute_huffman_codes(a)) return 0;
+ }
+ if (!stbi__parse_huffman_block(a)) return 0;
+ }
+ } while (!final);
+ return 1;
+}
+
+static int stbi__do_zlib(stbi__zbuf *a, char *obuf, int olen, int exp, int parse_header)
+{
+ a->zout_start = obuf;
+ a->zout = obuf;
+ a->zout_end = obuf + olen;
+ a->z_expandable = exp;
+
+ return stbi__parse_zlib(a, parse_header);
+}
+
+STBIDEF char *stbi_zlib_decode_malloc_guesssize(const char *buffer, int len, int initial_size, int *outlen)
+{
+ stbi__zbuf a;
+ char *p = (char *) stbi__malloc(initial_size);
+ if (p == NULL) return NULL;
+ a.zbuffer = (stbi_uc *) buffer;
+ a.zbuffer_end = (stbi_uc *) buffer + len;
+ if (stbi__do_zlib(&a, p, initial_size, 1, 1)) {
+ if (outlen) *outlen = (int) (a.zout - a.zout_start);
+ return a.zout_start;
+ } else {
+ STBI_FREE(a.zout_start);
+ return NULL;
+ }
+}
+
+STBIDEF char *stbi_zlib_decode_malloc(char const *buffer, int len, int *outlen)
+{
+ return stbi_zlib_decode_malloc_guesssize(buffer, len, 16384, outlen);
+}
+
+STBIDEF char *stbi_zlib_decode_malloc_guesssize_headerflag(const char *buffer, int len, int initial_size, int *outlen, int parse_header)
+{
+ stbi__zbuf a;
+ char *p = (char *) stbi__malloc(initial_size);
+ if (p == NULL) return NULL;
+ a.zbuffer = (stbi_uc *) buffer;
+ a.zbuffer_end = (stbi_uc *) buffer + len;
+ if (stbi__do_zlib(&a, p, initial_size, 1, parse_header)) {
+ if (outlen) *outlen = (int) (a.zout - a.zout_start);
+ return a.zout_start;
+ } else {
+ STBI_FREE(a.zout_start);
+ return NULL;
+ }
+}
+
+STBIDEF int stbi_zlib_decode_buffer(char *obuffer, int olen, char const *ibuffer, int ilen)
+{
+ stbi__zbuf a;
+ a.zbuffer = (stbi_uc *) ibuffer;
+ a.zbuffer_end = (stbi_uc *) ibuffer + ilen;
+ if (stbi__do_zlib(&a, obuffer, olen, 0, 1))
+ return (int) (a.zout - a.zout_start);
+ else
+ return -1;
+}
+
+STBIDEF char *stbi_zlib_decode_noheader_malloc(char const *buffer, int len, int *outlen)
+{
+ stbi__zbuf a;
+ char *p = (char *) stbi__malloc(16384);
+ if (p == NULL) return NULL;
+ a.zbuffer = (stbi_uc *) buffer;
+ a.zbuffer_end = (stbi_uc *) buffer+len;
+ if (stbi__do_zlib(&a, p, 16384, 1, 0)) {
+ if (outlen) *outlen = (int) (a.zout - a.zout_start);
+ return a.zout_start;
+ } else {
+ STBI_FREE(a.zout_start);
+ return NULL;
+ }
+}
+
+STBIDEF int stbi_zlib_decode_noheader_buffer(char *obuffer, int olen, const char *ibuffer, int ilen)
+{
+ stbi__zbuf a;
+ a.zbuffer = (stbi_uc *) ibuffer;
+ a.zbuffer_end = (stbi_uc *) ibuffer + ilen;
+ if (stbi__do_zlib(&a, obuffer, olen, 0, 0))
+ return (int) (a.zout - a.zout_start);
+ else
+ return -1;
+}
+#endif
+
+// public domain "baseline" PNG decoder v0.10 Sean Barrett 2006-11-18
+// simple implementation
+// - only 8-bit samples
+// - no CRC checking
+// - allocates lots of intermediate memory
+// - avoids problem of streaming data between subsystems
+// - avoids explicit window management
+// performance
+// - uses stb_zlib, a PD zlib implementation with fast huffman decoding
+
+#ifndef STBI_NO_PNG
+typedef struct
+{
+ stbi__uint32 length;
+ stbi__uint32 type;
+} stbi__pngchunk;
+
+static stbi__pngchunk stbi__get_chunk_header(stbi__context *s)
+{
+ stbi__pngchunk c;
+ c.length = stbi__get32be(s);
+ c.type = stbi__get32be(s);
+ return c;
+}
+
+static int stbi__check_png_header(stbi__context *s)
+{
+ static stbi_uc png_sig[8] = { 137,80,78,71,13,10,26,10 };
+ int i;
+ for (i=0; i < 8; ++i)
+ if (stbi__get8(s) != png_sig[i]) return stbi__err("bad png sig","Not a PNG");
+ return 1;
+}
+
+typedef struct
+{
+ stbi__context *s;
+ stbi_uc *idata, *expanded, *out;
+ int depth;
+} stbi__png;
+
+
+enum {
+ STBI__F_none=0,
+ STBI__F_sub=1,
+ STBI__F_up=2,
+ STBI__F_avg=3,
+ STBI__F_paeth=4,
+ // synthetic filters used for first scanline to avoid needing a dummy row of 0s
+ STBI__F_avg_first,
+ STBI__F_paeth_first
+};
+
+static stbi_uc first_row_filter[5] =
+{
+ STBI__F_none,
+ STBI__F_sub,
+ STBI__F_none,
+ STBI__F_avg_first,
+ STBI__F_paeth_first
+};
+
+static int stbi__paeth(int a, int b, int c)
+{
+ int p = a + b - c;
+ int pa = abs(p-a);
+ int pb = abs(p-b);
+ int pc = abs(p-c);
+ if (pa <= pb && pa <= pc) return a;
+ if (pb <= pc) return b;
+ return c;
+}
+
+static stbi_uc stbi__depth_scale_table[9] = { 0, 0xff, 0x55, 0, 0x11, 0,0,0, 0x01 };
+
+// create the png data from post-deflated data
+static int stbi__create_png_image_raw(stbi__png *a, stbi_uc *raw, stbi__uint32 raw_len, int out_n, stbi__uint32 x, stbi__uint32 y, int depth, int color)
+{
+ int bytes = (depth == 16? 2 : 1);
+ stbi__context *s = a->s;
+ stbi__uint32 i,j,stride = x*out_n*bytes;
+ stbi__uint32 img_len, img_width_bytes;
+ int k;
+ int img_n = s->img_n; // copy it into a local for later
+
+ int output_bytes = out_n*bytes;
+ int filter_bytes = img_n*bytes;
+ int width = x;
+
+ STBI_ASSERT(out_n == s->img_n || out_n == s->img_n+1);
+ a->out = (stbi_uc *) stbi__malloc_mad3(x, y, output_bytes, 0); // extra bytes to write off the end into
+ if (!a->out) return stbi__err("outofmem", "Out of memory");
+
+ img_width_bytes = (((img_n * x * depth) + 7) >> 3);
+ img_len = (img_width_bytes + 1) * y;
+ if (s->img_x == x && s->img_y == y) {
+ if (raw_len != img_len) return stbi__err("not enough pixels","Corrupt PNG");
+ } else { // interlaced:
+ if (raw_len < img_len) return stbi__err("not enough pixels","Corrupt PNG");
+ }
+
+ for (j=0; j < y; ++j) {
+ stbi_uc *cur = a->out + stride*j;
+ stbi_uc *prior;
+ int filter = *raw++;
+
+ if (filter > 4)
+ return stbi__err("invalid filter","Corrupt PNG");
+
+ if (depth < 8) {
+ STBI_ASSERT(img_width_bytes <= x);
+ cur += x*out_n - img_width_bytes; // store output to the rightmost img_len bytes, so we can decode in place
+ filter_bytes = 1;
+ width = img_width_bytes;
+ }
+ prior = cur - stride; // bugfix: need to compute this after 'cur +=' computation above
+
+ // if first row, use special filter that doesn't sample previous row
+ if (j == 0) filter = first_row_filter[filter];
+
+ // handle first byte explicitly
+ for (k=0; k < filter_bytes; ++k) {
+ switch (filter) {
+ case STBI__F_none : cur[k] = raw[k]; break;
+ case STBI__F_sub : cur[k] = raw[k]; break;
+ case STBI__F_up : cur[k] = STBI__BYTECAST(raw[k] + prior[k]); break;
+ case STBI__F_avg : cur[k] = STBI__BYTECAST(raw[k] + (prior[k]>>1)); break;
+ case STBI__F_paeth : cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(0,prior[k],0)); break;
+ case STBI__F_avg_first : cur[k] = raw[k]; break;
+ case STBI__F_paeth_first: cur[k] = raw[k]; break;
+ }
+ }
+
+ if (depth == 8) {
+ if (img_n != out_n)
+ cur[img_n] = 255; // first pixel
+ raw += img_n;
+ cur += out_n;
+ prior += out_n;
+ } else if (depth == 16) {
+ if (img_n != out_n) {
+ cur[filter_bytes] = 255; // first pixel top byte
+ cur[filter_bytes+1] = 255; // first pixel bottom byte
+ }
+ raw += filter_bytes;
+ cur += output_bytes;
+ prior += output_bytes;
+ } else {
+ raw += 1;
+ cur += 1;
+ prior += 1;
+ }
+
+ // this is a little gross, so that we don't switch per-pixel or per-component
+ if (depth < 8 || img_n == out_n) {
+ int nk = (width - 1)*filter_bytes;
+ #define STBI__CASE(f) \
+ case f: \
+ for (k=0; k < nk; ++k)
+ switch (filter) {
+ // "none" filter turns into a memcpy here; make that explicit.
+ case STBI__F_none: memcpy(cur, raw, nk); break;
+ STBI__CASE(STBI__F_sub) { cur[k] = STBI__BYTECAST(raw[k] + cur[k-filter_bytes]); } break;
+ STBI__CASE(STBI__F_up) { cur[k] = STBI__BYTECAST(raw[k] + prior[k]); } break;
+ STBI__CASE(STBI__F_avg) { cur[k] = STBI__BYTECAST(raw[k] + ((prior[k] + cur[k-filter_bytes])>>1)); } break;
+ STBI__CASE(STBI__F_paeth) { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k-filter_bytes],prior[k],prior[k-filter_bytes])); } break;
+ STBI__CASE(STBI__F_avg_first) { cur[k] = STBI__BYTECAST(raw[k] + (cur[k-filter_bytes] >> 1)); } break;
+ STBI__CASE(STBI__F_paeth_first) { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k-filter_bytes],0,0)); } break;
+ }
+ #undef STBI__CASE
+ raw += nk;
+ } else {
+ STBI_ASSERT(img_n+1 == out_n);
+ #define STBI__CASE(f) \
+ case f: \
+ for (i=x-1; i >= 1; --i, cur[filter_bytes]=255,raw+=filter_bytes,cur+=output_bytes,prior+=output_bytes) \
+ for (k=0; k < filter_bytes; ++k)
+ switch (filter) {
+ STBI__CASE(STBI__F_none) { cur[k] = raw[k]; } break;
+ STBI__CASE(STBI__F_sub) { cur[k] = STBI__BYTECAST(raw[k] + cur[k- output_bytes]); } break;
+ STBI__CASE(STBI__F_up) { cur[k] = STBI__BYTECAST(raw[k] + prior[k]); } break;
+ STBI__CASE(STBI__F_avg) { cur[k] = STBI__BYTECAST(raw[k] + ((prior[k] + cur[k- output_bytes])>>1)); } break;
+ STBI__CASE(STBI__F_paeth) { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k- output_bytes],prior[k],prior[k- output_bytes])); } break;
+ STBI__CASE(STBI__F_avg_first) { cur[k] = STBI__BYTECAST(raw[k] + (cur[k- output_bytes] >> 1)); } break;
+ STBI__CASE(STBI__F_paeth_first) { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k- output_bytes],0,0)); } break;
+ }
+ #undef STBI__CASE
+
+ // the loop above sets the high byte of the pixels' alpha, but for
+ // 16 bit png files we also need the low byte set. we'll do that here.
+ if (depth == 16) {
+ cur = a->out + stride*j; // start at the beginning of the row again
+ for (i=0; i < x; ++i,cur+=output_bytes) {
+ cur[filter_bytes+1] = 255;
+ }
+ }
+ }
+ }
+
+ // we make a separate pass to expand bits to pixels; for performance,
+ // this could run two scanlines behind the above code, so it won't
+ // intefere with filtering but will still be in the cache.
+ if (depth < 8) {
+ for (j=0; j < y; ++j) {
+ stbi_uc *cur = a->out + stride*j;
+ stbi_uc *in = a->out + stride*j + x*out_n - img_width_bytes;
+ // unpack 1/2/4-bit into a 8-bit buffer. allows us to keep the common 8-bit path optimal at minimal cost for 1/2/4-bit
+ // png guarante byte alignment, if width is not multiple of 8/4/2 we'll decode dummy trailing data that will be skipped in the later loop
+ stbi_uc scale = (color == 0) ? stbi__depth_scale_table[depth] : 1; // scale grayscale values to 0..255 range
+
+ // note that the final byte might overshoot and write more data than desired.
+ // we can allocate enough data that this never writes out of memory, but it
+ // could also overwrite the next scanline. can it overwrite non-empty data
+ // on the next scanline? yes, consider 1-pixel-wide scanlines with 1-bit-per-pixel.
+ // so we need to explicitly clamp the final ones
+
+ if (depth == 4) {
+ for (k=x*img_n; k >= 2; k-=2, ++in) {
+ *cur++ = scale * ((*in >> 4) );
+ *cur++ = scale * ((*in ) & 0x0f);
+ }
+ if (k > 0) *cur++ = scale * ((*in >> 4) );
+ } else if (depth == 2) {
+ for (k=x*img_n; k >= 4; k-=4, ++in) {
+ *cur++ = scale * ((*in >> 6) );
+ *cur++ = scale * ((*in >> 4) & 0x03);
+ *cur++ = scale * ((*in >> 2) & 0x03);
+ *cur++ = scale * ((*in ) & 0x03);
+ }
+ if (k > 0) *cur++ = scale * ((*in >> 6) );
+ if (k > 1) *cur++ = scale * ((*in >> 4) & 0x03);
+ if (k > 2) *cur++ = scale * ((*in >> 2) & 0x03);
+ } else if (depth == 1) {
+ for (k=x*img_n; k >= 8; k-=8, ++in) {
+ *cur++ = scale * ((*in >> 7) );
+ *cur++ = scale * ((*in >> 6) & 0x01);
+ *cur++ = scale * ((*in >> 5) & 0x01);
+ *cur++ = scale * ((*in >> 4) & 0x01);
+ *cur++ = scale * ((*in >> 3) & 0x01);
+ *cur++ = scale * ((*in >> 2) & 0x01);
+ *cur++ = scale * ((*in >> 1) & 0x01);
+ *cur++ = scale * ((*in ) & 0x01);
+ }
+ if (k > 0) *cur++ = scale * ((*in >> 7) );
+ if (k > 1) *cur++ = scale * ((*in >> 6) & 0x01);
+ if (k > 2) *cur++ = scale * ((*in >> 5) & 0x01);
+ if (k > 3) *cur++ = scale * ((*in >> 4) & 0x01);
+ if (k > 4) *cur++ = scale * ((*in >> 3) & 0x01);
+ if (k > 5) *cur++ = scale * ((*in >> 2) & 0x01);
+ if (k > 6) *cur++ = scale * ((*in >> 1) & 0x01);
+ }
+ if (img_n != out_n) {
+ int q;
+ // insert alpha = 255
+ cur = a->out + stride*j;
+ if (img_n == 1) {
+ for (q=x-1; q >= 0; --q) {
+ cur[q*2+1] = 255;
+ cur[q*2+0] = cur[q];
+ }
+ } else {
+ STBI_ASSERT(img_n == 3);
+ for (q=x-1; q >= 0; --q) {
+ cur[q*4+3] = 255;
+ cur[q*4+2] = cur[q*3+2];
+ cur[q*4+1] = cur[q*3+1];
+ cur[q*4+0] = cur[q*3+0];
+ }
+ }
+ }
+ }
+ } else if (depth == 16) {
+ // force the image data from big-endian to platform-native.
+ // this is done in a separate pass due to the decoding relying
+ // on the data being untouched, but could probably be done
+ // per-line during decode if care is taken.
+ stbi_uc *cur = a->out;
+ stbi__uint16 *cur16 = (stbi__uint16*)cur;
+
+ for(i=0; i < x*y*out_n; ++i,cur16++,cur+=2) {
+ *cur16 = (cur[0] << 8) | cur[1];
+ }
+ }
+
+ return 1;
+}
+
+static int stbi__create_png_image(stbi__png *a, stbi_uc *image_data, stbi__uint32 image_data_len, int out_n, int depth, int color, int interlaced)
+{
+ int bytes = (depth == 16 ? 2 : 1);
+ int out_bytes = out_n * bytes;
+ stbi_uc *final;
+ int p;
+ if (!interlaced)
+ return stbi__create_png_image_raw(a, image_data, image_data_len, out_n, a->s->img_x, a->s->img_y, depth, color);
+
+ // de-interlacing
+ final = (stbi_uc *) stbi__malloc_mad3(a->s->img_x, a->s->img_y, out_bytes, 0);
+ for (p=0; p < 7; ++p) {
+ int xorig[] = { 0,4,0,2,0,1,0 };
+ int yorig[] = { 0,0,4,0,2,0,1 };
+ int xspc[] = { 8,8,4,4,2,2,1 };
+ int yspc[] = { 8,8,8,4,4,2,2 };
+ int i,j,x,y;
+ // pass1_x[4] = 0, pass1_x[5] = 1, pass1_x[12] = 1
+ x = (a->s->img_x - xorig[p] + xspc[p]-1) / xspc[p];
+ y = (a->s->img_y - yorig[p] + yspc[p]-1) / yspc[p];
+ if (x && y) {
+ stbi__uint32 img_len = ((((a->s->img_n * x * depth) + 7) >> 3) + 1) * y;
+ if (!stbi__create_png_image_raw(a, image_data, image_data_len, out_n, x, y, depth, color)) {
+ STBI_FREE(final);
+ return 0;
+ }
+ for (j=0; j < y; ++j) {
+ for (i=0; i < x; ++i) {
+ int out_y = j*yspc[p]+yorig[p];
+ int out_x = i*xspc[p]+xorig[p];
+ memcpy(final + out_y*a->s->img_x*out_bytes + out_x*out_bytes,
+ a->out + (j*x+i)*out_bytes, out_bytes);
+ }
+ }
+ STBI_FREE(a->out);
+ image_data += img_len;
+ image_data_len -= img_len;
+ }
+ }
+ a->out = final;
+
+ return 1;
+}
+
+static int stbi__compute_transparency(stbi__png *z, stbi_uc tc[3], int out_n)
+{
+ stbi__context *s = z->s;
+ stbi__uint32 i, pixel_count = s->img_x * s->img_y;
+ stbi_uc *p = z->out;
+
+ // compute color-based transparency, assuming we've
+ // already got 255 as the alpha value in the output
+ STBI_ASSERT(out_n == 2 || out_n == 4);
+
+ if (out_n == 2) {
+ for (i=0; i < pixel_count; ++i) {
+ p[1] = (p[0] == tc[0] ? 0 : 255);
+ p += 2;
+ }
+ } else {
+ for (i=0; i < pixel_count; ++i) {
+ if (p[0] == tc[0] && p[1] == tc[1] && p[2] == tc[2])
+ p[3] = 0;
+ p += 4;
+ }
+ }
+ return 1;
+}
+
+static int stbi__compute_transparency16(stbi__png *z, stbi__uint16 tc[3], int out_n)
+{
+ stbi__context *s = z->s;
+ stbi__uint32 i, pixel_count = s->img_x * s->img_y;
+ stbi__uint16 *p = (stbi__uint16*) z->out;
+
+ // compute color-based transparency, assuming we've
+ // already got 65535 as the alpha value in the output
+ STBI_ASSERT(out_n == 2 || out_n == 4);
+
+ if (out_n == 2) {
+ for (i = 0; i < pixel_count; ++i) {
+ p[1] = (p[0] == tc[0] ? 0 : 65535);
+ p += 2;
+ }
+ } else {
+ for (i = 0; i < pixel_count; ++i) {
+ if (p[0] == tc[0] && p[1] == tc[1] && p[2] == tc[2])
+ p[3] = 0;
+ p += 4;
+ }
+ }
+ return 1;
+}
+
+static int stbi__expand_png_palette(stbi__png *a, stbi_uc *palette, int len, int pal_img_n)
+{
+ stbi__uint32 i, pixel_count = a->s->img_x * a->s->img_y;
+ stbi_uc *p, *temp_out, *orig = a->out;
+
+ p = (stbi_uc *) stbi__malloc_mad2(pixel_count, pal_img_n, 0);
+ if (p == NULL) return stbi__err("outofmem", "Out of memory");
+
+ // between here and free(out) below, exitting would leak
+ temp_out = p;
+
+ if (pal_img_n == 3) {
+ for (i=0; i < pixel_count; ++i) {
+ int n = orig[i]*4;
+ p[0] = palette[n ];
+ p[1] = palette[n+1];
+ p[2] = palette[n+2];
+ p += 3;
+ }
+ } else {
+ for (i=0; i < pixel_count; ++i) {
+ int n = orig[i]*4;
+ p[0] = palette[n ];
+ p[1] = palette[n+1];
+ p[2] = palette[n+2];
+ p[3] = palette[n+3];
+ p += 4;
+ }
+ }
+ STBI_FREE(a->out);
+ a->out = temp_out;
+
+ STBI_NOTUSED(len);
+
+ return 1;
+}
+
+static int stbi__unpremultiply_on_load = 0;
+static int stbi__de_iphone_flag = 0;
+
+STBIDEF void stbi_set_unpremultiply_on_load(int flag_true_if_should_unpremultiply)
+{
+ stbi__unpremultiply_on_load = flag_true_if_should_unpremultiply;
+}
+
+STBIDEF void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert)
+{
+ stbi__de_iphone_flag = flag_true_if_should_convert;
+}
+
+static void stbi__de_iphone(stbi__png *z)
+{
+ stbi__context *s = z->s;
+ stbi__uint32 i, pixel_count = s->img_x * s->img_y;
+ stbi_uc *p = z->out;
+
+ if (s->img_out_n == 3) { // convert bgr to rgb
+ for (i=0; i < pixel_count; ++i) {
+ stbi_uc t = p[0];
+ p[0] = p[2];
+ p[2] = t;
+ p += 3;
+ }
+ } else {
+ STBI_ASSERT(s->img_out_n == 4);
+ if (stbi__unpremultiply_on_load) {
+ // convert bgr to rgb and unpremultiply
+ for (i=0; i < pixel_count; ++i) {
+ stbi_uc a = p[3];
+ stbi_uc t = p[0];
+ if (a) {
+ p[0] = p[2] * 255 / a;
+ p[1] = p[1] * 255 / a;
+ p[2] = t * 255 / a;
+ } else {
+ p[0] = p[2];
+ p[2] = t;
+ }
+ p += 4;
+ }
+ } else {
+ // convert bgr to rgb
+ for (i=0; i < pixel_count; ++i) {
+ stbi_uc t = p[0];
+ p[0] = p[2];
+ p[2] = t;
+ p += 4;
+ }
+ }
+ }
+}
+
+#define STBI__PNG_TYPE(a,b,c,d) (((a) << 24) + ((b) << 16) + ((c) << 8) + (d))
+
+static int stbi__parse_png_file(stbi__png *z, int scan, int req_comp)
+{
+ stbi_uc palette[1024], pal_img_n=0;
+ stbi_uc has_trans=0, tc[3];
+ stbi__uint16 tc16[3];
+ stbi__uint32 ioff=0, idata_limit=0, i, pal_len=0;
+ int first=1,k,interlace=0, color=0, is_iphone=0;
+ stbi__context *s = z->s;
+
+ z->expanded = NULL;
+ z->idata = NULL;
+ z->out = NULL;
+
+ if (!stbi__check_png_header(s)) return 0;
+
+ if (scan == STBI__SCAN_type) return 1;
+
+ for (;;) {
+ stbi__pngchunk c = stbi__get_chunk_header(s);
+ switch (c.type) {
+ case STBI__PNG_TYPE('C','g','B','I'):
+ is_iphone = 1;
+ stbi__skip(s, c.length);
+ break;
+ case STBI__PNG_TYPE('I','H','D','R'): {
+ int comp,filter;
+ if (!first) return stbi__err("multiple IHDR","Corrupt PNG");
+ first = 0;
+ if (c.length != 13) return stbi__err("bad IHDR len","Corrupt PNG");
+ s->img_x = stbi__get32be(s); if (s->img_x > (1 << 24)) return stbi__err("too large","Very large image (corrupt?)");
+ s->img_y = stbi__get32be(s); if (s->img_y > (1 << 24)) return stbi__err("too large","Very large image (corrupt?)");
+ z->depth = stbi__get8(s); if (z->depth != 1 && z->depth != 2 && z->depth != 4 && z->depth != 8 && z->depth != 16) return stbi__err("1/2/4/8/16-bit only","PNG not supported: 1/2/4/8/16-bit only");
+ color = stbi__get8(s); if (color > 6) return stbi__err("bad ctype","Corrupt PNG");
+ if (color == 3 && z->depth == 16) return stbi__err("bad ctype","Corrupt PNG");
+ if (color == 3) pal_img_n = 3; else if (color & 1) return stbi__err("bad ctype","Corrupt PNG");
+ comp = stbi__get8(s); if (comp) return stbi__err("bad comp method","Corrupt PNG");
+ filter= stbi__get8(s); if (filter) return stbi__err("bad filter method","Corrupt PNG");
+ interlace = stbi__get8(s); if (interlace>1) return stbi__err("bad interlace method","Corrupt PNG");
+ if (!s->img_x || !s->img_y) return stbi__err("0-pixel image","Corrupt PNG");
+ if (!pal_img_n) {
+ s->img_n = (color & 2 ? 3 : 1) + (color & 4 ? 1 : 0);
+ if ((1 << 30) / s->img_x / s->img_n < s->img_y) return stbi__err("too large", "Image too large to decode");
+ if (scan == STBI__SCAN_header) return 1;
+ } else {
+ // if paletted, then pal_n is our final components, and
+ // img_n is # components to decompress/filter.
+ s->img_n = 1;
+ if ((1 << 30) / s->img_x / 4 < s->img_y) return stbi__err("too large","Corrupt PNG");
+ // if SCAN_header, have to scan to see if we have a tRNS
+ }
+ break;
+ }
+
+ case STBI__PNG_TYPE('P','L','T','E'): {
+ if (first) return stbi__err("first not IHDR", "Corrupt PNG");
+ if (c.length > 256*3) return stbi__err("invalid PLTE","Corrupt PNG");
+ pal_len = c.length / 3;
+ if (pal_len * 3 != c.length) return stbi__err("invalid PLTE","Corrupt PNG");
+ for (i=0; i < pal_len; ++i) {
+ palette[i*4+0] = stbi__get8(s);
+ palette[i*4+1] = stbi__get8(s);
+ palette[i*4+2] = stbi__get8(s);
+ palette[i*4+3] = 255;
+ }
+ break;
+ }
+
+ case STBI__PNG_TYPE('t','R','N','S'): {
+ if (first) return stbi__err("first not IHDR", "Corrupt PNG");
+ if (z->idata) return stbi__err("tRNS after IDAT","Corrupt PNG");
+ if (pal_img_n) {
+ if (scan == STBI__SCAN_header) { s->img_n = 4; return 1; }
+ if (pal_len == 0) return stbi__err("tRNS before PLTE","Corrupt PNG");
+ if (c.length > pal_len) return stbi__err("bad tRNS len","Corrupt PNG");
+ pal_img_n = 4;
+ for (i=0; i < c.length; ++i)
+ palette[i*4+3] = stbi__get8(s);
+ } else {
+ if (!(s->img_n & 1)) return stbi__err("tRNS with alpha","Corrupt PNG");
+ if (c.length != (stbi__uint32) s->img_n*2) return stbi__err("bad tRNS len","Corrupt PNG");
+ has_trans = 1;
+ if (z->depth == 16) {
+ for (k = 0; k < s->img_n; ++k) tc16[k] = (stbi__uint16)stbi__get16be(s); // copy the values as-is
+ } else {
+ for (k = 0; k < s->img_n; ++k) tc[k] = (stbi_uc)(stbi__get16be(s) & 255) * stbi__depth_scale_table[z->depth]; // non 8-bit images will be larger
+ }
+ }
+ break;
+ }
+
+ case STBI__PNG_TYPE('I','D','A','T'): {
+ if (first) return stbi__err("first not IHDR", "Corrupt PNG");
+ if (pal_img_n && !pal_len) return stbi__err("no PLTE","Corrupt PNG");
+ if (scan == STBI__SCAN_header) { s->img_n = pal_img_n; return 1; }
+ if ((int)(ioff + c.length) < (int)ioff) return 0;
+ if (ioff + c.length > idata_limit) {
+ stbi__uint32 idata_limit_old = idata_limit;
+ stbi_uc *p;
+ if (idata_limit == 0) idata_limit = c.length > 4096 ? c.length : 4096;
+ while (ioff + c.length > idata_limit)
+ idata_limit *= 2;
+ STBI_NOTUSED(idata_limit_old);
+ p = (stbi_uc *) STBI_REALLOC_SIZED(z->idata, idata_limit_old, idata_limit); if (p == NULL) return stbi__err("outofmem", "Out of memory");
+ z->idata = p;
+ }
+ if (!stbi__getn(s, z->idata+ioff,c.length)) return stbi__err("outofdata","Corrupt PNG");
+ ioff += c.length;
+ break;
+ }
+
+ case STBI__PNG_TYPE('I','E','N','D'): {
+ stbi__uint32 raw_len, bpl;
+ if (first) return stbi__err("first not IHDR", "Corrupt PNG");
+ if (scan != STBI__SCAN_load) return 1;
+ if (z->idata == NULL) return stbi__err("no IDAT","Corrupt PNG");
+ // initial guess for decoded data size to avoid unnecessary reallocs
+ bpl = (s->img_x * z->depth + 7) / 8; // bytes per line, per component
+ raw_len = bpl * s->img_y * s->img_n /* pixels */ + s->img_y /* filter mode per row */;
+ z->expanded = (stbi_uc *) stbi_zlib_decode_malloc_guesssize_headerflag((char *) z->idata, ioff, raw_len, (int *) &raw_len, !is_iphone);
+ if (z->expanded == NULL) return 0; // zlib should set error
+ STBI_FREE(z->idata); z->idata = NULL;
+ if ((req_comp == s->img_n+1 && req_comp != 3 && !pal_img_n) || has_trans)
+ s->img_out_n = s->img_n+1;
+ else
+ s->img_out_n = s->img_n;
+ if (!stbi__create_png_image(z, z->expanded, raw_len, s->img_out_n, z->depth, color, interlace)) return 0;
+ if (has_trans) {
+ if (z->depth == 16) {
+ if (!stbi__compute_transparency16(z, tc16, s->img_out_n)) return 0;
+ } else {
+ if (!stbi__compute_transparency(z, tc, s->img_out_n)) return 0;
+ }
+ }
+ if (is_iphone && stbi__de_iphone_flag && s->img_out_n > 2)
+ stbi__de_iphone(z);
+ if (pal_img_n) {
+ // pal_img_n == 3 or 4
+ s->img_n = pal_img_n; // record the actual colors we had
+ s->img_out_n = pal_img_n;
+ if (req_comp >= 3) s->img_out_n = req_comp;
+ if (!stbi__expand_png_palette(z, palette, pal_len, s->img_out_n))
+ return 0;
+ }
+ STBI_FREE(z->expanded); z->expanded = NULL;
+ return 1;
+ }
+
+ default:
+ // if critical, fail
+ if (first) return stbi__err("first not IHDR", "Corrupt PNG");
+ if ((c.type & (1 << 29)) == 0) {
+ #ifndef STBI_NO_FAILURE_STRINGS
+ // not threadsafe
+ static char invalid_chunk[] = "XXXX PNG chunk not known";
+ invalid_chunk[0] = STBI__BYTECAST(c.type >> 24);
+ invalid_chunk[1] = STBI__BYTECAST(c.type >> 16);
+ invalid_chunk[2] = STBI__BYTECAST(c.type >> 8);
+ invalid_chunk[3] = STBI__BYTECAST(c.type >> 0);
+ #endif
+ return stbi__err(invalid_chunk, "PNG not supported: unknown PNG chunk type");
+ }
+ stbi__skip(s, c.length);
+ break;
+ }
+ // end of PNG chunk, read and skip CRC
+ stbi__get32be(s);
+ }
+}
+
+static void *stbi__do_png(stbi__png *p, int *x, int *y, int *n, int req_comp, stbi__result_info *ri)
+{
+ void *result=NULL;
+ if (req_comp < 0 || req_comp > 4) return stbi__errpuc("bad req_comp", "Internal error");
+ if (stbi__parse_png_file(p, STBI__SCAN_load, req_comp)) {
+ if (p->depth < 8)
+ ri->bits_per_channel = 8;
+ else
+ ri->bits_per_channel = p->depth;
+ result = p->out;
+ p->out = NULL;
+ if (req_comp && req_comp != p->s->img_out_n) {
+ if (ri->bits_per_channel == 8)
+ result = stbi__convert_format((unsigned char *) result, p->s->img_out_n, req_comp, p->s->img_x, p->s->img_y);
+ else
+ result = stbi__convert_format16((stbi__uint16 *) result, p->s->img_out_n, req_comp, p->s->img_x, p->s->img_y);
+ p->s->img_out_n = req_comp;
+ if (result == NULL) return result;
+ }
+ *x = p->s->img_x;
+ *y = p->s->img_y;
+ if (n) *n = p->s->img_n;
+ }
+ STBI_FREE(p->out); p->out = NULL;
+ STBI_FREE(p->expanded); p->expanded = NULL;
+ STBI_FREE(p->idata); p->idata = NULL;
+
+ return result;
+}
+
+static void *stbi__png_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+ stbi__png p;
+ p.s = s;
+ return stbi__do_png(&p, x,y,comp,req_comp, ri);
+}
+
+static int stbi__png_test(stbi__context *s)
+{
+ int r;
+ r = stbi__check_png_header(s);
+ stbi__rewind(s);
+ return r;
+}
+
+static int stbi__png_info_raw(stbi__png *p, int *x, int *y, int *comp)
+{
+ if (!stbi__parse_png_file(p, STBI__SCAN_header, 0)) {
+ stbi__rewind( p->s );
+ return 0;
+ }
+ if (x) *x = p->s->img_x;
+ if (y) *y = p->s->img_y;
+ if (comp) *comp = p->s->img_n;
+ return 1;
+}
+
+static int stbi__png_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ stbi__png p;
+ p.s = s;
+ return stbi__png_info_raw(&p, x, y, comp);
+}
+#endif
+
+// Microsoft/Windows BMP image
+
+#ifndef STBI_NO_BMP
+static int stbi__bmp_test_raw(stbi__context *s)
+{
+ int r;
+ int sz;
+ if (stbi__get8(s) != 'B') return 0;
+ if (stbi__get8(s) != 'M') return 0;
+ stbi__get32le(s); // discard filesize
+ stbi__get16le(s); // discard reserved
+ stbi__get16le(s); // discard reserved
+ stbi__get32le(s); // discard data offset
+ sz = stbi__get32le(s);
+ r = (sz == 12 || sz == 40 || sz == 56 || sz == 108 || sz == 124);
+ return r;
+}
+
+static int stbi__bmp_test(stbi__context *s)
+{
+ int r = stbi__bmp_test_raw(s);
+ stbi__rewind(s);
+ return r;
+}
+
+
+// returns 0..31 for the highest set bit
+static int stbi__high_bit(unsigned int z)
+{
+ int n=0;
+ if (z == 0) return -1;
+ if (z >= 0x10000) n += 16, z >>= 16;
+ if (z >= 0x00100) n += 8, z >>= 8;
+ if (z >= 0x00010) n += 4, z >>= 4;
+ if (z >= 0x00004) n += 2, z >>= 2;
+ if (z >= 0x00002) n += 1, z >>= 1;
+ return n;
+}
+
+static int stbi__bitcount(unsigned int a)
+{
+ a = (a & 0x55555555) + ((a >> 1) & 0x55555555); // max 2
+ a = (a & 0x33333333) + ((a >> 2) & 0x33333333); // max 4
+ a = (a + (a >> 4)) & 0x0f0f0f0f; // max 8 per 4, now 8 bits
+ a = (a + (a >> 8)); // max 16 per 8 bits
+ a = (a + (a >> 16)); // max 32 per 8 bits
+ return a & 0xff;
+}
+
+static int stbi__shiftsigned(int v, int shift, int bits)
+{
+ int result;
+ int z=0;
+
+ if (shift < 0) v <<= -shift;
+ else v >>= shift;
+ result = v;
+
+ z = bits;
+ while (z < 8) {
+ result += v >> z;
+ z += bits;
+ }
+ return result;
+}
+
+typedef struct
+{
+ int bpp, offset, hsz;
+ unsigned int mr,mg,mb,ma, all_a;
+} stbi__bmp_data;
+
+static void *stbi__bmp_parse_header(stbi__context *s, stbi__bmp_data *info)
+{
+ int hsz;
+ if (stbi__get8(s) != 'B' || stbi__get8(s) != 'M') return stbi__errpuc("not BMP", "Corrupt BMP");
+ stbi__get32le(s); // discard filesize
+ stbi__get16le(s); // discard reserved
+ stbi__get16le(s); // discard reserved
+ info->offset = stbi__get32le(s);
+ info->hsz = hsz = stbi__get32le(s);
+ info->mr = info->mg = info->mb = info->ma = 0;
+
+ if (hsz != 12 && hsz != 40 && hsz != 56 && hsz != 108 && hsz != 124) return stbi__errpuc("unknown BMP", "BMP type not supported: unknown");
+ if (hsz == 12) {
+ s->img_x = stbi__get16le(s);
+ s->img_y = stbi__get16le(s);
+ } else {
+ s->img_x = stbi__get32le(s);
+ s->img_y = stbi__get32le(s);
+ }
+ if (stbi__get16le(s) != 1) return stbi__errpuc("bad BMP", "bad BMP");
+ info->bpp = stbi__get16le(s);
+ if (info->bpp == 1) return stbi__errpuc("monochrome", "BMP type not supported: 1-bit");
+ if (hsz != 12) {
+ int compress = stbi__get32le(s);
+ if (compress == 1 || compress == 2) return stbi__errpuc("BMP RLE", "BMP type not supported: RLE");
+ stbi__get32le(s); // discard sizeof
+ stbi__get32le(s); // discard hres
+ stbi__get32le(s); // discard vres
+ stbi__get32le(s); // discard colorsused
+ stbi__get32le(s); // discard max important
+ if (hsz == 40 || hsz == 56) {
+ if (hsz == 56) {
+ stbi__get32le(s);
+ stbi__get32le(s);
+ stbi__get32le(s);
+ stbi__get32le(s);
+ }
+ if (info->bpp == 16 || info->bpp == 32) {
+ if (compress == 0) {
+ if (info->bpp == 32) {
+ info->mr = 0xffu << 16;
+ info->mg = 0xffu << 8;
+ info->mb = 0xffu << 0;
+ info->ma = 0xffu << 24;
+ info->all_a = 0; // if all_a is 0 at end, then we loaded alpha channel but it was all 0
+ } else {
+ info->mr = 31u << 10;
+ info->mg = 31u << 5;
+ info->mb = 31u << 0;
+ }
+ } else if (compress == 3) {
+ info->mr = stbi__get32le(s);
+ info->mg = stbi__get32le(s);
+ info->mb = stbi__get32le(s);
+ // not documented, but generated by photoshop and handled by mspaint
+ if (info->mr == info->mg && info->mg == info->mb) {
+ // ?!?!?
+ return stbi__errpuc("bad BMP", "bad BMP");
+ }
+ } else
+ return stbi__errpuc("bad BMP", "bad BMP");
+ }
+ } else {
+ int i;
+ if (hsz != 108 && hsz != 124)
+ return stbi__errpuc("bad BMP", "bad BMP");
+ info->mr = stbi__get32le(s);
+ info->mg = stbi__get32le(s);
+ info->mb = stbi__get32le(s);
+ info->ma = stbi__get32le(s);
+ stbi__get32le(s); // discard color space
+ for (i=0; i < 12; ++i)
+ stbi__get32le(s); // discard color space parameters
+ if (hsz == 124) {
+ stbi__get32le(s); // discard rendering intent
+ stbi__get32le(s); // discard offset of profile data
+ stbi__get32le(s); // discard size of profile data
+ stbi__get32le(s); // discard reserved
+ }
+ }
+ }
+ return (void *) 1;
+}
+
+
+static void *stbi__bmp_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+ stbi_uc *out;
+ unsigned int mr=0,mg=0,mb=0,ma=0, all_a;
+ stbi_uc pal[256][4];
+ int psize=0,i,j,width;
+ int flip_vertically, pad, target;
+ stbi__bmp_data info;
+ STBI_NOTUSED(ri);
+
+ info.all_a = 255;
+ if (stbi__bmp_parse_header(s, &info) == NULL)
+ return NULL; // error code already set
+
+ flip_vertically = ((int) s->img_y) > 0;
+ s->img_y = abs((int) s->img_y);
+
+ mr = info.mr;
+ mg = info.mg;
+ mb = info.mb;
+ ma = info.ma;
+ all_a = info.all_a;
+
+ if (info.hsz == 12) {
+ if (info.bpp < 24)
+ psize = (info.offset - 14 - 24) / 3;
+ } else {
+ if (info.bpp < 16)
+ psize = (info.offset - 14 - info.hsz) >> 2;
+ }
+
+ s->img_n = ma ? 4 : 3;
+ if (req_comp && req_comp >= 3) // we can directly decode 3 or 4
+ target = req_comp;
+ else
+ target = s->img_n; // if they want monochrome, we'll post-convert
+
+ // sanity-check size
+ if (!stbi__mad3sizes_valid(target, s->img_x, s->img_y, 0))
+ return stbi__errpuc("too large", "Corrupt BMP");
+
+ out = (stbi_uc *) stbi__malloc_mad3(target, s->img_x, s->img_y, 0);
+ if (!out) return stbi__errpuc("outofmem", "Out of memory");
+ if (info.bpp < 16) {
+ int z=0;
+ if (psize == 0 || psize > 256) { STBI_FREE(out); return stbi__errpuc("invalid", "Corrupt BMP"); }
+ for (i=0; i < psize; ++i) {
+ pal[i][2] = stbi__get8(s);
+ pal[i][1] = stbi__get8(s);
+ pal[i][0] = stbi__get8(s);
+ if (info.hsz != 12) stbi__get8(s);
+ pal[i][3] = 255;
+ }
+ stbi__skip(s, info.offset - 14 - info.hsz - psize * (info.hsz == 12 ? 3 : 4));
+ if (info.bpp == 4) width = (s->img_x + 1) >> 1;
+ else if (info.bpp == 8) width = s->img_x;
+ else { STBI_FREE(out); return stbi__errpuc("bad bpp", "Corrupt BMP"); }
+ pad = (-width)&3;
+ for (j=0; j < (int) s->img_y; ++j) {
+ for (i=0; i < (int) s->img_x; i += 2) {
+ int v=stbi__get8(s),v2=0;
+ if (info.bpp == 4) {
+ v2 = v & 15;
+ v >>= 4;
+ }
+ out[z++] = pal[v][0];
+ out[z++] = pal[v][1];
+ out[z++] = pal[v][2];
+ if (target == 4) out[z++] = 255;
+ if (i+1 == (int) s->img_x) break;
+ v = (info.bpp == 8) ? stbi__get8(s) : v2;
+ out[z++] = pal[v][0];
+ out[z++] = pal[v][1];
+ out[z++] = pal[v][2];
+ if (target == 4) out[z++] = 255;
+ }
+ stbi__skip(s, pad);
+ }
+ } else {
+ int rshift=0,gshift=0,bshift=0,ashift=0,rcount=0,gcount=0,bcount=0,acount=0;
+ int z = 0;
+ int easy=0;
+ stbi__skip(s, info.offset - 14 - info.hsz);
+ if (info.bpp == 24) width = 3 * s->img_x;
+ else if (info.bpp == 16) width = 2*s->img_x;
+ else /* bpp = 32 and pad = 0 */ width=0;
+ pad = (-width) & 3;
+ if (info.bpp == 24) {
+ easy = 1;
+ } else if (info.bpp == 32) {
+ if (mb == 0xff && mg == 0xff00 && mr == 0x00ff0000 && ma == 0xff000000)
+ easy = 2;
+ }
+ if (!easy) {
+ if (!mr || !mg || !mb) { STBI_FREE(out); return stbi__errpuc("bad masks", "Corrupt BMP"); }
+ // right shift amt to put high bit in position #7
+ rshift = stbi__high_bit(mr)-7; rcount = stbi__bitcount(mr);
+ gshift = stbi__high_bit(mg)-7; gcount = stbi__bitcount(mg);
+ bshift = stbi__high_bit(mb)-7; bcount = stbi__bitcount(mb);
+ ashift = stbi__high_bit(ma)-7; acount = stbi__bitcount(ma);
+ }
+ for (j=0; j < (int) s->img_y; ++j) {
+ if (easy) {
+ for (i=0; i < (int) s->img_x; ++i) {
+ unsigned char a;
+ out[z+2] = stbi__get8(s);
+ out[z+1] = stbi__get8(s);
+ out[z+0] = stbi__get8(s);
+ z += 3;
+ a = (easy == 2 ? stbi__get8(s) : 255);
+ all_a |= a;
+ if (target == 4) out[z++] = a;
+ }
+ } else {
+ int bpp = info.bpp;
+ for (i=0; i < (int) s->img_x; ++i) {
+ stbi__uint32 v = (bpp == 16 ? (stbi__uint32) stbi__get16le(s) : stbi__get32le(s));
+ int a;
+ out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mr, rshift, rcount));
+ out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mg, gshift, gcount));
+ out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mb, bshift, bcount));
+ a = (ma ? stbi__shiftsigned(v & ma, ashift, acount) : 255);
+ all_a |= a;
+ if (target == 4) out[z++] = STBI__BYTECAST(a);
+ }
+ }
+ stbi__skip(s, pad);
+ }
+ }
+
+ // if alpha channel is all 0s, replace with all 255s
+ if (target == 4 && all_a == 0)
+ for (i=4*s->img_x*s->img_y-1; i >= 0; i -= 4)
+ out[i] = 255;
+
+ if (flip_vertically) {
+ stbi_uc t;
+ for (j=0; j < (int) s->img_y>>1; ++j) {
+ stbi_uc *p1 = out + j *s->img_x*target;
+ stbi_uc *p2 = out + (s->img_y-1-j)*s->img_x*target;
+ for (i=0; i < (int) s->img_x*target; ++i) {
+ t = p1[i], p1[i] = p2[i], p2[i] = t;
+ }
+ }
+ }
+
+ if (req_comp && req_comp != target) {
+ out = stbi__convert_format(out, target, req_comp, s->img_x, s->img_y);
+ if (out == NULL) return out; // stbi__convert_format frees input on failure
+ }
+
+ *x = s->img_x;
+ *y = s->img_y;
+ if (comp) *comp = s->img_n;
+ return out;
+}
+#endif
+
+// Targa Truevision - TGA
+// by Jonathan Dummer
+#ifndef STBI_NO_TGA
+// returns STBI_rgb or whatever, 0 on error
+static int stbi__tga_get_comp(int bits_per_pixel, int is_grey, int* is_rgb16)
+{
+ // only RGB or RGBA (incl. 16bit) or grey allowed
+ if(is_rgb16) *is_rgb16 = 0;
+ switch(bits_per_pixel) {
+ case 8: return STBI_grey;
+ case 16: if(is_grey) return STBI_grey_alpha;
+ // else: fall-through
+ case 15: if(is_rgb16) *is_rgb16 = 1;
+ return STBI_rgb;
+ case 24: // fall-through
+ case 32: return bits_per_pixel/8;
+ default: return 0;
+ }
+}
+
+static int stbi__tga_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ int tga_w, tga_h, tga_comp, tga_image_type, tga_bits_per_pixel, tga_colormap_bpp;
+ int sz, tga_colormap_type;
+ stbi__get8(s); // discard Offset
+ tga_colormap_type = stbi__get8(s); // colormap type
+ if( tga_colormap_type > 1 ) {
+ stbi__rewind(s);
+ return 0; // only RGB or indexed allowed
+ }
+ tga_image_type = stbi__get8(s); // image type
+ if ( tga_colormap_type == 1 ) { // colormapped (paletted) image
+ if (tga_image_type != 1 && tga_image_type != 9) {
+ stbi__rewind(s);
+ return 0;
+ }
+ stbi__skip(s,4); // skip index of first colormap entry and number of entries
+ sz = stbi__get8(s); // check bits per palette color entry
+ if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) {
+ stbi__rewind(s);
+ return 0;
+ }
+ stbi__skip(s,4); // skip image x and y origin
+ tga_colormap_bpp = sz;
+ } else { // "normal" image w/o colormap - only RGB or grey allowed, +/- RLE
+ if ( (tga_image_type != 2) && (tga_image_type != 3) && (tga_image_type != 10) && (tga_image_type != 11) ) {
+ stbi__rewind(s);
+ return 0; // only RGB or grey allowed, +/- RLE
+ }
+ stbi__skip(s,9); // skip colormap specification and image x/y origin
+ tga_colormap_bpp = 0;
+ }
+ tga_w = stbi__get16le(s);
+ if( tga_w < 1 ) {
+ stbi__rewind(s);
+ return 0; // test width
+ }
+ tga_h = stbi__get16le(s);
+ if( tga_h < 1 ) {
+ stbi__rewind(s);
+ return 0; // test height
+ }
+ tga_bits_per_pixel = stbi__get8(s); // bits per pixel
+ stbi__get8(s); // ignore alpha bits
+ if (tga_colormap_bpp != 0) {
+ if((tga_bits_per_pixel != 8) && (tga_bits_per_pixel != 16)) {
+ // when using a colormap, tga_bits_per_pixel is the size of the indexes
+ // I don't think anything but 8 or 16bit indexes makes sense
+ stbi__rewind(s);
+ return 0;
+ }
+ tga_comp = stbi__tga_get_comp(tga_colormap_bpp, 0, NULL);
+ } else {
+ tga_comp = stbi__tga_get_comp(tga_bits_per_pixel, (tga_image_type == 3) || (tga_image_type == 11), NULL);
+ }
+ if(!tga_comp) {
+ stbi__rewind(s);
+ return 0;
+ }
+ if (x) *x = tga_w;
+ if (y) *y = tga_h;
+ if (comp) *comp = tga_comp;
+ return 1; // seems to have passed everything
+}
+
+static int stbi__tga_test(stbi__context *s)
+{
+ int res = 0;
+ int sz, tga_color_type;
+ stbi__get8(s); // discard Offset
+ tga_color_type = stbi__get8(s); // color type
+ if ( tga_color_type > 1 ) goto errorEnd; // only RGB or indexed allowed
+ sz = stbi__get8(s); // image type
+ if ( tga_color_type == 1 ) { // colormapped (paletted) image
+ if (sz != 1 && sz != 9) goto errorEnd; // colortype 1 demands image type 1 or 9
+ stbi__skip(s,4); // skip index of first colormap entry and number of entries
+ sz = stbi__get8(s); // check bits per palette color entry
+ if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) goto errorEnd;
+ stbi__skip(s,4); // skip image x and y origin
+ } else { // "normal" image w/o colormap
+ if ( (sz != 2) && (sz != 3) && (sz != 10) && (sz != 11) ) goto errorEnd; // only RGB or grey allowed, +/- RLE
+ stbi__skip(s,9); // skip colormap specification and image x/y origin
+ }
+ if ( stbi__get16le(s) < 1 ) goto errorEnd; // test width
+ if ( stbi__get16le(s) < 1 ) goto errorEnd; // test height
+ sz = stbi__get8(s); // bits per pixel
+ if ( (tga_color_type == 1) && (sz != 8) && (sz != 16) ) goto errorEnd; // for colormapped images, bpp is size of an index
+ if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) goto errorEnd;
+
+ res = 1; // if we got this far, everything's good and we can return 1 instead of 0
+
+errorEnd:
+ stbi__rewind(s);
+ return res;
+}
+
+// read 16bit value and convert to 24bit RGB
+static void stbi__tga_read_rgb16(stbi__context *s, stbi_uc* out)
+{
+ stbi__uint16 px = (stbi__uint16)stbi__get16le(s);
+ stbi__uint16 fiveBitMask = 31;
+ // we have 3 channels with 5bits each
+ int r = (px >> 10) & fiveBitMask;
+ int g = (px >> 5) & fiveBitMask;
+ int b = px & fiveBitMask;
+ // Note that this saves the data in RGB(A) order, so it doesn't need to be swapped later
+ out[0] = (stbi_uc)((r * 255)/31);
+ out[1] = (stbi_uc)((g * 255)/31);
+ out[2] = (stbi_uc)((b * 255)/31);
+
+ // some people claim that the most significant bit might be used for alpha
+ // (possibly if an alpha-bit is set in the "image descriptor byte")
+ // but that only made 16bit test images completely translucent..
+ // so let's treat all 15 and 16bit TGAs as RGB with no alpha.
+}
+
+static void *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+ // read in the TGA header stuff
+ int tga_offset = stbi__get8(s);
+ int tga_indexed = stbi__get8(s);
+ int tga_image_type = stbi__get8(s);
+ int tga_is_RLE = 0;
+ int tga_palette_start = stbi__get16le(s);
+ int tga_palette_len = stbi__get16le(s);
+ int tga_palette_bits = stbi__get8(s);
+ int tga_x_origin = stbi__get16le(s);
+ int tga_y_origin = stbi__get16le(s);
+ int tga_width = stbi__get16le(s);
+ int tga_height = stbi__get16le(s);
+ int tga_bits_per_pixel = stbi__get8(s);
+ int tga_comp, tga_rgb16=0;
+ int tga_inverted = stbi__get8(s);
+ // int tga_alpha_bits = tga_inverted & 15; // the 4 lowest bits - unused (useless?)
+ // image data
+ unsigned char *tga_data;
+ unsigned char *tga_palette = NULL;
+ int i, j;
+ unsigned char raw_data[4] = {0};
+ int RLE_count = 0;
+ int RLE_repeating = 0;
+ int read_next_pixel = 1;
+ STBI_NOTUSED(ri);
+
+ // do a tiny bit of precessing
+ if ( tga_image_type >= 8 )
+ {
+ tga_image_type -= 8;
+ tga_is_RLE = 1;
+ }
+ tga_inverted = 1 - ((tga_inverted >> 5) & 1);
+
+ // If I'm paletted, then I'll use the number of bits from the palette
+ if ( tga_indexed ) tga_comp = stbi__tga_get_comp(tga_palette_bits, 0, &tga_rgb16);
+ else tga_comp = stbi__tga_get_comp(tga_bits_per_pixel, (tga_image_type == 3), &tga_rgb16);
+
+ if(!tga_comp) // shouldn't really happen, stbi__tga_test() should have ensured basic consistency
+ return stbi__errpuc("bad format", "Can't find out TGA pixelformat");
+
+ // tga info
+ *x = tga_width;
+ *y = tga_height;
+ if (comp) *comp = tga_comp;
+
+ if (!stbi__mad3sizes_valid(tga_width, tga_height, tga_comp, 0))
+ return stbi__errpuc("too large", "Corrupt TGA");
+
+ tga_data = (unsigned char*)stbi__malloc_mad3(tga_width, tga_height, tga_comp, 0);
+ if (!tga_data) return stbi__errpuc("outofmem", "Out of memory");
+
+ // skip to the data's starting position (offset usually = 0)
+ stbi__skip(s, tga_offset );
+
+ if ( !tga_indexed && !tga_is_RLE && !tga_rgb16 ) {
+ for (i=0; i < tga_height; ++i) {
+ int row = tga_inverted ? tga_height -i - 1 : i;
+ stbi_uc *tga_row = tga_data + row*tga_width*tga_comp;
+ stbi__getn(s, tga_row, tga_width * tga_comp);
+ }
+ } else {
+ // do I need to load a palette?
+ if ( tga_indexed)
+ {
+ // any data to skip? (offset usually = 0)
+ stbi__skip(s, tga_palette_start );
+ // load the palette
+ tga_palette = (unsigned char*)stbi__malloc_mad2(tga_palette_len, tga_comp, 0);
+ if (!tga_palette) {
+ STBI_FREE(tga_data);
+ return stbi__errpuc("outofmem", "Out of memory");
+ }
+ if (tga_rgb16) {
+ stbi_uc *pal_entry = tga_palette;
+ STBI_ASSERT(tga_comp == STBI_rgb);
+ for (i=0; i < tga_palette_len; ++i) {
+ stbi__tga_read_rgb16(s, pal_entry);
+ pal_entry += tga_comp;
+ }
+ } else if (!stbi__getn(s, tga_palette, tga_palette_len * tga_comp)) {
+ STBI_FREE(tga_data);
+ STBI_FREE(tga_palette);
+ return stbi__errpuc("bad palette", "Corrupt TGA");
+ }
+ }
+ // load the data
+ for (i=0; i < tga_width * tga_height; ++i)
+ {
+ // if I'm in RLE mode, do I need to get a RLE stbi__pngchunk?
+ if ( tga_is_RLE )
+ {
+ if ( RLE_count == 0 )
+ {
+ // yep, get the next byte as a RLE command
+ int RLE_cmd = stbi__get8(s);
+ RLE_count = 1 + (RLE_cmd & 127);
+ RLE_repeating = RLE_cmd >> 7;
+ read_next_pixel = 1;
+ } else if ( !RLE_repeating )
+ {
+ read_next_pixel = 1;
+ }
+ } else
+ {
+ read_next_pixel = 1;
+ }
+ // OK, if I need to read a pixel, do it now
+ if ( read_next_pixel )
+ {
+ // load however much data we did have
+ if ( tga_indexed )
+ {
+ // read in index, then perform the lookup
+ int pal_idx = (tga_bits_per_pixel == 8) ? stbi__get8(s) : stbi__get16le(s);
+ if ( pal_idx >= tga_palette_len ) {
+ // invalid index
+ pal_idx = 0;
+ }
+ pal_idx *= tga_comp;
+ for (j = 0; j < tga_comp; ++j) {
+ raw_data[j] = tga_palette[pal_idx+j];
+ }
+ } else if(tga_rgb16) {
+ STBI_ASSERT(tga_comp == STBI_rgb);
+ stbi__tga_read_rgb16(s, raw_data);
+ } else {
+ // read in the data raw
+ for (j = 0; j < tga_comp; ++j) {
+ raw_data[j] = stbi__get8(s);
+ }
+ }
+ // clear the reading flag for the next pixel
+ read_next_pixel = 0;
+ } // end of reading a pixel
+
+ // copy data
+ for (j = 0; j < tga_comp; ++j)
+ tga_data[i*tga_comp+j] = raw_data[j];
+
+ // in case we're in RLE mode, keep counting down
+ --RLE_count;
+ }
+ // do I need to invert the image?
+ if ( tga_inverted )
+ {
+ for (j = 0; j*2 < tga_height; ++j)
+ {
+ int index1 = j * tga_width * tga_comp;
+ int index2 = (tga_height - 1 - j) * tga_width * tga_comp;
+ for (i = tga_width * tga_comp; i > 0; --i)
+ {
+ unsigned char temp = tga_data[index1];
+ tga_data[index1] = tga_data[index2];
+ tga_data[index2] = temp;
+ ++index1;
+ ++index2;
+ }
+ }
+ }
+ // clear my palette, if I had one
+ if ( tga_palette != NULL )
+ {
+ STBI_FREE( tga_palette );
+ }
+ }
+
+ // swap RGB - if the source data was RGB16, it already is in the right order
+ if (tga_comp >= 3 && !tga_rgb16)
+ {
+ unsigned char* tga_pixel = tga_data;
+ for (i=0; i < tga_width * tga_height; ++i)
+ {
+ unsigned char temp = tga_pixel[0];
+ tga_pixel[0] = tga_pixel[2];
+ tga_pixel[2] = temp;
+ tga_pixel += tga_comp;
+ }
+ }
+
+ // convert to target component count
+ if (req_comp && req_comp != tga_comp)
+ tga_data = stbi__convert_format(tga_data, tga_comp, req_comp, tga_width, tga_height);
+
+ // the things I do to get rid of an error message, and yet keep
+ // Microsoft's C compilers happy... [8^(
+ tga_palette_start = tga_palette_len = tga_palette_bits =
+ tga_x_origin = tga_y_origin = 0;
+ // OK, done
+ return tga_data;
+}
+#endif
+
+// *************************************************************************************************
+// Photoshop PSD loader -- PD by Thatcher Ulrich, integration by Nicolas Schulz, tweaked by STB
+
+#ifndef STBI_NO_PSD
+static int stbi__psd_test(stbi__context *s)
+{
+ int r = (stbi__get32be(s) == 0x38425053);
+ stbi__rewind(s);
+ return r;
+}
+
+static int stbi__psd_decode_rle(stbi__context *s, stbi_uc *p, int pixelCount)
+{
+ int count, nleft, len;
+
+ count = 0;
+ while ((nleft = pixelCount - count) > 0) {
+ len = stbi__get8(s);
+ if (len == 128) {
+ // No-op.
+ } else if (len < 128) {
+ // Copy next len+1 bytes literally.
+ len++;
+ if (len > nleft) return 0; // corrupt data
+ count += len;
+ while (len) {
+ *p = stbi__get8(s);
+ p += 4;
+ len--;
+ }
+ } else if (len > 128) {
+ stbi_uc val;
+ // Next -len+1 bytes in the dest are replicated from next source byte.
+ // (Interpret len as a negative 8-bit int.)
+ len = 257 - len;
+ if (len > nleft) return 0; // corrupt data
+ val = stbi__get8(s);
+ count += len;
+ while (len) {
+ *p = val;
+ p += 4;
+ len--;
+ }
+ }
+ }
+
+ return 1;
+}
+
+static void *stbi__psd_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri, int bpc)
+{
+ int pixelCount;
+ int channelCount, compression;
+ int channel, i;
+ int bitdepth;
+ int w,h;
+ stbi_uc *out;
+ STBI_NOTUSED(ri);
+
+ // Check identifier
+ if (stbi__get32be(s) != 0x38425053) // "8BPS"
+ return stbi__errpuc("not PSD", "Corrupt PSD image");
+
+ // Check file type version.
+ if (stbi__get16be(s) != 1)
+ return stbi__errpuc("wrong version", "Unsupported version of PSD image");
+
+ // Skip 6 reserved bytes.
+ stbi__skip(s, 6 );
+
+ // Read the number of channels (R, G, B, A, etc).
+ channelCount = stbi__get16be(s);
+ if (channelCount < 0 || channelCount > 16)
+ return stbi__errpuc("wrong channel count", "Unsupported number of channels in PSD image");
+
+ // Read the rows and columns of the image.
+ h = stbi__get32be(s);
+ w = stbi__get32be(s);
+
+ // Make sure the depth is 8 bits.
+ bitdepth = stbi__get16be(s);
+ if (bitdepth != 8 && bitdepth != 16)
+ return stbi__errpuc("unsupported bit depth", "PSD bit depth is not 8 or 16 bit");
+
+ // Make sure the color mode is RGB.
+ // Valid options are:
+ // 0: Bitmap
+ // 1: Grayscale
+ // 2: Indexed color
+ // 3: RGB color
+ // 4: CMYK color
+ // 7: Multichannel
+ // 8: Duotone
+ // 9: Lab color
+ if (stbi__get16be(s) != 3)
+ return stbi__errpuc("wrong color format", "PSD is not in RGB color format");
+
+ // Skip the Mode Data. (It's the palette for indexed color; other info for other modes.)
+ stbi__skip(s,stbi__get32be(s) );
+
+ // Skip the image resources. (resolution, pen tool paths, etc)
+ stbi__skip(s, stbi__get32be(s) );
+
+ // Skip the reserved data.
+ stbi__skip(s, stbi__get32be(s) );
+
+ // Find out if the data is compressed.
+ // Known values:
+ // 0: no compression
+ // 1: RLE compressed
+ compression = stbi__get16be(s);
+ if (compression > 1)
+ return stbi__errpuc("bad compression", "PSD has an unknown compression format");
+
+ // Check size
+ if (!stbi__mad3sizes_valid(4, w, h, 0))
+ return stbi__errpuc("too large", "Corrupt PSD");
+
+ // Create the destination image.
+
+ if (!compression && bitdepth == 16 && bpc == 16) {
+ out = (stbi_uc *) stbi__malloc_mad3(8, w, h, 0);
+ ri->bits_per_channel = 16;
+ } else
+ out = (stbi_uc *) stbi__malloc(4 * w*h);
+
+ if (!out) return stbi__errpuc("outofmem", "Out of memory");
+ pixelCount = w*h;
+
+ // Initialize the data to zero.
+ //memset( out, 0, pixelCount * 4 );
+
+ // Finally, the image data.
+ if (compression) {
+ // RLE as used by .PSD and .TIFF
+ // Loop until you get the number of unpacked bytes you are expecting:
+ // Read the next source byte into n.
+ // If n is between 0 and 127 inclusive, copy the next n+1 bytes literally.
+ // Else if n is between -127 and -1 inclusive, copy the next byte -n+1 times.
+ // Else if n is 128, noop.
+ // Endloop
+
+ // The RLE-compressed data is preceeded by a 2-byte data count for each row in the data,
+ // which we're going to just skip.
+ stbi__skip(s, h * channelCount * 2 );
+
+ // Read the RLE data by channel.
+ for (channel = 0; channel < 4; channel++) {
+ stbi_uc *p;
+
+ p = out+channel;
+ if (channel >= channelCount) {
+ // Fill this channel with default data.
+ for (i = 0; i < pixelCount; i++, p += 4)
+ *p = (channel == 3 ? 255 : 0);
+ } else {
+ // Read the RLE data.
+ if (!stbi__psd_decode_rle(s, p, pixelCount)) {
+ STBI_FREE(out);
+ return stbi__errpuc("corrupt", "bad RLE data");
+ }
+ }
+ }
+
+ } else {
+ // We're at the raw image data. It's each channel in order (Red, Green, Blue, Alpha, ...)
+ // where each channel consists of an 8-bit (or 16-bit) value for each pixel in the image.
+
+ // Read the data by channel.
+ for (channel = 0; channel < 4; channel++) {
+ if (channel >= channelCount) {
+ // Fill this channel with default data.
+ if (bitdepth == 16 && bpc == 16) {
+ stbi__uint16 *q = ((stbi__uint16 *) out) + channel;
+ stbi__uint16 val = channel == 3 ? 65535 : 0;
+ for (i = 0; i < pixelCount; i++, q += 4)
+ *q = val;
+ } else {
+ stbi_uc *p = out+channel;
+ stbi_uc val = channel == 3 ? 255 : 0;
+ for (i = 0; i < pixelCount; i++, p += 4)
+ *p = val;
+ }
+ } else {
+ if (ri->bits_per_channel == 16) { // output bpc
+ stbi__uint16 *q = ((stbi__uint16 *) out) + channel;
+ for (i = 0; i < pixelCount; i++, q += 4)
+ *q = (stbi__uint16) stbi__get16be(s);
+ } else {
+ stbi_uc *p = out+channel;
+ if (bitdepth == 16) { // input bpc
+ for (i = 0; i < pixelCount; i++, p += 4)
+ *p = (stbi_uc) (stbi__get16be(s) >> 8);
+ } else {
+ for (i = 0; i < pixelCount; i++, p += 4)
+ *p = stbi__get8(s);
+ }
+ }
+ }
+ }
+ }
+
+ // remove weird white matte from PSD
+ if (channelCount >= 4) {
+ if (ri->bits_per_channel == 16) {
+ for (i=0; i < w*h; ++i) {
+ stbi__uint16 *pixel = (stbi__uint16 *) out + 4*i;
+ if (pixel[3] != 0 && pixel[3] != 65535) {
+ float a = pixel[3] / 65535.0f;
+ float ra = 1.0f / a;
+ float inv_a = 65535.0f * (1 - ra);
+ pixel[0] = (stbi__uint16) (pixel[0]*ra + inv_a);
+ pixel[1] = (stbi__uint16) (pixel[1]*ra + inv_a);
+ pixel[2] = (stbi__uint16) (pixel[2]*ra + inv_a);
+ }
+ }
+ } else {
+ for (i=0; i < w*h; ++i) {
+ unsigned char *pixel = out + 4*i;
+ if (pixel[3] != 0 && pixel[3] != 255) {
+ float a = pixel[3] / 255.0f;
+ float ra = 1.0f / a;
+ float inv_a = 255.0f * (1 - ra);
+ pixel[0] = (unsigned char) (pixel[0]*ra + inv_a);
+ pixel[1] = (unsigned char) (pixel[1]*ra + inv_a);
+ pixel[2] = (unsigned char) (pixel[2]*ra + inv_a);
+ }
+ }
+ }
+ }
+
+ // convert to desired output format
+ if (req_comp && req_comp != 4) {
+ if (ri->bits_per_channel == 16)
+ out = (stbi_uc *) stbi__convert_format16((stbi__uint16 *) out, 4, req_comp, w, h);
+ else
+ out = stbi__convert_format(out, 4, req_comp, w, h);
+ if (out == NULL) return out; // stbi__convert_format frees input on failure
+ }
+
+ if (comp) *comp = 4;
+ *y = h;
+ *x = w;
+
+ return out;
+}
+#endif
+
+// *************************************************************************************************
+// Softimage PIC loader
+// by Tom Seddon
+//
+// See http://softimage.wiki.softimage.com/index.php/INFO:_PIC_file_format
+// See http://ozviz.wasp.uwa.edu.au/~pbourke/dataformats/softimagepic/
+
+#ifndef STBI_NO_PIC
+static int stbi__pic_is4(stbi__context *s,const char *str)
+{
+ int i;
+ for (i=0; i<4; ++i)
+ if (stbi__get8(s) != (stbi_uc)str[i])
+ return 0;
+
+ return 1;
+}
+
+static int stbi__pic_test_core(stbi__context *s)
+{
+ int i;
+
+ if (!stbi__pic_is4(s,"\x53\x80\xF6\x34"))
+ return 0;
+
+ for(i=0;i<84;++i)
+ stbi__get8(s);
+
+ if (!stbi__pic_is4(s,"PICT"))
+ return 0;
+
+ return 1;
+}
+
+typedef struct
+{
+ stbi_uc size,type,channel;
+} stbi__pic_packet;
+
+static stbi_uc *stbi__readval(stbi__context *s, int channel, stbi_uc *dest)
+{
+ int mask=0x80, i;
+
+ for (i=0; i<4; ++i, mask>>=1) {
+ if (channel & mask) {
+ if (stbi__at_eof(s)) return stbi__errpuc("bad file","PIC file too short");
+ dest[i]=stbi__get8(s);
+ }
+ }
+
+ return dest;
+}
+
+static void stbi__copyval(int channel,stbi_uc *dest,const stbi_uc *src)
+{
+ int mask=0x80,i;
+
+ for (i=0;i<4; ++i, mask>>=1)
+ if (channel&mask)
+ dest[i]=src[i];
+}
+
+static stbi_uc *stbi__pic_load_core(stbi__context *s,int width,int height,int *comp, stbi_uc *result)
+{
+ int act_comp=0,num_packets=0,y,chained;
+ stbi__pic_packet packets[10];
+
+ // this will (should...) cater for even some bizarre stuff like having data
+ // for the same channel in multiple packets.
+ do {
+ stbi__pic_packet *packet;
+
+ if (num_packets==sizeof(packets)/sizeof(packets[0]))
+ return stbi__errpuc("bad format","too many packets");
+
+ packet = &packets[num_packets++];
+
+ chained = stbi__get8(s);
+ packet->size = stbi__get8(s);
+ packet->type = stbi__get8(s);
+ packet->channel = stbi__get8(s);
+
+ act_comp |= packet->channel;
+
+ if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (reading packets)");
+ if (packet->size != 8) return stbi__errpuc("bad format","packet isn't 8bpp");
+ } while (chained);
+
+ *comp = (act_comp & 0x10 ? 4 : 3); // has alpha channel?
+
+ for(y=0; y<height; ++y) {
+ int packet_idx;
+
+ for(packet_idx=0; packet_idx < num_packets; ++packet_idx) {
+ stbi__pic_packet *packet = &packets[packet_idx];
+ stbi_uc *dest = result+y*width*4;
+
+ switch (packet->type) {
+ default:
+ return stbi__errpuc("bad format","packet has bad compression type");
+
+ case 0: {//uncompressed
+ int x;
+
+ for(x=0;x<width;++x, dest+=4)
+ if (!stbi__readval(s,packet->channel,dest))
+ return 0;
+ break;
+ }
+
+ case 1://Pure RLE
+ {
+ int left=width, i;
+
+ while (left>0) {
+ stbi_uc count,value[4];
+
+ count=stbi__get8(s);
+ if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (pure read count)");
+
+ if (count > left)
+ count = (stbi_uc) left;
+
+ if (!stbi__readval(s,packet->channel,value)) return 0;
+
+ for(i=0; i<count; ++i,dest+=4)
+ stbi__copyval(packet->channel,dest,value);
+ left -= count;
+ }
+ }
+ break;
+
+ case 2: {//Mixed RLE
+ int left=width;
+ while (left>0) {
+ int count = stbi__get8(s), i;
+ if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (mixed read count)");
+
+ if (count >= 128) { // Repeated
+ stbi_uc value[4];
+
+ if (count==128)
+ count = stbi__get16be(s);
+ else
+ count -= 127;
+ if (count > left)
+ return stbi__errpuc("bad file","scanline overrun");
+
+ if (!stbi__readval(s,packet->channel,value))
+ return 0;
+
+ for(i=0;i<count;++i, dest += 4)
+ stbi__copyval(packet->channel,dest,value);
+ } else { // Raw
+ ++count;
+ if (count>left) return stbi__errpuc("bad file","scanline overrun");
+
+ for(i=0;i<count;++i, dest+=4)
+ if (!stbi__readval(s,packet->channel,dest))
+ return 0;
+ }
+ left-=count;
+ }
+ break;
+ }
+ }
+ }
+ }
+
+ return result;
+}
+
+static void *stbi__pic_load(stbi__context *s,int *px,int *py,int *comp,int req_comp, stbi__result_info *ri)
+{
+ stbi_uc *result;
+ int i, x,y, internal_comp;
+ STBI_NOTUSED(ri);
+
+ if (!comp) comp = &internal_comp;
+
+ for (i=0; i<92; ++i)
+ stbi__get8(s);
+
+ x = stbi__get16be(s);
+ y = stbi__get16be(s);
+ if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (pic header)");
+ if (!stbi__mad3sizes_valid(x, y, 4, 0)) return stbi__errpuc("too large", "PIC image too large to decode");
+
+ stbi__get32be(s); //skip `ratio'
+ stbi__get16be(s); //skip `fields'
+ stbi__get16be(s); //skip `pad'
+
+ // intermediate buffer is RGBA
+ result = (stbi_uc *) stbi__malloc_mad3(x, y, 4, 0);
+ memset(result, 0xff, x*y*4);
+
+ if (!stbi__pic_load_core(s,x,y,comp, result)) {
+ STBI_FREE(result);
+ result=0;
+ }
+ *px = x;
+ *py = y;
+ if (req_comp == 0) req_comp = *comp;
+ result=stbi__convert_format(result,4,req_comp,x,y);
+
+ return result;
+}
+
+static int stbi__pic_test(stbi__context *s)
+{
+ int r = stbi__pic_test_core(s);
+ stbi__rewind(s);
+ return r;
+}
+#endif
+
+// *************************************************************************************************
+// GIF loader -- public domain by Jean-Marc Lienher -- simplified/shrunk by stb
+
+#ifndef STBI_NO_GIF
+typedef struct
+{
+ stbi__int16 prefix;
+ stbi_uc first;
+ stbi_uc suffix;
+} stbi__gif_lzw;
+
+typedef struct
+{
+ int w,h;
+ stbi_uc *out, *old_out; // output buffer (always 4 components)
+ int flags, bgindex, ratio, transparent, eflags, delay;
+ stbi_uc pal[256][4];
+ stbi_uc lpal[256][4];
+ stbi__gif_lzw codes[4096];
+ stbi_uc *color_table;
+ int parse, step;
+ int lflags;
+ int start_x, start_y;
+ int max_x, max_y;
+ int cur_x, cur_y;
+ int line_size;
+} stbi__gif;
+
+static int stbi__gif_test_raw(stbi__context *s)
+{
+ int sz;
+ if (stbi__get8(s) != 'G' || stbi__get8(s) != 'I' || stbi__get8(s) != 'F' || stbi__get8(s) != '8') return 0;
+ sz = stbi__get8(s);
+ if (sz != '9' && sz != '7') return 0;
+ if (stbi__get8(s) != 'a') return 0;
+ return 1;
+}
+
+static int stbi__gif_test(stbi__context *s)
+{
+ int r = stbi__gif_test_raw(s);
+ stbi__rewind(s);
+ return r;
+}
+
+static void stbi__gif_parse_colortable(stbi__context *s, stbi_uc pal[256][4], int num_entries, int transp)
+{
+ int i;
+ for (i=0; i < num_entries; ++i) {
+ pal[i][2] = stbi__get8(s);
+ pal[i][1] = stbi__get8(s);
+ pal[i][0] = stbi__get8(s);
+ pal[i][3] = transp == i ? 0 : 255;
+ }
+}
+
+static int stbi__gif_header(stbi__context *s, stbi__gif *g, int *comp, int is_info)
+{
+ stbi_uc version;
+ if (stbi__get8(s) != 'G' || stbi__get8(s) != 'I' || stbi__get8(s) != 'F' || stbi__get8(s) != '8')
+ return stbi__err("not GIF", "Corrupt GIF");
+
+ version = stbi__get8(s);
+ if (version != '7' && version != '9') return stbi__err("not GIF", "Corrupt GIF");
+ if (stbi__get8(s) != 'a') return stbi__err("not GIF", "Corrupt GIF");
+
+ stbi__g_failure_reason = "";
+ g->w = stbi__get16le(s);
+ g->h = stbi__get16le(s);
+ g->flags = stbi__get8(s);
+ g->bgindex = stbi__get8(s);
+ g->ratio = stbi__get8(s);
+ g->transparent = -1;
+
+ if (comp != 0) *comp = 4; // can't actually tell whether it's 3 or 4 until we parse the comments
+
+ if (is_info) return 1;
+
+ if (g->flags & 0x80)
+ stbi__gif_parse_colortable(s,g->pal, 2 << (g->flags & 7), -1);
+
+ return 1;
+}
+
+static int stbi__gif_info_raw(stbi__context *s, int *x, int *y, int *comp)
+{
+ stbi__gif* g = (stbi__gif*) stbi__malloc(sizeof(stbi__gif));
+ if (!stbi__gif_header(s, g, comp, 1)) {
+ STBI_FREE(g);
+ stbi__rewind( s );
+ return 0;
+ }
+ if (x) *x = g->w;
+ if (y) *y = g->h;
+ STBI_FREE(g);
+ return 1;
+}
+
+static void stbi__out_gif_code(stbi__gif *g, stbi__uint16 code)
+{
+ stbi_uc *p, *c;
+
+ // recurse to decode the prefixes, since the linked-list is backwards,
+ // and working backwards through an interleaved image would be nasty
+ if (g->codes[code].prefix >= 0)
+ stbi__out_gif_code(g, g->codes[code].prefix);
+
+ if (g->cur_y >= g->max_y) return;
+
+ p = &g->out[g->cur_x + g->cur_y];
+ c = &g->color_table[g->codes[code].suffix * 4];
+
+ if (c[3] >= 128) {
+ p[0] = c[2];
+ p[1] = c[1];
+ p[2] = c[0];
+ p[3] = c[3];
+ }
+ g->cur_x += 4;
+
+ if (g->cur_x >= g->max_x) {
+ g->cur_x = g->start_x;
+ g->cur_y += g->step;
+
+ while (g->cur_y >= g->max_y && g->parse > 0) {
+ g->step = (1 << g->parse) * g->line_size;
+ g->cur_y = g->start_y + (g->step >> 1);
+ --g->parse;
+ }
+ }
+}
+
+static stbi_uc *stbi__process_gif_raster(stbi__context *s, stbi__gif *g)
+{
+ stbi_uc lzw_cs;
+ stbi__int32 len, init_code;
+ stbi__uint32 first;
+ stbi__int32 codesize, codemask, avail, oldcode, bits, valid_bits, clear;
+ stbi__gif_lzw *p;
+
+ lzw_cs = stbi__get8(s);
+ if (lzw_cs > 12) return NULL;
+ clear = 1 << lzw_cs;
+ first = 1;
+ codesize = lzw_cs + 1;
+ codemask = (1 << codesize) - 1;
+ bits = 0;
+ valid_bits = 0;
+ for (init_code = 0; init_code < clear; init_code++) {
+ g->codes[init_code].prefix = -1;
+ g->codes[init_code].first = (stbi_uc) init_code;
+ g->codes[init_code].suffix = (stbi_uc) init_code;
+ }
+
+ // support no starting clear code
+ avail = clear+2;
+ oldcode = -1;
+
+ len = 0;
+ for(;;) {
+ if (valid_bits < codesize) {
+ if (len == 0) {
+ len = stbi__get8(s); // start new block
+ if (len == 0)
+ return g->out;
+ }
+ --len;
+ bits |= (stbi__int32) stbi__get8(s) << valid_bits;
+ valid_bits += 8;
+ } else {
+ stbi__int32 code = bits & codemask;
+ bits >>= codesize;
+ valid_bits -= codesize;
+ // @OPTIMIZE: is there some way we can accelerate the non-clear path?
+ if (code == clear) { // clear code
+ codesize = lzw_cs + 1;
+ codemask = (1 << codesize) - 1;
+ avail = clear + 2;
+ oldcode = -1;
+ first = 0;
+ } else if (code == clear + 1) { // end of stream code
+ stbi__skip(s, len);
+ while ((len = stbi__get8(s)) > 0)
+ stbi__skip(s,len);
+ return g->out;
+ } else if (code <= avail) {
+ if (first) return stbi__errpuc("no clear code", "Corrupt GIF");
+
+ if (oldcode >= 0) {
+ p = &g->codes[avail++];
+ if (avail > 4096) return stbi__errpuc("too many codes", "Corrupt GIF");
+ p->prefix = (stbi__int16) oldcode;
+ p->first = g->codes[oldcode].first;
+ p->suffix = (code == avail) ? p->first : g->codes[code].first;
+ } else if (code == avail)
+ return stbi__errpuc("illegal code in raster", "Corrupt GIF");
+
+ stbi__out_gif_code(g, (stbi__uint16) code);
+
+ if ((avail & codemask) == 0 && avail <= 0x0FFF) {
+ codesize++;
+ codemask = (1 << codesize) - 1;
+ }
+
+ oldcode = code;
+ } else {
+ return stbi__errpuc("illegal code in raster", "Corrupt GIF");
+ }
+ }
+ }
+}
+
+static void stbi__fill_gif_background(stbi__gif *g, int x0, int y0, int x1, int y1)
+{
+ int x, y;
+ stbi_uc *c = g->pal[g->bgindex];
+ for (y = y0; y < y1; y += 4 * g->w) {
+ for (x = x0; x < x1; x += 4) {
+ stbi_uc *p = &g->out[y + x];
+ p[0] = c[2];
+ p[1] = c[1];
+ p[2] = c[0];
+ p[3] = 0;
+ }
+ }
+}
+
+// this function is designed to support animated gifs, although stb_image doesn't support it
+static stbi_uc *stbi__gif_load_next(stbi__context *s, stbi__gif *g, int *comp, int req_comp)
+{
+ int i;
+ stbi_uc *prev_out = 0;
+
+ if (g->out == 0 && !stbi__gif_header(s, g, comp,0))
+ return 0; // stbi__g_failure_reason set by stbi__gif_header
+
+ if (!stbi__mad3sizes_valid(g->w, g->h, 4, 0))
+ return stbi__errpuc("too large", "GIF too large");
+
+ prev_out = g->out;
+ g->out = (stbi_uc *) stbi__malloc_mad3(4, g->w, g->h, 0);
+ if (g->out == 0) return stbi__errpuc("outofmem", "Out of memory");
+
+ switch ((g->eflags & 0x1C) >> 2) {
+ case 0: // unspecified (also always used on 1st frame)
+ stbi__fill_gif_background(g, 0, 0, 4 * g->w, 4 * g->w * g->h);
+ break;
+ case 1: // do not dispose
+ if (prev_out) memcpy(g->out, prev_out, 4 * g->w * g->h);
+ g->old_out = prev_out;
+ break;
+ case 2: // dispose to background
+ if (prev_out) memcpy(g->out, prev_out, 4 * g->w * g->h);
+ stbi__fill_gif_background(g, g->start_x, g->start_y, g->max_x, g->max_y);
+ break;
+ case 3: // dispose to previous
+ if (g->old_out) {
+ for (i = g->start_y; i < g->max_y; i += 4 * g->w)
+ memcpy(&g->out[i + g->start_x], &g->old_out[i + g->start_x], g->max_x - g->start_x);
+ }
+ break;
+ }
+
+ for (;;) {
+ switch (stbi__get8(s)) {
+ case 0x2C: /* Image Descriptor */
+ {
+ int prev_trans = -1;
+ stbi__int32 x, y, w, h;
+ stbi_uc *o;
+
+ x = stbi__get16le(s);
+ y = stbi__get16le(s);
+ w = stbi__get16le(s);
+ h = stbi__get16le(s);
+ if (((x + w) > (g->w)) || ((y + h) > (g->h)))
+ return stbi__errpuc("bad Image Descriptor", "Corrupt GIF");
+
+ g->line_size = g->w * 4;
+ g->start_x = x * 4;
+ g->start_y = y * g->line_size;
+ g->max_x = g->start_x + w * 4;
+ g->max_y = g->start_y + h * g->line_size;
+ g->cur_x = g->start_x;
+ g->cur_y = g->start_y;
+
+ g->lflags = stbi__get8(s);
+
+ if (g->lflags & 0x40) {
+ g->step = 8 * g->line_size; // first interlaced spacing
+ g->parse = 3;
+ } else {
+ g->step = g->line_size;
+ g->parse = 0;
+ }
+
+ if (g->lflags & 0x80) {
+ stbi__gif_parse_colortable(s,g->lpal, 2 << (g->lflags & 7), g->eflags & 0x01 ? g->transparent : -1);
+ g->color_table = (stbi_uc *) g->lpal;
+ } else if (g->flags & 0x80) {
+ if (g->transparent >= 0 && (g->eflags & 0x01)) {
+ prev_trans = g->pal[g->transparent][3];
+ g->pal[g->transparent][3] = 0;
+ }
+ g->color_table = (stbi_uc *) g->pal;
+ } else
+ return stbi__errpuc("missing color table", "Corrupt GIF");
+
+ o = stbi__process_gif_raster(s, g);
+ if (o == NULL) return NULL;
+
+ if (prev_trans != -1)
+ g->pal[g->transparent][3] = (stbi_uc) prev_trans;
+
+ return o;
+ }
+
+ case 0x21: // Comment Extension.
+ {
+ int len;
+ if (stbi__get8(s) == 0xF9) { // Graphic Control Extension.
+ len = stbi__get8(s);
+ if (len == 4) {
+ g->eflags = stbi__get8(s);
+ g->delay = stbi__get16le(s);
+ g->transparent = stbi__get8(s);
+ } else {
+ stbi__skip(s, len);
+ break;
+ }
+ }
+ while ((len = stbi__get8(s)) != 0)
+ stbi__skip(s, len);
+ break;
+ }
+
+ case 0x3B: // gif stream termination code
+ return (stbi_uc *) s; // using '1' causes warning on some compilers
+
+ default:
+ return stbi__errpuc("unknown code", "Corrupt GIF");
+ }
+ }
+
+ STBI_NOTUSED(req_comp);
+}
+
+static void *stbi__gif_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+ stbi_uc *u = 0;
+ stbi__gif* g = (stbi__gif*) stbi__malloc(sizeof(stbi__gif));
+ memset(g, 0, sizeof(*g));
+ STBI_NOTUSED(ri);
+
+ u = stbi__gif_load_next(s, g, comp, req_comp);
+ if (u == (stbi_uc *) s) u = 0; // end of animated gif marker
+ if (u) {
+ *x = g->w;
+ *y = g->h;
+ if (req_comp && req_comp != 4)
+ u = stbi__convert_format(u, 4, req_comp, g->w, g->h);
+ }
+ else if (g->out)
+ STBI_FREE(g->out);
+ STBI_FREE(g);
+ return u;
+}
+
+static int stbi__gif_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ return stbi__gif_info_raw(s,x,y,comp);
+}
+#endif
+
+// *************************************************************************************************
+// Radiance RGBE HDR loader
+// originally by Nicolas Schulz
+#ifndef STBI_NO_HDR
+static int stbi__hdr_test_core(stbi__context *s, const char *signature)
+{
+ int i;
+ for (i=0; signature[i]; ++i)
+ if (stbi__get8(s) != signature[i])
+ return 0;
+ stbi__rewind(s);
+ return 1;
+}
+
+static int stbi__hdr_test(stbi__context* s)
+{
+ int r = stbi__hdr_test_core(s, "#?RADIANCE\n");
+ stbi__rewind(s);
+ if(!r) {
+ r = stbi__hdr_test_core(s, "#?RGBE\n");
+ stbi__rewind(s);
+ }
+ return r;
+}
+
+#define STBI__HDR_BUFLEN 1024
+static char *stbi__hdr_gettoken(stbi__context *z, char *buffer)
+{
+ int len=0;
+ char c = '\0';
+
+ c = (char) stbi__get8(z);
+
+ while (!stbi__at_eof(z) && c != '\n') {
+ buffer[len++] = c;
+ if (len == STBI__HDR_BUFLEN-1) {
+ // flush to end of line
+ while (!stbi__at_eof(z) && stbi__get8(z) != '\n')
+ ;
+ break;
+ }
+ c = (char) stbi__get8(z);
+ }
+
+ buffer[len] = 0;
+ return buffer;
+}
+
+static void stbi__hdr_convert(float *output, stbi_uc *input, int req_comp)
+{
+ if ( input[3] != 0 ) {
+ float f1;
+ // Exponent
+ f1 = (float) ldexp(1.0f, input[3] - (int)(128 + 8));
+ if (req_comp <= 2)
+ output[0] = (input[0] + input[1] + input[2]) * f1 / 3;
+ else {
+ output[0] = input[0] * f1;
+ output[1] = input[1] * f1;
+ output[2] = input[2] * f1;
+ }
+ if (req_comp == 2) output[1] = 1;
+ if (req_comp == 4) output[3] = 1;
+ } else {
+ switch (req_comp) {
+ case 4: output[3] = 1; /* fallthrough */
+ case 3: output[0] = output[1] = output[2] = 0;
+ break;
+ case 2: output[1] = 1; /* fallthrough */
+ case 1: output[0] = 0;
+ break;
+ }
+ }
+}
+
+static float *stbi__hdr_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+ char buffer[STBI__HDR_BUFLEN];
+ char *token;
+ int valid = 0;
+ int width, height;
+ stbi_uc *scanline;
+ float *hdr_data;
+ int len;
+ unsigned char count, value;
+ int i, j, k, c1,c2, z;
+ const char *headerToken;
+ STBI_NOTUSED(ri);
+
+ // Check identifier
+ headerToken = stbi__hdr_gettoken(s,buffer);
+ if (strcmp(headerToken, "#?RADIANCE") != 0 && strcmp(headerToken, "#?RGBE") != 0)
+ return stbi__errpf("not HDR", "Corrupt HDR image");
+
+ // Parse header
+ for(;;) {
+ token = stbi__hdr_gettoken(s,buffer);
+ if (token[0] == 0) break;
+ if (strcmp(token, "FORMAT=32-bit_rle_rgbe") == 0) valid = 1;
+ }
+
+ if (!valid) return stbi__errpf("unsupported format", "Unsupported HDR format");
+
+ // Parse width and height
+ // can't use sscanf() if we're not using stdio!
+ token = stbi__hdr_gettoken(s,buffer);
+ if (strncmp(token, "-Y ", 3)) return stbi__errpf("unsupported data layout", "Unsupported HDR format");
+ token += 3;
+ height = (int) strtol(token, &token, 10);
+ while (*token == ' ') ++token;
+ if (strncmp(token, "+X ", 3)) return stbi__errpf("unsupported data layout", "Unsupported HDR format");
+ token += 3;
+ width = (int) strtol(token, NULL, 10);
+
+ *x = width;
+ *y = height;
+
+ if (comp) *comp = 3;
+ if (req_comp == 0) req_comp = 3;
+
+ if (!stbi__mad4sizes_valid(width, height, req_comp, sizeof(float), 0))
+ return stbi__errpf("too large", "HDR image is too large");
+
+ // Read data
+ hdr_data = (float *) stbi__malloc_mad4(width, height, req_comp, sizeof(float), 0);
+ if (!hdr_data)
+ return stbi__errpf("outofmem", "Out of memory");
+
+ // Load image data
+ // image data is stored as some number of sca
+ if ( width < 8 || width >= 32768) {
+ // Read flat data
+ for (j=0; j < height; ++j) {
+ for (i=0; i < width; ++i) {
+ stbi_uc rgbe[4];
+ main_decode_loop:
+ stbi__getn(s, rgbe, 4);
+ stbi__hdr_convert(hdr_data + j * width * req_comp + i * req_comp, rgbe, req_comp);
+ }
+ }
+ } else {
+ // Read RLE-encoded data
+ scanline = NULL;
+
+ for (j = 0; j < height; ++j) {
+ c1 = stbi__get8(s);
+ c2 = stbi__get8(s);
+ len = stbi__get8(s);
+ if (c1 != 2 || c2 != 2 || (len & 0x80)) {
+ // not run-length encoded, so we have to actually use THIS data as a decoded
+ // pixel (note this can't be a valid pixel--one of RGB must be >= 128)
+ stbi_uc rgbe[4];
+ rgbe[0] = (stbi_uc) c1;
+ rgbe[1] = (stbi_uc) c2;
+ rgbe[2] = (stbi_uc) len;
+ rgbe[3] = (stbi_uc) stbi__get8(s);
+ stbi__hdr_convert(hdr_data, rgbe, req_comp);
+ i = 1;
+ j = 0;
+ STBI_FREE(scanline);
+ goto main_decode_loop; // yes, this makes no sense
+ }
+ len <<= 8;
+ len |= stbi__get8(s);
+ if (len != width) { STBI_FREE(hdr_data); STBI_FREE(scanline); return stbi__errpf("invalid decoded scanline length", "corrupt HDR"); }
+ if (scanline == NULL) {
+ scanline = (stbi_uc *) stbi__malloc_mad2(width, 4, 0);
+ if (!scanline) {
+ STBI_FREE(hdr_data);
+ return stbi__errpf("outofmem", "Out of memory");
+ }
+ }
+
+ for (k = 0; k < 4; ++k) {
+ int nleft;
+ i = 0;
+ while ((nleft = width - i) > 0) {
+ count = stbi__get8(s);
+ if (count > 128) {
+ // Run
+ value = stbi__get8(s);
+ count -= 128;
+ if (count > nleft) { STBI_FREE(hdr_data); STBI_FREE(scanline); return stbi__errpf("corrupt", "bad RLE data in HDR"); }
+ for (z = 0; z < count; ++z)
+ scanline[i++ * 4 + k] = value;
+ } else {
+ // Dump
+ if (count > nleft) { STBI_FREE(hdr_data); STBI_FREE(scanline); return stbi__errpf("corrupt", "bad RLE data in HDR"); }
+ for (z = 0; z < count; ++z)
+ scanline[i++ * 4 + k] = stbi__get8(s);
+ }
+ }
+ }
+ for (i=0; i < width; ++i)
+ stbi__hdr_convert(hdr_data+(j*width + i)*req_comp, scanline + i*4, req_comp);
+ }
+ if (scanline)
+ STBI_FREE(scanline);
+ }
+
+ return hdr_data;
+}
+
+static int stbi__hdr_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ char buffer[STBI__HDR_BUFLEN];
+ char *token;
+ int valid = 0;
+ int dummy;
+
+ if (!x) x = &dummy;
+ if (!y) y = &dummy;
+ if (!comp) comp = &dummy;
+
+ if (stbi__hdr_test(s) == 0) {
+ stbi__rewind( s );
+ return 0;
+ }
+
+ for(;;) {
+ token = stbi__hdr_gettoken(s,buffer);
+ if (token[0] == 0) break;
+ if (strcmp(token, "FORMAT=32-bit_rle_rgbe") == 0) valid = 1;
+ }
+
+ if (!valid) {
+ stbi__rewind( s );
+ return 0;
+ }
+ token = stbi__hdr_gettoken(s,buffer);
+ if (strncmp(token, "-Y ", 3)) {
+ stbi__rewind( s );
+ return 0;
+ }
+ token += 3;
+ *y = (int) strtol(token, &token, 10);
+ while (*token == ' ') ++token;
+ if (strncmp(token, "+X ", 3)) {
+ stbi__rewind( s );
+ return 0;
+ }
+ token += 3;
+ *x = (int) strtol(token, NULL, 10);
+ *comp = 3;
+ return 1;
+}
+#endif // STBI_NO_HDR
+
+#ifndef STBI_NO_BMP
+static int stbi__bmp_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ void *p;
+ stbi__bmp_data info;
+
+ info.all_a = 255;
+ p = stbi__bmp_parse_header(s, &info);
+ stbi__rewind( s );
+ if (p == NULL)
+ return 0;
+ if (x) *x = s->img_x;
+ if (y) *y = s->img_y;
+ if (comp) *comp = info.ma ? 4 : 3;
+ return 1;
+}
+#endif
+
+#ifndef STBI_NO_PSD
+static int stbi__psd_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ int channelCount, dummy;
+ if (!x) x = &dummy;
+ if (!y) y = &dummy;
+ if (!comp) comp = &dummy;
+ if (stbi__get32be(s) != 0x38425053) {
+ stbi__rewind( s );
+ return 0;
+ }
+ if (stbi__get16be(s) != 1) {
+ stbi__rewind( s );
+ return 0;
+ }
+ stbi__skip(s, 6);
+ channelCount = stbi__get16be(s);
+ if (channelCount < 0 || channelCount > 16) {
+ stbi__rewind( s );
+ return 0;
+ }
+ *y = stbi__get32be(s);
+ *x = stbi__get32be(s);
+ if (stbi__get16be(s) != 8) {
+ stbi__rewind( s );
+ return 0;
+ }
+ if (stbi__get16be(s) != 3) {
+ stbi__rewind( s );
+ return 0;
+ }
+ *comp = 4;
+ return 1;
+}
+#endif
+
+#ifndef STBI_NO_PIC
+static int stbi__pic_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ int act_comp=0,num_packets=0,chained,dummy;
+ stbi__pic_packet packets[10];
+
+ if (!x) x = &dummy;
+ if (!y) y = &dummy;
+ if (!comp) comp = &dummy;
+
+ if (!stbi__pic_is4(s,"\x53\x80\xF6\x34")) {
+ stbi__rewind(s);
+ return 0;
+ }
+
+ stbi__skip(s, 88);
+
+ *x = stbi__get16be(s);
+ *y = stbi__get16be(s);
+ if (stbi__at_eof(s)) {
+ stbi__rewind( s);
+ return 0;
+ }
+ if ( (*x) != 0 && (1 << 28) / (*x) < (*y)) {
+ stbi__rewind( s );
+ return 0;
+ }
+
+ stbi__skip(s, 8);
+
+ do {
+ stbi__pic_packet *packet;
+
+ if (num_packets==sizeof(packets)/sizeof(packets[0]))
+ return 0;
+
+ packet = &packets[num_packets++];
+ chained = stbi__get8(s);
+ packet->size = stbi__get8(s);
+ packet->type = stbi__get8(s);
+ packet->channel = stbi__get8(s);
+ act_comp |= packet->channel;
+
+ if (stbi__at_eof(s)) {
+ stbi__rewind( s );
+ return 0;
+ }
+ if (packet->size != 8) {
+ stbi__rewind( s );
+ return 0;
+ }
+ } while (chained);
+
+ *comp = (act_comp & 0x10 ? 4 : 3);
+
+ return 1;
+}
+#endif
+
+// *************************************************************************************************
+// Portable Gray Map and Portable Pixel Map loader
+// by Ken Miller
+//
+// PGM: http://netpbm.sourceforge.net/doc/pgm.html
+// PPM: http://netpbm.sourceforge.net/doc/ppm.html
+//
+// Known limitations:
+// Does not support comments in the header section
+// Does not support ASCII image data (formats P2 and P3)
+// Does not support 16-bit-per-channel
+
+#ifndef STBI_NO_PNM
+
+static int stbi__pnm_test(stbi__context *s)
+{
+ char p, t;
+ p = (char) stbi__get8(s);
+ t = (char) stbi__get8(s);
+ if (p != 'P' || (t != '5' && t != '6')) {
+ stbi__rewind( s );
+ return 0;
+ }
+ return 1;
+}
+
+static void *stbi__pnm_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+ stbi_uc *out;
+ STBI_NOTUSED(ri);
+
+ if (!stbi__pnm_info(s, (int *)&s->img_x, (int *)&s->img_y, (int *)&s->img_n))
+ return 0;
+
+ *x = s->img_x;
+ *y = s->img_y;
+ if (comp) *comp = s->img_n;
+
+ if (!stbi__mad3sizes_valid(s->img_n, s->img_x, s->img_y, 0))
+ return stbi__errpuc("too large", "PNM too large");
+
+ out = (stbi_uc *) stbi__malloc_mad3(s->img_n, s->img_x, s->img_y, 0);
+ if (!out) return stbi__errpuc("outofmem", "Out of memory");
+ stbi__getn(s, out, s->img_n * s->img_x * s->img_y);
+
+ if (req_comp && req_comp != s->img_n) {
+ out = stbi__convert_format(out, s->img_n, req_comp, s->img_x, s->img_y);
+ if (out == NULL) return out; // stbi__convert_format frees input on failure
+ }
+ return out;
+}
+
+static int stbi__pnm_isspace(char c)
+{
+ return c == ' ' || c == '\t' || c == '\n' || c == '\v' || c == '\f' || c == '\r';
+}
+
+static void stbi__pnm_skip_whitespace(stbi__context *s, char *c)
+{
+ for (;;) {
+ while (!stbi__at_eof(s) && stbi__pnm_isspace(*c))
+ *c = (char) stbi__get8(s);
+
+ if (stbi__at_eof(s) || *c != '#')
+ break;
+
+ while (!stbi__at_eof(s) && *c != '\n' && *c != '\r' )
+ *c = (char) stbi__get8(s);
+ }
+}
+
+static int stbi__pnm_isdigit(char c)
+{
+ return c >= '0' && c <= '9';
+}
+
+static int stbi__pnm_getinteger(stbi__context *s, char *c)
+{
+ int value = 0;
+
+ while (!stbi__at_eof(s) && stbi__pnm_isdigit(*c)) {
+ value = value*10 + (*c - '0');
+ *c = (char) stbi__get8(s);
+ }
+
+ return value;
+}
+
+static int stbi__pnm_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ int maxv, dummy;
+ char c, p, t;
+
+ if (!x) x = &dummy;
+ if (!y) y = &dummy;
+ if (!comp) comp = &dummy;
+
+ stbi__rewind(s);
+
+ // Get identifier
+ p = (char) stbi__get8(s);
+ t = (char) stbi__get8(s);
+ if (p != 'P' || (t != '5' && t != '6')) {
+ stbi__rewind(s);
+ return 0;
+ }
+
+ *comp = (t == '6') ? 3 : 1; // '5' is 1-component .pgm; '6' is 3-component .ppm
+
+ c = (char) stbi__get8(s);
+ stbi__pnm_skip_whitespace(s, &c);
+
+ *x = stbi__pnm_getinteger(s, &c); // read width
+ stbi__pnm_skip_whitespace(s, &c);
+
+ *y = stbi__pnm_getinteger(s, &c); // read height
+ stbi__pnm_skip_whitespace(s, &c);
+
+ maxv = stbi__pnm_getinteger(s, &c); // read max value
+
+ if (maxv > 255)
+ return stbi__err("max value > 255", "PPM image not 8-bit");
+ else
+ return 1;
+}
+#endif
+
+static int stbi__info_main(stbi__context *s, int *x, int *y, int *comp)
+{
+ #ifndef STBI_NO_JPEG
+ if (stbi__jpeg_info(s, x, y, comp)) return 1;
+ #endif
+
+ #ifndef STBI_NO_PNG
+ if (stbi__png_info(s, x, y, comp)) return 1;
+ #endif
+
+ #ifndef STBI_NO_GIF
+ if (stbi__gif_info(s, x, y, comp)) return 1;
+ #endif
+
+ #ifndef STBI_NO_BMP
+ if (stbi__bmp_info(s, x, y, comp)) return 1;
+ #endif
+
+ #ifndef STBI_NO_PSD
+ if (stbi__psd_info(s, x, y, comp)) return 1;
+ #endif
+
+ #ifndef STBI_NO_PIC
+ if (stbi__pic_info(s, x, y, comp)) return 1;
+ #endif
+
+ #ifndef STBI_NO_PNM
+ if (stbi__pnm_info(s, x, y, comp)) return 1;
+ #endif
+
+ #ifndef STBI_NO_HDR
+ if (stbi__hdr_info(s, x, y, comp)) return 1;
+ #endif
+
+ // test tga last because it's a crappy test!
+ #ifndef STBI_NO_TGA
+ if (stbi__tga_info(s, x, y, comp))
+ return 1;
+ #endif
+ return stbi__err("unknown image type", "Image not of any known type, or corrupt");
+}
+
+#ifndef STBI_NO_STDIO
+STBIDEF int stbi_info(char const *filename, int *x, int *y, int *comp)
+{
+ FILE *f = stbi__fopen(filename, "rb");
+ int result;
+ if (!f) return stbi__err("can't fopen", "Unable to open file");
+ result = stbi_info_from_file(f, x, y, comp);
+ fclose(f);
+ return result;
+}
+
+STBIDEF int stbi_info_from_file(FILE *f, int *x, int *y, int *comp)
+{
+ int r;
+ stbi__context s;
+ long pos = ftell(f);
+ stbi__start_file(&s, f);
+ r = stbi__info_main(&s,x,y,comp);
+ fseek(f,pos,SEEK_SET);
+ return r;
+}
+#endif // !STBI_NO_STDIO
+
+STBIDEF int stbi_info_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp)
+{
+ stbi__context s;
+ stbi__start_mem(&s,buffer,len);
+ return stbi__info_main(&s,x,y,comp);
+}
+
+STBIDEF int stbi_info_from_callbacks(stbi_io_callbacks const *c, void *user, int *x, int *y, int *comp)
+{
+ stbi__context s;
+ stbi__start_callbacks(&s, (stbi_io_callbacks *) c, user);
+ return stbi__info_main(&s,x,y,comp);
+}
+
+#endif // STB_IMAGE_IMPLEMENTATION
+
+/*
+ revision history:
+ 2.15 (2017-03-18) fix png-1,2,4 bug; now all Imagenet JPGs decode;
+ warning fixes; disable run-time SSE detection on gcc;
+ uniform handling of optional "return" values;
+ thread-safe initialization of zlib tables
+ 2.14 (2017-03-03) remove deprecated STBI_JPEG_OLD; fixes for Imagenet JPGs
+ 2.13 (2016-11-29) add 16-bit API, only supported for PNG right now
+ 2.12 (2016-04-02) fix typo in 2.11 PSD fix that caused crashes
+ 2.11 (2016-04-02) allocate large structures on the stack
+ remove white matting for transparent PSD
+ fix reported channel count for PNG & BMP
+ re-enable SSE2 in non-gcc 64-bit
+ support RGB-formatted JPEG
+ read 16-bit PNGs (only as 8-bit)
+ 2.10 (2016-01-22) avoid warning introduced in 2.09 by STBI_REALLOC_SIZED
+ 2.09 (2016-01-16) allow comments in PNM files
+ 16-bit-per-pixel TGA (not bit-per-component)
+ info() for TGA could break due to .hdr handling
+ info() for BMP to shares code instead of sloppy parse
+ can use STBI_REALLOC_SIZED if allocator doesn't support realloc
+ code cleanup
+ 2.08 (2015-09-13) fix to 2.07 cleanup, reading RGB PSD as RGBA
+ 2.07 (2015-09-13) fix compiler warnings
+ partial animated GIF support
+ limited 16-bpc PSD support
+ #ifdef unused functions
+ bug with < 92 byte PIC,PNM,HDR,TGA
+ 2.06 (2015-04-19) fix bug where PSD returns wrong '*comp' value
+ 2.05 (2015-04-19) fix bug in progressive JPEG handling, fix warning
+ 2.04 (2015-04-15) try to re-enable SIMD on MinGW 64-bit
+ 2.03 (2015-04-12) extra corruption checking (mmozeiko)
+ stbi_set_flip_vertically_on_load (nguillemot)
+ fix NEON support; fix mingw support
+ 2.02 (2015-01-19) fix incorrect assert, fix warning
+ 2.01 (2015-01-17) fix various warnings; suppress SIMD on gcc 32-bit without -msse2
+ 2.00b (2014-12-25) fix STBI_MALLOC in progressive JPEG
+ 2.00 (2014-12-25) optimize JPG, including x86 SSE2 & NEON SIMD (ryg)
+ progressive JPEG (stb)
+ PGM/PPM support (Ken Miller)
+ STBI_MALLOC,STBI_REALLOC,STBI_FREE
+ GIF bugfix -- seemingly never worked
+ STBI_NO_*, STBI_ONLY_*
+ 1.48 (2014-12-14) fix incorrectly-named assert()
+ 1.47 (2014-12-14) 1/2/4-bit PNG support, both direct and paletted (Omar Cornut & stb)
+ optimize PNG (ryg)
+ fix bug in interlaced PNG with user-specified channel count (stb)
+ 1.46 (2014-08-26)
+ fix broken tRNS chunk (colorkey-style transparency) in non-paletted PNG
+ 1.45 (2014-08-16)
+ fix MSVC-ARM internal compiler error by wrapping malloc
+ 1.44 (2014-08-07)
+ various warning fixes from Ronny Chevalier
+ 1.43 (2014-07-15)
+ fix MSVC-only compiler problem in code changed in 1.42
+ 1.42 (2014-07-09)
+ don't define _CRT_SECURE_NO_WARNINGS (affects user code)
+ fixes to stbi__cleanup_jpeg path
+ added STBI_ASSERT to avoid requiring assert.h
+ 1.41 (2014-06-25)
+ fix search&replace from 1.36 that messed up comments/error messages
+ 1.40 (2014-06-22)
+ fix gcc struct-initialization warning
+ 1.39 (2014-06-15)
+ fix to TGA optimization when req_comp != number of components in TGA;
+ fix to GIF loading because BMP wasn't rewinding (whoops, no GIFs in my test suite)
+ add support for BMP version 5 (more ignored fields)
+ 1.38 (2014-06-06)
+ suppress MSVC warnings on integer casts truncating values
+ fix accidental rename of 'skip' field of I/O
+ 1.37 (2014-06-04)
+ remove duplicate typedef
+ 1.36 (2014-06-03)
+ convert to header file single-file library
+ if de-iphone isn't set, load iphone images color-swapped instead of returning NULL
+ 1.35 (2014-05-27)
+ various warnings
+ fix broken STBI_SIMD path
+ fix bug where stbi_load_from_file no longer left file pointer in correct place
+ fix broken non-easy path for 32-bit BMP (possibly never used)
+ TGA optimization by Arseny Kapoulkine
+ 1.34 (unknown)
+ use STBI_NOTUSED in stbi__resample_row_generic(), fix one more leak in tga failure case
+ 1.33 (2011-07-14)
+ make stbi_is_hdr work in STBI_NO_HDR (as specified), minor compiler-friendly improvements
+ 1.32 (2011-07-13)
+ support for "info" function for all supported filetypes (SpartanJ)
+ 1.31 (2011-06-20)
+ a few more leak fixes, bug in PNG handling (SpartanJ)
+ 1.30 (2011-06-11)
+ added ability to load files via callbacks to accomidate custom input streams (Ben Wenger)
+ removed deprecated format-specific test/load functions
+ removed support for installable file formats (stbi_loader) -- would have been broken for IO callbacks anyway
+ error cases in bmp and tga give messages and don't leak (Raymond Barbiero, grisha)
+ fix inefficiency in decoding 32-bit BMP (David Woo)
+ 1.29 (2010-08-16)
+ various warning fixes from Aurelien Pocheville
+ 1.28 (2010-08-01)
+ fix bug in GIF palette transparency (SpartanJ)
+ 1.27 (2010-08-01)
+ cast-to-stbi_uc to fix warnings
+ 1.26 (2010-07-24)
+ fix bug in file buffering for PNG reported by SpartanJ
+ 1.25 (2010-07-17)
+ refix trans_data warning (Won Chun)
+ 1.24 (2010-07-12)
+ perf improvements reading from files on platforms with lock-heavy fgetc()
+ minor perf improvements for jpeg
+ deprecated type-specific functions so we'll get feedback if they're needed
+ attempt to fix trans_data warning (Won Chun)
+ 1.23 fixed bug in iPhone support
+ 1.22 (2010-07-10)
+ removed image *writing* support
+ stbi_info support from Jetro Lauha
+ GIF support from Jean-Marc Lienher
+ iPhone PNG-extensions from James Brown
+ warning-fixes from Nicolas Schulz and Janez Zemva (i.stbi__err. Janez (U+017D)emva)
+ 1.21 fix use of 'stbi_uc' in header (reported by jon blow)
+ 1.20 added support for Softimage PIC, by Tom Seddon
+ 1.19 bug in interlaced PNG corruption check (found by ryg)
+ 1.18 (2008-08-02)
+ fix a threading bug (local mutable static)
+ 1.17 support interlaced PNG
+ 1.16 major bugfix - stbi__convert_format converted one too many pixels
+ 1.15 initialize some fields for thread safety
+ 1.14 fix threadsafe conversion bug
+ header-file-only version (#define STBI_HEADER_FILE_ONLY before including)
+ 1.13 threadsafe
+ 1.12 const qualifiers in the API
+ 1.11 Support installable IDCT, colorspace conversion routines
+ 1.10 Fixes for 64-bit (don't use "unsigned long")
+ optimized upsampling by Fabian "ryg" Giesen
+ 1.09 Fix format-conversion for PSD code (bad global variables!)
+ 1.08 Thatcher Ulrich's PSD code integrated by Nicolas Schulz
+ 1.07 attempt to fix C++ warning/errors again
+ 1.06 attempt to fix C++ warning/errors again
+ 1.05 fix TGA loading to return correct *comp and use good luminance calc
+ 1.04 default float alpha is 1, not 255; use 'void *' for stbi_image_free
+ 1.03 bugfixes to STBI_NO_STDIO, STBI_NO_HDR
+ 1.02 support for (subset of) HDR files, float interface for preferred access to them
+ 1.01 fix bug: possible bug in handling right-side up bmps... not sure
+ fix bug: the stbi__bmp_load() and stbi__tga_load() functions didn't work at all
+ 1.00 interface to zlib that skips zlib header
+ 0.99 correct handling of alpha in palette
+ 0.98 TGA loader by lonesock; dynamically add loaders (untested)
+ 0.97 jpeg errors on too large a file; also catch another malloc failure
+ 0.96 fix detection of invalid v value - particleman@mollyrocket forum
+ 0.95 during header scan, seek to markers in case of padding
+ 0.94 STBI_NO_STDIO to disable stdio usage; rename all #defines the same
+ 0.93 handle jpegtran output; verbose errors
+ 0.92 read 4,8,16,24,32-bit BMP files of several formats
+ 0.91 output 24-bit Windows 3.0 BMP files
+ 0.90 fix a few more warnings; bump version number to approach 1.0
+ 0.61 bugfixes due to Marc LeBlanc, Christopher Lloyd
+ 0.60 fix compiling as c++
+ 0.59 fix warnings: merge Dave Moore's -Wall fixes
+ 0.58 fix bug: zlib uncompressed mode len/nlen was wrong endian
+ 0.57 fix bug: jpg last huffman symbol before marker was >9 bits but less than 16 available
+ 0.56 fix bug: zlib uncompressed mode len vs. nlen
+ 0.55 fix bug: restart_interval not initialized to 0
+ 0.54 allow NULL for 'int *comp'
+ 0.53 fix bug in png 3->4; speedup png decoding
+ 0.52 png handles req_comp=3,4 directly; minor cleanup; jpeg comments
+ 0.51 obey req_comp requests, 1-component jpegs return as 1-component,
+ on 'test' only check type, not whether we support this variant
+ 0.50 (2006-11-19)
+ first released version
+*/
+
+
+/*
+------------------------------------------------------------------------------
+This software is available under 2 licenses -- choose whichever you prefer.
+------------------------------------------------------------------------------
+ALTERNATIVE A - MIT License
+Copyright (c) 2017 Sean Barrett
+Permission is hereby granted, free of charge, to any person obtaining a copy of
+this software and associated documentation files (the "Software"), to deal in
+the Software without restriction, including without limitation the rights to
+use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
+of the Software, and to permit persons to whom the Software is furnished to do
+so, subject to the following conditions:
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+------------------------------------------------------------------------------
+ALTERNATIVE B - Public Domain (www.unlicense.org)
+This is free and unencumbered software released into the public domain.
+Anyone is free to copy, modify, publish, use, compile, sell, or distribute this
+software, either in source code form or as a compiled binary, for any purpose,
+commercial or non-commercial, and by any means.
+In jurisdictions that recognize copyright laws, the author or authors of this
+software dedicate any and all copyright interest in the software to the public
+domain. We make this dedication for the benefit of the public at large and to
+the detriment of our heirs and successors. We intend this dedication to be an
+overt act of relinquishment in perpetuity of all present and future rights to
+this software under copyright law.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+------------------------------------------------------------------------------
+*/
--- /dev/null
+/* stb_image_resize - v0.94 - public domain image resizing
+ by Jorge L Rodriguez (@VinoBS) - 2014
+ http://github.com/nothings/stb
+
+ Written with emphasis on usability, portability, and efficiency. (No
+ SIMD or threads, so it be easily outperformed by libs that use those.)
+ Only scaling and translation is supported, no rotations or shears.
+ Easy API downsamples w/Mitchell filter, upsamples w/cubic interpolation.
+
+ COMPILING & LINKING
+ In one C/C++ file that #includes this file, do this:
+ #define STB_IMAGE_RESIZE_IMPLEMENTATION
+ before the #include. That will create the implementation in that file.
+
+ QUICKSTART
+ stbir_resize_uint8( input_pixels , in_w , in_h , 0,
+ output_pixels, out_w, out_h, 0, num_channels)
+ stbir_resize_float(...)
+ stbir_resize_uint8_srgb( input_pixels , in_w , in_h , 0,
+ output_pixels, out_w, out_h, 0,
+ num_channels , alpha_chan , 0)
+ stbir_resize_uint8_srgb_edgemode(
+ input_pixels , in_w , in_h , 0,
+ output_pixels, out_w, out_h, 0,
+ num_channels , alpha_chan , 0, STBIR_EDGE_CLAMP)
+ // WRAP/REFLECT/ZERO
+
+ FULL API
+ See the "header file" section of the source for API documentation.
+
+ ADDITIONAL DOCUMENTATION
+
+ SRGB & FLOATING POINT REPRESENTATION
+ The sRGB functions presume IEEE floating point. If you do not have
+ IEEE floating point, define STBIR_NON_IEEE_FLOAT. This will use
+ a slower implementation.
+
+ MEMORY ALLOCATION
+ The resize functions here perform a single memory allocation using
+ malloc. To control the memory allocation, before the #include that
+ triggers the implementation, do:
+
+ #define STBIR_MALLOC(size,context) ...
+ #define STBIR_FREE(ptr,context) ...
+
+ Each resize function makes exactly one call to malloc/free, so to use
+ temp memory, store the temp memory in the context and return that.
+
+ ASSERT
+ Define STBIR_ASSERT(boolval) to override assert() and not use assert.h
+
+ OPTIMIZATION
+ Define STBIR_SATURATE_INT to compute clamp values in-range using
+ integer operations instead of float operations. This may be faster
+ on some platforms.
+
+ DEFAULT FILTERS
+ For functions which don't provide explicit control over what filters
+ to use, you can change the compile-time defaults with
+
+ #define STBIR_DEFAULT_FILTER_UPSAMPLE STBIR_FILTER_something
+ #define STBIR_DEFAULT_FILTER_DOWNSAMPLE STBIR_FILTER_something
+
+ See stbir_filter in the header-file section for the list of filters.
+
+ NEW FILTERS
+ A number of 1D filter kernels are used. For a list of
+ supported filters see the stbir_filter enum. To add a new filter,
+ write a filter function and add it to stbir__filter_info_table.
+
+ PROGRESS
+ For interactive use with slow resize operations, you can install
+ a progress-report callback:
+
+ #define STBIR_PROGRESS_REPORT(val) some_func(val)
+
+ The parameter val is a float which goes from 0 to 1 as progress is made.
+
+ For example:
+
+ static void my_progress_report(float progress);
+ #define STBIR_PROGRESS_REPORT(val) my_progress_report(val)
+
+ #define STB_IMAGE_RESIZE_IMPLEMENTATION
+ #include "stb_image_resize.h"
+
+ static void my_progress_report(float progress)
+ {
+ printf("Progress: %f%%\n", progress*100);
+ }
+
+ MAX CHANNELS
+ If your image has more than 64 channels, define STBIR_MAX_CHANNELS
+ to the max you'll have.
+
+ ALPHA CHANNEL
+ Most of the resizing functions provide the ability to control how
+ the alpha channel of an image is processed. The important things
+ to know about this:
+
+ 1. The best mathematically-behaved version of alpha to use is
+ called "premultiplied alpha", in which the other color channels
+ have had the alpha value multiplied in. If you use premultiplied
+ alpha, linear filtering (such as image resampling done by this
+ library, or performed in texture units on GPUs) does the "right
+ thing". While premultiplied alpha is standard in the movie CGI
+ industry, it is still uncommon in the videogame/real-time world.
+
+ If you linearly filter non-premultiplied alpha, strange effects
+ occur. (For example, the 50/50 average of 99% transparent bright green
+ and 1% transparent black produces 50% transparent dark green when
+ non-premultiplied, whereas premultiplied it produces 50%
+ transparent near-black. The former introduces green energy
+ that doesn't exist in the source image.)
+
+ 2. Artists should not edit premultiplied-alpha images; artists
+ want non-premultiplied alpha images. Thus, art tools generally output
+ non-premultiplied alpha images.
+
+ 3. You will get best results in most cases by converting images
+ to premultiplied alpha before processing them mathematically.
+
+ 4. If you pass the flag STBIR_FLAG_ALPHA_PREMULTIPLIED, the
+ resizer does not do anything special for the alpha channel;
+ it is resampled identically to other channels. This produces
+ the correct results for premultiplied-alpha images, but produces
+ less-than-ideal results for non-premultiplied-alpha images.
+
+ 5. If you do not pass the flag STBIR_FLAG_ALPHA_PREMULTIPLIED,
+ then the resizer weights the contribution of input pixels
+ based on their alpha values, or, equivalently, it multiplies
+ the alpha value into the color channels, resamples, then divides
+ by the resultant alpha value. Input pixels which have alpha=0 do
+ not contribute at all to output pixels unless _all_ of the input
+ pixels affecting that output pixel have alpha=0, in which case
+ the result for that pixel is the same as it would be without
+ STBIR_FLAG_ALPHA_PREMULTIPLIED. However, this is only true for
+ input images in integer formats. For input images in float format,
+ input pixels with alpha=0 have no effect, and output pixels
+ which have alpha=0 will be 0 in all channels. (For float images,
+ you can manually achieve the same result by adding a tiny epsilon
+ value to the alpha channel of every image, and then subtracting
+ or clamping it at the end.)
+
+ 6. You can suppress the behavior described in #5 and make
+ all-0-alpha pixels have 0 in all channels by #defining
+ STBIR_NO_ALPHA_EPSILON.
+
+ 7. You can separately control whether the alpha channel is
+ interpreted as linear or affected by the colorspace. By default
+ it is linear; you almost never want to apply the colorspace.
+ (For example, graphics hardware does not apply sRGB conversion
+ to the alpha channel.)
+
+ CONTRIBUTORS
+ Jorge L Rodriguez: Implementation
+ Sean Barrett: API design, optimizations
+ Aras Pranckevicius: bugfix
+
+ REVISIONS
+ 0.94 (2017-03-18) fixed warnings
+ 0.93 (2017-03-03) fixed bug with certain combinations of heights
+ 0.92 (2017-01-02) fix integer overflow on large (>2GB) images
+ 0.91 (2016-04-02) fix warnings; fix handling of subpixel regions
+ 0.90 (2014-09-17) first released version
+
+ LICENSE
+ See end of file for license information.
+
+ TODO
+ Don't decode all of the image data when only processing a partial tile
+ Don't use full-width decode buffers when only processing a partial tile
+ When processing wide images, break processing into tiles so data fits in L1 cache
+ Installable filters?
+ Resize that respects alpha test coverage
+ (Reference code: FloatImage::alphaTestCoverage and FloatImage::scaleAlphaToCoverage:
+ https://code.google.com/p/nvidia-texture-tools/source/browse/trunk/src/nvimage/FloatImage.cpp )
+*/
+
+#ifndef STBIR_INCLUDE_STB_IMAGE_RESIZE_H
+#define STBIR_INCLUDE_STB_IMAGE_RESIZE_H
+
+#ifdef _MSC_VER
+typedef unsigned char stbir_uint8;
+typedef unsigned short stbir_uint16;
+typedef unsigned int stbir_uint32;
+#else
+#include <stdint.h>
+typedef uint8_t stbir_uint8;
+typedef uint16_t stbir_uint16;
+typedef uint32_t stbir_uint32;
+#endif
+
+#ifdef STB_IMAGE_RESIZE_STATIC
+#define STBIRDEF static
+#else
+#ifdef __cplusplus
+#define STBIRDEF extern "C"
+#else
+#define STBIRDEF extern
+#endif
+#endif
+
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// Easy-to-use API:
+//
+// * "input pixels" points to an array of image data with 'num_channels' channels (e.g. RGB=3, RGBA=4)
+// * input_w is input image width (x-axis), input_h is input image height (y-axis)
+// * stride is the offset between successive rows of image data in memory, in bytes. you can
+// specify 0 to mean packed continuously in memory
+// * alpha channel is treated identically to other channels.
+// * colorspace is linear or sRGB as specified by function name
+// * returned result is 1 for success or 0 in case of an error.
+// #define STBIR_ASSERT() to trigger an assert on parameter validation errors.
+// * Memory required grows approximately linearly with input and output size, but with
+// discontinuities at input_w == output_w and input_h == output_h.
+// * These functions use a "default" resampling filter defined at compile time. To change the filter,
+// you can change the compile-time defaults by #defining STBIR_DEFAULT_FILTER_UPSAMPLE
+// and STBIR_DEFAULT_FILTER_DOWNSAMPLE, or you can use the medium-complexity API.
+
+STBIRDEF int stbir_resize_uint8( const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels);
+
+STBIRDEF int stbir_resize_float( const float *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ float *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels);
+
+
+// The following functions interpret image data as gamma-corrected sRGB.
+// Specify STBIR_ALPHA_CHANNEL_NONE if you have no alpha channel,
+// or otherwise provide the index of the alpha channel. Flags value
+// of 0 will probably do the right thing if you're not sure what
+// the flags mean.
+
+#define STBIR_ALPHA_CHANNEL_NONE -1
+
+// Set this flag if your texture has premultiplied alpha. Otherwise, stbir will
+// use alpha-weighted resampling (effectively premultiplying, resampling,
+// then unpremultiplying).
+#define STBIR_FLAG_ALPHA_PREMULTIPLIED (1 << 0)
+// The specified alpha channel should be handled as gamma-corrected value even
+// when doing sRGB operations.
+#define STBIR_FLAG_ALPHA_USES_COLORSPACE (1 << 1)
+
+STBIRDEF int stbir_resize_uint8_srgb(const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags);
+
+
+typedef enum
+{
+ STBIR_EDGE_CLAMP = 1,
+ STBIR_EDGE_REFLECT = 2,
+ STBIR_EDGE_WRAP = 3,
+ STBIR_EDGE_ZERO = 4,
+} stbir_edge;
+
+// This function adds the ability to specify how requests to sample off the edge of the image are handled.
+STBIRDEF int stbir_resize_uint8_srgb_edgemode(const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode);
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// Medium-complexity API
+//
+// This extends the easy-to-use API as follows:
+//
+// * Alpha-channel can be processed separately
+// * If alpha_channel is not STBIR_ALPHA_CHANNEL_NONE
+// * Alpha channel will not be gamma corrected (unless flags&STBIR_FLAG_GAMMA_CORRECT)
+// * Filters will be weighted by alpha channel (unless flags&STBIR_FLAG_ALPHA_PREMULTIPLIED)
+// * Filter can be selected explicitly
+// * uint16 image type
+// * sRGB colorspace available for all types
+// * context parameter for passing to STBIR_MALLOC
+
+typedef enum
+{
+ STBIR_FILTER_DEFAULT = 0, // use same filter type that easy-to-use API chooses
+ STBIR_FILTER_BOX = 1, // A trapezoid w/1-pixel wide ramps, same result as box for integer scale ratios
+ STBIR_FILTER_TRIANGLE = 2, // On upsampling, produces same results as bilinear texture filtering
+ STBIR_FILTER_CUBICBSPLINE = 3, // The cubic b-spline (aka Mitchell-Netrevalli with B=1,C=0), gaussian-esque
+ STBIR_FILTER_CATMULLROM = 4, // An interpolating cubic spline
+ STBIR_FILTER_MITCHELL = 5, // Mitchell-Netrevalli filter with B=1/3, C=1/3
+} stbir_filter;
+
+typedef enum
+{
+ STBIR_COLORSPACE_LINEAR,
+ STBIR_COLORSPACE_SRGB,
+
+ STBIR_MAX_COLORSPACES,
+} stbir_colorspace;
+
+// The following functions are all identical except for the type of the image data
+
+STBIRDEF int stbir_resize_uint8_generic( const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space,
+ void *alloc_context);
+
+STBIRDEF int stbir_resize_uint16_generic(const stbir_uint16 *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ stbir_uint16 *output_pixels , int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space,
+ void *alloc_context);
+
+STBIRDEF int stbir_resize_float_generic( const float *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ float *output_pixels , int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space,
+ void *alloc_context);
+
+
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// Full-complexity API
+//
+// This extends the medium API as follows:
+//
+// * uint32 image type
+// * not typesafe
+// * separate filter types for each axis
+// * separate edge modes for each axis
+// * can specify scale explicitly for subpixel correctness
+// * can specify image source tile using texture coordinates
+
+typedef enum
+{
+ STBIR_TYPE_UINT8 ,
+ STBIR_TYPE_UINT16,
+ STBIR_TYPE_UINT32,
+ STBIR_TYPE_FLOAT ,
+
+ STBIR_MAX_TYPES
+} stbir_datatype;
+
+STBIRDEF int stbir_resize( const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ stbir_datatype datatype,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical,
+ stbir_filter filter_horizontal, stbir_filter filter_vertical,
+ stbir_colorspace space, void *alloc_context);
+
+STBIRDEF int stbir_resize_subpixel(const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ stbir_datatype datatype,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical,
+ stbir_filter filter_horizontal, stbir_filter filter_vertical,
+ stbir_colorspace space, void *alloc_context,
+ float x_scale, float y_scale,
+ float x_offset, float y_offset);
+
+STBIRDEF int stbir_resize_region( const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ stbir_datatype datatype,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical,
+ stbir_filter filter_horizontal, stbir_filter filter_vertical,
+ stbir_colorspace space, void *alloc_context,
+ float s0, float t0, float s1, float t1);
+// (s0, t0) & (s1, t1) are the top-left and bottom right corner (uv addressing style: [0, 1]x[0, 1]) of a region of the input image to use.
+
+//
+//
+//// end header file /////////////////////////////////////////////////////
+#endif // STBIR_INCLUDE_STB_IMAGE_RESIZE_H
+
+
+
+
+
+#ifdef STB_IMAGE_RESIZE_IMPLEMENTATION
+
+#ifndef STBIR_ASSERT
+#include <assert.h>
+#define STBIR_ASSERT(x) assert(x)
+#endif
+
+// For memset
+#include <string.h>
+
+#include <math.h>
+
+#ifndef STBIR_MALLOC
+#include <stdlib.h>
+#define STBIR_MALLOC(size,c) malloc(size)
+#define STBIR_FREE(ptr,c) free(ptr)
+#endif
+
+#ifndef _MSC_VER
+#ifdef __cplusplus
+#define stbir__inline inline
+#else
+#define stbir__inline
+#endif
+#else
+#define stbir__inline __forceinline
+#endif
+
+
+// should produce compiler error if size is wrong
+typedef unsigned char stbir__validate_uint32[sizeof(stbir_uint32) == 4 ? 1 : -1];
+
+#ifdef _MSC_VER
+#define STBIR__NOTUSED(v) (void)(v)
+#else
+#define STBIR__NOTUSED(v) (void)sizeof(v)
+#endif
+
+#define STBIR__ARRAY_SIZE(a) (sizeof((a))/sizeof((a)[0]))
+
+#ifndef STBIR_DEFAULT_FILTER_UPSAMPLE
+#define STBIR_DEFAULT_FILTER_UPSAMPLE STBIR_FILTER_CATMULLROM
+#endif
+
+#ifndef STBIR_DEFAULT_FILTER_DOWNSAMPLE
+#define STBIR_DEFAULT_FILTER_DOWNSAMPLE STBIR_FILTER_MITCHELL
+#endif
+
+#ifndef STBIR_PROGRESS_REPORT
+#define STBIR_PROGRESS_REPORT(float_0_to_1)
+#endif
+
+#ifndef STBIR_MAX_CHANNELS
+#define STBIR_MAX_CHANNELS 64
+#endif
+
+#if STBIR_MAX_CHANNELS > 65536
+#error "Too many channels; STBIR_MAX_CHANNELS must be no more than 65536."
+// because we store the indices in 16-bit variables
+#endif
+
+// This value is added to alpha just before premultiplication to avoid
+// zeroing out color values. It is equivalent to 2^-80. If you don't want
+// that behavior (it may interfere if you have floating point images with
+// very small alpha values) then you can define STBIR_NO_ALPHA_EPSILON to
+// disable it.
+#ifndef STBIR_ALPHA_EPSILON
+#define STBIR_ALPHA_EPSILON ((float)1 / (1 << 20) / (1 << 20) / (1 << 20) / (1 << 20))
+#endif
+
+
+
+#ifdef _MSC_VER
+#define STBIR__UNUSED_PARAM(v) (void)(v)
+#else
+#define STBIR__UNUSED_PARAM(v) (void)sizeof(v)
+#endif
+
+// must match stbir_datatype
+static unsigned char stbir__type_size[] = {
+ 1, // STBIR_TYPE_UINT8
+ 2, // STBIR_TYPE_UINT16
+ 4, // STBIR_TYPE_UINT32
+ 4, // STBIR_TYPE_FLOAT
+};
+
+// Kernel function centered at 0
+typedef float (stbir__kernel_fn)(float x, float scale);
+typedef float (stbir__support_fn)(float scale);
+
+typedef struct
+{
+ stbir__kernel_fn* kernel;
+ stbir__support_fn* support;
+} stbir__filter_info;
+
+// When upsampling, the contributors are which source pixels contribute.
+// When downsampling, the contributors are which destination pixels are contributed to.
+typedef struct
+{
+ int n0; // First contributing pixel
+ int n1; // Last contributing pixel
+} stbir__contributors;
+
+typedef struct
+{
+ const void* input_data;
+ int input_w;
+ int input_h;
+ int input_stride_bytes;
+
+ void* output_data;
+ int output_w;
+ int output_h;
+ int output_stride_bytes;
+
+ float s0, t0, s1, t1;
+
+ float horizontal_shift; // Units: output pixels
+ float vertical_shift; // Units: output pixels
+ float horizontal_scale;
+ float vertical_scale;
+
+ int channels;
+ int alpha_channel;
+ stbir_uint32 flags;
+ stbir_datatype type;
+ stbir_filter horizontal_filter;
+ stbir_filter vertical_filter;
+ stbir_edge edge_horizontal;
+ stbir_edge edge_vertical;
+ stbir_colorspace colorspace;
+
+ stbir__contributors* horizontal_contributors;
+ float* horizontal_coefficients;
+
+ stbir__contributors* vertical_contributors;
+ float* vertical_coefficients;
+
+ int decode_buffer_pixels;
+ float* decode_buffer;
+
+ float* horizontal_buffer;
+
+ // cache these because ceil/floor are inexplicably showing up in profile
+ int horizontal_coefficient_width;
+ int vertical_coefficient_width;
+ int horizontal_filter_pixel_width;
+ int vertical_filter_pixel_width;
+ int horizontal_filter_pixel_margin;
+ int vertical_filter_pixel_margin;
+ int horizontal_num_contributors;
+ int vertical_num_contributors;
+
+ int ring_buffer_length_bytes; // The length of an individual entry in the ring buffer. The total number of ring buffers is stbir__get_filter_pixel_width(filter)
+ int ring_buffer_num_entries; // Total number of entries in the ring buffer.
+ int ring_buffer_first_scanline;
+ int ring_buffer_last_scanline;
+ int ring_buffer_begin_index; // first_scanline is at this index in the ring buffer
+ float* ring_buffer;
+
+ float* encode_buffer; // A temporary buffer to store floats so we don't lose precision while we do multiply-adds.
+
+ int horizontal_contributors_size;
+ int horizontal_coefficients_size;
+ int vertical_contributors_size;
+ int vertical_coefficients_size;
+ int decode_buffer_size;
+ int horizontal_buffer_size;
+ int ring_buffer_size;
+ int encode_buffer_size;
+} stbir__info;
+
+
+static const float stbir__max_uint8_as_float = 255.0f;
+static const float stbir__max_uint16_as_float = 65535.0f;
+static const double stbir__max_uint32_as_float = 4294967295.0;
+
+
+static stbir__inline int stbir__min(int a, int b)
+{
+ return a < b ? a : b;
+}
+
+static stbir__inline float stbir__saturate(float x)
+{
+ if (x < 0)
+ return 0;
+
+ if (x > 1)
+ return 1;
+
+ return x;
+}
+
+#ifdef STBIR_SATURATE_INT
+static stbir__inline stbir_uint8 stbir__saturate8(int x)
+{
+ if ((unsigned int) x <= 255)
+ return x;
+
+ if (x < 0)
+ return 0;
+
+ return 255;
+}
+
+static stbir__inline stbir_uint16 stbir__saturate16(int x)
+{
+ if ((unsigned int) x <= 65535)
+ return x;
+
+ if (x < 0)
+ return 0;
+
+ return 65535;
+}
+#endif
+
+static float stbir__srgb_uchar_to_linear_float[256] = {
+ 0.000000f, 0.000304f, 0.000607f, 0.000911f, 0.001214f, 0.001518f, 0.001821f, 0.002125f, 0.002428f, 0.002732f, 0.003035f,
+ 0.003347f, 0.003677f, 0.004025f, 0.004391f, 0.004777f, 0.005182f, 0.005605f, 0.006049f, 0.006512f, 0.006995f, 0.007499f,
+ 0.008023f, 0.008568f, 0.009134f, 0.009721f, 0.010330f, 0.010960f, 0.011612f, 0.012286f, 0.012983f, 0.013702f, 0.014444f,
+ 0.015209f, 0.015996f, 0.016807f, 0.017642f, 0.018500f, 0.019382f, 0.020289f, 0.021219f, 0.022174f, 0.023153f, 0.024158f,
+ 0.025187f, 0.026241f, 0.027321f, 0.028426f, 0.029557f, 0.030713f, 0.031896f, 0.033105f, 0.034340f, 0.035601f, 0.036889f,
+ 0.038204f, 0.039546f, 0.040915f, 0.042311f, 0.043735f, 0.045186f, 0.046665f, 0.048172f, 0.049707f, 0.051269f, 0.052861f,
+ 0.054480f, 0.056128f, 0.057805f, 0.059511f, 0.061246f, 0.063010f, 0.064803f, 0.066626f, 0.068478f, 0.070360f, 0.072272f,
+ 0.074214f, 0.076185f, 0.078187f, 0.080220f, 0.082283f, 0.084376f, 0.086500f, 0.088656f, 0.090842f, 0.093059f, 0.095307f,
+ 0.097587f, 0.099899f, 0.102242f, 0.104616f, 0.107023f, 0.109462f, 0.111932f, 0.114435f, 0.116971f, 0.119538f, 0.122139f,
+ 0.124772f, 0.127438f, 0.130136f, 0.132868f, 0.135633f, 0.138432f, 0.141263f, 0.144128f, 0.147027f, 0.149960f, 0.152926f,
+ 0.155926f, 0.158961f, 0.162029f, 0.165132f, 0.168269f, 0.171441f, 0.174647f, 0.177888f, 0.181164f, 0.184475f, 0.187821f,
+ 0.191202f, 0.194618f, 0.198069f, 0.201556f, 0.205079f, 0.208637f, 0.212231f, 0.215861f, 0.219526f, 0.223228f, 0.226966f,
+ 0.230740f, 0.234551f, 0.238398f, 0.242281f, 0.246201f, 0.250158f, 0.254152f, 0.258183f, 0.262251f, 0.266356f, 0.270498f,
+ 0.274677f, 0.278894f, 0.283149f, 0.287441f, 0.291771f, 0.296138f, 0.300544f, 0.304987f, 0.309469f, 0.313989f, 0.318547f,
+ 0.323143f, 0.327778f, 0.332452f, 0.337164f, 0.341914f, 0.346704f, 0.351533f, 0.356400f, 0.361307f, 0.366253f, 0.371238f,
+ 0.376262f, 0.381326f, 0.386430f, 0.391573f, 0.396755f, 0.401978f, 0.407240f, 0.412543f, 0.417885f, 0.423268f, 0.428691f,
+ 0.434154f, 0.439657f, 0.445201f, 0.450786f, 0.456411f, 0.462077f, 0.467784f, 0.473532f, 0.479320f, 0.485150f, 0.491021f,
+ 0.496933f, 0.502887f, 0.508881f, 0.514918f, 0.520996f, 0.527115f, 0.533276f, 0.539480f, 0.545725f, 0.552011f, 0.558340f,
+ 0.564712f, 0.571125f, 0.577581f, 0.584078f, 0.590619f, 0.597202f, 0.603827f, 0.610496f, 0.617207f, 0.623960f, 0.630757f,
+ 0.637597f, 0.644480f, 0.651406f, 0.658375f, 0.665387f, 0.672443f, 0.679543f, 0.686685f, 0.693872f, 0.701102f, 0.708376f,
+ 0.715694f, 0.723055f, 0.730461f, 0.737911f, 0.745404f, 0.752942f, 0.760525f, 0.768151f, 0.775822f, 0.783538f, 0.791298f,
+ 0.799103f, 0.806952f, 0.814847f, 0.822786f, 0.830770f, 0.838799f, 0.846873f, 0.854993f, 0.863157f, 0.871367f, 0.879622f,
+ 0.887923f, 0.896269f, 0.904661f, 0.913099f, 0.921582f, 0.930111f, 0.938686f, 0.947307f, 0.955974f, 0.964686f, 0.973445f,
+ 0.982251f, 0.991102f, 1.0f
+};
+
+static float stbir__srgb_to_linear(float f)
+{
+ if (f <= 0.04045f)
+ return f / 12.92f;
+ else
+ return (float)pow((f + 0.055f) / 1.055f, 2.4f);
+}
+
+static float stbir__linear_to_srgb(float f)
+{
+ if (f <= 0.0031308f)
+ return f * 12.92f;
+ else
+ return 1.055f * (float)pow(f, 1 / 2.4f) - 0.055f;
+}
+
+#ifndef STBIR_NON_IEEE_FLOAT
+// From https://gist.github.com/rygorous/2203834
+
+typedef union
+{
+ stbir_uint32 u;
+ float f;
+} stbir__FP32;
+
+static const stbir_uint32 fp32_to_srgb8_tab4[104] = {
+ 0x0073000d, 0x007a000d, 0x0080000d, 0x0087000d, 0x008d000d, 0x0094000d, 0x009a000d, 0x00a1000d,
+ 0x00a7001a, 0x00b4001a, 0x00c1001a, 0x00ce001a, 0x00da001a, 0x00e7001a, 0x00f4001a, 0x0101001a,
+ 0x010e0033, 0x01280033, 0x01410033, 0x015b0033, 0x01750033, 0x018f0033, 0x01a80033, 0x01c20033,
+ 0x01dc0067, 0x020f0067, 0x02430067, 0x02760067, 0x02aa0067, 0x02dd0067, 0x03110067, 0x03440067,
+ 0x037800ce, 0x03df00ce, 0x044600ce, 0x04ad00ce, 0x051400ce, 0x057b00c5, 0x05dd00bc, 0x063b00b5,
+ 0x06970158, 0x07420142, 0x07e30130, 0x087b0120, 0x090b0112, 0x09940106, 0x0a1700fc, 0x0a9500f2,
+ 0x0b0f01cb, 0x0bf401ae, 0x0ccb0195, 0x0d950180, 0x0e56016e, 0x0f0d015e, 0x0fbc0150, 0x10630143,
+ 0x11070264, 0x1238023e, 0x1357021d, 0x14660201, 0x156601e9, 0x165a01d3, 0x174401c0, 0x182401af,
+ 0x18fe0331, 0x1a9602fe, 0x1c1502d2, 0x1d7e02ad, 0x1ed4028d, 0x201a0270, 0x21520256, 0x227d0240,
+ 0x239f0443, 0x25c003fe, 0x27bf03c4, 0x29a10392, 0x2b6a0367, 0x2d1d0341, 0x2ebe031f, 0x304d0300,
+ 0x31d105b0, 0x34a80555, 0x37520507, 0x39d504c5, 0x3c37048b, 0x3e7c0458, 0x40a8042a, 0x42bd0401,
+ 0x44c20798, 0x488e071e, 0x4c1c06b6, 0x4f76065d, 0x52a50610, 0x55ac05cc, 0x5892058f, 0x5b590559,
+ 0x5e0c0a23, 0x631c0980, 0x67db08f6, 0x6c55087f, 0x70940818, 0x74a007bd, 0x787d076c, 0x7c330723,
+};
+
+static stbir_uint8 stbir__linear_to_srgb_uchar(float in)
+{
+ static const stbir__FP32 almostone = { 0x3f7fffff }; // 1-eps
+ static const stbir__FP32 minval = { (127-13) << 23 };
+ stbir_uint32 tab,bias,scale,t;
+ stbir__FP32 f;
+
+ // Clamp to [2^(-13), 1-eps]; these two values map to 0 and 1, respectively.
+ // The tests are carefully written so that NaNs map to 0, same as in the reference
+ // implementation.
+ if (!(in > minval.f)) // written this way to catch NaNs
+ in = minval.f;
+ if (in > almostone.f)
+ in = almostone.f;
+
+ // Do the table lookup and unpack bias, scale
+ f.f = in;
+ tab = fp32_to_srgb8_tab4[(f.u - minval.u) >> 20];
+ bias = (tab >> 16) << 9;
+ scale = tab & 0xffff;
+
+ // Grab next-highest mantissa bits and perform linear interpolation
+ t = (f.u >> 12) & 0xff;
+ return (unsigned char) ((bias + scale*t) >> 16);
+}
+
+#else
+// sRGB transition values, scaled by 1<<28
+static int stbir__srgb_offset_to_linear_scaled[256] =
+{
+ 0, 40738, 122216, 203693, 285170, 366648, 448125, 529603,
+ 611080, 692557, 774035, 855852, 942009, 1033024, 1128971, 1229926,
+ 1335959, 1447142, 1563542, 1685229, 1812268, 1944725, 2082664, 2226148,
+ 2375238, 2529996, 2690481, 2856753, 3028870, 3206888, 3390865, 3580856,
+ 3776916, 3979100, 4187460, 4402049, 4622919, 4850123, 5083710, 5323731,
+ 5570236, 5823273, 6082892, 6349140, 6622065, 6901714, 7188133, 7481369,
+ 7781466, 8088471, 8402427, 8723380, 9051372, 9386448, 9728650, 10078021,
+ 10434603, 10798439, 11169569, 11548036, 11933879, 12327139, 12727857, 13136073,
+ 13551826, 13975156, 14406100, 14844697, 15290987, 15745007, 16206795, 16676389,
+ 17153826, 17639142, 18132374, 18633560, 19142734, 19659934, 20185196, 20718552,
+ 21260042, 21809696, 22367554, 22933648, 23508010, 24090680, 24681686, 25281066,
+ 25888850, 26505076, 27129772, 27762974, 28404716, 29055026, 29713942, 30381490,
+ 31057708, 31742624, 32436272, 33138682, 33849884, 34569912, 35298800, 36036568,
+ 36783260, 37538896, 38303512, 39077136, 39859796, 40651528, 41452360, 42262316,
+ 43081432, 43909732, 44747252, 45594016, 46450052, 47315392, 48190064, 49074096,
+ 49967516, 50870356, 51782636, 52704392, 53635648, 54576432, 55526772, 56486700,
+ 57456236, 58435408, 59424248, 60422780, 61431036, 62449032, 63476804, 64514376,
+ 65561776, 66619028, 67686160, 68763192, 69850160, 70947088, 72053992, 73170912,
+ 74297864, 75434880, 76581976, 77739184, 78906536, 80084040, 81271736, 82469648,
+ 83677792, 84896192, 86124888, 87363888, 88613232, 89872928, 91143016, 92423512,
+ 93714432, 95015816, 96327688, 97650056, 98982952, 100326408, 101680440, 103045072,
+ 104420320, 105806224, 107202800, 108610064, 110028048, 111456776, 112896264, 114346544,
+ 115807632, 117279552, 118762328, 120255976, 121760536, 123276016, 124802440, 126339832,
+ 127888216, 129447616, 131018048, 132599544, 134192112, 135795792, 137410592, 139036528,
+ 140673648, 142321952, 143981456, 145652208, 147334208, 149027488, 150732064, 152447968,
+ 154175200, 155913792, 157663776, 159425168, 161197984, 162982240, 164777968, 166585184,
+ 168403904, 170234160, 172075968, 173929344, 175794320, 177670896, 179559120, 181458992,
+ 183370528, 185293776, 187228736, 189175424, 191133888, 193104112, 195086128, 197079968,
+ 199085648, 201103184, 203132592, 205173888, 207227120, 209292272, 211369392, 213458480,
+ 215559568, 217672656, 219797792, 221934976, 224084240, 226245600, 228419056, 230604656,
+ 232802400, 235012320, 237234432, 239468736, 241715280, 243974080, 246245120, 248528464,
+ 250824112, 253132064, 255452368, 257785040, 260130080, 262487520, 264857376, 267239664,
+};
+
+static stbir_uint8 stbir__linear_to_srgb_uchar(float f)
+{
+ int x = (int) (f * (1 << 28)); // has headroom so you don't need to clamp
+ int v = 0;
+ int i;
+
+ // Refine the guess with a short binary search.
+ i = v + 128; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+ i = v + 64; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+ i = v + 32; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+ i = v + 16; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+ i = v + 8; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+ i = v + 4; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+ i = v + 2; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+ i = v + 1; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+
+ return (stbir_uint8) v;
+}
+#endif
+
+static float stbir__filter_trapezoid(float x, float scale)
+{
+ float halfscale = scale / 2;
+ float t = 0.5f + halfscale;
+ STBIR_ASSERT(scale <= 1);
+
+ x = (float)fabs(x);
+
+ if (x >= t)
+ return 0;
+ else
+ {
+ float r = 0.5f - halfscale;
+ if (x <= r)
+ return 1;
+ else
+ return (t - x) / scale;
+ }
+}
+
+static float stbir__support_trapezoid(float scale)
+{
+ STBIR_ASSERT(scale <= 1);
+ return 0.5f + scale / 2;
+}
+
+static float stbir__filter_triangle(float x, float s)
+{
+ STBIR__UNUSED_PARAM(s);
+
+ x = (float)fabs(x);
+
+ if (x <= 1.0f)
+ return 1 - x;
+ else
+ return 0;
+}
+
+static float stbir__filter_cubic(float x, float s)
+{
+ STBIR__UNUSED_PARAM(s);
+
+ x = (float)fabs(x);
+
+ if (x < 1.0f)
+ return (4 + x*x*(3*x - 6))/6;
+ else if (x < 2.0f)
+ return (8 + x*(-12 + x*(6 - x)))/6;
+
+ return (0.0f);
+}
+
+static float stbir__filter_catmullrom(float x, float s)
+{
+ STBIR__UNUSED_PARAM(s);
+
+ x = (float)fabs(x);
+
+ if (x < 1.0f)
+ return 1 - x*x*(2.5f - 1.5f*x);
+ else if (x < 2.0f)
+ return 2 - x*(4 + x*(0.5f*x - 2.5f));
+
+ return (0.0f);
+}
+
+static float stbir__filter_mitchell(float x, float s)
+{
+ STBIR__UNUSED_PARAM(s);
+
+ x = (float)fabs(x);
+
+ if (x < 1.0f)
+ return (16 + x*x*(21 * x - 36))/18;
+ else if (x < 2.0f)
+ return (32 + x*(-60 + x*(36 - 7*x)))/18;
+
+ return (0.0f);
+}
+
+static float stbir__support_zero(float s)
+{
+ STBIR__UNUSED_PARAM(s);
+ return 0;
+}
+
+static float stbir__support_one(float s)
+{
+ STBIR__UNUSED_PARAM(s);
+ return 1;
+}
+
+static float stbir__support_two(float s)
+{
+ STBIR__UNUSED_PARAM(s);
+ return 2;
+}
+
+static stbir__filter_info stbir__filter_info_table[] = {
+ { NULL, stbir__support_zero },
+ { stbir__filter_trapezoid, stbir__support_trapezoid },
+ { stbir__filter_triangle, stbir__support_one },
+ { stbir__filter_cubic, stbir__support_two },
+ { stbir__filter_catmullrom, stbir__support_two },
+ { stbir__filter_mitchell, stbir__support_two },
+};
+
+stbir__inline static int stbir__use_upsampling(float ratio)
+{
+ return ratio > 1;
+}
+
+stbir__inline static int stbir__use_width_upsampling(stbir__info* stbir_info)
+{
+ return stbir__use_upsampling(stbir_info->horizontal_scale);
+}
+
+stbir__inline static int stbir__use_height_upsampling(stbir__info* stbir_info)
+{
+ return stbir__use_upsampling(stbir_info->vertical_scale);
+}
+
+// This is the maximum number of input samples that can affect an output sample
+// with the given filter
+static int stbir__get_filter_pixel_width(stbir_filter filter, float scale)
+{
+ STBIR_ASSERT(filter != 0);
+ STBIR_ASSERT(filter < STBIR__ARRAY_SIZE(stbir__filter_info_table));
+
+ if (stbir__use_upsampling(scale))
+ return (int)ceil(stbir__filter_info_table[filter].support(1/scale) * 2);
+ else
+ return (int)ceil(stbir__filter_info_table[filter].support(scale) * 2 / scale);
+}
+
+// This is how much to expand buffers to account for filters seeking outside
+// the image boundaries.
+static int stbir__get_filter_pixel_margin(stbir_filter filter, float scale)
+{
+ return stbir__get_filter_pixel_width(filter, scale) / 2;
+}
+
+static int stbir__get_coefficient_width(stbir_filter filter, float scale)
+{
+ if (stbir__use_upsampling(scale))
+ return (int)ceil(stbir__filter_info_table[filter].support(1 / scale) * 2);
+ else
+ return (int)ceil(stbir__filter_info_table[filter].support(scale) * 2);
+}
+
+static int stbir__get_contributors(float scale, stbir_filter filter, int input_size, int output_size)
+{
+ if (stbir__use_upsampling(scale))
+ return output_size;
+ else
+ return (input_size + stbir__get_filter_pixel_margin(filter, scale) * 2);
+}
+
+static int stbir__get_total_horizontal_coefficients(stbir__info* info)
+{
+ return info->horizontal_num_contributors
+ * stbir__get_coefficient_width (info->horizontal_filter, info->horizontal_scale);
+}
+
+static int stbir__get_total_vertical_coefficients(stbir__info* info)
+{
+ return info->vertical_num_contributors
+ * stbir__get_coefficient_width (info->vertical_filter, info->vertical_scale);
+}
+
+static stbir__contributors* stbir__get_contributor(stbir__contributors* contributors, int n)
+{
+ return &contributors[n];
+}
+
+// For perf reasons this code is duplicated in stbir__resample_horizontal_upsample/downsample,
+// if you change it here change it there too.
+static float* stbir__get_coefficient(float* coefficients, stbir_filter filter, float scale, int n, int c)
+{
+ int width = stbir__get_coefficient_width(filter, scale);
+ return &coefficients[width*n + c];
+}
+
+static int stbir__edge_wrap_slow(stbir_edge edge, int n, int max)
+{
+ switch (edge)
+ {
+ case STBIR_EDGE_ZERO:
+ return 0; // we'll decode the wrong pixel here, and then overwrite with 0s later
+
+ case STBIR_EDGE_CLAMP:
+ if (n < 0)
+ return 0;
+
+ if (n >= max)
+ return max - 1;
+
+ return n; // NOTREACHED
+
+ case STBIR_EDGE_REFLECT:
+ {
+ if (n < 0)
+ {
+ if (n < max)
+ return -n;
+ else
+ return max - 1;
+ }
+
+ if (n >= max)
+ {
+ int max2 = max * 2;
+ if (n >= max2)
+ return 0;
+ else
+ return max2 - n - 1;
+ }
+
+ return n; // NOTREACHED
+ }
+
+ case STBIR_EDGE_WRAP:
+ if (n >= 0)
+ return (n % max);
+ else
+ {
+ int m = (-n) % max;
+
+ if (m != 0)
+ m = max - m;
+
+ return (m);
+ }
+ return n; // NOTREACHED
+
+ default:
+ STBIR_ASSERT(!"Unimplemented edge type");
+ return 0;
+ }
+}
+
+stbir__inline static int stbir__edge_wrap(stbir_edge edge, int n, int max)
+{
+ // avoid per-pixel switch
+ if (n >= 0 && n < max)
+ return n;
+ return stbir__edge_wrap_slow(edge, n, max);
+}
+
+// What input pixels contribute to this output pixel?
+static void stbir__calculate_sample_range_upsample(int n, float out_filter_radius, float scale_ratio, float out_shift, int* in_first_pixel, int* in_last_pixel, float* in_center_of_out)
+{
+ float out_pixel_center = (float)n + 0.5f;
+ float out_pixel_influence_lowerbound = out_pixel_center - out_filter_radius;
+ float out_pixel_influence_upperbound = out_pixel_center + out_filter_radius;
+
+ float in_pixel_influence_lowerbound = (out_pixel_influence_lowerbound + out_shift) / scale_ratio;
+ float in_pixel_influence_upperbound = (out_pixel_influence_upperbound + out_shift) / scale_ratio;
+
+ *in_center_of_out = (out_pixel_center + out_shift) / scale_ratio;
+ *in_first_pixel = (int)(floor(in_pixel_influence_lowerbound + 0.5));
+ *in_last_pixel = (int)(floor(in_pixel_influence_upperbound - 0.5));
+}
+
+// What output pixels does this input pixel contribute to?
+static void stbir__calculate_sample_range_downsample(int n, float in_pixels_radius, float scale_ratio, float out_shift, int* out_first_pixel, int* out_last_pixel, float* out_center_of_in)
+{
+ float in_pixel_center = (float)n + 0.5f;
+ float in_pixel_influence_lowerbound = in_pixel_center - in_pixels_radius;
+ float in_pixel_influence_upperbound = in_pixel_center + in_pixels_radius;
+
+ float out_pixel_influence_lowerbound = in_pixel_influence_lowerbound * scale_ratio - out_shift;
+ float out_pixel_influence_upperbound = in_pixel_influence_upperbound * scale_ratio - out_shift;
+
+ *out_center_of_in = in_pixel_center * scale_ratio - out_shift;
+ *out_first_pixel = (int)(floor(out_pixel_influence_lowerbound + 0.5));
+ *out_last_pixel = (int)(floor(out_pixel_influence_upperbound - 0.5));
+}
+
+static void stbir__calculate_coefficients_upsample(stbir_filter filter, float scale, int in_first_pixel, int in_last_pixel, float in_center_of_out, stbir__contributors* contributor, float* coefficient_group)
+{
+ int i;
+ float total_filter = 0;
+ float filter_scale;
+
+ STBIR_ASSERT(in_last_pixel - in_first_pixel <= (int)ceil(stbir__filter_info_table[filter].support(1/scale) * 2)); // Taken directly from stbir__get_coefficient_width() which we can't call because we don't know if we're horizontal or vertical.
+
+ contributor->n0 = in_first_pixel;
+ contributor->n1 = in_last_pixel;
+
+ STBIR_ASSERT(contributor->n1 >= contributor->n0);
+
+ for (i = 0; i <= in_last_pixel - in_first_pixel; i++)
+ {
+ float in_pixel_center = (float)(i + in_first_pixel) + 0.5f;
+ coefficient_group[i] = stbir__filter_info_table[filter].kernel(in_center_of_out - in_pixel_center, 1 / scale);
+
+ // If the coefficient is zero, skip it. (Don't do the <0 check here, we want the influence of those outside pixels.)
+ if (i == 0 && !coefficient_group[i])
+ {
+ contributor->n0 = ++in_first_pixel;
+ i--;
+ continue;
+ }
+
+ total_filter += coefficient_group[i];
+ }
+
+ STBIR_ASSERT(stbir__filter_info_table[filter].kernel((float)(in_last_pixel + 1) + 0.5f - in_center_of_out, 1/scale) == 0);
+
+ STBIR_ASSERT(total_filter > 0.9);
+ STBIR_ASSERT(total_filter < 1.1f); // Make sure it's not way off.
+
+ // Make sure the sum of all coefficients is 1.
+ filter_scale = 1 / total_filter;
+
+ for (i = 0; i <= in_last_pixel - in_first_pixel; i++)
+ coefficient_group[i] *= filter_scale;
+
+ for (i = in_last_pixel - in_first_pixel; i >= 0; i--)
+ {
+ if (coefficient_group[i])
+ break;
+
+ // This line has no weight. We can skip it.
+ contributor->n1 = contributor->n0 + i - 1;
+ }
+}
+
+static void stbir__calculate_coefficients_downsample(stbir_filter filter, float scale_ratio, int out_first_pixel, int out_last_pixel, float out_center_of_in, stbir__contributors* contributor, float* coefficient_group)
+{
+ int i;
+
+ STBIR_ASSERT(out_last_pixel - out_first_pixel <= (int)ceil(stbir__filter_info_table[filter].support(scale_ratio) * 2)); // Taken directly from stbir__get_coefficient_width() which we can't call because we don't know if we're horizontal or vertical.
+
+ contributor->n0 = out_first_pixel;
+ contributor->n1 = out_last_pixel;
+
+ STBIR_ASSERT(contributor->n1 >= contributor->n0);
+
+ for (i = 0; i <= out_last_pixel - out_first_pixel; i++)
+ {
+ float out_pixel_center = (float)(i + out_first_pixel) + 0.5f;
+ float x = out_pixel_center - out_center_of_in;
+ coefficient_group[i] = stbir__filter_info_table[filter].kernel(x, scale_ratio) * scale_ratio;
+ }
+
+ STBIR_ASSERT(stbir__filter_info_table[filter].kernel((float)(out_last_pixel + 1) + 0.5f - out_center_of_in, scale_ratio) == 0);
+
+ for (i = out_last_pixel - out_first_pixel; i >= 0; i--)
+ {
+ if (coefficient_group[i])
+ break;
+
+ // This line has no weight. We can skip it.
+ contributor->n1 = contributor->n0 + i - 1;
+ }
+}
+
+static void stbir__normalize_downsample_coefficients(stbir__contributors* contributors, float* coefficients, stbir_filter filter, float scale_ratio, int input_size, int output_size)
+{
+ int num_contributors = stbir__get_contributors(scale_ratio, filter, input_size, output_size);
+ int num_coefficients = stbir__get_coefficient_width(filter, scale_ratio);
+ int i, j;
+ int skip;
+
+ for (i = 0; i < output_size; i++)
+ {
+ float scale;
+ float total = 0;
+
+ for (j = 0; j < num_contributors; j++)
+ {
+ if (i >= contributors[j].n0 && i <= contributors[j].n1)
+ {
+ float coefficient = *stbir__get_coefficient(coefficients, filter, scale_ratio, j, i - contributors[j].n0);
+ total += coefficient;
+ }
+ else if (i < contributors[j].n0)
+ break;
+ }
+
+ STBIR_ASSERT(total > 0.9f);
+ STBIR_ASSERT(total < 1.1f);
+
+ scale = 1 / total;
+
+ for (j = 0; j < num_contributors; j++)
+ {
+ if (i >= contributors[j].n0 && i <= contributors[j].n1)
+ *stbir__get_coefficient(coefficients, filter, scale_ratio, j, i - contributors[j].n0) *= scale;
+ else if (i < contributors[j].n0)
+ break;
+ }
+ }
+
+ // Optimize: Skip zero coefficients and contributions outside of image bounds.
+ // Do this after normalizing because normalization depends on the n0/n1 values.
+ for (j = 0; j < num_contributors; j++)
+ {
+ int range, max, width;
+
+ skip = 0;
+ while (*stbir__get_coefficient(coefficients, filter, scale_ratio, j, skip) == 0)
+ skip++;
+
+ contributors[j].n0 += skip;
+
+ while (contributors[j].n0 < 0)
+ {
+ contributors[j].n0++;
+ skip++;
+ }
+
+ range = contributors[j].n1 - contributors[j].n0 + 1;
+ max = stbir__min(num_coefficients, range);
+
+ width = stbir__get_coefficient_width(filter, scale_ratio);
+ for (i = 0; i < max; i++)
+ {
+ if (i + skip >= width)
+ break;
+
+ *stbir__get_coefficient(coefficients, filter, scale_ratio, j, i) = *stbir__get_coefficient(coefficients, filter, scale_ratio, j, i + skip);
+ }
+
+ continue;
+ }
+
+ // Using min to avoid writing into invalid pixels.
+ for (i = 0; i < num_contributors; i++)
+ contributors[i].n1 = stbir__min(contributors[i].n1, output_size - 1);
+}
+
+// Each scan line uses the same kernel values so we should calculate the kernel
+// values once and then we can use them for every scan line.
+static void stbir__calculate_filters(stbir__contributors* contributors, float* coefficients, stbir_filter filter, float scale_ratio, float shift, int input_size, int output_size)
+{
+ int n;
+ int total_contributors = stbir__get_contributors(scale_ratio, filter, input_size, output_size);
+
+ if (stbir__use_upsampling(scale_ratio))
+ {
+ float out_pixels_radius = stbir__filter_info_table[filter].support(1 / scale_ratio) * scale_ratio;
+
+ // Looping through out pixels
+ for (n = 0; n < total_contributors; n++)
+ {
+ float in_center_of_out; // Center of the current out pixel in the in pixel space
+ int in_first_pixel, in_last_pixel;
+
+ stbir__calculate_sample_range_upsample(n, out_pixels_radius, scale_ratio, shift, &in_first_pixel, &in_last_pixel, &in_center_of_out);
+
+ stbir__calculate_coefficients_upsample(filter, scale_ratio, in_first_pixel, in_last_pixel, in_center_of_out, stbir__get_contributor(contributors, n), stbir__get_coefficient(coefficients, filter, scale_ratio, n, 0));
+ }
+ }
+ else
+ {
+ float in_pixels_radius = stbir__filter_info_table[filter].support(scale_ratio) / scale_ratio;
+
+ // Looping through in pixels
+ for (n = 0; n < total_contributors; n++)
+ {
+ float out_center_of_in; // Center of the current out pixel in the in pixel space
+ int out_first_pixel, out_last_pixel;
+ int n_adjusted = n - stbir__get_filter_pixel_margin(filter, scale_ratio);
+
+ stbir__calculate_sample_range_downsample(n_adjusted, in_pixels_radius, scale_ratio, shift, &out_first_pixel, &out_last_pixel, &out_center_of_in);
+
+ stbir__calculate_coefficients_downsample(filter, scale_ratio, out_first_pixel, out_last_pixel, out_center_of_in, stbir__get_contributor(contributors, n), stbir__get_coefficient(coefficients, filter, scale_ratio, n, 0));
+ }
+
+ stbir__normalize_downsample_coefficients(contributors, coefficients, filter, scale_ratio, input_size, output_size);
+ }
+}
+
+static float* stbir__get_decode_buffer(stbir__info* stbir_info)
+{
+ // The 0 index of the decode buffer starts after the margin. This makes
+ // it okay to use negative indexes on the decode buffer.
+ return &stbir_info->decode_buffer[stbir_info->horizontal_filter_pixel_margin * stbir_info->channels];
+}
+
+#define STBIR__DECODE(type, colorspace) ((type) * (STBIR_MAX_COLORSPACES) + (colorspace))
+
+static void stbir__decode_scanline(stbir__info* stbir_info, int n)
+{
+ int c;
+ int channels = stbir_info->channels;
+ int alpha_channel = stbir_info->alpha_channel;
+ int type = stbir_info->type;
+ int colorspace = stbir_info->colorspace;
+ int input_w = stbir_info->input_w;
+ size_t input_stride_bytes = stbir_info->input_stride_bytes;
+ float* decode_buffer = stbir__get_decode_buffer(stbir_info);
+ stbir_edge edge_horizontal = stbir_info->edge_horizontal;
+ stbir_edge edge_vertical = stbir_info->edge_vertical;
+ size_t in_buffer_row_offset = stbir__edge_wrap(edge_vertical, n, stbir_info->input_h) * input_stride_bytes;
+ const void* input_data = (char *) stbir_info->input_data + in_buffer_row_offset;
+ int max_x = input_w + stbir_info->horizontal_filter_pixel_margin;
+ int decode = STBIR__DECODE(type, colorspace);
+
+ int x = -stbir_info->horizontal_filter_pixel_margin;
+
+ // special handling for STBIR_EDGE_ZERO because it needs to return an item that doesn't appear in the input,
+ // and we want to avoid paying overhead on every pixel if not STBIR_EDGE_ZERO
+ if (edge_vertical == STBIR_EDGE_ZERO && (n < 0 || n >= stbir_info->input_h))
+ {
+ for (; x < max_x; x++)
+ for (c = 0; c < channels; c++)
+ decode_buffer[x*channels + c] = 0;
+ return;
+ }
+
+ switch (decode)
+ {
+ case STBIR__DECODE(STBIR_TYPE_UINT8, STBIR_COLORSPACE_LINEAR):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = ((float)((const unsigned char*)input_data)[input_pixel_index + c]) / stbir__max_uint8_as_float;
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT8, STBIR_COLORSPACE_SRGB):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = stbir__srgb_uchar_to_linear_float[((const unsigned char*)input_data)[input_pixel_index + c]];
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ decode_buffer[decode_pixel_index + alpha_channel] = ((float)((const unsigned char*)input_data)[input_pixel_index + alpha_channel]) / stbir__max_uint8_as_float;
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT16, STBIR_COLORSPACE_LINEAR):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = ((float)((const unsigned short*)input_data)[input_pixel_index + c]) / stbir__max_uint16_as_float;
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT16, STBIR_COLORSPACE_SRGB):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = stbir__srgb_to_linear(((float)((const unsigned short*)input_data)[input_pixel_index + c]) / stbir__max_uint16_as_float);
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ decode_buffer[decode_pixel_index + alpha_channel] = ((float)((const unsigned short*)input_data)[input_pixel_index + alpha_channel]) / stbir__max_uint16_as_float;
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT32, STBIR_COLORSPACE_LINEAR):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = (float)(((double)((const unsigned int*)input_data)[input_pixel_index + c]) / stbir__max_uint32_as_float);
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT32, STBIR_COLORSPACE_SRGB):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = stbir__srgb_to_linear((float)(((double)((const unsigned int*)input_data)[input_pixel_index + c]) / stbir__max_uint32_as_float));
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ decode_buffer[decode_pixel_index + alpha_channel] = (float)(((double)((const unsigned int*)input_data)[input_pixel_index + alpha_channel]) / stbir__max_uint32_as_float);
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_FLOAT, STBIR_COLORSPACE_LINEAR):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = ((const float*)input_data)[input_pixel_index + c];
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_FLOAT, STBIR_COLORSPACE_SRGB):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = stbir__srgb_to_linear(((const float*)input_data)[input_pixel_index + c]);
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ decode_buffer[decode_pixel_index + alpha_channel] = ((const float*)input_data)[input_pixel_index + alpha_channel];
+ }
+
+ break;
+
+ default:
+ STBIR_ASSERT(!"Unknown type/colorspace/channels combination.");
+ break;
+ }
+
+ if (!(stbir_info->flags & STBIR_FLAG_ALPHA_PREMULTIPLIED))
+ {
+ for (x = -stbir_info->horizontal_filter_pixel_margin; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+
+ // If the alpha value is 0 it will clobber the color values. Make sure it's not.
+ float alpha = decode_buffer[decode_pixel_index + alpha_channel];
+#ifndef STBIR_NO_ALPHA_EPSILON
+ if (stbir_info->type != STBIR_TYPE_FLOAT) {
+ alpha += STBIR_ALPHA_EPSILON;
+ decode_buffer[decode_pixel_index + alpha_channel] = alpha;
+ }
+#endif
+ for (c = 0; c < channels; c++)
+ {
+ if (c == alpha_channel)
+ continue;
+
+ decode_buffer[decode_pixel_index + c] *= alpha;
+ }
+ }
+ }
+
+ if (edge_horizontal == STBIR_EDGE_ZERO)
+ {
+ for (x = -stbir_info->horizontal_filter_pixel_margin; x < 0; x++)
+ {
+ for (c = 0; c < channels; c++)
+ decode_buffer[x*channels + c] = 0;
+ }
+ for (x = input_w; x < max_x; x++)
+ {
+ for (c = 0; c < channels; c++)
+ decode_buffer[x*channels + c] = 0;
+ }
+ }
+}
+
+static float* stbir__get_ring_buffer_entry(float* ring_buffer, int index, int ring_buffer_length)
+{
+ return &ring_buffer[index * ring_buffer_length];
+}
+
+static float* stbir__add_empty_ring_buffer_entry(stbir__info* stbir_info, int n)
+{
+ int ring_buffer_index;
+ float* ring_buffer;
+
+ stbir_info->ring_buffer_last_scanline = n;
+
+ if (stbir_info->ring_buffer_begin_index < 0)
+ {
+ ring_buffer_index = stbir_info->ring_buffer_begin_index = 0;
+ stbir_info->ring_buffer_first_scanline = n;
+ }
+ else
+ {
+ ring_buffer_index = (stbir_info->ring_buffer_begin_index + (stbir_info->ring_buffer_last_scanline - stbir_info->ring_buffer_first_scanline)) % stbir_info->ring_buffer_num_entries;
+ STBIR_ASSERT(ring_buffer_index != stbir_info->ring_buffer_begin_index);
+ }
+
+ ring_buffer = stbir__get_ring_buffer_entry(stbir_info->ring_buffer, ring_buffer_index, stbir_info->ring_buffer_length_bytes / sizeof(float));
+ memset(ring_buffer, 0, stbir_info->ring_buffer_length_bytes);
+
+ return ring_buffer;
+}
+
+
+static void stbir__resample_horizontal_upsample(stbir__info* stbir_info, float* output_buffer)
+{
+ int x, k;
+ int output_w = stbir_info->output_w;
+ int channels = stbir_info->channels;
+ float* decode_buffer = stbir__get_decode_buffer(stbir_info);
+ stbir__contributors* horizontal_contributors = stbir_info->horizontal_contributors;
+ float* horizontal_coefficients = stbir_info->horizontal_coefficients;
+ int coefficient_width = stbir_info->horizontal_coefficient_width;
+
+ for (x = 0; x < output_w; x++)
+ {
+ int n0 = horizontal_contributors[x].n0;
+ int n1 = horizontal_contributors[x].n1;
+
+ int out_pixel_index = x * channels;
+ int coefficient_group = coefficient_width * x;
+ int coefficient_counter = 0;
+
+ STBIR_ASSERT(n1 >= n0);
+ STBIR_ASSERT(n0 >= -stbir_info->horizontal_filter_pixel_margin);
+ STBIR_ASSERT(n1 >= -stbir_info->horizontal_filter_pixel_margin);
+ STBIR_ASSERT(n0 < stbir_info->input_w + stbir_info->horizontal_filter_pixel_margin);
+ STBIR_ASSERT(n1 < stbir_info->input_w + stbir_info->horizontal_filter_pixel_margin);
+
+ switch (channels) {
+ case 1:
+ for (k = n0; k <= n1; k++)
+ {
+ int in_pixel_index = k * 1;
+ float coefficient = horizontal_coefficients[coefficient_group + coefficient_counter++];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ }
+ break;
+ case 2:
+ for (k = n0; k <= n1; k++)
+ {
+ int in_pixel_index = k * 2;
+ float coefficient = horizontal_coefficients[coefficient_group + coefficient_counter++];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
+ }
+ break;
+ case 3:
+ for (k = n0; k <= n1; k++)
+ {
+ int in_pixel_index = k * 3;
+ float coefficient = horizontal_coefficients[coefficient_group + coefficient_counter++];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
+ output_buffer[out_pixel_index + 2] += decode_buffer[in_pixel_index + 2] * coefficient;
+ }
+ break;
+ case 4:
+ for (k = n0; k <= n1; k++)
+ {
+ int in_pixel_index = k * 4;
+ float coefficient = horizontal_coefficients[coefficient_group + coefficient_counter++];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
+ output_buffer[out_pixel_index + 2] += decode_buffer[in_pixel_index + 2] * coefficient;
+ output_buffer[out_pixel_index + 3] += decode_buffer[in_pixel_index + 3] * coefficient;
+ }
+ break;
+ default:
+ for (k = n0; k <= n1; k++)
+ {
+ int in_pixel_index = k * channels;
+ float coefficient = horizontal_coefficients[coefficient_group + coefficient_counter++];
+ int c;
+ STBIR_ASSERT(coefficient != 0);
+ for (c = 0; c < channels; c++)
+ output_buffer[out_pixel_index + c] += decode_buffer[in_pixel_index + c] * coefficient;
+ }
+ break;
+ }
+ }
+}
+
+static void stbir__resample_horizontal_downsample(stbir__info* stbir_info, float* output_buffer)
+{
+ int x, k;
+ int input_w = stbir_info->input_w;
+ int channels = stbir_info->channels;
+ float* decode_buffer = stbir__get_decode_buffer(stbir_info);
+ stbir__contributors* horizontal_contributors = stbir_info->horizontal_contributors;
+ float* horizontal_coefficients = stbir_info->horizontal_coefficients;
+ int coefficient_width = stbir_info->horizontal_coefficient_width;
+ int filter_pixel_margin = stbir_info->horizontal_filter_pixel_margin;
+ int max_x = input_w + filter_pixel_margin * 2;
+
+ STBIR_ASSERT(!stbir__use_width_upsampling(stbir_info));
+
+ switch (channels) {
+ case 1:
+ for (x = 0; x < max_x; x++)
+ {
+ int n0 = horizontal_contributors[x].n0;
+ int n1 = horizontal_contributors[x].n1;
+
+ int in_x = x - filter_pixel_margin;
+ int in_pixel_index = in_x * 1;
+ int max_n = n1;
+ int coefficient_group = coefficient_width * x;
+
+ for (k = n0; k <= max_n; k++)
+ {
+ int out_pixel_index = k * 1;
+ float coefficient = horizontal_coefficients[coefficient_group + k - n0];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ }
+ }
+ break;
+
+ case 2:
+ for (x = 0; x < max_x; x++)
+ {
+ int n0 = horizontal_contributors[x].n0;
+ int n1 = horizontal_contributors[x].n1;
+
+ int in_x = x - filter_pixel_margin;
+ int in_pixel_index = in_x * 2;
+ int max_n = n1;
+ int coefficient_group = coefficient_width * x;
+
+ for (k = n0; k <= max_n; k++)
+ {
+ int out_pixel_index = k * 2;
+ float coefficient = horizontal_coefficients[coefficient_group + k - n0];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
+ }
+ }
+ break;
+
+ case 3:
+ for (x = 0; x < max_x; x++)
+ {
+ int n0 = horizontal_contributors[x].n0;
+ int n1 = horizontal_contributors[x].n1;
+
+ int in_x = x - filter_pixel_margin;
+ int in_pixel_index = in_x * 3;
+ int max_n = n1;
+ int coefficient_group = coefficient_width * x;
+
+ for (k = n0; k <= max_n; k++)
+ {
+ int out_pixel_index = k * 3;
+ float coefficient = horizontal_coefficients[coefficient_group + k - n0];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
+ output_buffer[out_pixel_index + 2] += decode_buffer[in_pixel_index + 2] * coefficient;
+ }
+ }
+ break;
+
+ case 4:
+ for (x = 0; x < max_x; x++)
+ {
+ int n0 = horizontal_contributors[x].n0;
+ int n1 = horizontal_contributors[x].n1;
+
+ int in_x = x - filter_pixel_margin;
+ int in_pixel_index = in_x * 4;
+ int max_n = n1;
+ int coefficient_group = coefficient_width * x;
+
+ for (k = n0; k <= max_n; k++)
+ {
+ int out_pixel_index = k * 4;
+ float coefficient = horizontal_coefficients[coefficient_group + k - n0];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
+ output_buffer[out_pixel_index + 2] += decode_buffer[in_pixel_index + 2] * coefficient;
+ output_buffer[out_pixel_index + 3] += decode_buffer[in_pixel_index + 3] * coefficient;
+ }
+ }
+ break;
+
+ default:
+ for (x = 0; x < max_x; x++)
+ {
+ int n0 = horizontal_contributors[x].n0;
+ int n1 = horizontal_contributors[x].n1;
+
+ int in_x = x - filter_pixel_margin;
+ int in_pixel_index = in_x * channels;
+ int max_n = n1;
+ int coefficient_group = coefficient_width * x;
+
+ for (k = n0; k <= max_n; k++)
+ {
+ int c;
+ int out_pixel_index = k * channels;
+ float coefficient = horizontal_coefficients[coefficient_group + k - n0];
+ STBIR_ASSERT(coefficient != 0);
+ for (c = 0; c < channels; c++)
+ output_buffer[out_pixel_index + c] += decode_buffer[in_pixel_index + c] * coefficient;
+ }
+ }
+ break;
+ }
+}
+
+static void stbir__decode_and_resample_upsample(stbir__info* stbir_info, int n)
+{
+ // Decode the nth scanline from the source image into the decode buffer.
+ stbir__decode_scanline(stbir_info, n);
+
+ // Now resample it into the ring buffer.
+ if (stbir__use_width_upsampling(stbir_info))
+ stbir__resample_horizontal_upsample(stbir_info, stbir__add_empty_ring_buffer_entry(stbir_info, n));
+ else
+ stbir__resample_horizontal_downsample(stbir_info, stbir__add_empty_ring_buffer_entry(stbir_info, n));
+
+ // Now it's sitting in the ring buffer ready to be used as source for the vertical sampling.
+}
+
+static void stbir__decode_and_resample_downsample(stbir__info* stbir_info, int n)
+{
+ // Decode the nth scanline from the source image into the decode buffer.
+ stbir__decode_scanline(stbir_info, n);
+
+ memset(stbir_info->horizontal_buffer, 0, stbir_info->output_w * stbir_info->channels * sizeof(float));
+
+ // Now resample it into the horizontal buffer.
+ if (stbir__use_width_upsampling(stbir_info))
+ stbir__resample_horizontal_upsample(stbir_info, stbir_info->horizontal_buffer);
+ else
+ stbir__resample_horizontal_downsample(stbir_info, stbir_info->horizontal_buffer);
+
+ // Now it's sitting in the horizontal buffer ready to be distributed into the ring buffers.
+}
+
+// Get the specified scan line from the ring buffer.
+static float* stbir__get_ring_buffer_scanline(int get_scanline, float* ring_buffer, int begin_index, int first_scanline, int ring_buffer_num_entries, int ring_buffer_length)
+{
+ int ring_buffer_index = (begin_index + (get_scanline - first_scanline)) % ring_buffer_num_entries;
+ return stbir__get_ring_buffer_entry(ring_buffer, ring_buffer_index, ring_buffer_length);
+}
+
+
+static void stbir__encode_scanline(stbir__info* stbir_info, int num_pixels, void *output_buffer, float *encode_buffer, int channels, int alpha_channel, int decode)
+{
+ int x;
+ int n;
+ int num_nonalpha;
+ stbir_uint16 nonalpha[STBIR_MAX_CHANNELS];
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_PREMULTIPLIED))
+ {
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ float alpha = encode_buffer[pixel_index + alpha_channel];
+ float reciprocal_alpha = alpha ? 1.0f / alpha : 0;
+
+ // unrolling this produced a 1% slowdown upscaling a large RGBA linear-space image on my machine - stb
+ for (n = 0; n < channels; n++)
+ if (n != alpha_channel)
+ encode_buffer[pixel_index + n] *= reciprocal_alpha;
+
+ // We added in a small epsilon to prevent the color channel from being deleted with zero alpha.
+ // Because we only add it for integer types, it will automatically be discarded on integer
+ // conversion, so we don't need to subtract it back out (which would be problematic for
+ // numeric precision reasons).
+ }
+ }
+
+ // build a table of all channels that need colorspace correction, so
+ // we don't perform colorspace correction on channels that don't need it.
+ for (x = 0, num_nonalpha = 0; x < channels; ++x)
+ {
+ if (x != alpha_channel || (stbir_info->flags & STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ {
+ nonalpha[num_nonalpha++] = (stbir_uint16)x;
+ }
+ }
+
+ #define STBIR__ROUND_INT(f) ((int) ((f)+0.5))
+ #define STBIR__ROUND_UINT(f) ((stbir_uint32) ((f)+0.5))
+
+ #ifdef STBIR__SATURATE_INT
+ #define STBIR__ENCODE_LINEAR8(f) stbir__saturate8 (STBIR__ROUND_INT((f) * stbir__max_uint8_as_float ))
+ #define STBIR__ENCODE_LINEAR16(f) stbir__saturate16(STBIR__ROUND_INT((f) * stbir__max_uint16_as_float))
+ #else
+ #define STBIR__ENCODE_LINEAR8(f) (unsigned char ) STBIR__ROUND_INT(stbir__saturate(f) * stbir__max_uint8_as_float )
+ #define STBIR__ENCODE_LINEAR16(f) (unsigned short) STBIR__ROUND_INT(stbir__saturate(f) * stbir__max_uint16_as_float)
+ #endif
+
+ switch (decode)
+ {
+ case STBIR__DECODE(STBIR_TYPE_UINT8, STBIR_COLORSPACE_LINEAR):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < channels; n++)
+ {
+ int index = pixel_index + n;
+ ((unsigned char*)output_buffer)[index] = STBIR__ENCODE_LINEAR8(encode_buffer[index]);
+ }
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT8, STBIR_COLORSPACE_SRGB):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < num_nonalpha; n++)
+ {
+ int index = pixel_index + nonalpha[n];
+ ((unsigned char*)output_buffer)[index] = stbir__linear_to_srgb_uchar(encode_buffer[index]);
+ }
+
+ if (!(stbir_info->flags & STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ ((unsigned char *)output_buffer)[pixel_index + alpha_channel] = STBIR__ENCODE_LINEAR8(encode_buffer[pixel_index+alpha_channel]);
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT16, STBIR_COLORSPACE_LINEAR):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < channels; n++)
+ {
+ int index = pixel_index + n;
+ ((unsigned short*)output_buffer)[index] = STBIR__ENCODE_LINEAR16(encode_buffer[index]);
+ }
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT16, STBIR_COLORSPACE_SRGB):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < num_nonalpha; n++)
+ {
+ int index = pixel_index + nonalpha[n];
+ ((unsigned short*)output_buffer)[index] = (unsigned short)STBIR__ROUND_INT(stbir__linear_to_srgb(stbir__saturate(encode_buffer[index])) * stbir__max_uint16_as_float);
+ }
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ ((unsigned short*)output_buffer)[pixel_index + alpha_channel] = STBIR__ENCODE_LINEAR16(encode_buffer[pixel_index + alpha_channel]);
+ }
+
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT32, STBIR_COLORSPACE_LINEAR):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < channels; n++)
+ {
+ int index = pixel_index + n;
+ ((unsigned int*)output_buffer)[index] = (unsigned int)STBIR__ROUND_UINT(((double)stbir__saturate(encode_buffer[index])) * stbir__max_uint32_as_float);
+ }
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT32, STBIR_COLORSPACE_SRGB):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < num_nonalpha; n++)
+ {
+ int index = pixel_index + nonalpha[n];
+ ((unsigned int*)output_buffer)[index] = (unsigned int)STBIR__ROUND_UINT(((double)stbir__linear_to_srgb(stbir__saturate(encode_buffer[index]))) * stbir__max_uint32_as_float);
+ }
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ ((unsigned int*)output_buffer)[pixel_index + alpha_channel] = (unsigned int)STBIR__ROUND_INT(((double)stbir__saturate(encode_buffer[pixel_index + alpha_channel])) * stbir__max_uint32_as_float);
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_FLOAT, STBIR_COLORSPACE_LINEAR):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < channels; n++)
+ {
+ int index = pixel_index + n;
+ ((float*)output_buffer)[index] = encode_buffer[index];
+ }
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_FLOAT, STBIR_COLORSPACE_SRGB):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < num_nonalpha; n++)
+ {
+ int index = pixel_index + nonalpha[n];
+ ((float*)output_buffer)[index] = stbir__linear_to_srgb(encode_buffer[index]);
+ }
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ ((float*)output_buffer)[pixel_index + alpha_channel] = encode_buffer[pixel_index + alpha_channel];
+ }
+ break;
+
+ default:
+ STBIR_ASSERT(!"Unknown type/colorspace/channels combination.");
+ break;
+ }
+}
+
+static void stbir__resample_vertical_upsample(stbir__info* stbir_info, int n)
+{
+ int x, k;
+ int output_w = stbir_info->output_w;
+ stbir__contributors* vertical_contributors = stbir_info->vertical_contributors;
+ float* vertical_coefficients = stbir_info->vertical_coefficients;
+ int channels = stbir_info->channels;
+ int alpha_channel = stbir_info->alpha_channel;
+ int type = stbir_info->type;
+ int colorspace = stbir_info->colorspace;
+ int ring_buffer_entries = stbir_info->ring_buffer_num_entries;
+ void* output_data = stbir_info->output_data;
+ float* encode_buffer = stbir_info->encode_buffer;
+ int decode = STBIR__DECODE(type, colorspace);
+ int coefficient_width = stbir_info->vertical_coefficient_width;
+ int coefficient_counter;
+ int contributor = n;
+
+ float* ring_buffer = stbir_info->ring_buffer;
+ int ring_buffer_begin_index = stbir_info->ring_buffer_begin_index;
+ int ring_buffer_first_scanline = stbir_info->ring_buffer_first_scanline;
+ int ring_buffer_length = stbir_info->ring_buffer_length_bytes/sizeof(float);
+
+ int n0,n1, output_row_start;
+ int coefficient_group = coefficient_width * contributor;
+
+ n0 = vertical_contributors[contributor].n0;
+ n1 = vertical_contributors[contributor].n1;
+
+ output_row_start = n * stbir_info->output_stride_bytes;
+
+ STBIR_ASSERT(stbir__use_height_upsampling(stbir_info));
+
+ memset(encode_buffer, 0, output_w * sizeof(float) * channels);
+
+ // I tried reblocking this for better cache usage of encode_buffer
+ // (using x_outer, k, x_inner), but it lost speed. -- stb
+
+ coefficient_counter = 0;
+ switch (channels) {
+ case 1:
+ for (k = n0; k <= n1; k++)
+ {
+ int coefficient_index = coefficient_counter++;
+ float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length);
+ float coefficient = vertical_coefficients[coefficient_group + coefficient_index];
+ for (x = 0; x < output_w; ++x)
+ {
+ int in_pixel_index = x * 1;
+ encode_buffer[in_pixel_index + 0] += ring_buffer_entry[in_pixel_index + 0] * coefficient;
+ }
+ }
+ break;
+ case 2:
+ for (k = n0; k <= n1; k++)
+ {
+ int coefficient_index = coefficient_counter++;
+ float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length);
+ float coefficient = vertical_coefficients[coefficient_group + coefficient_index];
+ for (x = 0; x < output_w; ++x)
+ {
+ int in_pixel_index = x * 2;
+ encode_buffer[in_pixel_index + 0] += ring_buffer_entry[in_pixel_index + 0] * coefficient;
+ encode_buffer[in_pixel_index + 1] += ring_buffer_entry[in_pixel_index + 1] * coefficient;
+ }
+ }
+ break;
+ case 3:
+ for (k = n0; k <= n1; k++)
+ {
+ int coefficient_index = coefficient_counter++;
+ float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length);
+ float coefficient = vertical_coefficients[coefficient_group + coefficient_index];
+ for (x = 0; x < output_w; ++x)
+ {
+ int in_pixel_index = x * 3;
+ encode_buffer[in_pixel_index + 0] += ring_buffer_entry[in_pixel_index + 0] * coefficient;
+ encode_buffer[in_pixel_index + 1] += ring_buffer_entry[in_pixel_index + 1] * coefficient;
+ encode_buffer[in_pixel_index + 2] += ring_buffer_entry[in_pixel_index + 2] * coefficient;
+ }
+ }
+ break;
+ case 4:
+ for (k = n0; k <= n1; k++)
+ {
+ int coefficient_index = coefficient_counter++;
+ float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length);
+ float coefficient = vertical_coefficients[coefficient_group + coefficient_index];
+ for (x = 0; x < output_w; ++x)
+ {
+ int in_pixel_index = x * 4;
+ encode_buffer[in_pixel_index + 0] += ring_buffer_entry[in_pixel_index + 0] * coefficient;
+ encode_buffer[in_pixel_index + 1] += ring_buffer_entry[in_pixel_index + 1] * coefficient;
+ encode_buffer[in_pixel_index + 2] += ring_buffer_entry[in_pixel_index + 2] * coefficient;
+ encode_buffer[in_pixel_index + 3] += ring_buffer_entry[in_pixel_index + 3] * coefficient;
+ }
+ }
+ break;
+ default:
+ for (k = n0; k <= n1; k++)
+ {
+ int coefficient_index = coefficient_counter++;
+ float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length);
+ float coefficient = vertical_coefficients[coefficient_group + coefficient_index];
+ for (x = 0; x < output_w; ++x)
+ {
+ int in_pixel_index = x * channels;
+ int c;
+ for (c = 0; c < channels; c++)
+ encode_buffer[in_pixel_index + c] += ring_buffer_entry[in_pixel_index + c] * coefficient;
+ }
+ }
+ break;
+ }
+ stbir__encode_scanline(stbir_info, output_w, (char *) output_data + output_row_start, encode_buffer, channels, alpha_channel, decode);
+}
+
+static void stbir__resample_vertical_downsample(stbir__info* stbir_info, int n)
+{
+ int x, k;
+ int output_w = stbir_info->output_w;
+ stbir__contributors* vertical_contributors = stbir_info->vertical_contributors;
+ float* vertical_coefficients = stbir_info->vertical_coefficients;
+ int channels = stbir_info->channels;
+ int ring_buffer_entries = stbir_info->ring_buffer_num_entries;
+ float* horizontal_buffer = stbir_info->horizontal_buffer;
+ int coefficient_width = stbir_info->vertical_coefficient_width;
+ int contributor = n + stbir_info->vertical_filter_pixel_margin;
+
+ float* ring_buffer = stbir_info->ring_buffer;
+ int ring_buffer_begin_index = stbir_info->ring_buffer_begin_index;
+ int ring_buffer_first_scanline = stbir_info->ring_buffer_first_scanline;
+ int ring_buffer_length = stbir_info->ring_buffer_length_bytes/sizeof(float);
+ int n0,n1;
+
+ n0 = vertical_contributors[contributor].n0;
+ n1 = vertical_contributors[contributor].n1;
+
+ STBIR_ASSERT(!stbir__use_height_upsampling(stbir_info));
+
+ for (k = n0; k <= n1; k++)
+ {
+ int coefficient_index = k - n0;
+ int coefficient_group = coefficient_width * contributor;
+ float coefficient = vertical_coefficients[coefficient_group + coefficient_index];
+
+ float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length);
+
+ switch (channels) {
+ case 1:
+ for (x = 0; x < output_w; x++)
+ {
+ int in_pixel_index = x * 1;
+ ring_buffer_entry[in_pixel_index + 0] += horizontal_buffer[in_pixel_index + 0] * coefficient;
+ }
+ break;
+ case 2:
+ for (x = 0; x < output_w; x++)
+ {
+ int in_pixel_index = x * 2;
+ ring_buffer_entry[in_pixel_index + 0] += horizontal_buffer[in_pixel_index + 0] * coefficient;
+ ring_buffer_entry[in_pixel_index + 1] += horizontal_buffer[in_pixel_index + 1] * coefficient;
+ }
+ break;
+ case 3:
+ for (x = 0; x < output_w; x++)
+ {
+ int in_pixel_index = x * 3;
+ ring_buffer_entry[in_pixel_index + 0] += horizontal_buffer[in_pixel_index + 0] * coefficient;
+ ring_buffer_entry[in_pixel_index + 1] += horizontal_buffer[in_pixel_index + 1] * coefficient;
+ ring_buffer_entry[in_pixel_index + 2] += horizontal_buffer[in_pixel_index + 2] * coefficient;
+ }
+ break;
+ case 4:
+ for (x = 0; x < output_w; x++)
+ {
+ int in_pixel_index = x * 4;
+ ring_buffer_entry[in_pixel_index + 0] += horizontal_buffer[in_pixel_index + 0] * coefficient;
+ ring_buffer_entry[in_pixel_index + 1] += horizontal_buffer[in_pixel_index + 1] * coefficient;
+ ring_buffer_entry[in_pixel_index + 2] += horizontal_buffer[in_pixel_index + 2] * coefficient;
+ ring_buffer_entry[in_pixel_index + 3] += horizontal_buffer[in_pixel_index + 3] * coefficient;
+ }
+ break;
+ default:
+ for (x = 0; x < output_w; x++)
+ {
+ int in_pixel_index = x * channels;
+
+ int c;
+ for (c = 0; c < channels; c++)
+ ring_buffer_entry[in_pixel_index + c] += horizontal_buffer[in_pixel_index + c] * coefficient;
+ }
+ break;
+ }
+ }
+}
+
+static void stbir__buffer_loop_upsample(stbir__info* stbir_info)
+{
+ int y;
+ float scale_ratio = stbir_info->vertical_scale;
+ float out_scanlines_radius = stbir__filter_info_table[stbir_info->vertical_filter].support(1/scale_ratio) * scale_ratio;
+
+ STBIR_ASSERT(stbir__use_height_upsampling(stbir_info));
+
+ for (y = 0; y < stbir_info->output_h; y++)
+ {
+ float in_center_of_out = 0; // Center of the current out scanline in the in scanline space
+ int in_first_scanline = 0, in_last_scanline = 0;
+
+ stbir__calculate_sample_range_upsample(y, out_scanlines_radius, scale_ratio, stbir_info->vertical_shift, &in_first_scanline, &in_last_scanline, &in_center_of_out);
+
+ STBIR_ASSERT(in_last_scanline - in_first_scanline + 1 <= stbir_info->ring_buffer_num_entries);
+
+ if (stbir_info->ring_buffer_begin_index >= 0)
+ {
+ // Get rid of whatever we don't need anymore.
+ while (in_first_scanline > stbir_info->ring_buffer_first_scanline)
+ {
+ if (stbir_info->ring_buffer_first_scanline == stbir_info->ring_buffer_last_scanline)
+ {
+ // We just popped the last scanline off the ring buffer.
+ // Reset it to the empty state.
+ stbir_info->ring_buffer_begin_index = -1;
+ stbir_info->ring_buffer_first_scanline = 0;
+ stbir_info->ring_buffer_last_scanline = 0;
+ break;
+ }
+ else
+ {
+ stbir_info->ring_buffer_first_scanline++;
+ stbir_info->ring_buffer_begin_index = (stbir_info->ring_buffer_begin_index + 1) % stbir_info->ring_buffer_num_entries;
+ }
+ }
+ }
+
+ // Load in new ones.
+ if (stbir_info->ring_buffer_begin_index < 0)
+ stbir__decode_and_resample_upsample(stbir_info, in_first_scanline);
+
+ while (in_last_scanline > stbir_info->ring_buffer_last_scanline)
+ stbir__decode_and_resample_upsample(stbir_info, stbir_info->ring_buffer_last_scanline + 1);
+
+ // Now all buffers should be ready to write a row of vertical sampling.
+ stbir__resample_vertical_upsample(stbir_info, y);
+
+ STBIR_PROGRESS_REPORT((float)y / stbir_info->output_h);
+ }
+}
+
+static void stbir__empty_ring_buffer(stbir__info* stbir_info, int first_necessary_scanline)
+{
+ int output_stride_bytes = stbir_info->output_stride_bytes;
+ int channels = stbir_info->channels;
+ int alpha_channel = stbir_info->alpha_channel;
+ int type = stbir_info->type;
+ int colorspace = stbir_info->colorspace;
+ int output_w = stbir_info->output_w;
+ void* output_data = stbir_info->output_data;
+ int decode = STBIR__DECODE(type, colorspace);
+
+ float* ring_buffer = stbir_info->ring_buffer;
+ int ring_buffer_length = stbir_info->ring_buffer_length_bytes/sizeof(float);
+
+ if (stbir_info->ring_buffer_begin_index >= 0)
+ {
+ // Get rid of whatever we don't need anymore.
+ while (first_necessary_scanline > stbir_info->ring_buffer_first_scanline)
+ {
+ if (stbir_info->ring_buffer_first_scanline >= 0 && stbir_info->ring_buffer_first_scanline < stbir_info->output_h)
+ {
+ int output_row_start = stbir_info->ring_buffer_first_scanline * output_stride_bytes;
+ float* ring_buffer_entry = stbir__get_ring_buffer_entry(ring_buffer, stbir_info->ring_buffer_begin_index, ring_buffer_length);
+ stbir__encode_scanline(stbir_info, output_w, (char *) output_data + output_row_start, ring_buffer_entry, channels, alpha_channel, decode);
+ STBIR_PROGRESS_REPORT((float)stbir_info->ring_buffer_first_scanline / stbir_info->output_h);
+ }
+
+ if (stbir_info->ring_buffer_first_scanline == stbir_info->ring_buffer_last_scanline)
+ {
+ // We just popped the last scanline off the ring buffer.
+ // Reset it to the empty state.
+ stbir_info->ring_buffer_begin_index = -1;
+ stbir_info->ring_buffer_first_scanline = 0;
+ stbir_info->ring_buffer_last_scanline = 0;
+ break;
+ }
+ else
+ {
+ stbir_info->ring_buffer_first_scanline++;
+ stbir_info->ring_buffer_begin_index = (stbir_info->ring_buffer_begin_index + 1) % stbir_info->ring_buffer_num_entries;
+ }
+ }
+ }
+}
+
+static void stbir__buffer_loop_downsample(stbir__info* stbir_info)
+{
+ int y;
+ float scale_ratio = stbir_info->vertical_scale;
+ int output_h = stbir_info->output_h;
+ float in_pixels_radius = stbir__filter_info_table[stbir_info->vertical_filter].support(scale_ratio) / scale_ratio;
+ int pixel_margin = stbir_info->vertical_filter_pixel_margin;
+ int max_y = stbir_info->input_h + pixel_margin;
+
+ STBIR_ASSERT(!stbir__use_height_upsampling(stbir_info));
+
+ for (y = -pixel_margin; y < max_y; y++)
+ {
+ float out_center_of_in; // Center of the current out scanline in the in scanline space
+ int out_first_scanline, out_last_scanline;
+
+ stbir__calculate_sample_range_downsample(y, in_pixels_radius, scale_ratio, stbir_info->vertical_shift, &out_first_scanline, &out_last_scanline, &out_center_of_in);
+
+ STBIR_ASSERT(out_last_scanline - out_first_scanline + 1 <= stbir_info->ring_buffer_num_entries);
+
+ if (out_last_scanline < 0 || out_first_scanline >= output_h)
+ continue;
+
+ stbir__empty_ring_buffer(stbir_info, out_first_scanline);
+
+ stbir__decode_and_resample_downsample(stbir_info, y);
+
+ // Load in new ones.
+ if (stbir_info->ring_buffer_begin_index < 0)
+ stbir__add_empty_ring_buffer_entry(stbir_info, out_first_scanline);
+
+ while (out_last_scanline > stbir_info->ring_buffer_last_scanline)
+ stbir__add_empty_ring_buffer_entry(stbir_info, stbir_info->ring_buffer_last_scanline + 1);
+
+ // Now the horizontal buffer is ready to write to all ring buffer rows.
+ stbir__resample_vertical_downsample(stbir_info, y);
+ }
+
+ stbir__empty_ring_buffer(stbir_info, stbir_info->output_h);
+}
+
+static void stbir__setup(stbir__info *info, int input_w, int input_h, int output_w, int output_h, int channels)
+{
+ info->input_w = input_w;
+ info->input_h = input_h;
+ info->output_w = output_w;
+ info->output_h = output_h;
+ info->channels = channels;
+}
+
+static void stbir__calculate_transform(stbir__info *info, float s0, float t0, float s1, float t1, float *transform)
+{
+ info->s0 = s0;
+ info->t0 = t0;
+ info->s1 = s1;
+ info->t1 = t1;
+
+ if (transform)
+ {
+ info->horizontal_scale = transform[0];
+ info->vertical_scale = transform[1];
+ info->horizontal_shift = transform[2];
+ info->vertical_shift = transform[3];
+ }
+ else
+ {
+ info->horizontal_scale = ((float)info->output_w / info->input_w) / (s1 - s0);
+ info->vertical_scale = ((float)info->output_h / info->input_h) / (t1 - t0);
+
+ info->horizontal_shift = s0 * info->output_w / (s1 - s0);
+ info->vertical_shift = t0 * info->output_h / (t1 - t0);
+ }
+}
+
+static void stbir__choose_filter(stbir__info *info, stbir_filter h_filter, stbir_filter v_filter)
+{
+ if (h_filter == 0)
+ h_filter = stbir__use_upsampling(info->horizontal_scale) ? STBIR_DEFAULT_FILTER_UPSAMPLE : STBIR_DEFAULT_FILTER_DOWNSAMPLE;
+ if (v_filter == 0)
+ v_filter = stbir__use_upsampling(info->vertical_scale) ? STBIR_DEFAULT_FILTER_UPSAMPLE : STBIR_DEFAULT_FILTER_DOWNSAMPLE;
+ info->horizontal_filter = h_filter;
+ info->vertical_filter = v_filter;
+}
+
+static stbir_uint32 stbir__calculate_memory(stbir__info *info)
+{
+ int pixel_margin = stbir__get_filter_pixel_margin(info->horizontal_filter, info->horizontal_scale);
+ int filter_height = stbir__get_filter_pixel_width(info->vertical_filter, info->vertical_scale);
+
+ info->horizontal_num_contributors = stbir__get_contributors(info->horizontal_scale, info->horizontal_filter, info->input_w, info->output_w);
+ info->vertical_num_contributors = stbir__get_contributors(info->vertical_scale , info->vertical_filter , info->input_h, info->output_h);
+
+ // One extra entry because floating point precision problems sometimes cause an extra to be necessary.
+ info->ring_buffer_num_entries = filter_height + 1;
+
+ info->horizontal_contributors_size = info->horizontal_num_contributors * sizeof(stbir__contributors);
+ info->horizontal_coefficients_size = stbir__get_total_horizontal_coefficients(info) * sizeof(float);
+ info->vertical_contributors_size = info->vertical_num_contributors * sizeof(stbir__contributors);
+ info->vertical_coefficients_size = stbir__get_total_vertical_coefficients(info) * sizeof(float);
+ info->decode_buffer_size = (info->input_w + pixel_margin * 2) * info->channels * sizeof(float);
+ info->horizontal_buffer_size = info->output_w * info->channels * sizeof(float);
+ info->ring_buffer_size = info->output_w * info->channels * info->ring_buffer_num_entries * sizeof(float);
+ info->encode_buffer_size = info->output_w * info->channels * sizeof(float);
+
+ STBIR_ASSERT(info->horizontal_filter != 0);
+ STBIR_ASSERT(info->horizontal_filter < STBIR__ARRAY_SIZE(stbir__filter_info_table)); // this now happens too late
+ STBIR_ASSERT(info->vertical_filter != 0);
+ STBIR_ASSERT(info->vertical_filter < STBIR__ARRAY_SIZE(stbir__filter_info_table)); // this now happens too late
+
+ if (stbir__use_height_upsampling(info))
+ // The horizontal buffer is for when we're downsampling the height and we
+ // can't output the result of sampling the decode buffer directly into the
+ // ring buffers.
+ info->horizontal_buffer_size = 0;
+ else
+ // The encode buffer is to retain precision in the height upsampling method
+ // and isn't used when height downsampling.
+ info->encode_buffer_size = 0;
+
+ return info->horizontal_contributors_size + info->horizontal_coefficients_size
+ + info->vertical_contributors_size + info->vertical_coefficients_size
+ + info->decode_buffer_size + info->horizontal_buffer_size
+ + info->ring_buffer_size + info->encode_buffer_size;
+}
+
+static int stbir__resize_allocated(stbir__info *info,
+ const void* input_data, int input_stride_in_bytes,
+ void* output_data, int output_stride_in_bytes,
+ int alpha_channel, stbir_uint32 flags, stbir_datatype type,
+ stbir_edge edge_horizontal, stbir_edge edge_vertical, stbir_colorspace colorspace,
+ void* tempmem, size_t tempmem_size_in_bytes)
+{
+ size_t memory_required = stbir__calculate_memory(info);
+
+ int width_stride_input = input_stride_in_bytes ? input_stride_in_bytes : info->channels * info->input_w * stbir__type_size[type];
+ int width_stride_output = output_stride_in_bytes ? output_stride_in_bytes : info->channels * info->output_w * stbir__type_size[type];
+
+#ifdef STBIR_DEBUG_OVERWRITE_TEST
+#define OVERWRITE_ARRAY_SIZE 8
+ unsigned char overwrite_output_before_pre[OVERWRITE_ARRAY_SIZE];
+ unsigned char overwrite_tempmem_before_pre[OVERWRITE_ARRAY_SIZE];
+ unsigned char overwrite_output_after_pre[OVERWRITE_ARRAY_SIZE];
+ unsigned char overwrite_tempmem_after_pre[OVERWRITE_ARRAY_SIZE];
+
+ size_t begin_forbidden = width_stride_output * (info->output_h - 1) + info->output_w * info->channels * stbir__type_size[type];
+ memcpy(overwrite_output_before_pre, &((unsigned char*)output_data)[-OVERWRITE_ARRAY_SIZE], OVERWRITE_ARRAY_SIZE);
+ memcpy(overwrite_output_after_pre, &((unsigned char*)output_data)[begin_forbidden], OVERWRITE_ARRAY_SIZE);
+ memcpy(overwrite_tempmem_before_pre, &((unsigned char*)tempmem)[-OVERWRITE_ARRAY_SIZE], OVERWRITE_ARRAY_SIZE);
+ memcpy(overwrite_tempmem_after_pre, &((unsigned char*)tempmem)[tempmem_size_in_bytes], OVERWRITE_ARRAY_SIZE);
+#endif
+
+ STBIR_ASSERT(info->channels >= 0);
+ STBIR_ASSERT(info->channels <= STBIR_MAX_CHANNELS);
+
+ if (info->channels < 0 || info->channels > STBIR_MAX_CHANNELS)
+ return 0;
+
+ STBIR_ASSERT(info->horizontal_filter < STBIR__ARRAY_SIZE(stbir__filter_info_table));
+ STBIR_ASSERT(info->vertical_filter < STBIR__ARRAY_SIZE(stbir__filter_info_table));
+
+ if (info->horizontal_filter >= STBIR__ARRAY_SIZE(stbir__filter_info_table))
+ return 0;
+ if (info->vertical_filter >= STBIR__ARRAY_SIZE(stbir__filter_info_table))
+ return 0;
+
+ if (alpha_channel < 0)
+ flags |= STBIR_FLAG_ALPHA_USES_COLORSPACE | STBIR_FLAG_ALPHA_PREMULTIPLIED;
+
+ if (!(flags&STBIR_FLAG_ALPHA_USES_COLORSPACE) || !(flags&STBIR_FLAG_ALPHA_PREMULTIPLIED))
+ STBIR_ASSERT(alpha_channel >= 0 && alpha_channel < info->channels);
+
+ if (alpha_channel >= info->channels)
+ return 0;
+
+ STBIR_ASSERT(tempmem);
+
+ if (!tempmem)
+ return 0;
+
+ STBIR_ASSERT(tempmem_size_in_bytes >= memory_required);
+
+ if (tempmem_size_in_bytes < memory_required)
+ return 0;
+
+ memset(tempmem, 0, tempmem_size_in_bytes);
+
+ info->input_data = input_data;
+ info->input_stride_bytes = width_stride_input;
+
+ info->output_data = output_data;
+ info->output_stride_bytes = width_stride_output;
+
+ info->alpha_channel = alpha_channel;
+ info->flags = flags;
+ info->type = type;
+ info->edge_horizontal = edge_horizontal;
+ info->edge_vertical = edge_vertical;
+ info->colorspace = colorspace;
+
+ info->horizontal_coefficient_width = stbir__get_coefficient_width (info->horizontal_filter, info->horizontal_scale);
+ info->vertical_coefficient_width = stbir__get_coefficient_width (info->vertical_filter , info->vertical_scale );
+ info->horizontal_filter_pixel_width = stbir__get_filter_pixel_width (info->horizontal_filter, info->horizontal_scale);
+ info->vertical_filter_pixel_width = stbir__get_filter_pixel_width (info->vertical_filter , info->vertical_scale );
+ info->horizontal_filter_pixel_margin = stbir__get_filter_pixel_margin(info->horizontal_filter, info->horizontal_scale);
+ info->vertical_filter_pixel_margin = stbir__get_filter_pixel_margin(info->vertical_filter , info->vertical_scale );
+
+ info->ring_buffer_length_bytes = info->output_w * info->channels * sizeof(float);
+ info->decode_buffer_pixels = info->input_w + info->horizontal_filter_pixel_margin * 2;
+
+#define STBIR__NEXT_MEMPTR(current, newtype) (newtype*)(((unsigned char*)current) + current##_size)
+
+ info->horizontal_contributors = (stbir__contributors *) tempmem;
+ info->horizontal_coefficients = STBIR__NEXT_MEMPTR(info->horizontal_contributors, float);
+ info->vertical_contributors = STBIR__NEXT_MEMPTR(info->horizontal_coefficients, stbir__contributors);
+ info->vertical_coefficients = STBIR__NEXT_MEMPTR(info->vertical_contributors, float);
+ info->decode_buffer = STBIR__NEXT_MEMPTR(info->vertical_coefficients, float);
+
+ if (stbir__use_height_upsampling(info))
+ {
+ info->horizontal_buffer = NULL;
+ info->ring_buffer = STBIR__NEXT_MEMPTR(info->decode_buffer, float);
+ info->encode_buffer = STBIR__NEXT_MEMPTR(info->ring_buffer, float);
+
+ STBIR_ASSERT((size_t)STBIR__NEXT_MEMPTR(info->encode_buffer, unsigned char) == (size_t)tempmem + tempmem_size_in_bytes);
+ }
+ else
+ {
+ info->horizontal_buffer = STBIR__NEXT_MEMPTR(info->decode_buffer, float);
+ info->ring_buffer = STBIR__NEXT_MEMPTR(info->horizontal_buffer, float);
+ info->encode_buffer = NULL;
+
+ STBIR_ASSERT((size_t)STBIR__NEXT_MEMPTR(info->ring_buffer, unsigned char) == (size_t)tempmem + tempmem_size_in_bytes);
+ }
+
+#undef STBIR__NEXT_MEMPTR
+
+ // This signals that the ring buffer is empty
+ info->ring_buffer_begin_index = -1;
+
+ stbir__calculate_filters(info->horizontal_contributors, info->horizontal_coefficients, info->horizontal_filter, info->horizontal_scale, info->horizontal_shift, info->input_w, info->output_w);
+ stbir__calculate_filters(info->vertical_contributors, info->vertical_coefficients, info->vertical_filter, info->vertical_scale, info->vertical_shift, info->input_h, info->output_h);
+
+ STBIR_PROGRESS_REPORT(0);
+
+ if (stbir__use_height_upsampling(info))
+ stbir__buffer_loop_upsample(info);
+ else
+ stbir__buffer_loop_downsample(info);
+
+ STBIR_PROGRESS_REPORT(1);
+
+#ifdef STBIR_DEBUG_OVERWRITE_TEST
+ STBIR_ASSERT(memcmp(overwrite_output_before_pre, &((unsigned char*)output_data)[-OVERWRITE_ARRAY_SIZE], OVERWRITE_ARRAY_SIZE) == 0);
+ STBIR_ASSERT(memcmp(overwrite_output_after_pre, &((unsigned char*)output_data)[begin_forbidden], OVERWRITE_ARRAY_SIZE) == 0);
+ STBIR_ASSERT(memcmp(overwrite_tempmem_before_pre, &((unsigned char*)tempmem)[-OVERWRITE_ARRAY_SIZE], OVERWRITE_ARRAY_SIZE) == 0);
+ STBIR_ASSERT(memcmp(overwrite_tempmem_after_pre, &((unsigned char*)tempmem)[tempmem_size_in_bytes], OVERWRITE_ARRAY_SIZE) == 0);
+#endif
+
+ return 1;
+}
+
+
+static int stbir__resize_arbitrary(
+ void *alloc_context,
+ const void* input_data, int input_w, int input_h, int input_stride_in_bytes,
+ void* output_data, int output_w, int output_h, int output_stride_in_bytes,
+ float s0, float t0, float s1, float t1, float *transform,
+ int channels, int alpha_channel, stbir_uint32 flags, stbir_datatype type,
+ stbir_filter h_filter, stbir_filter v_filter,
+ stbir_edge edge_horizontal, stbir_edge edge_vertical, stbir_colorspace colorspace)
+{
+ stbir__info info;
+ int result;
+ size_t memory_required;
+ void* extra_memory;
+
+ stbir__setup(&info, input_w, input_h, output_w, output_h, channels);
+ stbir__calculate_transform(&info, s0,t0,s1,t1,transform);
+ stbir__choose_filter(&info, h_filter, v_filter);
+ memory_required = stbir__calculate_memory(&info);
+ extra_memory = STBIR_MALLOC(memory_required, alloc_context);
+
+ if (!extra_memory)
+ return 0;
+
+ result = stbir__resize_allocated(&info, input_data, input_stride_in_bytes,
+ output_data, output_stride_in_bytes,
+ alpha_channel, flags, type,
+ edge_horizontal, edge_vertical,
+ colorspace, extra_memory, memory_required);
+
+ STBIR_FREE(extra_memory, alloc_context);
+
+ return result;
+}
+
+STBIRDEF int stbir_resize_uint8( const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels)
+{
+ return stbir__resize_arbitrary(NULL, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,-1,0, STBIR_TYPE_UINT8, STBIR_FILTER_DEFAULT, STBIR_FILTER_DEFAULT,
+ STBIR_EDGE_CLAMP, STBIR_EDGE_CLAMP, STBIR_COLORSPACE_LINEAR);
+}
+
+STBIRDEF int stbir_resize_float( const float *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ float *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels)
+{
+ return stbir__resize_arbitrary(NULL, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,-1,0, STBIR_TYPE_FLOAT, STBIR_FILTER_DEFAULT, STBIR_FILTER_DEFAULT,
+ STBIR_EDGE_CLAMP, STBIR_EDGE_CLAMP, STBIR_COLORSPACE_LINEAR);
+}
+
+STBIRDEF int stbir_resize_uint8_srgb(const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags)
+{
+ return stbir__resize_arbitrary(NULL, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,alpha_channel,flags, STBIR_TYPE_UINT8, STBIR_FILTER_DEFAULT, STBIR_FILTER_DEFAULT,
+ STBIR_EDGE_CLAMP, STBIR_EDGE_CLAMP, STBIR_COLORSPACE_SRGB);
+}
+
+STBIRDEF int stbir_resize_uint8_srgb_edgemode(const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode)
+{
+ return stbir__resize_arbitrary(NULL, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,alpha_channel,flags, STBIR_TYPE_UINT8, STBIR_FILTER_DEFAULT, STBIR_FILTER_DEFAULT,
+ edge_wrap_mode, edge_wrap_mode, STBIR_COLORSPACE_SRGB);
+}
+
+STBIRDEF int stbir_resize_uint8_generic( const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space,
+ void *alloc_context)
+{
+ return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,alpha_channel,flags, STBIR_TYPE_UINT8, filter, filter,
+ edge_wrap_mode, edge_wrap_mode, space);
+}
+
+STBIRDEF int stbir_resize_uint16_generic(const stbir_uint16 *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ stbir_uint16 *output_pixels , int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space,
+ void *alloc_context)
+{
+ return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,alpha_channel,flags, STBIR_TYPE_UINT16, filter, filter,
+ edge_wrap_mode, edge_wrap_mode, space);
+}
+
+
+STBIRDEF int stbir_resize_float_generic( const float *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ float *output_pixels , int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space,
+ void *alloc_context)
+{
+ return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,alpha_channel,flags, STBIR_TYPE_FLOAT, filter, filter,
+ edge_wrap_mode, edge_wrap_mode, space);
+}
+
+
+STBIRDEF int stbir_resize( const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ stbir_datatype datatype,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical,
+ stbir_filter filter_horizontal, stbir_filter filter_vertical,
+ stbir_colorspace space, void *alloc_context)
+{
+ return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,alpha_channel,flags, datatype, filter_horizontal, filter_vertical,
+ edge_mode_horizontal, edge_mode_vertical, space);
+}
+
+
+STBIRDEF int stbir_resize_subpixel(const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ stbir_datatype datatype,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical,
+ stbir_filter filter_horizontal, stbir_filter filter_vertical,
+ stbir_colorspace space, void *alloc_context,
+ float x_scale, float y_scale,
+ float x_offset, float y_offset)
+{
+ float transform[4];
+ transform[0] = x_scale;
+ transform[1] = y_scale;
+ transform[2] = x_offset;
+ transform[3] = y_offset;
+ return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,transform,num_channels,alpha_channel,flags, datatype, filter_horizontal, filter_vertical,
+ edge_mode_horizontal, edge_mode_vertical, space);
+}
+
+STBIRDEF int stbir_resize_region( const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ stbir_datatype datatype,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical,
+ stbir_filter filter_horizontal, stbir_filter filter_vertical,
+ stbir_colorspace space, void *alloc_context,
+ float s0, float t0, float s1, float t1)
+{
+ return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ s0,t0,s1,t1,NULL,num_channels,alpha_channel,flags, datatype, filter_horizontal, filter_vertical,
+ edge_mode_horizontal, edge_mode_vertical, space);
+}
+
+#endif // STB_IMAGE_RESIZE_IMPLEMENTATION
+
+/*
+------------------------------------------------------------------------------
+This software is available under 2 licenses -- choose whichever you prefer.
+------------------------------------------------------------------------------
+ALTERNATIVE A - MIT License
+Copyright (c) 2017 Sean Barrett
+Permission is hereby granted, free of charge, to any person obtaining a copy of
+this software and associated documentation files (the "Software"), to deal in
+the Software without restriction, including without limitation the rights to
+use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
+of the Software, and to permit persons to whom the Software is furnished to do
+so, subject to the following conditions:
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+------------------------------------------------------------------------------
+ALTERNATIVE B - Public Domain (www.unlicense.org)
+This is free and unencumbered software released into the public domain.
+Anyone is free to copy, modify, publish, use, compile, sell, or distribute this
+software, either in source code form or as a compiled binary, for any purpose,
+commercial or non-commercial, and by any means.
+In jurisdictions that recognize copyright laws, the author or authors of this
+software dedicate any and all copyright interest in the software to the public
+domain. We make this dedication for the benefit of the public at large and to
+the detriment of our heirs and successors. We intend this dedication to be an
+overt act of relinquishment in perpetuity of all present and future rights to
+this software under copyright law.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+------------------------------------------------------------------------------
+*/
--- /dev/null
+# multistick_cpp: A Movidius Neural Compute Stick example for multiple devices in C++
+
+This directory contains a C++ example that shows how to program for mulitiple NCS devices. The program opens two NCS devices and uses one device to run GoogLeNet inferences and the other device to run SqueezeNet inferences.
+
+## Prerequisites
+
+This code example requires that the following components are available:
+1. Movidius Neural Compute Stick
+2. Movidius Neural Compute SDK
+
+
+## Building the example
+To run the example code do the following :
+1. Open a terminal and change directory to the multistick_cpp example base directory
+2. Type the following command in the terminal: make
+
+
+## Running the Example
+After building the example you can run the example code by doing the following :
+1. Open a terminal and change directory to the multistick_cpp base directory
+2. Type the following command in the terminal: make run
+
+When the application runs normally and is able to connect to the NCS device the output will be similar to this:
+
+~~~
+TBD
+~~~
+
+
+
--- /dev/null
+
+ifneq ($(findstring movidius, $(PYTHONPATH)), movidius)
+ export PYTHONPATH:=/opt/movidius/caffe/python:/opt/movidius/mvnc/python:$(PYTHONPATH)
+endif
+
+NCCOMPILE = mvNCCompile
+NCPROFILE = mvNCProfile
+NCCHECK = mvNCCheck
+
+PROTOTXT_FILENAME= deploy.prototxt
+GET_PROTOTXT = wget -P . https://raw.githubusercontent.com/BVLC/caffe/master/models/bvlc_alexnet/${PROTOTXT_FILENAME}
+
+CAFFEMODEL_FILENAME = bvlc_alexnet.caffemodel
+GET_CAFFEMODEL = wget -P . -N http://dl.caffe.berkeleyvision.org/${CAFFEMODEL_FILENAME}
+
+
+.PHONY: all
+all: profile check compile cpp
+
+.PHONY: prereqs
+prereqs:
+ @echo "\nmaking prereqs"
+ (cd ../../data/ilsvrc12; make)
+ @sed -i 's/\r//' run.py
+ chmod +x run.py
+
+.PHONY: prototxt
+prototxt: prereqs
+ @echo "\nmaking prototxt"
+ @if [ -e ${PROTOTXT_FILENAME} ] ; \
+ then \
+ echo "Prototxt file already exists"; \
+ else \
+ echo "Downloading Prototxt file"; \
+ ${GET_PROTOTXT}; \
+ if [ -e ${PROTOTXT_FILENAME} ] ; \
+ then \
+ echo "Adding input shape to prototxt file."; \
+ awk 'NR <2 {print}' < ${PROTOTXT_FILENAME} > temp; cat input_shape.prototxt >> temp; awk 'NR > 7 {print}' < ${PROTOTXT_FILENAME} >> temp; mv temp ${PROTOTXT_FILENAME}; \
+ else \
+ echo "***\nError - Could not download prototxt file. Check network and proxy settings \n***\n"; \
+ exit 1; \
+ fi ; \
+ fi
+
+.PHONY: caffemodel
+caffemodel:
+ @echo "\nmaking caffemodel"
+ @if [ -e ${CAFFEMODEL_FILENAME} ] ; \
+ then \
+ echo "caffemodel file already exists"; \
+ else \
+ echo "Downloading caffemodel file"; \
+ ${GET_CAFFEMODEL}; \
+ if ! [ -e ${CAFFEMODEL_FILENAME} ] ; \
+ then \
+ echo "***\nError - Could not download caffemodel file. Check network and proxy settings \n***\n"; \
+ exit 1; \
+ fi ; \
+ fi
+
+.PHONY: profile
+profile: prototxt
+ @echo "\nmaking profile"
+ ${NCPROFILE} ${PROTOTXT_FILENAME} -s 12
+
+.PHONY: browse_profile
+browse_profile: profile
+ @echo "\nmaking browse_profile"
+ @if [ -e output_report.html ] ; \
+ then \
+ firefox output_report.html & \
+ else \
+ @echo "***\nError - output_report.html not found" ; \
+ fi ;
+
+.PHONY: compile
+compile: prototxt caffemodel
+ @echo "\nmaking compile"
+ ${NCCOMPILE} -w ${CAFFEMODEL_FILENAME} -s 12 ${PROTOTXT_FILENAME}
+
+.PHONY: check
+check: prototxt caffemodel
+ @echo "\nmaking check"
+#-${NCCHECK} -w ${CAFFEMODEL_FILENAME} -i ../../data/images/cat.jpg -s 12 -id 281 ${PROTOTXT_FILENAME} -S 255 -M ../../data/ilsvrc12/ilsvrc_2012_mean.npy
+ -${NCCHECK} -w ${CAFFEMODEL_FILENAME} -i ../../data/images/cat.jpg -s 12 -id 281 ${PROTOTXT_FILENAME} -M 110 -S 255
+
+.PHONY: run
+run: compile
+ @echo "\nmaking run"
+ ./run.py
+
+.PHONY: run_py
+run_py: compile
+ @echo "\nmaking run_py"
+ ./run.py
+
+.PHONY: cpp
+cpp: cpp/run.cpp cpp/fp16.c
+ @echo "\nmaking cpp"
+ g++ cpp/run.cpp cpp/fp16.c -o cpp/run_cpp -lmvnc
+
+.PHONY: run_cpp
+run_cpp: compile cpp
+ @echo "\nmaking run_cpp"
+ cd cpp; ./run_cpp; cd ..
+
+.PHONY: help
+help:
+ @echo "possible make targets: ";
+ @echo " make help - shows this message";
+ @echo " make all - makes the following: prototxt, profile, compile, check, cpp, run_py, run_cpp";
+ @echo " make prototxt - downloads and adds input shape to Caffe prototxt file";
+ @echo " make caffemodel - downloads the caffemodel for the network"
+ @echo " make compile - runs SDK compiler tool to compile the NCS graph file for the network";
+ @echo " make check - runs SDK checker tool to verify an NCS graph file";
+ @echo " make profile - runs the SDK profiler tool to profile the network creating output_report.html";
+ @echo " make browse_profile - runs the SDK profiler tool and brings up report in browser.";
+ @echo " make cpp - builds the run_cpp executable example";
+ @echo " make run_cpp - runs the run_cpp executable example program";
+ @echo " make run_py - runs the run.py python example program:";
+ @echo " make clean - removes all created content"
+
+clean_caffe_model:
+ @echo "\nmaking clean_caffe_model"
+ rm -f ${PROTOTXT_FILENAME}
+ rm -f ${CAFFEMODEL_FILENAME}
+
+clean: clean_caffe_model
+ @echo "\nmaking clean"
+ rm -f graph
+ rm -f output.gv
+ rm -f output.gv.svg
+ rm -f output_report.html
+ rm -f output_expected.npy
+ rm -f zero_weights.caffemodel
+ rm -f output_result.npy
+ rm -f output_val.csv
+ rm -f cpp/run_cpp
--- /dev/null
+# Introduction
+The [AlexNet](https://github.com/BVLC/caffe/tree/master/models/bvlc_alexnet) network can be used for image classification. The provided Makefile does the following
+1. Downloads the Caffe prototxt file and makes any changes necessary to work with the Movidius Neural Compute SDK
+2. Downloads and generates the required ilsvrc12 data
+3. Downloads the .caffemodel file which was trained and provided by BVLC.
+3. Profiles, Compiles and Checks the network using the Neural Compute SDK.
+4. There is a python example (run.py) and a C++ example (cpp/run.cpp) which both do a single inference on an image as an example of how to use this network with the Neural Compute API thats provided in the Neural Compute SDK.
+
+# Makefile
+Provided Makefile describes various targets that help with the above mentioned tasks.
+
+## make help
+Shows makefile possible targets and brief descriptions.
+
+## make all
+Makes the following: prototxt, caffemodel, profile, compile, check, cpp, run, run_cpp.
+
+## make prototxt
+Downloads the Caffe prototxt file and makes a few changes necessary to work with the Movidius Neural Compute SDK.
+
+## make caffemodel
+Downloads the Caffe model file
+
+## make profile
+Runs the provided network on the NCS and generates per layer statistics that are helpful for understanding the performance of the network on the Neural Compute Stick. Output diplayed on terminal and the output_report.html file is also created. Demonstrates NCSDK tool: cnprofile
+
+## make browse_profile
+profiles the network similar to make profile and then brings up output_report.html in a browser. Demonstrates NCSDK tool: mvNCProfile
+
+## make compile
+Uses the network description and the trained weights files to generate a Movidius internal 'graph' format file. This file is later loaded on the Neural Compute Stick where the inferences on the network can be executed. Demonstrates NCSDK tool: mvNCCompile
+
+## make check
+Runs the network on Caffe on the CPU and compares results when run on the Neural Compute Stick. Consistency results are output to the terminal. Demonstrates the NCSDK tool: mvNCCheck.
+
+## make run_py
+Runs the provided run.py python script which sends a single image to the Neural Compute Stick and receives and displays the inference results.
+
+## make cpp
+Builds the C++ example program run_cpp which can be executed with make run_cpp.
+
+## make run_cpp
+Runs the provided run_cpp executable program that is built via make cpp. This program sends a single image to the Neural Compute Stick and receives and displays the inference results.
+
+## make clean
+Removes all the temporary and target files that are created by the Makefile.
--- /dev/null
+#! /bin/bash
+
+g++ run.cpp fp16.c -o run_cpp -lmvnc
+
+
+
--- /dev/null
+#include "fp16.h"
+
+// Copied from Numpy
+
+static unsigned half2float(unsigned short h)
+{
+ unsigned short h_exp, h_sig;
+ unsigned f_sgn, f_exp, f_sig;
+
+ h_exp = (h&0x7c00u);
+ f_sgn = ((unsigned)h&0x8000u) << 16;
+ switch (h_exp) {
+ case 0x0000u: /* 0 or subnormal */
+ h_sig = (h&0x03ffu);
+ /* Signed zero */
+ if (h_sig == 0) {
+ return f_sgn;
+ }
+ /* Subnormal */
+ h_sig <<= 1;
+ while ((h_sig&0x0400u) == 0) {
+ h_sig <<= 1;
+ h_exp++;
+ }
+ f_exp = ((unsigned)(127 - 15 - h_exp)) << 23;
+ f_sig = ((unsigned)(h_sig&0x03ffu)) << 13;
+ return f_sgn + f_exp + f_sig;
+ case 0x7c00u: /* inf or NaN */
+ /* All-ones exponent and a copy of the significand */
+ return f_sgn + 0x7f800000u + (((unsigned)(h&0x03ffu)) << 13);
+ default: /* normalized */
+ /* Just need to adjust the exponent and shift */
+ return f_sgn + (((unsigned)(h&0x7fffu) + 0x1c000u) << 13);
+ }
+}
+
+unsigned short float2half(unsigned f)
+{
+ unsigned f_exp, f_sig;
+ unsigned short h_sgn, h_exp, h_sig;
+
+ h_sgn = (unsigned short) ((f&0x80000000u) >> 16);
+ f_exp = (f&0x7f800000u);
+
+ /* Exponent overflow/NaN converts to signed inf/NaN */
+ if (f_exp >= 0x47800000u) {
+ if (f_exp == 0x7f800000u) {
+ /* Inf or NaN */
+ f_sig = (f&0x007fffffu);
+ if (f_sig != 0) {
+ /* NaN - propagate the flag in the significand... */
+ unsigned short ret = (unsigned short) (0x7c00u + (f_sig >> 13));
+ /* ...but make sure it stays a NaN */
+ if (ret == 0x7c00u) {
+ ret++;
+ }
+ return h_sgn + ret;
+ } else {
+ /* signed inf */
+ return (unsigned short) (h_sgn + 0x7c00u);
+ }
+ } else {
+ /* overflow to signed inf */
+#if NPY_HALF_GENERATE_OVERFLOW
+ npy_set_floatstatus_overflow();
+#endif
+ return (unsigned short) (h_sgn + 0x7c00u);
+ }
+ }
+
+ /* Exponent underflow converts to a subnormal half or signed zero */
+ if (f_exp <= 0x38000000u) {
+ /*
+ * Signed zeros, subnormal floats, and floats with small
+ * exponents all convert to signed zero halfs.
+ */
+ if (f_exp < 0x33000000u) {
+#if NPY_HALF_GENERATE_UNDERFLOW
+ /* If f != 0, it underflowed to 0 */
+ if ((f&0x7fffffff) != 0) {
+ npy_set_floatstatus_underflow();
+ }
+#endif
+ return h_sgn;
+ }
+ /* Make the subnormal significand */
+ f_exp >>= 23;
+ f_sig = (0x00800000u + (f&0x007fffffu));
+#if NPY_HALF_GENERATE_UNDERFLOW
+ /* If it's not exactly represented, it underflowed */
+ if ((f_sig&(((unsigned)1 << (126 - f_exp)) - 1)) != 0) {
+ npy_set_floatstatus_underflow();
+ }
+#endif
+ f_sig >>= (113 - f_exp);
+ /* Handle rounding by adding 1 to the bit beyond half precision */
+#if NPY_HALF_ROUND_TIES_TO_EVEN
+ /*
+ * If the last bit in the half significand is 0 (already even), and
+ * the remaining bit pattern is 1000...0, then we do not add one
+ * to the bit after the half significand. In all other cases, we do.
+ */
+ if ((f_sig&0x00003fffu) != 0x00001000u) {
+ f_sig += 0x00001000u;
+ }
+#else
+ f_sig += 0x00001000u;
+#endif
+ h_sig = (unsigned short) (f_sig >> 13);
+ /*
+ * If the rounding causes a bit to spill into h_exp, it will
+ * increment h_exp from zero to one and h_sig will be zero.
+ * This is the correct result.
+ */
+ return (unsigned short) (h_sgn + h_sig);
+ }
+
+ /* Regular case with no overflow or underflow */
+ h_exp = (unsigned short) ((f_exp - 0x38000000u) >> 13);
+ /* Handle rounding by adding 1 to the bit beyond half precision */
+ f_sig = (f&0x007fffffu);
+#if NPY_HALF_ROUND_TIES_TO_EVEN
+ /*
+ * If the last bit in the half significand is 0 (already even), and
+ * the remaining bit pattern is 1000...0, then we do not add one
+ * to the bit after the half significand. In all other cases, we do.
+ */
+ if ((f_sig&0x00003fffu) != 0x00001000u) {
+ f_sig += 0x00001000u;
+ }
+#else
+ f_sig += 0x00001000u;
+#endif
+ h_sig = (unsigned short) (f_sig >> 13);
+ /*
+ * If the rounding causes a bit to spill into h_exp, it will
+ * increment h_exp by one and h_sig will be zero. This is the
+ * correct result. h_exp may increment to 15, at greatest, in
+ * which case the result overflows to a signed inf.
+ */
+#if NPY_HALF_GENERATE_OVERFLOW
+ h_sig += h_exp;
+ if (h_sig == 0x7c00u) {
+ npy_set_floatstatus_overflow();
+ }
+ return h_sgn + h_sig;
+#else
+ return h_sgn + h_exp + h_sig;
+#endif
+}
+
+void floattofp16(unsigned char *dst, float *src, unsigned nelem)
+{
+ unsigned i;
+ unsigned short *_dst = (unsigned short *)dst;
+ unsigned *_src = (unsigned *)src;
+
+ for(i = 0; i < nelem; i++)
+ _dst[i] = float2half(_src[i]);
+}
+
+void fp16tofloat(float *dst, unsigned char *src, unsigned nelem)
+{
+ unsigned i;
+ unsigned *_dst = (unsigned *)dst;
+ unsigned short *_src = (unsigned short *)src;
+
+ for(i = 0; i < nelem; i++)
+ _dst[i] = half2float(_src[i]);
+}
+
--- /dev/null
+// Copied from Numpy
+
+
+static unsigned half2float(unsigned short h);
+unsigned short float2half(unsigned f);
+void floattofp16(unsigned char *dst, float *src, unsigned nelem);
+void fp16tofloat(float *dst, unsigned char *src, unsigned nelem);
+
+
--- /dev/null
+// Copyright 2017 Intel Corporation.
+// The source code, information and material ("Material") contained herein is
+// owned by Intel Corporation or its suppliers or licensors, and title to such
+// Material remains with Intel Corporation or its suppliers or licensors.
+// The Material contains proprietary information of Intel or its suppliers and
+// licensors. The Material is protected by worldwide copyright laws and treaty
+// provisions.
+// No part of the Material may be used, copied, reproduced, modified, published,
+// uploaded, posted, transmitted, distributed or disclosed in any way without
+// Intel's prior express written permission. No license under any patent,
+// copyright or other intellectual property rights in the Material is granted to
+// or conferred upon you, either expressly, by implication, inducement, estoppel
+// or otherwise.
+// Any license under such intellectual property rights must be express and
+// approved by Intel in writing.
+
+#include <stdio.h>
+#include <stdlib.h>
+
+#define STB_IMAGE_IMPLEMENTATION
+#include "stb_image.h"
+
+#define STB_IMAGE_RESIZE_IMPLEMENTATION
+#include "stb_image_resize.h"
+
+#include "fp16.h"
+#include <mvnc.h>
+
+
+// somewhat arbitrary buffer size for the device name
+#define NAME_SIZE 100
+
+// graph file name - assume we are running in this directory: ncsdk/examples/caffe/GoogLeNet/cpp
+#define GRAPH_FILE_NAME "../graph"
+
+// image file name - assume we are running in this directory: ncsdk/examples/caffe/GoogLeNet/cpp
+#define IMAGE_FILE_NAME "../../../data/images/nps_electric_guitar.png"
+
+
+// 16 bits. will use this to store half precision floats since C++ has no
+// built in support for it.
+typedef unsigned short half;
+
+// GoogleNet image dimensions, network mean values for each channel in BGR order.
+const int networkDim = 227;
+float networkMean[] = {0.40787054*255.0, 0.45752458*255.0, 0.48109378*255.0};
+
+// Load a graph file
+// caller must free the buffer returned.
+void *LoadFile(const char *path, unsigned int *length)
+{
+ FILE *fp;
+ char *buf;
+
+ fp = fopen(path, "rb");
+ if(fp == NULL)
+ return 0;
+ fseek(fp, 0, SEEK_END);
+ *length = ftell(fp);
+ rewind(fp);
+ if(!(buf = (char*) malloc(*length)))
+ {
+ fclose(fp);
+ return 0;
+ }
+ if(fread(buf, 1, *length, fp) != *length)
+ {
+ fclose(fp);
+ free(buf);
+ return 0;
+ }
+ fclose(fp);
+ return buf;
+}
+
+
+half *LoadImage(const char *path, int reqsize, float *mean)
+{
+ int width, height, cp, i;
+ unsigned char *img, *imgresized;
+ float *imgfp32;
+ half *imgfp16;
+
+ img = stbi_load(path, &width, &height, &cp, 3);
+ if(!img)
+ {
+ printf("The picture %s could not be loaded\n", path);
+ return 0;
+ }
+ imgresized = (unsigned char*) malloc(3*reqsize*reqsize);
+ if(!imgresized)
+ {
+ free(img);
+ perror("malloc");
+ return 0;
+ }
+ stbir_resize_uint8(img, width, height, 0, imgresized, reqsize, reqsize, 0, 3);
+ free(img);
+ imgfp32 = (float*) malloc(sizeof(*imgfp32) * reqsize * reqsize * 3);
+ if(!imgfp32)
+ {
+ free(imgresized);
+ perror("malloc");
+ return 0;
+ }
+ for(i = 0; i < reqsize * reqsize * 3; i++)
+ imgfp32[i] = imgresized[i];
+ free(imgresized);
+ imgfp16 = (half*) malloc(sizeof(*imgfp16) * reqsize * reqsize * 3);
+ if(!imgfp16)
+ {
+ free(imgfp32);
+ perror("malloc");
+ return 0;
+ }
+ for(i = 0; i < reqsize*reqsize; i++)
+ {
+ float blue, green, red;
+ blue = imgfp32[3*i+2];
+ green = imgfp32[3*i+1];
+ red = imgfp32[3*i+0];
+
+ imgfp32[3*i+0] = blue-mean[0];
+ imgfp32[3*i+1] = green-mean[1];
+ imgfp32[3*i+2] = red-mean[2];
+
+ // uncomment to see what values are getting passed to mvncLoadTensor() before conversion to half float
+ //printf("Blue: %f, Grean: %f, Red: %f \n", imgfp32[3*i+0], imgfp32[3*i+1], imgfp32[3*i+2]);
+ }
+ floattofp16((unsigned char *)imgfp16, imgfp32, 3*reqsize*reqsize);
+ free(imgfp32);
+ return imgfp16;
+}
+
+
+int main(int argc, char** argv)
+{
+ mvncStatus retCode;
+ void *deviceHandle;
+ char devName[NAME_SIZE];
+ retCode = mvncGetDeviceName(0, devName, NAME_SIZE);
+ if (retCode != MVNC_OK)
+ { // failed to get device name, maybe none plugged in.
+ printf("No NCS devices found\n");
+ exit(-1);
+ }
+
+ // Try to open the NCS device via the device name
+ retCode = mvncOpenDevice(devName, &deviceHandle);
+ if (retCode != MVNC_OK)
+ { // failed to open the device.
+ printf("Could not open NCS device\n");
+ exit(-1);
+ }
+
+ // deviceHandle is ready to use now.
+ // Pass it to other NC API calls as needed and close it when finished.
+ printf("Successfully opened NCS device!\n");
+
+ // Now read in a graph file
+ unsigned int graphFileLen;
+ void* graphFileBuf = LoadFile(GRAPH_FILE_NAME, &graphFileLen);
+
+ // allocate the graph
+ void* graphHandle;
+ retCode = mvncAllocateGraph(deviceHandle, &graphHandle, graphFileBuf, graphFileLen);
+ if (retCode != MVNC_OK)
+ { // error allocating graph
+ printf("Could not allocate graph for file: %s\n", GRAPH_FILE_NAME);
+ printf("Error from mvncAllocateGraph is: %d\n", retCode);
+ }
+ else
+ { // successfully allocated graph. Now graphHandle is ready to go.
+ // use graphHandle for other API calls and call mvncDeallocateGraph
+ // when done with it.
+ printf("Successfully allocated graph for %s\n", GRAPH_FILE_NAME);
+
+ // LoadImage will read image from disk, convert channels to floats
+ // subtract network mean for each value in each channel. Then, convert
+ // floats to half precision floats and return pointer to the buffer
+ // of half precision floats (Fp16s)
+ half* imageBufFp16 = LoadImage(IMAGE_FILE_NAME, networkDim, networkMean);
+
+ // calculate the length of the buffer that contains the half precision floats.
+ // 3 channels * width * height * sizeof a 16bit float
+ unsigned int lenBufFp16 = 3*networkDim*networkDim*sizeof(*imageBufFp16);
+
+ // start the inference with mvncLoadTensor()
+ retCode = mvncLoadTensor(graphHandle, imageBufFp16, lenBufFp16, NULL);
+ if (retCode != MVNC_OK)
+ { // error loading tensor
+ printf("Could not load tensor\n");
+ printf("Error from mvncLoadTensor is: %d\n", retCode);
+ }
+ else
+ { // the inference has been started, now call mvncGetResult() for the
+ // inference result
+ printf("Successfully loaded the tensor for image %s\n", IMAGE_FILE_NAME);
+
+ void* resultData16;
+ void* userParam;
+ unsigned int lenResultData;
+ retCode = mvncGetResult(graphHandle, &resultData16, &lenResultData, &userParam);
+ if (retCode == MVNC_OK)
+ { // Successfully got the result. The inference result is in the buffer pointed to by resultData
+ printf("Successfully got the inference result for image %s\n", IMAGE_FILE_NAME);
+ printf("resultData is %d bytes which is %d 16-bit floats.\n", lenResultData, lenResultData/(int)sizeof(half));
+
+ // convert half precision floats to full floats
+ int numResults = lenResultData / sizeof(half);
+ float* resultData32;
+ resultData32 = (float*)malloc(numResults * sizeof(*resultData32));
+ fp16tofloat(resultData32, (unsigned char*)resultData16, numResults);
+
+ float maxResult = 0.0;
+ int maxIndex = -1;
+ for (int index = 0; index < numResults; index++)
+ {
+ // printf("Category %d is: %f\n", index, resultData32[index]);
+ if (resultData32[index] > maxResult)
+ {
+ maxResult = resultData32[index];
+ maxIndex = index;
+ }
+ }
+ printf("Index of top result is: %d\n", maxIndex);
+ printf("Probability of top result is: %f\n", resultData32[maxIndex]);
+ }
+ }
+
+ retCode = mvncDeallocateGraph(graphHandle);
+ graphHandle = NULL;
+ }
+
+ free(graphFileBuf);
+ retCode = mvncCloseDevice(deviceHandle);
+ deviceHandle = NULL;
+}
--- /dev/null
+/* stb_image - v2.15 - public domain image loader - http://nothings.org/stb_image.h
+ no warranty implied; use at your own risk
+
+ Do this:
+ #define STB_IMAGE_IMPLEMENTATION
+ before you include this file in *one* C or C++ file to create the implementation.
+
+ // i.e. it should look like this:
+ #include ...
+ #include ...
+ #include ...
+ #define STB_IMAGE_IMPLEMENTATION
+ #include "stb_image.h"
+
+ You can #define STBI_ASSERT(x) before the #include to avoid using assert.h.
+ And #define STBI_MALLOC, STBI_REALLOC, and STBI_FREE to avoid using malloc,realloc,free
+
+
+ QUICK NOTES:
+ Primarily of interest to game developers and other people who can
+ avoid problematic images and only need the trivial interface
+
+ JPEG baseline & progressive (12 bpc/arithmetic not supported, same as stock IJG lib)
+ PNG 1/2/4/8/16-bit-per-channel
+
+ TGA (not sure what subset, if a subset)
+ BMP non-1bpp, non-RLE
+ PSD (composited view only, no extra channels, 8/16 bit-per-channel)
+
+ GIF (*comp always reports as 4-channel)
+ HDR (radiance rgbE format)
+ PIC (Softimage PIC)
+ PNM (PPM and PGM binary only)
+
+ Animated GIF still needs a proper API, but here's one way to do it:
+ http://gist.github.com/urraka/685d9a6340b26b830d49
+
+ - decode from memory or through FILE (define STBI_NO_STDIO to remove code)
+ - decode from arbitrary I/O callbacks
+ - SIMD acceleration on x86/x64 (SSE2) and ARM (NEON)
+
+ Full documentation under "DOCUMENTATION" below.
+
+
+LICENSE
+
+ See end of file for license information.
+
+RECENT REVISION HISTORY:
+
+ 2.15 (2017-03-18) fix png-1,2,4; all Imagenet JPGs; no runtime SSE detection on GCC
+ 2.14 (2017-03-03) remove deprecated STBI_JPEG_OLD; fixes for Imagenet JPGs
+ 2.13 (2016-12-04) experimental 16-bit API, only for PNG so far; fixes
+ 2.12 (2016-04-02) fix typo in 2.11 PSD fix that caused crashes
+ 2.11 (2016-04-02) 16-bit PNGS; enable SSE2 in non-gcc x64
+ RGB-format JPEG; remove white matting in PSD;
+ allocate large structures on the stack;
+ correct channel count for PNG & BMP
+ 2.10 (2016-01-22) avoid warning introduced in 2.09
+ 2.09 (2016-01-16) 16-bit TGA; comments in PNM files; STBI_REALLOC_SIZED
+ 2.08 (2015-09-13) fix to 2.07 cleanup, reading RGB PSD as RGBA
+ 2.07 (2015-09-13) partial animated GIF support
+ limited 16-bit PSD support
+ minor bugs, code cleanup, and compiler warnings
+
+ See end of file for full revision history.
+
+
+ ============================ Contributors =========================
+
+ Image formats Extensions, features
+ Sean Barrett (jpeg, png, bmp) Jetro Lauha (stbi_info)
+ Nicolas Schulz (hdr, psd) Martin "SpartanJ" Golini (stbi_info)
+ Jonathan Dummer (tga) James "moose2000" Brown (iPhone PNG)
+ Jean-Marc Lienher (gif) Ben "Disch" Wenger (io callbacks)
+ Tom Seddon (pic) Omar Cornut (1/2/4-bit PNG)
+ Thatcher Ulrich (psd) Nicolas Guillemot (vertical flip)
+ Ken Miller (pgm, ppm) Richard Mitton (16-bit PSD)
+ github:urraka (animated gif) Junggon Kim (PNM comments)
+ Daniel Gibson (16-bit TGA)
+ socks-the-fox (16-bit PNG)
+ Jeremy Sawicki (handle all ImageNet JPGs)
+ Optimizations & bugfixes
+ Fabian "ryg" Giesen
+ Arseny Kapoulkine
+
+ Bug & warning fixes
+ Marc LeBlanc David Woo Guillaume George Martins Mozeiko
+ Christpher Lloyd Jerry Jansson Joseph Thomson Phil Jordan
+ Dave Moore Roy Eltham Hayaki Saito Nathan Reed
+ Won Chun Luke Graham Johan Duparc Nick Verigakis
+ the Horde3D community Thomas Ruf Ronny Chevalier Baldur Karlsson
+ Janez Zemva John Bartholomew Michal Cichon github:rlyeh
+ Jonathan Blow Ken Hamada Tero Hanninen github:romigrou
+ Laurent Gomila Cort Stratton Sergio Gonzalez github:svdijk
+ Aruelien Pocheville Thibault Reuille Cass Everitt github:snagar
+ Ryamond Barbiero Paul Du Bois Engin Manap github:Zelex
+ Michaelangel007@github Philipp Wiesemann Dale Weiler github:grim210
+ Oriol Ferrer Mesia Josh Tobin Matthew Gregan github:sammyhw
+ Blazej Dariusz Roszkowski Gregory Mullen github:phprus
+
+*/
+
+#ifndef STBI_INCLUDE_STB_IMAGE_H
+#define STBI_INCLUDE_STB_IMAGE_H
+
+// DOCUMENTATION
+//
+// Limitations:
+// - no 16-bit-per-channel PNG
+// - no 12-bit-per-channel JPEG
+// - no JPEGs with arithmetic coding
+// - no 1-bit BMP
+// - GIF always returns *comp=4
+//
+// Basic usage (see HDR discussion below for HDR usage):
+// int x,y,n;
+// unsigned char *data = stbi_load(filename, &x, &y, &n, 0);
+// // ... process data if not NULL ...
+// // ... x = width, y = height, n = # 8-bit components per pixel ...
+// // ... replace '0' with '1'..'4' to force that many components per pixel
+// // ... but 'n' will always be the number that it would have been if you said 0
+// stbi_image_free(data)
+//
+// Standard parameters:
+// int *x -- outputs image width in pixels
+// int *y -- outputs image height in pixels
+// int *channels_in_file -- outputs # of image components in image file
+// int desired_channels -- if non-zero, # of image components requested in result
+//
+// The return value from an image loader is an 'unsigned char *' which points
+// to the pixel data, or NULL on an allocation failure or if the image is
+// corrupt or invalid. The pixel data consists of *y scanlines of *x pixels,
+// with each pixel consisting of N interleaved 8-bit components; the first
+// pixel pointed to is top-left-most in the image. There is no padding between
+// image scanlines or between pixels, regardless of format. The number of
+// components N is 'req_comp' if req_comp is non-zero, or *comp otherwise.
+// If req_comp is non-zero, *comp has the number of components that _would_
+// have been output otherwise. E.g. if you set req_comp to 4, you will always
+// get RGBA output, but you can check *comp to see if it's trivially opaque
+// because e.g. there were only 3 channels in the source image.
+//
+// An output image with N components has the following components interleaved
+// in this order in each pixel:
+//
+// N=#comp components
+// 1 grey
+// 2 grey, alpha
+// 3 red, green, blue
+// 4 red, green, blue, alpha
+//
+// If image loading fails for any reason, the return value will be NULL,
+// and *x, *y, *comp will be unchanged. The function stbi_failure_reason()
+// can be queried for an extremely brief, end-user unfriendly explanation
+// of why the load failed. Define STBI_NO_FAILURE_STRINGS to avoid
+// compiling these strings at all, and STBI_FAILURE_USERMSG to get slightly
+// more user-friendly ones.
+//
+// Paletted PNG, BMP, GIF, and PIC images are automatically depalettized.
+//
+// ===========================================================================
+//
+// Philosophy
+//
+// stb libraries are designed with the following priorities:
+//
+// 1. easy to use
+// 2. easy to maintain
+// 3. good performance
+//
+// Sometimes I let "good performance" creep up in priority over "easy to maintain",
+// and for best performance I may provide less-easy-to-use APIs that give higher
+// performance, in addition to the easy to use ones. Nevertheless, it's important
+// to keep in mind that from the standpoint of you, a client of this library,
+// all you care about is #1 and #3, and stb libraries DO NOT emphasize #3 above all.
+//
+// Some secondary priorities arise directly from the first two, some of which
+// make more explicit reasons why performance can't be emphasized.
+//
+// - Portable ("ease of use")
+// - Small source code footprint ("easy to maintain")
+// - No dependencies ("ease of use")
+//
+// ===========================================================================
+//
+// I/O callbacks
+//
+// I/O callbacks allow you to read from arbitrary sources, like packaged
+// files or some other source. Data read from callbacks are processed
+// through a small internal buffer (currently 128 bytes) to try to reduce
+// overhead.
+//
+// The three functions you must define are "read" (reads some bytes of data),
+// "skip" (skips some bytes of data), "eof" (reports if the stream is at the end).
+//
+// ===========================================================================
+//
+// SIMD support
+//
+// The JPEG decoder will try to automatically use SIMD kernels on x86 when
+// supported by the compiler. For ARM Neon support, you must explicitly
+// request it.
+//
+// (The old do-it-yourself SIMD API is no longer supported in the current
+// code.)
+//
+// On x86, SSE2 will automatically be used when available based on a run-time
+// test; if not, the generic C versions are used as a fall-back. On ARM targets,
+// the typical path is to have separate builds for NEON and non-NEON devices
+// (at least this is true for iOS and Android). Therefore, the NEON support is
+// toggled by a build flag: define STBI_NEON to get NEON loops.
+//
+// If for some reason you do not want to use any of SIMD code, or if
+// you have issues compiling it, you can disable it entirely by
+// defining STBI_NO_SIMD.
+//
+// ===========================================================================
+//
+// HDR image support (disable by defining STBI_NO_HDR)
+//
+// stb_image now supports loading HDR images in general, and currently
+// the Radiance .HDR file format, although the support is provided
+// generically. You can still load any file through the existing interface;
+// if you attempt to load an HDR file, it will be automatically remapped to
+// LDR, assuming gamma 2.2 and an arbitrary scale factor defaulting to 1;
+// both of these constants can be reconfigured through this interface:
+//
+// stbi_hdr_to_ldr_gamma(2.2f);
+// stbi_hdr_to_ldr_scale(1.0f);
+//
+// (note, do not use _inverse_ constants; stbi_image will invert them
+// appropriately).
+//
+// Additionally, there is a new, parallel interface for loading files as
+// (linear) floats to preserve the full dynamic range:
+//
+// float *data = stbi_loadf(filename, &x, &y, &n, 0);
+//
+// If you load LDR images through this interface, those images will
+// be promoted to floating point values, run through the inverse of
+// constants corresponding to the above:
+//
+// stbi_ldr_to_hdr_scale(1.0f);
+// stbi_ldr_to_hdr_gamma(2.2f);
+//
+// Finally, given a filename (or an open file or memory block--see header
+// file for details) containing image data, you can query for the "most
+// appropriate" interface to use (that is, whether the image is HDR or
+// not), using:
+//
+// stbi_is_hdr(char *filename);
+//
+// ===========================================================================
+//
+// iPhone PNG support:
+//
+// By default we convert iphone-formatted PNGs back to RGB, even though
+// they are internally encoded differently. You can disable this conversion
+// by by calling stbi_convert_iphone_png_to_rgb(0), in which case
+// you will always just get the native iphone "format" through (which
+// is BGR stored in RGB).
+//
+// Call stbi_set_unpremultiply_on_load(1) as well to force a divide per
+// pixel to remove any premultiplied alpha *only* if the image file explicitly
+// says there's premultiplied data (currently only happens in iPhone images,
+// and only if iPhone convert-to-rgb processing is on).
+//
+// ===========================================================================
+//
+// ADDITIONAL CONFIGURATION
+//
+// - You can suppress implementation of any of the decoders to reduce
+// your code footprint by #defining one or more of the following
+// symbols before creating the implementation.
+//
+// STBI_NO_JPEG
+// STBI_NO_PNG
+// STBI_NO_BMP
+// STBI_NO_PSD
+// STBI_NO_TGA
+// STBI_NO_GIF
+// STBI_NO_HDR
+// STBI_NO_PIC
+// STBI_NO_PNM (.ppm and .pgm)
+//
+// - You can request *only* certain decoders and suppress all other ones
+// (this will be more forward-compatible, as addition of new decoders
+// doesn't require you to disable them explicitly):
+//
+// STBI_ONLY_JPEG
+// STBI_ONLY_PNG
+// STBI_ONLY_BMP
+// STBI_ONLY_PSD
+// STBI_ONLY_TGA
+// STBI_ONLY_GIF
+// STBI_ONLY_HDR
+// STBI_ONLY_PIC
+// STBI_ONLY_PNM (.ppm and .pgm)
+//
+// - If you use STBI_NO_PNG (or _ONLY_ without PNG), and you still
+// want the zlib decoder to be available, #define STBI_SUPPORT_ZLIB
+//
+
+
+#ifndef STBI_NO_STDIO
+#include <stdio.h>
+#endif // STBI_NO_STDIO
+
+#define STBI_VERSION 1
+
+enum
+{
+ STBI_default = 0, // only used for req_comp
+
+ STBI_grey = 1,
+ STBI_grey_alpha = 2,
+ STBI_rgb = 3,
+ STBI_rgb_alpha = 4
+};
+
+typedef unsigned char stbi_uc;
+typedef unsigned short stbi_us;
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifdef STB_IMAGE_STATIC
+#define STBIDEF static
+#else
+#define STBIDEF extern
+#endif
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// PRIMARY API - works on images of any type
+//
+
+//
+// load image by filename, open file, or memory buffer
+//
+
+typedef struct
+{
+ int (*read) (void *user,char *data,int size); // fill 'data' with 'size' bytes. return number of bytes actually read
+ void (*skip) (void *user,int n); // skip the next 'n' bytes, or 'unget' the last -n bytes if negative
+ int (*eof) (void *user); // returns nonzero if we are at end of file/data
+} stbi_io_callbacks;
+
+////////////////////////////////////
+//
+// 8-bits-per-channel interface
+//
+
+STBIDEF stbi_uc *stbi_load (char const *filename, int *x, int *y, int *channels_in_file, int desired_channels);
+STBIDEF stbi_uc *stbi_load_from_memory (stbi_uc const *buffer, int len , int *x, int *y, int *channels_in_file, int desired_channels);
+STBIDEF stbi_uc *stbi_load_from_callbacks(stbi_io_callbacks const *clbk , void *user, int *x, int *y, int *channels_in_file, int desired_channels);
+
+#ifndef STBI_NO_STDIO
+STBIDEF stbi_uc *stbi_load_from_file (FILE *f, int *x, int *y, int *channels_in_file, int desired_channels);
+// for stbi_load_from_file, file pointer is left pointing immediately after image
+#endif
+
+////////////////////////////////////
+//
+// 16-bits-per-channel interface
+//
+
+STBIDEF stbi_us *stbi_load_16(char const *filename, int *x, int *y, int *channels_in_file, int desired_channels);
+#ifndef STBI_NO_STDIO
+STBIDEF stbi_us *stbi_load_from_file_16(FILE *f, int *x, int *y, int *channels_in_file, int desired_channels);
+#endif
+// @TODO the other variants
+
+////////////////////////////////////
+//
+// float-per-channel interface
+//
+#ifndef STBI_NO_LINEAR
+ STBIDEF float *stbi_loadf (char const *filename, int *x, int *y, int *channels_in_file, int desired_channels);
+ STBIDEF float *stbi_loadf_from_memory (stbi_uc const *buffer, int len, int *x, int *y, int *channels_in_file, int desired_channels);
+ STBIDEF float *stbi_loadf_from_callbacks (stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *channels_in_file, int desired_channels);
+
+ #ifndef STBI_NO_STDIO
+ STBIDEF float *stbi_loadf_from_file (FILE *f, int *x, int *y, int *channels_in_file, int desired_channels);
+ #endif
+#endif
+
+#ifndef STBI_NO_HDR
+ STBIDEF void stbi_hdr_to_ldr_gamma(float gamma);
+ STBIDEF void stbi_hdr_to_ldr_scale(float scale);
+#endif // STBI_NO_HDR
+
+#ifndef STBI_NO_LINEAR
+ STBIDEF void stbi_ldr_to_hdr_gamma(float gamma);
+ STBIDEF void stbi_ldr_to_hdr_scale(float scale);
+#endif // STBI_NO_LINEAR
+
+// stbi_is_hdr is always defined, but always returns false if STBI_NO_HDR
+STBIDEF int stbi_is_hdr_from_callbacks(stbi_io_callbacks const *clbk, void *user);
+STBIDEF int stbi_is_hdr_from_memory(stbi_uc const *buffer, int len);
+#ifndef STBI_NO_STDIO
+STBIDEF int stbi_is_hdr (char const *filename);
+STBIDEF int stbi_is_hdr_from_file(FILE *f);
+#endif // STBI_NO_STDIO
+
+
+// get a VERY brief reason for failure
+// NOT THREADSAFE
+STBIDEF const char *stbi_failure_reason (void);
+
+// free the loaded image -- this is just free()
+STBIDEF void stbi_image_free (void *retval_from_stbi_load);
+
+// get image dimensions & components without fully decoding
+STBIDEF int stbi_info_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp);
+STBIDEF int stbi_info_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp);
+
+#ifndef STBI_NO_STDIO
+STBIDEF int stbi_info (char const *filename, int *x, int *y, int *comp);
+STBIDEF int stbi_info_from_file (FILE *f, int *x, int *y, int *comp);
+
+#endif
+
+
+
+// for image formats that explicitly notate that they have premultiplied alpha,
+// we just return the colors as stored in the file. set this flag to force
+// unpremultiplication. results are undefined if the unpremultiply overflow.
+STBIDEF void stbi_set_unpremultiply_on_load(int flag_true_if_should_unpremultiply);
+
+// indicate whether we should process iphone images back to canonical format,
+// or just pass them through "as-is"
+STBIDEF void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert);
+
+// flip the image vertically, so the first pixel in the output array is the bottom left
+STBIDEF void stbi_set_flip_vertically_on_load(int flag_true_if_should_flip);
+
+// ZLIB client - used by PNG, available for other purposes
+
+STBIDEF char *stbi_zlib_decode_malloc_guesssize(const char *buffer, int len, int initial_size, int *outlen);
+STBIDEF char *stbi_zlib_decode_malloc_guesssize_headerflag(const char *buffer, int len, int initial_size, int *outlen, int parse_header);
+STBIDEF char *stbi_zlib_decode_malloc(const char *buffer, int len, int *outlen);
+STBIDEF int stbi_zlib_decode_buffer(char *obuffer, int olen, const char *ibuffer, int ilen);
+
+STBIDEF char *stbi_zlib_decode_noheader_malloc(const char *buffer, int len, int *outlen);
+STBIDEF int stbi_zlib_decode_noheader_buffer(char *obuffer, int olen, const char *ibuffer, int ilen);
+
+
+#ifdef __cplusplus
+}
+#endif
+
+//
+//
+//// end header file /////////////////////////////////////////////////////
+#endif // STBI_INCLUDE_STB_IMAGE_H
+
+#ifdef STB_IMAGE_IMPLEMENTATION
+
+#if defined(STBI_ONLY_JPEG) || defined(STBI_ONLY_PNG) || defined(STBI_ONLY_BMP) \
+ || defined(STBI_ONLY_TGA) || defined(STBI_ONLY_GIF) || defined(STBI_ONLY_PSD) \
+ || defined(STBI_ONLY_HDR) || defined(STBI_ONLY_PIC) || defined(STBI_ONLY_PNM) \
+ || defined(STBI_ONLY_ZLIB)
+ #ifndef STBI_ONLY_JPEG
+ #define STBI_NO_JPEG
+ #endif
+ #ifndef STBI_ONLY_PNG
+ #define STBI_NO_PNG
+ #endif
+ #ifndef STBI_ONLY_BMP
+ #define STBI_NO_BMP
+ #endif
+ #ifndef STBI_ONLY_PSD
+ #define STBI_NO_PSD
+ #endif
+ #ifndef STBI_ONLY_TGA
+ #define STBI_NO_TGA
+ #endif
+ #ifndef STBI_ONLY_GIF
+ #define STBI_NO_GIF
+ #endif
+ #ifndef STBI_ONLY_HDR
+ #define STBI_NO_HDR
+ #endif
+ #ifndef STBI_ONLY_PIC
+ #define STBI_NO_PIC
+ #endif
+ #ifndef STBI_ONLY_PNM
+ #define STBI_NO_PNM
+ #endif
+#endif
+
+#if defined(STBI_NO_PNG) && !defined(STBI_SUPPORT_ZLIB) && !defined(STBI_NO_ZLIB)
+#define STBI_NO_ZLIB
+#endif
+
+
+#include <stdarg.h>
+#include <stddef.h> // ptrdiff_t on osx
+#include <stdlib.h>
+#include <string.h>
+#include <limits.h>
+
+#if !defined(STBI_NO_LINEAR) || !defined(STBI_NO_HDR)
+#include <math.h> // ldexp
+#endif
+
+#ifndef STBI_NO_STDIO
+#include <stdio.h>
+#endif
+
+#ifndef STBI_ASSERT
+#include <assert.h>
+#define STBI_ASSERT(x) assert(x)
+#endif
+
+
+#ifndef _MSC_VER
+ #ifdef __cplusplus
+ #define stbi_inline inline
+ #else
+ #define stbi_inline
+ #endif
+#else
+ #define stbi_inline __forceinline
+#endif
+
+
+#ifdef _MSC_VER
+typedef unsigned short stbi__uint16;
+typedef signed short stbi__int16;
+typedef unsigned int stbi__uint32;
+typedef signed int stbi__int32;
+#else
+#include <stdint.h>
+typedef uint16_t stbi__uint16;
+typedef int16_t stbi__int16;
+typedef uint32_t stbi__uint32;
+typedef int32_t stbi__int32;
+#endif
+
+// should produce compiler error if size is wrong
+typedef unsigned char validate_uint32[sizeof(stbi__uint32)==4 ? 1 : -1];
+
+#ifdef _MSC_VER
+#define STBI_NOTUSED(v) (void)(v)
+#else
+#define STBI_NOTUSED(v) (void)sizeof(v)
+#endif
+
+#ifdef _MSC_VER
+#define STBI_HAS_LROTL
+#endif
+
+#ifdef STBI_HAS_LROTL
+ #define stbi_lrot(x,y) _lrotl(x,y)
+#else
+ #define stbi_lrot(x,y) (((x) << (y)) | ((x) >> (32 - (y))))
+#endif
+
+#if defined(STBI_MALLOC) && defined(STBI_FREE) && (defined(STBI_REALLOC) || defined(STBI_REALLOC_SIZED))
+// ok
+#elif !defined(STBI_MALLOC) && !defined(STBI_FREE) && !defined(STBI_REALLOC) && !defined(STBI_REALLOC_SIZED)
+// ok
+#else
+#error "Must define all or none of STBI_MALLOC, STBI_FREE, and STBI_REALLOC (or STBI_REALLOC_SIZED)."
+#endif
+
+#ifndef STBI_MALLOC
+#define STBI_MALLOC(sz) malloc(sz)
+#define STBI_REALLOC(p,newsz) realloc(p,newsz)
+#define STBI_FREE(p) free(p)
+#endif
+
+#ifndef STBI_REALLOC_SIZED
+#define STBI_REALLOC_SIZED(p,oldsz,newsz) STBI_REALLOC(p,newsz)
+#endif
+
+// x86/x64 detection
+#if defined(__x86_64__) || defined(_M_X64)
+#define STBI__X64_TARGET
+#elif defined(__i386) || defined(_M_IX86)
+#define STBI__X86_TARGET
+#endif
+
+#if defined(__GNUC__) && defined(STBI__X86_TARGET) && !defined(__SSE2__) && !defined(STBI_NO_SIMD)
+// gcc doesn't support sse2 intrinsics unless you compile with -msse2,
+// which in turn means it gets to use SSE2 everywhere. This is unfortunate,
+// but previous attempts to provide the SSE2 functions with runtime
+// detection caused numerous issues. The way architecture extensions are
+// exposed in GCC/Clang is, sadly, not really suited for one-file libs.
+// New behavior: if compiled with -msse2, we use SSE2 without any
+// detection; if not, we don't use it at all.
+#define STBI_NO_SIMD
+#endif
+
+#if defined(__MINGW32__) && defined(STBI__X86_TARGET) && !defined(STBI_MINGW_ENABLE_SSE2) && !defined(STBI_NO_SIMD)
+// Note that __MINGW32__ doesn't actually mean 32-bit, so we have to avoid STBI__X64_TARGET
+//
+// 32-bit MinGW wants ESP to be 16-byte aligned, but this is not in the
+// Windows ABI and VC++ as well as Windows DLLs don't maintain that invariant.
+// As a result, enabling SSE2 on 32-bit MinGW is dangerous when not
+// simultaneously enabling "-mstackrealign".
+//
+// See https://github.com/nothings/stb/issues/81 for more information.
+//
+// So default to no SSE2 on 32-bit MinGW. If you've read this far and added
+// -mstackrealign to your build settings, feel free to #define STBI_MINGW_ENABLE_SSE2.
+#define STBI_NO_SIMD
+#endif
+
+#if !defined(STBI_NO_SIMD) && (defined(STBI__X86_TARGET) || defined(STBI__X64_TARGET))
+#define STBI_SSE2
+#include <emmintrin.h>
+
+#ifdef _MSC_VER
+
+#if _MSC_VER >= 1400 // not VC6
+#include <intrin.h> // __cpuid
+static int stbi__cpuid3(void)
+{
+ int info[4];
+ __cpuid(info,1);
+ return info[3];
+}
+#else
+static int stbi__cpuid3(void)
+{
+ int res;
+ __asm {
+ mov eax,1
+ cpuid
+ mov res,edx
+ }
+ return res;
+}
+#endif
+
+#define STBI_SIMD_ALIGN(type, name) __declspec(align(16)) type name
+
+static int stbi__sse2_available()
+{
+ int info3 = stbi__cpuid3();
+ return ((info3 >> 26) & 1) != 0;
+}
+#else // assume GCC-style if not VC++
+#define STBI_SIMD_ALIGN(type, name) type name __attribute__((aligned(16)))
+
+static int stbi__sse2_available()
+{
+ // If we're even attempting to compile this on GCC/Clang, that means
+ // -msse2 is on, which means the compiler is allowed to use SSE2
+ // instructions at will, and so are we.
+ return 1;
+}
+#endif
+#endif
+
+// ARM NEON
+#if defined(STBI_NO_SIMD) && defined(STBI_NEON)
+#undef STBI_NEON
+#endif
+
+#ifdef STBI_NEON
+#include <arm_neon.h>
+// assume GCC or Clang on ARM targets
+#define STBI_SIMD_ALIGN(type, name) type name __attribute__((aligned(16)))
+#endif
+
+#ifndef STBI_SIMD_ALIGN
+#define STBI_SIMD_ALIGN(type, name) type name
+#endif
+
+///////////////////////////////////////////////
+//
+// stbi__context struct and start_xxx functions
+
+// stbi__context structure is our basic context used by all images, so it
+// contains all the IO context, plus some basic image information
+typedef struct
+{
+ stbi__uint32 img_x, img_y;
+ int img_n, img_out_n;
+
+ stbi_io_callbacks io;
+ void *io_user_data;
+
+ int read_from_callbacks;
+ int buflen;
+ stbi_uc buffer_start[128];
+
+ stbi_uc *img_buffer, *img_buffer_end;
+ stbi_uc *img_buffer_original, *img_buffer_original_end;
+} stbi__context;
+
+
+static void stbi__refill_buffer(stbi__context *s);
+
+// initialize a memory-decode context
+static void stbi__start_mem(stbi__context *s, stbi_uc const *buffer, int len)
+{
+ s->io.read = NULL;
+ s->read_from_callbacks = 0;
+ s->img_buffer = s->img_buffer_original = (stbi_uc *) buffer;
+ s->img_buffer_end = s->img_buffer_original_end = (stbi_uc *) buffer+len;
+}
+
+// initialize a callback-based context
+static void stbi__start_callbacks(stbi__context *s, stbi_io_callbacks *c, void *user)
+{
+ s->io = *c;
+ s->io_user_data = user;
+ s->buflen = sizeof(s->buffer_start);
+ s->read_from_callbacks = 1;
+ s->img_buffer_original = s->buffer_start;
+ stbi__refill_buffer(s);
+ s->img_buffer_original_end = s->img_buffer_end;
+}
+
+#ifndef STBI_NO_STDIO
+
+static int stbi__stdio_read(void *user, char *data, int size)
+{
+ return (int) fread(data,1,size,(FILE*) user);
+}
+
+static void stbi__stdio_skip(void *user, int n)
+{
+ fseek((FILE*) user, n, SEEK_CUR);
+}
+
+static int stbi__stdio_eof(void *user)
+{
+ return feof((FILE*) user);
+}
+
+static stbi_io_callbacks stbi__stdio_callbacks =
+{
+ stbi__stdio_read,
+ stbi__stdio_skip,
+ stbi__stdio_eof,
+};
+
+static void stbi__start_file(stbi__context *s, FILE *f)
+{
+ stbi__start_callbacks(s, &stbi__stdio_callbacks, (void *) f);
+}
+
+//static void stop_file(stbi__context *s) { }
+
+#endif // !STBI_NO_STDIO
+
+static void stbi__rewind(stbi__context *s)
+{
+ // conceptually rewind SHOULD rewind to the beginning of the stream,
+ // but we just rewind to the beginning of the initial buffer, because
+ // we only use it after doing 'test', which only ever looks at at most 92 bytes
+ s->img_buffer = s->img_buffer_original;
+ s->img_buffer_end = s->img_buffer_original_end;
+}
+
+enum
+{
+ STBI_ORDER_RGB,
+ STBI_ORDER_BGR
+};
+
+typedef struct
+{
+ int bits_per_channel;
+ int num_channels;
+ int channel_order;
+} stbi__result_info;
+
+#ifndef STBI_NO_JPEG
+static int stbi__jpeg_test(stbi__context *s);
+static void *stbi__jpeg_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__jpeg_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_PNG
+static int stbi__png_test(stbi__context *s);
+static void *stbi__png_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__png_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_BMP
+static int stbi__bmp_test(stbi__context *s);
+static void *stbi__bmp_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__bmp_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_TGA
+static int stbi__tga_test(stbi__context *s);
+static void *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__tga_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_PSD
+static int stbi__psd_test(stbi__context *s);
+static void *stbi__psd_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri, int bpc);
+static int stbi__psd_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_HDR
+static int stbi__hdr_test(stbi__context *s);
+static float *stbi__hdr_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__hdr_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_PIC
+static int stbi__pic_test(stbi__context *s);
+static void *stbi__pic_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__pic_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_GIF
+static int stbi__gif_test(stbi__context *s);
+static void *stbi__gif_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__gif_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_PNM
+static int stbi__pnm_test(stbi__context *s);
+static void *stbi__pnm_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__pnm_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+// this is not threadsafe
+static const char *stbi__g_failure_reason;
+
+STBIDEF const char *stbi_failure_reason(void)
+{
+ return stbi__g_failure_reason;
+}
+
+static int stbi__err(const char *str)
+{
+ stbi__g_failure_reason = str;
+ return 0;
+}
+
+static void *stbi__malloc(size_t size)
+{
+ return STBI_MALLOC(size);
+}
+
+// stb_image uses ints pervasively, including for offset calculations.
+// therefore the largest decoded image size we can support with the
+// current code, even on 64-bit targets, is INT_MAX. this is not a
+// significant limitation for the intended use case.
+//
+// we do, however, need to make sure our size calculations don't
+// overflow. hence a few helper functions for size calculations that
+// multiply integers together, making sure that they're non-negative
+// and no overflow occurs.
+
+// return 1 if the sum is valid, 0 on overflow.
+// negative terms are considered invalid.
+static int stbi__addsizes_valid(int a, int b)
+{
+ if (b < 0) return 0;
+ // now 0 <= b <= INT_MAX, hence also
+ // 0 <= INT_MAX - b <= INTMAX.
+ // And "a + b <= INT_MAX" (which might overflow) is the
+ // same as a <= INT_MAX - b (no overflow)
+ return a <= INT_MAX - b;
+}
+
+// returns 1 if the product is valid, 0 on overflow.
+// negative factors are considered invalid.
+static int stbi__mul2sizes_valid(int a, int b)
+{
+ if (a < 0 || b < 0) return 0;
+ if (b == 0) return 1; // mul-by-0 is always safe
+ // portable way to check for no overflows in a*b
+ return a <= INT_MAX/b;
+}
+
+// returns 1 if "a*b + add" has no negative terms/factors and doesn't overflow
+static int stbi__mad2sizes_valid(int a, int b, int add)
+{
+ return stbi__mul2sizes_valid(a, b) && stbi__addsizes_valid(a*b, add);
+}
+
+// returns 1 if "a*b*c + add" has no negative terms/factors and doesn't overflow
+static int stbi__mad3sizes_valid(int a, int b, int c, int add)
+{
+ return stbi__mul2sizes_valid(a, b) && stbi__mul2sizes_valid(a*b, c) &&
+ stbi__addsizes_valid(a*b*c, add);
+}
+
+// returns 1 if "a*b*c*d + add" has no negative terms/factors and doesn't overflow
+static int stbi__mad4sizes_valid(int a, int b, int c, int d, int add)
+{
+ return stbi__mul2sizes_valid(a, b) && stbi__mul2sizes_valid(a*b, c) &&
+ stbi__mul2sizes_valid(a*b*c, d) && stbi__addsizes_valid(a*b*c*d, add);
+}
+
+// mallocs with size overflow checking
+static void *stbi__malloc_mad2(int a, int b, int add)
+{
+ if (!stbi__mad2sizes_valid(a, b, add)) return NULL;
+ return stbi__malloc(a*b + add);
+}
+
+static void *stbi__malloc_mad3(int a, int b, int c, int add)
+{
+ if (!stbi__mad3sizes_valid(a, b, c, add)) return NULL;
+ return stbi__malloc(a*b*c + add);
+}
+
+static void *stbi__malloc_mad4(int a, int b, int c, int d, int add)
+{
+ if (!stbi__mad4sizes_valid(a, b, c, d, add)) return NULL;
+ return stbi__malloc(a*b*c*d + add);
+}
+
+// stbi__err - error
+// stbi__errpf - error returning pointer to float
+// stbi__errpuc - error returning pointer to unsigned char
+
+#ifdef STBI_NO_FAILURE_STRINGS
+ #define stbi__err(x,y) 0
+#elif defined(STBI_FAILURE_USERMSG)
+ #define stbi__err(x,y) stbi__err(y)
+#else
+ #define stbi__err(x,y) stbi__err(x)
+#endif
+
+#define stbi__errpf(x,y) ((float *)(size_t) (stbi__err(x,y)?NULL:NULL))
+#define stbi__errpuc(x,y) ((unsigned char *)(size_t) (stbi__err(x,y)?NULL:NULL))
+
+STBIDEF void stbi_image_free(void *retval_from_stbi_load)
+{
+ STBI_FREE(retval_from_stbi_load);
+}
+
+#ifndef STBI_NO_LINEAR
+static float *stbi__ldr_to_hdr(stbi_uc *data, int x, int y, int comp);
+#endif
+
+#ifndef STBI_NO_HDR
+static stbi_uc *stbi__hdr_to_ldr(float *data, int x, int y, int comp);
+#endif
+
+static int stbi__vertically_flip_on_load = 0;
+
+STBIDEF void stbi_set_flip_vertically_on_load(int flag_true_if_should_flip)
+{
+ stbi__vertically_flip_on_load = flag_true_if_should_flip;
+}
+
+static void *stbi__load_main(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri, int bpc)
+{
+ memset(ri, 0, sizeof(*ri)); // make sure it's initialized if we add new fields
+ ri->bits_per_channel = 8; // default is 8 so most paths don't have to be changed
+ ri->channel_order = STBI_ORDER_RGB; // all current input & output are this, but this is here so we can add BGR order
+ ri->num_channels = 0;
+
+ #ifndef STBI_NO_JPEG
+ if (stbi__jpeg_test(s)) return stbi__jpeg_load(s,x,y,comp,req_comp, ri);
+ #endif
+ #ifndef STBI_NO_PNG
+ if (stbi__png_test(s)) return stbi__png_load(s,x,y,comp,req_comp, ri);
+ #endif
+ #ifndef STBI_NO_BMP
+ if (stbi__bmp_test(s)) return stbi__bmp_load(s,x,y,comp,req_comp, ri);
+ #endif
+ #ifndef STBI_NO_GIF
+ if (stbi__gif_test(s)) return stbi__gif_load(s,x,y,comp,req_comp, ri);
+ #endif
+ #ifndef STBI_NO_PSD
+ if (stbi__psd_test(s)) return stbi__psd_load(s,x,y,comp,req_comp, ri, bpc);
+ #endif
+ #ifndef STBI_NO_PIC
+ if (stbi__pic_test(s)) return stbi__pic_load(s,x,y,comp,req_comp, ri);
+ #endif
+ #ifndef STBI_NO_PNM
+ if (stbi__pnm_test(s)) return stbi__pnm_load(s,x,y,comp,req_comp, ri);
+ #endif
+
+ #ifndef STBI_NO_HDR
+ if (stbi__hdr_test(s)) {
+ float *hdr = stbi__hdr_load(s, x,y,comp,req_comp, ri);
+ return stbi__hdr_to_ldr(hdr, *x, *y, req_comp ? req_comp : *comp);
+ }
+ #endif
+
+ #ifndef STBI_NO_TGA
+ // test tga last because it's a crappy test!
+ if (stbi__tga_test(s))
+ return stbi__tga_load(s,x,y,comp,req_comp, ri);
+ #endif
+
+ return stbi__errpuc("unknown image type", "Image not of any known type, or corrupt");
+}
+
+static stbi_uc *stbi__convert_16_to_8(stbi__uint16 *orig, int w, int h, int channels)
+{
+ int i;
+ int img_len = w * h * channels;
+ stbi_uc *reduced;
+
+ reduced = (stbi_uc *) stbi__malloc(img_len);
+ if (reduced == NULL) return stbi__errpuc("outofmem", "Out of memory");
+
+ for (i = 0; i < img_len; ++i)
+ reduced[i] = (stbi_uc)((orig[i] >> 8) & 0xFF); // top half of each byte is sufficient approx of 16->8 bit scaling
+
+ STBI_FREE(orig);
+ return reduced;
+}
+
+static stbi__uint16 *stbi__convert_8_to_16(stbi_uc *orig, int w, int h, int channels)
+{
+ int i;
+ int img_len = w * h * channels;
+ stbi__uint16 *enlarged;
+
+ enlarged = (stbi__uint16 *) stbi__malloc(img_len*2);
+ if (enlarged == NULL) return (stbi__uint16 *) stbi__errpuc("outofmem", "Out of memory");
+
+ for (i = 0; i < img_len; ++i)
+ enlarged[i] = (stbi__uint16)((orig[i] << 8) + orig[i]); // replicate to high and low byte, maps 0->0, 255->0xffff
+
+ STBI_FREE(orig);
+ return enlarged;
+}
+
+static unsigned char *stbi__load_and_postprocess_8bit(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__result_info ri;
+ void *result = stbi__load_main(s, x, y, comp, req_comp, &ri, 8);
+
+ if (result == NULL)
+ return NULL;
+
+ if (ri.bits_per_channel != 8) {
+ STBI_ASSERT(ri.bits_per_channel == 16);
+ result = stbi__convert_16_to_8((stbi__uint16 *) result, *x, *y, req_comp == 0 ? *comp : req_comp);
+ ri.bits_per_channel = 8;
+ }
+
+ // @TODO: move stbi__convert_format to here
+
+ if (stbi__vertically_flip_on_load) {
+ int w = *x, h = *y;
+ int channels = req_comp ? req_comp : *comp;
+ int row,col,z;
+ stbi_uc *image = (stbi_uc *) result;
+
+ // @OPTIMIZE: use a bigger temp buffer and memcpy multiple pixels at once
+ for (row = 0; row < (h>>1); row++) {
+ for (col = 0; col < w; col++) {
+ for (z = 0; z < channels; z++) {
+ stbi_uc temp = image[(row * w + col) * channels + z];
+ image[(row * w + col) * channels + z] = image[((h - row - 1) * w + col) * channels + z];
+ image[((h - row - 1) * w + col) * channels + z] = temp;
+ }
+ }
+ }
+ }
+
+ return (unsigned char *) result;
+}
+
+static stbi__uint16 *stbi__load_and_postprocess_16bit(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__result_info ri;
+ void *result = stbi__load_main(s, x, y, comp, req_comp, &ri, 16);
+
+ if (result == NULL)
+ return NULL;
+
+ if (ri.bits_per_channel != 16) {
+ STBI_ASSERT(ri.bits_per_channel == 8);
+ result = stbi__convert_8_to_16((stbi_uc *) result, *x, *y, req_comp == 0 ? *comp : req_comp);
+ ri.bits_per_channel = 16;
+ }
+
+ // @TODO: move stbi__convert_format16 to here
+ // @TODO: special case RGB-to-Y (and RGBA-to-YA) for 8-bit-to-16-bit case to keep more precision
+
+ if (stbi__vertically_flip_on_load) {
+ int w = *x, h = *y;
+ int channels = req_comp ? req_comp : *comp;
+ int row,col,z;
+ stbi__uint16 *image = (stbi__uint16 *) result;
+
+ // @OPTIMIZE: use a bigger temp buffer and memcpy multiple pixels at once
+ for (row = 0; row < (h>>1); row++) {
+ for (col = 0; col < w; col++) {
+ for (z = 0; z < channels; z++) {
+ stbi__uint16 temp = image[(row * w + col) * channels + z];
+ image[(row * w + col) * channels + z] = image[((h - row - 1) * w + col) * channels + z];
+ image[((h - row - 1) * w + col) * channels + z] = temp;
+ }
+ }
+ }
+ }
+
+ return (stbi__uint16 *) result;
+}
+
+#ifndef STBI_NO_HDR
+static void stbi__float_postprocess(float *result, int *x, int *y, int *comp, int req_comp)
+{
+ if (stbi__vertically_flip_on_load && result != NULL) {
+ int w = *x, h = *y;
+ int depth = req_comp ? req_comp : *comp;
+ int row,col,z;
+ float temp;
+
+ // @OPTIMIZE: use a bigger temp buffer and memcpy multiple pixels at once
+ for (row = 0; row < (h>>1); row++) {
+ for (col = 0; col < w; col++) {
+ for (z = 0; z < depth; z++) {
+ temp = result[(row * w + col) * depth + z];
+ result[(row * w + col) * depth + z] = result[((h - row - 1) * w + col) * depth + z];
+ result[((h - row - 1) * w + col) * depth + z] = temp;
+ }
+ }
+ }
+ }
+}
+#endif
+
+#ifndef STBI_NO_STDIO
+
+static FILE *stbi__fopen(char const *filename, char const *mode)
+{
+ FILE *f;
+#if defined(_MSC_VER) && _MSC_VER >= 1400
+ if (0 != fopen_s(&f, filename, mode))
+ f=0;
+#else
+ f = fopen(filename, mode);
+#endif
+ return f;
+}
+
+
+STBIDEF stbi_uc *stbi_load(char const *filename, int *x, int *y, int *comp, int req_comp)
+{
+ FILE *f = stbi__fopen(filename, "rb");
+ unsigned char *result;
+ if (!f) return stbi__errpuc("can't fopen", "Unable to open file");
+ result = stbi_load_from_file(f,x,y,comp,req_comp);
+ fclose(f);
+ return result;
+}
+
+STBIDEF stbi_uc *stbi_load_from_file(FILE *f, int *x, int *y, int *comp, int req_comp)
+{
+ unsigned char *result;
+ stbi__context s;
+ stbi__start_file(&s,f);
+ result = stbi__load_and_postprocess_8bit(&s,x,y,comp,req_comp);
+ if (result) {
+ // need to 'unget' all the characters in the IO buffer
+ fseek(f, - (int) (s.img_buffer_end - s.img_buffer), SEEK_CUR);
+ }
+ return result;
+}
+
+STBIDEF stbi__uint16 *stbi_load_from_file_16(FILE *f, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__uint16 *result;
+ stbi__context s;
+ stbi__start_file(&s,f);
+ result = stbi__load_and_postprocess_16bit(&s,x,y,comp,req_comp);
+ if (result) {
+ // need to 'unget' all the characters in the IO buffer
+ fseek(f, - (int) (s.img_buffer_end - s.img_buffer), SEEK_CUR);
+ }
+ return result;
+}
+
+STBIDEF stbi_us *stbi_load_16(char const *filename, int *x, int *y, int *comp, int req_comp)
+{
+ FILE *f = stbi__fopen(filename, "rb");
+ stbi__uint16 *result;
+ if (!f) return (stbi_us *) stbi__errpuc("can't fopen", "Unable to open file");
+ result = stbi_load_from_file_16(f,x,y,comp,req_comp);
+ fclose(f);
+ return result;
+}
+
+
+#endif //!STBI_NO_STDIO
+
+STBIDEF stbi_uc *stbi_load_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__context s;
+ stbi__start_mem(&s,buffer,len);
+ return stbi__load_and_postprocess_8bit(&s,x,y,comp,req_comp);
+}
+
+STBIDEF stbi_uc *stbi_load_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__context s;
+ stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user);
+ return stbi__load_and_postprocess_8bit(&s,x,y,comp,req_comp);
+}
+
+#ifndef STBI_NO_LINEAR
+static float *stbi__loadf_main(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+{
+ unsigned char *data;
+ #ifndef STBI_NO_HDR
+ if (stbi__hdr_test(s)) {
+ stbi__result_info ri;
+ float *hdr_data = stbi__hdr_load(s,x,y,comp,req_comp, &ri);
+ if (hdr_data)
+ stbi__float_postprocess(hdr_data,x,y,comp,req_comp);
+ return hdr_data;
+ }
+ #endif
+ data = stbi__load_and_postprocess_8bit(s, x, y, comp, req_comp);
+ if (data)
+ return stbi__ldr_to_hdr(data, *x, *y, req_comp ? req_comp : *comp);
+ return stbi__errpf("unknown image type", "Image not of any known type, or corrupt");
+}
+
+STBIDEF float *stbi_loadf_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__context s;
+ stbi__start_mem(&s,buffer,len);
+ return stbi__loadf_main(&s,x,y,comp,req_comp);
+}
+
+STBIDEF float *stbi_loadf_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__context s;
+ stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user);
+ return stbi__loadf_main(&s,x,y,comp,req_comp);
+}
+
+#ifndef STBI_NO_STDIO
+STBIDEF float *stbi_loadf(char const *filename, int *x, int *y, int *comp, int req_comp)
+{
+ float *result;
+ FILE *f = stbi__fopen(filename, "rb");
+ if (!f) return stbi__errpf("can't fopen", "Unable to open file");
+ result = stbi_loadf_from_file(f,x,y,comp,req_comp);
+ fclose(f);
+ return result;
+}
+
+STBIDEF float *stbi_loadf_from_file(FILE *f, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__context s;
+ stbi__start_file(&s,f);
+ return stbi__loadf_main(&s,x,y,comp,req_comp);
+}
+#endif // !STBI_NO_STDIO
+
+#endif // !STBI_NO_LINEAR
+
+// these is-hdr-or-not is defined independent of whether STBI_NO_LINEAR is
+// defined, for API simplicity; if STBI_NO_LINEAR is defined, it always
+// reports false!
+
+STBIDEF int stbi_is_hdr_from_memory(stbi_uc const *buffer, int len)
+{
+ #ifndef STBI_NO_HDR
+ stbi__context s;
+ stbi__start_mem(&s,buffer,len);
+ return stbi__hdr_test(&s);
+ #else
+ STBI_NOTUSED(buffer);
+ STBI_NOTUSED(len);
+ return 0;
+ #endif
+}
+
+#ifndef STBI_NO_STDIO
+STBIDEF int stbi_is_hdr (char const *filename)
+{
+ FILE *f = stbi__fopen(filename, "rb");
+ int result=0;
+ if (f) {
+ result = stbi_is_hdr_from_file(f);
+ fclose(f);
+ }
+ return result;
+}
+
+STBIDEF int stbi_is_hdr_from_file(FILE *f)
+{
+ #ifndef STBI_NO_HDR
+ stbi__context s;
+ stbi__start_file(&s,f);
+ return stbi__hdr_test(&s);
+ #else
+ STBI_NOTUSED(f);
+ return 0;
+ #endif
+}
+#endif // !STBI_NO_STDIO
+
+STBIDEF int stbi_is_hdr_from_callbacks(stbi_io_callbacks const *clbk, void *user)
+{
+ #ifndef STBI_NO_HDR
+ stbi__context s;
+ stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user);
+ return stbi__hdr_test(&s);
+ #else
+ STBI_NOTUSED(clbk);
+ STBI_NOTUSED(user);
+ return 0;
+ #endif
+}
+
+#ifndef STBI_NO_LINEAR
+static float stbi__l2h_gamma=2.2f, stbi__l2h_scale=1.0f;
+
+STBIDEF void stbi_ldr_to_hdr_gamma(float gamma) { stbi__l2h_gamma = gamma; }
+STBIDEF void stbi_ldr_to_hdr_scale(float scale) { stbi__l2h_scale = scale; }
+#endif
+
+static float stbi__h2l_gamma_i=1.0f/2.2f, stbi__h2l_scale_i=1.0f;
+
+STBIDEF void stbi_hdr_to_ldr_gamma(float gamma) { stbi__h2l_gamma_i = 1/gamma; }
+STBIDEF void stbi_hdr_to_ldr_scale(float scale) { stbi__h2l_scale_i = 1/scale; }
+
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// Common code used by all image loaders
+//
+
+enum
+{
+ STBI__SCAN_load=0,
+ STBI__SCAN_type,
+ STBI__SCAN_header
+};
+
+static void stbi__refill_buffer(stbi__context *s)
+{
+ int n = (s->io.read)(s->io_user_data,(char*)s->buffer_start,s->buflen);
+ if (n == 0) {
+ // at end of file, treat same as if from memory, but need to handle case
+ // where s->img_buffer isn't pointing to safe memory, e.g. 0-byte file
+ s->read_from_callbacks = 0;
+ s->img_buffer = s->buffer_start;
+ s->img_buffer_end = s->buffer_start+1;
+ *s->img_buffer = 0;
+ } else {
+ s->img_buffer = s->buffer_start;
+ s->img_buffer_end = s->buffer_start + n;
+ }
+}
+
+stbi_inline static stbi_uc stbi__get8(stbi__context *s)
+{
+ if (s->img_buffer < s->img_buffer_end)
+ return *s->img_buffer++;
+ if (s->read_from_callbacks) {
+ stbi__refill_buffer(s);
+ return *s->img_buffer++;
+ }
+ return 0;
+}
+
+stbi_inline static int stbi__at_eof(stbi__context *s)
+{
+ if (s->io.read) {
+ if (!(s->io.eof)(s->io_user_data)) return 0;
+ // if feof() is true, check if buffer = end
+ // special case: we've only got the special 0 character at the end
+ if (s->read_from_callbacks == 0) return 1;
+ }
+
+ return s->img_buffer >= s->img_buffer_end;
+}
+
+static void stbi__skip(stbi__context *s, int n)
+{
+ if (n < 0) {
+ s->img_buffer = s->img_buffer_end;
+ return;
+ }
+ if (s->io.read) {
+ int blen = (int) (s->img_buffer_end - s->img_buffer);
+ if (blen < n) {
+ s->img_buffer = s->img_buffer_end;
+ (s->io.skip)(s->io_user_data, n - blen);
+ return;
+ }
+ }
+ s->img_buffer += n;
+}
+
+static int stbi__getn(stbi__context *s, stbi_uc *buffer, int n)
+{
+ if (s->io.read) {
+ int blen = (int) (s->img_buffer_end - s->img_buffer);
+ if (blen < n) {
+ int res, count;
+
+ memcpy(buffer, s->img_buffer, blen);
+
+ count = (s->io.read)(s->io_user_data, (char*) buffer + blen, n - blen);
+ res = (count == (n-blen));
+ s->img_buffer = s->img_buffer_end;
+ return res;
+ }
+ }
+
+ if (s->img_buffer+n <= s->img_buffer_end) {
+ memcpy(buffer, s->img_buffer, n);
+ s->img_buffer += n;
+ return 1;
+ } else
+ return 0;
+}
+
+static int stbi__get16be(stbi__context *s)
+{
+ int z = stbi__get8(s);
+ return (z << 8) + stbi__get8(s);
+}
+
+static stbi__uint32 stbi__get32be(stbi__context *s)
+{
+ stbi__uint32 z = stbi__get16be(s);
+ return (z << 16) + stbi__get16be(s);
+}
+
+#if defined(STBI_NO_BMP) && defined(STBI_NO_TGA) && defined(STBI_NO_GIF)
+// nothing
+#else
+static int stbi__get16le(stbi__context *s)
+{
+ int z = stbi__get8(s);
+ return z + (stbi__get8(s) << 8);
+}
+#endif
+
+#ifndef STBI_NO_BMP
+static stbi__uint32 stbi__get32le(stbi__context *s)
+{
+ stbi__uint32 z = stbi__get16le(s);
+ return z + (stbi__get16le(s) << 16);
+}
+#endif
+
+#define STBI__BYTECAST(x) ((stbi_uc) ((x) & 255)) // truncate int to byte without warnings
+
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// generic converter from built-in img_n to req_comp
+// individual types do this automatically as much as possible (e.g. jpeg
+// does all cases internally since it needs to colorspace convert anyway,
+// and it never has alpha, so very few cases ). png can automatically
+// interleave an alpha=255 channel, but falls back to this for other cases
+//
+// assume data buffer is malloced, so malloc a new one and free that one
+// only failure mode is malloc failing
+
+static stbi_uc stbi__compute_y(int r, int g, int b)
+{
+ return (stbi_uc) (((r*77) + (g*150) + (29*b)) >> 8);
+}
+
+static unsigned char *stbi__convert_format(unsigned char *data, int img_n, int req_comp, unsigned int x, unsigned int y)
+{
+ int i,j;
+ unsigned char *good;
+
+ if (req_comp == img_n) return data;
+ STBI_ASSERT(req_comp >= 1 && req_comp <= 4);
+
+ good = (unsigned char *) stbi__malloc_mad3(req_comp, x, y, 0);
+ if (good == NULL) {
+ STBI_FREE(data);
+ return stbi__errpuc("outofmem", "Out of memory");
+ }
+
+ for (j=0; j < (int) y; ++j) {
+ unsigned char *src = data + j * x * img_n ;
+ unsigned char *dest = good + j * x * req_comp;
+
+ #define STBI__COMBO(a,b) ((a)*8+(b))
+ #define STBI__CASE(a,b) case STBI__COMBO(a,b): for(i=x-1; i >= 0; --i, src += a, dest += b)
+ // convert source image with img_n components to one with req_comp components;
+ // avoid switch per pixel, so use switch per scanline and massive macros
+ switch (STBI__COMBO(img_n, req_comp)) {
+ STBI__CASE(1,2) { dest[0]=src[0], dest[1]=255; } break;
+ STBI__CASE(1,3) { dest[0]=dest[1]=dest[2]=src[0]; } break;
+ STBI__CASE(1,4) { dest[0]=dest[1]=dest[2]=src[0], dest[3]=255; } break;
+ STBI__CASE(2,1) { dest[0]=src[0]; } break;
+ STBI__CASE(2,3) { dest[0]=dest[1]=dest[2]=src[0]; } break;
+ STBI__CASE(2,4) { dest[0]=dest[1]=dest[2]=src[0], dest[3]=src[1]; } break;
+ STBI__CASE(3,4) { dest[0]=src[0],dest[1]=src[1],dest[2]=src[2],dest[3]=255; } break;
+ STBI__CASE(3,1) { dest[0]=stbi__compute_y(src[0],src[1],src[2]); } break;
+ STBI__CASE(3,2) { dest[0]=stbi__compute_y(src[0],src[1],src[2]), dest[1] = 255; } break;
+ STBI__CASE(4,1) { dest[0]=stbi__compute_y(src[0],src[1],src[2]); } break;
+ STBI__CASE(4,2) { dest[0]=stbi__compute_y(src[0],src[1],src[2]), dest[1] = src[3]; } break;
+ STBI__CASE(4,3) { dest[0]=src[0],dest[1]=src[1],dest[2]=src[2]; } break;
+ default: STBI_ASSERT(0);
+ }
+ #undef STBI__CASE
+ }
+
+ STBI_FREE(data);
+ return good;
+}
+
+static stbi__uint16 stbi__compute_y_16(int r, int g, int b)
+{
+ return (stbi__uint16) (((r*77) + (g*150) + (29*b)) >> 8);
+}
+
+static stbi__uint16 *stbi__convert_format16(stbi__uint16 *data, int img_n, int req_comp, unsigned int x, unsigned int y)
+{
+ int i,j;
+ stbi__uint16 *good;
+
+ if (req_comp == img_n) return data;
+ STBI_ASSERT(req_comp >= 1 && req_comp <= 4);
+
+ good = (stbi__uint16 *) stbi__malloc(req_comp * x * y * 2);
+ if (good == NULL) {
+ STBI_FREE(data);
+ return (stbi__uint16 *) stbi__errpuc("outofmem", "Out of memory");
+ }
+
+ for (j=0; j < (int) y; ++j) {
+ stbi__uint16 *src = data + j * x * img_n ;
+ stbi__uint16 *dest = good + j * x * req_comp;
+
+ #define STBI__COMBO(a,b) ((a)*8+(b))
+ #define STBI__CASE(a,b) case STBI__COMBO(a,b): for(i=x-1; i >= 0; --i, src += a, dest += b)
+ // convert source image with img_n components to one with req_comp components;
+ // avoid switch per pixel, so use switch per scanline and massive macros
+ switch (STBI__COMBO(img_n, req_comp)) {
+ STBI__CASE(1,2) { dest[0]=src[0], dest[1]=0xffff; } break;
+ STBI__CASE(1,3) { dest[0]=dest[1]=dest[2]=src[0]; } break;
+ STBI__CASE(1,4) { dest[0]=dest[1]=dest[2]=src[0], dest[3]=0xffff; } break;
+ STBI__CASE(2,1) { dest[0]=src[0]; } break;
+ STBI__CASE(2,3) { dest[0]=dest[1]=dest[2]=src[0]; } break;
+ STBI__CASE(2,4) { dest[0]=dest[1]=dest[2]=src[0], dest[3]=src[1]; } break;
+ STBI__CASE(3,4) { dest[0]=src[0],dest[1]=src[1],dest[2]=src[2],dest[3]=0xffff; } break;
+ STBI__CASE(3,1) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]); } break;
+ STBI__CASE(3,2) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]), dest[1] = 0xffff; } break;
+ STBI__CASE(4,1) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]); } break;
+ STBI__CASE(4,2) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]), dest[1] = src[3]; } break;
+ STBI__CASE(4,3) { dest[0]=src[0],dest[1]=src[1],dest[2]=src[2]; } break;
+ default: STBI_ASSERT(0);
+ }
+ #undef STBI__CASE
+ }
+
+ STBI_FREE(data);
+ return good;
+}
+
+#ifndef STBI_NO_LINEAR
+static float *stbi__ldr_to_hdr(stbi_uc *data, int x, int y, int comp)
+{
+ int i,k,n;
+ float *output;
+ if (!data) return NULL;
+ output = (float *) stbi__malloc_mad4(x, y, comp, sizeof(float), 0);
+ if (output == NULL) { STBI_FREE(data); return stbi__errpf("outofmem", "Out of memory"); }
+ // compute number of non-alpha components
+ if (comp & 1) n = comp; else n = comp-1;
+ for (i=0; i < x*y; ++i) {
+ for (k=0; k < n; ++k) {
+ output[i*comp + k] = (float) (pow(data[i*comp+k]/255.0f, stbi__l2h_gamma) * stbi__l2h_scale);
+ }
+ if (k < comp) output[i*comp + k] = data[i*comp+k]/255.0f;
+ }
+ STBI_FREE(data);
+ return output;
+}
+#endif
+
+#ifndef STBI_NO_HDR
+#define stbi__float2int(x) ((int) (x))
+static stbi_uc *stbi__hdr_to_ldr(float *data, int x, int y, int comp)
+{
+ int i,k,n;
+ stbi_uc *output;
+ if (!data) return NULL;
+ output = (stbi_uc *) stbi__malloc_mad3(x, y, comp, 0);
+ if (output == NULL) { STBI_FREE(data); return stbi__errpuc("outofmem", "Out of memory"); }
+ // compute number of non-alpha components
+ if (comp & 1) n = comp; else n = comp-1;
+ for (i=0; i < x*y; ++i) {
+ for (k=0; k < n; ++k) {
+ float z = (float) pow(data[i*comp+k]*stbi__h2l_scale_i, stbi__h2l_gamma_i) * 255 + 0.5f;
+ if (z < 0) z = 0;
+ if (z > 255) z = 255;
+ output[i*comp + k] = (stbi_uc) stbi__float2int(z);
+ }
+ if (k < comp) {
+ float z = data[i*comp+k] * 255 + 0.5f;
+ if (z < 0) z = 0;
+ if (z > 255) z = 255;
+ output[i*comp + k] = (stbi_uc) stbi__float2int(z);
+ }
+ }
+ STBI_FREE(data);
+ return output;
+}
+#endif
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// "baseline" JPEG/JFIF decoder
+//
+// simple implementation
+// - doesn't support delayed output of y-dimension
+// - simple interface (only one output format: 8-bit interleaved RGB)
+// - doesn't try to recover corrupt jpegs
+// - doesn't allow partial loading, loading multiple at once
+// - still fast on x86 (copying globals into locals doesn't help x86)
+// - allocates lots of intermediate memory (full size of all components)
+// - non-interleaved case requires this anyway
+// - allows good upsampling (see next)
+// high-quality
+// - upsampled channels are bilinearly interpolated, even across blocks
+// - quality integer IDCT derived from IJG's 'slow'
+// performance
+// - fast huffman; reasonable integer IDCT
+// - some SIMD kernels for common paths on targets with SSE2/NEON
+// - uses a lot of intermediate memory, could cache poorly
+
+#ifndef STBI_NO_JPEG
+
+// huffman decoding acceleration
+#define FAST_BITS 9 // larger handles more cases; smaller stomps less cache
+
+typedef struct
+{
+ stbi_uc fast[1 << FAST_BITS];
+ // weirdly, repacking this into AoS is a 10% speed loss, instead of a win
+ stbi__uint16 code[256];
+ stbi_uc values[256];
+ stbi_uc size[257];
+ unsigned int maxcode[18];
+ int delta[17]; // old 'firstsymbol' - old 'firstcode'
+} stbi__huffman;
+
+typedef struct
+{
+ stbi__context *s;
+ stbi__huffman huff_dc[4];
+ stbi__huffman huff_ac[4];
+ stbi__uint16 dequant[4][64];
+ stbi__int16 fast_ac[4][1 << FAST_BITS];
+
+// sizes for components, interleaved MCUs
+ int img_h_max, img_v_max;
+ int img_mcu_x, img_mcu_y;
+ int img_mcu_w, img_mcu_h;
+
+// definition of jpeg image component
+ struct
+ {
+ int id;
+ int h,v;
+ int tq;
+ int hd,ha;
+ int dc_pred;
+
+ int x,y,w2,h2;
+ stbi_uc *data;
+ void *raw_data, *raw_coeff;
+ stbi_uc *linebuf;
+ short *coeff; // progressive only
+ int coeff_w, coeff_h; // number of 8x8 coefficient blocks
+ } img_comp[4];
+
+ stbi__uint32 code_buffer; // jpeg entropy-coded buffer
+ int code_bits; // number of valid bits
+ unsigned char marker; // marker seen while filling entropy buffer
+ int nomore; // flag if we saw a marker so must stop
+
+ int progressive;
+ int spec_start;
+ int spec_end;
+ int succ_high;
+ int succ_low;
+ int eob_run;
+ int jfif;
+ int app14_color_transform; // Adobe APP14 tag
+ int rgb;
+
+ int scan_n, order[4];
+ int restart_interval, todo;
+
+// kernels
+ void (*idct_block_kernel)(stbi_uc *out, int out_stride, short data[64]);
+ void (*YCbCr_to_RGB_kernel)(stbi_uc *out, const stbi_uc *y, const stbi_uc *pcb, const stbi_uc *pcr, int count, int step);
+ stbi_uc *(*resample_row_hv_2_kernel)(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs);
+} stbi__jpeg;
+
+static int stbi__build_huffman(stbi__huffman *h, int *count)
+{
+ int i,j,k=0,code;
+ // build size list for each symbol (from JPEG spec)
+ for (i=0; i < 16; ++i)
+ for (j=0; j < count[i]; ++j)
+ h->size[k++] = (stbi_uc) (i+1);
+ h->size[k] = 0;
+
+ // compute actual symbols (from jpeg spec)
+ code = 0;
+ k = 0;
+ for(j=1; j <= 16; ++j) {
+ // compute delta to add to code to compute symbol id
+ h->delta[j] = k - code;
+ if (h->size[k] == j) {
+ while (h->size[k] == j)
+ h->code[k++] = (stbi__uint16) (code++);
+ if (code-1 >= (1 << j)) return stbi__err("bad code lengths","Corrupt JPEG");
+ }
+ // compute largest code + 1 for this size, preshifted as needed later
+ h->maxcode[j] = code << (16-j);
+ code <<= 1;
+ }
+ h->maxcode[j] = 0xffffffff;
+
+ // build non-spec acceleration table; 255 is flag for not-accelerated
+ memset(h->fast, 255, 1 << FAST_BITS);
+ for (i=0; i < k; ++i) {
+ int s = h->size[i];
+ if (s <= FAST_BITS) {
+ int c = h->code[i] << (FAST_BITS-s);
+ int m = 1 << (FAST_BITS-s);
+ for (j=0; j < m; ++j) {
+ h->fast[c+j] = (stbi_uc) i;
+ }
+ }
+ }
+ return 1;
+}
+
+// build a table that decodes both magnitude and value of small ACs in
+// one go.
+static void stbi__build_fast_ac(stbi__int16 *fast_ac, stbi__huffman *h)
+{
+ int i;
+ for (i=0; i < (1 << FAST_BITS); ++i) {
+ stbi_uc fast = h->fast[i];
+ fast_ac[i] = 0;
+ if (fast < 255) {
+ int rs = h->values[fast];
+ int run = (rs >> 4) & 15;
+ int magbits = rs & 15;
+ int len = h->size[fast];
+
+ if (magbits && len + magbits <= FAST_BITS) {
+ // magnitude code followed by receive_extend code
+ int k = ((i << len) & ((1 << FAST_BITS) - 1)) >> (FAST_BITS - magbits);
+ int m = 1 << (magbits - 1);
+ if (k < m) k += (~0U << magbits) + 1;
+ // if the result is small enough, we can fit it in fast_ac table
+ if (k >= -128 && k <= 127)
+ fast_ac[i] = (stbi__int16) ((k << 8) + (run << 4) + (len + magbits));
+ }
+ }
+ }
+}
+
+static void stbi__grow_buffer_unsafe(stbi__jpeg *j)
+{
+ do {
+ int b = j->nomore ? 0 : stbi__get8(j->s);
+ if (b == 0xff) {
+ int c = stbi__get8(j->s);
+ while (c == 0xff) c = stbi__get8(j->s); // consume fill bytes
+ if (c != 0) {
+ j->marker = (unsigned char) c;
+ j->nomore = 1;
+ return;
+ }
+ }
+ j->code_buffer |= b << (24 - j->code_bits);
+ j->code_bits += 8;
+ } while (j->code_bits <= 24);
+}
+
+// (1 << n) - 1
+static stbi__uint32 stbi__bmask[17]={0,1,3,7,15,31,63,127,255,511,1023,2047,4095,8191,16383,32767,65535};
+
+// decode a jpeg huffman value from the bitstream
+stbi_inline static int stbi__jpeg_huff_decode(stbi__jpeg *j, stbi__huffman *h)
+{
+ unsigned int temp;
+ int c,k;
+
+ if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
+
+ // look at the top FAST_BITS and determine what symbol ID it is,
+ // if the code is <= FAST_BITS
+ c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1);
+ k = h->fast[c];
+ if (k < 255) {
+ int s = h->size[k];
+ if (s > j->code_bits)
+ return -1;
+ j->code_buffer <<= s;
+ j->code_bits -= s;
+ return h->values[k];
+ }
+
+ // naive test is to shift the code_buffer down so k bits are
+ // valid, then test against maxcode. To speed this up, we've
+ // preshifted maxcode left so that it has (16-k) 0s at the
+ // end; in other words, regardless of the number of bits, it
+ // wants to be compared against something shifted to have 16;
+ // that way we don't need to shift inside the loop.
+ temp = j->code_buffer >> 16;
+ for (k=FAST_BITS+1 ; ; ++k)
+ if (temp < h->maxcode[k])
+ break;
+ if (k == 17) {
+ // error! code not found
+ j->code_bits -= 16;
+ return -1;
+ }
+
+ if (k > j->code_bits)
+ return -1;
+
+ // convert the huffman code to the symbol id
+ c = ((j->code_buffer >> (32 - k)) & stbi__bmask[k]) + h->delta[k];
+ STBI_ASSERT((((j->code_buffer) >> (32 - h->size[c])) & stbi__bmask[h->size[c]]) == h->code[c]);
+
+ // convert the id to a symbol
+ j->code_bits -= k;
+ j->code_buffer <<= k;
+ return h->values[c];
+}
+
+// bias[n] = (-1<<n) + 1
+static int const stbi__jbias[16] = {0,-1,-3,-7,-15,-31,-63,-127,-255,-511,-1023,-2047,-4095,-8191,-16383,-32767};
+
+// combined JPEG 'receive' and JPEG 'extend', since baseline
+// always extends everything it receives.
+stbi_inline static int stbi__extend_receive(stbi__jpeg *j, int n)
+{
+ unsigned int k;
+ int sgn;
+ if (j->code_bits < n) stbi__grow_buffer_unsafe(j);
+
+ sgn = (stbi__int32)j->code_buffer >> 31; // sign bit is always in MSB
+ k = stbi_lrot(j->code_buffer, n);
+ STBI_ASSERT(n >= 0 && n < (int) (sizeof(stbi__bmask)/sizeof(*stbi__bmask)));
+ j->code_buffer = k & ~stbi__bmask[n];
+ k &= stbi__bmask[n];
+ j->code_bits -= n;
+ return k + (stbi__jbias[n] & ~sgn);
+}
+
+// get some unsigned bits
+stbi_inline static int stbi__jpeg_get_bits(stbi__jpeg *j, int n)
+{
+ unsigned int k;
+ if (j->code_bits < n) stbi__grow_buffer_unsafe(j);
+ k = stbi_lrot(j->code_buffer, n);
+ j->code_buffer = k & ~stbi__bmask[n];
+ k &= stbi__bmask[n];
+ j->code_bits -= n;
+ return k;
+}
+
+stbi_inline static int stbi__jpeg_get_bit(stbi__jpeg *j)
+{
+ unsigned int k;
+ if (j->code_bits < 1) stbi__grow_buffer_unsafe(j);
+ k = j->code_buffer;
+ j->code_buffer <<= 1;
+ --j->code_bits;
+ return k & 0x80000000;
+}
+
+// given a value that's at position X in the zigzag stream,
+// where does it appear in the 8x8 matrix coded as row-major?
+static stbi_uc stbi__jpeg_dezigzag[64+15] =
+{
+ 0, 1, 8, 16, 9, 2, 3, 10,
+ 17, 24, 32, 25, 18, 11, 4, 5,
+ 12, 19, 26, 33, 40, 48, 41, 34,
+ 27, 20, 13, 6, 7, 14, 21, 28,
+ 35, 42, 49, 56, 57, 50, 43, 36,
+ 29, 22, 15, 23, 30, 37, 44, 51,
+ 58, 59, 52, 45, 38, 31, 39, 46,
+ 53, 60, 61, 54, 47, 55, 62, 63,
+ // let corrupt input sample past end
+ 63, 63, 63, 63, 63, 63, 63, 63,
+ 63, 63, 63, 63, 63, 63, 63
+};
+
+// decode one 64-entry block--
+static int stbi__jpeg_decode_block(stbi__jpeg *j, short data[64], stbi__huffman *hdc, stbi__huffman *hac, stbi__int16 *fac, int b, stbi__uint16 *dequant)
+{
+ int diff,dc,k;
+ int t;
+
+ if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
+ t = stbi__jpeg_huff_decode(j, hdc);
+ if (t < 0) return stbi__err("bad huffman code","Corrupt JPEG");
+
+ // 0 all the ac values now so we can do it 32-bits at a time
+ memset(data,0,64*sizeof(data[0]));
+
+ diff = t ? stbi__extend_receive(j, t) : 0;
+ dc = j->img_comp[b].dc_pred + diff;
+ j->img_comp[b].dc_pred = dc;
+ data[0] = (short) (dc * dequant[0]);
+
+ // decode AC components, see JPEG spec
+ k = 1;
+ do {
+ unsigned int zig;
+ int c,r,s;
+ if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
+ c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1);
+ r = fac[c];
+ if (r) { // fast-AC path
+ k += (r >> 4) & 15; // run
+ s = r & 15; // combined length
+ j->code_buffer <<= s;
+ j->code_bits -= s;
+ // decode into unzigzag'd location
+ zig = stbi__jpeg_dezigzag[k++];
+ data[zig] = (short) ((r >> 8) * dequant[zig]);
+ } else {
+ int rs = stbi__jpeg_huff_decode(j, hac);
+ if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG");
+ s = rs & 15;
+ r = rs >> 4;
+ if (s == 0) {
+ if (rs != 0xf0) break; // end block
+ k += 16;
+ } else {
+ k += r;
+ // decode into unzigzag'd location
+ zig = stbi__jpeg_dezigzag[k++];
+ data[zig] = (short) (stbi__extend_receive(j,s) * dequant[zig]);
+ }
+ }
+ } while (k < 64);
+ return 1;
+}
+
+static int stbi__jpeg_decode_block_prog_dc(stbi__jpeg *j, short data[64], stbi__huffman *hdc, int b)
+{
+ int diff,dc;
+ int t;
+ if (j->spec_end != 0) return stbi__err("can't merge dc and ac", "Corrupt JPEG");
+
+ if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
+
+ if (j->succ_high == 0) {
+ // first scan for DC coefficient, must be first
+ memset(data,0,64*sizeof(data[0])); // 0 all the ac values now
+ t = stbi__jpeg_huff_decode(j, hdc);
+ diff = t ? stbi__extend_receive(j, t) : 0;
+
+ dc = j->img_comp[b].dc_pred + diff;
+ j->img_comp[b].dc_pred = dc;
+ data[0] = (short) (dc << j->succ_low);
+ } else {
+ // refinement scan for DC coefficient
+ if (stbi__jpeg_get_bit(j))
+ data[0] += (short) (1 << j->succ_low);
+ }
+ return 1;
+}
+
+// @OPTIMIZE: store non-zigzagged during the decode passes,
+// and only de-zigzag when dequantizing
+static int stbi__jpeg_decode_block_prog_ac(stbi__jpeg *j, short data[64], stbi__huffman *hac, stbi__int16 *fac)
+{
+ int k;
+ if (j->spec_start == 0) return stbi__err("can't merge dc and ac", "Corrupt JPEG");
+
+ if (j->succ_high == 0) {
+ int shift = j->succ_low;
+
+ if (j->eob_run) {
+ --j->eob_run;
+ return 1;
+ }
+
+ k = j->spec_start;
+ do {
+ unsigned int zig;
+ int c,r,s;
+ if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
+ c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1);
+ r = fac[c];
+ if (r) { // fast-AC path
+ k += (r >> 4) & 15; // run
+ s = r & 15; // combined length
+ j->code_buffer <<= s;
+ j->code_bits -= s;
+ zig = stbi__jpeg_dezigzag[k++];
+ data[zig] = (short) ((r >> 8) << shift);
+ } else {
+ int rs = stbi__jpeg_huff_decode(j, hac);
+ if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG");
+ s = rs & 15;
+ r = rs >> 4;
+ if (s == 0) {
+ if (r < 15) {
+ j->eob_run = (1 << r);
+ if (r)
+ j->eob_run += stbi__jpeg_get_bits(j, r);
+ --j->eob_run;
+ break;
+ }
+ k += 16;
+ } else {
+ k += r;
+ zig = stbi__jpeg_dezigzag[k++];
+ data[zig] = (short) (stbi__extend_receive(j,s) << shift);
+ }
+ }
+ } while (k <= j->spec_end);
+ } else {
+ // refinement scan for these AC coefficients
+
+ short bit = (short) (1 << j->succ_low);
+
+ if (j->eob_run) {
+ --j->eob_run;
+ for (k = j->spec_start; k <= j->spec_end; ++k) {
+ short *p = &data[stbi__jpeg_dezigzag[k]];
+ if (*p != 0)
+ if (stbi__jpeg_get_bit(j))
+ if ((*p & bit)==0) {
+ if (*p > 0)
+ *p += bit;
+ else
+ *p -= bit;
+ }
+ }
+ } else {
+ k = j->spec_start;
+ do {
+ int r,s;
+ int rs = stbi__jpeg_huff_decode(j, hac); // @OPTIMIZE see if we can use the fast path here, advance-by-r is so slow, eh
+ if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG");
+ s = rs & 15;
+ r = rs >> 4;
+ if (s == 0) {
+ if (r < 15) {
+ j->eob_run = (1 << r) - 1;
+ if (r)
+ j->eob_run += stbi__jpeg_get_bits(j, r);
+ r = 64; // force end of block
+ } else {
+ // r=15 s=0 should write 16 0s, so we just do
+ // a run of 15 0s and then write s (which is 0),
+ // so we don't have to do anything special here
+ }
+ } else {
+ if (s != 1) return stbi__err("bad huffman code", "Corrupt JPEG");
+ // sign bit
+ if (stbi__jpeg_get_bit(j))
+ s = bit;
+ else
+ s = -bit;
+ }
+
+ // advance by r
+ while (k <= j->spec_end) {
+ short *p = &data[stbi__jpeg_dezigzag[k++]];
+ if (*p != 0) {
+ if (stbi__jpeg_get_bit(j))
+ if ((*p & bit)==0) {
+ if (*p > 0)
+ *p += bit;
+ else
+ *p -= bit;
+ }
+ } else {
+ if (r == 0) {
+ *p = (short) s;
+ break;
+ }
+ --r;
+ }
+ }
+ } while (k <= j->spec_end);
+ }
+ }
+ return 1;
+}
+
+// take a -128..127 value and stbi__clamp it and convert to 0..255
+stbi_inline static stbi_uc stbi__clamp(int x)
+{
+ // trick to use a single test to catch both cases
+ if ((unsigned int) x > 255) {
+ if (x < 0) return 0;
+ if (x > 255) return 255;
+ }
+ return (stbi_uc) x;
+}
+
+#define stbi__f2f(x) ((int) (((x) * 4096 + 0.5)))
+#define stbi__fsh(x) ((x) << 12)
+
+// derived from jidctint -- DCT_ISLOW
+#define STBI__IDCT_1D(s0,s1,s2,s3,s4,s5,s6,s7) \
+ int t0,t1,t2,t3,p1,p2,p3,p4,p5,x0,x1,x2,x3; \
+ p2 = s2; \
+ p3 = s6; \
+ p1 = (p2+p3) * stbi__f2f(0.5411961f); \
+ t2 = p1 + p3*stbi__f2f(-1.847759065f); \
+ t3 = p1 + p2*stbi__f2f( 0.765366865f); \
+ p2 = s0; \
+ p3 = s4; \
+ t0 = stbi__fsh(p2+p3); \
+ t1 = stbi__fsh(p2-p3); \
+ x0 = t0+t3; \
+ x3 = t0-t3; \
+ x1 = t1+t2; \
+ x2 = t1-t2; \
+ t0 = s7; \
+ t1 = s5; \
+ t2 = s3; \
+ t3 = s1; \
+ p3 = t0+t2; \
+ p4 = t1+t3; \
+ p1 = t0+t3; \
+ p2 = t1+t2; \
+ p5 = (p3+p4)*stbi__f2f( 1.175875602f); \
+ t0 = t0*stbi__f2f( 0.298631336f); \
+ t1 = t1*stbi__f2f( 2.053119869f); \
+ t2 = t2*stbi__f2f( 3.072711026f); \
+ t3 = t3*stbi__f2f( 1.501321110f); \
+ p1 = p5 + p1*stbi__f2f(-0.899976223f); \
+ p2 = p5 + p2*stbi__f2f(-2.562915447f); \
+ p3 = p3*stbi__f2f(-1.961570560f); \
+ p4 = p4*stbi__f2f(-0.390180644f); \
+ t3 += p1+p4; \
+ t2 += p2+p3; \
+ t1 += p2+p4; \
+ t0 += p1+p3;
+
+static void stbi__idct_block(stbi_uc *out, int out_stride, short data[64])
+{
+ int i,val[64],*v=val;
+ stbi_uc *o;
+ short *d = data;
+
+ // columns
+ for (i=0; i < 8; ++i,++d, ++v) {
+ // if all zeroes, shortcut -- this avoids dequantizing 0s and IDCTing
+ if (d[ 8]==0 && d[16]==0 && d[24]==0 && d[32]==0
+ && d[40]==0 && d[48]==0 && d[56]==0) {
+ // no shortcut 0 seconds
+ // (1|2|3|4|5|6|7)==0 0 seconds
+ // all separate -0.047 seconds
+ // 1 && 2|3 && 4|5 && 6|7: -0.047 seconds
+ int dcterm = d[0] << 2;
+ v[0] = v[8] = v[16] = v[24] = v[32] = v[40] = v[48] = v[56] = dcterm;
+ } else {
+ STBI__IDCT_1D(d[ 0],d[ 8],d[16],d[24],d[32],d[40],d[48],d[56])
+ // constants scaled things up by 1<<12; let's bring them back
+ // down, but keep 2 extra bits of precision
+ x0 += 512; x1 += 512; x2 += 512; x3 += 512;
+ v[ 0] = (x0+t3) >> 10;
+ v[56] = (x0-t3) >> 10;
+ v[ 8] = (x1+t2) >> 10;
+ v[48] = (x1-t2) >> 10;
+ v[16] = (x2+t1) >> 10;
+ v[40] = (x2-t1) >> 10;
+ v[24] = (x3+t0) >> 10;
+ v[32] = (x3-t0) >> 10;
+ }
+ }
+
+ for (i=0, v=val, o=out; i < 8; ++i,v+=8,o+=out_stride) {
+ // no fast case since the first 1D IDCT spread components out
+ STBI__IDCT_1D(v[0],v[1],v[2],v[3],v[4],v[5],v[6],v[7])
+ // constants scaled things up by 1<<12, plus we had 1<<2 from first
+ // loop, plus horizontal and vertical each scale by sqrt(8) so together
+ // we've got an extra 1<<3, so 1<<17 total we need to remove.
+ // so we want to round that, which means adding 0.5 * 1<<17,
+ // aka 65536. Also, we'll end up with -128 to 127 that we want
+ // to encode as 0..255 by adding 128, so we'll add that before the shift
+ x0 += 65536 + (128<<17);
+ x1 += 65536 + (128<<17);
+ x2 += 65536 + (128<<17);
+ x3 += 65536 + (128<<17);
+ // tried computing the shifts into temps, or'ing the temps to see
+ // if any were out of range, but that was slower
+ o[0] = stbi__clamp((x0+t3) >> 17);
+ o[7] = stbi__clamp((x0-t3) >> 17);
+ o[1] = stbi__clamp((x1+t2) >> 17);
+ o[6] = stbi__clamp((x1-t2) >> 17);
+ o[2] = stbi__clamp((x2+t1) >> 17);
+ o[5] = stbi__clamp((x2-t1) >> 17);
+ o[3] = stbi__clamp((x3+t0) >> 17);
+ o[4] = stbi__clamp((x3-t0) >> 17);
+ }
+}
+
+#ifdef STBI_SSE2
+// sse2 integer IDCT. not the fastest possible implementation but it
+// produces bit-identical results to the generic C version so it's
+// fully "transparent".
+static void stbi__idct_simd(stbi_uc *out, int out_stride, short data[64])
+{
+ // This is constructed to match our regular (generic) integer IDCT exactly.
+ __m128i row0, row1, row2, row3, row4, row5, row6, row7;
+ __m128i tmp;
+
+ // dot product constant: even elems=x, odd elems=y
+ #define dct_const(x,y) _mm_setr_epi16((x),(y),(x),(y),(x),(y),(x),(y))
+
+ // out(0) = c0[even]*x + c0[odd]*y (c0, x, y 16-bit, out 32-bit)
+ // out(1) = c1[even]*x + c1[odd]*y
+ #define dct_rot(out0,out1, x,y,c0,c1) \
+ __m128i c0##lo = _mm_unpacklo_epi16((x),(y)); \
+ __m128i c0##hi = _mm_unpackhi_epi16((x),(y)); \
+ __m128i out0##_l = _mm_madd_epi16(c0##lo, c0); \
+ __m128i out0##_h = _mm_madd_epi16(c0##hi, c0); \
+ __m128i out1##_l = _mm_madd_epi16(c0##lo, c1); \
+ __m128i out1##_h = _mm_madd_epi16(c0##hi, c1)
+
+ // out = in << 12 (in 16-bit, out 32-bit)
+ #define dct_widen(out, in) \
+ __m128i out##_l = _mm_srai_epi32(_mm_unpacklo_epi16(_mm_setzero_si128(), (in)), 4); \
+ __m128i out##_h = _mm_srai_epi32(_mm_unpackhi_epi16(_mm_setzero_si128(), (in)), 4)
+
+ // wide add
+ #define dct_wadd(out, a, b) \
+ __m128i out##_l = _mm_add_epi32(a##_l, b##_l); \
+ __m128i out##_h = _mm_add_epi32(a##_h, b##_h)
+
+ // wide sub
+ #define dct_wsub(out, a, b) \
+ __m128i out##_l = _mm_sub_epi32(a##_l, b##_l); \
+ __m128i out##_h = _mm_sub_epi32(a##_h, b##_h)
+
+ // butterfly a/b, add bias, then shift by "s" and pack
+ #define dct_bfly32o(out0, out1, a,b,bias,s) \
+ { \
+ __m128i abiased_l = _mm_add_epi32(a##_l, bias); \
+ __m128i abiased_h = _mm_add_epi32(a##_h, bias); \
+ dct_wadd(sum, abiased, b); \
+ dct_wsub(dif, abiased, b); \
+ out0 = _mm_packs_epi32(_mm_srai_epi32(sum_l, s), _mm_srai_epi32(sum_h, s)); \
+ out1 = _mm_packs_epi32(_mm_srai_epi32(dif_l, s), _mm_srai_epi32(dif_h, s)); \
+ }
+
+ // 8-bit interleave step (for transposes)
+ #define dct_interleave8(a, b) \
+ tmp = a; \
+ a = _mm_unpacklo_epi8(a, b); \
+ b = _mm_unpackhi_epi8(tmp, b)
+
+ // 16-bit interleave step (for transposes)
+ #define dct_interleave16(a, b) \
+ tmp = a; \
+ a = _mm_unpacklo_epi16(a, b); \
+ b = _mm_unpackhi_epi16(tmp, b)
+
+ #define dct_pass(bias,shift) \
+ { \
+ /* even part */ \
+ dct_rot(t2e,t3e, row2,row6, rot0_0,rot0_1); \
+ __m128i sum04 = _mm_add_epi16(row0, row4); \
+ __m128i dif04 = _mm_sub_epi16(row0, row4); \
+ dct_widen(t0e, sum04); \
+ dct_widen(t1e, dif04); \
+ dct_wadd(x0, t0e, t3e); \
+ dct_wsub(x3, t0e, t3e); \
+ dct_wadd(x1, t1e, t2e); \
+ dct_wsub(x2, t1e, t2e); \
+ /* odd part */ \
+ dct_rot(y0o,y2o, row7,row3, rot2_0,rot2_1); \
+ dct_rot(y1o,y3o, row5,row1, rot3_0,rot3_1); \
+ __m128i sum17 = _mm_add_epi16(row1, row7); \
+ __m128i sum35 = _mm_add_epi16(row3, row5); \
+ dct_rot(y4o,y5o, sum17,sum35, rot1_0,rot1_1); \
+ dct_wadd(x4, y0o, y4o); \
+ dct_wadd(x5, y1o, y5o); \
+ dct_wadd(x6, y2o, y5o); \
+ dct_wadd(x7, y3o, y4o); \
+ dct_bfly32o(row0,row7, x0,x7,bias,shift); \
+ dct_bfly32o(row1,row6, x1,x6,bias,shift); \
+ dct_bfly32o(row2,row5, x2,x5,bias,shift); \
+ dct_bfly32o(row3,row4, x3,x4,bias,shift); \
+ }
+
+ __m128i rot0_0 = dct_const(stbi__f2f(0.5411961f), stbi__f2f(0.5411961f) + stbi__f2f(-1.847759065f));
+ __m128i rot0_1 = dct_const(stbi__f2f(0.5411961f) + stbi__f2f( 0.765366865f), stbi__f2f(0.5411961f));
+ __m128i rot1_0 = dct_const(stbi__f2f(1.175875602f) + stbi__f2f(-0.899976223f), stbi__f2f(1.175875602f));
+ __m128i rot1_1 = dct_const(stbi__f2f(1.175875602f), stbi__f2f(1.175875602f) + stbi__f2f(-2.562915447f));
+ __m128i rot2_0 = dct_const(stbi__f2f(-1.961570560f) + stbi__f2f( 0.298631336f), stbi__f2f(-1.961570560f));
+ __m128i rot2_1 = dct_const(stbi__f2f(-1.961570560f), stbi__f2f(-1.961570560f) + stbi__f2f( 3.072711026f));
+ __m128i rot3_0 = dct_const(stbi__f2f(-0.390180644f) + stbi__f2f( 2.053119869f), stbi__f2f(-0.390180644f));
+ __m128i rot3_1 = dct_const(stbi__f2f(-0.390180644f), stbi__f2f(-0.390180644f) + stbi__f2f( 1.501321110f));
+
+ // rounding biases in column/row passes, see stbi__idct_block for explanation.
+ __m128i bias_0 = _mm_set1_epi32(512);
+ __m128i bias_1 = _mm_set1_epi32(65536 + (128<<17));
+
+ // load
+ row0 = _mm_load_si128((const __m128i *) (data + 0*8));
+ row1 = _mm_load_si128((const __m128i *) (data + 1*8));
+ row2 = _mm_load_si128((const __m128i *) (data + 2*8));
+ row3 = _mm_load_si128((const __m128i *) (data + 3*8));
+ row4 = _mm_load_si128((const __m128i *) (data + 4*8));
+ row5 = _mm_load_si128((const __m128i *) (data + 5*8));
+ row6 = _mm_load_si128((const __m128i *) (data + 6*8));
+ row7 = _mm_load_si128((const __m128i *) (data + 7*8));
+
+ // column pass
+ dct_pass(bias_0, 10);
+
+ {
+ // 16bit 8x8 transpose pass 1
+ dct_interleave16(row0, row4);
+ dct_interleave16(row1, row5);
+ dct_interleave16(row2, row6);
+ dct_interleave16(row3, row7);
+
+ // transpose pass 2
+ dct_interleave16(row0, row2);
+ dct_interleave16(row1, row3);
+ dct_interleave16(row4, row6);
+ dct_interleave16(row5, row7);
+
+ // transpose pass 3
+ dct_interleave16(row0, row1);
+ dct_interleave16(row2, row3);
+ dct_interleave16(row4, row5);
+ dct_interleave16(row6, row7);
+ }
+
+ // row pass
+ dct_pass(bias_1, 17);
+
+ {
+ // pack
+ __m128i p0 = _mm_packus_epi16(row0, row1); // a0a1a2a3...a7b0b1b2b3...b7
+ __m128i p1 = _mm_packus_epi16(row2, row3);
+ __m128i p2 = _mm_packus_epi16(row4, row5);
+ __m128i p3 = _mm_packus_epi16(row6, row7);
+
+ // 8bit 8x8 transpose pass 1
+ dct_interleave8(p0, p2); // a0e0a1e1...
+ dct_interleave8(p1, p3); // c0g0c1g1...
+
+ // transpose pass 2
+ dct_interleave8(p0, p1); // a0c0e0g0...
+ dct_interleave8(p2, p3); // b0d0f0h0...
+
+ // transpose pass 3
+ dct_interleave8(p0, p2); // a0b0c0d0...
+ dct_interleave8(p1, p3); // a4b4c4d4...
+
+ // store
+ _mm_storel_epi64((__m128i *) out, p0); out += out_stride;
+ _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p0, 0x4e)); out += out_stride;
+ _mm_storel_epi64((__m128i *) out, p2); out += out_stride;
+ _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p2, 0x4e)); out += out_stride;
+ _mm_storel_epi64((__m128i *) out, p1); out += out_stride;
+ _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p1, 0x4e)); out += out_stride;
+ _mm_storel_epi64((__m128i *) out, p3); out += out_stride;
+ _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p3, 0x4e));
+ }
+
+#undef dct_const
+#undef dct_rot
+#undef dct_widen
+#undef dct_wadd
+#undef dct_wsub
+#undef dct_bfly32o
+#undef dct_interleave8
+#undef dct_interleave16
+#undef dct_pass
+}
+
+#endif // STBI_SSE2
+
+#ifdef STBI_NEON
+
+// NEON integer IDCT. should produce bit-identical
+// results to the generic C version.
+static void stbi__idct_simd(stbi_uc *out, int out_stride, short data[64])
+{
+ int16x8_t row0, row1, row2, row3, row4, row5, row6, row7;
+
+ int16x4_t rot0_0 = vdup_n_s16(stbi__f2f(0.5411961f));
+ int16x4_t rot0_1 = vdup_n_s16(stbi__f2f(-1.847759065f));
+ int16x4_t rot0_2 = vdup_n_s16(stbi__f2f( 0.765366865f));
+ int16x4_t rot1_0 = vdup_n_s16(stbi__f2f( 1.175875602f));
+ int16x4_t rot1_1 = vdup_n_s16(stbi__f2f(-0.899976223f));
+ int16x4_t rot1_2 = vdup_n_s16(stbi__f2f(-2.562915447f));
+ int16x4_t rot2_0 = vdup_n_s16(stbi__f2f(-1.961570560f));
+ int16x4_t rot2_1 = vdup_n_s16(stbi__f2f(-0.390180644f));
+ int16x4_t rot3_0 = vdup_n_s16(stbi__f2f( 0.298631336f));
+ int16x4_t rot3_1 = vdup_n_s16(stbi__f2f( 2.053119869f));
+ int16x4_t rot3_2 = vdup_n_s16(stbi__f2f( 3.072711026f));
+ int16x4_t rot3_3 = vdup_n_s16(stbi__f2f( 1.501321110f));
+
+#define dct_long_mul(out, inq, coeff) \
+ int32x4_t out##_l = vmull_s16(vget_low_s16(inq), coeff); \
+ int32x4_t out##_h = vmull_s16(vget_high_s16(inq), coeff)
+
+#define dct_long_mac(out, acc, inq, coeff) \
+ int32x4_t out##_l = vmlal_s16(acc##_l, vget_low_s16(inq), coeff); \
+ int32x4_t out##_h = vmlal_s16(acc##_h, vget_high_s16(inq), coeff)
+
+#define dct_widen(out, inq) \
+ int32x4_t out##_l = vshll_n_s16(vget_low_s16(inq), 12); \
+ int32x4_t out##_h = vshll_n_s16(vget_high_s16(inq), 12)
+
+// wide add
+#define dct_wadd(out, a, b) \
+ int32x4_t out##_l = vaddq_s32(a##_l, b##_l); \
+ int32x4_t out##_h = vaddq_s32(a##_h, b##_h)
+
+// wide sub
+#define dct_wsub(out, a, b) \
+ int32x4_t out##_l = vsubq_s32(a##_l, b##_l); \
+ int32x4_t out##_h = vsubq_s32(a##_h, b##_h)
+
+// butterfly a/b, then shift using "shiftop" by "s" and pack
+#define dct_bfly32o(out0,out1, a,b,shiftop,s) \
+ { \
+ dct_wadd(sum, a, b); \
+ dct_wsub(dif, a, b); \
+ out0 = vcombine_s16(shiftop(sum_l, s), shiftop(sum_h, s)); \
+ out1 = vcombine_s16(shiftop(dif_l, s), shiftop(dif_h, s)); \
+ }
+
+#define dct_pass(shiftop, shift) \
+ { \
+ /* even part */ \
+ int16x8_t sum26 = vaddq_s16(row2, row6); \
+ dct_long_mul(p1e, sum26, rot0_0); \
+ dct_long_mac(t2e, p1e, row6, rot0_1); \
+ dct_long_mac(t3e, p1e, row2, rot0_2); \
+ int16x8_t sum04 = vaddq_s16(row0, row4); \
+ int16x8_t dif04 = vsubq_s16(row0, row4); \
+ dct_widen(t0e, sum04); \
+ dct_widen(t1e, dif04); \
+ dct_wadd(x0, t0e, t3e); \
+ dct_wsub(x3, t0e, t3e); \
+ dct_wadd(x1, t1e, t2e); \
+ dct_wsub(x2, t1e, t2e); \
+ /* odd part */ \
+ int16x8_t sum15 = vaddq_s16(row1, row5); \
+ int16x8_t sum17 = vaddq_s16(row1, row7); \
+ int16x8_t sum35 = vaddq_s16(row3, row5); \
+ int16x8_t sum37 = vaddq_s16(row3, row7); \
+ int16x8_t sumodd = vaddq_s16(sum17, sum35); \
+ dct_long_mul(p5o, sumodd, rot1_0); \
+ dct_long_mac(p1o, p5o, sum17, rot1_1); \
+ dct_long_mac(p2o, p5o, sum35, rot1_2); \
+ dct_long_mul(p3o, sum37, rot2_0); \
+ dct_long_mul(p4o, sum15, rot2_1); \
+ dct_wadd(sump13o, p1o, p3o); \
+ dct_wadd(sump24o, p2o, p4o); \
+ dct_wadd(sump23o, p2o, p3o); \
+ dct_wadd(sump14o, p1o, p4o); \
+ dct_long_mac(x4, sump13o, row7, rot3_0); \
+ dct_long_mac(x5, sump24o, row5, rot3_1); \
+ dct_long_mac(x6, sump23o, row3, rot3_2); \
+ dct_long_mac(x7, sump14o, row1, rot3_3); \
+ dct_bfly32o(row0,row7, x0,x7,shiftop,shift); \
+ dct_bfly32o(row1,row6, x1,x6,shiftop,shift); \
+ dct_bfly32o(row2,row5, x2,x5,shiftop,shift); \
+ dct_bfly32o(row3,row4, x3,x4,shiftop,shift); \
+ }
+
+ // load
+ row0 = vld1q_s16(data + 0*8);
+ row1 = vld1q_s16(data + 1*8);
+ row2 = vld1q_s16(data + 2*8);
+ row3 = vld1q_s16(data + 3*8);
+ row4 = vld1q_s16(data + 4*8);
+ row5 = vld1q_s16(data + 5*8);
+ row6 = vld1q_s16(data + 6*8);
+ row7 = vld1q_s16(data + 7*8);
+
+ // add DC bias
+ row0 = vaddq_s16(row0, vsetq_lane_s16(1024, vdupq_n_s16(0), 0));
+
+ // column pass
+ dct_pass(vrshrn_n_s32, 10);
+
+ // 16bit 8x8 transpose
+ {
+// these three map to a single VTRN.16, VTRN.32, and VSWP, respectively.
+// whether compilers actually get this is another story, sadly.
+#define dct_trn16(x, y) { int16x8x2_t t = vtrnq_s16(x, y); x = t.val[0]; y = t.val[1]; }
+#define dct_trn32(x, y) { int32x4x2_t t = vtrnq_s32(vreinterpretq_s32_s16(x), vreinterpretq_s32_s16(y)); x = vreinterpretq_s16_s32(t.val[0]); y = vreinterpretq_s16_s32(t.val[1]); }
+#define dct_trn64(x, y) { int16x8_t x0 = x; int16x8_t y0 = y; x = vcombine_s16(vget_low_s16(x0), vget_low_s16(y0)); y = vcombine_s16(vget_high_s16(x0), vget_high_s16(y0)); }
+
+ // pass 1
+ dct_trn16(row0, row1); // a0b0a2b2a4b4a6b6
+ dct_trn16(row2, row3);
+ dct_trn16(row4, row5);
+ dct_trn16(row6, row7);
+
+ // pass 2
+ dct_trn32(row0, row2); // a0b0c0d0a4b4c4d4
+ dct_trn32(row1, row3);
+ dct_trn32(row4, row6);
+ dct_trn32(row5, row7);
+
+ // pass 3
+ dct_trn64(row0, row4); // a0b0c0d0e0f0g0h0
+ dct_trn64(row1, row5);
+ dct_trn64(row2, row6);
+ dct_trn64(row3, row7);
+
+#undef dct_trn16
+#undef dct_trn32
+#undef dct_trn64
+ }
+
+ // row pass
+ // vrshrn_n_s32 only supports shifts up to 16, we need
+ // 17. so do a non-rounding shift of 16 first then follow
+ // up with a rounding shift by 1.
+ dct_pass(vshrn_n_s32, 16);
+
+ {
+ // pack and round
+ uint8x8_t p0 = vqrshrun_n_s16(row0, 1);
+ uint8x8_t p1 = vqrshrun_n_s16(row1, 1);
+ uint8x8_t p2 = vqrshrun_n_s16(row2, 1);
+ uint8x8_t p3 = vqrshrun_n_s16(row3, 1);
+ uint8x8_t p4 = vqrshrun_n_s16(row4, 1);
+ uint8x8_t p5 = vqrshrun_n_s16(row5, 1);
+ uint8x8_t p6 = vqrshrun_n_s16(row6, 1);
+ uint8x8_t p7 = vqrshrun_n_s16(row7, 1);
+
+ // again, these can translate into one instruction, but often don't.
+#define dct_trn8_8(x, y) { uint8x8x2_t t = vtrn_u8(x, y); x = t.val[0]; y = t.val[1]; }
+#define dct_trn8_16(x, y) { uint16x4x2_t t = vtrn_u16(vreinterpret_u16_u8(x), vreinterpret_u16_u8(y)); x = vreinterpret_u8_u16(t.val[0]); y = vreinterpret_u8_u16(t.val[1]); }
+#define dct_trn8_32(x, y) { uint32x2x2_t t = vtrn_u32(vreinterpret_u32_u8(x), vreinterpret_u32_u8(y)); x = vreinterpret_u8_u32(t.val[0]); y = vreinterpret_u8_u32(t.val[1]); }
+
+ // sadly can't use interleaved stores here since we only write
+ // 8 bytes to each scan line!
+
+ // 8x8 8-bit transpose pass 1
+ dct_trn8_8(p0, p1);
+ dct_trn8_8(p2, p3);
+ dct_trn8_8(p4, p5);
+ dct_trn8_8(p6, p7);
+
+ // pass 2
+ dct_trn8_16(p0, p2);
+ dct_trn8_16(p1, p3);
+ dct_trn8_16(p4, p6);
+ dct_trn8_16(p5, p7);
+
+ // pass 3
+ dct_trn8_32(p0, p4);
+ dct_trn8_32(p1, p5);
+ dct_trn8_32(p2, p6);
+ dct_trn8_32(p3, p7);
+
+ // store
+ vst1_u8(out, p0); out += out_stride;
+ vst1_u8(out, p1); out += out_stride;
+ vst1_u8(out, p2); out += out_stride;
+ vst1_u8(out, p3); out += out_stride;
+ vst1_u8(out, p4); out += out_stride;
+ vst1_u8(out, p5); out += out_stride;
+ vst1_u8(out, p6); out += out_stride;
+ vst1_u8(out, p7);
+
+#undef dct_trn8_8
+#undef dct_trn8_16
+#undef dct_trn8_32
+ }
+
+#undef dct_long_mul
+#undef dct_long_mac
+#undef dct_widen
+#undef dct_wadd
+#undef dct_wsub
+#undef dct_bfly32o
+#undef dct_pass
+}
+
+#endif // STBI_NEON
+
+#define STBI__MARKER_none 0xff
+// if there's a pending marker from the entropy stream, return that
+// otherwise, fetch from the stream and get a marker. if there's no
+// marker, return 0xff, which is never a valid marker value
+static stbi_uc stbi__get_marker(stbi__jpeg *j)
+{
+ stbi_uc x;
+ if (j->marker != STBI__MARKER_none) { x = j->marker; j->marker = STBI__MARKER_none; return x; }
+ x = stbi__get8(j->s);
+ if (x != 0xff) return STBI__MARKER_none;
+ while (x == 0xff)
+ x = stbi__get8(j->s); // consume repeated 0xff fill bytes
+ return x;
+}
+
+// in each scan, we'll have scan_n components, and the order
+// of the components is specified by order[]
+#define STBI__RESTART(x) ((x) >= 0xd0 && (x) <= 0xd7)
+
+// after a restart interval, stbi__jpeg_reset the entropy decoder and
+// the dc prediction
+static void stbi__jpeg_reset(stbi__jpeg *j)
+{
+ j->code_bits = 0;
+ j->code_buffer = 0;
+ j->nomore = 0;
+ j->img_comp[0].dc_pred = j->img_comp[1].dc_pred = j->img_comp[2].dc_pred = j->img_comp[3].dc_pred = 0;
+ j->marker = STBI__MARKER_none;
+ j->todo = j->restart_interval ? j->restart_interval : 0x7fffffff;
+ j->eob_run = 0;
+ // no more than 1<<31 MCUs if no restart_interal? that's plenty safe,
+ // since we don't even allow 1<<30 pixels
+}
+
+static int stbi__parse_entropy_coded_data(stbi__jpeg *z)
+{
+ stbi__jpeg_reset(z);
+ if (!z->progressive) {
+ if (z->scan_n == 1) {
+ int i,j;
+ STBI_SIMD_ALIGN(short, data[64]);
+ int n = z->order[0];
+ // non-interleaved data, we just need to process one block at a time,
+ // in trivial scanline order
+ // number of blocks to do just depends on how many actual "pixels" this
+ // component has, independent of interleaved MCU blocking and such
+ int w = (z->img_comp[n].x+7) >> 3;
+ int h = (z->img_comp[n].y+7) >> 3;
+ for (j=0; j < h; ++j) {
+ for (i=0; i < w; ++i) {
+ int ha = z->img_comp[n].ha;
+ if (!stbi__jpeg_decode_block(z, data, z->huff_dc+z->img_comp[n].hd, z->huff_ac+ha, z->fast_ac[ha], n, z->dequant[z->img_comp[n].tq])) return 0;
+ z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*j*8+i*8, z->img_comp[n].w2, data);
+ // every data block is an MCU, so countdown the restart interval
+ if (--z->todo <= 0) {
+ if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
+ // if it's NOT a restart, then just bail, so we get corrupt data
+ // rather than no data
+ if (!STBI__RESTART(z->marker)) return 1;
+ stbi__jpeg_reset(z);
+ }
+ }
+ }
+ return 1;
+ } else { // interleaved
+ int i,j,k,x,y;
+ STBI_SIMD_ALIGN(short, data[64]);
+ for (j=0; j < z->img_mcu_y; ++j) {
+ for (i=0; i < z->img_mcu_x; ++i) {
+ // scan an interleaved mcu... process scan_n components in order
+ for (k=0; k < z->scan_n; ++k) {
+ int n = z->order[k];
+ // scan out an mcu's worth of this component; that's just determined
+ // by the basic H and V specified for the component
+ for (y=0; y < z->img_comp[n].v; ++y) {
+ for (x=0; x < z->img_comp[n].h; ++x) {
+ int x2 = (i*z->img_comp[n].h + x)*8;
+ int y2 = (j*z->img_comp[n].v + y)*8;
+ int ha = z->img_comp[n].ha;
+ if (!stbi__jpeg_decode_block(z, data, z->huff_dc+z->img_comp[n].hd, z->huff_ac+ha, z->fast_ac[ha], n, z->dequant[z->img_comp[n].tq])) return 0;
+ z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*y2+x2, z->img_comp[n].w2, data);
+ }
+ }
+ }
+ // after all interleaved components, that's an interleaved MCU,
+ // so now count down the restart interval
+ if (--z->todo <= 0) {
+ if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
+ if (!STBI__RESTART(z->marker)) return 1;
+ stbi__jpeg_reset(z);
+ }
+ }
+ }
+ return 1;
+ }
+ } else {
+ if (z->scan_n == 1) {
+ int i,j;
+ int n = z->order[0];
+ // non-interleaved data, we just need to process one block at a time,
+ // in trivial scanline order
+ // number of blocks to do just depends on how many actual "pixels" this
+ // component has, independent of interleaved MCU blocking and such
+ int w = (z->img_comp[n].x+7) >> 3;
+ int h = (z->img_comp[n].y+7) >> 3;
+ for (j=0; j < h; ++j) {
+ for (i=0; i < w; ++i) {
+ short *data = z->img_comp[n].coeff + 64 * (i + j * z->img_comp[n].coeff_w);
+ if (z->spec_start == 0) {
+ if (!stbi__jpeg_decode_block_prog_dc(z, data, &z->huff_dc[z->img_comp[n].hd], n))
+ return 0;
+ } else {
+ int ha = z->img_comp[n].ha;
+ if (!stbi__jpeg_decode_block_prog_ac(z, data, &z->huff_ac[ha], z->fast_ac[ha]))
+ return 0;
+ }
+ // every data block is an MCU, so countdown the restart interval
+ if (--z->todo <= 0) {
+ if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
+ if (!STBI__RESTART(z->marker)) return 1;
+ stbi__jpeg_reset(z);
+ }
+ }
+ }
+ return 1;
+ } else { // interleaved
+ int i,j,k,x,y;
+ for (j=0; j < z->img_mcu_y; ++j) {
+ for (i=0; i < z->img_mcu_x; ++i) {
+ // scan an interleaved mcu... process scan_n components in order
+ for (k=0; k < z->scan_n; ++k) {
+ int n = z->order[k];
+ // scan out an mcu's worth of this component; that's just determined
+ // by the basic H and V specified for the component
+ for (y=0; y < z->img_comp[n].v; ++y) {
+ for (x=0; x < z->img_comp[n].h; ++x) {
+ int x2 = (i*z->img_comp[n].h + x);
+ int y2 = (j*z->img_comp[n].v + y);
+ short *data = z->img_comp[n].coeff + 64 * (x2 + y2 * z->img_comp[n].coeff_w);
+ if (!stbi__jpeg_decode_block_prog_dc(z, data, &z->huff_dc[z->img_comp[n].hd], n))
+ return 0;
+ }
+ }
+ }
+ // after all interleaved components, that's an interleaved MCU,
+ // so now count down the restart interval
+ if (--z->todo <= 0) {
+ if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
+ if (!STBI__RESTART(z->marker)) return 1;
+ stbi__jpeg_reset(z);
+ }
+ }
+ }
+ return 1;
+ }
+ }
+}
+
+static void stbi__jpeg_dequantize(short *data, stbi__uint16 *dequant)
+{
+ int i;
+ for (i=0; i < 64; ++i)
+ data[i] *= dequant[i];
+}
+
+static void stbi__jpeg_finish(stbi__jpeg *z)
+{
+ if (z->progressive) {
+ // dequantize and idct the data
+ int i,j,n;
+ for (n=0; n < z->s->img_n; ++n) {
+ int w = (z->img_comp[n].x+7) >> 3;
+ int h = (z->img_comp[n].y+7) >> 3;
+ for (j=0; j < h; ++j) {
+ for (i=0; i < w; ++i) {
+ short *data = z->img_comp[n].coeff + 64 * (i + j * z->img_comp[n].coeff_w);
+ stbi__jpeg_dequantize(data, z->dequant[z->img_comp[n].tq]);
+ z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*j*8+i*8, z->img_comp[n].w2, data);
+ }
+ }
+ }
+ }
+}
+
+static int stbi__process_marker(stbi__jpeg *z, int m)
+{
+ int L;
+ switch (m) {
+ case STBI__MARKER_none: // no marker found
+ return stbi__err("expected marker","Corrupt JPEG");
+
+ case 0xDD: // DRI - specify restart interval
+ if (stbi__get16be(z->s) != 4) return stbi__err("bad DRI len","Corrupt JPEG");
+ z->restart_interval = stbi__get16be(z->s);
+ return 1;
+
+ case 0xDB: // DQT - define quantization table
+ L = stbi__get16be(z->s)-2;
+ while (L > 0) {
+ int q = stbi__get8(z->s);
+ int p = q >> 4, sixteen = (p != 0);
+ int t = q & 15,i;
+ if (p != 0 && p != 1) return stbi__err("bad DQT type","Corrupt JPEG");
+ if (t > 3) return stbi__err("bad DQT table","Corrupt JPEG");
+
+ for (i=0; i < 64; ++i)
+ z->dequant[t][stbi__jpeg_dezigzag[i]] = sixteen ? stbi__get16be(z->s) : stbi__get8(z->s);
+ L -= (sixteen ? 129 : 65);
+ }
+ return L==0;
+
+ case 0xC4: // DHT - define huffman table
+ L = stbi__get16be(z->s)-2;
+ while (L > 0) {
+ stbi_uc *v;
+ int sizes[16],i,n=0;
+ int q = stbi__get8(z->s);
+ int tc = q >> 4;
+ int th = q & 15;
+ if (tc > 1 || th > 3) return stbi__err("bad DHT header","Corrupt JPEG");
+ for (i=0; i < 16; ++i) {
+ sizes[i] = stbi__get8(z->s);
+ n += sizes[i];
+ }
+ L -= 17;
+ if (tc == 0) {
+ if (!stbi__build_huffman(z->huff_dc+th, sizes)) return 0;
+ v = z->huff_dc[th].values;
+ } else {
+ if (!stbi__build_huffman(z->huff_ac+th, sizes)) return 0;
+ v = z->huff_ac[th].values;
+ }
+ for (i=0; i < n; ++i)
+ v[i] = stbi__get8(z->s);
+ if (tc != 0)
+ stbi__build_fast_ac(z->fast_ac[th], z->huff_ac + th);
+ L -= n;
+ }
+ return L==0;
+ }
+
+ // check for comment block or APP blocks
+ if ((m >= 0xE0 && m <= 0xEF) || m == 0xFE) {
+ L = stbi__get16be(z->s);
+ if (L < 2) {
+ if (m == 0xFE)
+ return stbi__err("bad COM len","Corrupt JPEG");
+ else
+ return stbi__err("bad APP len","Corrupt JPEG");
+ }
+ L -= 2;
+
+ if (m == 0xE0 && L >= 5) { // JFIF APP0 segment
+ static const unsigned char tag[5] = {'J','F','I','F','\0'};
+ int ok = 1;
+ int i;
+ for (i=0; i < 5; ++i)
+ if (stbi__get8(z->s) != tag[i])
+ ok = 0;
+ L -= 5;
+ if (ok)
+ z->jfif = 1;
+ } else if (m == 0xEE && L >= 12) { // Adobe APP14 segment
+ static const unsigned char tag[6] = {'A','d','o','b','e','\0'};
+ int ok = 1;
+ int i;
+ for (i=0; i < 6; ++i)
+ if (stbi__get8(z->s) != tag[i])
+ ok = 0;
+ L -= 6;
+ if (ok) {
+ stbi__get8(z->s); // version
+ stbi__get16be(z->s); // flags0
+ stbi__get16be(z->s); // flags1
+ z->app14_color_transform = stbi__get8(z->s); // color transform
+ L -= 6;
+ }
+ }
+
+ stbi__skip(z->s, L);
+ return 1;
+ }
+
+ return stbi__err("unknown marker","Corrupt JPEG");
+}
+
+// after we see SOS
+static int stbi__process_scan_header(stbi__jpeg *z)
+{
+ int i;
+ int Ls = stbi__get16be(z->s);
+ z->scan_n = stbi__get8(z->s);
+ if (z->scan_n < 1 || z->scan_n > 4 || z->scan_n > (int) z->s->img_n) return stbi__err("bad SOS component count","Corrupt JPEG");
+ if (Ls != 6+2*z->scan_n) return stbi__err("bad SOS len","Corrupt JPEG");
+ for (i=0; i < z->scan_n; ++i) {
+ int id = stbi__get8(z->s), which;
+ int q = stbi__get8(z->s);
+ for (which = 0; which < z->s->img_n; ++which)
+ if (z->img_comp[which].id == id)
+ break;
+ if (which == z->s->img_n) return 0; // no match
+ z->img_comp[which].hd = q >> 4; if (z->img_comp[which].hd > 3) return stbi__err("bad DC huff","Corrupt JPEG");
+ z->img_comp[which].ha = q & 15; if (z->img_comp[which].ha > 3) return stbi__err("bad AC huff","Corrupt JPEG");
+ z->order[i] = which;
+ }
+
+ {
+ int aa;
+ z->spec_start = stbi__get8(z->s);
+ z->spec_end = stbi__get8(z->s); // should be 63, but might be 0
+ aa = stbi__get8(z->s);
+ z->succ_high = (aa >> 4);
+ z->succ_low = (aa & 15);
+ if (z->progressive) {
+ if (z->spec_start > 63 || z->spec_end > 63 || z->spec_start > z->spec_end || z->succ_high > 13 || z->succ_low > 13)
+ return stbi__err("bad SOS", "Corrupt JPEG");
+ } else {
+ if (z->spec_start != 0) return stbi__err("bad SOS","Corrupt JPEG");
+ if (z->succ_high != 0 || z->succ_low != 0) return stbi__err("bad SOS","Corrupt JPEG");
+ z->spec_end = 63;
+ }
+ }
+
+ return 1;
+}
+
+static int stbi__free_jpeg_components(stbi__jpeg *z, int ncomp, int why)
+{
+ int i;
+ for (i=0; i < ncomp; ++i) {
+ if (z->img_comp[i].raw_data) {
+ STBI_FREE(z->img_comp[i].raw_data);
+ z->img_comp[i].raw_data = NULL;
+ z->img_comp[i].data = NULL;
+ }
+ if (z->img_comp[i].raw_coeff) {
+ STBI_FREE(z->img_comp[i].raw_coeff);
+ z->img_comp[i].raw_coeff = 0;
+ z->img_comp[i].coeff = 0;
+ }
+ if (z->img_comp[i].linebuf) {
+ STBI_FREE(z->img_comp[i].linebuf);
+ z->img_comp[i].linebuf = NULL;
+ }
+ }
+ return why;
+}
+
+static int stbi__process_frame_header(stbi__jpeg *z, int scan)
+{
+ stbi__context *s = z->s;
+ int Lf,p,i,q, h_max=1,v_max=1,c;
+ Lf = stbi__get16be(s); if (Lf < 11) return stbi__err("bad SOF len","Corrupt JPEG"); // JPEG
+ p = stbi__get8(s); if (p != 8) return stbi__err("only 8-bit","JPEG format not supported: 8-bit only"); // JPEG baseline
+ s->img_y = stbi__get16be(s); if (s->img_y == 0) return stbi__err("no header height", "JPEG format not supported: delayed height"); // Legal, but we don't handle it--but neither does IJG
+ s->img_x = stbi__get16be(s); if (s->img_x == 0) return stbi__err("0 width","Corrupt JPEG"); // JPEG requires
+ c = stbi__get8(s);
+ if (c != 3 && c != 1 && c != 4) return stbi__err("bad component count","Corrupt JPEG");
+ s->img_n = c;
+ for (i=0; i < c; ++i) {
+ z->img_comp[i].data = NULL;
+ z->img_comp[i].linebuf = NULL;
+ }
+
+ if (Lf != 8+3*s->img_n) return stbi__err("bad SOF len","Corrupt JPEG");
+
+ z->rgb = 0;
+ for (i=0; i < s->img_n; ++i) {
+ static unsigned char rgb[3] = { 'R', 'G', 'B' };
+ z->img_comp[i].id = stbi__get8(s);
+ if (s->img_n == 3 && z->img_comp[i].id == rgb[i])
+ ++z->rgb;
+ q = stbi__get8(s);
+ z->img_comp[i].h = (q >> 4); if (!z->img_comp[i].h || z->img_comp[i].h > 4) return stbi__err("bad H","Corrupt JPEG");
+ z->img_comp[i].v = q & 15; if (!z->img_comp[i].v || z->img_comp[i].v > 4) return stbi__err("bad V","Corrupt JPEG");
+ z->img_comp[i].tq = stbi__get8(s); if (z->img_comp[i].tq > 3) return stbi__err("bad TQ","Corrupt JPEG");
+ }
+
+ if (scan != STBI__SCAN_load) return 1;
+
+ if (!stbi__mad3sizes_valid(s->img_x, s->img_y, s->img_n, 0)) return stbi__err("too large", "Image too large to decode");
+
+ for (i=0; i < s->img_n; ++i) {
+ if (z->img_comp[i].h > h_max) h_max = z->img_comp[i].h;
+ if (z->img_comp[i].v > v_max) v_max = z->img_comp[i].v;
+ }
+
+ // compute interleaved mcu info
+ z->img_h_max = h_max;
+ z->img_v_max = v_max;
+ z->img_mcu_w = h_max * 8;
+ z->img_mcu_h = v_max * 8;
+ // these sizes can't be more than 17 bits
+ z->img_mcu_x = (s->img_x + z->img_mcu_w-1) / z->img_mcu_w;
+ z->img_mcu_y = (s->img_y + z->img_mcu_h-1) / z->img_mcu_h;
+
+ for (i=0; i < s->img_n; ++i) {
+ // number of effective pixels (e.g. for non-interleaved MCU)
+ z->img_comp[i].x = (s->img_x * z->img_comp[i].h + h_max-1) / h_max;
+ z->img_comp[i].y = (s->img_y * z->img_comp[i].v + v_max-1) / v_max;
+ // to simplify generation, we'll allocate enough memory to decode
+ // the bogus oversized data from using interleaved MCUs and their
+ // big blocks (e.g. a 16x16 iMCU on an image of width 33); we won't
+ // discard the extra data until colorspace conversion
+ //
+ // img_mcu_x, img_mcu_y: <=17 bits; comp[i].h and .v are <=4 (checked earlier)
+ // so these muls can't overflow with 32-bit ints (which we require)
+ z->img_comp[i].w2 = z->img_mcu_x * z->img_comp[i].h * 8;
+ z->img_comp[i].h2 = z->img_mcu_y * z->img_comp[i].v * 8;
+ z->img_comp[i].coeff = 0;
+ z->img_comp[i].raw_coeff = 0;
+ z->img_comp[i].linebuf = NULL;
+ z->img_comp[i].raw_data = stbi__malloc_mad2(z->img_comp[i].w2, z->img_comp[i].h2, 15);
+ if (z->img_comp[i].raw_data == NULL)
+ return stbi__free_jpeg_components(z, i+1, stbi__err("outofmem", "Out of memory"));
+ // align blocks for idct using mmx/sse
+ z->img_comp[i].data = (stbi_uc*) (((size_t) z->img_comp[i].raw_data + 15) & ~15);
+ if (z->progressive) {
+ // w2, h2 are multiples of 8 (see above)
+ z->img_comp[i].coeff_w = z->img_comp[i].w2 / 8;
+ z->img_comp[i].coeff_h = z->img_comp[i].h2 / 8;
+ z->img_comp[i].raw_coeff = stbi__malloc_mad3(z->img_comp[i].w2, z->img_comp[i].h2, sizeof(short), 15);
+ if (z->img_comp[i].raw_coeff == NULL)
+ return stbi__free_jpeg_components(z, i+1, stbi__err("outofmem", "Out of memory"));
+ z->img_comp[i].coeff = (short*) (((size_t) z->img_comp[i].raw_coeff + 15) & ~15);
+ }
+ }
+
+ return 1;
+}
+
+// use comparisons since in some cases we handle more than one case (e.g. SOF)
+#define stbi__DNL(x) ((x) == 0xdc)
+#define stbi__SOI(x) ((x) == 0xd8)
+#define stbi__EOI(x) ((x) == 0xd9)
+#define stbi__SOF(x) ((x) == 0xc0 || (x) == 0xc1 || (x) == 0xc2)
+#define stbi__SOS(x) ((x) == 0xda)
+
+#define stbi__SOF_progressive(x) ((x) == 0xc2)
+
+static int stbi__decode_jpeg_header(stbi__jpeg *z, int scan)
+{
+ int m;
+ z->jfif = 0;
+ z->app14_color_transform = -1; // valid values are 0,1,2
+ z->marker = STBI__MARKER_none; // initialize cached marker to empty
+ m = stbi__get_marker(z);
+ if (!stbi__SOI(m)) return stbi__err("no SOI","Corrupt JPEG");
+ if (scan == STBI__SCAN_type) return 1;
+ m = stbi__get_marker(z);
+ while (!stbi__SOF(m)) {
+ if (!stbi__process_marker(z,m)) return 0;
+ m = stbi__get_marker(z);
+ while (m == STBI__MARKER_none) {
+ // some files have extra padding after their blocks, so ok, we'll scan
+ if (stbi__at_eof(z->s)) return stbi__err("no SOF", "Corrupt JPEG");
+ m = stbi__get_marker(z);
+ }
+ }
+ z->progressive = stbi__SOF_progressive(m);
+ if (!stbi__process_frame_header(z, scan)) return 0;
+ return 1;
+}
+
+// decode image to YCbCr format
+static int stbi__decode_jpeg_image(stbi__jpeg *j)
+{
+ int m;
+ for (m = 0; m < 4; m++) {
+ j->img_comp[m].raw_data = NULL;
+ j->img_comp[m].raw_coeff = NULL;
+ }
+ j->restart_interval = 0;
+ if (!stbi__decode_jpeg_header(j, STBI__SCAN_load)) return 0;
+ m = stbi__get_marker(j);
+ while (!stbi__EOI(m)) {
+ if (stbi__SOS(m)) {
+ if (!stbi__process_scan_header(j)) return 0;
+ if (!stbi__parse_entropy_coded_data(j)) return 0;
+ if (j->marker == STBI__MARKER_none ) {
+ // handle 0s at the end of image data from IP Kamera 9060
+ while (!stbi__at_eof(j->s)) {
+ int x = stbi__get8(j->s);
+ if (x == 255) {
+ j->marker = stbi__get8(j->s);
+ break;
+ }
+ }
+ // if we reach eof without hitting a marker, stbi__get_marker() below will fail and we'll eventually return 0
+ }
+ } else if (stbi__DNL(m)) {
+ int Ld = stbi__get16be(j->s);
+ stbi__uint32 NL = stbi__get16be(j->s);
+ if (Ld != 4) stbi__err("bad DNL len", "Corrupt JPEG");
+ if (NL != j->s->img_y) stbi__err("bad DNL height", "Corrupt JPEG");
+ } else {
+ if (!stbi__process_marker(j, m)) return 0;
+ }
+ m = stbi__get_marker(j);
+ }
+ if (j->progressive)
+ stbi__jpeg_finish(j);
+ return 1;
+}
+
+// static jfif-centered resampling (across block boundaries)
+
+typedef stbi_uc *(*resample_row_func)(stbi_uc *out, stbi_uc *in0, stbi_uc *in1,
+ int w, int hs);
+
+#define stbi__div4(x) ((stbi_uc) ((x) >> 2))
+
+static stbi_uc *resample_row_1(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+{
+ STBI_NOTUSED(out);
+ STBI_NOTUSED(in_far);
+ STBI_NOTUSED(w);
+ STBI_NOTUSED(hs);
+ return in_near;
+}
+
+static stbi_uc* stbi__resample_row_v_2(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+{
+ // need to generate two samples vertically for every one in input
+ int i;
+ STBI_NOTUSED(hs);
+ for (i=0; i < w; ++i)
+ out[i] = stbi__div4(3*in_near[i] + in_far[i] + 2);
+ return out;
+}
+
+static stbi_uc* stbi__resample_row_h_2(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+{
+ // need to generate two samples horizontally for every one in input
+ int i;
+ stbi_uc *input = in_near;
+
+ if (w == 1) {
+ // if only one sample, can't do any interpolation
+ out[0] = out[1] = input[0];
+ return out;
+ }
+
+ out[0] = input[0];
+ out[1] = stbi__div4(input[0]*3 + input[1] + 2);
+ for (i=1; i < w-1; ++i) {
+ int n = 3*input[i]+2;
+ out[i*2+0] = stbi__div4(n+input[i-1]);
+ out[i*2+1] = stbi__div4(n+input[i+1]);
+ }
+ out[i*2+0] = stbi__div4(input[w-2]*3 + input[w-1] + 2);
+ out[i*2+1] = input[w-1];
+
+ STBI_NOTUSED(in_far);
+ STBI_NOTUSED(hs);
+
+ return out;
+}
+
+#define stbi__div16(x) ((stbi_uc) ((x) >> 4))
+
+static stbi_uc *stbi__resample_row_hv_2(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+{
+ // need to generate 2x2 samples for every one in input
+ int i,t0,t1;
+ if (w == 1) {
+ out[0] = out[1] = stbi__div4(3*in_near[0] + in_far[0] + 2);
+ return out;
+ }
+
+ t1 = 3*in_near[0] + in_far[0];
+ out[0] = stbi__div4(t1+2);
+ for (i=1; i < w; ++i) {
+ t0 = t1;
+ t1 = 3*in_near[i]+in_far[i];
+ out[i*2-1] = stbi__div16(3*t0 + t1 + 8);
+ out[i*2 ] = stbi__div16(3*t1 + t0 + 8);
+ }
+ out[w*2-1] = stbi__div4(t1+2);
+
+ STBI_NOTUSED(hs);
+
+ return out;
+}
+
+#if defined(STBI_SSE2) || defined(STBI_NEON)
+static stbi_uc *stbi__resample_row_hv_2_simd(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+{
+ // need to generate 2x2 samples for every one in input
+ int i=0,t0,t1;
+
+ if (w == 1) {
+ out[0] = out[1] = stbi__div4(3*in_near[0] + in_far[0] + 2);
+ return out;
+ }
+
+ t1 = 3*in_near[0] + in_far[0];
+ // process groups of 8 pixels for as long as we can.
+ // note we can't handle the last pixel in a row in this loop
+ // because we need to handle the filter boundary conditions.
+ for (; i < ((w-1) & ~7); i += 8) {
+#if defined(STBI_SSE2)
+ // load and perform the vertical filtering pass
+ // this uses 3*x + y = 4*x + (y - x)
+ __m128i zero = _mm_setzero_si128();
+ __m128i farb = _mm_loadl_epi64((__m128i *) (in_far + i));
+ __m128i nearb = _mm_loadl_epi64((__m128i *) (in_near + i));
+ __m128i farw = _mm_unpacklo_epi8(farb, zero);
+ __m128i nearw = _mm_unpacklo_epi8(nearb, zero);
+ __m128i diff = _mm_sub_epi16(farw, nearw);
+ __m128i nears = _mm_slli_epi16(nearw, 2);
+ __m128i curr = _mm_add_epi16(nears, diff); // current row
+
+ // horizontal filter works the same based on shifted vers of current
+ // row. "prev" is current row shifted right by 1 pixel; we need to
+ // insert the previous pixel value (from t1).
+ // "next" is current row shifted left by 1 pixel, with first pixel
+ // of next block of 8 pixels added in.
+ __m128i prv0 = _mm_slli_si128(curr, 2);
+ __m128i nxt0 = _mm_srli_si128(curr, 2);
+ __m128i prev = _mm_insert_epi16(prv0, t1, 0);
+ __m128i next = _mm_insert_epi16(nxt0, 3*in_near[i+8] + in_far[i+8], 7);
+
+ // horizontal filter, polyphase implementation since it's convenient:
+ // even pixels = 3*cur + prev = cur*4 + (prev - cur)
+ // odd pixels = 3*cur + next = cur*4 + (next - cur)
+ // note the shared term.
+ __m128i bias = _mm_set1_epi16(8);
+ __m128i curs = _mm_slli_epi16(curr, 2);
+ __m128i prvd = _mm_sub_epi16(prev, curr);
+ __m128i nxtd = _mm_sub_epi16(next, curr);
+ __m128i curb = _mm_add_epi16(curs, bias);
+ __m128i even = _mm_add_epi16(prvd, curb);
+ __m128i odd = _mm_add_epi16(nxtd, curb);
+
+ // interleave even and odd pixels, then undo scaling.
+ __m128i int0 = _mm_unpacklo_epi16(even, odd);
+ __m128i int1 = _mm_unpackhi_epi16(even, odd);
+ __m128i de0 = _mm_srli_epi16(int0, 4);
+ __m128i de1 = _mm_srli_epi16(int1, 4);
+
+ // pack and write output
+ __m128i outv = _mm_packus_epi16(de0, de1);
+ _mm_storeu_si128((__m128i *) (out + i*2), outv);
+#elif defined(STBI_NEON)
+ // load and perform the vertical filtering pass
+ // this uses 3*x + y = 4*x + (y - x)
+ uint8x8_t farb = vld1_u8(in_far + i);
+ uint8x8_t nearb = vld1_u8(in_near + i);
+ int16x8_t diff = vreinterpretq_s16_u16(vsubl_u8(farb, nearb));
+ int16x8_t nears = vreinterpretq_s16_u16(vshll_n_u8(nearb, 2));
+ int16x8_t curr = vaddq_s16(nears, diff); // current row
+
+ // horizontal filter works the same based on shifted vers of current
+ // row. "prev" is current row shifted right by 1 pixel; we need to
+ // insert the previous pixel value (from t1).
+ // "next" is current row shifted left by 1 pixel, with first pixel
+ // of next block of 8 pixels added in.
+ int16x8_t prv0 = vextq_s16(curr, curr, 7);
+ int16x8_t nxt0 = vextq_s16(curr, curr, 1);
+ int16x8_t prev = vsetq_lane_s16(t1, prv0, 0);
+ int16x8_t next = vsetq_lane_s16(3*in_near[i+8] + in_far[i+8], nxt0, 7);
+
+ // horizontal filter, polyphase implementation since it's convenient:
+ // even pixels = 3*cur + prev = cur*4 + (prev - cur)
+ // odd pixels = 3*cur + next = cur*4 + (next - cur)
+ // note the shared term.
+ int16x8_t curs = vshlq_n_s16(curr, 2);
+ int16x8_t prvd = vsubq_s16(prev, curr);
+ int16x8_t nxtd = vsubq_s16(next, curr);
+ int16x8_t even = vaddq_s16(curs, prvd);
+ int16x8_t odd = vaddq_s16(curs, nxtd);
+
+ // undo scaling and round, then store with even/odd phases interleaved
+ uint8x8x2_t o;
+ o.val[0] = vqrshrun_n_s16(even, 4);
+ o.val[1] = vqrshrun_n_s16(odd, 4);
+ vst2_u8(out + i*2, o);
+#endif
+
+ // "previous" value for next iter
+ t1 = 3*in_near[i+7] + in_far[i+7];
+ }
+
+ t0 = t1;
+ t1 = 3*in_near[i] + in_far[i];
+ out[i*2] = stbi__div16(3*t1 + t0 + 8);
+
+ for (++i; i < w; ++i) {
+ t0 = t1;
+ t1 = 3*in_near[i]+in_far[i];
+ out[i*2-1] = stbi__div16(3*t0 + t1 + 8);
+ out[i*2 ] = stbi__div16(3*t1 + t0 + 8);
+ }
+ out[w*2-1] = stbi__div4(t1+2);
+
+ STBI_NOTUSED(hs);
+
+ return out;
+}
+#endif
+
+static stbi_uc *stbi__resample_row_generic(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+{
+ // resample with nearest-neighbor
+ int i,j;
+ STBI_NOTUSED(in_far);
+ for (i=0; i < w; ++i)
+ for (j=0; j < hs; ++j)
+ out[i*hs+j] = in_near[i];
+ return out;
+}
+
+// this is a reduced-precision calculation of YCbCr-to-RGB introduced
+// to make sure the code produces the same results in both SIMD and scalar
+#define stbi__float2fixed(x) (((int) ((x) * 4096.0f + 0.5f)) << 8)
+static void stbi__YCbCr_to_RGB_row(stbi_uc *out, const stbi_uc *y, const stbi_uc *pcb, const stbi_uc *pcr, int count, int step)
+{
+ int i;
+ for (i=0; i < count; ++i) {
+ int y_fixed = (y[i] << 20) + (1<<19); // rounding
+ int r,g,b;
+ int cr = pcr[i] - 128;
+ int cb = pcb[i] - 128;
+ r = y_fixed + cr* stbi__float2fixed(1.40200f);
+ g = y_fixed + (cr*-stbi__float2fixed(0.71414f)) + ((cb*-stbi__float2fixed(0.34414f)) & 0xffff0000);
+ b = y_fixed + cb* stbi__float2fixed(1.77200f);
+ r >>= 20;
+ g >>= 20;
+ b >>= 20;
+ if ((unsigned) r > 255) { if (r < 0) r = 0; else r = 255; }
+ if ((unsigned) g > 255) { if (g < 0) g = 0; else g = 255; }
+ if ((unsigned) b > 255) { if (b < 0) b = 0; else b = 255; }
+ out[0] = (stbi_uc)r;
+ out[1] = (stbi_uc)g;
+ out[2] = (stbi_uc)b;
+ out[3] = 255;
+ out += step;
+ }
+}
+
+#if defined(STBI_SSE2) || defined(STBI_NEON)
+static void stbi__YCbCr_to_RGB_simd(stbi_uc *out, stbi_uc const *y, stbi_uc const *pcb, stbi_uc const *pcr, int count, int step)
+{
+ int i = 0;
+
+#ifdef STBI_SSE2
+ // step == 3 is pretty ugly on the final interleave, and i'm not convinced
+ // it's useful in practice (you wouldn't use it for textures, for example).
+ // so just accelerate step == 4 case.
+ if (step == 4) {
+ // this is a fairly straightforward implementation and not super-optimized.
+ __m128i signflip = _mm_set1_epi8(-0x80);
+ __m128i cr_const0 = _mm_set1_epi16( (short) ( 1.40200f*4096.0f+0.5f));
+ __m128i cr_const1 = _mm_set1_epi16( - (short) ( 0.71414f*4096.0f+0.5f));
+ __m128i cb_const0 = _mm_set1_epi16( - (short) ( 0.34414f*4096.0f+0.5f));
+ __m128i cb_const1 = _mm_set1_epi16( (short) ( 1.77200f*4096.0f+0.5f));
+ __m128i y_bias = _mm_set1_epi8((char) (unsigned char) 128);
+ __m128i xw = _mm_set1_epi16(255); // alpha channel
+
+ for (; i+7 < count; i += 8) {
+ // load
+ __m128i y_bytes = _mm_loadl_epi64((__m128i *) (y+i));
+ __m128i cr_bytes = _mm_loadl_epi64((__m128i *) (pcr+i));
+ __m128i cb_bytes = _mm_loadl_epi64((__m128i *) (pcb+i));
+ __m128i cr_biased = _mm_xor_si128(cr_bytes, signflip); // -128
+ __m128i cb_biased = _mm_xor_si128(cb_bytes, signflip); // -128
+
+ // unpack to short (and left-shift cr, cb by 8)
+ __m128i yw = _mm_unpacklo_epi8(y_bias, y_bytes);
+ __m128i crw = _mm_unpacklo_epi8(_mm_setzero_si128(), cr_biased);
+ __m128i cbw = _mm_unpacklo_epi8(_mm_setzero_si128(), cb_biased);
+
+ // color transform
+ __m128i yws = _mm_srli_epi16(yw, 4);
+ __m128i cr0 = _mm_mulhi_epi16(cr_const0, crw);
+ __m128i cb0 = _mm_mulhi_epi16(cb_const0, cbw);
+ __m128i cb1 = _mm_mulhi_epi16(cbw, cb_const1);
+ __m128i cr1 = _mm_mulhi_epi16(crw, cr_const1);
+ __m128i rws = _mm_add_epi16(cr0, yws);
+ __m128i gwt = _mm_add_epi16(cb0, yws);
+ __m128i bws = _mm_add_epi16(yws, cb1);
+ __m128i gws = _mm_add_epi16(gwt, cr1);
+
+ // descale
+ __m128i rw = _mm_srai_epi16(rws, 4);
+ __m128i bw = _mm_srai_epi16(bws, 4);
+ __m128i gw = _mm_srai_epi16(gws, 4);
+
+ // back to byte, set up for transpose
+ __m128i brb = _mm_packus_epi16(rw, bw);
+ __m128i gxb = _mm_packus_epi16(gw, xw);
+
+ // transpose to interleave channels
+ __m128i t0 = _mm_unpacklo_epi8(brb, gxb);
+ __m128i t1 = _mm_unpackhi_epi8(brb, gxb);
+ __m128i o0 = _mm_unpacklo_epi16(t0, t1);
+ __m128i o1 = _mm_unpackhi_epi16(t0, t1);
+
+ // store
+ _mm_storeu_si128((__m128i *) (out + 0), o0);
+ _mm_storeu_si128((__m128i *) (out + 16), o1);
+ out += 32;
+ }
+ }
+#endif
+
+#ifdef STBI_NEON
+ // in this version, step=3 support would be easy to add. but is there demand?
+ if (step == 4) {
+ // this is a fairly straightforward implementation and not super-optimized.
+ uint8x8_t signflip = vdup_n_u8(0x80);
+ int16x8_t cr_const0 = vdupq_n_s16( (short) ( 1.40200f*4096.0f+0.5f));
+ int16x8_t cr_const1 = vdupq_n_s16( - (short) ( 0.71414f*4096.0f+0.5f));
+ int16x8_t cb_const0 = vdupq_n_s16( - (short) ( 0.34414f*4096.0f+0.5f));
+ int16x8_t cb_const1 = vdupq_n_s16( (short) ( 1.77200f*4096.0f+0.5f));
+
+ for (; i+7 < count; i += 8) {
+ // load
+ uint8x8_t y_bytes = vld1_u8(y + i);
+ uint8x8_t cr_bytes = vld1_u8(pcr + i);
+ uint8x8_t cb_bytes = vld1_u8(pcb + i);
+ int8x8_t cr_biased = vreinterpret_s8_u8(vsub_u8(cr_bytes, signflip));
+ int8x8_t cb_biased = vreinterpret_s8_u8(vsub_u8(cb_bytes, signflip));
+
+ // expand to s16
+ int16x8_t yws = vreinterpretq_s16_u16(vshll_n_u8(y_bytes, 4));
+ int16x8_t crw = vshll_n_s8(cr_biased, 7);
+ int16x8_t cbw = vshll_n_s8(cb_biased, 7);
+
+ // color transform
+ int16x8_t cr0 = vqdmulhq_s16(crw, cr_const0);
+ int16x8_t cb0 = vqdmulhq_s16(cbw, cb_const0);
+ int16x8_t cr1 = vqdmulhq_s16(crw, cr_const1);
+ int16x8_t cb1 = vqdmulhq_s16(cbw, cb_const1);
+ int16x8_t rws = vaddq_s16(yws, cr0);
+ int16x8_t gws = vaddq_s16(vaddq_s16(yws, cb0), cr1);
+ int16x8_t bws = vaddq_s16(yws, cb1);
+
+ // undo scaling, round, convert to byte
+ uint8x8x4_t o;
+ o.val[0] = vqrshrun_n_s16(rws, 4);
+ o.val[1] = vqrshrun_n_s16(gws, 4);
+ o.val[2] = vqrshrun_n_s16(bws, 4);
+ o.val[3] = vdup_n_u8(255);
+
+ // store, interleaving r/g/b/a
+ vst4_u8(out, o);
+ out += 8*4;
+ }
+ }
+#endif
+
+ for (; i < count; ++i) {
+ int y_fixed = (y[i] << 20) + (1<<19); // rounding
+ int r,g,b;
+ int cr = pcr[i] - 128;
+ int cb = pcb[i] - 128;
+ r = y_fixed + cr* stbi__float2fixed(1.40200f);
+ g = y_fixed + cr*-stbi__float2fixed(0.71414f) + ((cb*-stbi__float2fixed(0.34414f)) & 0xffff0000);
+ b = y_fixed + cb* stbi__float2fixed(1.77200f);
+ r >>= 20;
+ g >>= 20;
+ b >>= 20;
+ if ((unsigned) r > 255) { if (r < 0) r = 0; else r = 255; }
+ if ((unsigned) g > 255) { if (g < 0) g = 0; else g = 255; }
+ if ((unsigned) b > 255) { if (b < 0) b = 0; else b = 255; }
+ out[0] = (stbi_uc)r;
+ out[1] = (stbi_uc)g;
+ out[2] = (stbi_uc)b;
+ out[3] = 255;
+ out += step;
+ }
+}
+#endif
+
+// set up the kernels
+static void stbi__setup_jpeg(stbi__jpeg *j)
+{
+ j->idct_block_kernel = stbi__idct_block;
+ j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_row;
+ j->resample_row_hv_2_kernel = stbi__resample_row_hv_2;
+
+#ifdef STBI_SSE2
+ if (stbi__sse2_available()) {
+ j->idct_block_kernel = stbi__idct_simd;
+ j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_simd;
+ j->resample_row_hv_2_kernel = stbi__resample_row_hv_2_simd;
+ }
+#endif
+
+#ifdef STBI_NEON
+ j->idct_block_kernel = stbi__idct_simd;
+ j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_simd;
+ j->resample_row_hv_2_kernel = stbi__resample_row_hv_2_simd;
+#endif
+}
+
+// clean up the temporary component buffers
+static void stbi__cleanup_jpeg(stbi__jpeg *j)
+{
+ stbi__free_jpeg_components(j, j->s->img_n, 0);
+}
+
+typedef struct
+{
+ resample_row_func resample;
+ stbi_uc *line0,*line1;
+ int hs,vs; // expansion factor in each axis
+ int w_lores; // horizontal pixels pre-expansion
+ int ystep; // how far through vertical expansion we are
+ int ypos; // which pre-expansion row we're on
+} stbi__resample;
+
+// fast 0..255 * 0..255 => 0..255 rounded multiplication
+static stbi_uc stbi__blinn_8x8(stbi_uc x, stbi_uc y)
+{
+ unsigned int t = x*y + 128;
+ return (stbi_uc) ((t + (t >>8)) >> 8);
+}
+
+static stbi_uc *load_jpeg_image(stbi__jpeg *z, int *out_x, int *out_y, int *comp, int req_comp)
+{
+ int n, decode_n, is_rgb;
+ z->s->img_n = 0; // make stbi__cleanup_jpeg safe
+
+ // validate req_comp
+ if (req_comp < 0 || req_comp > 4) return stbi__errpuc("bad req_comp", "Internal error");
+
+ // load a jpeg image from whichever source, but leave in YCbCr format
+ if (!stbi__decode_jpeg_image(z)) { stbi__cleanup_jpeg(z); return NULL; }
+
+ // determine actual number of components to generate
+ n = req_comp ? req_comp : z->s->img_n >= 3 ? 3 : 1;
+
+ is_rgb = z->s->img_n == 3 && (z->rgb == 3 || (z->app14_color_transform == 0 && !z->jfif));
+
+ if (z->s->img_n == 3 && n < 3 && !is_rgb)
+ decode_n = 1;
+ else
+ decode_n = z->s->img_n;
+
+ // resample and color-convert
+ {
+ int k;
+ unsigned int i,j;
+ stbi_uc *output;
+ stbi_uc *coutput[4];
+
+ stbi__resample res_comp[4];
+
+ for (k=0; k < decode_n; ++k) {
+ stbi__resample *r = &res_comp[k];
+
+ // allocate line buffer big enough for upsampling off the edges
+ // with upsample factor of 4
+ z->img_comp[k].linebuf = (stbi_uc *) stbi__malloc(z->s->img_x + 3);
+ if (!z->img_comp[k].linebuf) { stbi__cleanup_jpeg(z); return stbi__errpuc("outofmem", "Out of memory"); }
+
+ r->hs = z->img_h_max / z->img_comp[k].h;
+ r->vs = z->img_v_max / z->img_comp[k].v;
+ r->ystep = r->vs >> 1;
+ r->w_lores = (z->s->img_x + r->hs-1) / r->hs;
+ r->ypos = 0;
+ r->line0 = r->line1 = z->img_comp[k].data;
+
+ if (r->hs == 1 && r->vs == 1) r->resample = resample_row_1;
+ else if (r->hs == 1 && r->vs == 2) r->resample = stbi__resample_row_v_2;
+ else if (r->hs == 2 && r->vs == 1) r->resample = stbi__resample_row_h_2;
+ else if (r->hs == 2 && r->vs == 2) r->resample = z->resample_row_hv_2_kernel;
+ else r->resample = stbi__resample_row_generic;
+ }
+
+ // can't error after this so, this is safe
+ output = (stbi_uc *) stbi__malloc_mad3(n, z->s->img_x, z->s->img_y, 1);
+ if (!output) { stbi__cleanup_jpeg(z); return stbi__errpuc("outofmem", "Out of memory"); }
+
+ // now go ahead and resample
+ for (j=0; j < z->s->img_y; ++j) {
+ stbi_uc *out = output + n * z->s->img_x * j;
+ for (k=0; k < decode_n; ++k) {
+ stbi__resample *r = &res_comp[k];
+ int y_bot = r->ystep >= (r->vs >> 1);
+ coutput[k] = r->resample(z->img_comp[k].linebuf,
+ y_bot ? r->line1 : r->line0,
+ y_bot ? r->line0 : r->line1,
+ r->w_lores, r->hs);
+ if (++r->ystep >= r->vs) {
+ r->ystep = 0;
+ r->line0 = r->line1;
+ if (++r->ypos < z->img_comp[k].y)
+ r->line1 += z->img_comp[k].w2;
+ }
+ }
+ if (n >= 3) {
+ stbi_uc *y = coutput[0];
+ if (z->s->img_n == 3) {
+ if (is_rgb) {
+ for (i=0; i < z->s->img_x; ++i) {
+ out[0] = y[i];
+ out[1] = coutput[1][i];
+ out[2] = coutput[2][i];
+ out[3] = 255;
+ out += n;
+ }
+ } else {
+ z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n);
+ }
+ } else if (z->s->img_n == 4) {
+ if (z->app14_color_transform == 0) { // CMYK
+ for (i=0; i < z->s->img_x; ++i) {
+ stbi_uc k = coutput[3][i];
+ out[0] = stbi__blinn_8x8(coutput[0][i], k);
+ out[1] = stbi__blinn_8x8(coutput[1][i], k);
+ out[2] = stbi__blinn_8x8(coutput[2][i], k);
+ out[3] = 255;
+ out += n;
+ }
+ } else if (z->app14_color_transform == 2) { // YCCK
+ z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n);
+ for (i=0; i < z->s->img_x; ++i) {
+ stbi_uc k = coutput[3][i];
+ out[0] = stbi__blinn_8x8(255 - out[0], k);
+ out[1] = stbi__blinn_8x8(255 - out[1], k);
+ out[2] = stbi__blinn_8x8(255 - out[2], k);
+ out += n;
+ }
+ } else { // YCbCr + alpha? Ignore the fourth channel for now
+ z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n);
+ }
+ } else
+ for (i=0; i < z->s->img_x; ++i) {
+ out[0] = out[1] = out[2] = y[i];
+ out[3] = 255; // not used if n==3
+ out += n;
+ }
+ } else {
+ if (is_rgb) {
+ if (n == 1)
+ for (i=0; i < z->s->img_x; ++i)
+ *out++ = stbi__compute_y(coutput[0][i], coutput[1][i], coutput[2][i]);
+ else {
+ for (i=0; i < z->s->img_x; ++i, out += 2) {
+ out[0] = stbi__compute_y(coutput[0][i], coutput[1][i], coutput[2][i]);
+ out[1] = 255;
+ }
+ }
+ } else if (z->s->img_n == 4 && z->app14_color_transform == 0) {
+ for (i=0; i < z->s->img_x; ++i) {
+ stbi_uc k = coutput[3][i];
+ stbi_uc r = stbi__blinn_8x8(coutput[0][i], k);
+ stbi_uc g = stbi__blinn_8x8(coutput[1][i], k);
+ stbi_uc b = stbi__blinn_8x8(coutput[2][i], k);
+ out[0] = stbi__compute_y(r, g, b);
+ out[1] = 255;
+ out += n;
+ }
+ } else if (z->s->img_n == 4 && z->app14_color_transform == 2) {
+ for (i=0; i < z->s->img_x; ++i) {
+ out[0] = stbi__blinn_8x8(255 - coutput[0][i], coutput[3][i]);
+ out[1] = 255;
+ out += n;
+ }
+ } else {
+ stbi_uc *y = coutput[0];
+ if (n == 1)
+ for (i=0; i < z->s->img_x; ++i) out[i] = y[i];
+ else
+ for (i=0; i < z->s->img_x; ++i) *out++ = y[i], *out++ = 255;
+ }
+ }
+ }
+ stbi__cleanup_jpeg(z);
+ *out_x = z->s->img_x;
+ *out_y = z->s->img_y;
+ if (comp) *comp = z->s->img_n >= 3 ? 3 : 1; // report original components, not output
+ return output;
+ }
+}
+
+static void *stbi__jpeg_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+ unsigned char* result;
+ stbi__jpeg* j = (stbi__jpeg*) stbi__malloc(sizeof(stbi__jpeg));
+ STBI_NOTUSED(ri);
+ j->s = s;
+ stbi__setup_jpeg(j);
+ result = load_jpeg_image(j, x,y,comp,req_comp);
+ STBI_FREE(j);
+ return result;
+}
+
+static int stbi__jpeg_test(stbi__context *s)
+{
+ int r;
+ stbi__jpeg* j = (stbi__jpeg*)stbi__malloc(sizeof(stbi__jpeg));
+ j->s = s;
+ stbi__setup_jpeg(j);
+ r = stbi__decode_jpeg_header(j, STBI__SCAN_type);
+ stbi__rewind(s);
+ STBI_FREE(j);
+ return r;
+}
+
+static int stbi__jpeg_info_raw(stbi__jpeg *j, int *x, int *y, int *comp)
+{
+ if (!stbi__decode_jpeg_header(j, STBI__SCAN_header)) {
+ stbi__rewind( j->s );
+ return 0;
+ }
+ if (x) *x = j->s->img_x;
+ if (y) *y = j->s->img_y;
+ if (comp) *comp = j->s->img_n >= 3 ? 3 : 1;
+ return 1;
+}
+
+static int stbi__jpeg_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ int result;
+ stbi__jpeg* j = (stbi__jpeg*) (stbi__malloc(sizeof(stbi__jpeg)));
+ j->s = s;
+ result = stbi__jpeg_info_raw(j, x, y, comp);
+ STBI_FREE(j);
+ return result;
+}
+#endif
+
+// public domain zlib decode v0.2 Sean Barrett 2006-11-18
+// simple implementation
+// - all input must be provided in an upfront buffer
+// - all output is written to a single output buffer (can malloc/realloc)
+// performance
+// - fast huffman
+
+#ifndef STBI_NO_ZLIB
+
+// fast-way is faster to check than jpeg huffman, but slow way is slower
+#define STBI__ZFAST_BITS 9 // accelerate all cases in default tables
+#define STBI__ZFAST_MASK ((1 << STBI__ZFAST_BITS) - 1)
+
+// zlib-style huffman encoding
+// (jpegs packs from left, zlib from right, so can't share code)
+typedef struct
+{
+ stbi__uint16 fast[1 << STBI__ZFAST_BITS];
+ stbi__uint16 firstcode[16];
+ int maxcode[17];
+ stbi__uint16 firstsymbol[16];
+ stbi_uc size[288];
+ stbi__uint16 value[288];
+} stbi__zhuffman;
+
+stbi_inline static int stbi__bitreverse16(int n)
+{
+ n = ((n & 0xAAAA) >> 1) | ((n & 0x5555) << 1);
+ n = ((n & 0xCCCC) >> 2) | ((n & 0x3333) << 2);
+ n = ((n & 0xF0F0) >> 4) | ((n & 0x0F0F) << 4);
+ n = ((n & 0xFF00) >> 8) | ((n & 0x00FF) << 8);
+ return n;
+}
+
+stbi_inline static int stbi__bit_reverse(int v, int bits)
+{
+ STBI_ASSERT(bits <= 16);
+ // to bit reverse n bits, reverse 16 and shift
+ // e.g. 11 bits, bit reverse and shift away 5
+ return stbi__bitreverse16(v) >> (16-bits);
+}
+
+static int stbi__zbuild_huffman(stbi__zhuffman *z, const stbi_uc *sizelist, int num)
+{
+ int i,k=0;
+ int code, next_code[16], sizes[17];
+
+ // DEFLATE spec for generating codes
+ memset(sizes, 0, sizeof(sizes));
+ memset(z->fast, 0, sizeof(z->fast));
+ for (i=0; i < num; ++i)
+ ++sizes[sizelist[i]];
+ sizes[0] = 0;
+ for (i=1; i < 16; ++i)
+ if (sizes[i] > (1 << i))
+ return stbi__err("bad sizes", "Corrupt PNG");
+ code = 0;
+ for (i=1; i < 16; ++i) {
+ next_code[i] = code;
+ z->firstcode[i] = (stbi__uint16) code;
+ z->firstsymbol[i] = (stbi__uint16) k;
+ code = (code + sizes[i]);
+ if (sizes[i])
+ if (code-1 >= (1 << i)) return stbi__err("bad codelengths","Corrupt PNG");
+ z->maxcode[i] = code << (16-i); // preshift for inner loop
+ code <<= 1;
+ k += sizes[i];
+ }
+ z->maxcode[16] = 0x10000; // sentinel
+ for (i=0; i < num; ++i) {
+ int s = sizelist[i];
+ if (s) {
+ int c = next_code[s] - z->firstcode[s] + z->firstsymbol[s];
+ stbi__uint16 fastv = (stbi__uint16) ((s << 9) | i);
+ z->size [c] = (stbi_uc ) s;
+ z->value[c] = (stbi__uint16) i;
+ if (s <= STBI__ZFAST_BITS) {
+ int j = stbi__bit_reverse(next_code[s],s);
+ while (j < (1 << STBI__ZFAST_BITS)) {
+ z->fast[j] = fastv;
+ j += (1 << s);
+ }
+ }
+ ++next_code[s];
+ }
+ }
+ return 1;
+}
+
+// zlib-from-memory implementation for PNG reading
+// because PNG allows splitting the zlib stream arbitrarily,
+// and it's annoying structurally to have PNG call ZLIB call PNG,
+// we require PNG read all the IDATs and combine them into a single
+// memory buffer
+
+typedef struct
+{
+ stbi_uc *zbuffer, *zbuffer_end;
+ int num_bits;
+ stbi__uint32 code_buffer;
+
+ char *zout;
+ char *zout_start;
+ char *zout_end;
+ int z_expandable;
+
+ stbi__zhuffman z_length, z_distance;
+} stbi__zbuf;
+
+stbi_inline static stbi_uc stbi__zget8(stbi__zbuf *z)
+{
+ if (z->zbuffer >= z->zbuffer_end) return 0;
+ return *z->zbuffer++;
+}
+
+static void stbi__fill_bits(stbi__zbuf *z)
+{
+ do {
+ STBI_ASSERT(z->code_buffer < (1U << z->num_bits));
+ z->code_buffer |= (unsigned int) stbi__zget8(z) << z->num_bits;
+ z->num_bits += 8;
+ } while (z->num_bits <= 24);
+}
+
+stbi_inline static unsigned int stbi__zreceive(stbi__zbuf *z, int n)
+{
+ unsigned int k;
+ if (z->num_bits < n) stbi__fill_bits(z);
+ k = z->code_buffer & ((1 << n) - 1);
+ z->code_buffer >>= n;
+ z->num_bits -= n;
+ return k;
+}
+
+static int stbi__zhuffman_decode_slowpath(stbi__zbuf *a, stbi__zhuffman *z)
+{
+ int b,s,k;
+ // not resolved by fast table, so compute it the slow way
+ // use jpeg approach, which requires MSbits at top
+ k = stbi__bit_reverse(a->code_buffer, 16);
+ for (s=STBI__ZFAST_BITS+1; ; ++s)
+ if (k < z->maxcode[s])
+ break;
+ if (s == 16) return -1; // invalid code!
+ // code size is s, so:
+ b = (k >> (16-s)) - z->firstcode[s] + z->firstsymbol[s];
+ STBI_ASSERT(z->size[b] == s);
+ a->code_buffer >>= s;
+ a->num_bits -= s;
+ return z->value[b];
+}
+
+stbi_inline static int stbi__zhuffman_decode(stbi__zbuf *a, stbi__zhuffman *z)
+{
+ int b,s;
+ if (a->num_bits < 16) stbi__fill_bits(a);
+ b = z->fast[a->code_buffer & STBI__ZFAST_MASK];
+ if (b) {
+ s = b >> 9;
+ a->code_buffer >>= s;
+ a->num_bits -= s;
+ return b & 511;
+ }
+ return stbi__zhuffman_decode_slowpath(a, z);
+}
+
+static int stbi__zexpand(stbi__zbuf *z, char *zout, int n) // need to make room for n bytes
+{
+ char *q;
+ int cur, limit, old_limit;
+ z->zout = zout;
+ if (!z->z_expandable) return stbi__err("output buffer limit","Corrupt PNG");
+ cur = (int) (z->zout - z->zout_start);
+ limit = old_limit = (int) (z->zout_end - z->zout_start);
+ while (cur + n > limit)
+ limit *= 2;
+ q = (char *) STBI_REALLOC_SIZED(z->zout_start, old_limit, limit);
+ STBI_NOTUSED(old_limit);
+ if (q == NULL) return stbi__err("outofmem", "Out of memory");
+ z->zout_start = q;
+ z->zout = q + cur;
+ z->zout_end = q + limit;
+ return 1;
+}
+
+static int stbi__zlength_base[31] = {
+ 3,4,5,6,7,8,9,10,11,13,
+ 15,17,19,23,27,31,35,43,51,59,
+ 67,83,99,115,131,163,195,227,258,0,0 };
+
+static int stbi__zlength_extra[31]=
+{ 0,0,0,0,0,0,0,0,1,1,1,1,2,2,2,2,3,3,3,3,4,4,4,4,5,5,5,5,0,0,0 };
+
+static int stbi__zdist_base[32] = { 1,2,3,4,5,7,9,13,17,25,33,49,65,97,129,193,
+257,385,513,769,1025,1537,2049,3073,4097,6145,8193,12289,16385,24577,0,0};
+
+static int stbi__zdist_extra[32] =
+{ 0,0,0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,8,8,9,9,10,10,11,11,12,12,13,13};
+
+static int stbi__parse_huffman_block(stbi__zbuf *a)
+{
+ char *zout = a->zout;
+ for(;;) {
+ int z = stbi__zhuffman_decode(a, &a->z_length);
+ if (z < 256) {
+ if (z < 0) return stbi__err("bad huffman code","Corrupt PNG"); // error in huffman codes
+ if (zout >= a->zout_end) {
+ if (!stbi__zexpand(a, zout, 1)) return 0;
+ zout = a->zout;
+ }
+ *zout++ = (char) z;
+ } else {
+ stbi_uc *p;
+ int len,dist;
+ if (z == 256) {
+ a->zout = zout;
+ return 1;
+ }
+ z -= 257;
+ len = stbi__zlength_base[z];
+ if (stbi__zlength_extra[z]) len += stbi__zreceive(a, stbi__zlength_extra[z]);
+ z = stbi__zhuffman_decode(a, &a->z_distance);
+ if (z < 0) return stbi__err("bad huffman code","Corrupt PNG");
+ dist = stbi__zdist_base[z];
+ if (stbi__zdist_extra[z]) dist += stbi__zreceive(a, stbi__zdist_extra[z]);
+ if (zout - a->zout_start < dist) return stbi__err("bad dist","Corrupt PNG");
+ if (zout + len > a->zout_end) {
+ if (!stbi__zexpand(a, zout, len)) return 0;
+ zout = a->zout;
+ }
+ p = (stbi_uc *) (zout - dist);
+ if (dist == 1) { // run of one byte; common in images.
+ stbi_uc v = *p;
+ if (len) { do *zout++ = v; while (--len); }
+ } else {
+ if (len) { do *zout++ = *p++; while (--len); }
+ }
+ }
+ }
+}
+
+static int stbi__compute_huffman_codes(stbi__zbuf *a)
+{
+ static stbi_uc length_dezigzag[19] = { 16,17,18,0,8,7,9,6,10,5,11,4,12,3,13,2,14,1,15 };
+ stbi__zhuffman z_codelength;
+ stbi_uc lencodes[286+32+137];//padding for maximum single op
+ stbi_uc codelength_sizes[19];
+ int i,n;
+
+ int hlit = stbi__zreceive(a,5) + 257;
+ int hdist = stbi__zreceive(a,5) + 1;
+ int hclen = stbi__zreceive(a,4) + 4;
+ int ntot = hlit + hdist;
+
+ memset(codelength_sizes, 0, sizeof(codelength_sizes));
+ for (i=0; i < hclen; ++i) {
+ int s = stbi__zreceive(a,3);
+ codelength_sizes[length_dezigzag[i]] = (stbi_uc) s;
+ }
+ if (!stbi__zbuild_huffman(&z_codelength, codelength_sizes, 19)) return 0;
+
+ n = 0;
+ while (n < ntot) {
+ int c = stbi__zhuffman_decode(a, &z_codelength);
+ if (c < 0 || c >= 19) return stbi__err("bad codelengths", "Corrupt PNG");
+ if (c < 16)
+ lencodes[n++] = (stbi_uc) c;
+ else {
+ stbi_uc fill = 0;
+ if (c == 16) {
+ c = stbi__zreceive(a,2)+3;
+ if (n == 0) return stbi__err("bad codelengths", "Corrupt PNG");
+ fill = lencodes[n-1];
+ } else if (c == 17)
+ c = stbi__zreceive(a,3)+3;
+ else {
+ STBI_ASSERT(c == 18);
+ c = stbi__zreceive(a,7)+11;
+ }
+ if (ntot - n < c) return stbi__err("bad codelengths", "Corrupt PNG");
+ memset(lencodes+n, fill, c);
+ n += c;
+ }
+ }
+ if (n != ntot) return stbi__err("bad codelengths","Corrupt PNG");
+ if (!stbi__zbuild_huffman(&a->z_length, lencodes, hlit)) return 0;
+ if (!stbi__zbuild_huffman(&a->z_distance, lencodes+hlit, hdist)) return 0;
+ return 1;
+}
+
+static int stbi__parse_uncompressed_block(stbi__zbuf *a)
+{
+ stbi_uc header[4];
+ int len,nlen,k;
+ if (a->num_bits & 7)
+ stbi__zreceive(a, a->num_bits & 7); // discard
+ // drain the bit-packed data into header
+ k = 0;
+ while (a->num_bits > 0) {
+ header[k++] = (stbi_uc) (a->code_buffer & 255); // suppress MSVC run-time check
+ a->code_buffer >>= 8;
+ a->num_bits -= 8;
+ }
+ STBI_ASSERT(a->num_bits == 0);
+ // now fill header the normal way
+ while (k < 4)
+ header[k++] = stbi__zget8(a);
+ len = header[1] * 256 + header[0];
+ nlen = header[3] * 256 + header[2];
+ if (nlen != (len ^ 0xffff)) return stbi__err("zlib corrupt","Corrupt PNG");
+ if (a->zbuffer + len > a->zbuffer_end) return stbi__err("read past buffer","Corrupt PNG");
+ if (a->zout + len > a->zout_end)
+ if (!stbi__zexpand(a, a->zout, len)) return 0;
+ memcpy(a->zout, a->zbuffer, len);
+ a->zbuffer += len;
+ a->zout += len;
+ return 1;
+}
+
+static int stbi__parse_zlib_header(stbi__zbuf *a)
+{
+ int cmf = stbi__zget8(a);
+ int cm = cmf & 15;
+ /* int cinfo = cmf >> 4; */
+ int flg = stbi__zget8(a);
+ if ((cmf*256+flg) % 31 != 0) return stbi__err("bad zlib header","Corrupt PNG"); // zlib spec
+ if (flg & 32) return stbi__err("no preset dict","Corrupt PNG"); // preset dictionary not allowed in png
+ if (cm != 8) return stbi__err("bad compression","Corrupt PNG"); // DEFLATE required for png
+ // window = 1 << (8 + cinfo)... but who cares, we fully buffer output
+ return 1;
+}
+
+static const stbi_uc stbi__zdefault_length[288] =
+{
+ 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
+ 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
+ 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
+ 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
+ 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,
+ 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,
+ 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,
+ 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,
+ 7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7, 7,7,7,7,7,7,7,7,8,8,8,8,8,8,8,8
+};
+static const stbi_uc stbi__zdefault_distance[32] =
+{
+ 5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5
+};
+/*
+Init algorithm:
+{
+ int i; // use <= to match clearly with spec
+ for (i=0; i <= 143; ++i) stbi__zdefault_length[i] = 8;
+ for ( ; i <= 255; ++i) stbi__zdefault_length[i] = 9;
+ for ( ; i <= 279; ++i) stbi__zdefault_length[i] = 7;
+ for ( ; i <= 287; ++i) stbi__zdefault_length[i] = 8;
+
+ for (i=0; i <= 31; ++i) stbi__zdefault_distance[i] = 5;
+}
+*/
+
+static int stbi__parse_zlib(stbi__zbuf *a, int parse_header)
+{
+ int final, type;
+ if (parse_header)
+ if (!stbi__parse_zlib_header(a)) return 0;
+ a->num_bits = 0;
+ a->code_buffer = 0;
+ do {
+ final = stbi__zreceive(a,1);
+ type = stbi__zreceive(a,2);
+ if (type == 0) {
+ if (!stbi__parse_uncompressed_block(a)) return 0;
+ } else if (type == 3) {
+ return 0;
+ } else {
+ if (type == 1) {
+ // use fixed code lengths
+ if (!stbi__zbuild_huffman(&a->z_length , stbi__zdefault_length , 288)) return 0;
+ if (!stbi__zbuild_huffman(&a->z_distance, stbi__zdefault_distance, 32)) return 0;
+ } else {
+ if (!stbi__compute_huffman_codes(a)) return 0;
+ }
+ if (!stbi__parse_huffman_block(a)) return 0;
+ }
+ } while (!final);
+ return 1;
+}
+
+static int stbi__do_zlib(stbi__zbuf *a, char *obuf, int olen, int exp, int parse_header)
+{
+ a->zout_start = obuf;
+ a->zout = obuf;
+ a->zout_end = obuf + olen;
+ a->z_expandable = exp;
+
+ return stbi__parse_zlib(a, parse_header);
+}
+
+STBIDEF char *stbi_zlib_decode_malloc_guesssize(const char *buffer, int len, int initial_size, int *outlen)
+{
+ stbi__zbuf a;
+ char *p = (char *) stbi__malloc(initial_size);
+ if (p == NULL) return NULL;
+ a.zbuffer = (stbi_uc *) buffer;
+ a.zbuffer_end = (stbi_uc *) buffer + len;
+ if (stbi__do_zlib(&a, p, initial_size, 1, 1)) {
+ if (outlen) *outlen = (int) (a.zout - a.zout_start);
+ return a.zout_start;
+ } else {
+ STBI_FREE(a.zout_start);
+ return NULL;
+ }
+}
+
+STBIDEF char *stbi_zlib_decode_malloc(char const *buffer, int len, int *outlen)
+{
+ return stbi_zlib_decode_malloc_guesssize(buffer, len, 16384, outlen);
+}
+
+STBIDEF char *stbi_zlib_decode_malloc_guesssize_headerflag(const char *buffer, int len, int initial_size, int *outlen, int parse_header)
+{
+ stbi__zbuf a;
+ char *p = (char *) stbi__malloc(initial_size);
+ if (p == NULL) return NULL;
+ a.zbuffer = (stbi_uc *) buffer;
+ a.zbuffer_end = (stbi_uc *) buffer + len;
+ if (stbi__do_zlib(&a, p, initial_size, 1, parse_header)) {
+ if (outlen) *outlen = (int) (a.zout - a.zout_start);
+ return a.zout_start;
+ } else {
+ STBI_FREE(a.zout_start);
+ return NULL;
+ }
+}
+
+STBIDEF int stbi_zlib_decode_buffer(char *obuffer, int olen, char const *ibuffer, int ilen)
+{
+ stbi__zbuf a;
+ a.zbuffer = (stbi_uc *) ibuffer;
+ a.zbuffer_end = (stbi_uc *) ibuffer + ilen;
+ if (stbi__do_zlib(&a, obuffer, olen, 0, 1))
+ return (int) (a.zout - a.zout_start);
+ else
+ return -1;
+}
+
+STBIDEF char *stbi_zlib_decode_noheader_malloc(char const *buffer, int len, int *outlen)
+{
+ stbi__zbuf a;
+ char *p = (char *) stbi__malloc(16384);
+ if (p == NULL) return NULL;
+ a.zbuffer = (stbi_uc *) buffer;
+ a.zbuffer_end = (stbi_uc *) buffer+len;
+ if (stbi__do_zlib(&a, p, 16384, 1, 0)) {
+ if (outlen) *outlen = (int) (a.zout - a.zout_start);
+ return a.zout_start;
+ } else {
+ STBI_FREE(a.zout_start);
+ return NULL;
+ }
+}
+
+STBIDEF int stbi_zlib_decode_noheader_buffer(char *obuffer, int olen, const char *ibuffer, int ilen)
+{
+ stbi__zbuf a;
+ a.zbuffer = (stbi_uc *) ibuffer;
+ a.zbuffer_end = (stbi_uc *) ibuffer + ilen;
+ if (stbi__do_zlib(&a, obuffer, olen, 0, 0))
+ return (int) (a.zout - a.zout_start);
+ else
+ return -1;
+}
+#endif
+
+// public domain "baseline" PNG decoder v0.10 Sean Barrett 2006-11-18
+// simple implementation
+// - only 8-bit samples
+// - no CRC checking
+// - allocates lots of intermediate memory
+// - avoids problem of streaming data between subsystems
+// - avoids explicit window management
+// performance
+// - uses stb_zlib, a PD zlib implementation with fast huffman decoding
+
+#ifndef STBI_NO_PNG
+typedef struct
+{
+ stbi__uint32 length;
+ stbi__uint32 type;
+} stbi__pngchunk;
+
+static stbi__pngchunk stbi__get_chunk_header(stbi__context *s)
+{
+ stbi__pngchunk c;
+ c.length = stbi__get32be(s);
+ c.type = stbi__get32be(s);
+ return c;
+}
+
+static int stbi__check_png_header(stbi__context *s)
+{
+ static stbi_uc png_sig[8] = { 137,80,78,71,13,10,26,10 };
+ int i;
+ for (i=0; i < 8; ++i)
+ if (stbi__get8(s) != png_sig[i]) return stbi__err("bad png sig","Not a PNG");
+ return 1;
+}
+
+typedef struct
+{
+ stbi__context *s;
+ stbi_uc *idata, *expanded, *out;
+ int depth;
+} stbi__png;
+
+
+enum {
+ STBI__F_none=0,
+ STBI__F_sub=1,
+ STBI__F_up=2,
+ STBI__F_avg=3,
+ STBI__F_paeth=4,
+ // synthetic filters used for first scanline to avoid needing a dummy row of 0s
+ STBI__F_avg_first,
+ STBI__F_paeth_first
+};
+
+static stbi_uc first_row_filter[5] =
+{
+ STBI__F_none,
+ STBI__F_sub,
+ STBI__F_none,
+ STBI__F_avg_first,
+ STBI__F_paeth_first
+};
+
+static int stbi__paeth(int a, int b, int c)
+{
+ int p = a + b - c;
+ int pa = abs(p-a);
+ int pb = abs(p-b);
+ int pc = abs(p-c);
+ if (pa <= pb && pa <= pc) return a;
+ if (pb <= pc) return b;
+ return c;
+}
+
+static stbi_uc stbi__depth_scale_table[9] = { 0, 0xff, 0x55, 0, 0x11, 0,0,0, 0x01 };
+
+// create the png data from post-deflated data
+static int stbi__create_png_image_raw(stbi__png *a, stbi_uc *raw, stbi__uint32 raw_len, int out_n, stbi__uint32 x, stbi__uint32 y, int depth, int color)
+{
+ int bytes = (depth == 16? 2 : 1);
+ stbi__context *s = a->s;
+ stbi__uint32 i,j,stride = x*out_n*bytes;
+ stbi__uint32 img_len, img_width_bytes;
+ int k;
+ int img_n = s->img_n; // copy it into a local for later
+
+ int output_bytes = out_n*bytes;
+ int filter_bytes = img_n*bytes;
+ int width = x;
+
+ STBI_ASSERT(out_n == s->img_n || out_n == s->img_n+1);
+ a->out = (stbi_uc *) stbi__malloc_mad3(x, y, output_bytes, 0); // extra bytes to write off the end into
+ if (!a->out) return stbi__err("outofmem", "Out of memory");
+
+ img_width_bytes = (((img_n * x * depth) + 7) >> 3);
+ img_len = (img_width_bytes + 1) * y;
+ if (s->img_x == x && s->img_y == y) {
+ if (raw_len != img_len) return stbi__err("not enough pixels","Corrupt PNG");
+ } else { // interlaced:
+ if (raw_len < img_len) return stbi__err("not enough pixels","Corrupt PNG");
+ }
+
+ for (j=0; j < y; ++j) {
+ stbi_uc *cur = a->out + stride*j;
+ stbi_uc *prior;
+ int filter = *raw++;
+
+ if (filter > 4)
+ return stbi__err("invalid filter","Corrupt PNG");
+
+ if (depth < 8) {
+ STBI_ASSERT(img_width_bytes <= x);
+ cur += x*out_n - img_width_bytes; // store output to the rightmost img_len bytes, so we can decode in place
+ filter_bytes = 1;
+ width = img_width_bytes;
+ }
+ prior = cur - stride; // bugfix: need to compute this after 'cur +=' computation above
+
+ // if first row, use special filter that doesn't sample previous row
+ if (j == 0) filter = first_row_filter[filter];
+
+ // handle first byte explicitly
+ for (k=0; k < filter_bytes; ++k) {
+ switch (filter) {
+ case STBI__F_none : cur[k] = raw[k]; break;
+ case STBI__F_sub : cur[k] = raw[k]; break;
+ case STBI__F_up : cur[k] = STBI__BYTECAST(raw[k] + prior[k]); break;
+ case STBI__F_avg : cur[k] = STBI__BYTECAST(raw[k] + (prior[k]>>1)); break;
+ case STBI__F_paeth : cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(0,prior[k],0)); break;
+ case STBI__F_avg_first : cur[k] = raw[k]; break;
+ case STBI__F_paeth_first: cur[k] = raw[k]; break;
+ }
+ }
+
+ if (depth == 8) {
+ if (img_n != out_n)
+ cur[img_n] = 255; // first pixel
+ raw += img_n;
+ cur += out_n;
+ prior += out_n;
+ } else if (depth == 16) {
+ if (img_n != out_n) {
+ cur[filter_bytes] = 255; // first pixel top byte
+ cur[filter_bytes+1] = 255; // first pixel bottom byte
+ }
+ raw += filter_bytes;
+ cur += output_bytes;
+ prior += output_bytes;
+ } else {
+ raw += 1;
+ cur += 1;
+ prior += 1;
+ }
+
+ // this is a little gross, so that we don't switch per-pixel or per-component
+ if (depth < 8 || img_n == out_n) {
+ int nk = (width - 1)*filter_bytes;
+ #define STBI__CASE(f) \
+ case f: \
+ for (k=0; k < nk; ++k)
+ switch (filter) {
+ // "none" filter turns into a memcpy here; make that explicit.
+ case STBI__F_none: memcpy(cur, raw, nk); break;
+ STBI__CASE(STBI__F_sub) { cur[k] = STBI__BYTECAST(raw[k] + cur[k-filter_bytes]); } break;
+ STBI__CASE(STBI__F_up) { cur[k] = STBI__BYTECAST(raw[k] + prior[k]); } break;
+ STBI__CASE(STBI__F_avg) { cur[k] = STBI__BYTECAST(raw[k] + ((prior[k] + cur[k-filter_bytes])>>1)); } break;
+ STBI__CASE(STBI__F_paeth) { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k-filter_bytes],prior[k],prior[k-filter_bytes])); } break;
+ STBI__CASE(STBI__F_avg_first) { cur[k] = STBI__BYTECAST(raw[k] + (cur[k-filter_bytes] >> 1)); } break;
+ STBI__CASE(STBI__F_paeth_first) { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k-filter_bytes],0,0)); } break;
+ }
+ #undef STBI__CASE
+ raw += nk;
+ } else {
+ STBI_ASSERT(img_n+1 == out_n);
+ #define STBI__CASE(f) \
+ case f: \
+ for (i=x-1; i >= 1; --i, cur[filter_bytes]=255,raw+=filter_bytes,cur+=output_bytes,prior+=output_bytes) \
+ for (k=0; k < filter_bytes; ++k)
+ switch (filter) {
+ STBI__CASE(STBI__F_none) { cur[k] = raw[k]; } break;
+ STBI__CASE(STBI__F_sub) { cur[k] = STBI__BYTECAST(raw[k] + cur[k- output_bytes]); } break;
+ STBI__CASE(STBI__F_up) { cur[k] = STBI__BYTECAST(raw[k] + prior[k]); } break;
+ STBI__CASE(STBI__F_avg) { cur[k] = STBI__BYTECAST(raw[k] + ((prior[k] + cur[k- output_bytes])>>1)); } break;
+ STBI__CASE(STBI__F_paeth) { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k- output_bytes],prior[k],prior[k- output_bytes])); } break;
+ STBI__CASE(STBI__F_avg_first) { cur[k] = STBI__BYTECAST(raw[k] + (cur[k- output_bytes] >> 1)); } break;
+ STBI__CASE(STBI__F_paeth_first) { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k- output_bytes],0,0)); } break;
+ }
+ #undef STBI__CASE
+
+ // the loop above sets the high byte of the pixels' alpha, but for
+ // 16 bit png files we also need the low byte set. we'll do that here.
+ if (depth == 16) {
+ cur = a->out + stride*j; // start at the beginning of the row again
+ for (i=0; i < x; ++i,cur+=output_bytes) {
+ cur[filter_bytes+1] = 255;
+ }
+ }
+ }
+ }
+
+ // we make a separate pass to expand bits to pixels; for performance,
+ // this could run two scanlines behind the above code, so it won't
+ // intefere with filtering but will still be in the cache.
+ if (depth < 8) {
+ for (j=0; j < y; ++j) {
+ stbi_uc *cur = a->out + stride*j;
+ stbi_uc *in = a->out + stride*j + x*out_n - img_width_bytes;
+ // unpack 1/2/4-bit into a 8-bit buffer. allows us to keep the common 8-bit path optimal at minimal cost for 1/2/4-bit
+ // png guarante byte alignment, if width is not multiple of 8/4/2 we'll decode dummy trailing data that will be skipped in the later loop
+ stbi_uc scale = (color == 0) ? stbi__depth_scale_table[depth] : 1; // scale grayscale values to 0..255 range
+
+ // note that the final byte might overshoot and write more data than desired.
+ // we can allocate enough data that this never writes out of memory, but it
+ // could also overwrite the next scanline. can it overwrite non-empty data
+ // on the next scanline? yes, consider 1-pixel-wide scanlines with 1-bit-per-pixel.
+ // so we need to explicitly clamp the final ones
+
+ if (depth == 4) {
+ for (k=x*img_n; k >= 2; k-=2, ++in) {
+ *cur++ = scale * ((*in >> 4) );
+ *cur++ = scale * ((*in ) & 0x0f);
+ }
+ if (k > 0) *cur++ = scale * ((*in >> 4) );
+ } else if (depth == 2) {
+ for (k=x*img_n; k >= 4; k-=4, ++in) {
+ *cur++ = scale * ((*in >> 6) );
+ *cur++ = scale * ((*in >> 4) & 0x03);
+ *cur++ = scale * ((*in >> 2) & 0x03);
+ *cur++ = scale * ((*in ) & 0x03);
+ }
+ if (k > 0) *cur++ = scale * ((*in >> 6) );
+ if (k > 1) *cur++ = scale * ((*in >> 4) & 0x03);
+ if (k > 2) *cur++ = scale * ((*in >> 2) & 0x03);
+ } else if (depth == 1) {
+ for (k=x*img_n; k >= 8; k-=8, ++in) {
+ *cur++ = scale * ((*in >> 7) );
+ *cur++ = scale * ((*in >> 6) & 0x01);
+ *cur++ = scale * ((*in >> 5) & 0x01);
+ *cur++ = scale * ((*in >> 4) & 0x01);
+ *cur++ = scale * ((*in >> 3) & 0x01);
+ *cur++ = scale * ((*in >> 2) & 0x01);
+ *cur++ = scale * ((*in >> 1) & 0x01);
+ *cur++ = scale * ((*in ) & 0x01);
+ }
+ if (k > 0) *cur++ = scale * ((*in >> 7) );
+ if (k > 1) *cur++ = scale * ((*in >> 6) & 0x01);
+ if (k > 2) *cur++ = scale * ((*in >> 5) & 0x01);
+ if (k > 3) *cur++ = scale * ((*in >> 4) & 0x01);
+ if (k > 4) *cur++ = scale * ((*in >> 3) & 0x01);
+ if (k > 5) *cur++ = scale * ((*in >> 2) & 0x01);
+ if (k > 6) *cur++ = scale * ((*in >> 1) & 0x01);
+ }
+ if (img_n != out_n) {
+ int q;
+ // insert alpha = 255
+ cur = a->out + stride*j;
+ if (img_n == 1) {
+ for (q=x-1; q >= 0; --q) {
+ cur[q*2+1] = 255;
+ cur[q*2+0] = cur[q];
+ }
+ } else {
+ STBI_ASSERT(img_n == 3);
+ for (q=x-1; q >= 0; --q) {
+ cur[q*4+3] = 255;
+ cur[q*4+2] = cur[q*3+2];
+ cur[q*4+1] = cur[q*3+1];
+ cur[q*4+0] = cur[q*3+0];
+ }
+ }
+ }
+ }
+ } else if (depth == 16) {
+ // force the image data from big-endian to platform-native.
+ // this is done in a separate pass due to the decoding relying
+ // on the data being untouched, but could probably be done
+ // per-line during decode if care is taken.
+ stbi_uc *cur = a->out;
+ stbi__uint16 *cur16 = (stbi__uint16*)cur;
+
+ for(i=0; i < x*y*out_n; ++i,cur16++,cur+=2) {
+ *cur16 = (cur[0] << 8) | cur[1];
+ }
+ }
+
+ return 1;
+}
+
+static int stbi__create_png_image(stbi__png *a, stbi_uc *image_data, stbi__uint32 image_data_len, int out_n, int depth, int color, int interlaced)
+{
+ int bytes = (depth == 16 ? 2 : 1);
+ int out_bytes = out_n * bytes;
+ stbi_uc *final;
+ int p;
+ if (!interlaced)
+ return stbi__create_png_image_raw(a, image_data, image_data_len, out_n, a->s->img_x, a->s->img_y, depth, color);
+
+ // de-interlacing
+ final = (stbi_uc *) stbi__malloc_mad3(a->s->img_x, a->s->img_y, out_bytes, 0);
+ for (p=0; p < 7; ++p) {
+ int xorig[] = { 0,4,0,2,0,1,0 };
+ int yorig[] = { 0,0,4,0,2,0,1 };
+ int xspc[] = { 8,8,4,4,2,2,1 };
+ int yspc[] = { 8,8,8,4,4,2,2 };
+ int i,j,x,y;
+ // pass1_x[4] = 0, pass1_x[5] = 1, pass1_x[12] = 1
+ x = (a->s->img_x - xorig[p] + xspc[p]-1) / xspc[p];
+ y = (a->s->img_y - yorig[p] + yspc[p]-1) / yspc[p];
+ if (x && y) {
+ stbi__uint32 img_len = ((((a->s->img_n * x * depth) + 7) >> 3) + 1) * y;
+ if (!stbi__create_png_image_raw(a, image_data, image_data_len, out_n, x, y, depth, color)) {
+ STBI_FREE(final);
+ return 0;
+ }
+ for (j=0; j < y; ++j) {
+ for (i=0; i < x; ++i) {
+ int out_y = j*yspc[p]+yorig[p];
+ int out_x = i*xspc[p]+xorig[p];
+ memcpy(final + out_y*a->s->img_x*out_bytes + out_x*out_bytes,
+ a->out + (j*x+i)*out_bytes, out_bytes);
+ }
+ }
+ STBI_FREE(a->out);
+ image_data += img_len;
+ image_data_len -= img_len;
+ }
+ }
+ a->out = final;
+
+ return 1;
+}
+
+static int stbi__compute_transparency(stbi__png *z, stbi_uc tc[3], int out_n)
+{
+ stbi__context *s = z->s;
+ stbi__uint32 i, pixel_count = s->img_x * s->img_y;
+ stbi_uc *p = z->out;
+
+ // compute color-based transparency, assuming we've
+ // already got 255 as the alpha value in the output
+ STBI_ASSERT(out_n == 2 || out_n == 4);
+
+ if (out_n == 2) {
+ for (i=0; i < pixel_count; ++i) {
+ p[1] = (p[0] == tc[0] ? 0 : 255);
+ p += 2;
+ }
+ } else {
+ for (i=0; i < pixel_count; ++i) {
+ if (p[0] == tc[0] && p[1] == tc[1] && p[2] == tc[2])
+ p[3] = 0;
+ p += 4;
+ }
+ }
+ return 1;
+}
+
+static int stbi__compute_transparency16(stbi__png *z, stbi__uint16 tc[3], int out_n)
+{
+ stbi__context *s = z->s;
+ stbi__uint32 i, pixel_count = s->img_x * s->img_y;
+ stbi__uint16 *p = (stbi__uint16*) z->out;
+
+ // compute color-based transparency, assuming we've
+ // already got 65535 as the alpha value in the output
+ STBI_ASSERT(out_n == 2 || out_n == 4);
+
+ if (out_n == 2) {
+ for (i = 0; i < pixel_count; ++i) {
+ p[1] = (p[0] == tc[0] ? 0 : 65535);
+ p += 2;
+ }
+ } else {
+ for (i = 0; i < pixel_count; ++i) {
+ if (p[0] == tc[0] && p[1] == tc[1] && p[2] == tc[2])
+ p[3] = 0;
+ p += 4;
+ }
+ }
+ return 1;
+}
+
+static int stbi__expand_png_palette(stbi__png *a, stbi_uc *palette, int len, int pal_img_n)
+{
+ stbi__uint32 i, pixel_count = a->s->img_x * a->s->img_y;
+ stbi_uc *p, *temp_out, *orig = a->out;
+
+ p = (stbi_uc *) stbi__malloc_mad2(pixel_count, pal_img_n, 0);
+ if (p == NULL) return stbi__err("outofmem", "Out of memory");
+
+ // between here and free(out) below, exitting would leak
+ temp_out = p;
+
+ if (pal_img_n == 3) {
+ for (i=0; i < pixel_count; ++i) {
+ int n = orig[i]*4;
+ p[0] = palette[n ];
+ p[1] = palette[n+1];
+ p[2] = palette[n+2];
+ p += 3;
+ }
+ } else {
+ for (i=0; i < pixel_count; ++i) {
+ int n = orig[i]*4;
+ p[0] = palette[n ];
+ p[1] = palette[n+1];
+ p[2] = palette[n+2];
+ p[3] = palette[n+3];
+ p += 4;
+ }
+ }
+ STBI_FREE(a->out);
+ a->out = temp_out;
+
+ STBI_NOTUSED(len);
+
+ return 1;
+}
+
+static int stbi__unpremultiply_on_load = 0;
+static int stbi__de_iphone_flag = 0;
+
+STBIDEF void stbi_set_unpremultiply_on_load(int flag_true_if_should_unpremultiply)
+{
+ stbi__unpremultiply_on_load = flag_true_if_should_unpremultiply;
+}
+
+STBIDEF void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert)
+{
+ stbi__de_iphone_flag = flag_true_if_should_convert;
+}
+
+static void stbi__de_iphone(stbi__png *z)
+{
+ stbi__context *s = z->s;
+ stbi__uint32 i, pixel_count = s->img_x * s->img_y;
+ stbi_uc *p = z->out;
+
+ if (s->img_out_n == 3) { // convert bgr to rgb
+ for (i=0; i < pixel_count; ++i) {
+ stbi_uc t = p[0];
+ p[0] = p[2];
+ p[2] = t;
+ p += 3;
+ }
+ } else {
+ STBI_ASSERT(s->img_out_n == 4);
+ if (stbi__unpremultiply_on_load) {
+ // convert bgr to rgb and unpremultiply
+ for (i=0; i < pixel_count; ++i) {
+ stbi_uc a = p[3];
+ stbi_uc t = p[0];
+ if (a) {
+ p[0] = p[2] * 255 / a;
+ p[1] = p[1] * 255 / a;
+ p[2] = t * 255 / a;
+ } else {
+ p[0] = p[2];
+ p[2] = t;
+ }
+ p += 4;
+ }
+ } else {
+ // convert bgr to rgb
+ for (i=0; i < pixel_count; ++i) {
+ stbi_uc t = p[0];
+ p[0] = p[2];
+ p[2] = t;
+ p += 4;
+ }
+ }
+ }
+}
+
+#define STBI__PNG_TYPE(a,b,c,d) (((a) << 24) + ((b) << 16) + ((c) << 8) + (d))
+
+static int stbi__parse_png_file(stbi__png *z, int scan, int req_comp)
+{
+ stbi_uc palette[1024], pal_img_n=0;
+ stbi_uc has_trans=0, tc[3];
+ stbi__uint16 tc16[3];
+ stbi__uint32 ioff=0, idata_limit=0, i, pal_len=0;
+ int first=1,k,interlace=0, color=0, is_iphone=0;
+ stbi__context *s = z->s;
+
+ z->expanded = NULL;
+ z->idata = NULL;
+ z->out = NULL;
+
+ if (!stbi__check_png_header(s)) return 0;
+
+ if (scan == STBI__SCAN_type) return 1;
+
+ for (;;) {
+ stbi__pngchunk c = stbi__get_chunk_header(s);
+ switch (c.type) {
+ case STBI__PNG_TYPE('C','g','B','I'):
+ is_iphone = 1;
+ stbi__skip(s, c.length);
+ break;
+ case STBI__PNG_TYPE('I','H','D','R'): {
+ int comp,filter;
+ if (!first) return stbi__err("multiple IHDR","Corrupt PNG");
+ first = 0;
+ if (c.length != 13) return stbi__err("bad IHDR len","Corrupt PNG");
+ s->img_x = stbi__get32be(s); if (s->img_x > (1 << 24)) return stbi__err("too large","Very large image (corrupt?)");
+ s->img_y = stbi__get32be(s); if (s->img_y > (1 << 24)) return stbi__err("too large","Very large image (corrupt?)");
+ z->depth = stbi__get8(s); if (z->depth != 1 && z->depth != 2 && z->depth != 4 && z->depth != 8 && z->depth != 16) return stbi__err("1/2/4/8/16-bit only","PNG not supported: 1/2/4/8/16-bit only");
+ color = stbi__get8(s); if (color > 6) return stbi__err("bad ctype","Corrupt PNG");
+ if (color == 3 && z->depth == 16) return stbi__err("bad ctype","Corrupt PNG");
+ if (color == 3) pal_img_n = 3; else if (color & 1) return stbi__err("bad ctype","Corrupt PNG");
+ comp = stbi__get8(s); if (comp) return stbi__err("bad comp method","Corrupt PNG");
+ filter= stbi__get8(s); if (filter) return stbi__err("bad filter method","Corrupt PNG");
+ interlace = stbi__get8(s); if (interlace>1) return stbi__err("bad interlace method","Corrupt PNG");
+ if (!s->img_x || !s->img_y) return stbi__err("0-pixel image","Corrupt PNG");
+ if (!pal_img_n) {
+ s->img_n = (color & 2 ? 3 : 1) + (color & 4 ? 1 : 0);
+ if ((1 << 30) / s->img_x / s->img_n < s->img_y) return stbi__err("too large", "Image too large to decode");
+ if (scan == STBI__SCAN_header) return 1;
+ } else {
+ // if paletted, then pal_n is our final components, and
+ // img_n is # components to decompress/filter.
+ s->img_n = 1;
+ if ((1 << 30) / s->img_x / 4 < s->img_y) return stbi__err("too large","Corrupt PNG");
+ // if SCAN_header, have to scan to see if we have a tRNS
+ }
+ break;
+ }
+
+ case STBI__PNG_TYPE('P','L','T','E'): {
+ if (first) return stbi__err("first not IHDR", "Corrupt PNG");
+ if (c.length > 256*3) return stbi__err("invalid PLTE","Corrupt PNG");
+ pal_len = c.length / 3;
+ if (pal_len * 3 != c.length) return stbi__err("invalid PLTE","Corrupt PNG");
+ for (i=0; i < pal_len; ++i) {
+ palette[i*4+0] = stbi__get8(s);
+ palette[i*4+1] = stbi__get8(s);
+ palette[i*4+2] = stbi__get8(s);
+ palette[i*4+3] = 255;
+ }
+ break;
+ }
+
+ case STBI__PNG_TYPE('t','R','N','S'): {
+ if (first) return stbi__err("first not IHDR", "Corrupt PNG");
+ if (z->idata) return stbi__err("tRNS after IDAT","Corrupt PNG");
+ if (pal_img_n) {
+ if (scan == STBI__SCAN_header) { s->img_n = 4; return 1; }
+ if (pal_len == 0) return stbi__err("tRNS before PLTE","Corrupt PNG");
+ if (c.length > pal_len) return stbi__err("bad tRNS len","Corrupt PNG");
+ pal_img_n = 4;
+ for (i=0; i < c.length; ++i)
+ palette[i*4+3] = stbi__get8(s);
+ } else {
+ if (!(s->img_n & 1)) return stbi__err("tRNS with alpha","Corrupt PNG");
+ if (c.length != (stbi__uint32) s->img_n*2) return stbi__err("bad tRNS len","Corrupt PNG");
+ has_trans = 1;
+ if (z->depth == 16) {
+ for (k = 0; k < s->img_n; ++k) tc16[k] = (stbi__uint16)stbi__get16be(s); // copy the values as-is
+ } else {
+ for (k = 0; k < s->img_n; ++k) tc[k] = (stbi_uc)(stbi__get16be(s) & 255) * stbi__depth_scale_table[z->depth]; // non 8-bit images will be larger
+ }
+ }
+ break;
+ }
+
+ case STBI__PNG_TYPE('I','D','A','T'): {
+ if (first) return stbi__err("first not IHDR", "Corrupt PNG");
+ if (pal_img_n && !pal_len) return stbi__err("no PLTE","Corrupt PNG");
+ if (scan == STBI__SCAN_header) { s->img_n = pal_img_n; return 1; }
+ if ((int)(ioff + c.length) < (int)ioff) return 0;
+ if (ioff + c.length > idata_limit) {
+ stbi__uint32 idata_limit_old = idata_limit;
+ stbi_uc *p;
+ if (idata_limit == 0) idata_limit = c.length > 4096 ? c.length : 4096;
+ while (ioff + c.length > idata_limit)
+ idata_limit *= 2;
+ STBI_NOTUSED(idata_limit_old);
+ p = (stbi_uc *) STBI_REALLOC_SIZED(z->idata, idata_limit_old, idata_limit); if (p == NULL) return stbi__err("outofmem", "Out of memory");
+ z->idata = p;
+ }
+ if (!stbi__getn(s, z->idata+ioff,c.length)) return stbi__err("outofdata","Corrupt PNG");
+ ioff += c.length;
+ break;
+ }
+
+ case STBI__PNG_TYPE('I','E','N','D'): {
+ stbi__uint32 raw_len, bpl;
+ if (first) return stbi__err("first not IHDR", "Corrupt PNG");
+ if (scan != STBI__SCAN_load) return 1;
+ if (z->idata == NULL) return stbi__err("no IDAT","Corrupt PNG");
+ // initial guess for decoded data size to avoid unnecessary reallocs
+ bpl = (s->img_x * z->depth + 7) / 8; // bytes per line, per component
+ raw_len = bpl * s->img_y * s->img_n /* pixels */ + s->img_y /* filter mode per row */;
+ z->expanded = (stbi_uc *) stbi_zlib_decode_malloc_guesssize_headerflag((char *) z->idata, ioff, raw_len, (int *) &raw_len, !is_iphone);
+ if (z->expanded == NULL) return 0; // zlib should set error
+ STBI_FREE(z->idata); z->idata = NULL;
+ if ((req_comp == s->img_n+1 && req_comp != 3 && !pal_img_n) || has_trans)
+ s->img_out_n = s->img_n+1;
+ else
+ s->img_out_n = s->img_n;
+ if (!stbi__create_png_image(z, z->expanded, raw_len, s->img_out_n, z->depth, color, interlace)) return 0;
+ if (has_trans) {
+ if (z->depth == 16) {
+ if (!stbi__compute_transparency16(z, tc16, s->img_out_n)) return 0;
+ } else {
+ if (!stbi__compute_transparency(z, tc, s->img_out_n)) return 0;
+ }
+ }
+ if (is_iphone && stbi__de_iphone_flag && s->img_out_n > 2)
+ stbi__de_iphone(z);
+ if (pal_img_n) {
+ // pal_img_n == 3 or 4
+ s->img_n = pal_img_n; // record the actual colors we had
+ s->img_out_n = pal_img_n;
+ if (req_comp >= 3) s->img_out_n = req_comp;
+ if (!stbi__expand_png_palette(z, palette, pal_len, s->img_out_n))
+ return 0;
+ }
+ STBI_FREE(z->expanded); z->expanded = NULL;
+ return 1;
+ }
+
+ default:
+ // if critical, fail
+ if (first) return stbi__err("first not IHDR", "Corrupt PNG");
+ if ((c.type & (1 << 29)) == 0) {
+ #ifndef STBI_NO_FAILURE_STRINGS
+ // not threadsafe
+ static char invalid_chunk[] = "XXXX PNG chunk not known";
+ invalid_chunk[0] = STBI__BYTECAST(c.type >> 24);
+ invalid_chunk[1] = STBI__BYTECAST(c.type >> 16);
+ invalid_chunk[2] = STBI__BYTECAST(c.type >> 8);
+ invalid_chunk[3] = STBI__BYTECAST(c.type >> 0);
+ #endif
+ return stbi__err(invalid_chunk, "PNG not supported: unknown PNG chunk type");
+ }
+ stbi__skip(s, c.length);
+ break;
+ }
+ // end of PNG chunk, read and skip CRC
+ stbi__get32be(s);
+ }
+}
+
+static void *stbi__do_png(stbi__png *p, int *x, int *y, int *n, int req_comp, stbi__result_info *ri)
+{
+ void *result=NULL;
+ if (req_comp < 0 || req_comp > 4) return stbi__errpuc("bad req_comp", "Internal error");
+ if (stbi__parse_png_file(p, STBI__SCAN_load, req_comp)) {
+ if (p->depth < 8)
+ ri->bits_per_channel = 8;
+ else
+ ri->bits_per_channel = p->depth;
+ result = p->out;
+ p->out = NULL;
+ if (req_comp && req_comp != p->s->img_out_n) {
+ if (ri->bits_per_channel == 8)
+ result = stbi__convert_format((unsigned char *) result, p->s->img_out_n, req_comp, p->s->img_x, p->s->img_y);
+ else
+ result = stbi__convert_format16((stbi__uint16 *) result, p->s->img_out_n, req_comp, p->s->img_x, p->s->img_y);
+ p->s->img_out_n = req_comp;
+ if (result == NULL) return result;
+ }
+ *x = p->s->img_x;
+ *y = p->s->img_y;
+ if (n) *n = p->s->img_n;
+ }
+ STBI_FREE(p->out); p->out = NULL;
+ STBI_FREE(p->expanded); p->expanded = NULL;
+ STBI_FREE(p->idata); p->idata = NULL;
+
+ return result;
+}
+
+static void *stbi__png_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+ stbi__png p;
+ p.s = s;
+ return stbi__do_png(&p, x,y,comp,req_comp, ri);
+}
+
+static int stbi__png_test(stbi__context *s)
+{
+ int r;
+ r = stbi__check_png_header(s);
+ stbi__rewind(s);
+ return r;
+}
+
+static int stbi__png_info_raw(stbi__png *p, int *x, int *y, int *comp)
+{
+ if (!stbi__parse_png_file(p, STBI__SCAN_header, 0)) {
+ stbi__rewind( p->s );
+ return 0;
+ }
+ if (x) *x = p->s->img_x;
+ if (y) *y = p->s->img_y;
+ if (comp) *comp = p->s->img_n;
+ return 1;
+}
+
+static int stbi__png_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ stbi__png p;
+ p.s = s;
+ return stbi__png_info_raw(&p, x, y, comp);
+}
+#endif
+
+// Microsoft/Windows BMP image
+
+#ifndef STBI_NO_BMP
+static int stbi__bmp_test_raw(stbi__context *s)
+{
+ int r;
+ int sz;
+ if (stbi__get8(s) != 'B') return 0;
+ if (stbi__get8(s) != 'M') return 0;
+ stbi__get32le(s); // discard filesize
+ stbi__get16le(s); // discard reserved
+ stbi__get16le(s); // discard reserved
+ stbi__get32le(s); // discard data offset
+ sz = stbi__get32le(s);
+ r = (sz == 12 || sz == 40 || sz == 56 || sz == 108 || sz == 124);
+ return r;
+}
+
+static int stbi__bmp_test(stbi__context *s)
+{
+ int r = stbi__bmp_test_raw(s);
+ stbi__rewind(s);
+ return r;
+}
+
+
+// returns 0..31 for the highest set bit
+static int stbi__high_bit(unsigned int z)
+{
+ int n=0;
+ if (z == 0) return -1;
+ if (z >= 0x10000) n += 16, z >>= 16;
+ if (z >= 0x00100) n += 8, z >>= 8;
+ if (z >= 0x00010) n += 4, z >>= 4;
+ if (z >= 0x00004) n += 2, z >>= 2;
+ if (z >= 0x00002) n += 1, z >>= 1;
+ return n;
+}
+
+static int stbi__bitcount(unsigned int a)
+{
+ a = (a & 0x55555555) + ((a >> 1) & 0x55555555); // max 2
+ a = (a & 0x33333333) + ((a >> 2) & 0x33333333); // max 4
+ a = (a + (a >> 4)) & 0x0f0f0f0f; // max 8 per 4, now 8 bits
+ a = (a + (a >> 8)); // max 16 per 8 bits
+ a = (a + (a >> 16)); // max 32 per 8 bits
+ return a & 0xff;
+}
+
+static int stbi__shiftsigned(int v, int shift, int bits)
+{
+ int result;
+ int z=0;
+
+ if (shift < 0) v <<= -shift;
+ else v >>= shift;
+ result = v;
+
+ z = bits;
+ while (z < 8) {
+ result += v >> z;
+ z += bits;
+ }
+ return result;
+}
+
+typedef struct
+{
+ int bpp, offset, hsz;
+ unsigned int mr,mg,mb,ma, all_a;
+} stbi__bmp_data;
+
+static void *stbi__bmp_parse_header(stbi__context *s, stbi__bmp_data *info)
+{
+ int hsz;
+ if (stbi__get8(s) != 'B' || stbi__get8(s) != 'M') return stbi__errpuc("not BMP", "Corrupt BMP");
+ stbi__get32le(s); // discard filesize
+ stbi__get16le(s); // discard reserved
+ stbi__get16le(s); // discard reserved
+ info->offset = stbi__get32le(s);
+ info->hsz = hsz = stbi__get32le(s);
+ info->mr = info->mg = info->mb = info->ma = 0;
+
+ if (hsz != 12 && hsz != 40 && hsz != 56 && hsz != 108 && hsz != 124) return stbi__errpuc("unknown BMP", "BMP type not supported: unknown");
+ if (hsz == 12) {
+ s->img_x = stbi__get16le(s);
+ s->img_y = stbi__get16le(s);
+ } else {
+ s->img_x = stbi__get32le(s);
+ s->img_y = stbi__get32le(s);
+ }
+ if (stbi__get16le(s) != 1) return stbi__errpuc("bad BMP", "bad BMP");
+ info->bpp = stbi__get16le(s);
+ if (info->bpp == 1) return stbi__errpuc("monochrome", "BMP type not supported: 1-bit");
+ if (hsz != 12) {
+ int compress = stbi__get32le(s);
+ if (compress == 1 || compress == 2) return stbi__errpuc("BMP RLE", "BMP type not supported: RLE");
+ stbi__get32le(s); // discard sizeof
+ stbi__get32le(s); // discard hres
+ stbi__get32le(s); // discard vres
+ stbi__get32le(s); // discard colorsused
+ stbi__get32le(s); // discard max important
+ if (hsz == 40 || hsz == 56) {
+ if (hsz == 56) {
+ stbi__get32le(s);
+ stbi__get32le(s);
+ stbi__get32le(s);
+ stbi__get32le(s);
+ }
+ if (info->bpp == 16 || info->bpp == 32) {
+ if (compress == 0) {
+ if (info->bpp == 32) {
+ info->mr = 0xffu << 16;
+ info->mg = 0xffu << 8;
+ info->mb = 0xffu << 0;
+ info->ma = 0xffu << 24;
+ info->all_a = 0; // if all_a is 0 at end, then we loaded alpha channel but it was all 0
+ } else {
+ info->mr = 31u << 10;
+ info->mg = 31u << 5;
+ info->mb = 31u << 0;
+ }
+ } else if (compress == 3) {
+ info->mr = stbi__get32le(s);
+ info->mg = stbi__get32le(s);
+ info->mb = stbi__get32le(s);
+ // not documented, but generated by photoshop and handled by mspaint
+ if (info->mr == info->mg && info->mg == info->mb) {
+ // ?!?!?
+ return stbi__errpuc("bad BMP", "bad BMP");
+ }
+ } else
+ return stbi__errpuc("bad BMP", "bad BMP");
+ }
+ } else {
+ int i;
+ if (hsz != 108 && hsz != 124)
+ return stbi__errpuc("bad BMP", "bad BMP");
+ info->mr = stbi__get32le(s);
+ info->mg = stbi__get32le(s);
+ info->mb = stbi__get32le(s);
+ info->ma = stbi__get32le(s);
+ stbi__get32le(s); // discard color space
+ for (i=0; i < 12; ++i)
+ stbi__get32le(s); // discard color space parameters
+ if (hsz == 124) {
+ stbi__get32le(s); // discard rendering intent
+ stbi__get32le(s); // discard offset of profile data
+ stbi__get32le(s); // discard size of profile data
+ stbi__get32le(s); // discard reserved
+ }
+ }
+ }
+ return (void *) 1;
+}
+
+
+static void *stbi__bmp_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+ stbi_uc *out;
+ unsigned int mr=0,mg=0,mb=0,ma=0, all_a;
+ stbi_uc pal[256][4];
+ int psize=0,i,j,width;
+ int flip_vertically, pad, target;
+ stbi__bmp_data info;
+ STBI_NOTUSED(ri);
+
+ info.all_a = 255;
+ if (stbi__bmp_parse_header(s, &info) == NULL)
+ return NULL; // error code already set
+
+ flip_vertically = ((int) s->img_y) > 0;
+ s->img_y = abs((int) s->img_y);
+
+ mr = info.mr;
+ mg = info.mg;
+ mb = info.mb;
+ ma = info.ma;
+ all_a = info.all_a;
+
+ if (info.hsz == 12) {
+ if (info.bpp < 24)
+ psize = (info.offset - 14 - 24) / 3;
+ } else {
+ if (info.bpp < 16)
+ psize = (info.offset - 14 - info.hsz) >> 2;
+ }
+
+ s->img_n = ma ? 4 : 3;
+ if (req_comp && req_comp >= 3) // we can directly decode 3 or 4
+ target = req_comp;
+ else
+ target = s->img_n; // if they want monochrome, we'll post-convert
+
+ // sanity-check size
+ if (!stbi__mad3sizes_valid(target, s->img_x, s->img_y, 0))
+ return stbi__errpuc("too large", "Corrupt BMP");
+
+ out = (stbi_uc *) stbi__malloc_mad3(target, s->img_x, s->img_y, 0);
+ if (!out) return stbi__errpuc("outofmem", "Out of memory");
+ if (info.bpp < 16) {
+ int z=0;
+ if (psize == 0 || psize > 256) { STBI_FREE(out); return stbi__errpuc("invalid", "Corrupt BMP"); }
+ for (i=0; i < psize; ++i) {
+ pal[i][2] = stbi__get8(s);
+ pal[i][1] = stbi__get8(s);
+ pal[i][0] = stbi__get8(s);
+ if (info.hsz != 12) stbi__get8(s);
+ pal[i][3] = 255;
+ }
+ stbi__skip(s, info.offset - 14 - info.hsz - psize * (info.hsz == 12 ? 3 : 4));
+ if (info.bpp == 4) width = (s->img_x + 1) >> 1;
+ else if (info.bpp == 8) width = s->img_x;
+ else { STBI_FREE(out); return stbi__errpuc("bad bpp", "Corrupt BMP"); }
+ pad = (-width)&3;
+ for (j=0; j < (int) s->img_y; ++j) {
+ for (i=0; i < (int) s->img_x; i += 2) {
+ int v=stbi__get8(s),v2=0;
+ if (info.bpp == 4) {
+ v2 = v & 15;
+ v >>= 4;
+ }
+ out[z++] = pal[v][0];
+ out[z++] = pal[v][1];
+ out[z++] = pal[v][2];
+ if (target == 4) out[z++] = 255;
+ if (i+1 == (int) s->img_x) break;
+ v = (info.bpp == 8) ? stbi__get8(s) : v2;
+ out[z++] = pal[v][0];
+ out[z++] = pal[v][1];
+ out[z++] = pal[v][2];
+ if (target == 4) out[z++] = 255;
+ }
+ stbi__skip(s, pad);
+ }
+ } else {
+ int rshift=0,gshift=0,bshift=0,ashift=0,rcount=0,gcount=0,bcount=0,acount=0;
+ int z = 0;
+ int easy=0;
+ stbi__skip(s, info.offset - 14 - info.hsz);
+ if (info.bpp == 24) width = 3 * s->img_x;
+ else if (info.bpp == 16) width = 2*s->img_x;
+ else /* bpp = 32 and pad = 0 */ width=0;
+ pad = (-width) & 3;
+ if (info.bpp == 24) {
+ easy = 1;
+ } else if (info.bpp == 32) {
+ if (mb == 0xff && mg == 0xff00 && mr == 0x00ff0000 && ma == 0xff000000)
+ easy = 2;
+ }
+ if (!easy) {
+ if (!mr || !mg || !mb) { STBI_FREE(out); return stbi__errpuc("bad masks", "Corrupt BMP"); }
+ // right shift amt to put high bit in position #7
+ rshift = stbi__high_bit(mr)-7; rcount = stbi__bitcount(mr);
+ gshift = stbi__high_bit(mg)-7; gcount = stbi__bitcount(mg);
+ bshift = stbi__high_bit(mb)-7; bcount = stbi__bitcount(mb);
+ ashift = stbi__high_bit(ma)-7; acount = stbi__bitcount(ma);
+ }
+ for (j=0; j < (int) s->img_y; ++j) {
+ if (easy) {
+ for (i=0; i < (int) s->img_x; ++i) {
+ unsigned char a;
+ out[z+2] = stbi__get8(s);
+ out[z+1] = stbi__get8(s);
+ out[z+0] = stbi__get8(s);
+ z += 3;
+ a = (easy == 2 ? stbi__get8(s) : 255);
+ all_a |= a;
+ if (target == 4) out[z++] = a;
+ }
+ } else {
+ int bpp = info.bpp;
+ for (i=0; i < (int) s->img_x; ++i) {
+ stbi__uint32 v = (bpp == 16 ? (stbi__uint32) stbi__get16le(s) : stbi__get32le(s));
+ int a;
+ out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mr, rshift, rcount));
+ out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mg, gshift, gcount));
+ out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mb, bshift, bcount));
+ a = (ma ? stbi__shiftsigned(v & ma, ashift, acount) : 255);
+ all_a |= a;
+ if (target == 4) out[z++] = STBI__BYTECAST(a);
+ }
+ }
+ stbi__skip(s, pad);
+ }
+ }
+
+ // if alpha channel is all 0s, replace with all 255s
+ if (target == 4 && all_a == 0)
+ for (i=4*s->img_x*s->img_y-1; i >= 0; i -= 4)
+ out[i] = 255;
+
+ if (flip_vertically) {
+ stbi_uc t;
+ for (j=0; j < (int) s->img_y>>1; ++j) {
+ stbi_uc *p1 = out + j *s->img_x*target;
+ stbi_uc *p2 = out + (s->img_y-1-j)*s->img_x*target;
+ for (i=0; i < (int) s->img_x*target; ++i) {
+ t = p1[i], p1[i] = p2[i], p2[i] = t;
+ }
+ }
+ }
+
+ if (req_comp && req_comp != target) {
+ out = stbi__convert_format(out, target, req_comp, s->img_x, s->img_y);
+ if (out == NULL) return out; // stbi__convert_format frees input on failure
+ }
+
+ *x = s->img_x;
+ *y = s->img_y;
+ if (comp) *comp = s->img_n;
+ return out;
+}
+#endif
+
+// Targa Truevision - TGA
+// by Jonathan Dummer
+#ifndef STBI_NO_TGA
+// returns STBI_rgb or whatever, 0 on error
+static int stbi__tga_get_comp(int bits_per_pixel, int is_grey, int* is_rgb16)
+{
+ // only RGB or RGBA (incl. 16bit) or grey allowed
+ if(is_rgb16) *is_rgb16 = 0;
+ switch(bits_per_pixel) {
+ case 8: return STBI_grey;
+ case 16: if(is_grey) return STBI_grey_alpha;
+ // else: fall-through
+ case 15: if(is_rgb16) *is_rgb16 = 1;
+ return STBI_rgb;
+ case 24: // fall-through
+ case 32: return bits_per_pixel/8;
+ default: return 0;
+ }
+}
+
+static int stbi__tga_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ int tga_w, tga_h, tga_comp, tga_image_type, tga_bits_per_pixel, tga_colormap_bpp;
+ int sz, tga_colormap_type;
+ stbi__get8(s); // discard Offset
+ tga_colormap_type = stbi__get8(s); // colormap type
+ if( tga_colormap_type > 1 ) {
+ stbi__rewind(s);
+ return 0; // only RGB or indexed allowed
+ }
+ tga_image_type = stbi__get8(s); // image type
+ if ( tga_colormap_type == 1 ) { // colormapped (paletted) image
+ if (tga_image_type != 1 && tga_image_type != 9) {
+ stbi__rewind(s);
+ return 0;
+ }
+ stbi__skip(s,4); // skip index of first colormap entry and number of entries
+ sz = stbi__get8(s); // check bits per palette color entry
+ if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) {
+ stbi__rewind(s);
+ return 0;
+ }
+ stbi__skip(s,4); // skip image x and y origin
+ tga_colormap_bpp = sz;
+ } else { // "normal" image w/o colormap - only RGB or grey allowed, +/- RLE
+ if ( (tga_image_type != 2) && (tga_image_type != 3) && (tga_image_type != 10) && (tga_image_type != 11) ) {
+ stbi__rewind(s);
+ return 0; // only RGB or grey allowed, +/- RLE
+ }
+ stbi__skip(s,9); // skip colormap specification and image x/y origin
+ tga_colormap_bpp = 0;
+ }
+ tga_w = stbi__get16le(s);
+ if( tga_w < 1 ) {
+ stbi__rewind(s);
+ return 0; // test width
+ }
+ tga_h = stbi__get16le(s);
+ if( tga_h < 1 ) {
+ stbi__rewind(s);
+ return 0; // test height
+ }
+ tga_bits_per_pixel = stbi__get8(s); // bits per pixel
+ stbi__get8(s); // ignore alpha bits
+ if (tga_colormap_bpp != 0) {
+ if((tga_bits_per_pixel != 8) && (tga_bits_per_pixel != 16)) {
+ // when using a colormap, tga_bits_per_pixel is the size of the indexes
+ // I don't think anything but 8 or 16bit indexes makes sense
+ stbi__rewind(s);
+ return 0;
+ }
+ tga_comp = stbi__tga_get_comp(tga_colormap_bpp, 0, NULL);
+ } else {
+ tga_comp = stbi__tga_get_comp(tga_bits_per_pixel, (tga_image_type == 3) || (tga_image_type == 11), NULL);
+ }
+ if(!tga_comp) {
+ stbi__rewind(s);
+ return 0;
+ }
+ if (x) *x = tga_w;
+ if (y) *y = tga_h;
+ if (comp) *comp = tga_comp;
+ return 1; // seems to have passed everything
+}
+
+static int stbi__tga_test(stbi__context *s)
+{
+ int res = 0;
+ int sz, tga_color_type;
+ stbi__get8(s); // discard Offset
+ tga_color_type = stbi__get8(s); // color type
+ if ( tga_color_type > 1 ) goto errorEnd; // only RGB or indexed allowed
+ sz = stbi__get8(s); // image type
+ if ( tga_color_type == 1 ) { // colormapped (paletted) image
+ if (sz != 1 && sz != 9) goto errorEnd; // colortype 1 demands image type 1 or 9
+ stbi__skip(s,4); // skip index of first colormap entry and number of entries
+ sz = stbi__get8(s); // check bits per palette color entry
+ if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) goto errorEnd;
+ stbi__skip(s,4); // skip image x and y origin
+ } else { // "normal" image w/o colormap
+ if ( (sz != 2) && (sz != 3) && (sz != 10) && (sz != 11) ) goto errorEnd; // only RGB or grey allowed, +/- RLE
+ stbi__skip(s,9); // skip colormap specification and image x/y origin
+ }
+ if ( stbi__get16le(s) < 1 ) goto errorEnd; // test width
+ if ( stbi__get16le(s) < 1 ) goto errorEnd; // test height
+ sz = stbi__get8(s); // bits per pixel
+ if ( (tga_color_type == 1) && (sz != 8) && (sz != 16) ) goto errorEnd; // for colormapped images, bpp is size of an index
+ if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) goto errorEnd;
+
+ res = 1; // if we got this far, everything's good and we can return 1 instead of 0
+
+errorEnd:
+ stbi__rewind(s);
+ return res;
+}
+
+// read 16bit value and convert to 24bit RGB
+static void stbi__tga_read_rgb16(stbi__context *s, stbi_uc* out)
+{
+ stbi__uint16 px = (stbi__uint16)stbi__get16le(s);
+ stbi__uint16 fiveBitMask = 31;
+ // we have 3 channels with 5bits each
+ int r = (px >> 10) & fiveBitMask;
+ int g = (px >> 5) & fiveBitMask;
+ int b = px & fiveBitMask;
+ // Note that this saves the data in RGB(A) order, so it doesn't need to be swapped later
+ out[0] = (stbi_uc)((r * 255)/31);
+ out[1] = (stbi_uc)((g * 255)/31);
+ out[2] = (stbi_uc)((b * 255)/31);
+
+ // some people claim that the most significant bit might be used for alpha
+ // (possibly if an alpha-bit is set in the "image descriptor byte")
+ // but that only made 16bit test images completely translucent..
+ // so let's treat all 15 and 16bit TGAs as RGB with no alpha.
+}
+
+static void *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+ // read in the TGA header stuff
+ int tga_offset = stbi__get8(s);
+ int tga_indexed = stbi__get8(s);
+ int tga_image_type = stbi__get8(s);
+ int tga_is_RLE = 0;
+ int tga_palette_start = stbi__get16le(s);
+ int tga_palette_len = stbi__get16le(s);
+ int tga_palette_bits = stbi__get8(s);
+ int tga_x_origin = stbi__get16le(s);
+ int tga_y_origin = stbi__get16le(s);
+ int tga_width = stbi__get16le(s);
+ int tga_height = stbi__get16le(s);
+ int tga_bits_per_pixel = stbi__get8(s);
+ int tga_comp, tga_rgb16=0;
+ int tga_inverted = stbi__get8(s);
+ // int tga_alpha_bits = tga_inverted & 15; // the 4 lowest bits - unused (useless?)
+ // image data
+ unsigned char *tga_data;
+ unsigned char *tga_palette = NULL;
+ int i, j;
+ unsigned char raw_data[4] = {0};
+ int RLE_count = 0;
+ int RLE_repeating = 0;
+ int read_next_pixel = 1;
+ STBI_NOTUSED(ri);
+
+ // do a tiny bit of precessing
+ if ( tga_image_type >= 8 )
+ {
+ tga_image_type -= 8;
+ tga_is_RLE = 1;
+ }
+ tga_inverted = 1 - ((tga_inverted >> 5) & 1);
+
+ // If I'm paletted, then I'll use the number of bits from the palette
+ if ( tga_indexed ) tga_comp = stbi__tga_get_comp(tga_palette_bits, 0, &tga_rgb16);
+ else tga_comp = stbi__tga_get_comp(tga_bits_per_pixel, (tga_image_type == 3), &tga_rgb16);
+
+ if(!tga_comp) // shouldn't really happen, stbi__tga_test() should have ensured basic consistency
+ return stbi__errpuc("bad format", "Can't find out TGA pixelformat");
+
+ // tga info
+ *x = tga_width;
+ *y = tga_height;
+ if (comp) *comp = tga_comp;
+
+ if (!stbi__mad3sizes_valid(tga_width, tga_height, tga_comp, 0))
+ return stbi__errpuc("too large", "Corrupt TGA");
+
+ tga_data = (unsigned char*)stbi__malloc_mad3(tga_width, tga_height, tga_comp, 0);
+ if (!tga_data) return stbi__errpuc("outofmem", "Out of memory");
+
+ // skip to the data's starting position (offset usually = 0)
+ stbi__skip(s, tga_offset );
+
+ if ( !tga_indexed && !tga_is_RLE && !tga_rgb16 ) {
+ for (i=0; i < tga_height; ++i) {
+ int row = tga_inverted ? tga_height -i - 1 : i;
+ stbi_uc *tga_row = tga_data + row*tga_width*tga_comp;
+ stbi__getn(s, tga_row, tga_width * tga_comp);
+ }
+ } else {
+ // do I need to load a palette?
+ if ( tga_indexed)
+ {
+ // any data to skip? (offset usually = 0)
+ stbi__skip(s, tga_palette_start );
+ // load the palette
+ tga_palette = (unsigned char*)stbi__malloc_mad2(tga_palette_len, tga_comp, 0);
+ if (!tga_palette) {
+ STBI_FREE(tga_data);
+ return stbi__errpuc("outofmem", "Out of memory");
+ }
+ if (tga_rgb16) {
+ stbi_uc *pal_entry = tga_palette;
+ STBI_ASSERT(tga_comp == STBI_rgb);
+ for (i=0; i < tga_palette_len; ++i) {
+ stbi__tga_read_rgb16(s, pal_entry);
+ pal_entry += tga_comp;
+ }
+ } else if (!stbi__getn(s, tga_palette, tga_palette_len * tga_comp)) {
+ STBI_FREE(tga_data);
+ STBI_FREE(tga_palette);
+ return stbi__errpuc("bad palette", "Corrupt TGA");
+ }
+ }
+ // load the data
+ for (i=0; i < tga_width * tga_height; ++i)
+ {
+ // if I'm in RLE mode, do I need to get a RLE stbi__pngchunk?
+ if ( tga_is_RLE )
+ {
+ if ( RLE_count == 0 )
+ {
+ // yep, get the next byte as a RLE command
+ int RLE_cmd = stbi__get8(s);
+ RLE_count = 1 + (RLE_cmd & 127);
+ RLE_repeating = RLE_cmd >> 7;
+ read_next_pixel = 1;
+ } else if ( !RLE_repeating )
+ {
+ read_next_pixel = 1;
+ }
+ } else
+ {
+ read_next_pixel = 1;
+ }
+ // OK, if I need to read a pixel, do it now
+ if ( read_next_pixel )
+ {
+ // load however much data we did have
+ if ( tga_indexed )
+ {
+ // read in index, then perform the lookup
+ int pal_idx = (tga_bits_per_pixel == 8) ? stbi__get8(s) : stbi__get16le(s);
+ if ( pal_idx >= tga_palette_len ) {
+ // invalid index
+ pal_idx = 0;
+ }
+ pal_idx *= tga_comp;
+ for (j = 0; j < tga_comp; ++j) {
+ raw_data[j] = tga_palette[pal_idx+j];
+ }
+ } else if(tga_rgb16) {
+ STBI_ASSERT(tga_comp == STBI_rgb);
+ stbi__tga_read_rgb16(s, raw_data);
+ } else {
+ // read in the data raw
+ for (j = 0; j < tga_comp; ++j) {
+ raw_data[j] = stbi__get8(s);
+ }
+ }
+ // clear the reading flag for the next pixel
+ read_next_pixel = 0;
+ } // end of reading a pixel
+
+ // copy data
+ for (j = 0; j < tga_comp; ++j)
+ tga_data[i*tga_comp+j] = raw_data[j];
+
+ // in case we're in RLE mode, keep counting down
+ --RLE_count;
+ }
+ // do I need to invert the image?
+ if ( tga_inverted )
+ {
+ for (j = 0; j*2 < tga_height; ++j)
+ {
+ int index1 = j * tga_width * tga_comp;
+ int index2 = (tga_height - 1 - j) * tga_width * tga_comp;
+ for (i = tga_width * tga_comp; i > 0; --i)
+ {
+ unsigned char temp = tga_data[index1];
+ tga_data[index1] = tga_data[index2];
+ tga_data[index2] = temp;
+ ++index1;
+ ++index2;
+ }
+ }
+ }
+ // clear my palette, if I had one
+ if ( tga_palette != NULL )
+ {
+ STBI_FREE( tga_palette );
+ }
+ }
+
+ // swap RGB - if the source data was RGB16, it already is in the right order
+ if (tga_comp >= 3 && !tga_rgb16)
+ {
+ unsigned char* tga_pixel = tga_data;
+ for (i=0; i < tga_width * tga_height; ++i)
+ {
+ unsigned char temp = tga_pixel[0];
+ tga_pixel[0] = tga_pixel[2];
+ tga_pixel[2] = temp;
+ tga_pixel += tga_comp;
+ }
+ }
+
+ // convert to target component count
+ if (req_comp && req_comp != tga_comp)
+ tga_data = stbi__convert_format(tga_data, tga_comp, req_comp, tga_width, tga_height);
+
+ // the things I do to get rid of an error message, and yet keep
+ // Microsoft's C compilers happy... [8^(
+ tga_palette_start = tga_palette_len = tga_palette_bits =
+ tga_x_origin = tga_y_origin = 0;
+ // OK, done
+ return tga_data;
+}
+#endif
+
+// *************************************************************************************************
+// Photoshop PSD loader -- PD by Thatcher Ulrich, integration by Nicolas Schulz, tweaked by STB
+
+#ifndef STBI_NO_PSD
+static int stbi__psd_test(stbi__context *s)
+{
+ int r = (stbi__get32be(s) == 0x38425053);
+ stbi__rewind(s);
+ return r;
+}
+
+static int stbi__psd_decode_rle(stbi__context *s, stbi_uc *p, int pixelCount)
+{
+ int count, nleft, len;
+
+ count = 0;
+ while ((nleft = pixelCount - count) > 0) {
+ len = stbi__get8(s);
+ if (len == 128) {
+ // No-op.
+ } else if (len < 128) {
+ // Copy next len+1 bytes literally.
+ len++;
+ if (len > nleft) return 0; // corrupt data
+ count += len;
+ while (len) {
+ *p = stbi__get8(s);
+ p += 4;
+ len--;
+ }
+ } else if (len > 128) {
+ stbi_uc val;
+ // Next -len+1 bytes in the dest are replicated from next source byte.
+ // (Interpret len as a negative 8-bit int.)
+ len = 257 - len;
+ if (len > nleft) return 0; // corrupt data
+ val = stbi__get8(s);
+ count += len;
+ while (len) {
+ *p = val;
+ p += 4;
+ len--;
+ }
+ }
+ }
+
+ return 1;
+}
+
+static void *stbi__psd_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri, int bpc)
+{
+ int pixelCount;
+ int channelCount, compression;
+ int channel, i;
+ int bitdepth;
+ int w,h;
+ stbi_uc *out;
+ STBI_NOTUSED(ri);
+
+ // Check identifier
+ if (stbi__get32be(s) != 0x38425053) // "8BPS"
+ return stbi__errpuc("not PSD", "Corrupt PSD image");
+
+ // Check file type version.
+ if (stbi__get16be(s) != 1)
+ return stbi__errpuc("wrong version", "Unsupported version of PSD image");
+
+ // Skip 6 reserved bytes.
+ stbi__skip(s, 6 );
+
+ // Read the number of channels (R, G, B, A, etc).
+ channelCount = stbi__get16be(s);
+ if (channelCount < 0 || channelCount > 16)
+ return stbi__errpuc("wrong channel count", "Unsupported number of channels in PSD image");
+
+ // Read the rows and columns of the image.
+ h = stbi__get32be(s);
+ w = stbi__get32be(s);
+
+ // Make sure the depth is 8 bits.
+ bitdepth = stbi__get16be(s);
+ if (bitdepth != 8 && bitdepth != 16)
+ return stbi__errpuc("unsupported bit depth", "PSD bit depth is not 8 or 16 bit");
+
+ // Make sure the color mode is RGB.
+ // Valid options are:
+ // 0: Bitmap
+ // 1: Grayscale
+ // 2: Indexed color
+ // 3: RGB color
+ // 4: CMYK color
+ // 7: Multichannel
+ // 8: Duotone
+ // 9: Lab color
+ if (stbi__get16be(s) != 3)
+ return stbi__errpuc("wrong color format", "PSD is not in RGB color format");
+
+ // Skip the Mode Data. (It's the palette for indexed color; other info for other modes.)
+ stbi__skip(s,stbi__get32be(s) );
+
+ // Skip the image resources. (resolution, pen tool paths, etc)
+ stbi__skip(s, stbi__get32be(s) );
+
+ // Skip the reserved data.
+ stbi__skip(s, stbi__get32be(s) );
+
+ // Find out if the data is compressed.
+ // Known values:
+ // 0: no compression
+ // 1: RLE compressed
+ compression = stbi__get16be(s);
+ if (compression > 1)
+ return stbi__errpuc("bad compression", "PSD has an unknown compression format");
+
+ // Check size
+ if (!stbi__mad3sizes_valid(4, w, h, 0))
+ return stbi__errpuc("too large", "Corrupt PSD");
+
+ // Create the destination image.
+
+ if (!compression && bitdepth == 16 && bpc == 16) {
+ out = (stbi_uc *) stbi__malloc_mad3(8, w, h, 0);
+ ri->bits_per_channel = 16;
+ } else
+ out = (stbi_uc *) stbi__malloc(4 * w*h);
+
+ if (!out) return stbi__errpuc("outofmem", "Out of memory");
+ pixelCount = w*h;
+
+ // Initialize the data to zero.
+ //memset( out, 0, pixelCount * 4 );
+
+ // Finally, the image data.
+ if (compression) {
+ // RLE as used by .PSD and .TIFF
+ // Loop until you get the number of unpacked bytes you are expecting:
+ // Read the next source byte into n.
+ // If n is between 0 and 127 inclusive, copy the next n+1 bytes literally.
+ // Else if n is between -127 and -1 inclusive, copy the next byte -n+1 times.
+ // Else if n is 128, noop.
+ // Endloop
+
+ // The RLE-compressed data is preceeded by a 2-byte data count for each row in the data,
+ // which we're going to just skip.
+ stbi__skip(s, h * channelCount * 2 );
+
+ // Read the RLE data by channel.
+ for (channel = 0; channel < 4; channel++) {
+ stbi_uc *p;
+
+ p = out+channel;
+ if (channel >= channelCount) {
+ // Fill this channel with default data.
+ for (i = 0; i < pixelCount; i++, p += 4)
+ *p = (channel == 3 ? 255 : 0);
+ } else {
+ // Read the RLE data.
+ if (!stbi__psd_decode_rle(s, p, pixelCount)) {
+ STBI_FREE(out);
+ return stbi__errpuc("corrupt", "bad RLE data");
+ }
+ }
+ }
+
+ } else {
+ // We're at the raw image data. It's each channel in order (Red, Green, Blue, Alpha, ...)
+ // where each channel consists of an 8-bit (or 16-bit) value for each pixel in the image.
+
+ // Read the data by channel.
+ for (channel = 0; channel < 4; channel++) {
+ if (channel >= channelCount) {
+ // Fill this channel with default data.
+ if (bitdepth == 16 && bpc == 16) {
+ stbi__uint16 *q = ((stbi__uint16 *) out) + channel;
+ stbi__uint16 val = channel == 3 ? 65535 : 0;
+ for (i = 0; i < pixelCount; i++, q += 4)
+ *q = val;
+ } else {
+ stbi_uc *p = out+channel;
+ stbi_uc val = channel == 3 ? 255 : 0;
+ for (i = 0; i < pixelCount; i++, p += 4)
+ *p = val;
+ }
+ } else {
+ if (ri->bits_per_channel == 16) { // output bpc
+ stbi__uint16 *q = ((stbi__uint16 *) out) + channel;
+ for (i = 0; i < pixelCount; i++, q += 4)
+ *q = (stbi__uint16) stbi__get16be(s);
+ } else {
+ stbi_uc *p = out+channel;
+ if (bitdepth == 16) { // input bpc
+ for (i = 0; i < pixelCount; i++, p += 4)
+ *p = (stbi_uc) (stbi__get16be(s) >> 8);
+ } else {
+ for (i = 0; i < pixelCount; i++, p += 4)
+ *p = stbi__get8(s);
+ }
+ }
+ }
+ }
+ }
+
+ // remove weird white matte from PSD
+ if (channelCount >= 4) {
+ if (ri->bits_per_channel == 16) {
+ for (i=0; i < w*h; ++i) {
+ stbi__uint16 *pixel = (stbi__uint16 *) out + 4*i;
+ if (pixel[3] != 0 && pixel[3] != 65535) {
+ float a = pixel[3] / 65535.0f;
+ float ra = 1.0f / a;
+ float inv_a = 65535.0f * (1 - ra);
+ pixel[0] = (stbi__uint16) (pixel[0]*ra + inv_a);
+ pixel[1] = (stbi__uint16) (pixel[1]*ra + inv_a);
+ pixel[2] = (stbi__uint16) (pixel[2]*ra + inv_a);
+ }
+ }
+ } else {
+ for (i=0; i < w*h; ++i) {
+ unsigned char *pixel = out + 4*i;
+ if (pixel[3] != 0 && pixel[3] != 255) {
+ float a = pixel[3] / 255.0f;
+ float ra = 1.0f / a;
+ float inv_a = 255.0f * (1 - ra);
+ pixel[0] = (unsigned char) (pixel[0]*ra + inv_a);
+ pixel[1] = (unsigned char) (pixel[1]*ra + inv_a);
+ pixel[2] = (unsigned char) (pixel[2]*ra + inv_a);
+ }
+ }
+ }
+ }
+
+ // convert to desired output format
+ if (req_comp && req_comp != 4) {
+ if (ri->bits_per_channel == 16)
+ out = (stbi_uc *) stbi__convert_format16((stbi__uint16 *) out, 4, req_comp, w, h);
+ else
+ out = stbi__convert_format(out, 4, req_comp, w, h);
+ if (out == NULL) return out; // stbi__convert_format frees input on failure
+ }
+
+ if (comp) *comp = 4;
+ *y = h;
+ *x = w;
+
+ return out;
+}
+#endif
+
+// *************************************************************************************************
+// Softimage PIC loader
+// by Tom Seddon
+//
+// See http://softimage.wiki.softimage.com/index.php/INFO:_PIC_file_format
+// See http://ozviz.wasp.uwa.edu.au/~pbourke/dataformats/softimagepic/
+
+#ifndef STBI_NO_PIC
+static int stbi__pic_is4(stbi__context *s,const char *str)
+{
+ int i;
+ for (i=0; i<4; ++i)
+ if (stbi__get8(s) != (stbi_uc)str[i])
+ return 0;
+
+ return 1;
+}
+
+static int stbi__pic_test_core(stbi__context *s)
+{
+ int i;
+
+ if (!stbi__pic_is4(s,"\x53\x80\xF6\x34"))
+ return 0;
+
+ for(i=0;i<84;++i)
+ stbi__get8(s);
+
+ if (!stbi__pic_is4(s,"PICT"))
+ return 0;
+
+ return 1;
+}
+
+typedef struct
+{
+ stbi_uc size,type,channel;
+} stbi__pic_packet;
+
+static stbi_uc *stbi__readval(stbi__context *s, int channel, stbi_uc *dest)
+{
+ int mask=0x80, i;
+
+ for (i=0; i<4; ++i, mask>>=1) {
+ if (channel & mask) {
+ if (stbi__at_eof(s)) return stbi__errpuc("bad file","PIC file too short");
+ dest[i]=stbi__get8(s);
+ }
+ }
+
+ return dest;
+}
+
+static void stbi__copyval(int channel,stbi_uc *dest,const stbi_uc *src)
+{
+ int mask=0x80,i;
+
+ for (i=0;i<4; ++i, mask>>=1)
+ if (channel&mask)
+ dest[i]=src[i];
+}
+
+static stbi_uc *stbi__pic_load_core(stbi__context *s,int width,int height,int *comp, stbi_uc *result)
+{
+ int act_comp=0,num_packets=0,y,chained;
+ stbi__pic_packet packets[10];
+
+ // this will (should...) cater for even some bizarre stuff like having data
+ // for the same channel in multiple packets.
+ do {
+ stbi__pic_packet *packet;
+
+ if (num_packets==sizeof(packets)/sizeof(packets[0]))
+ return stbi__errpuc("bad format","too many packets");
+
+ packet = &packets[num_packets++];
+
+ chained = stbi__get8(s);
+ packet->size = stbi__get8(s);
+ packet->type = stbi__get8(s);
+ packet->channel = stbi__get8(s);
+
+ act_comp |= packet->channel;
+
+ if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (reading packets)");
+ if (packet->size != 8) return stbi__errpuc("bad format","packet isn't 8bpp");
+ } while (chained);
+
+ *comp = (act_comp & 0x10 ? 4 : 3); // has alpha channel?
+
+ for(y=0; y<height; ++y) {
+ int packet_idx;
+
+ for(packet_idx=0; packet_idx < num_packets; ++packet_idx) {
+ stbi__pic_packet *packet = &packets[packet_idx];
+ stbi_uc *dest = result+y*width*4;
+
+ switch (packet->type) {
+ default:
+ return stbi__errpuc("bad format","packet has bad compression type");
+
+ case 0: {//uncompressed
+ int x;
+
+ for(x=0;x<width;++x, dest+=4)
+ if (!stbi__readval(s,packet->channel,dest))
+ return 0;
+ break;
+ }
+
+ case 1://Pure RLE
+ {
+ int left=width, i;
+
+ while (left>0) {
+ stbi_uc count,value[4];
+
+ count=stbi__get8(s);
+ if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (pure read count)");
+
+ if (count > left)
+ count = (stbi_uc) left;
+
+ if (!stbi__readval(s,packet->channel,value)) return 0;
+
+ for(i=0; i<count; ++i,dest+=4)
+ stbi__copyval(packet->channel,dest,value);
+ left -= count;
+ }
+ }
+ break;
+
+ case 2: {//Mixed RLE
+ int left=width;
+ while (left>0) {
+ int count = stbi__get8(s), i;
+ if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (mixed read count)");
+
+ if (count >= 128) { // Repeated
+ stbi_uc value[4];
+
+ if (count==128)
+ count = stbi__get16be(s);
+ else
+ count -= 127;
+ if (count > left)
+ return stbi__errpuc("bad file","scanline overrun");
+
+ if (!stbi__readval(s,packet->channel,value))
+ return 0;
+
+ for(i=0;i<count;++i, dest += 4)
+ stbi__copyval(packet->channel,dest,value);
+ } else { // Raw
+ ++count;
+ if (count>left) return stbi__errpuc("bad file","scanline overrun");
+
+ for(i=0;i<count;++i, dest+=4)
+ if (!stbi__readval(s,packet->channel,dest))
+ return 0;
+ }
+ left-=count;
+ }
+ break;
+ }
+ }
+ }
+ }
+
+ return result;
+}
+
+static void *stbi__pic_load(stbi__context *s,int *px,int *py,int *comp,int req_comp, stbi__result_info *ri)
+{
+ stbi_uc *result;
+ int i, x,y, internal_comp;
+ STBI_NOTUSED(ri);
+
+ if (!comp) comp = &internal_comp;
+
+ for (i=0; i<92; ++i)
+ stbi__get8(s);
+
+ x = stbi__get16be(s);
+ y = stbi__get16be(s);
+ if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (pic header)");
+ if (!stbi__mad3sizes_valid(x, y, 4, 0)) return stbi__errpuc("too large", "PIC image too large to decode");
+
+ stbi__get32be(s); //skip `ratio'
+ stbi__get16be(s); //skip `fields'
+ stbi__get16be(s); //skip `pad'
+
+ // intermediate buffer is RGBA
+ result = (stbi_uc *) stbi__malloc_mad3(x, y, 4, 0);
+ memset(result, 0xff, x*y*4);
+
+ if (!stbi__pic_load_core(s,x,y,comp, result)) {
+ STBI_FREE(result);
+ result=0;
+ }
+ *px = x;
+ *py = y;
+ if (req_comp == 0) req_comp = *comp;
+ result=stbi__convert_format(result,4,req_comp,x,y);
+
+ return result;
+}
+
+static int stbi__pic_test(stbi__context *s)
+{
+ int r = stbi__pic_test_core(s);
+ stbi__rewind(s);
+ return r;
+}
+#endif
+
+// *************************************************************************************************
+// GIF loader -- public domain by Jean-Marc Lienher -- simplified/shrunk by stb
+
+#ifndef STBI_NO_GIF
+typedef struct
+{
+ stbi__int16 prefix;
+ stbi_uc first;
+ stbi_uc suffix;
+} stbi__gif_lzw;
+
+typedef struct
+{
+ int w,h;
+ stbi_uc *out, *old_out; // output buffer (always 4 components)
+ int flags, bgindex, ratio, transparent, eflags, delay;
+ stbi_uc pal[256][4];
+ stbi_uc lpal[256][4];
+ stbi__gif_lzw codes[4096];
+ stbi_uc *color_table;
+ int parse, step;
+ int lflags;
+ int start_x, start_y;
+ int max_x, max_y;
+ int cur_x, cur_y;
+ int line_size;
+} stbi__gif;
+
+static int stbi__gif_test_raw(stbi__context *s)
+{
+ int sz;
+ if (stbi__get8(s) != 'G' || stbi__get8(s) != 'I' || stbi__get8(s) != 'F' || stbi__get8(s) != '8') return 0;
+ sz = stbi__get8(s);
+ if (sz != '9' && sz != '7') return 0;
+ if (stbi__get8(s) != 'a') return 0;
+ return 1;
+}
+
+static int stbi__gif_test(stbi__context *s)
+{
+ int r = stbi__gif_test_raw(s);
+ stbi__rewind(s);
+ return r;
+}
+
+static void stbi__gif_parse_colortable(stbi__context *s, stbi_uc pal[256][4], int num_entries, int transp)
+{
+ int i;
+ for (i=0; i < num_entries; ++i) {
+ pal[i][2] = stbi__get8(s);
+ pal[i][1] = stbi__get8(s);
+ pal[i][0] = stbi__get8(s);
+ pal[i][3] = transp == i ? 0 : 255;
+ }
+}
+
+static int stbi__gif_header(stbi__context *s, stbi__gif *g, int *comp, int is_info)
+{
+ stbi_uc version;
+ if (stbi__get8(s) != 'G' || stbi__get8(s) != 'I' || stbi__get8(s) != 'F' || stbi__get8(s) != '8')
+ return stbi__err("not GIF", "Corrupt GIF");
+
+ version = stbi__get8(s);
+ if (version != '7' && version != '9') return stbi__err("not GIF", "Corrupt GIF");
+ if (stbi__get8(s) != 'a') return stbi__err("not GIF", "Corrupt GIF");
+
+ stbi__g_failure_reason = "";
+ g->w = stbi__get16le(s);
+ g->h = stbi__get16le(s);
+ g->flags = stbi__get8(s);
+ g->bgindex = stbi__get8(s);
+ g->ratio = stbi__get8(s);
+ g->transparent = -1;
+
+ if (comp != 0) *comp = 4; // can't actually tell whether it's 3 or 4 until we parse the comments
+
+ if (is_info) return 1;
+
+ if (g->flags & 0x80)
+ stbi__gif_parse_colortable(s,g->pal, 2 << (g->flags & 7), -1);
+
+ return 1;
+}
+
+static int stbi__gif_info_raw(stbi__context *s, int *x, int *y, int *comp)
+{
+ stbi__gif* g = (stbi__gif*) stbi__malloc(sizeof(stbi__gif));
+ if (!stbi__gif_header(s, g, comp, 1)) {
+ STBI_FREE(g);
+ stbi__rewind( s );
+ return 0;
+ }
+ if (x) *x = g->w;
+ if (y) *y = g->h;
+ STBI_FREE(g);
+ return 1;
+}
+
+static void stbi__out_gif_code(stbi__gif *g, stbi__uint16 code)
+{
+ stbi_uc *p, *c;
+
+ // recurse to decode the prefixes, since the linked-list is backwards,
+ // and working backwards through an interleaved image would be nasty
+ if (g->codes[code].prefix >= 0)
+ stbi__out_gif_code(g, g->codes[code].prefix);
+
+ if (g->cur_y >= g->max_y) return;
+
+ p = &g->out[g->cur_x + g->cur_y];
+ c = &g->color_table[g->codes[code].suffix * 4];
+
+ if (c[3] >= 128) {
+ p[0] = c[2];
+ p[1] = c[1];
+ p[2] = c[0];
+ p[3] = c[3];
+ }
+ g->cur_x += 4;
+
+ if (g->cur_x >= g->max_x) {
+ g->cur_x = g->start_x;
+ g->cur_y += g->step;
+
+ while (g->cur_y >= g->max_y && g->parse > 0) {
+ g->step = (1 << g->parse) * g->line_size;
+ g->cur_y = g->start_y + (g->step >> 1);
+ --g->parse;
+ }
+ }
+}
+
+static stbi_uc *stbi__process_gif_raster(stbi__context *s, stbi__gif *g)
+{
+ stbi_uc lzw_cs;
+ stbi__int32 len, init_code;
+ stbi__uint32 first;
+ stbi__int32 codesize, codemask, avail, oldcode, bits, valid_bits, clear;
+ stbi__gif_lzw *p;
+
+ lzw_cs = stbi__get8(s);
+ if (lzw_cs > 12) return NULL;
+ clear = 1 << lzw_cs;
+ first = 1;
+ codesize = lzw_cs + 1;
+ codemask = (1 << codesize) - 1;
+ bits = 0;
+ valid_bits = 0;
+ for (init_code = 0; init_code < clear; init_code++) {
+ g->codes[init_code].prefix = -1;
+ g->codes[init_code].first = (stbi_uc) init_code;
+ g->codes[init_code].suffix = (stbi_uc) init_code;
+ }
+
+ // support no starting clear code
+ avail = clear+2;
+ oldcode = -1;
+
+ len = 0;
+ for(;;) {
+ if (valid_bits < codesize) {
+ if (len == 0) {
+ len = stbi__get8(s); // start new block
+ if (len == 0)
+ return g->out;
+ }
+ --len;
+ bits |= (stbi__int32) stbi__get8(s) << valid_bits;
+ valid_bits += 8;
+ } else {
+ stbi__int32 code = bits & codemask;
+ bits >>= codesize;
+ valid_bits -= codesize;
+ // @OPTIMIZE: is there some way we can accelerate the non-clear path?
+ if (code == clear) { // clear code
+ codesize = lzw_cs + 1;
+ codemask = (1 << codesize) - 1;
+ avail = clear + 2;
+ oldcode = -1;
+ first = 0;
+ } else if (code == clear + 1) { // end of stream code
+ stbi__skip(s, len);
+ while ((len = stbi__get8(s)) > 0)
+ stbi__skip(s,len);
+ return g->out;
+ } else if (code <= avail) {
+ if (first) return stbi__errpuc("no clear code", "Corrupt GIF");
+
+ if (oldcode >= 0) {
+ p = &g->codes[avail++];
+ if (avail > 4096) return stbi__errpuc("too many codes", "Corrupt GIF");
+ p->prefix = (stbi__int16) oldcode;
+ p->first = g->codes[oldcode].first;
+ p->suffix = (code == avail) ? p->first : g->codes[code].first;
+ } else if (code == avail)
+ return stbi__errpuc("illegal code in raster", "Corrupt GIF");
+
+ stbi__out_gif_code(g, (stbi__uint16) code);
+
+ if ((avail & codemask) == 0 && avail <= 0x0FFF) {
+ codesize++;
+ codemask = (1 << codesize) - 1;
+ }
+
+ oldcode = code;
+ } else {
+ return stbi__errpuc("illegal code in raster", "Corrupt GIF");
+ }
+ }
+ }
+}
+
+static void stbi__fill_gif_background(stbi__gif *g, int x0, int y0, int x1, int y1)
+{
+ int x, y;
+ stbi_uc *c = g->pal[g->bgindex];
+ for (y = y0; y < y1; y += 4 * g->w) {
+ for (x = x0; x < x1; x += 4) {
+ stbi_uc *p = &g->out[y + x];
+ p[0] = c[2];
+ p[1] = c[1];
+ p[2] = c[0];
+ p[3] = 0;
+ }
+ }
+}
+
+// this function is designed to support animated gifs, although stb_image doesn't support it
+static stbi_uc *stbi__gif_load_next(stbi__context *s, stbi__gif *g, int *comp, int req_comp)
+{
+ int i;
+ stbi_uc *prev_out = 0;
+
+ if (g->out == 0 && !stbi__gif_header(s, g, comp,0))
+ return 0; // stbi__g_failure_reason set by stbi__gif_header
+
+ if (!stbi__mad3sizes_valid(g->w, g->h, 4, 0))
+ return stbi__errpuc("too large", "GIF too large");
+
+ prev_out = g->out;
+ g->out = (stbi_uc *) stbi__malloc_mad3(4, g->w, g->h, 0);
+ if (g->out == 0) return stbi__errpuc("outofmem", "Out of memory");
+
+ switch ((g->eflags & 0x1C) >> 2) {
+ case 0: // unspecified (also always used on 1st frame)
+ stbi__fill_gif_background(g, 0, 0, 4 * g->w, 4 * g->w * g->h);
+ break;
+ case 1: // do not dispose
+ if (prev_out) memcpy(g->out, prev_out, 4 * g->w * g->h);
+ g->old_out = prev_out;
+ break;
+ case 2: // dispose to background
+ if (prev_out) memcpy(g->out, prev_out, 4 * g->w * g->h);
+ stbi__fill_gif_background(g, g->start_x, g->start_y, g->max_x, g->max_y);
+ break;
+ case 3: // dispose to previous
+ if (g->old_out) {
+ for (i = g->start_y; i < g->max_y; i += 4 * g->w)
+ memcpy(&g->out[i + g->start_x], &g->old_out[i + g->start_x], g->max_x - g->start_x);
+ }
+ break;
+ }
+
+ for (;;) {
+ switch (stbi__get8(s)) {
+ case 0x2C: /* Image Descriptor */
+ {
+ int prev_trans = -1;
+ stbi__int32 x, y, w, h;
+ stbi_uc *o;
+
+ x = stbi__get16le(s);
+ y = stbi__get16le(s);
+ w = stbi__get16le(s);
+ h = stbi__get16le(s);
+ if (((x + w) > (g->w)) || ((y + h) > (g->h)))
+ return stbi__errpuc("bad Image Descriptor", "Corrupt GIF");
+
+ g->line_size = g->w * 4;
+ g->start_x = x * 4;
+ g->start_y = y * g->line_size;
+ g->max_x = g->start_x + w * 4;
+ g->max_y = g->start_y + h * g->line_size;
+ g->cur_x = g->start_x;
+ g->cur_y = g->start_y;
+
+ g->lflags = stbi__get8(s);
+
+ if (g->lflags & 0x40) {
+ g->step = 8 * g->line_size; // first interlaced spacing
+ g->parse = 3;
+ } else {
+ g->step = g->line_size;
+ g->parse = 0;
+ }
+
+ if (g->lflags & 0x80) {
+ stbi__gif_parse_colortable(s,g->lpal, 2 << (g->lflags & 7), g->eflags & 0x01 ? g->transparent : -1);
+ g->color_table = (stbi_uc *) g->lpal;
+ } else if (g->flags & 0x80) {
+ if (g->transparent >= 0 && (g->eflags & 0x01)) {
+ prev_trans = g->pal[g->transparent][3];
+ g->pal[g->transparent][3] = 0;
+ }
+ g->color_table = (stbi_uc *) g->pal;
+ } else
+ return stbi__errpuc("missing color table", "Corrupt GIF");
+
+ o = stbi__process_gif_raster(s, g);
+ if (o == NULL) return NULL;
+
+ if (prev_trans != -1)
+ g->pal[g->transparent][3] = (stbi_uc) prev_trans;
+
+ return o;
+ }
+
+ case 0x21: // Comment Extension.
+ {
+ int len;
+ if (stbi__get8(s) == 0xF9) { // Graphic Control Extension.
+ len = stbi__get8(s);
+ if (len == 4) {
+ g->eflags = stbi__get8(s);
+ g->delay = stbi__get16le(s);
+ g->transparent = stbi__get8(s);
+ } else {
+ stbi__skip(s, len);
+ break;
+ }
+ }
+ while ((len = stbi__get8(s)) != 0)
+ stbi__skip(s, len);
+ break;
+ }
+
+ case 0x3B: // gif stream termination code
+ return (stbi_uc *) s; // using '1' causes warning on some compilers
+
+ default:
+ return stbi__errpuc("unknown code", "Corrupt GIF");
+ }
+ }
+
+ STBI_NOTUSED(req_comp);
+}
+
+static void *stbi__gif_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+ stbi_uc *u = 0;
+ stbi__gif* g = (stbi__gif*) stbi__malloc(sizeof(stbi__gif));
+ memset(g, 0, sizeof(*g));
+ STBI_NOTUSED(ri);
+
+ u = stbi__gif_load_next(s, g, comp, req_comp);
+ if (u == (stbi_uc *) s) u = 0; // end of animated gif marker
+ if (u) {
+ *x = g->w;
+ *y = g->h;
+ if (req_comp && req_comp != 4)
+ u = stbi__convert_format(u, 4, req_comp, g->w, g->h);
+ }
+ else if (g->out)
+ STBI_FREE(g->out);
+ STBI_FREE(g);
+ return u;
+}
+
+static int stbi__gif_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ return stbi__gif_info_raw(s,x,y,comp);
+}
+#endif
+
+// *************************************************************************************************
+// Radiance RGBE HDR loader
+// originally by Nicolas Schulz
+#ifndef STBI_NO_HDR
+static int stbi__hdr_test_core(stbi__context *s, const char *signature)
+{
+ int i;
+ for (i=0; signature[i]; ++i)
+ if (stbi__get8(s) != signature[i])
+ return 0;
+ stbi__rewind(s);
+ return 1;
+}
+
+static int stbi__hdr_test(stbi__context* s)
+{
+ int r = stbi__hdr_test_core(s, "#?RADIANCE\n");
+ stbi__rewind(s);
+ if(!r) {
+ r = stbi__hdr_test_core(s, "#?RGBE\n");
+ stbi__rewind(s);
+ }
+ return r;
+}
+
+#define STBI__HDR_BUFLEN 1024
+static char *stbi__hdr_gettoken(stbi__context *z, char *buffer)
+{
+ int len=0;
+ char c = '\0';
+
+ c = (char) stbi__get8(z);
+
+ while (!stbi__at_eof(z) && c != '\n') {
+ buffer[len++] = c;
+ if (len == STBI__HDR_BUFLEN-1) {
+ // flush to end of line
+ while (!stbi__at_eof(z) && stbi__get8(z) != '\n')
+ ;
+ break;
+ }
+ c = (char) stbi__get8(z);
+ }
+
+ buffer[len] = 0;
+ return buffer;
+}
+
+static void stbi__hdr_convert(float *output, stbi_uc *input, int req_comp)
+{
+ if ( input[3] != 0 ) {
+ float f1;
+ // Exponent
+ f1 = (float) ldexp(1.0f, input[3] - (int)(128 + 8));
+ if (req_comp <= 2)
+ output[0] = (input[0] + input[1] + input[2]) * f1 / 3;
+ else {
+ output[0] = input[0] * f1;
+ output[1] = input[1] * f1;
+ output[2] = input[2] * f1;
+ }
+ if (req_comp == 2) output[1] = 1;
+ if (req_comp == 4) output[3] = 1;
+ } else {
+ switch (req_comp) {
+ case 4: output[3] = 1; /* fallthrough */
+ case 3: output[0] = output[1] = output[2] = 0;
+ break;
+ case 2: output[1] = 1; /* fallthrough */
+ case 1: output[0] = 0;
+ break;
+ }
+ }
+}
+
+static float *stbi__hdr_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+ char buffer[STBI__HDR_BUFLEN];
+ char *token;
+ int valid = 0;
+ int width, height;
+ stbi_uc *scanline;
+ float *hdr_data;
+ int len;
+ unsigned char count, value;
+ int i, j, k, c1,c2, z;
+ const char *headerToken;
+ STBI_NOTUSED(ri);
+
+ // Check identifier
+ headerToken = stbi__hdr_gettoken(s,buffer);
+ if (strcmp(headerToken, "#?RADIANCE") != 0 && strcmp(headerToken, "#?RGBE") != 0)
+ return stbi__errpf("not HDR", "Corrupt HDR image");
+
+ // Parse header
+ for(;;) {
+ token = stbi__hdr_gettoken(s,buffer);
+ if (token[0] == 0) break;
+ if (strcmp(token, "FORMAT=32-bit_rle_rgbe") == 0) valid = 1;
+ }
+
+ if (!valid) return stbi__errpf("unsupported format", "Unsupported HDR format");
+
+ // Parse width and height
+ // can't use sscanf() if we're not using stdio!
+ token = stbi__hdr_gettoken(s,buffer);
+ if (strncmp(token, "-Y ", 3)) return stbi__errpf("unsupported data layout", "Unsupported HDR format");
+ token += 3;
+ height = (int) strtol(token, &token, 10);
+ while (*token == ' ') ++token;
+ if (strncmp(token, "+X ", 3)) return stbi__errpf("unsupported data layout", "Unsupported HDR format");
+ token += 3;
+ width = (int) strtol(token, NULL, 10);
+
+ *x = width;
+ *y = height;
+
+ if (comp) *comp = 3;
+ if (req_comp == 0) req_comp = 3;
+
+ if (!stbi__mad4sizes_valid(width, height, req_comp, sizeof(float), 0))
+ return stbi__errpf("too large", "HDR image is too large");
+
+ // Read data
+ hdr_data = (float *) stbi__malloc_mad4(width, height, req_comp, sizeof(float), 0);
+ if (!hdr_data)
+ return stbi__errpf("outofmem", "Out of memory");
+
+ // Load image data
+ // image data is stored as some number of sca
+ if ( width < 8 || width >= 32768) {
+ // Read flat data
+ for (j=0; j < height; ++j) {
+ for (i=0; i < width; ++i) {
+ stbi_uc rgbe[4];
+ main_decode_loop:
+ stbi__getn(s, rgbe, 4);
+ stbi__hdr_convert(hdr_data + j * width * req_comp + i * req_comp, rgbe, req_comp);
+ }
+ }
+ } else {
+ // Read RLE-encoded data
+ scanline = NULL;
+
+ for (j = 0; j < height; ++j) {
+ c1 = stbi__get8(s);
+ c2 = stbi__get8(s);
+ len = stbi__get8(s);
+ if (c1 != 2 || c2 != 2 || (len & 0x80)) {
+ // not run-length encoded, so we have to actually use THIS data as a decoded
+ // pixel (note this can't be a valid pixel--one of RGB must be >= 128)
+ stbi_uc rgbe[4];
+ rgbe[0] = (stbi_uc) c1;
+ rgbe[1] = (stbi_uc) c2;
+ rgbe[2] = (stbi_uc) len;
+ rgbe[3] = (stbi_uc) stbi__get8(s);
+ stbi__hdr_convert(hdr_data, rgbe, req_comp);
+ i = 1;
+ j = 0;
+ STBI_FREE(scanline);
+ goto main_decode_loop; // yes, this makes no sense
+ }
+ len <<= 8;
+ len |= stbi__get8(s);
+ if (len != width) { STBI_FREE(hdr_data); STBI_FREE(scanline); return stbi__errpf("invalid decoded scanline length", "corrupt HDR"); }
+ if (scanline == NULL) {
+ scanline = (stbi_uc *) stbi__malloc_mad2(width, 4, 0);
+ if (!scanline) {
+ STBI_FREE(hdr_data);
+ return stbi__errpf("outofmem", "Out of memory");
+ }
+ }
+
+ for (k = 0; k < 4; ++k) {
+ int nleft;
+ i = 0;
+ while ((nleft = width - i) > 0) {
+ count = stbi__get8(s);
+ if (count > 128) {
+ // Run
+ value = stbi__get8(s);
+ count -= 128;
+ if (count > nleft) { STBI_FREE(hdr_data); STBI_FREE(scanline); return stbi__errpf("corrupt", "bad RLE data in HDR"); }
+ for (z = 0; z < count; ++z)
+ scanline[i++ * 4 + k] = value;
+ } else {
+ // Dump
+ if (count > nleft) { STBI_FREE(hdr_data); STBI_FREE(scanline); return stbi__errpf("corrupt", "bad RLE data in HDR"); }
+ for (z = 0; z < count; ++z)
+ scanline[i++ * 4 + k] = stbi__get8(s);
+ }
+ }
+ }
+ for (i=0; i < width; ++i)
+ stbi__hdr_convert(hdr_data+(j*width + i)*req_comp, scanline + i*4, req_comp);
+ }
+ if (scanline)
+ STBI_FREE(scanline);
+ }
+
+ return hdr_data;
+}
+
+static int stbi__hdr_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ char buffer[STBI__HDR_BUFLEN];
+ char *token;
+ int valid = 0;
+ int dummy;
+
+ if (!x) x = &dummy;
+ if (!y) y = &dummy;
+ if (!comp) comp = &dummy;
+
+ if (stbi__hdr_test(s) == 0) {
+ stbi__rewind( s );
+ return 0;
+ }
+
+ for(;;) {
+ token = stbi__hdr_gettoken(s,buffer);
+ if (token[0] == 0) break;
+ if (strcmp(token, "FORMAT=32-bit_rle_rgbe") == 0) valid = 1;
+ }
+
+ if (!valid) {
+ stbi__rewind( s );
+ return 0;
+ }
+ token = stbi__hdr_gettoken(s,buffer);
+ if (strncmp(token, "-Y ", 3)) {
+ stbi__rewind( s );
+ return 0;
+ }
+ token += 3;
+ *y = (int) strtol(token, &token, 10);
+ while (*token == ' ') ++token;
+ if (strncmp(token, "+X ", 3)) {
+ stbi__rewind( s );
+ return 0;
+ }
+ token += 3;
+ *x = (int) strtol(token, NULL, 10);
+ *comp = 3;
+ return 1;
+}
+#endif // STBI_NO_HDR
+
+#ifndef STBI_NO_BMP
+static int stbi__bmp_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ void *p;
+ stbi__bmp_data info;
+
+ info.all_a = 255;
+ p = stbi__bmp_parse_header(s, &info);
+ stbi__rewind( s );
+ if (p == NULL)
+ return 0;
+ if (x) *x = s->img_x;
+ if (y) *y = s->img_y;
+ if (comp) *comp = info.ma ? 4 : 3;
+ return 1;
+}
+#endif
+
+#ifndef STBI_NO_PSD
+static int stbi__psd_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ int channelCount, dummy;
+ if (!x) x = &dummy;
+ if (!y) y = &dummy;
+ if (!comp) comp = &dummy;
+ if (stbi__get32be(s) != 0x38425053) {
+ stbi__rewind( s );
+ return 0;
+ }
+ if (stbi__get16be(s) != 1) {
+ stbi__rewind( s );
+ return 0;
+ }
+ stbi__skip(s, 6);
+ channelCount = stbi__get16be(s);
+ if (channelCount < 0 || channelCount > 16) {
+ stbi__rewind( s );
+ return 0;
+ }
+ *y = stbi__get32be(s);
+ *x = stbi__get32be(s);
+ if (stbi__get16be(s) != 8) {
+ stbi__rewind( s );
+ return 0;
+ }
+ if (stbi__get16be(s) != 3) {
+ stbi__rewind( s );
+ return 0;
+ }
+ *comp = 4;
+ return 1;
+}
+#endif
+
+#ifndef STBI_NO_PIC
+static int stbi__pic_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ int act_comp=0,num_packets=0,chained,dummy;
+ stbi__pic_packet packets[10];
+
+ if (!x) x = &dummy;
+ if (!y) y = &dummy;
+ if (!comp) comp = &dummy;
+
+ if (!stbi__pic_is4(s,"\x53\x80\xF6\x34")) {
+ stbi__rewind(s);
+ return 0;
+ }
+
+ stbi__skip(s, 88);
+
+ *x = stbi__get16be(s);
+ *y = stbi__get16be(s);
+ if (stbi__at_eof(s)) {
+ stbi__rewind( s);
+ return 0;
+ }
+ if ( (*x) != 0 && (1 << 28) / (*x) < (*y)) {
+ stbi__rewind( s );
+ return 0;
+ }
+
+ stbi__skip(s, 8);
+
+ do {
+ stbi__pic_packet *packet;
+
+ if (num_packets==sizeof(packets)/sizeof(packets[0]))
+ return 0;
+
+ packet = &packets[num_packets++];
+ chained = stbi__get8(s);
+ packet->size = stbi__get8(s);
+ packet->type = stbi__get8(s);
+ packet->channel = stbi__get8(s);
+ act_comp |= packet->channel;
+
+ if (stbi__at_eof(s)) {
+ stbi__rewind( s );
+ return 0;
+ }
+ if (packet->size != 8) {
+ stbi__rewind( s );
+ return 0;
+ }
+ } while (chained);
+
+ *comp = (act_comp & 0x10 ? 4 : 3);
+
+ return 1;
+}
+#endif
+
+// *************************************************************************************************
+// Portable Gray Map and Portable Pixel Map loader
+// by Ken Miller
+//
+// PGM: http://netpbm.sourceforge.net/doc/pgm.html
+// PPM: http://netpbm.sourceforge.net/doc/ppm.html
+//
+// Known limitations:
+// Does not support comments in the header section
+// Does not support ASCII image data (formats P2 and P3)
+// Does not support 16-bit-per-channel
+
+#ifndef STBI_NO_PNM
+
+static int stbi__pnm_test(stbi__context *s)
+{
+ char p, t;
+ p = (char) stbi__get8(s);
+ t = (char) stbi__get8(s);
+ if (p != 'P' || (t != '5' && t != '6')) {
+ stbi__rewind( s );
+ return 0;
+ }
+ return 1;
+}
+
+static void *stbi__pnm_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+ stbi_uc *out;
+ STBI_NOTUSED(ri);
+
+ if (!stbi__pnm_info(s, (int *)&s->img_x, (int *)&s->img_y, (int *)&s->img_n))
+ return 0;
+
+ *x = s->img_x;
+ *y = s->img_y;
+ if (comp) *comp = s->img_n;
+
+ if (!stbi__mad3sizes_valid(s->img_n, s->img_x, s->img_y, 0))
+ return stbi__errpuc("too large", "PNM too large");
+
+ out = (stbi_uc *) stbi__malloc_mad3(s->img_n, s->img_x, s->img_y, 0);
+ if (!out) return stbi__errpuc("outofmem", "Out of memory");
+ stbi__getn(s, out, s->img_n * s->img_x * s->img_y);
+
+ if (req_comp && req_comp != s->img_n) {
+ out = stbi__convert_format(out, s->img_n, req_comp, s->img_x, s->img_y);
+ if (out == NULL) return out; // stbi__convert_format frees input on failure
+ }
+ return out;
+}
+
+static int stbi__pnm_isspace(char c)
+{
+ return c == ' ' || c == '\t' || c == '\n' || c == '\v' || c == '\f' || c == '\r';
+}
+
+static void stbi__pnm_skip_whitespace(stbi__context *s, char *c)
+{
+ for (;;) {
+ while (!stbi__at_eof(s) && stbi__pnm_isspace(*c))
+ *c = (char) stbi__get8(s);
+
+ if (stbi__at_eof(s) || *c != '#')
+ break;
+
+ while (!stbi__at_eof(s) && *c != '\n' && *c != '\r' )
+ *c = (char) stbi__get8(s);
+ }
+}
+
+static int stbi__pnm_isdigit(char c)
+{
+ return c >= '0' && c <= '9';
+}
+
+static int stbi__pnm_getinteger(stbi__context *s, char *c)
+{
+ int value = 0;
+
+ while (!stbi__at_eof(s) && stbi__pnm_isdigit(*c)) {
+ value = value*10 + (*c - '0');
+ *c = (char) stbi__get8(s);
+ }
+
+ return value;
+}
+
+static int stbi__pnm_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ int maxv, dummy;
+ char c, p, t;
+
+ if (!x) x = &dummy;
+ if (!y) y = &dummy;
+ if (!comp) comp = &dummy;
+
+ stbi__rewind(s);
+
+ // Get identifier
+ p = (char) stbi__get8(s);
+ t = (char) stbi__get8(s);
+ if (p != 'P' || (t != '5' && t != '6')) {
+ stbi__rewind(s);
+ return 0;
+ }
+
+ *comp = (t == '6') ? 3 : 1; // '5' is 1-component .pgm; '6' is 3-component .ppm
+
+ c = (char) stbi__get8(s);
+ stbi__pnm_skip_whitespace(s, &c);
+
+ *x = stbi__pnm_getinteger(s, &c); // read width
+ stbi__pnm_skip_whitespace(s, &c);
+
+ *y = stbi__pnm_getinteger(s, &c); // read height
+ stbi__pnm_skip_whitespace(s, &c);
+
+ maxv = stbi__pnm_getinteger(s, &c); // read max value
+
+ if (maxv > 255)
+ return stbi__err("max value > 255", "PPM image not 8-bit");
+ else
+ return 1;
+}
+#endif
+
+static int stbi__info_main(stbi__context *s, int *x, int *y, int *comp)
+{
+ #ifndef STBI_NO_JPEG
+ if (stbi__jpeg_info(s, x, y, comp)) return 1;
+ #endif
+
+ #ifndef STBI_NO_PNG
+ if (stbi__png_info(s, x, y, comp)) return 1;
+ #endif
+
+ #ifndef STBI_NO_GIF
+ if (stbi__gif_info(s, x, y, comp)) return 1;
+ #endif
+
+ #ifndef STBI_NO_BMP
+ if (stbi__bmp_info(s, x, y, comp)) return 1;
+ #endif
+
+ #ifndef STBI_NO_PSD
+ if (stbi__psd_info(s, x, y, comp)) return 1;
+ #endif
+
+ #ifndef STBI_NO_PIC
+ if (stbi__pic_info(s, x, y, comp)) return 1;
+ #endif
+
+ #ifndef STBI_NO_PNM
+ if (stbi__pnm_info(s, x, y, comp)) return 1;
+ #endif
+
+ #ifndef STBI_NO_HDR
+ if (stbi__hdr_info(s, x, y, comp)) return 1;
+ #endif
+
+ // test tga last because it's a crappy test!
+ #ifndef STBI_NO_TGA
+ if (stbi__tga_info(s, x, y, comp))
+ return 1;
+ #endif
+ return stbi__err("unknown image type", "Image not of any known type, or corrupt");
+}
+
+#ifndef STBI_NO_STDIO
+STBIDEF int stbi_info(char const *filename, int *x, int *y, int *comp)
+{
+ FILE *f = stbi__fopen(filename, "rb");
+ int result;
+ if (!f) return stbi__err("can't fopen", "Unable to open file");
+ result = stbi_info_from_file(f, x, y, comp);
+ fclose(f);
+ return result;
+}
+
+STBIDEF int stbi_info_from_file(FILE *f, int *x, int *y, int *comp)
+{
+ int r;
+ stbi__context s;
+ long pos = ftell(f);
+ stbi__start_file(&s, f);
+ r = stbi__info_main(&s,x,y,comp);
+ fseek(f,pos,SEEK_SET);
+ return r;
+}
+#endif // !STBI_NO_STDIO
+
+STBIDEF int stbi_info_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp)
+{
+ stbi__context s;
+ stbi__start_mem(&s,buffer,len);
+ return stbi__info_main(&s,x,y,comp);
+}
+
+STBIDEF int stbi_info_from_callbacks(stbi_io_callbacks const *c, void *user, int *x, int *y, int *comp)
+{
+ stbi__context s;
+ stbi__start_callbacks(&s, (stbi_io_callbacks *) c, user);
+ return stbi__info_main(&s,x,y,comp);
+}
+
+#endif // STB_IMAGE_IMPLEMENTATION
+
+/*
+ revision history:
+ 2.15 (2017-03-18) fix png-1,2,4 bug; now all Imagenet JPGs decode;
+ warning fixes; disable run-time SSE detection on gcc;
+ uniform handling of optional "return" values;
+ thread-safe initialization of zlib tables
+ 2.14 (2017-03-03) remove deprecated STBI_JPEG_OLD; fixes for Imagenet JPGs
+ 2.13 (2016-11-29) add 16-bit API, only supported for PNG right now
+ 2.12 (2016-04-02) fix typo in 2.11 PSD fix that caused crashes
+ 2.11 (2016-04-02) allocate large structures on the stack
+ remove white matting for transparent PSD
+ fix reported channel count for PNG & BMP
+ re-enable SSE2 in non-gcc 64-bit
+ support RGB-formatted JPEG
+ read 16-bit PNGs (only as 8-bit)
+ 2.10 (2016-01-22) avoid warning introduced in 2.09 by STBI_REALLOC_SIZED
+ 2.09 (2016-01-16) allow comments in PNM files
+ 16-bit-per-pixel TGA (not bit-per-component)
+ info() for TGA could break due to .hdr handling
+ info() for BMP to shares code instead of sloppy parse
+ can use STBI_REALLOC_SIZED if allocator doesn't support realloc
+ code cleanup
+ 2.08 (2015-09-13) fix to 2.07 cleanup, reading RGB PSD as RGBA
+ 2.07 (2015-09-13) fix compiler warnings
+ partial animated GIF support
+ limited 16-bpc PSD support
+ #ifdef unused functions
+ bug with < 92 byte PIC,PNM,HDR,TGA
+ 2.06 (2015-04-19) fix bug where PSD returns wrong '*comp' value
+ 2.05 (2015-04-19) fix bug in progressive JPEG handling, fix warning
+ 2.04 (2015-04-15) try to re-enable SIMD on MinGW 64-bit
+ 2.03 (2015-04-12) extra corruption checking (mmozeiko)
+ stbi_set_flip_vertically_on_load (nguillemot)
+ fix NEON support; fix mingw support
+ 2.02 (2015-01-19) fix incorrect assert, fix warning
+ 2.01 (2015-01-17) fix various warnings; suppress SIMD on gcc 32-bit without -msse2
+ 2.00b (2014-12-25) fix STBI_MALLOC in progressive JPEG
+ 2.00 (2014-12-25) optimize JPG, including x86 SSE2 & NEON SIMD (ryg)
+ progressive JPEG (stb)
+ PGM/PPM support (Ken Miller)
+ STBI_MALLOC,STBI_REALLOC,STBI_FREE
+ GIF bugfix -- seemingly never worked
+ STBI_NO_*, STBI_ONLY_*
+ 1.48 (2014-12-14) fix incorrectly-named assert()
+ 1.47 (2014-12-14) 1/2/4-bit PNG support, both direct and paletted (Omar Cornut & stb)
+ optimize PNG (ryg)
+ fix bug in interlaced PNG with user-specified channel count (stb)
+ 1.46 (2014-08-26)
+ fix broken tRNS chunk (colorkey-style transparency) in non-paletted PNG
+ 1.45 (2014-08-16)
+ fix MSVC-ARM internal compiler error by wrapping malloc
+ 1.44 (2014-08-07)
+ various warning fixes from Ronny Chevalier
+ 1.43 (2014-07-15)
+ fix MSVC-only compiler problem in code changed in 1.42
+ 1.42 (2014-07-09)
+ don't define _CRT_SECURE_NO_WARNINGS (affects user code)
+ fixes to stbi__cleanup_jpeg path
+ added STBI_ASSERT to avoid requiring assert.h
+ 1.41 (2014-06-25)
+ fix search&replace from 1.36 that messed up comments/error messages
+ 1.40 (2014-06-22)
+ fix gcc struct-initialization warning
+ 1.39 (2014-06-15)
+ fix to TGA optimization when req_comp != number of components in TGA;
+ fix to GIF loading because BMP wasn't rewinding (whoops, no GIFs in my test suite)
+ add support for BMP version 5 (more ignored fields)
+ 1.38 (2014-06-06)
+ suppress MSVC warnings on integer casts truncating values
+ fix accidental rename of 'skip' field of I/O
+ 1.37 (2014-06-04)
+ remove duplicate typedef
+ 1.36 (2014-06-03)
+ convert to header file single-file library
+ if de-iphone isn't set, load iphone images color-swapped instead of returning NULL
+ 1.35 (2014-05-27)
+ various warnings
+ fix broken STBI_SIMD path
+ fix bug where stbi_load_from_file no longer left file pointer in correct place
+ fix broken non-easy path for 32-bit BMP (possibly never used)
+ TGA optimization by Arseny Kapoulkine
+ 1.34 (unknown)
+ use STBI_NOTUSED in stbi__resample_row_generic(), fix one more leak in tga failure case
+ 1.33 (2011-07-14)
+ make stbi_is_hdr work in STBI_NO_HDR (as specified), minor compiler-friendly improvements
+ 1.32 (2011-07-13)
+ support for "info" function for all supported filetypes (SpartanJ)
+ 1.31 (2011-06-20)
+ a few more leak fixes, bug in PNG handling (SpartanJ)
+ 1.30 (2011-06-11)
+ added ability to load files via callbacks to accomidate custom input streams (Ben Wenger)
+ removed deprecated format-specific test/load functions
+ removed support for installable file formats (stbi_loader) -- would have been broken for IO callbacks anyway
+ error cases in bmp and tga give messages and don't leak (Raymond Barbiero, grisha)
+ fix inefficiency in decoding 32-bit BMP (David Woo)
+ 1.29 (2010-08-16)
+ various warning fixes from Aurelien Pocheville
+ 1.28 (2010-08-01)
+ fix bug in GIF palette transparency (SpartanJ)
+ 1.27 (2010-08-01)
+ cast-to-stbi_uc to fix warnings
+ 1.26 (2010-07-24)
+ fix bug in file buffering for PNG reported by SpartanJ
+ 1.25 (2010-07-17)
+ refix trans_data warning (Won Chun)
+ 1.24 (2010-07-12)
+ perf improvements reading from files on platforms with lock-heavy fgetc()
+ minor perf improvements for jpeg
+ deprecated type-specific functions so we'll get feedback if they're needed
+ attempt to fix trans_data warning (Won Chun)
+ 1.23 fixed bug in iPhone support
+ 1.22 (2010-07-10)
+ removed image *writing* support
+ stbi_info support from Jetro Lauha
+ GIF support from Jean-Marc Lienher
+ iPhone PNG-extensions from James Brown
+ warning-fixes from Nicolas Schulz and Janez Zemva (i.stbi__err. Janez (U+017D)emva)
+ 1.21 fix use of 'stbi_uc' in header (reported by jon blow)
+ 1.20 added support for Softimage PIC, by Tom Seddon
+ 1.19 bug in interlaced PNG corruption check (found by ryg)
+ 1.18 (2008-08-02)
+ fix a threading bug (local mutable static)
+ 1.17 support interlaced PNG
+ 1.16 major bugfix - stbi__convert_format converted one too many pixels
+ 1.15 initialize some fields for thread safety
+ 1.14 fix threadsafe conversion bug
+ header-file-only version (#define STBI_HEADER_FILE_ONLY before including)
+ 1.13 threadsafe
+ 1.12 const qualifiers in the API
+ 1.11 Support installable IDCT, colorspace conversion routines
+ 1.10 Fixes for 64-bit (don't use "unsigned long")
+ optimized upsampling by Fabian "ryg" Giesen
+ 1.09 Fix format-conversion for PSD code (bad global variables!)
+ 1.08 Thatcher Ulrich's PSD code integrated by Nicolas Schulz
+ 1.07 attempt to fix C++ warning/errors again
+ 1.06 attempt to fix C++ warning/errors again
+ 1.05 fix TGA loading to return correct *comp and use good luminance calc
+ 1.04 default float alpha is 1, not 255; use 'void *' for stbi_image_free
+ 1.03 bugfixes to STBI_NO_STDIO, STBI_NO_HDR
+ 1.02 support for (subset of) HDR files, float interface for preferred access to them
+ 1.01 fix bug: possible bug in handling right-side up bmps... not sure
+ fix bug: the stbi__bmp_load() and stbi__tga_load() functions didn't work at all
+ 1.00 interface to zlib that skips zlib header
+ 0.99 correct handling of alpha in palette
+ 0.98 TGA loader by lonesock; dynamically add loaders (untested)
+ 0.97 jpeg errors on too large a file; also catch another malloc failure
+ 0.96 fix detection of invalid v value - particleman@mollyrocket forum
+ 0.95 during header scan, seek to markers in case of padding
+ 0.94 STBI_NO_STDIO to disable stdio usage; rename all #defines the same
+ 0.93 handle jpegtran output; verbose errors
+ 0.92 read 4,8,16,24,32-bit BMP files of several formats
+ 0.91 output 24-bit Windows 3.0 BMP files
+ 0.90 fix a few more warnings; bump version number to approach 1.0
+ 0.61 bugfixes due to Marc LeBlanc, Christopher Lloyd
+ 0.60 fix compiling as c++
+ 0.59 fix warnings: merge Dave Moore's -Wall fixes
+ 0.58 fix bug: zlib uncompressed mode len/nlen was wrong endian
+ 0.57 fix bug: jpg last huffman symbol before marker was >9 bits but less than 16 available
+ 0.56 fix bug: zlib uncompressed mode len vs. nlen
+ 0.55 fix bug: restart_interval not initialized to 0
+ 0.54 allow NULL for 'int *comp'
+ 0.53 fix bug in png 3->4; speedup png decoding
+ 0.52 png handles req_comp=3,4 directly; minor cleanup; jpeg comments
+ 0.51 obey req_comp requests, 1-component jpegs return as 1-component,
+ on 'test' only check type, not whether we support this variant
+ 0.50 (2006-11-19)
+ first released version
+*/
+
+
+/*
+------------------------------------------------------------------------------
+This software is available under 2 licenses -- choose whichever you prefer.
+------------------------------------------------------------------------------
+ALTERNATIVE A - MIT License
+Copyright (c) 2017 Sean Barrett
+Permission is hereby granted, free of charge, to any person obtaining a copy of
+this software and associated documentation files (the "Software"), to deal in
+the Software without restriction, including without limitation the rights to
+use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
+of the Software, and to permit persons to whom the Software is furnished to do
+so, subject to the following conditions:
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+------------------------------------------------------------------------------
+ALTERNATIVE B - Public Domain (www.unlicense.org)
+This is free and unencumbered software released into the public domain.
+Anyone is free to copy, modify, publish, use, compile, sell, or distribute this
+software, either in source code form or as a compiled binary, for any purpose,
+commercial or non-commercial, and by any means.
+In jurisdictions that recognize copyright laws, the author or authors of this
+software dedicate any and all copyright interest in the software to the public
+domain. We make this dedication for the benefit of the public at large and to
+the detriment of our heirs and successors. We intend this dedication to be an
+overt act of relinquishment in perpetuity of all present and future rights to
+this software under copyright law.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+------------------------------------------------------------------------------
+*/
--- /dev/null
+/* stb_image_resize - v0.94 - public domain image resizing
+ by Jorge L Rodriguez (@VinoBS) - 2014
+ http://github.com/nothings/stb
+
+ Written with emphasis on usability, portability, and efficiency. (No
+ SIMD or threads, so it be easily outperformed by libs that use those.)
+ Only scaling and translation is supported, no rotations or shears.
+ Easy API downsamples w/Mitchell filter, upsamples w/cubic interpolation.
+
+ COMPILING & LINKING
+ In one C/C++ file that #includes this file, do this:
+ #define STB_IMAGE_RESIZE_IMPLEMENTATION
+ before the #include. That will create the implementation in that file.
+
+ QUICKSTART
+ stbir_resize_uint8( input_pixels , in_w , in_h , 0,
+ output_pixels, out_w, out_h, 0, num_channels)
+ stbir_resize_float(...)
+ stbir_resize_uint8_srgb( input_pixels , in_w , in_h , 0,
+ output_pixels, out_w, out_h, 0,
+ num_channels , alpha_chan , 0)
+ stbir_resize_uint8_srgb_edgemode(
+ input_pixels , in_w , in_h , 0,
+ output_pixels, out_w, out_h, 0,
+ num_channels , alpha_chan , 0, STBIR_EDGE_CLAMP)
+ // WRAP/REFLECT/ZERO
+
+ FULL API
+ See the "header file" section of the source for API documentation.
+
+ ADDITIONAL DOCUMENTATION
+
+ SRGB & FLOATING POINT REPRESENTATION
+ The sRGB functions presume IEEE floating point. If you do not have
+ IEEE floating point, define STBIR_NON_IEEE_FLOAT. This will use
+ a slower implementation.
+
+ MEMORY ALLOCATION
+ The resize functions here perform a single memory allocation using
+ malloc. To control the memory allocation, before the #include that
+ triggers the implementation, do:
+
+ #define STBIR_MALLOC(size,context) ...
+ #define STBIR_FREE(ptr,context) ...
+
+ Each resize function makes exactly one call to malloc/free, so to use
+ temp memory, store the temp memory in the context and return that.
+
+ ASSERT
+ Define STBIR_ASSERT(boolval) to override assert() and not use assert.h
+
+ OPTIMIZATION
+ Define STBIR_SATURATE_INT to compute clamp values in-range using
+ integer operations instead of float operations. This may be faster
+ on some platforms.
+
+ DEFAULT FILTERS
+ For functions which don't provide explicit control over what filters
+ to use, you can change the compile-time defaults with
+
+ #define STBIR_DEFAULT_FILTER_UPSAMPLE STBIR_FILTER_something
+ #define STBIR_DEFAULT_FILTER_DOWNSAMPLE STBIR_FILTER_something
+
+ See stbir_filter in the header-file section for the list of filters.
+
+ NEW FILTERS
+ A number of 1D filter kernels are used. For a list of
+ supported filters see the stbir_filter enum. To add a new filter,
+ write a filter function and add it to stbir__filter_info_table.
+
+ PROGRESS
+ For interactive use with slow resize operations, you can install
+ a progress-report callback:
+
+ #define STBIR_PROGRESS_REPORT(val) some_func(val)
+
+ The parameter val is a float which goes from 0 to 1 as progress is made.
+
+ For example:
+
+ static void my_progress_report(float progress);
+ #define STBIR_PROGRESS_REPORT(val) my_progress_report(val)
+
+ #define STB_IMAGE_RESIZE_IMPLEMENTATION
+ #include "stb_image_resize.h"
+
+ static void my_progress_report(float progress)
+ {
+ printf("Progress: %f%%\n", progress*100);
+ }
+
+ MAX CHANNELS
+ If your image has more than 64 channels, define STBIR_MAX_CHANNELS
+ to the max you'll have.
+
+ ALPHA CHANNEL
+ Most of the resizing functions provide the ability to control how
+ the alpha channel of an image is processed. The important things
+ to know about this:
+
+ 1. The best mathematically-behaved version of alpha to use is
+ called "premultiplied alpha", in which the other color channels
+ have had the alpha value multiplied in. If you use premultiplied
+ alpha, linear filtering (such as image resampling done by this
+ library, or performed in texture units on GPUs) does the "right
+ thing". While premultiplied alpha is standard in the movie CGI
+ industry, it is still uncommon in the videogame/real-time world.
+
+ If you linearly filter non-premultiplied alpha, strange effects
+ occur. (For example, the 50/50 average of 99% transparent bright green
+ and 1% transparent black produces 50% transparent dark green when
+ non-premultiplied, whereas premultiplied it produces 50%
+ transparent near-black. The former introduces green energy
+ that doesn't exist in the source image.)
+
+ 2. Artists should not edit premultiplied-alpha images; artists
+ want non-premultiplied alpha images. Thus, art tools generally output
+ non-premultiplied alpha images.
+
+ 3. You will get best results in most cases by converting images
+ to premultiplied alpha before processing them mathematically.
+
+ 4. If you pass the flag STBIR_FLAG_ALPHA_PREMULTIPLIED, the
+ resizer does not do anything special for the alpha channel;
+ it is resampled identically to other channels. This produces
+ the correct results for premultiplied-alpha images, but produces
+ less-than-ideal results for non-premultiplied-alpha images.
+
+ 5. If you do not pass the flag STBIR_FLAG_ALPHA_PREMULTIPLIED,
+ then the resizer weights the contribution of input pixels
+ based on their alpha values, or, equivalently, it multiplies
+ the alpha value into the color channels, resamples, then divides
+ by the resultant alpha value. Input pixels which have alpha=0 do
+ not contribute at all to output pixels unless _all_ of the input
+ pixels affecting that output pixel have alpha=0, in which case
+ the result for that pixel is the same as it would be without
+ STBIR_FLAG_ALPHA_PREMULTIPLIED. However, this is only true for
+ input images in integer formats. For input images in float format,
+ input pixels with alpha=0 have no effect, and output pixels
+ which have alpha=0 will be 0 in all channels. (For float images,
+ you can manually achieve the same result by adding a tiny epsilon
+ value to the alpha channel of every image, and then subtracting
+ or clamping it at the end.)
+
+ 6. You can suppress the behavior described in #5 and make
+ all-0-alpha pixels have 0 in all channels by #defining
+ STBIR_NO_ALPHA_EPSILON.
+
+ 7. You can separately control whether the alpha channel is
+ interpreted as linear or affected by the colorspace. By default
+ it is linear; you almost never want to apply the colorspace.
+ (For example, graphics hardware does not apply sRGB conversion
+ to the alpha channel.)
+
+ CONTRIBUTORS
+ Jorge L Rodriguez: Implementation
+ Sean Barrett: API design, optimizations
+ Aras Pranckevicius: bugfix
+
+ REVISIONS
+ 0.94 (2017-03-18) fixed warnings
+ 0.93 (2017-03-03) fixed bug with certain combinations of heights
+ 0.92 (2017-01-02) fix integer overflow on large (>2GB) images
+ 0.91 (2016-04-02) fix warnings; fix handling of subpixel regions
+ 0.90 (2014-09-17) first released version
+
+ LICENSE
+ See end of file for license information.
+
+ TODO
+ Don't decode all of the image data when only processing a partial tile
+ Don't use full-width decode buffers when only processing a partial tile
+ When processing wide images, break processing into tiles so data fits in L1 cache
+ Installable filters?
+ Resize that respects alpha test coverage
+ (Reference code: FloatImage::alphaTestCoverage and FloatImage::scaleAlphaToCoverage:
+ https://code.google.com/p/nvidia-texture-tools/source/browse/trunk/src/nvimage/FloatImage.cpp )
+*/
+
+#ifndef STBIR_INCLUDE_STB_IMAGE_RESIZE_H
+#define STBIR_INCLUDE_STB_IMAGE_RESIZE_H
+
+#ifdef _MSC_VER
+typedef unsigned char stbir_uint8;
+typedef unsigned short stbir_uint16;
+typedef unsigned int stbir_uint32;
+#else
+#include <stdint.h>
+typedef uint8_t stbir_uint8;
+typedef uint16_t stbir_uint16;
+typedef uint32_t stbir_uint32;
+#endif
+
+#ifdef STB_IMAGE_RESIZE_STATIC
+#define STBIRDEF static
+#else
+#ifdef __cplusplus
+#define STBIRDEF extern "C"
+#else
+#define STBIRDEF extern
+#endif
+#endif
+
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// Easy-to-use API:
+//
+// * "input pixels" points to an array of image data with 'num_channels' channels (e.g. RGB=3, RGBA=4)
+// * input_w is input image width (x-axis), input_h is input image height (y-axis)
+// * stride is the offset between successive rows of image data in memory, in bytes. you can
+// specify 0 to mean packed continuously in memory
+// * alpha channel is treated identically to other channels.
+// * colorspace is linear or sRGB as specified by function name
+// * returned result is 1 for success or 0 in case of an error.
+// #define STBIR_ASSERT() to trigger an assert on parameter validation errors.
+// * Memory required grows approximately linearly with input and output size, but with
+// discontinuities at input_w == output_w and input_h == output_h.
+// * These functions use a "default" resampling filter defined at compile time. To change the filter,
+// you can change the compile-time defaults by #defining STBIR_DEFAULT_FILTER_UPSAMPLE
+// and STBIR_DEFAULT_FILTER_DOWNSAMPLE, or you can use the medium-complexity API.
+
+STBIRDEF int stbir_resize_uint8( const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels);
+
+STBIRDEF int stbir_resize_float( const float *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ float *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels);
+
+
+// The following functions interpret image data as gamma-corrected sRGB.
+// Specify STBIR_ALPHA_CHANNEL_NONE if you have no alpha channel,
+// or otherwise provide the index of the alpha channel. Flags value
+// of 0 will probably do the right thing if you're not sure what
+// the flags mean.
+
+#define STBIR_ALPHA_CHANNEL_NONE -1
+
+// Set this flag if your texture has premultiplied alpha. Otherwise, stbir will
+// use alpha-weighted resampling (effectively premultiplying, resampling,
+// then unpremultiplying).
+#define STBIR_FLAG_ALPHA_PREMULTIPLIED (1 << 0)
+// The specified alpha channel should be handled as gamma-corrected value even
+// when doing sRGB operations.
+#define STBIR_FLAG_ALPHA_USES_COLORSPACE (1 << 1)
+
+STBIRDEF int stbir_resize_uint8_srgb(const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags);
+
+
+typedef enum
+{
+ STBIR_EDGE_CLAMP = 1,
+ STBIR_EDGE_REFLECT = 2,
+ STBIR_EDGE_WRAP = 3,
+ STBIR_EDGE_ZERO = 4,
+} stbir_edge;
+
+// This function adds the ability to specify how requests to sample off the edge of the image are handled.
+STBIRDEF int stbir_resize_uint8_srgb_edgemode(const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode);
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// Medium-complexity API
+//
+// This extends the easy-to-use API as follows:
+//
+// * Alpha-channel can be processed separately
+// * If alpha_channel is not STBIR_ALPHA_CHANNEL_NONE
+// * Alpha channel will not be gamma corrected (unless flags&STBIR_FLAG_GAMMA_CORRECT)
+// * Filters will be weighted by alpha channel (unless flags&STBIR_FLAG_ALPHA_PREMULTIPLIED)
+// * Filter can be selected explicitly
+// * uint16 image type
+// * sRGB colorspace available for all types
+// * context parameter for passing to STBIR_MALLOC
+
+typedef enum
+{
+ STBIR_FILTER_DEFAULT = 0, // use same filter type that easy-to-use API chooses
+ STBIR_FILTER_BOX = 1, // A trapezoid w/1-pixel wide ramps, same result as box for integer scale ratios
+ STBIR_FILTER_TRIANGLE = 2, // On upsampling, produces same results as bilinear texture filtering
+ STBIR_FILTER_CUBICBSPLINE = 3, // The cubic b-spline (aka Mitchell-Netrevalli with B=1,C=0), gaussian-esque
+ STBIR_FILTER_CATMULLROM = 4, // An interpolating cubic spline
+ STBIR_FILTER_MITCHELL = 5, // Mitchell-Netrevalli filter with B=1/3, C=1/3
+} stbir_filter;
+
+typedef enum
+{
+ STBIR_COLORSPACE_LINEAR,
+ STBIR_COLORSPACE_SRGB,
+
+ STBIR_MAX_COLORSPACES,
+} stbir_colorspace;
+
+// The following functions are all identical except for the type of the image data
+
+STBIRDEF int stbir_resize_uint8_generic( const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space,
+ void *alloc_context);
+
+STBIRDEF int stbir_resize_uint16_generic(const stbir_uint16 *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ stbir_uint16 *output_pixels , int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space,
+ void *alloc_context);
+
+STBIRDEF int stbir_resize_float_generic( const float *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ float *output_pixels , int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space,
+ void *alloc_context);
+
+
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// Full-complexity API
+//
+// This extends the medium API as follows:
+//
+// * uint32 image type
+// * not typesafe
+// * separate filter types for each axis
+// * separate edge modes for each axis
+// * can specify scale explicitly for subpixel correctness
+// * can specify image source tile using texture coordinates
+
+typedef enum
+{
+ STBIR_TYPE_UINT8 ,
+ STBIR_TYPE_UINT16,
+ STBIR_TYPE_UINT32,
+ STBIR_TYPE_FLOAT ,
+
+ STBIR_MAX_TYPES
+} stbir_datatype;
+
+STBIRDEF int stbir_resize( const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ stbir_datatype datatype,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical,
+ stbir_filter filter_horizontal, stbir_filter filter_vertical,
+ stbir_colorspace space, void *alloc_context);
+
+STBIRDEF int stbir_resize_subpixel(const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ stbir_datatype datatype,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical,
+ stbir_filter filter_horizontal, stbir_filter filter_vertical,
+ stbir_colorspace space, void *alloc_context,
+ float x_scale, float y_scale,
+ float x_offset, float y_offset);
+
+STBIRDEF int stbir_resize_region( const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ stbir_datatype datatype,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical,
+ stbir_filter filter_horizontal, stbir_filter filter_vertical,
+ stbir_colorspace space, void *alloc_context,
+ float s0, float t0, float s1, float t1);
+// (s0, t0) & (s1, t1) are the top-left and bottom right corner (uv addressing style: [0, 1]x[0, 1]) of a region of the input image to use.
+
+//
+//
+//// end header file /////////////////////////////////////////////////////
+#endif // STBIR_INCLUDE_STB_IMAGE_RESIZE_H
+
+
+
+
+
+#ifdef STB_IMAGE_RESIZE_IMPLEMENTATION
+
+#ifndef STBIR_ASSERT
+#include <assert.h>
+#define STBIR_ASSERT(x) assert(x)
+#endif
+
+// For memset
+#include <string.h>
+
+#include <math.h>
+
+#ifndef STBIR_MALLOC
+#include <stdlib.h>
+#define STBIR_MALLOC(size,c) malloc(size)
+#define STBIR_FREE(ptr,c) free(ptr)
+#endif
+
+#ifndef _MSC_VER
+#ifdef __cplusplus
+#define stbir__inline inline
+#else
+#define stbir__inline
+#endif
+#else
+#define stbir__inline __forceinline
+#endif
+
+
+// should produce compiler error if size is wrong
+typedef unsigned char stbir__validate_uint32[sizeof(stbir_uint32) == 4 ? 1 : -1];
+
+#ifdef _MSC_VER
+#define STBIR__NOTUSED(v) (void)(v)
+#else
+#define STBIR__NOTUSED(v) (void)sizeof(v)
+#endif
+
+#define STBIR__ARRAY_SIZE(a) (sizeof((a))/sizeof((a)[0]))
+
+#ifndef STBIR_DEFAULT_FILTER_UPSAMPLE
+#define STBIR_DEFAULT_FILTER_UPSAMPLE STBIR_FILTER_CATMULLROM
+#endif
+
+#ifndef STBIR_DEFAULT_FILTER_DOWNSAMPLE
+#define STBIR_DEFAULT_FILTER_DOWNSAMPLE STBIR_FILTER_MITCHELL
+#endif
+
+#ifndef STBIR_PROGRESS_REPORT
+#define STBIR_PROGRESS_REPORT(float_0_to_1)
+#endif
+
+#ifndef STBIR_MAX_CHANNELS
+#define STBIR_MAX_CHANNELS 64
+#endif
+
+#if STBIR_MAX_CHANNELS > 65536
+#error "Too many channels; STBIR_MAX_CHANNELS must be no more than 65536."
+// because we store the indices in 16-bit variables
+#endif
+
+// This value is added to alpha just before premultiplication to avoid
+// zeroing out color values. It is equivalent to 2^-80. If you don't want
+// that behavior (it may interfere if you have floating point images with
+// very small alpha values) then you can define STBIR_NO_ALPHA_EPSILON to
+// disable it.
+#ifndef STBIR_ALPHA_EPSILON
+#define STBIR_ALPHA_EPSILON ((float)1 / (1 << 20) / (1 << 20) / (1 << 20) / (1 << 20))
+#endif
+
+
+
+#ifdef _MSC_VER
+#define STBIR__UNUSED_PARAM(v) (void)(v)
+#else
+#define STBIR__UNUSED_PARAM(v) (void)sizeof(v)
+#endif
+
+// must match stbir_datatype
+static unsigned char stbir__type_size[] = {
+ 1, // STBIR_TYPE_UINT8
+ 2, // STBIR_TYPE_UINT16
+ 4, // STBIR_TYPE_UINT32
+ 4, // STBIR_TYPE_FLOAT
+};
+
+// Kernel function centered at 0
+typedef float (stbir__kernel_fn)(float x, float scale);
+typedef float (stbir__support_fn)(float scale);
+
+typedef struct
+{
+ stbir__kernel_fn* kernel;
+ stbir__support_fn* support;
+} stbir__filter_info;
+
+// When upsampling, the contributors are which source pixels contribute.
+// When downsampling, the contributors are which destination pixels are contributed to.
+typedef struct
+{
+ int n0; // First contributing pixel
+ int n1; // Last contributing pixel
+} stbir__contributors;
+
+typedef struct
+{
+ const void* input_data;
+ int input_w;
+ int input_h;
+ int input_stride_bytes;
+
+ void* output_data;
+ int output_w;
+ int output_h;
+ int output_stride_bytes;
+
+ float s0, t0, s1, t1;
+
+ float horizontal_shift; // Units: output pixels
+ float vertical_shift; // Units: output pixels
+ float horizontal_scale;
+ float vertical_scale;
+
+ int channels;
+ int alpha_channel;
+ stbir_uint32 flags;
+ stbir_datatype type;
+ stbir_filter horizontal_filter;
+ stbir_filter vertical_filter;
+ stbir_edge edge_horizontal;
+ stbir_edge edge_vertical;
+ stbir_colorspace colorspace;
+
+ stbir__contributors* horizontal_contributors;
+ float* horizontal_coefficients;
+
+ stbir__contributors* vertical_contributors;
+ float* vertical_coefficients;
+
+ int decode_buffer_pixels;
+ float* decode_buffer;
+
+ float* horizontal_buffer;
+
+ // cache these because ceil/floor are inexplicably showing up in profile
+ int horizontal_coefficient_width;
+ int vertical_coefficient_width;
+ int horizontal_filter_pixel_width;
+ int vertical_filter_pixel_width;
+ int horizontal_filter_pixel_margin;
+ int vertical_filter_pixel_margin;
+ int horizontal_num_contributors;
+ int vertical_num_contributors;
+
+ int ring_buffer_length_bytes; // The length of an individual entry in the ring buffer. The total number of ring buffers is stbir__get_filter_pixel_width(filter)
+ int ring_buffer_num_entries; // Total number of entries in the ring buffer.
+ int ring_buffer_first_scanline;
+ int ring_buffer_last_scanline;
+ int ring_buffer_begin_index; // first_scanline is at this index in the ring buffer
+ float* ring_buffer;
+
+ float* encode_buffer; // A temporary buffer to store floats so we don't lose precision while we do multiply-adds.
+
+ int horizontal_contributors_size;
+ int horizontal_coefficients_size;
+ int vertical_contributors_size;
+ int vertical_coefficients_size;
+ int decode_buffer_size;
+ int horizontal_buffer_size;
+ int ring_buffer_size;
+ int encode_buffer_size;
+} stbir__info;
+
+
+static const float stbir__max_uint8_as_float = 255.0f;
+static const float stbir__max_uint16_as_float = 65535.0f;
+static const double stbir__max_uint32_as_float = 4294967295.0;
+
+
+static stbir__inline int stbir__min(int a, int b)
+{
+ return a < b ? a : b;
+}
+
+static stbir__inline float stbir__saturate(float x)
+{
+ if (x < 0)
+ return 0;
+
+ if (x > 1)
+ return 1;
+
+ return x;
+}
+
+#ifdef STBIR_SATURATE_INT
+static stbir__inline stbir_uint8 stbir__saturate8(int x)
+{
+ if ((unsigned int) x <= 255)
+ return x;
+
+ if (x < 0)
+ return 0;
+
+ return 255;
+}
+
+static stbir__inline stbir_uint16 stbir__saturate16(int x)
+{
+ if ((unsigned int) x <= 65535)
+ return x;
+
+ if (x < 0)
+ return 0;
+
+ return 65535;
+}
+#endif
+
+static float stbir__srgb_uchar_to_linear_float[256] = {
+ 0.000000f, 0.000304f, 0.000607f, 0.000911f, 0.001214f, 0.001518f, 0.001821f, 0.002125f, 0.002428f, 0.002732f, 0.003035f,
+ 0.003347f, 0.003677f, 0.004025f, 0.004391f, 0.004777f, 0.005182f, 0.005605f, 0.006049f, 0.006512f, 0.006995f, 0.007499f,
+ 0.008023f, 0.008568f, 0.009134f, 0.009721f, 0.010330f, 0.010960f, 0.011612f, 0.012286f, 0.012983f, 0.013702f, 0.014444f,
+ 0.015209f, 0.015996f, 0.016807f, 0.017642f, 0.018500f, 0.019382f, 0.020289f, 0.021219f, 0.022174f, 0.023153f, 0.024158f,
+ 0.025187f, 0.026241f, 0.027321f, 0.028426f, 0.029557f, 0.030713f, 0.031896f, 0.033105f, 0.034340f, 0.035601f, 0.036889f,
+ 0.038204f, 0.039546f, 0.040915f, 0.042311f, 0.043735f, 0.045186f, 0.046665f, 0.048172f, 0.049707f, 0.051269f, 0.052861f,
+ 0.054480f, 0.056128f, 0.057805f, 0.059511f, 0.061246f, 0.063010f, 0.064803f, 0.066626f, 0.068478f, 0.070360f, 0.072272f,
+ 0.074214f, 0.076185f, 0.078187f, 0.080220f, 0.082283f, 0.084376f, 0.086500f, 0.088656f, 0.090842f, 0.093059f, 0.095307f,
+ 0.097587f, 0.099899f, 0.102242f, 0.104616f, 0.107023f, 0.109462f, 0.111932f, 0.114435f, 0.116971f, 0.119538f, 0.122139f,
+ 0.124772f, 0.127438f, 0.130136f, 0.132868f, 0.135633f, 0.138432f, 0.141263f, 0.144128f, 0.147027f, 0.149960f, 0.152926f,
+ 0.155926f, 0.158961f, 0.162029f, 0.165132f, 0.168269f, 0.171441f, 0.174647f, 0.177888f, 0.181164f, 0.184475f, 0.187821f,
+ 0.191202f, 0.194618f, 0.198069f, 0.201556f, 0.205079f, 0.208637f, 0.212231f, 0.215861f, 0.219526f, 0.223228f, 0.226966f,
+ 0.230740f, 0.234551f, 0.238398f, 0.242281f, 0.246201f, 0.250158f, 0.254152f, 0.258183f, 0.262251f, 0.266356f, 0.270498f,
+ 0.274677f, 0.278894f, 0.283149f, 0.287441f, 0.291771f, 0.296138f, 0.300544f, 0.304987f, 0.309469f, 0.313989f, 0.318547f,
+ 0.323143f, 0.327778f, 0.332452f, 0.337164f, 0.341914f, 0.346704f, 0.351533f, 0.356400f, 0.361307f, 0.366253f, 0.371238f,
+ 0.376262f, 0.381326f, 0.386430f, 0.391573f, 0.396755f, 0.401978f, 0.407240f, 0.412543f, 0.417885f, 0.423268f, 0.428691f,
+ 0.434154f, 0.439657f, 0.445201f, 0.450786f, 0.456411f, 0.462077f, 0.467784f, 0.473532f, 0.479320f, 0.485150f, 0.491021f,
+ 0.496933f, 0.502887f, 0.508881f, 0.514918f, 0.520996f, 0.527115f, 0.533276f, 0.539480f, 0.545725f, 0.552011f, 0.558340f,
+ 0.564712f, 0.571125f, 0.577581f, 0.584078f, 0.590619f, 0.597202f, 0.603827f, 0.610496f, 0.617207f, 0.623960f, 0.630757f,
+ 0.637597f, 0.644480f, 0.651406f, 0.658375f, 0.665387f, 0.672443f, 0.679543f, 0.686685f, 0.693872f, 0.701102f, 0.708376f,
+ 0.715694f, 0.723055f, 0.730461f, 0.737911f, 0.745404f, 0.752942f, 0.760525f, 0.768151f, 0.775822f, 0.783538f, 0.791298f,
+ 0.799103f, 0.806952f, 0.814847f, 0.822786f, 0.830770f, 0.838799f, 0.846873f, 0.854993f, 0.863157f, 0.871367f, 0.879622f,
+ 0.887923f, 0.896269f, 0.904661f, 0.913099f, 0.921582f, 0.930111f, 0.938686f, 0.947307f, 0.955974f, 0.964686f, 0.973445f,
+ 0.982251f, 0.991102f, 1.0f
+};
+
+static float stbir__srgb_to_linear(float f)
+{
+ if (f <= 0.04045f)
+ return f / 12.92f;
+ else
+ return (float)pow((f + 0.055f) / 1.055f, 2.4f);
+}
+
+static float stbir__linear_to_srgb(float f)
+{
+ if (f <= 0.0031308f)
+ return f * 12.92f;
+ else
+ return 1.055f * (float)pow(f, 1 / 2.4f) - 0.055f;
+}
+
+#ifndef STBIR_NON_IEEE_FLOAT
+// From https://gist.github.com/rygorous/2203834
+
+typedef union
+{
+ stbir_uint32 u;
+ float f;
+} stbir__FP32;
+
+static const stbir_uint32 fp32_to_srgb8_tab4[104] = {
+ 0x0073000d, 0x007a000d, 0x0080000d, 0x0087000d, 0x008d000d, 0x0094000d, 0x009a000d, 0x00a1000d,
+ 0x00a7001a, 0x00b4001a, 0x00c1001a, 0x00ce001a, 0x00da001a, 0x00e7001a, 0x00f4001a, 0x0101001a,
+ 0x010e0033, 0x01280033, 0x01410033, 0x015b0033, 0x01750033, 0x018f0033, 0x01a80033, 0x01c20033,
+ 0x01dc0067, 0x020f0067, 0x02430067, 0x02760067, 0x02aa0067, 0x02dd0067, 0x03110067, 0x03440067,
+ 0x037800ce, 0x03df00ce, 0x044600ce, 0x04ad00ce, 0x051400ce, 0x057b00c5, 0x05dd00bc, 0x063b00b5,
+ 0x06970158, 0x07420142, 0x07e30130, 0x087b0120, 0x090b0112, 0x09940106, 0x0a1700fc, 0x0a9500f2,
+ 0x0b0f01cb, 0x0bf401ae, 0x0ccb0195, 0x0d950180, 0x0e56016e, 0x0f0d015e, 0x0fbc0150, 0x10630143,
+ 0x11070264, 0x1238023e, 0x1357021d, 0x14660201, 0x156601e9, 0x165a01d3, 0x174401c0, 0x182401af,
+ 0x18fe0331, 0x1a9602fe, 0x1c1502d2, 0x1d7e02ad, 0x1ed4028d, 0x201a0270, 0x21520256, 0x227d0240,
+ 0x239f0443, 0x25c003fe, 0x27bf03c4, 0x29a10392, 0x2b6a0367, 0x2d1d0341, 0x2ebe031f, 0x304d0300,
+ 0x31d105b0, 0x34a80555, 0x37520507, 0x39d504c5, 0x3c37048b, 0x3e7c0458, 0x40a8042a, 0x42bd0401,
+ 0x44c20798, 0x488e071e, 0x4c1c06b6, 0x4f76065d, 0x52a50610, 0x55ac05cc, 0x5892058f, 0x5b590559,
+ 0x5e0c0a23, 0x631c0980, 0x67db08f6, 0x6c55087f, 0x70940818, 0x74a007bd, 0x787d076c, 0x7c330723,
+};
+
+static stbir_uint8 stbir__linear_to_srgb_uchar(float in)
+{
+ static const stbir__FP32 almostone = { 0x3f7fffff }; // 1-eps
+ static const stbir__FP32 minval = { (127-13) << 23 };
+ stbir_uint32 tab,bias,scale,t;
+ stbir__FP32 f;
+
+ // Clamp to [2^(-13), 1-eps]; these two values map to 0 and 1, respectively.
+ // The tests are carefully written so that NaNs map to 0, same as in the reference
+ // implementation.
+ if (!(in > minval.f)) // written this way to catch NaNs
+ in = minval.f;
+ if (in > almostone.f)
+ in = almostone.f;
+
+ // Do the table lookup and unpack bias, scale
+ f.f = in;
+ tab = fp32_to_srgb8_tab4[(f.u - minval.u) >> 20];
+ bias = (tab >> 16) << 9;
+ scale = tab & 0xffff;
+
+ // Grab next-highest mantissa bits and perform linear interpolation
+ t = (f.u >> 12) & 0xff;
+ return (unsigned char) ((bias + scale*t) >> 16);
+}
+
+#else
+// sRGB transition values, scaled by 1<<28
+static int stbir__srgb_offset_to_linear_scaled[256] =
+{
+ 0, 40738, 122216, 203693, 285170, 366648, 448125, 529603,
+ 611080, 692557, 774035, 855852, 942009, 1033024, 1128971, 1229926,
+ 1335959, 1447142, 1563542, 1685229, 1812268, 1944725, 2082664, 2226148,
+ 2375238, 2529996, 2690481, 2856753, 3028870, 3206888, 3390865, 3580856,
+ 3776916, 3979100, 4187460, 4402049, 4622919, 4850123, 5083710, 5323731,
+ 5570236, 5823273, 6082892, 6349140, 6622065, 6901714, 7188133, 7481369,
+ 7781466, 8088471, 8402427, 8723380, 9051372, 9386448, 9728650, 10078021,
+ 10434603, 10798439, 11169569, 11548036, 11933879, 12327139, 12727857, 13136073,
+ 13551826, 13975156, 14406100, 14844697, 15290987, 15745007, 16206795, 16676389,
+ 17153826, 17639142, 18132374, 18633560, 19142734, 19659934, 20185196, 20718552,
+ 21260042, 21809696, 22367554, 22933648, 23508010, 24090680, 24681686, 25281066,
+ 25888850, 26505076, 27129772, 27762974, 28404716, 29055026, 29713942, 30381490,
+ 31057708, 31742624, 32436272, 33138682, 33849884, 34569912, 35298800, 36036568,
+ 36783260, 37538896, 38303512, 39077136, 39859796, 40651528, 41452360, 42262316,
+ 43081432, 43909732, 44747252, 45594016, 46450052, 47315392, 48190064, 49074096,
+ 49967516, 50870356, 51782636, 52704392, 53635648, 54576432, 55526772, 56486700,
+ 57456236, 58435408, 59424248, 60422780, 61431036, 62449032, 63476804, 64514376,
+ 65561776, 66619028, 67686160, 68763192, 69850160, 70947088, 72053992, 73170912,
+ 74297864, 75434880, 76581976, 77739184, 78906536, 80084040, 81271736, 82469648,
+ 83677792, 84896192, 86124888, 87363888, 88613232, 89872928, 91143016, 92423512,
+ 93714432, 95015816, 96327688, 97650056, 98982952, 100326408, 101680440, 103045072,
+ 104420320, 105806224, 107202800, 108610064, 110028048, 111456776, 112896264, 114346544,
+ 115807632, 117279552, 118762328, 120255976, 121760536, 123276016, 124802440, 126339832,
+ 127888216, 129447616, 131018048, 132599544, 134192112, 135795792, 137410592, 139036528,
+ 140673648, 142321952, 143981456, 145652208, 147334208, 149027488, 150732064, 152447968,
+ 154175200, 155913792, 157663776, 159425168, 161197984, 162982240, 164777968, 166585184,
+ 168403904, 170234160, 172075968, 173929344, 175794320, 177670896, 179559120, 181458992,
+ 183370528, 185293776, 187228736, 189175424, 191133888, 193104112, 195086128, 197079968,
+ 199085648, 201103184, 203132592, 205173888, 207227120, 209292272, 211369392, 213458480,
+ 215559568, 217672656, 219797792, 221934976, 224084240, 226245600, 228419056, 230604656,
+ 232802400, 235012320, 237234432, 239468736, 241715280, 243974080, 246245120, 248528464,
+ 250824112, 253132064, 255452368, 257785040, 260130080, 262487520, 264857376, 267239664,
+};
+
+static stbir_uint8 stbir__linear_to_srgb_uchar(float f)
+{
+ int x = (int) (f * (1 << 28)); // has headroom so you don't need to clamp
+ int v = 0;
+ int i;
+
+ // Refine the guess with a short binary search.
+ i = v + 128; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+ i = v + 64; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+ i = v + 32; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+ i = v + 16; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+ i = v + 8; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+ i = v + 4; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+ i = v + 2; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+ i = v + 1; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+
+ return (stbir_uint8) v;
+}
+#endif
+
+static float stbir__filter_trapezoid(float x, float scale)
+{
+ float halfscale = scale / 2;
+ float t = 0.5f + halfscale;
+ STBIR_ASSERT(scale <= 1);
+
+ x = (float)fabs(x);
+
+ if (x >= t)
+ return 0;
+ else
+ {
+ float r = 0.5f - halfscale;
+ if (x <= r)
+ return 1;
+ else
+ return (t - x) / scale;
+ }
+}
+
+static float stbir__support_trapezoid(float scale)
+{
+ STBIR_ASSERT(scale <= 1);
+ return 0.5f + scale / 2;
+}
+
+static float stbir__filter_triangle(float x, float s)
+{
+ STBIR__UNUSED_PARAM(s);
+
+ x = (float)fabs(x);
+
+ if (x <= 1.0f)
+ return 1 - x;
+ else
+ return 0;
+}
+
+static float stbir__filter_cubic(float x, float s)
+{
+ STBIR__UNUSED_PARAM(s);
+
+ x = (float)fabs(x);
+
+ if (x < 1.0f)
+ return (4 + x*x*(3*x - 6))/6;
+ else if (x < 2.0f)
+ return (8 + x*(-12 + x*(6 - x)))/6;
+
+ return (0.0f);
+}
+
+static float stbir__filter_catmullrom(float x, float s)
+{
+ STBIR__UNUSED_PARAM(s);
+
+ x = (float)fabs(x);
+
+ if (x < 1.0f)
+ return 1 - x*x*(2.5f - 1.5f*x);
+ else if (x < 2.0f)
+ return 2 - x*(4 + x*(0.5f*x - 2.5f));
+
+ return (0.0f);
+}
+
+static float stbir__filter_mitchell(float x, float s)
+{
+ STBIR__UNUSED_PARAM(s);
+
+ x = (float)fabs(x);
+
+ if (x < 1.0f)
+ return (16 + x*x*(21 * x - 36))/18;
+ else if (x < 2.0f)
+ return (32 + x*(-60 + x*(36 - 7*x)))/18;
+
+ return (0.0f);
+}
+
+static float stbir__support_zero(float s)
+{
+ STBIR__UNUSED_PARAM(s);
+ return 0;
+}
+
+static float stbir__support_one(float s)
+{
+ STBIR__UNUSED_PARAM(s);
+ return 1;
+}
+
+static float stbir__support_two(float s)
+{
+ STBIR__UNUSED_PARAM(s);
+ return 2;
+}
+
+static stbir__filter_info stbir__filter_info_table[] = {
+ { NULL, stbir__support_zero },
+ { stbir__filter_trapezoid, stbir__support_trapezoid },
+ { stbir__filter_triangle, stbir__support_one },
+ { stbir__filter_cubic, stbir__support_two },
+ { stbir__filter_catmullrom, stbir__support_two },
+ { stbir__filter_mitchell, stbir__support_two },
+};
+
+stbir__inline static int stbir__use_upsampling(float ratio)
+{
+ return ratio > 1;
+}
+
+stbir__inline static int stbir__use_width_upsampling(stbir__info* stbir_info)
+{
+ return stbir__use_upsampling(stbir_info->horizontal_scale);
+}
+
+stbir__inline static int stbir__use_height_upsampling(stbir__info* stbir_info)
+{
+ return stbir__use_upsampling(stbir_info->vertical_scale);
+}
+
+// This is the maximum number of input samples that can affect an output sample
+// with the given filter
+static int stbir__get_filter_pixel_width(stbir_filter filter, float scale)
+{
+ STBIR_ASSERT(filter != 0);
+ STBIR_ASSERT(filter < STBIR__ARRAY_SIZE(stbir__filter_info_table));
+
+ if (stbir__use_upsampling(scale))
+ return (int)ceil(stbir__filter_info_table[filter].support(1/scale) * 2);
+ else
+ return (int)ceil(stbir__filter_info_table[filter].support(scale) * 2 / scale);
+}
+
+// This is how much to expand buffers to account for filters seeking outside
+// the image boundaries.
+static int stbir__get_filter_pixel_margin(stbir_filter filter, float scale)
+{
+ return stbir__get_filter_pixel_width(filter, scale) / 2;
+}
+
+static int stbir__get_coefficient_width(stbir_filter filter, float scale)
+{
+ if (stbir__use_upsampling(scale))
+ return (int)ceil(stbir__filter_info_table[filter].support(1 / scale) * 2);
+ else
+ return (int)ceil(stbir__filter_info_table[filter].support(scale) * 2);
+}
+
+static int stbir__get_contributors(float scale, stbir_filter filter, int input_size, int output_size)
+{
+ if (stbir__use_upsampling(scale))
+ return output_size;
+ else
+ return (input_size + stbir__get_filter_pixel_margin(filter, scale) * 2);
+}
+
+static int stbir__get_total_horizontal_coefficients(stbir__info* info)
+{
+ return info->horizontal_num_contributors
+ * stbir__get_coefficient_width (info->horizontal_filter, info->horizontal_scale);
+}
+
+static int stbir__get_total_vertical_coefficients(stbir__info* info)
+{
+ return info->vertical_num_contributors
+ * stbir__get_coefficient_width (info->vertical_filter, info->vertical_scale);
+}
+
+static stbir__contributors* stbir__get_contributor(stbir__contributors* contributors, int n)
+{
+ return &contributors[n];
+}
+
+// For perf reasons this code is duplicated in stbir__resample_horizontal_upsample/downsample,
+// if you change it here change it there too.
+static float* stbir__get_coefficient(float* coefficients, stbir_filter filter, float scale, int n, int c)
+{
+ int width = stbir__get_coefficient_width(filter, scale);
+ return &coefficients[width*n + c];
+}
+
+static int stbir__edge_wrap_slow(stbir_edge edge, int n, int max)
+{
+ switch (edge)
+ {
+ case STBIR_EDGE_ZERO:
+ return 0; // we'll decode the wrong pixel here, and then overwrite with 0s later
+
+ case STBIR_EDGE_CLAMP:
+ if (n < 0)
+ return 0;
+
+ if (n >= max)
+ return max - 1;
+
+ return n; // NOTREACHED
+
+ case STBIR_EDGE_REFLECT:
+ {
+ if (n < 0)
+ {
+ if (n < max)
+ return -n;
+ else
+ return max - 1;
+ }
+
+ if (n >= max)
+ {
+ int max2 = max * 2;
+ if (n >= max2)
+ return 0;
+ else
+ return max2 - n - 1;
+ }
+
+ return n; // NOTREACHED
+ }
+
+ case STBIR_EDGE_WRAP:
+ if (n >= 0)
+ return (n % max);
+ else
+ {
+ int m = (-n) % max;
+
+ if (m != 0)
+ m = max - m;
+
+ return (m);
+ }
+ return n; // NOTREACHED
+
+ default:
+ STBIR_ASSERT(!"Unimplemented edge type");
+ return 0;
+ }
+}
+
+stbir__inline static int stbir__edge_wrap(stbir_edge edge, int n, int max)
+{
+ // avoid per-pixel switch
+ if (n >= 0 && n < max)
+ return n;
+ return stbir__edge_wrap_slow(edge, n, max);
+}
+
+// What input pixels contribute to this output pixel?
+static void stbir__calculate_sample_range_upsample(int n, float out_filter_radius, float scale_ratio, float out_shift, int* in_first_pixel, int* in_last_pixel, float* in_center_of_out)
+{
+ float out_pixel_center = (float)n + 0.5f;
+ float out_pixel_influence_lowerbound = out_pixel_center - out_filter_radius;
+ float out_pixel_influence_upperbound = out_pixel_center + out_filter_radius;
+
+ float in_pixel_influence_lowerbound = (out_pixel_influence_lowerbound + out_shift) / scale_ratio;
+ float in_pixel_influence_upperbound = (out_pixel_influence_upperbound + out_shift) / scale_ratio;
+
+ *in_center_of_out = (out_pixel_center + out_shift) / scale_ratio;
+ *in_first_pixel = (int)(floor(in_pixel_influence_lowerbound + 0.5));
+ *in_last_pixel = (int)(floor(in_pixel_influence_upperbound - 0.5));
+}
+
+// What output pixels does this input pixel contribute to?
+static void stbir__calculate_sample_range_downsample(int n, float in_pixels_radius, float scale_ratio, float out_shift, int* out_first_pixel, int* out_last_pixel, float* out_center_of_in)
+{
+ float in_pixel_center = (float)n + 0.5f;
+ float in_pixel_influence_lowerbound = in_pixel_center - in_pixels_radius;
+ float in_pixel_influence_upperbound = in_pixel_center + in_pixels_radius;
+
+ float out_pixel_influence_lowerbound = in_pixel_influence_lowerbound * scale_ratio - out_shift;
+ float out_pixel_influence_upperbound = in_pixel_influence_upperbound * scale_ratio - out_shift;
+
+ *out_center_of_in = in_pixel_center * scale_ratio - out_shift;
+ *out_first_pixel = (int)(floor(out_pixel_influence_lowerbound + 0.5));
+ *out_last_pixel = (int)(floor(out_pixel_influence_upperbound - 0.5));
+}
+
+static void stbir__calculate_coefficients_upsample(stbir_filter filter, float scale, int in_first_pixel, int in_last_pixel, float in_center_of_out, stbir__contributors* contributor, float* coefficient_group)
+{
+ int i;
+ float total_filter = 0;
+ float filter_scale;
+
+ STBIR_ASSERT(in_last_pixel - in_first_pixel <= (int)ceil(stbir__filter_info_table[filter].support(1/scale) * 2)); // Taken directly from stbir__get_coefficient_width() which we can't call because we don't know if we're horizontal or vertical.
+
+ contributor->n0 = in_first_pixel;
+ contributor->n1 = in_last_pixel;
+
+ STBIR_ASSERT(contributor->n1 >= contributor->n0);
+
+ for (i = 0; i <= in_last_pixel - in_first_pixel; i++)
+ {
+ float in_pixel_center = (float)(i + in_first_pixel) + 0.5f;
+ coefficient_group[i] = stbir__filter_info_table[filter].kernel(in_center_of_out - in_pixel_center, 1 / scale);
+
+ // If the coefficient is zero, skip it. (Don't do the <0 check here, we want the influence of those outside pixels.)
+ if (i == 0 && !coefficient_group[i])
+ {
+ contributor->n0 = ++in_first_pixel;
+ i--;
+ continue;
+ }
+
+ total_filter += coefficient_group[i];
+ }
+
+ STBIR_ASSERT(stbir__filter_info_table[filter].kernel((float)(in_last_pixel + 1) + 0.5f - in_center_of_out, 1/scale) == 0);
+
+ STBIR_ASSERT(total_filter > 0.9);
+ STBIR_ASSERT(total_filter < 1.1f); // Make sure it's not way off.
+
+ // Make sure the sum of all coefficients is 1.
+ filter_scale = 1 / total_filter;
+
+ for (i = 0; i <= in_last_pixel - in_first_pixel; i++)
+ coefficient_group[i] *= filter_scale;
+
+ for (i = in_last_pixel - in_first_pixel; i >= 0; i--)
+ {
+ if (coefficient_group[i])
+ break;
+
+ // This line has no weight. We can skip it.
+ contributor->n1 = contributor->n0 + i - 1;
+ }
+}
+
+static void stbir__calculate_coefficients_downsample(stbir_filter filter, float scale_ratio, int out_first_pixel, int out_last_pixel, float out_center_of_in, stbir__contributors* contributor, float* coefficient_group)
+{
+ int i;
+
+ STBIR_ASSERT(out_last_pixel - out_first_pixel <= (int)ceil(stbir__filter_info_table[filter].support(scale_ratio) * 2)); // Taken directly from stbir__get_coefficient_width() which we can't call because we don't know if we're horizontal or vertical.
+
+ contributor->n0 = out_first_pixel;
+ contributor->n1 = out_last_pixel;
+
+ STBIR_ASSERT(contributor->n1 >= contributor->n0);
+
+ for (i = 0; i <= out_last_pixel - out_first_pixel; i++)
+ {
+ float out_pixel_center = (float)(i + out_first_pixel) + 0.5f;
+ float x = out_pixel_center - out_center_of_in;
+ coefficient_group[i] = stbir__filter_info_table[filter].kernel(x, scale_ratio) * scale_ratio;
+ }
+
+ STBIR_ASSERT(stbir__filter_info_table[filter].kernel((float)(out_last_pixel + 1) + 0.5f - out_center_of_in, scale_ratio) == 0);
+
+ for (i = out_last_pixel - out_first_pixel; i >= 0; i--)
+ {
+ if (coefficient_group[i])
+ break;
+
+ // This line has no weight. We can skip it.
+ contributor->n1 = contributor->n0 + i - 1;
+ }
+}
+
+static void stbir__normalize_downsample_coefficients(stbir__contributors* contributors, float* coefficients, stbir_filter filter, float scale_ratio, int input_size, int output_size)
+{
+ int num_contributors = stbir__get_contributors(scale_ratio, filter, input_size, output_size);
+ int num_coefficients = stbir__get_coefficient_width(filter, scale_ratio);
+ int i, j;
+ int skip;
+
+ for (i = 0; i < output_size; i++)
+ {
+ float scale;
+ float total = 0;
+
+ for (j = 0; j < num_contributors; j++)
+ {
+ if (i >= contributors[j].n0 && i <= contributors[j].n1)
+ {
+ float coefficient = *stbir__get_coefficient(coefficients, filter, scale_ratio, j, i - contributors[j].n0);
+ total += coefficient;
+ }
+ else if (i < contributors[j].n0)
+ break;
+ }
+
+ STBIR_ASSERT(total > 0.9f);
+ STBIR_ASSERT(total < 1.1f);
+
+ scale = 1 / total;
+
+ for (j = 0; j < num_contributors; j++)
+ {
+ if (i >= contributors[j].n0 && i <= contributors[j].n1)
+ *stbir__get_coefficient(coefficients, filter, scale_ratio, j, i - contributors[j].n0) *= scale;
+ else if (i < contributors[j].n0)
+ break;
+ }
+ }
+
+ // Optimize: Skip zero coefficients and contributions outside of image bounds.
+ // Do this after normalizing because normalization depends on the n0/n1 values.
+ for (j = 0; j < num_contributors; j++)
+ {
+ int range, max, width;
+
+ skip = 0;
+ while (*stbir__get_coefficient(coefficients, filter, scale_ratio, j, skip) == 0)
+ skip++;
+
+ contributors[j].n0 += skip;
+
+ while (contributors[j].n0 < 0)
+ {
+ contributors[j].n0++;
+ skip++;
+ }
+
+ range = contributors[j].n1 - contributors[j].n0 + 1;
+ max = stbir__min(num_coefficients, range);
+
+ width = stbir__get_coefficient_width(filter, scale_ratio);
+ for (i = 0; i < max; i++)
+ {
+ if (i + skip >= width)
+ break;
+
+ *stbir__get_coefficient(coefficients, filter, scale_ratio, j, i) = *stbir__get_coefficient(coefficients, filter, scale_ratio, j, i + skip);
+ }
+
+ continue;
+ }
+
+ // Using min to avoid writing into invalid pixels.
+ for (i = 0; i < num_contributors; i++)
+ contributors[i].n1 = stbir__min(contributors[i].n1, output_size - 1);
+}
+
+// Each scan line uses the same kernel values so we should calculate the kernel
+// values once and then we can use them for every scan line.
+static void stbir__calculate_filters(stbir__contributors* contributors, float* coefficients, stbir_filter filter, float scale_ratio, float shift, int input_size, int output_size)
+{
+ int n;
+ int total_contributors = stbir__get_contributors(scale_ratio, filter, input_size, output_size);
+
+ if (stbir__use_upsampling(scale_ratio))
+ {
+ float out_pixels_radius = stbir__filter_info_table[filter].support(1 / scale_ratio) * scale_ratio;
+
+ // Looping through out pixels
+ for (n = 0; n < total_contributors; n++)
+ {
+ float in_center_of_out; // Center of the current out pixel in the in pixel space
+ int in_first_pixel, in_last_pixel;
+
+ stbir__calculate_sample_range_upsample(n, out_pixels_radius, scale_ratio, shift, &in_first_pixel, &in_last_pixel, &in_center_of_out);
+
+ stbir__calculate_coefficients_upsample(filter, scale_ratio, in_first_pixel, in_last_pixel, in_center_of_out, stbir__get_contributor(contributors, n), stbir__get_coefficient(coefficients, filter, scale_ratio, n, 0));
+ }
+ }
+ else
+ {
+ float in_pixels_radius = stbir__filter_info_table[filter].support(scale_ratio) / scale_ratio;
+
+ // Looping through in pixels
+ for (n = 0; n < total_contributors; n++)
+ {
+ float out_center_of_in; // Center of the current out pixel in the in pixel space
+ int out_first_pixel, out_last_pixel;
+ int n_adjusted = n - stbir__get_filter_pixel_margin(filter, scale_ratio);
+
+ stbir__calculate_sample_range_downsample(n_adjusted, in_pixels_radius, scale_ratio, shift, &out_first_pixel, &out_last_pixel, &out_center_of_in);
+
+ stbir__calculate_coefficients_downsample(filter, scale_ratio, out_first_pixel, out_last_pixel, out_center_of_in, stbir__get_contributor(contributors, n), stbir__get_coefficient(coefficients, filter, scale_ratio, n, 0));
+ }
+
+ stbir__normalize_downsample_coefficients(contributors, coefficients, filter, scale_ratio, input_size, output_size);
+ }
+}
+
+static float* stbir__get_decode_buffer(stbir__info* stbir_info)
+{
+ // The 0 index of the decode buffer starts after the margin. This makes
+ // it okay to use negative indexes on the decode buffer.
+ return &stbir_info->decode_buffer[stbir_info->horizontal_filter_pixel_margin * stbir_info->channels];
+}
+
+#define STBIR__DECODE(type, colorspace) ((type) * (STBIR_MAX_COLORSPACES) + (colorspace))
+
+static void stbir__decode_scanline(stbir__info* stbir_info, int n)
+{
+ int c;
+ int channels = stbir_info->channels;
+ int alpha_channel = stbir_info->alpha_channel;
+ int type = stbir_info->type;
+ int colorspace = stbir_info->colorspace;
+ int input_w = stbir_info->input_w;
+ size_t input_stride_bytes = stbir_info->input_stride_bytes;
+ float* decode_buffer = stbir__get_decode_buffer(stbir_info);
+ stbir_edge edge_horizontal = stbir_info->edge_horizontal;
+ stbir_edge edge_vertical = stbir_info->edge_vertical;
+ size_t in_buffer_row_offset = stbir__edge_wrap(edge_vertical, n, stbir_info->input_h) * input_stride_bytes;
+ const void* input_data = (char *) stbir_info->input_data + in_buffer_row_offset;
+ int max_x = input_w + stbir_info->horizontal_filter_pixel_margin;
+ int decode = STBIR__DECODE(type, colorspace);
+
+ int x = -stbir_info->horizontal_filter_pixel_margin;
+
+ // special handling for STBIR_EDGE_ZERO because it needs to return an item that doesn't appear in the input,
+ // and we want to avoid paying overhead on every pixel if not STBIR_EDGE_ZERO
+ if (edge_vertical == STBIR_EDGE_ZERO && (n < 0 || n >= stbir_info->input_h))
+ {
+ for (; x < max_x; x++)
+ for (c = 0; c < channels; c++)
+ decode_buffer[x*channels + c] = 0;
+ return;
+ }
+
+ switch (decode)
+ {
+ case STBIR__DECODE(STBIR_TYPE_UINT8, STBIR_COLORSPACE_LINEAR):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = ((float)((const unsigned char*)input_data)[input_pixel_index + c]) / stbir__max_uint8_as_float;
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT8, STBIR_COLORSPACE_SRGB):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = stbir__srgb_uchar_to_linear_float[((const unsigned char*)input_data)[input_pixel_index + c]];
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ decode_buffer[decode_pixel_index + alpha_channel] = ((float)((const unsigned char*)input_data)[input_pixel_index + alpha_channel]) / stbir__max_uint8_as_float;
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT16, STBIR_COLORSPACE_LINEAR):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = ((float)((const unsigned short*)input_data)[input_pixel_index + c]) / stbir__max_uint16_as_float;
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT16, STBIR_COLORSPACE_SRGB):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = stbir__srgb_to_linear(((float)((const unsigned short*)input_data)[input_pixel_index + c]) / stbir__max_uint16_as_float);
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ decode_buffer[decode_pixel_index + alpha_channel] = ((float)((const unsigned short*)input_data)[input_pixel_index + alpha_channel]) / stbir__max_uint16_as_float;
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT32, STBIR_COLORSPACE_LINEAR):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = (float)(((double)((const unsigned int*)input_data)[input_pixel_index + c]) / stbir__max_uint32_as_float);
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT32, STBIR_COLORSPACE_SRGB):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = stbir__srgb_to_linear((float)(((double)((const unsigned int*)input_data)[input_pixel_index + c]) / stbir__max_uint32_as_float));
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ decode_buffer[decode_pixel_index + alpha_channel] = (float)(((double)((const unsigned int*)input_data)[input_pixel_index + alpha_channel]) / stbir__max_uint32_as_float);
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_FLOAT, STBIR_COLORSPACE_LINEAR):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = ((const float*)input_data)[input_pixel_index + c];
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_FLOAT, STBIR_COLORSPACE_SRGB):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = stbir__srgb_to_linear(((const float*)input_data)[input_pixel_index + c]);
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ decode_buffer[decode_pixel_index + alpha_channel] = ((const float*)input_data)[input_pixel_index + alpha_channel];
+ }
+
+ break;
+
+ default:
+ STBIR_ASSERT(!"Unknown type/colorspace/channels combination.");
+ break;
+ }
+
+ if (!(stbir_info->flags & STBIR_FLAG_ALPHA_PREMULTIPLIED))
+ {
+ for (x = -stbir_info->horizontal_filter_pixel_margin; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+
+ // If the alpha value is 0 it will clobber the color values. Make sure it's not.
+ float alpha = decode_buffer[decode_pixel_index + alpha_channel];
+#ifndef STBIR_NO_ALPHA_EPSILON
+ if (stbir_info->type != STBIR_TYPE_FLOAT) {
+ alpha += STBIR_ALPHA_EPSILON;
+ decode_buffer[decode_pixel_index + alpha_channel] = alpha;
+ }
+#endif
+ for (c = 0; c < channels; c++)
+ {
+ if (c == alpha_channel)
+ continue;
+
+ decode_buffer[decode_pixel_index + c] *= alpha;
+ }
+ }
+ }
+
+ if (edge_horizontal == STBIR_EDGE_ZERO)
+ {
+ for (x = -stbir_info->horizontal_filter_pixel_margin; x < 0; x++)
+ {
+ for (c = 0; c < channels; c++)
+ decode_buffer[x*channels + c] = 0;
+ }
+ for (x = input_w; x < max_x; x++)
+ {
+ for (c = 0; c < channels; c++)
+ decode_buffer[x*channels + c] = 0;
+ }
+ }
+}
+
+static float* stbir__get_ring_buffer_entry(float* ring_buffer, int index, int ring_buffer_length)
+{
+ return &ring_buffer[index * ring_buffer_length];
+}
+
+static float* stbir__add_empty_ring_buffer_entry(stbir__info* stbir_info, int n)
+{
+ int ring_buffer_index;
+ float* ring_buffer;
+
+ stbir_info->ring_buffer_last_scanline = n;
+
+ if (stbir_info->ring_buffer_begin_index < 0)
+ {
+ ring_buffer_index = stbir_info->ring_buffer_begin_index = 0;
+ stbir_info->ring_buffer_first_scanline = n;
+ }
+ else
+ {
+ ring_buffer_index = (stbir_info->ring_buffer_begin_index + (stbir_info->ring_buffer_last_scanline - stbir_info->ring_buffer_first_scanline)) % stbir_info->ring_buffer_num_entries;
+ STBIR_ASSERT(ring_buffer_index != stbir_info->ring_buffer_begin_index);
+ }
+
+ ring_buffer = stbir__get_ring_buffer_entry(stbir_info->ring_buffer, ring_buffer_index, stbir_info->ring_buffer_length_bytes / sizeof(float));
+ memset(ring_buffer, 0, stbir_info->ring_buffer_length_bytes);
+
+ return ring_buffer;
+}
+
+
+static void stbir__resample_horizontal_upsample(stbir__info* stbir_info, float* output_buffer)
+{
+ int x, k;
+ int output_w = stbir_info->output_w;
+ int channels = stbir_info->channels;
+ float* decode_buffer = stbir__get_decode_buffer(stbir_info);
+ stbir__contributors* horizontal_contributors = stbir_info->horizontal_contributors;
+ float* horizontal_coefficients = stbir_info->horizontal_coefficients;
+ int coefficient_width = stbir_info->horizontal_coefficient_width;
+
+ for (x = 0; x < output_w; x++)
+ {
+ int n0 = horizontal_contributors[x].n0;
+ int n1 = horizontal_contributors[x].n1;
+
+ int out_pixel_index = x * channels;
+ int coefficient_group = coefficient_width * x;
+ int coefficient_counter = 0;
+
+ STBIR_ASSERT(n1 >= n0);
+ STBIR_ASSERT(n0 >= -stbir_info->horizontal_filter_pixel_margin);
+ STBIR_ASSERT(n1 >= -stbir_info->horizontal_filter_pixel_margin);
+ STBIR_ASSERT(n0 < stbir_info->input_w + stbir_info->horizontal_filter_pixel_margin);
+ STBIR_ASSERT(n1 < stbir_info->input_w + stbir_info->horizontal_filter_pixel_margin);
+
+ switch (channels) {
+ case 1:
+ for (k = n0; k <= n1; k++)
+ {
+ int in_pixel_index = k * 1;
+ float coefficient = horizontal_coefficients[coefficient_group + coefficient_counter++];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ }
+ break;
+ case 2:
+ for (k = n0; k <= n1; k++)
+ {
+ int in_pixel_index = k * 2;
+ float coefficient = horizontal_coefficients[coefficient_group + coefficient_counter++];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
+ }
+ break;
+ case 3:
+ for (k = n0; k <= n1; k++)
+ {
+ int in_pixel_index = k * 3;
+ float coefficient = horizontal_coefficients[coefficient_group + coefficient_counter++];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
+ output_buffer[out_pixel_index + 2] += decode_buffer[in_pixel_index + 2] * coefficient;
+ }
+ break;
+ case 4:
+ for (k = n0; k <= n1; k++)
+ {
+ int in_pixel_index = k * 4;
+ float coefficient = horizontal_coefficients[coefficient_group + coefficient_counter++];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
+ output_buffer[out_pixel_index + 2] += decode_buffer[in_pixel_index + 2] * coefficient;
+ output_buffer[out_pixel_index + 3] += decode_buffer[in_pixel_index + 3] * coefficient;
+ }
+ break;
+ default:
+ for (k = n0; k <= n1; k++)
+ {
+ int in_pixel_index = k * channels;
+ float coefficient = horizontal_coefficients[coefficient_group + coefficient_counter++];
+ int c;
+ STBIR_ASSERT(coefficient != 0);
+ for (c = 0; c < channels; c++)
+ output_buffer[out_pixel_index + c] += decode_buffer[in_pixel_index + c] * coefficient;
+ }
+ break;
+ }
+ }
+}
+
+static void stbir__resample_horizontal_downsample(stbir__info* stbir_info, float* output_buffer)
+{
+ int x, k;
+ int input_w = stbir_info->input_w;
+ int channels = stbir_info->channels;
+ float* decode_buffer = stbir__get_decode_buffer(stbir_info);
+ stbir__contributors* horizontal_contributors = stbir_info->horizontal_contributors;
+ float* horizontal_coefficients = stbir_info->horizontal_coefficients;
+ int coefficient_width = stbir_info->horizontal_coefficient_width;
+ int filter_pixel_margin = stbir_info->horizontal_filter_pixel_margin;
+ int max_x = input_w + filter_pixel_margin * 2;
+
+ STBIR_ASSERT(!stbir__use_width_upsampling(stbir_info));
+
+ switch (channels) {
+ case 1:
+ for (x = 0; x < max_x; x++)
+ {
+ int n0 = horizontal_contributors[x].n0;
+ int n1 = horizontal_contributors[x].n1;
+
+ int in_x = x - filter_pixel_margin;
+ int in_pixel_index = in_x * 1;
+ int max_n = n1;
+ int coefficient_group = coefficient_width * x;
+
+ for (k = n0; k <= max_n; k++)
+ {
+ int out_pixel_index = k * 1;
+ float coefficient = horizontal_coefficients[coefficient_group + k - n0];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ }
+ }
+ break;
+
+ case 2:
+ for (x = 0; x < max_x; x++)
+ {
+ int n0 = horizontal_contributors[x].n0;
+ int n1 = horizontal_contributors[x].n1;
+
+ int in_x = x - filter_pixel_margin;
+ int in_pixel_index = in_x * 2;
+ int max_n = n1;
+ int coefficient_group = coefficient_width * x;
+
+ for (k = n0; k <= max_n; k++)
+ {
+ int out_pixel_index = k * 2;
+ float coefficient = horizontal_coefficients[coefficient_group + k - n0];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
+ }
+ }
+ break;
+
+ case 3:
+ for (x = 0; x < max_x; x++)
+ {
+ int n0 = horizontal_contributors[x].n0;
+ int n1 = horizontal_contributors[x].n1;
+
+ int in_x = x - filter_pixel_margin;
+ int in_pixel_index = in_x * 3;
+ int max_n = n1;
+ int coefficient_group = coefficient_width * x;
+
+ for (k = n0; k <= max_n; k++)
+ {
+ int out_pixel_index = k * 3;
+ float coefficient = horizontal_coefficients[coefficient_group + k - n0];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
+ output_buffer[out_pixel_index + 2] += decode_buffer[in_pixel_index + 2] * coefficient;
+ }
+ }
+ break;
+
+ case 4:
+ for (x = 0; x < max_x; x++)
+ {
+ int n0 = horizontal_contributors[x].n0;
+ int n1 = horizontal_contributors[x].n1;
+
+ int in_x = x - filter_pixel_margin;
+ int in_pixel_index = in_x * 4;
+ int max_n = n1;
+ int coefficient_group = coefficient_width * x;
+
+ for (k = n0; k <= max_n; k++)
+ {
+ int out_pixel_index = k * 4;
+ float coefficient = horizontal_coefficients[coefficient_group + k - n0];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
+ output_buffer[out_pixel_index + 2] += decode_buffer[in_pixel_index + 2] * coefficient;
+ output_buffer[out_pixel_index + 3] += decode_buffer[in_pixel_index + 3] * coefficient;
+ }
+ }
+ break;
+
+ default:
+ for (x = 0; x < max_x; x++)
+ {
+ int n0 = horizontal_contributors[x].n0;
+ int n1 = horizontal_contributors[x].n1;
+
+ int in_x = x - filter_pixel_margin;
+ int in_pixel_index = in_x * channels;
+ int max_n = n1;
+ int coefficient_group = coefficient_width * x;
+
+ for (k = n0; k <= max_n; k++)
+ {
+ int c;
+ int out_pixel_index = k * channels;
+ float coefficient = horizontal_coefficients[coefficient_group + k - n0];
+ STBIR_ASSERT(coefficient != 0);
+ for (c = 0; c < channels; c++)
+ output_buffer[out_pixel_index + c] += decode_buffer[in_pixel_index + c] * coefficient;
+ }
+ }
+ break;
+ }
+}
+
+static void stbir__decode_and_resample_upsample(stbir__info* stbir_info, int n)
+{
+ // Decode the nth scanline from the source image into the decode buffer.
+ stbir__decode_scanline(stbir_info, n);
+
+ // Now resample it into the ring buffer.
+ if (stbir__use_width_upsampling(stbir_info))
+ stbir__resample_horizontal_upsample(stbir_info, stbir__add_empty_ring_buffer_entry(stbir_info, n));
+ else
+ stbir__resample_horizontal_downsample(stbir_info, stbir__add_empty_ring_buffer_entry(stbir_info, n));
+
+ // Now it's sitting in the ring buffer ready to be used as source for the vertical sampling.
+}
+
+static void stbir__decode_and_resample_downsample(stbir__info* stbir_info, int n)
+{
+ // Decode the nth scanline from the source image into the decode buffer.
+ stbir__decode_scanline(stbir_info, n);
+
+ memset(stbir_info->horizontal_buffer, 0, stbir_info->output_w * stbir_info->channels * sizeof(float));
+
+ // Now resample it into the horizontal buffer.
+ if (stbir__use_width_upsampling(stbir_info))
+ stbir__resample_horizontal_upsample(stbir_info, stbir_info->horizontal_buffer);
+ else
+ stbir__resample_horizontal_downsample(stbir_info, stbir_info->horizontal_buffer);
+
+ // Now it's sitting in the horizontal buffer ready to be distributed into the ring buffers.
+}
+
+// Get the specified scan line from the ring buffer.
+static float* stbir__get_ring_buffer_scanline(int get_scanline, float* ring_buffer, int begin_index, int first_scanline, int ring_buffer_num_entries, int ring_buffer_length)
+{
+ int ring_buffer_index = (begin_index + (get_scanline - first_scanline)) % ring_buffer_num_entries;
+ return stbir__get_ring_buffer_entry(ring_buffer, ring_buffer_index, ring_buffer_length);
+}
+
+
+static void stbir__encode_scanline(stbir__info* stbir_info, int num_pixels, void *output_buffer, float *encode_buffer, int channels, int alpha_channel, int decode)
+{
+ int x;
+ int n;
+ int num_nonalpha;
+ stbir_uint16 nonalpha[STBIR_MAX_CHANNELS];
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_PREMULTIPLIED))
+ {
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ float alpha = encode_buffer[pixel_index + alpha_channel];
+ float reciprocal_alpha = alpha ? 1.0f / alpha : 0;
+
+ // unrolling this produced a 1% slowdown upscaling a large RGBA linear-space image on my machine - stb
+ for (n = 0; n < channels; n++)
+ if (n != alpha_channel)
+ encode_buffer[pixel_index + n] *= reciprocal_alpha;
+
+ // We added in a small epsilon to prevent the color channel from being deleted with zero alpha.
+ // Because we only add it for integer types, it will automatically be discarded on integer
+ // conversion, so we don't need to subtract it back out (which would be problematic for
+ // numeric precision reasons).
+ }
+ }
+
+ // build a table of all channels that need colorspace correction, so
+ // we don't perform colorspace correction on channels that don't need it.
+ for (x = 0, num_nonalpha = 0; x < channels; ++x)
+ {
+ if (x != alpha_channel || (stbir_info->flags & STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ {
+ nonalpha[num_nonalpha++] = (stbir_uint16)x;
+ }
+ }
+
+ #define STBIR__ROUND_INT(f) ((int) ((f)+0.5))
+ #define STBIR__ROUND_UINT(f) ((stbir_uint32) ((f)+0.5))
+
+ #ifdef STBIR__SATURATE_INT
+ #define STBIR__ENCODE_LINEAR8(f) stbir__saturate8 (STBIR__ROUND_INT((f) * stbir__max_uint8_as_float ))
+ #define STBIR__ENCODE_LINEAR16(f) stbir__saturate16(STBIR__ROUND_INT((f) * stbir__max_uint16_as_float))
+ #else
+ #define STBIR__ENCODE_LINEAR8(f) (unsigned char ) STBIR__ROUND_INT(stbir__saturate(f) * stbir__max_uint8_as_float )
+ #define STBIR__ENCODE_LINEAR16(f) (unsigned short) STBIR__ROUND_INT(stbir__saturate(f) * stbir__max_uint16_as_float)
+ #endif
+
+ switch (decode)
+ {
+ case STBIR__DECODE(STBIR_TYPE_UINT8, STBIR_COLORSPACE_LINEAR):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < channels; n++)
+ {
+ int index = pixel_index + n;
+ ((unsigned char*)output_buffer)[index] = STBIR__ENCODE_LINEAR8(encode_buffer[index]);
+ }
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT8, STBIR_COLORSPACE_SRGB):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < num_nonalpha; n++)
+ {
+ int index = pixel_index + nonalpha[n];
+ ((unsigned char*)output_buffer)[index] = stbir__linear_to_srgb_uchar(encode_buffer[index]);
+ }
+
+ if (!(stbir_info->flags & STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ ((unsigned char *)output_buffer)[pixel_index + alpha_channel] = STBIR__ENCODE_LINEAR8(encode_buffer[pixel_index+alpha_channel]);
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT16, STBIR_COLORSPACE_LINEAR):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < channels; n++)
+ {
+ int index = pixel_index + n;
+ ((unsigned short*)output_buffer)[index] = STBIR__ENCODE_LINEAR16(encode_buffer[index]);
+ }
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT16, STBIR_COLORSPACE_SRGB):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < num_nonalpha; n++)
+ {
+ int index = pixel_index + nonalpha[n];
+ ((unsigned short*)output_buffer)[index] = (unsigned short)STBIR__ROUND_INT(stbir__linear_to_srgb(stbir__saturate(encode_buffer[index])) * stbir__max_uint16_as_float);
+ }
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ ((unsigned short*)output_buffer)[pixel_index + alpha_channel] = STBIR__ENCODE_LINEAR16(encode_buffer[pixel_index + alpha_channel]);
+ }
+
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT32, STBIR_COLORSPACE_LINEAR):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < channels; n++)
+ {
+ int index = pixel_index + n;
+ ((unsigned int*)output_buffer)[index] = (unsigned int)STBIR__ROUND_UINT(((double)stbir__saturate(encode_buffer[index])) * stbir__max_uint32_as_float);
+ }
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT32, STBIR_COLORSPACE_SRGB):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < num_nonalpha; n++)
+ {
+ int index = pixel_index + nonalpha[n];
+ ((unsigned int*)output_buffer)[index] = (unsigned int)STBIR__ROUND_UINT(((double)stbir__linear_to_srgb(stbir__saturate(encode_buffer[index]))) * stbir__max_uint32_as_float);
+ }
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ ((unsigned int*)output_buffer)[pixel_index + alpha_channel] = (unsigned int)STBIR__ROUND_INT(((double)stbir__saturate(encode_buffer[pixel_index + alpha_channel])) * stbir__max_uint32_as_float);
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_FLOAT, STBIR_COLORSPACE_LINEAR):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < channels; n++)
+ {
+ int index = pixel_index + n;
+ ((float*)output_buffer)[index] = encode_buffer[index];
+ }
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_FLOAT, STBIR_COLORSPACE_SRGB):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < num_nonalpha; n++)
+ {
+ int index = pixel_index + nonalpha[n];
+ ((float*)output_buffer)[index] = stbir__linear_to_srgb(encode_buffer[index]);
+ }
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ ((float*)output_buffer)[pixel_index + alpha_channel] = encode_buffer[pixel_index + alpha_channel];
+ }
+ break;
+
+ default:
+ STBIR_ASSERT(!"Unknown type/colorspace/channels combination.");
+ break;
+ }
+}
+
+static void stbir__resample_vertical_upsample(stbir__info* stbir_info, int n)
+{
+ int x, k;
+ int output_w = stbir_info->output_w;
+ stbir__contributors* vertical_contributors = stbir_info->vertical_contributors;
+ float* vertical_coefficients = stbir_info->vertical_coefficients;
+ int channels = stbir_info->channels;
+ int alpha_channel = stbir_info->alpha_channel;
+ int type = stbir_info->type;
+ int colorspace = stbir_info->colorspace;
+ int ring_buffer_entries = stbir_info->ring_buffer_num_entries;
+ void* output_data = stbir_info->output_data;
+ float* encode_buffer = stbir_info->encode_buffer;
+ int decode = STBIR__DECODE(type, colorspace);
+ int coefficient_width = stbir_info->vertical_coefficient_width;
+ int coefficient_counter;
+ int contributor = n;
+
+ float* ring_buffer = stbir_info->ring_buffer;
+ int ring_buffer_begin_index = stbir_info->ring_buffer_begin_index;
+ int ring_buffer_first_scanline = stbir_info->ring_buffer_first_scanline;
+ int ring_buffer_length = stbir_info->ring_buffer_length_bytes/sizeof(float);
+
+ int n0,n1, output_row_start;
+ int coefficient_group = coefficient_width * contributor;
+
+ n0 = vertical_contributors[contributor].n0;
+ n1 = vertical_contributors[contributor].n1;
+
+ output_row_start = n * stbir_info->output_stride_bytes;
+
+ STBIR_ASSERT(stbir__use_height_upsampling(stbir_info));
+
+ memset(encode_buffer, 0, output_w * sizeof(float) * channels);
+
+ // I tried reblocking this for better cache usage of encode_buffer
+ // (using x_outer, k, x_inner), but it lost speed. -- stb
+
+ coefficient_counter = 0;
+ switch (channels) {
+ case 1:
+ for (k = n0; k <= n1; k++)
+ {
+ int coefficient_index = coefficient_counter++;
+ float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length);
+ float coefficient = vertical_coefficients[coefficient_group + coefficient_index];
+ for (x = 0; x < output_w; ++x)
+ {
+ int in_pixel_index = x * 1;
+ encode_buffer[in_pixel_index + 0] += ring_buffer_entry[in_pixel_index + 0] * coefficient;
+ }
+ }
+ break;
+ case 2:
+ for (k = n0; k <= n1; k++)
+ {
+ int coefficient_index = coefficient_counter++;
+ float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length);
+ float coefficient = vertical_coefficients[coefficient_group + coefficient_index];
+ for (x = 0; x < output_w; ++x)
+ {
+ int in_pixel_index = x * 2;
+ encode_buffer[in_pixel_index + 0] += ring_buffer_entry[in_pixel_index + 0] * coefficient;
+ encode_buffer[in_pixel_index + 1] += ring_buffer_entry[in_pixel_index + 1] * coefficient;
+ }
+ }
+ break;
+ case 3:
+ for (k = n0; k <= n1; k++)
+ {
+ int coefficient_index = coefficient_counter++;
+ float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length);
+ float coefficient = vertical_coefficients[coefficient_group + coefficient_index];
+ for (x = 0; x < output_w; ++x)
+ {
+ int in_pixel_index = x * 3;
+ encode_buffer[in_pixel_index + 0] += ring_buffer_entry[in_pixel_index + 0] * coefficient;
+ encode_buffer[in_pixel_index + 1] += ring_buffer_entry[in_pixel_index + 1] * coefficient;
+ encode_buffer[in_pixel_index + 2] += ring_buffer_entry[in_pixel_index + 2] * coefficient;
+ }
+ }
+ break;
+ case 4:
+ for (k = n0; k <= n1; k++)
+ {
+ int coefficient_index = coefficient_counter++;
+ float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length);
+ float coefficient = vertical_coefficients[coefficient_group + coefficient_index];
+ for (x = 0; x < output_w; ++x)
+ {
+ int in_pixel_index = x * 4;
+ encode_buffer[in_pixel_index + 0] += ring_buffer_entry[in_pixel_index + 0] * coefficient;
+ encode_buffer[in_pixel_index + 1] += ring_buffer_entry[in_pixel_index + 1] * coefficient;
+ encode_buffer[in_pixel_index + 2] += ring_buffer_entry[in_pixel_index + 2] * coefficient;
+ encode_buffer[in_pixel_index + 3] += ring_buffer_entry[in_pixel_index + 3] * coefficient;
+ }
+ }
+ break;
+ default:
+ for (k = n0; k <= n1; k++)
+ {
+ int coefficient_index = coefficient_counter++;
+ float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length);
+ float coefficient = vertical_coefficients[coefficient_group + coefficient_index];
+ for (x = 0; x < output_w; ++x)
+ {
+ int in_pixel_index = x * channels;
+ int c;
+ for (c = 0; c < channels; c++)
+ encode_buffer[in_pixel_index + c] += ring_buffer_entry[in_pixel_index + c] * coefficient;
+ }
+ }
+ break;
+ }
+ stbir__encode_scanline(stbir_info, output_w, (char *) output_data + output_row_start, encode_buffer, channels, alpha_channel, decode);
+}
+
+static void stbir__resample_vertical_downsample(stbir__info* stbir_info, int n)
+{
+ int x, k;
+ int output_w = stbir_info->output_w;
+ stbir__contributors* vertical_contributors = stbir_info->vertical_contributors;
+ float* vertical_coefficients = stbir_info->vertical_coefficients;
+ int channels = stbir_info->channels;
+ int ring_buffer_entries = stbir_info->ring_buffer_num_entries;
+ float* horizontal_buffer = stbir_info->horizontal_buffer;
+ int coefficient_width = stbir_info->vertical_coefficient_width;
+ int contributor = n + stbir_info->vertical_filter_pixel_margin;
+
+ float* ring_buffer = stbir_info->ring_buffer;
+ int ring_buffer_begin_index = stbir_info->ring_buffer_begin_index;
+ int ring_buffer_first_scanline = stbir_info->ring_buffer_first_scanline;
+ int ring_buffer_length = stbir_info->ring_buffer_length_bytes/sizeof(float);
+ int n0,n1;
+
+ n0 = vertical_contributors[contributor].n0;
+ n1 = vertical_contributors[contributor].n1;
+
+ STBIR_ASSERT(!stbir__use_height_upsampling(stbir_info));
+
+ for (k = n0; k <= n1; k++)
+ {
+ int coefficient_index = k - n0;
+ int coefficient_group = coefficient_width * contributor;
+ float coefficient = vertical_coefficients[coefficient_group + coefficient_index];
+
+ float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length);
+
+ switch (channels) {
+ case 1:
+ for (x = 0; x < output_w; x++)
+ {
+ int in_pixel_index = x * 1;
+ ring_buffer_entry[in_pixel_index + 0] += horizontal_buffer[in_pixel_index + 0] * coefficient;
+ }
+ break;
+ case 2:
+ for (x = 0; x < output_w; x++)
+ {
+ int in_pixel_index = x * 2;
+ ring_buffer_entry[in_pixel_index + 0] += horizontal_buffer[in_pixel_index + 0] * coefficient;
+ ring_buffer_entry[in_pixel_index + 1] += horizontal_buffer[in_pixel_index + 1] * coefficient;
+ }
+ break;
+ case 3:
+ for (x = 0; x < output_w; x++)
+ {
+ int in_pixel_index = x * 3;
+ ring_buffer_entry[in_pixel_index + 0] += horizontal_buffer[in_pixel_index + 0] * coefficient;
+ ring_buffer_entry[in_pixel_index + 1] += horizontal_buffer[in_pixel_index + 1] * coefficient;
+ ring_buffer_entry[in_pixel_index + 2] += horizontal_buffer[in_pixel_index + 2] * coefficient;
+ }
+ break;
+ case 4:
+ for (x = 0; x < output_w; x++)
+ {
+ int in_pixel_index = x * 4;
+ ring_buffer_entry[in_pixel_index + 0] += horizontal_buffer[in_pixel_index + 0] * coefficient;
+ ring_buffer_entry[in_pixel_index + 1] += horizontal_buffer[in_pixel_index + 1] * coefficient;
+ ring_buffer_entry[in_pixel_index + 2] += horizontal_buffer[in_pixel_index + 2] * coefficient;
+ ring_buffer_entry[in_pixel_index + 3] += horizontal_buffer[in_pixel_index + 3] * coefficient;
+ }
+ break;
+ default:
+ for (x = 0; x < output_w; x++)
+ {
+ int in_pixel_index = x * channels;
+
+ int c;
+ for (c = 0; c < channels; c++)
+ ring_buffer_entry[in_pixel_index + c] += horizontal_buffer[in_pixel_index + c] * coefficient;
+ }
+ break;
+ }
+ }
+}
+
+static void stbir__buffer_loop_upsample(stbir__info* stbir_info)
+{
+ int y;
+ float scale_ratio = stbir_info->vertical_scale;
+ float out_scanlines_radius = stbir__filter_info_table[stbir_info->vertical_filter].support(1/scale_ratio) * scale_ratio;
+
+ STBIR_ASSERT(stbir__use_height_upsampling(stbir_info));
+
+ for (y = 0; y < stbir_info->output_h; y++)
+ {
+ float in_center_of_out = 0; // Center of the current out scanline in the in scanline space
+ int in_first_scanline = 0, in_last_scanline = 0;
+
+ stbir__calculate_sample_range_upsample(y, out_scanlines_radius, scale_ratio, stbir_info->vertical_shift, &in_first_scanline, &in_last_scanline, &in_center_of_out);
+
+ STBIR_ASSERT(in_last_scanline - in_first_scanline + 1 <= stbir_info->ring_buffer_num_entries);
+
+ if (stbir_info->ring_buffer_begin_index >= 0)
+ {
+ // Get rid of whatever we don't need anymore.
+ while (in_first_scanline > stbir_info->ring_buffer_first_scanline)
+ {
+ if (stbir_info->ring_buffer_first_scanline == stbir_info->ring_buffer_last_scanline)
+ {
+ // We just popped the last scanline off the ring buffer.
+ // Reset it to the empty state.
+ stbir_info->ring_buffer_begin_index = -1;
+ stbir_info->ring_buffer_first_scanline = 0;
+ stbir_info->ring_buffer_last_scanline = 0;
+ break;
+ }
+ else
+ {
+ stbir_info->ring_buffer_first_scanline++;
+ stbir_info->ring_buffer_begin_index = (stbir_info->ring_buffer_begin_index + 1) % stbir_info->ring_buffer_num_entries;
+ }
+ }
+ }
+
+ // Load in new ones.
+ if (stbir_info->ring_buffer_begin_index < 0)
+ stbir__decode_and_resample_upsample(stbir_info, in_first_scanline);
+
+ while (in_last_scanline > stbir_info->ring_buffer_last_scanline)
+ stbir__decode_and_resample_upsample(stbir_info, stbir_info->ring_buffer_last_scanline + 1);
+
+ // Now all buffers should be ready to write a row of vertical sampling.
+ stbir__resample_vertical_upsample(stbir_info, y);
+
+ STBIR_PROGRESS_REPORT((float)y / stbir_info->output_h);
+ }
+}
+
+static void stbir__empty_ring_buffer(stbir__info* stbir_info, int first_necessary_scanline)
+{
+ int output_stride_bytes = stbir_info->output_stride_bytes;
+ int channels = stbir_info->channels;
+ int alpha_channel = stbir_info->alpha_channel;
+ int type = stbir_info->type;
+ int colorspace = stbir_info->colorspace;
+ int output_w = stbir_info->output_w;
+ void* output_data = stbir_info->output_data;
+ int decode = STBIR__DECODE(type, colorspace);
+
+ float* ring_buffer = stbir_info->ring_buffer;
+ int ring_buffer_length = stbir_info->ring_buffer_length_bytes/sizeof(float);
+
+ if (stbir_info->ring_buffer_begin_index >= 0)
+ {
+ // Get rid of whatever we don't need anymore.
+ while (first_necessary_scanline > stbir_info->ring_buffer_first_scanline)
+ {
+ if (stbir_info->ring_buffer_first_scanline >= 0 && stbir_info->ring_buffer_first_scanline < stbir_info->output_h)
+ {
+ int output_row_start = stbir_info->ring_buffer_first_scanline * output_stride_bytes;
+ float* ring_buffer_entry = stbir__get_ring_buffer_entry(ring_buffer, stbir_info->ring_buffer_begin_index, ring_buffer_length);
+ stbir__encode_scanline(stbir_info, output_w, (char *) output_data + output_row_start, ring_buffer_entry, channels, alpha_channel, decode);
+ STBIR_PROGRESS_REPORT((float)stbir_info->ring_buffer_first_scanline / stbir_info->output_h);
+ }
+
+ if (stbir_info->ring_buffer_first_scanline == stbir_info->ring_buffer_last_scanline)
+ {
+ // We just popped the last scanline off the ring buffer.
+ // Reset it to the empty state.
+ stbir_info->ring_buffer_begin_index = -1;
+ stbir_info->ring_buffer_first_scanline = 0;
+ stbir_info->ring_buffer_last_scanline = 0;
+ break;
+ }
+ else
+ {
+ stbir_info->ring_buffer_first_scanline++;
+ stbir_info->ring_buffer_begin_index = (stbir_info->ring_buffer_begin_index + 1) % stbir_info->ring_buffer_num_entries;
+ }
+ }
+ }
+}
+
+static void stbir__buffer_loop_downsample(stbir__info* stbir_info)
+{
+ int y;
+ float scale_ratio = stbir_info->vertical_scale;
+ int output_h = stbir_info->output_h;
+ float in_pixels_radius = stbir__filter_info_table[stbir_info->vertical_filter].support(scale_ratio) / scale_ratio;
+ int pixel_margin = stbir_info->vertical_filter_pixel_margin;
+ int max_y = stbir_info->input_h + pixel_margin;
+
+ STBIR_ASSERT(!stbir__use_height_upsampling(stbir_info));
+
+ for (y = -pixel_margin; y < max_y; y++)
+ {
+ float out_center_of_in; // Center of the current out scanline in the in scanline space
+ int out_first_scanline, out_last_scanline;
+
+ stbir__calculate_sample_range_downsample(y, in_pixels_radius, scale_ratio, stbir_info->vertical_shift, &out_first_scanline, &out_last_scanline, &out_center_of_in);
+
+ STBIR_ASSERT(out_last_scanline - out_first_scanline + 1 <= stbir_info->ring_buffer_num_entries);
+
+ if (out_last_scanline < 0 || out_first_scanline >= output_h)
+ continue;
+
+ stbir__empty_ring_buffer(stbir_info, out_first_scanline);
+
+ stbir__decode_and_resample_downsample(stbir_info, y);
+
+ // Load in new ones.
+ if (stbir_info->ring_buffer_begin_index < 0)
+ stbir__add_empty_ring_buffer_entry(stbir_info, out_first_scanline);
+
+ while (out_last_scanline > stbir_info->ring_buffer_last_scanline)
+ stbir__add_empty_ring_buffer_entry(stbir_info, stbir_info->ring_buffer_last_scanline + 1);
+
+ // Now the horizontal buffer is ready to write to all ring buffer rows.
+ stbir__resample_vertical_downsample(stbir_info, y);
+ }
+
+ stbir__empty_ring_buffer(stbir_info, stbir_info->output_h);
+}
+
+static void stbir__setup(stbir__info *info, int input_w, int input_h, int output_w, int output_h, int channels)
+{
+ info->input_w = input_w;
+ info->input_h = input_h;
+ info->output_w = output_w;
+ info->output_h = output_h;
+ info->channels = channels;
+}
+
+static void stbir__calculate_transform(stbir__info *info, float s0, float t0, float s1, float t1, float *transform)
+{
+ info->s0 = s0;
+ info->t0 = t0;
+ info->s1 = s1;
+ info->t1 = t1;
+
+ if (transform)
+ {
+ info->horizontal_scale = transform[0];
+ info->vertical_scale = transform[1];
+ info->horizontal_shift = transform[2];
+ info->vertical_shift = transform[3];
+ }
+ else
+ {
+ info->horizontal_scale = ((float)info->output_w / info->input_w) / (s1 - s0);
+ info->vertical_scale = ((float)info->output_h / info->input_h) / (t1 - t0);
+
+ info->horizontal_shift = s0 * info->output_w / (s1 - s0);
+ info->vertical_shift = t0 * info->output_h / (t1 - t0);
+ }
+}
+
+static void stbir__choose_filter(stbir__info *info, stbir_filter h_filter, stbir_filter v_filter)
+{
+ if (h_filter == 0)
+ h_filter = stbir__use_upsampling(info->horizontal_scale) ? STBIR_DEFAULT_FILTER_UPSAMPLE : STBIR_DEFAULT_FILTER_DOWNSAMPLE;
+ if (v_filter == 0)
+ v_filter = stbir__use_upsampling(info->vertical_scale) ? STBIR_DEFAULT_FILTER_UPSAMPLE : STBIR_DEFAULT_FILTER_DOWNSAMPLE;
+ info->horizontal_filter = h_filter;
+ info->vertical_filter = v_filter;
+}
+
+static stbir_uint32 stbir__calculate_memory(stbir__info *info)
+{
+ int pixel_margin = stbir__get_filter_pixel_margin(info->horizontal_filter, info->horizontal_scale);
+ int filter_height = stbir__get_filter_pixel_width(info->vertical_filter, info->vertical_scale);
+
+ info->horizontal_num_contributors = stbir__get_contributors(info->horizontal_scale, info->horizontal_filter, info->input_w, info->output_w);
+ info->vertical_num_contributors = stbir__get_contributors(info->vertical_scale , info->vertical_filter , info->input_h, info->output_h);
+
+ // One extra entry because floating point precision problems sometimes cause an extra to be necessary.
+ info->ring_buffer_num_entries = filter_height + 1;
+
+ info->horizontal_contributors_size = info->horizontal_num_contributors * sizeof(stbir__contributors);
+ info->horizontal_coefficients_size = stbir__get_total_horizontal_coefficients(info) * sizeof(float);
+ info->vertical_contributors_size = info->vertical_num_contributors * sizeof(stbir__contributors);
+ info->vertical_coefficients_size = stbir__get_total_vertical_coefficients(info) * sizeof(float);
+ info->decode_buffer_size = (info->input_w + pixel_margin * 2) * info->channels * sizeof(float);
+ info->horizontal_buffer_size = info->output_w * info->channels * sizeof(float);
+ info->ring_buffer_size = info->output_w * info->channels * info->ring_buffer_num_entries * sizeof(float);
+ info->encode_buffer_size = info->output_w * info->channels * sizeof(float);
+
+ STBIR_ASSERT(info->horizontal_filter != 0);
+ STBIR_ASSERT(info->horizontal_filter < STBIR__ARRAY_SIZE(stbir__filter_info_table)); // this now happens too late
+ STBIR_ASSERT(info->vertical_filter != 0);
+ STBIR_ASSERT(info->vertical_filter < STBIR__ARRAY_SIZE(stbir__filter_info_table)); // this now happens too late
+
+ if (stbir__use_height_upsampling(info))
+ // The horizontal buffer is for when we're downsampling the height and we
+ // can't output the result of sampling the decode buffer directly into the
+ // ring buffers.
+ info->horizontal_buffer_size = 0;
+ else
+ // The encode buffer is to retain precision in the height upsampling method
+ // and isn't used when height downsampling.
+ info->encode_buffer_size = 0;
+
+ return info->horizontal_contributors_size + info->horizontal_coefficients_size
+ + info->vertical_contributors_size + info->vertical_coefficients_size
+ + info->decode_buffer_size + info->horizontal_buffer_size
+ + info->ring_buffer_size + info->encode_buffer_size;
+}
+
+static int stbir__resize_allocated(stbir__info *info,
+ const void* input_data, int input_stride_in_bytes,
+ void* output_data, int output_stride_in_bytes,
+ int alpha_channel, stbir_uint32 flags, stbir_datatype type,
+ stbir_edge edge_horizontal, stbir_edge edge_vertical, stbir_colorspace colorspace,
+ void* tempmem, size_t tempmem_size_in_bytes)
+{
+ size_t memory_required = stbir__calculate_memory(info);
+
+ int width_stride_input = input_stride_in_bytes ? input_stride_in_bytes : info->channels * info->input_w * stbir__type_size[type];
+ int width_stride_output = output_stride_in_bytes ? output_stride_in_bytes : info->channels * info->output_w * stbir__type_size[type];
+
+#ifdef STBIR_DEBUG_OVERWRITE_TEST
+#define OVERWRITE_ARRAY_SIZE 8
+ unsigned char overwrite_output_before_pre[OVERWRITE_ARRAY_SIZE];
+ unsigned char overwrite_tempmem_before_pre[OVERWRITE_ARRAY_SIZE];
+ unsigned char overwrite_output_after_pre[OVERWRITE_ARRAY_SIZE];
+ unsigned char overwrite_tempmem_after_pre[OVERWRITE_ARRAY_SIZE];
+
+ size_t begin_forbidden = width_stride_output * (info->output_h - 1) + info->output_w * info->channels * stbir__type_size[type];
+ memcpy(overwrite_output_before_pre, &((unsigned char*)output_data)[-OVERWRITE_ARRAY_SIZE], OVERWRITE_ARRAY_SIZE);
+ memcpy(overwrite_output_after_pre, &((unsigned char*)output_data)[begin_forbidden], OVERWRITE_ARRAY_SIZE);
+ memcpy(overwrite_tempmem_before_pre, &((unsigned char*)tempmem)[-OVERWRITE_ARRAY_SIZE], OVERWRITE_ARRAY_SIZE);
+ memcpy(overwrite_tempmem_after_pre, &((unsigned char*)tempmem)[tempmem_size_in_bytes], OVERWRITE_ARRAY_SIZE);
+#endif
+
+ STBIR_ASSERT(info->channels >= 0);
+ STBIR_ASSERT(info->channels <= STBIR_MAX_CHANNELS);
+
+ if (info->channels < 0 || info->channels > STBIR_MAX_CHANNELS)
+ return 0;
+
+ STBIR_ASSERT(info->horizontal_filter < STBIR__ARRAY_SIZE(stbir__filter_info_table));
+ STBIR_ASSERT(info->vertical_filter < STBIR__ARRAY_SIZE(stbir__filter_info_table));
+
+ if (info->horizontal_filter >= STBIR__ARRAY_SIZE(stbir__filter_info_table))
+ return 0;
+ if (info->vertical_filter >= STBIR__ARRAY_SIZE(stbir__filter_info_table))
+ return 0;
+
+ if (alpha_channel < 0)
+ flags |= STBIR_FLAG_ALPHA_USES_COLORSPACE | STBIR_FLAG_ALPHA_PREMULTIPLIED;
+
+ if (!(flags&STBIR_FLAG_ALPHA_USES_COLORSPACE) || !(flags&STBIR_FLAG_ALPHA_PREMULTIPLIED))
+ STBIR_ASSERT(alpha_channel >= 0 && alpha_channel < info->channels);
+
+ if (alpha_channel >= info->channels)
+ return 0;
+
+ STBIR_ASSERT(tempmem);
+
+ if (!tempmem)
+ return 0;
+
+ STBIR_ASSERT(tempmem_size_in_bytes >= memory_required);
+
+ if (tempmem_size_in_bytes < memory_required)
+ return 0;
+
+ memset(tempmem, 0, tempmem_size_in_bytes);
+
+ info->input_data = input_data;
+ info->input_stride_bytes = width_stride_input;
+
+ info->output_data = output_data;
+ info->output_stride_bytes = width_stride_output;
+
+ info->alpha_channel = alpha_channel;
+ info->flags = flags;
+ info->type = type;
+ info->edge_horizontal = edge_horizontal;
+ info->edge_vertical = edge_vertical;
+ info->colorspace = colorspace;
+
+ info->horizontal_coefficient_width = stbir__get_coefficient_width (info->horizontal_filter, info->horizontal_scale);
+ info->vertical_coefficient_width = stbir__get_coefficient_width (info->vertical_filter , info->vertical_scale );
+ info->horizontal_filter_pixel_width = stbir__get_filter_pixel_width (info->horizontal_filter, info->horizontal_scale);
+ info->vertical_filter_pixel_width = stbir__get_filter_pixel_width (info->vertical_filter , info->vertical_scale );
+ info->horizontal_filter_pixel_margin = stbir__get_filter_pixel_margin(info->horizontal_filter, info->horizontal_scale);
+ info->vertical_filter_pixel_margin = stbir__get_filter_pixel_margin(info->vertical_filter , info->vertical_scale );
+
+ info->ring_buffer_length_bytes = info->output_w * info->channels * sizeof(float);
+ info->decode_buffer_pixels = info->input_w + info->horizontal_filter_pixel_margin * 2;
+
+#define STBIR__NEXT_MEMPTR(current, newtype) (newtype*)(((unsigned char*)current) + current##_size)
+
+ info->horizontal_contributors = (stbir__contributors *) tempmem;
+ info->horizontal_coefficients = STBIR__NEXT_MEMPTR(info->horizontal_contributors, float);
+ info->vertical_contributors = STBIR__NEXT_MEMPTR(info->horizontal_coefficients, stbir__contributors);
+ info->vertical_coefficients = STBIR__NEXT_MEMPTR(info->vertical_contributors, float);
+ info->decode_buffer = STBIR__NEXT_MEMPTR(info->vertical_coefficients, float);
+
+ if (stbir__use_height_upsampling(info))
+ {
+ info->horizontal_buffer = NULL;
+ info->ring_buffer = STBIR__NEXT_MEMPTR(info->decode_buffer, float);
+ info->encode_buffer = STBIR__NEXT_MEMPTR(info->ring_buffer, float);
+
+ STBIR_ASSERT((size_t)STBIR__NEXT_MEMPTR(info->encode_buffer, unsigned char) == (size_t)tempmem + tempmem_size_in_bytes);
+ }
+ else
+ {
+ info->horizontal_buffer = STBIR__NEXT_MEMPTR(info->decode_buffer, float);
+ info->ring_buffer = STBIR__NEXT_MEMPTR(info->horizontal_buffer, float);
+ info->encode_buffer = NULL;
+
+ STBIR_ASSERT((size_t)STBIR__NEXT_MEMPTR(info->ring_buffer, unsigned char) == (size_t)tempmem + tempmem_size_in_bytes);
+ }
+
+#undef STBIR__NEXT_MEMPTR
+
+ // This signals that the ring buffer is empty
+ info->ring_buffer_begin_index = -1;
+
+ stbir__calculate_filters(info->horizontal_contributors, info->horizontal_coefficients, info->horizontal_filter, info->horizontal_scale, info->horizontal_shift, info->input_w, info->output_w);
+ stbir__calculate_filters(info->vertical_contributors, info->vertical_coefficients, info->vertical_filter, info->vertical_scale, info->vertical_shift, info->input_h, info->output_h);
+
+ STBIR_PROGRESS_REPORT(0);
+
+ if (stbir__use_height_upsampling(info))
+ stbir__buffer_loop_upsample(info);
+ else
+ stbir__buffer_loop_downsample(info);
+
+ STBIR_PROGRESS_REPORT(1);
+
+#ifdef STBIR_DEBUG_OVERWRITE_TEST
+ STBIR_ASSERT(memcmp(overwrite_output_before_pre, &((unsigned char*)output_data)[-OVERWRITE_ARRAY_SIZE], OVERWRITE_ARRAY_SIZE) == 0);
+ STBIR_ASSERT(memcmp(overwrite_output_after_pre, &((unsigned char*)output_data)[begin_forbidden], OVERWRITE_ARRAY_SIZE) == 0);
+ STBIR_ASSERT(memcmp(overwrite_tempmem_before_pre, &((unsigned char*)tempmem)[-OVERWRITE_ARRAY_SIZE], OVERWRITE_ARRAY_SIZE) == 0);
+ STBIR_ASSERT(memcmp(overwrite_tempmem_after_pre, &((unsigned char*)tempmem)[tempmem_size_in_bytes], OVERWRITE_ARRAY_SIZE) == 0);
+#endif
+
+ return 1;
+}
+
+
+static int stbir__resize_arbitrary(
+ void *alloc_context,
+ const void* input_data, int input_w, int input_h, int input_stride_in_bytes,
+ void* output_data, int output_w, int output_h, int output_stride_in_bytes,
+ float s0, float t0, float s1, float t1, float *transform,
+ int channels, int alpha_channel, stbir_uint32 flags, stbir_datatype type,
+ stbir_filter h_filter, stbir_filter v_filter,
+ stbir_edge edge_horizontal, stbir_edge edge_vertical, stbir_colorspace colorspace)
+{
+ stbir__info info;
+ int result;
+ size_t memory_required;
+ void* extra_memory;
+
+ stbir__setup(&info, input_w, input_h, output_w, output_h, channels);
+ stbir__calculate_transform(&info, s0,t0,s1,t1,transform);
+ stbir__choose_filter(&info, h_filter, v_filter);
+ memory_required = stbir__calculate_memory(&info);
+ extra_memory = STBIR_MALLOC(memory_required, alloc_context);
+
+ if (!extra_memory)
+ return 0;
+
+ result = stbir__resize_allocated(&info, input_data, input_stride_in_bytes,
+ output_data, output_stride_in_bytes,
+ alpha_channel, flags, type,
+ edge_horizontal, edge_vertical,
+ colorspace, extra_memory, memory_required);
+
+ STBIR_FREE(extra_memory, alloc_context);
+
+ return result;
+}
+
+STBIRDEF int stbir_resize_uint8( const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels)
+{
+ return stbir__resize_arbitrary(NULL, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,-1,0, STBIR_TYPE_UINT8, STBIR_FILTER_DEFAULT, STBIR_FILTER_DEFAULT,
+ STBIR_EDGE_CLAMP, STBIR_EDGE_CLAMP, STBIR_COLORSPACE_LINEAR);
+}
+
+STBIRDEF int stbir_resize_float( const float *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ float *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels)
+{
+ return stbir__resize_arbitrary(NULL, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,-1,0, STBIR_TYPE_FLOAT, STBIR_FILTER_DEFAULT, STBIR_FILTER_DEFAULT,
+ STBIR_EDGE_CLAMP, STBIR_EDGE_CLAMP, STBIR_COLORSPACE_LINEAR);
+}
+
+STBIRDEF int stbir_resize_uint8_srgb(const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags)
+{
+ return stbir__resize_arbitrary(NULL, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,alpha_channel,flags, STBIR_TYPE_UINT8, STBIR_FILTER_DEFAULT, STBIR_FILTER_DEFAULT,
+ STBIR_EDGE_CLAMP, STBIR_EDGE_CLAMP, STBIR_COLORSPACE_SRGB);
+}
+
+STBIRDEF int stbir_resize_uint8_srgb_edgemode(const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode)
+{
+ return stbir__resize_arbitrary(NULL, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,alpha_channel,flags, STBIR_TYPE_UINT8, STBIR_FILTER_DEFAULT, STBIR_FILTER_DEFAULT,
+ edge_wrap_mode, edge_wrap_mode, STBIR_COLORSPACE_SRGB);
+}
+
+STBIRDEF int stbir_resize_uint8_generic( const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space,
+ void *alloc_context)
+{
+ return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,alpha_channel,flags, STBIR_TYPE_UINT8, filter, filter,
+ edge_wrap_mode, edge_wrap_mode, space);
+}
+
+STBIRDEF int stbir_resize_uint16_generic(const stbir_uint16 *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ stbir_uint16 *output_pixels , int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space,
+ void *alloc_context)
+{
+ return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,alpha_channel,flags, STBIR_TYPE_UINT16, filter, filter,
+ edge_wrap_mode, edge_wrap_mode, space);
+}
+
+
+STBIRDEF int stbir_resize_float_generic( const float *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ float *output_pixels , int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space,
+ void *alloc_context)
+{
+ return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,alpha_channel,flags, STBIR_TYPE_FLOAT, filter, filter,
+ edge_wrap_mode, edge_wrap_mode, space);
+}
+
+
+STBIRDEF int stbir_resize( const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ stbir_datatype datatype,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical,
+ stbir_filter filter_horizontal, stbir_filter filter_vertical,
+ stbir_colorspace space, void *alloc_context)
+{
+ return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,alpha_channel,flags, datatype, filter_horizontal, filter_vertical,
+ edge_mode_horizontal, edge_mode_vertical, space);
+}
+
+
+STBIRDEF int stbir_resize_subpixel(const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ stbir_datatype datatype,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical,
+ stbir_filter filter_horizontal, stbir_filter filter_vertical,
+ stbir_colorspace space, void *alloc_context,
+ float x_scale, float y_scale,
+ float x_offset, float y_offset)
+{
+ float transform[4];
+ transform[0] = x_scale;
+ transform[1] = y_scale;
+ transform[2] = x_offset;
+ transform[3] = y_offset;
+ return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,transform,num_channels,alpha_channel,flags, datatype, filter_horizontal, filter_vertical,
+ edge_mode_horizontal, edge_mode_vertical, space);
+}
+
+STBIRDEF int stbir_resize_region( const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ stbir_datatype datatype,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical,
+ stbir_filter filter_horizontal, stbir_filter filter_vertical,
+ stbir_colorspace space, void *alloc_context,
+ float s0, float t0, float s1, float t1)
+{
+ return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ s0,t0,s1,t1,NULL,num_channels,alpha_channel,flags, datatype, filter_horizontal, filter_vertical,
+ edge_mode_horizontal, edge_mode_vertical, space);
+}
+
+#endif // STB_IMAGE_RESIZE_IMPLEMENTATION
+
+/*
+------------------------------------------------------------------------------
+This software is available under 2 licenses -- choose whichever you prefer.
+------------------------------------------------------------------------------
+ALTERNATIVE A - MIT License
+Copyright (c) 2017 Sean Barrett
+Permission is hereby granted, free of charge, to any person obtaining a copy of
+this software and associated documentation files (the "Software"), to deal in
+the Software without restriction, including without limitation the rights to
+use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
+of the Software, and to permit persons to whom the Software is furnished to do
+so, subject to the following conditions:
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+------------------------------------------------------------------------------
+ALTERNATIVE B - Public Domain (www.unlicense.org)
+This is free and unencumbered software released into the public domain.
+Anyone is free to copy, modify, publish, use, compile, sell, or distribute this
+software, either in source code form or as a compiled binary, for any purpose,
+commercial or non-commercial, and by any means.
+In jurisdictions that recognize copyright laws, the author or authors of this
+software dedicate any and all copyright interest in the software to the public
+domain. We make this dedication for the benefit of the public at large and to
+the detriment of our heirs and successors. We intend this dedication to be an
+overt act of relinquishment in perpetuity of all present and future rights to
+this software under copyright law.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+------------------------------------------------------------------------------
+*/
--- /dev/null
+input: "data"
+input_shape {
+ dim: 1
+ dim: 3
+ dim: 227
+ dim: 227
+}
--- /dev/null
+#! /usr/bin/env python3
+
+# Copyright 2017 Intel Corporation.
+# The source code, information and material ("Material") contained herein is
+# owned by Intel Corporation or its suppliers or licensors, and title to such
+# Material remains with Intel Corporation or its suppliers or licensors.
+# The Material contains proprietary information of Intel or its suppliers and
+# licensors. The Material is protected by worldwide copyright laws and treaty
+# provisions.
+# No part of the Material may be used, copied, reproduced, modified, published,
+# uploaded, posted, transmitted, distributed or disclosed in any way without
+# Intel's prior express written permission. No license under any patent,
+# copyright or other intellectual property rights in the Material is granted to
+# or conferred upon you, either expressly, by implication, inducement, estoppel
+# or otherwise.
+# Any license under such intellectual property rights must be express and
+# approved by Intel in writing.
+
+from mvnc import mvncapi as mvnc
+import sys
+import numpy
+import cv2
+import time
+import csv
+import os
+import sys
+
+dim=(227,227)
+EXAMPLES_BASE_DIR='../../'
+
+# ***************************************************************
+# get labels
+# ***************************************************************
+labels_file=EXAMPLES_BASE_DIR+'data/ilsvrc12/synset_words.txt'
+labels=numpy.loadtxt(labels_file,str,delimiter='\t')
+
+# ***************************************************************
+# configure the NCS
+# ***************************************************************
+mvnc.SetGlobalOption(mvnc.GlobalOption.LOG_LEVEL, 2)
+
+# ***************************************************************
+# Get a list of ALL the sticks that are plugged in
+# ***************************************************************
+devices = mvnc.EnumerateDevices()
+if len(devices) == 0:
+ print('No devices found')
+ quit()
+
+# ***************************************************************
+# Pick the first stick to run the network
+# ***************************************************************
+device = mvnc.Device(devices[0])
+
+# ***************************************************************
+# Open the NCS
+# ***************************************************************
+device.OpenDevice()
+
+network_blob='graph'
+
+#Load blob
+with open(network_blob, mode='rb') as f:
+ blob = f.read()
+
+graph = device.AllocateGraph(blob)
+
+# ***************************************************************
+# Load the image
+# ***************************************************************
+ilsvrc_mean = numpy.load(EXAMPLES_BASE_DIR+'data/ilsvrc12/ilsvrc_2012_mean.npy').mean(1).mean(1) #loading the mean file
+img = cv2.imread(EXAMPLES_BASE_DIR+'data/images/nps_electric_guitar.png')
+img=cv2.resize(img,dim)
+img = img.astype(numpy.float32)
+img[:,:,0] = (img[:,:,0] - ilsvrc_mean[0])
+img[:,:,1] = (img[:,:,1] - ilsvrc_mean[1])
+img[:,:,2] = (img[:,:,2] - ilsvrc_mean[2])
+
+# ***************************************************************
+# Send the image to the NCS
+# ***************************************************************
+graph.LoadTensor(img.astype(numpy.float16), 'user object')
+
+# ***************************************************************
+# Get the result from the NCS
+# ***************************************************************
+output, userobj = graph.GetResult()
+
+# ***************************************************************
+# Print the results of the inference form the NCS
+# ***************************************************************
+order = output.argsort()[::-1][:6]
+print('\n------- predictions --------')
+for i in range(0,5):
+ print ('prediction ' + str(i) + ' (probability ' + str(output[order[i]]*100) + '%) is ' + labels[order[i]] + ' label index is: ' + str(order[i]) )
+
+
+# ***************************************************************
+# Clean up the graph and the device
+# ***************************************************************
+graph.DeallocateGraph()
+device.CloseDevice()
+
+
+
+
--- /dev/null
+
+ifneq ($(findstring movidius, $(PYTHONPATH)), movidius)
+ export PYTHONPATH:=/opt/movidius/caffe/python:/opt/movidius/mvnc/python:$(PYTHONPATH)
+endif
+
+NCCOMPILE = mvNCCompile
+NCPROFILE = mvNCProfile
+NCCHECK = mvNCCheck
+
+
+PROTOTXT_FILENAME= deploy.prototxt
+GET_PROTOTXT = wget -P . https://raw.githubusercontent.com/BVLC/caffe/master/models/bvlc_googlenet/deploy.prototxt
+
+CAFFEMODEL_FILENAME = bvlc_googlenet.caffemodel
+GET_CAFFEMODEL = wget -P . -N http://dl.caffe.berkeleyvision.org/${CAFFEMODEL_FILENAME}
+
+
+.PHONY: all
+all: profile check compile cpp
+
+.PHONY: prereqs
+prereqs:
+ @echo "\nmaking prereqs"
+ (cd ../../data/ilsvrc12; make)
+ @sed -i 's/\r//' run.py
+ @chmod +x run.py
+
+.PHONY: prototxt
+prototxt: prereqs
+ @echo "\nmaking prototxt"
+ @if [ -e ${PROTOTXT_FILENAME} ] ; \
+ then \
+ echo "Prototxt file already exists"; \
+ else \
+ echo "Downloading Prototxt file"; \
+ ${GET_PROTOTXT}; \
+ if [ -e ${PROTOTXT_FILENAME} ] ; \
+ then \
+ echo "Adding input shape to prototxt file."; \
+ awk 'NR <2 {print}' < ${PROTOTXT_FILENAME} > temp; cat input_shape.prototxt >> temp; awk 'NR > 7 {print}' < ${PROTOTXT_FILENAME} >> temp; mv temp ${PROTOTXT_FILENAME}; \
+ else \
+ echo "***\nError - Could not download prototxt file. Check network and proxy settings \n***\n"; \
+ exit 1; \
+ fi ; \
+ fi
+
+.PHONY: caffemodel
+caffemodel:
+ @echo "\nmaking caffemodel"
+ @if [ -e ${CAFFEMODEL_FILENAME} ] ; \
+ then \
+ echo "caffemodel file already exists"; \
+ else \
+ echo "Downloading caffemodel file"; \
+ ${GET_CAFFEMODEL}; \
+ if ! [ -e ${CAFFEMODEL_FILENAME} ] ; \
+ then \
+ echo "***\nError - Could not download caffemodel file. Check network and proxy settings \n***\n"; \
+ exit 1; \
+ fi ; \
+ fi
+
+.PHONY: profile
+profile: prototxt
+ @echo "\nmaking profile"
+ ${NCPROFILE} ${PROTOTXT_FILENAME} -s 12
+
+.PHONY: browse_profile
+browse_profile: profile
+ @echo "\nmaking browse_profile"
+ @if [ -e output_report.html ] ; \
+ then \
+ firefox output_report.html & \
+ else \
+ @echo "***\nError - output_report.html not found" ; \
+ fi ;
+
+.PHONY: compile
+compile: prototxt caffemodel
+ @echo "\nmaking compile"
+ ${NCCOMPILE} -w ${CAFFEMODEL_FILENAME} -s 12 ${PROTOTXT_FILENAME}
+
+.PHONY: check
+check: prototxt caffemodel
+ @echo "\nmaking check"
+#-${NCCHECK} -w ${CAFFEMODEL_FILENAME} -i ../../data/images/cat.jpg -s 12 -id 281 ${PROTOTXT_FILENAME} -S 255 -M ../../data/ilsvrc12/ilsvrc_2012_mean.npy
+ ${NCCHECK} -w ${CAFFEMODEL_FILENAME} -i ../../data/images/nps_electric_guitar.png -s 12 -id 546 ${PROTOTXT_FILENAME} -S 255 -M 110
+
+.PHONY: run
+run: compile
+ @echo "\nmaking run"
+ ./run.py
+
+.PHONY: run_py
+run_py: compile
+ @echo "\nmaking run_py"
+ ./run.py
+
+.PHONY: cpp
+cpp: cpp/run.cpp cpp/fp16.c
+ @echo "\nmaking cpp"
+ g++ cpp/run.cpp cpp/fp16.c -o cpp/run_cpp -lmvnc
+
+.PHONY: run_cpp
+run_cpp: compile cpp
+ @echo "\nmaking run_cpp"
+ cd cpp; ./run_cpp; cd ..
+
+.PHONY: help
+help:
+ @echo "possible make targets: ";
+ @echo " make help - shows this message";
+ @echo " make all - makes the following: prototxt, profile, compile, check, cpp, run_py, run_cpp";
+ @echo " make prototxt - downloads and adds input shape to Caffe prototxt file";
+ @echo " make caffemodel - downloads the caffemodel for the network"
+ @echo " make compile - runs SDK compiler tool to compile the NCS graph file for the network";
+ @echo " make check - runs SDK checker tool to verify an NCS graph file";
+ @echo " make profile - runs the SDK profiler tool to profile the network creating output_report.html";
+ @echo " make browse_profile - runs the SDK profiler tool and brings up report in browser.";
+ @echo " make cpp - builds the run_cpp executable example";
+ @echo " make run_cpp - runs the run_cpp executable example program";
+ @echo " make run_py - runs the run.py python example program";
+ @echo " make clean - removes all created content"
+
+clean_caffe_model:
+ @echo "\nmaking clean_caffe_model"
+ rm -f ${PROTOTXT_FILENAME}
+ rm -f ${CAFFEMODEL_FILENAME}
+
+clean: clean_caffe_model
+ @echo "\nmaking clean"
+ rm -f graph
+ rm -f output.gv
+ rm -f output.gv.svg
+ rm -f output_report.html
+ rm -f output_expected.npy
+ rm -f zero_weights.caffemodel
+ rm -f output_result.npy
+ rm -f output_val.csv
+ rm -f cpp/run_cpp
--- /dev/null
+# Introduction
+The [GoogLeNet](https://github.com/BVLC/caffe/tree/master/models/bvlc_googlenet) network can be used for image classification. The provided Makefile does the following
+1. Downloads the Caffe prototxt file and makes any changes necessary to work with the Movidius Neural Compute SDK
+2. Downloads and generates the required ilsvrc12 data
+3. Downloads the .caffemodel file which was trained and provided by BVLC.
+3. Profiles, Compiles and Checks the network using the Neural Compute SDK.
+4. There is a python example (run.py) and a C++ example (cpp/run.cpp) which both do a single inference on an image as an example of how to use the network with the Neural Compute API thats provided in the Neural Compute SDK.
+
+# Makefile
+Provided Makefile describes various targets that help with the above mentioned tasks.
+
+## make help
+Shows makefile possible targets and brief descriptions.
+
+## make all
+Makes the following: prototxt, caffemodel, profile, compile, check, cpp, run, run_cpp.
+
+## make prototxt
+Downloads the Caffe prototxt file and makes a few changes necessary to work with the Movidius Neural Compute SDK.
+
+## make caffemodel
+Downloads the Caffe model file
+
+## make profile
+Runs the provided network on the NCS and generates per layer statistics that are helpful for understanding the performance of the network on the Neural Compute Stick. Output diplayed on terminal and the output_report.html file is also created. Demonstrates NCSDK tool: mvNCProfile
+
+## make browse_profile
+profiles the network similar to make profile and then brings up output_report.html in a browser. Demonstrates NCSDK tool: mvNCProfile
+
+## make compile
+Uses the network description and the trained weights files to generate a Movidius internal 'graph' format file. This file is later loaded on the Neural Compute Stick where the inferences on the network can be executed. Demonstrates NCSDK tool: mvNCCompile
+
+## make check
+Runs the network on Caffe on the CPU and compares results when run on the Neural Compute Stick. Consistency results are output to the terminal. Demonstrates the NCSDK tool: mvNCCheck.
+
+## make run_py
+Runs the provided run.py python script which sends a single image to the Neural Compute Stick and receives and displays the inference results.
+
+## make cpp
+Builds the C++ example program run_cpp which can be executed with make run_cpp.
+
+## make run_cpp
+Runs the provided run_cpp executable program that is built via make cpp. This program sends a single image to the Neural Compute Stick and receives and displays the inference results.
+
+## make clean
+Removes all the temporary and target files that are created by the Makefile.
--- /dev/null
+#! /bin/bash
+
+g++ run.cpp fp16.c -o run_cpp -lmvnc
+
+
+
--- /dev/null
+#include "fp16.h"
+
+// Copied from Numpy
+
+static unsigned half2float(unsigned short h)
+{
+ unsigned short h_exp, h_sig;
+ unsigned f_sgn, f_exp, f_sig;
+
+ h_exp = (h&0x7c00u);
+ f_sgn = ((unsigned)h&0x8000u) << 16;
+ switch (h_exp) {
+ case 0x0000u: /* 0 or subnormal */
+ h_sig = (h&0x03ffu);
+ /* Signed zero */
+ if (h_sig == 0) {
+ return f_sgn;
+ }
+ /* Subnormal */
+ h_sig <<= 1;
+ while ((h_sig&0x0400u) == 0) {
+ h_sig <<= 1;
+ h_exp++;
+ }
+ f_exp = ((unsigned)(127 - 15 - h_exp)) << 23;
+ f_sig = ((unsigned)(h_sig&0x03ffu)) << 13;
+ return f_sgn + f_exp + f_sig;
+ case 0x7c00u: /* inf or NaN */
+ /* All-ones exponent and a copy of the significand */
+ return f_sgn + 0x7f800000u + (((unsigned)(h&0x03ffu)) << 13);
+ default: /* normalized */
+ /* Just need to adjust the exponent and shift */
+ return f_sgn + (((unsigned)(h&0x7fffu) + 0x1c000u) << 13);
+ }
+}
+
+unsigned short float2half(unsigned f)
+{
+ unsigned f_exp, f_sig;
+ unsigned short h_sgn, h_exp, h_sig;
+
+ h_sgn = (unsigned short) ((f&0x80000000u) >> 16);
+ f_exp = (f&0x7f800000u);
+
+ /* Exponent overflow/NaN converts to signed inf/NaN */
+ if (f_exp >= 0x47800000u) {
+ if (f_exp == 0x7f800000u) {
+ /* Inf or NaN */
+ f_sig = (f&0x007fffffu);
+ if (f_sig != 0) {
+ /* NaN - propagate the flag in the significand... */
+ unsigned short ret = (unsigned short) (0x7c00u + (f_sig >> 13));
+ /* ...but make sure it stays a NaN */
+ if (ret == 0x7c00u) {
+ ret++;
+ }
+ return h_sgn + ret;
+ } else {
+ /* signed inf */
+ return (unsigned short) (h_sgn + 0x7c00u);
+ }
+ } else {
+ /* overflow to signed inf */
+#if NPY_HALF_GENERATE_OVERFLOW
+ npy_set_floatstatus_overflow();
+#endif
+ return (unsigned short) (h_sgn + 0x7c00u);
+ }
+ }
+
+ /* Exponent underflow converts to a subnormal half or signed zero */
+ if (f_exp <= 0x38000000u) {
+ /*
+ * Signed zeros, subnormal floats, and floats with small
+ * exponents all convert to signed zero halfs.
+ */
+ if (f_exp < 0x33000000u) {
+#if NPY_HALF_GENERATE_UNDERFLOW
+ /* If f != 0, it underflowed to 0 */
+ if ((f&0x7fffffff) != 0) {
+ npy_set_floatstatus_underflow();
+ }
+#endif
+ return h_sgn;
+ }
+ /* Make the subnormal significand */
+ f_exp >>= 23;
+ f_sig = (0x00800000u + (f&0x007fffffu));
+#if NPY_HALF_GENERATE_UNDERFLOW
+ /* If it's not exactly represented, it underflowed */
+ if ((f_sig&(((unsigned)1 << (126 - f_exp)) - 1)) != 0) {
+ npy_set_floatstatus_underflow();
+ }
+#endif
+ f_sig >>= (113 - f_exp);
+ /* Handle rounding by adding 1 to the bit beyond half precision */
+#if NPY_HALF_ROUND_TIES_TO_EVEN
+ /*
+ * If the last bit in the half significand is 0 (already even), and
+ * the remaining bit pattern is 1000...0, then we do not add one
+ * to the bit after the half significand. In all other cases, we do.
+ */
+ if ((f_sig&0x00003fffu) != 0x00001000u) {
+ f_sig += 0x00001000u;
+ }
+#else
+ f_sig += 0x00001000u;
+#endif
+ h_sig = (unsigned short) (f_sig >> 13);
+ /*
+ * If the rounding causes a bit to spill into h_exp, it will
+ * increment h_exp from zero to one and h_sig will be zero.
+ * This is the correct result.
+ */
+ return (unsigned short) (h_sgn + h_sig);
+ }
+
+ /* Regular case with no overflow or underflow */
+ h_exp = (unsigned short) ((f_exp - 0x38000000u) >> 13);
+ /* Handle rounding by adding 1 to the bit beyond half precision */
+ f_sig = (f&0x007fffffu);
+#if NPY_HALF_ROUND_TIES_TO_EVEN
+ /*
+ * If the last bit in the half significand is 0 (already even), and
+ * the remaining bit pattern is 1000...0, then we do not add one
+ * to the bit after the half significand. In all other cases, we do.
+ */
+ if ((f_sig&0x00003fffu) != 0x00001000u) {
+ f_sig += 0x00001000u;
+ }
+#else
+ f_sig += 0x00001000u;
+#endif
+ h_sig = (unsigned short) (f_sig >> 13);
+ /*
+ * If the rounding causes a bit to spill into h_exp, it will
+ * increment h_exp by one and h_sig will be zero. This is the
+ * correct result. h_exp may increment to 15, at greatest, in
+ * which case the result overflows to a signed inf.
+ */
+#if NPY_HALF_GENERATE_OVERFLOW
+ h_sig += h_exp;
+ if (h_sig == 0x7c00u) {
+ npy_set_floatstatus_overflow();
+ }
+ return h_sgn + h_sig;
+#else
+ return h_sgn + h_exp + h_sig;
+#endif
+}
+
+void floattofp16(unsigned char *dst, float *src, unsigned nelem)
+{
+ unsigned i;
+ unsigned short *_dst = (unsigned short *)dst;
+ unsigned *_src = (unsigned *)src;
+
+ for(i = 0; i < nelem; i++)
+ _dst[i] = float2half(_src[i]);
+}
+
+void fp16tofloat(float *dst, unsigned char *src, unsigned nelem)
+{
+ unsigned i;
+ unsigned *_dst = (unsigned *)dst;
+ unsigned short *_src = (unsigned short *)src;
+
+ for(i = 0; i < nelem; i++)
+ _dst[i] = half2float(_src[i]);
+}
+
--- /dev/null
+// Copied from Numpy
+
+
+static unsigned half2float(unsigned short h);
+unsigned short float2half(unsigned f);
+void floattofp16(unsigned char *dst, float *src, unsigned nelem);
+void fp16tofloat(float *dst, unsigned char *src, unsigned nelem);
+
+
--- /dev/null
+// Copyright 2017 Intel Corporation.
+// The source code, information and material ("Material") contained herein is
+// owned by Intel Corporation or its suppliers or licensors, and title to such
+// Material remains with Intel Corporation or its suppliers or licensors.
+// The Material contains proprietary information of Intel or its suppliers and
+// licensors. The Material is protected by worldwide copyright laws and treaty
+// provisions.
+// No part of the Material may be used, copied, reproduced, modified, published,
+// uploaded, posted, transmitted, distributed or disclosed in any way without
+// Intel's prior express written permission. No license under any patent,
+// copyright or other intellectual property rights in the Material is granted to
+// or conferred upon you, either expressly, by implication, inducement, estoppel
+// or otherwise.
+// Any license under such intellectual property rights must be express and
+// approved by Intel in writing.
+
+#include <stdio.h>
+#include <stdlib.h>
+
+#define STB_IMAGE_IMPLEMENTATION
+#include "stb_image.h"
+
+#define STB_IMAGE_RESIZE_IMPLEMENTATION
+#include "stb_image_resize.h"
+
+#include "fp16.h"
+#include <mvnc.h>
+
+
+// somewhat arbitrary buffer size for the device name
+#define NAME_SIZE 100
+
+// graph file name - assume we are running in this directory: ncsdk/examples/caffe/GoogLeNet/cpp
+#define GRAPH_FILE_NAME "../graph"
+
+// image file name - assume we are running in this directory: ncsdk/examples/caffe/GoogLeNet/cpp
+#define IMAGE_FILE_NAME "../../../data/images/nps_electric_guitar.png"
+
+
+// 16 bits. will use this to store half precision floats since C++ has no
+// built in support for it.
+typedef unsigned short half;
+
+// GoogleNet image dimensions, network mean values for each channel in BGR order.
+const int networkDim = 224;
+float networkMean[] = {0.40787054*255.0, 0.45752458*255.0, 0.48109378*255.0};
+
+// Load a graph file
+// caller must free the buffer returned.
+void *LoadFile(const char *path, unsigned int *length)
+{
+ FILE *fp;
+ char *buf;
+
+ fp = fopen(path, "rb");
+ if(fp == NULL)
+ return 0;
+ fseek(fp, 0, SEEK_END);
+ *length = ftell(fp);
+ rewind(fp);
+ if(!(buf = (char*) malloc(*length)))
+ {
+ fclose(fp);
+ return 0;
+ }
+ if(fread(buf, 1, *length, fp) != *length)
+ {
+ fclose(fp);
+ free(buf);
+ return 0;
+ }
+ fclose(fp);
+ return buf;
+}
+
+
+half *LoadImage(const char *path, int reqsize, float *mean)
+{
+ int width, height, cp, i;
+ unsigned char *img, *imgresized;
+ float *imgfp32;
+ half *imgfp16;
+
+ img = stbi_load(path, &width, &height, &cp, 3);
+ if(!img)
+ {
+ printf("The picture %s could not be loaded\n", path);
+ return 0;
+ }
+ imgresized = (unsigned char*) malloc(3*reqsize*reqsize);
+ if(!imgresized)
+ {
+ free(img);
+ perror("malloc");
+ return 0;
+ }
+ stbir_resize_uint8(img, width, height, 0, imgresized, reqsize, reqsize, 0, 3);
+ free(img);
+ imgfp32 = (float*) malloc(sizeof(*imgfp32) * reqsize * reqsize * 3);
+ if(!imgfp32)
+ {
+ free(imgresized);
+ perror("malloc");
+ return 0;
+ }
+ for(i = 0; i < reqsize * reqsize * 3; i++)
+ imgfp32[i] = imgresized[i];
+ free(imgresized);
+ imgfp16 = (half*) malloc(sizeof(*imgfp16) * reqsize * reqsize * 3);
+ if(!imgfp16)
+ {
+ free(imgfp32);
+ perror("malloc");
+ return 0;
+ }
+ for(i = 0; i < reqsize*reqsize; i++)
+ {
+ float blue, green, red;
+ blue = imgfp32[3*i+2];
+ green = imgfp32[3*i+1];
+ red = imgfp32[3*i+0];
+
+ imgfp32[3*i+0] = blue-mean[0];
+ imgfp32[3*i+1] = green-mean[1];
+ imgfp32[3*i+2] = red-mean[2];
+
+ // uncomment to see what values are getting passed to mvncLoadTensor() before conversion to half float
+ //printf("Blue: %f, Grean: %f, Red: %f \n", imgfp32[3*i+0], imgfp32[3*i+1], imgfp32[3*i+2]);
+ }
+ floattofp16((unsigned char *)imgfp16, imgfp32, 3*reqsize*reqsize);
+ free(imgfp32);
+ return imgfp16;
+}
+
+
+int main(int argc, char** argv)
+{
+ mvncStatus retCode;
+ void *deviceHandle;
+ char devName[NAME_SIZE];
+ retCode = mvncGetDeviceName(0, devName, NAME_SIZE);
+ if (retCode != MVNC_OK)
+ { // failed to get device name, maybe none plugged in.
+ printf("No NCS devices found\n");
+ exit(-1);
+ }
+
+ // Try to open the NCS device via the device name
+ retCode = mvncOpenDevice(devName, &deviceHandle);
+ if (retCode != MVNC_OK)
+ { // failed to open the device.
+ printf("Could not open NCS device\n");
+ exit(-1);
+ }
+
+ // deviceHandle is ready to use now.
+ // Pass it to other NC API calls as needed and close it when finished.
+ printf("Successfully opened NCS device!\n");
+
+ // Now read in a graph file
+ unsigned int graphFileLen;
+ void* graphFileBuf = LoadFile(GRAPH_FILE_NAME, &graphFileLen);
+
+ // allocate the graph
+ void* graphHandle;
+ retCode = mvncAllocateGraph(deviceHandle, &graphHandle, graphFileBuf, graphFileLen);
+ if (retCode != MVNC_OK)
+ { // error allocating graph
+ printf("Could not allocate graph for file: %s\n", GRAPH_FILE_NAME);
+ printf("Error from mvncAllocateGraph is: %d\n", retCode);
+ }
+ else
+ { // successfully allocated graph. Now graphHandle is ready to go.
+ // use graphHandle for other API calls and call mvncDeallocateGraph
+ // when done with it.
+ printf("Successfully allocated graph for %s\n", GRAPH_FILE_NAME);
+
+ // LoadImage will read image from disk, convert channels to floats
+ // subtract network mean for each value in each channel. Then, convert
+ // floats to half precision floats and return pointer to the buffer
+ // of half precision floats (Fp16s)
+ half* imageBufFp16 = LoadImage(IMAGE_FILE_NAME, networkDim, networkMean);
+
+ // calculate the length of the buffer that contains the half precision floats.
+ // 3 channels * width * height * sizeof a 16bit float
+ unsigned int lenBufFp16 = 3*networkDim*networkDim*sizeof(*imageBufFp16);
+
+ // start the inference with mvncLoadTensor()
+ retCode = mvncLoadTensor(graphHandle, imageBufFp16, lenBufFp16, NULL);
+ if (retCode != MVNC_OK)
+ { // error loading tensor
+ printf("Could not load tensor\n");
+ printf("Error from mvncLoadTensor is: %d\n", retCode);
+ }
+ else
+ { // the inference has been started, now call mvncGetResult() for the
+ // inference result
+ printf("Successfully loaded the tensor for image %s\n", IMAGE_FILE_NAME);
+
+ void* resultData16;
+ void* userParam;
+ unsigned int lenResultData;
+ retCode = mvncGetResult(graphHandle, &resultData16, &lenResultData, &userParam);
+ if (retCode == MVNC_OK)
+ { // Successfully got the result. The inference result is in the buffer pointed to by resultData
+ printf("Successfully got the inference result for image %s\n", IMAGE_FILE_NAME);
+ printf("resultData is %d bytes which is %d 16-bit floats.\n", lenResultData, lenResultData/(int)sizeof(half));
+
+ // convert half precision floats to full floats
+ int numResults = lenResultData / sizeof(half);
+ float* resultData32;
+ resultData32 = (float*)malloc(numResults * sizeof(*resultData32));
+ fp16tofloat(resultData32, (unsigned char*)resultData16, numResults);
+
+ float maxResult = 0.0;
+ int maxIndex = -1;
+ for (int index = 0; index < numResults; index++)
+ {
+ // printf("Category %d is: %f\n", index, resultData32[index]);
+ if (resultData32[index] > maxResult)
+ {
+ maxResult = resultData32[index];
+ maxIndex = index;
+ }
+ }
+ printf("Index of top result is: %d\n", maxIndex);
+ printf("Probability of top result is: %f\n", resultData32[maxIndex]);
+ }
+ }
+
+ retCode = mvncDeallocateGraph(graphHandle);
+ graphHandle = NULL;
+ }
+
+ free(graphFileBuf);
+ retCode = mvncCloseDevice(deviceHandle);
+ deviceHandle = NULL;
+}
--- /dev/null
+/* stb_image - v2.15 - public domain image loader - http://nothings.org/stb_image.h
+ no warranty implied; use at your own risk
+
+ Do this:
+ #define STB_IMAGE_IMPLEMENTATION
+ before you include this file in *one* C or C++ file to create the implementation.
+
+ // i.e. it should look like this:
+ #include ...
+ #include ...
+ #include ...
+ #define STB_IMAGE_IMPLEMENTATION
+ #include "stb_image.h"
+
+ You can #define STBI_ASSERT(x) before the #include to avoid using assert.h.
+ And #define STBI_MALLOC, STBI_REALLOC, and STBI_FREE to avoid using malloc,realloc,free
+
+
+ QUICK NOTES:
+ Primarily of interest to game developers and other people who can
+ avoid problematic images and only need the trivial interface
+
+ JPEG baseline & progressive (12 bpc/arithmetic not supported, same as stock IJG lib)
+ PNG 1/2/4/8/16-bit-per-channel
+
+ TGA (not sure what subset, if a subset)
+ BMP non-1bpp, non-RLE
+ PSD (composited view only, no extra channels, 8/16 bit-per-channel)
+
+ GIF (*comp always reports as 4-channel)
+ HDR (radiance rgbE format)
+ PIC (Softimage PIC)
+ PNM (PPM and PGM binary only)
+
+ Animated GIF still needs a proper API, but here's one way to do it:
+ http://gist.github.com/urraka/685d9a6340b26b830d49
+
+ - decode from memory or through FILE (define STBI_NO_STDIO to remove code)
+ - decode from arbitrary I/O callbacks
+ - SIMD acceleration on x86/x64 (SSE2) and ARM (NEON)
+
+ Full documentation under "DOCUMENTATION" below.
+
+
+LICENSE
+
+ See end of file for license information.
+
+RECENT REVISION HISTORY:
+
+ 2.15 (2017-03-18) fix png-1,2,4; all Imagenet JPGs; no runtime SSE detection on GCC
+ 2.14 (2017-03-03) remove deprecated STBI_JPEG_OLD; fixes for Imagenet JPGs
+ 2.13 (2016-12-04) experimental 16-bit API, only for PNG so far; fixes
+ 2.12 (2016-04-02) fix typo in 2.11 PSD fix that caused crashes
+ 2.11 (2016-04-02) 16-bit PNGS; enable SSE2 in non-gcc x64
+ RGB-format JPEG; remove white matting in PSD;
+ allocate large structures on the stack;
+ correct channel count for PNG & BMP
+ 2.10 (2016-01-22) avoid warning introduced in 2.09
+ 2.09 (2016-01-16) 16-bit TGA; comments in PNM files; STBI_REALLOC_SIZED
+ 2.08 (2015-09-13) fix to 2.07 cleanup, reading RGB PSD as RGBA
+ 2.07 (2015-09-13) partial animated GIF support
+ limited 16-bit PSD support
+ minor bugs, code cleanup, and compiler warnings
+
+ See end of file for full revision history.
+
+
+ ============================ Contributors =========================
+
+ Image formats Extensions, features
+ Sean Barrett (jpeg, png, bmp) Jetro Lauha (stbi_info)
+ Nicolas Schulz (hdr, psd) Martin "SpartanJ" Golini (stbi_info)
+ Jonathan Dummer (tga) James "moose2000" Brown (iPhone PNG)
+ Jean-Marc Lienher (gif) Ben "Disch" Wenger (io callbacks)
+ Tom Seddon (pic) Omar Cornut (1/2/4-bit PNG)
+ Thatcher Ulrich (psd) Nicolas Guillemot (vertical flip)
+ Ken Miller (pgm, ppm) Richard Mitton (16-bit PSD)
+ github:urraka (animated gif) Junggon Kim (PNM comments)
+ Daniel Gibson (16-bit TGA)
+ socks-the-fox (16-bit PNG)
+ Jeremy Sawicki (handle all ImageNet JPGs)
+ Optimizations & bugfixes
+ Fabian "ryg" Giesen
+ Arseny Kapoulkine
+
+ Bug & warning fixes
+ Marc LeBlanc David Woo Guillaume George Martins Mozeiko
+ Christpher Lloyd Jerry Jansson Joseph Thomson Phil Jordan
+ Dave Moore Roy Eltham Hayaki Saito Nathan Reed
+ Won Chun Luke Graham Johan Duparc Nick Verigakis
+ the Horde3D community Thomas Ruf Ronny Chevalier Baldur Karlsson
+ Janez Zemva John Bartholomew Michal Cichon github:rlyeh
+ Jonathan Blow Ken Hamada Tero Hanninen github:romigrou
+ Laurent Gomila Cort Stratton Sergio Gonzalez github:svdijk
+ Aruelien Pocheville Thibault Reuille Cass Everitt github:snagar
+ Ryamond Barbiero Paul Du Bois Engin Manap github:Zelex
+ Michaelangel007@github Philipp Wiesemann Dale Weiler github:grim210
+ Oriol Ferrer Mesia Josh Tobin Matthew Gregan github:sammyhw
+ Blazej Dariusz Roszkowski Gregory Mullen github:phprus
+
+*/
+
+#ifndef STBI_INCLUDE_STB_IMAGE_H
+#define STBI_INCLUDE_STB_IMAGE_H
+
+// DOCUMENTATION
+//
+// Limitations:
+// - no 16-bit-per-channel PNG
+// - no 12-bit-per-channel JPEG
+// - no JPEGs with arithmetic coding
+// - no 1-bit BMP
+// - GIF always returns *comp=4
+//
+// Basic usage (see HDR discussion below for HDR usage):
+// int x,y,n;
+// unsigned char *data = stbi_load(filename, &x, &y, &n, 0);
+// // ... process data if not NULL ...
+// // ... x = width, y = height, n = # 8-bit components per pixel ...
+// // ... replace '0' with '1'..'4' to force that many components per pixel
+// // ... but 'n' will always be the number that it would have been if you said 0
+// stbi_image_free(data)
+//
+// Standard parameters:
+// int *x -- outputs image width in pixels
+// int *y -- outputs image height in pixels
+// int *channels_in_file -- outputs # of image components in image file
+// int desired_channels -- if non-zero, # of image components requested in result
+//
+// The return value from an image loader is an 'unsigned char *' which points
+// to the pixel data, or NULL on an allocation failure or if the image is
+// corrupt or invalid. The pixel data consists of *y scanlines of *x pixels,
+// with each pixel consisting of N interleaved 8-bit components; the first
+// pixel pointed to is top-left-most in the image. There is no padding between
+// image scanlines or between pixels, regardless of format. The number of
+// components N is 'req_comp' if req_comp is non-zero, or *comp otherwise.
+// If req_comp is non-zero, *comp has the number of components that _would_
+// have been output otherwise. E.g. if you set req_comp to 4, you will always
+// get RGBA output, but you can check *comp to see if it's trivially opaque
+// because e.g. there were only 3 channels in the source image.
+//
+// An output image with N components has the following components interleaved
+// in this order in each pixel:
+//
+// N=#comp components
+// 1 grey
+// 2 grey, alpha
+// 3 red, green, blue
+// 4 red, green, blue, alpha
+//
+// If image loading fails for any reason, the return value will be NULL,
+// and *x, *y, *comp will be unchanged. The function stbi_failure_reason()
+// can be queried for an extremely brief, end-user unfriendly explanation
+// of why the load failed. Define STBI_NO_FAILURE_STRINGS to avoid
+// compiling these strings at all, and STBI_FAILURE_USERMSG to get slightly
+// more user-friendly ones.
+//
+// Paletted PNG, BMP, GIF, and PIC images are automatically depalettized.
+//
+// ===========================================================================
+//
+// Philosophy
+//
+// stb libraries are designed with the following priorities:
+//
+// 1. easy to use
+// 2. easy to maintain
+// 3. good performance
+//
+// Sometimes I let "good performance" creep up in priority over "easy to maintain",
+// and for best performance I may provide less-easy-to-use APIs that give higher
+// performance, in addition to the easy to use ones. Nevertheless, it's important
+// to keep in mind that from the standpoint of you, a client of this library,
+// all you care about is #1 and #3, and stb libraries DO NOT emphasize #3 above all.
+//
+// Some secondary priorities arise directly from the first two, some of which
+// make more explicit reasons why performance can't be emphasized.
+//
+// - Portable ("ease of use")
+// - Small source code footprint ("easy to maintain")
+// - No dependencies ("ease of use")
+//
+// ===========================================================================
+//
+// I/O callbacks
+//
+// I/O callbacks allow you to read from arbitrary sources, like packaged
+// files or some other source. Data read from callbacks are processed
+// through a small internal buffer (currently 128 bytes) to try to reduce
+// overhead.
+//
+// The three functions you must define are "read" (reads some bytes of data),
+// "skip" (skips some bytes of data), "eof" (reports if the stream is at the end).
+//
+// ===========================================================================
+//
+// SIMD support
+//
+// The JPEG decoder will try to automatically use SIMD kernels on x86 when
+// supported by the compiler. For ARM Neon support, you must explicitly
+// request it.
+//
+// (The old do-it-yourself SIMD API is no longer supported in the current
+// code.)
+//
+// On x86, SSE2 will automatically be used when available based on a run-time
+// test; if not, the generic C versions are used as a fall-back. On ARM targets,
+// the typical path is to have separate builds for NEON and non-NEON devices
+// (at least this is true for iOS and Android). Therefore, the NEON support is
+// toggled by a build flag: define STBI_NEON to get NEON loops.
+//
+// If for some reason you do not want to use any of SIMD code, or if
+// you have issues compiling it, you can disable it entirely by
+// defining STBI_NO_SIMD.
+//
+// ===========================================================================
+//
+// HDR image support (disable by defining STBI_NO_HDR)
+//
+// stb_image now supports loading HDR images in general, and currently
+// the Radiance .HDR file format, although the support is provided
+// generically. You can still load any file through the existing interface;
+// if you attempt to load an HDR file, it will be automatically remapped to
+// LDR, assuming gamma 2.2 and an arbitrary scale factor defaulting to 1;
+// both of these constants can be reconfigured through this interface:
+//
+// stbi_hdr_to_ldr_gamma(2.2f);
+// stbi_hdr_to_ldr_scale(1.0f);
+//
+// (note, do not use _inverse_ constants; stbi_image will invert them
+// appropriately).
+//
+// Additionally, there is a new, parallel interface for loading files as
+// (linear) floats to preserve the full dynamic range:
+//
+// float *data = stbi_loadf(filename, &x, &y, &n, 0);
+//
+// If you load LDR images through this interface, those images will
+// be promoted to floating point values, run through the inverse of
+// constants corresponding to the above:
+//
+// stbi_ldr_to_hdr_scale(1.0f);
+// stbi_ldr_to_hdr_gamma(2.2f);
+//
+// Finally, given a filename (or an open file or memory block--see header
+// file for details) containing image data, you can query for the "most
+// appropriate" interface to use (that is, whether the image is HDR or
+// not), using:
+//
+// stbi_is_hdr(char *filename);
+//
+// ===========================================================================
+//
+// iPhone PNG support:
+//
+// By default we convert iphone-formatted PNGs back to RGB, even though
+// they are internally encoded differently. You can disable this conversion
+// by by calling stbi_convert_iphone_png_to_rgb(0), in which case
+// you will always just get the native iphone "format" through (which
+// is BGR stored in RGB).
+//
+// Call stbi_set_unpremultiply_on_load(1) as well to force a divide per
+// pixel to remove any premultiplied alpha *only* if the image file explicitly
+// says there's premultiplied data (currently only happens in iPhone images,
+// and only if iPhone convert-to-rgb processing is on).
+//
+// ===========================================================================
+//
+// ADDITIONAL CONFIGURATION
+//
+// - You can suppress implementation of any of the decoders to reduce
+// your code footprint by #defining one or more of the following
+// symbols before creating the implementation.
+//
+// STBI_NO_JPEG
+// STBI_NO_PNG
+// STBI_NO_BMP
+// STBI_NO_PSD
+// STBI_NO_TGA
+// STBI_NO_GIF
+// STBI_NO_HDR
+// STBI_NO_PIC
+// STBI_NO_PNM (.ppm and .pgm)
+//
+// - You can request *only* certain decoders and suppress all other ones
+// (this will be more forward-compatible, as addition of new decoders
+// doesn't require you to disable them explicitly):
+//
+// STBI_ONLY_JPEG
+// STBI_ONLY_PNG
+// STBI_ONLY_BMP
+// STBI_ONLY_PSD
+// STBI_ONLY_TGA
+// STBI_ONLY_GIF
+// STBI_ONLY_HDR
+// STBI_ONLY_PIC
+// STBI_ONLY_PNM (.ppm and .pgm)
+//
+// - If you use STBI_NO_PNG (or _ONLY_ without PNG), and you still
+// want the zlib decoder to be available, #define STBI_SUPPORT_ZLIB
+//
+
+
+#ifndef STBI_NO_STDIO
+#include <stdio.h>
+#endif // STBI_NO_STDIO
+
+#define STBI_VERSION 1
+
+enum
+{
+ STBI_default = 0, // only used for req_comp
+
+ STBI_grey = 1,
+ STBI_grey_alpha = 2,
+ STBI_rgb = 3,
+ STBI_rgb_alpha = 4
+};
+
+typedef unsigned char stbi_uc;
+typedef unsigned short stbi_us;
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifdef STB_IMAGE_STATIC
+#define STBIDEF static
+#else
+#define STBIDEF extern
+#endif
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// PRIMARY API - works on images of any type
+//
+
+//
+// load image by filename, open file, or memory buffer
+//
+
+typedef struct
+{
+ int (*read) (void *user,char *data,int size); // fill 'data' with 'size' bytes. return number of bytes actually read
+ void (*skip) (void *user,int n); // skip the next 'n' bytes, or 'unget' the last -n bytes if negative
+ int (*eof) (void *user); // returns nonzero if we are at end of file/data
+} stbi_io_callbacks;
+
+////////////////////////////////////
+//
+// 8-bits-per-channel interface
+//
+
+STBIDEF stbi_uc *stbi_load (char const *filename, int *x, int *y, int *channels_in_file, int desired_channels);
+STBIDEF stbi_uc *stbi_load_from_memory (stbi_uc const *buffer, int len , int *x, int *y, int *channels_in_file, int desired_channels);
+STBIDEF stbi_uc *stbi_load_from_callbacks(stbi_io_callbacks const *clbk , void *user, int *x, int *y, int *channels_in_file, int desired_channels);
+
+#ifndef STBI_NO_STDIO
+STBIDEF stbi_uc *stbi_load_from_file (FILE *f, int *x, int *y, int *channels_in_file, int desired_channels);
+// for stbi_load_from_file, file pointer is left pointing immediately after image
+#endif
+
+////////////////////////////////////
+//
+// 16-bits-per-channel interface
+//
+
+STBIDEF stbi_us *stbi_load_16(char const *filename, int *x, int *y, int *channels_in_file, int desired_channels);
+#ifndef STBI_NO_STDIO
+STBIDEF stbi_us *stbi_load_from_file_16(FILE *f, int *x, int *y, int *channels_in_file, int desired_channels);
+#endif
+// @TODO the other variants
+
+////////////////////////////////////
+//
+// float-per-channel interface
+//
+#ifndef STBI_NO_LINEAR
+ STBIDEF float *stbi_loadf (char const *filename, int *x, int *y, int *channels_in_file, int desired_channels);
+ STBIDEF float *stbi_loadf_from_memory (stbi_uc const *buffer, int len, int *x, int *y, int *channels_in_file, int desired_channels);
+ STBIDEF float *stbi_loadf_from_callbacks (stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *channels_in_file, int desired_channels);
+
+ #ifndef STBI_NO_STDIO
+ STBIDEF float *stbi_loadf_from_file (FILE *f, int *x, int *y, int *channels_in_file, int desired_channels);
+ #endif
+#endif
+
+#ifndef STBI_NO_HDR
+ STBIDEF void stbi_hdr_to_ldr_gamma(float gamma);
+ STBIDEF void stbi_hdr_to_ldr_scale(float scale);
+#endif // STBI_NO_HDR
+
+#ifndef STBI_NO_LINEAR
+ STBIDEF void stbi_ldr_to_hdr_gamma(float gamma);
+ STBIDEF void stbi_ldr_to_hdr_scale(float scale);
+#endif // STBI_NO_LINEAR
+
+// stbi_is_hdr is always defined, but always returns false if STBI_NO_HDR
+STBIDEF int stbi_is_hdr_from_callbacks(stbi_io_callbacks const *clbk, void *user);
+STBIDEF int stbi_is_hdr_from_memory(stbi_uc const *buffer, int len);
+#ifndef STBI_NO_STDIO
+STBIDEF int stbi_is_hdr (char const *filename);
+STBIDEF int stbi_is_hdr_from_file(FILE *f);
+#endif // STBI_NO_STDIO
+
+
+// get a VERY brief reason for failure
+// NOT THREADSAFE
+STBIDEF const char *stbi_failure_reason (void);
+
+// free the loaded image -- this is just free()
+STBIDEF void stbi_image_free (void *retval_from_stbi_load);
+
+// get image dimensions & components without fully decoding
+STBIDEF int stbi_info_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp);
+STBIDEF int stbi_info_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp);
+
+#ifndef STBI_NO_STDIO
+STBIDEF int stbi_info (char const *filename, int *x, int *y, int *comp);
+STBIDEF int stbi_info_from_file (FILE *f, int *x, int *y, int *comp);
+
+#endif
+
+
+
+// for image formats that explicitly notate that they have premultiplied alpha,
+// we just return the colors as stored in the file. set this flag to force
+// unpremultiplication. results are undefined if the unpremultiply overflow.
+STBIDEF void stbi_set_unpremultiply_on_load(int flag_true_if_should_unpremultiply);
+
+// indicate whether we should process iphone images back to canonical format,
+// or just pass them through "as-is"
+STBIDEF void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert);
+
+// flip the image vertically, so the first pixel in the output array is the bottom left
+STBIDEF void stbi_set_flip_vertically_on_load(int flag_true_if_should_flip);
+
+// ZLIB client - used by PNG, available for other purposes
+
+STBIDEF char *stbi_zlib_decode_malloc_guesssize(const char *buffer, int len, int initial_size, int *outlen);
+STBIDEF char *stbi_zlib_decode_malloc_guesssize_headerflag(const char *buffer, int len, int initial_size, int *outlen, int parse_header);
+STBIDEF char *stbi_zlib_decode_malloc(const char *buffer, int len, int *outlen);
+STBIDEF int stbi_zlib_decode_buffer(char *obuffer, int olen, const char *ibuffer, int ilen);
+
+STBIDEF char *stbi_zlib_decode_noheader_malloc(const char *buffer, int len, int *outlen);
+STBIDEF int stbi_zlib_decode_noheader_buffer(char *obuffer, int olen, const char *ibuffer, int ilen);
+
+
+#ifdef __cplusplus
+}
+#endif
+
+//
+//
+//// end header file /////////////////////////////////////////////////////
+#endif // STBI_INCLUDE_STB_IMAGE_H
+
+#ifdef STB_IMAGE_IMPLEMENTATION
+
+#if defined(STBI_ONLY_JPEG) || defined(STBI_ONLY_PNG) || defined(STBI_ONLY_BMP) \
+ || defined(STBI_ONLY_TGA) || defined(STBI_ONLY_GIF) || defined(STBI_ONLY_PSD) \
+ || defined(STBI_ONLY_HDR) || defined(STBI_ONLY_PIC) || defined(STBI_ONLY_PNM) \
+ || defined(STBI_ONLY_ZLIB)
+ #ifndef STBI_ONLY_JPEG
+ #define STBI_NO_JPEG
+ #endif
+ #ifndef STBI_ONLY_PNG
+ #define STBI_NO_PNG
+ #endif
+ #ifndef STBI_ONLY_BMP
+ #define STBI_NO_BMP
+ #endif
+ #ifndef STBI_ONLY_PSD
+ #define STBI_NO_PSD
+ #endif
+ #ifndef STBI_ONLY_TGA
+ #define STBI_NO_TGA
+ #endif
+ #ifndef STBI_ONLY_GIF
+ #define STBI_NO_GIF
+ #endif
+ #ifndef STBI_ONLY_HDR
+ #define STBI_NO_HDR
+ #endif
+ #ifndef STBI_ONLY_PIC
+ #define STBI_NO_PIC
+ #endif
+ #ifndef STBI_ONLY_PNM
+ #define STBI_NO_PNM
+ #endif
+#endif
+
+#if defined(STBI_NO_PNG) && !defined(STBI_SUPPORT_ZLIB) && !defined(STBI_NO_ZLIB)
+#define STBI_NO_ZLIB
+#endif
+
+
+#include <stdarg.h>
+#include <stddef.h> // ptrdiff_t on osx
+#include <stdlib.h>
+#include <string.h>
+#include <limits.h>
+
+#if !defined(STBI_NO_LINEAR) || !defined(STBI_NO_HDR)
+#include <math.h> // ldexp
+#endif
+
+#ifndef STBI_NO_STDIO
+#include <stdio.h>
+#endif
+
+#ifndef STBI_ASSERT
+#include <assert.h>
+#define STBI_ASSERT(x) assert(x)
+#endif
+
+
+#ifndef _MSC_VER
+ #ifdef __cplusplus
+ #define stbi_inline inline
+ #else
+ #define stbi_inline
+ #endif
+#else
+ #define stbi_inline __forceinline
+#endif
+
+
+#ifdef _MSC_VER
+typedef unsigned short stbi__uint16;
+typedef signed short stbi__int16;
+typedef unsigned int stbi__uint32;
+typedef signed int stbi__int32;
+#else
+#include <stdint.h>
+typedef uint16_t stbi__uint16;
+typedef int16_t stbi__int16;
+typedef uint32_t stbi__uint32;
+typedef int32_t stbi__int32;
+#endif
+
+// should produce compiler error if size is wrong
+typedef unsigned char validate_uint32[sizeof(stbi__uint32)==4 ? 1 : -1];
+
+#ifdef _MSC_VER
+#define STBI_NOTUSED(v) (void)(v)
+#else
+#define STBI_NOTUSED(v) (void)sizeof(v)
+#endif
+
+#ifdef _MSC_VER
+#define STBI_HAS_LROTL
+#endif
+
+#ifdef STBI_HAS_LROTL
+ #define stbi_lrot(x,y) _lrotl(x,y)
+#else
+ #define stbi_lrot(x,y) (((x) << (y)) | ((x) >> (32 - (y))))
+#endif
+
+#if defined(STBI_MALLOC) && defined(STBI_FREE) && (defined(STBI_REALLOC) || defined(STBI_REALLOC_SIZED))
+// ok
+#elif !defined(STBI_MALLOC) && !defined(STBI_FREE) && !defined(STBI_REALLOC) && !defined(STBI_REALLOC_SIZED)
+// ok
+#else
+#error "Must define all or none of STBI_MALLOC, STBI_FREE, and STBI_REALLOC (or STBI_REALLOC_SIZED)."
+#endif
+
+#ifndef STBI_MALLOC
+#define STBI_MALLOC(sz) malloc(sz)
+#define STBI_REALLOC(p,newsz) realloc(p,newsz)
+#define STBI_FREE(p) free(p)
+#endif
+
+#ifndef STBI_REALLOC_SIZED
+#define STBI_REALLOC_SIZED(p,oldsz,newsz) STBI_REALLOC(p,newsz)
+#endif
+
+// x86/x64 detection
+#if defined(__x86_64__) || defined(_M_X64)
+#define STBI__X64_TARGET
+#elif defined(__i386) || defined(_M_IX86)
+#define STBI__X86_TARGET
+#endif
+
+#if defined(__GNUC__) && defined(STBI__X86_TARGET) && !defined(__SSE2__) && !defined(STBI_NO_SIMD)
+// gcc doesn't support sse2 intrinsics unless you compile with -msse2,
+// which in turn means it gets to use SSE2 everywhere. This is unfortunate,
+// but previous attempts to provide the SSE2 functions with runtime
+// detection caused numerous issues. The way architecture extensions are
+// exposed in GCC/Clang is, sadly, not really suited for one-file libs.
+// New behavior: if compiled with -msse2, we use SSE2 without any
+// detection; if not, we don't use it at all.
+#define STBI_NO_SIMD
+#endif
+
+#if defined(__MINGW32__) && defined(STBI__X86_TARGET) && !defined(STBI_MINGW_ENABLE_SSE2) && !defined(STBI_NO_SIMD)
+// Note that __MINGW32__ doesn't actually mean 32-bit, so we have to avoid STBI__X64_TARGET
+//
+// 32-bit MinGW wants ESP to be 16-byte aligned, but this is not in the
+// Windows ABI and VC++ as well as Windows DLLs don't maintain that invariant.
+// As a result, enabling SSE2 on 32-bit MinGW is dangerous when not
+// simultaneously enabling "-mstackrealign".
+//
+// See https://github.com/nothings/stb/issues/81 for more information.
+//
+// So default to no SSE2 on 32-bit MinGW. If you've read this far and added
+// -mstackrealign to your build settings, feel free to #define STBI_MINGW_ENABLE_SSE2.
+#define STBI_NO_SIMD
+#endif
+
+#if !defined(STBI_NO_SIMD) && (defined(STBI__X86_TARGET) || defined(STBI__X64_TARGET))
+#define STBI_SSE2
+#include <emmintrin.h>
+
+#ifdef _MSC_VER
+
+#if _MSC_VER >= 1400 // not VC6
+#include <intrin.h> // __cpuid
+static int stbi__cpuid3(void)
+{
+ int info[4];
+ __cpuid(info,1);
+ return info[3];
+}
+#else
+static int stbi__cpuid3(void)
+{
+ int res;
+ __asm {
+ mov eax,1
+ cpuid
+ mov res,edx
+ }
+ return res;
+}
+#endif
+
+#define STBI_SIMD_ALIGN(type, name) __declspec(align(16)) type name
+
+static int stbi__sse2_available()
+{
+ int info3 = stbi__cpuid3();
+ return ((info3 >> 26) & 1) != 0;
+}
+#else // assume GCC-style if not VC++
+#define STBI_SIMD_ALIGN(type, name) type name __attribute__((aligned(16)))
+
+static int stbi__sse2_available()
+{
+ // If we're even attempting to compile this on GCC/Clang, that means
+ // -msse2 is on, which means the compiler is allowed to use SSE2
+ // instructions at will, and so are we.
+ return 1;
+}
+#endif
+#endif
+
+// ARM NEON
+#if defined(STBI_NO_SIMD) && defined(STBI_NEON)
+#undef STBI_NEON
+#endif
+
+#ifdef STBI_NEON
+#include <arm_neon.h>
+// assume GCC or Clang on ARM targets
+#define STBI_SIMD_ALIGN(type, name) type name __attribute__((aligned(16)))
+#endif
+
+#ifndef STBI_SIMD_ALIGN
+#define STBI_SIMD_ALIGN(type, name) type name
+#endif
+
+///////////////////////////////////////////////
+//
+// stbi__context struct and start_xxx functions
+
+// stbi__context structure is our basic context used by all images, so it
+// contains all the IO context, plus some basic image information
+typedef struct
+{
+ stbi__uint32 img_x, img_y;
+ int img_n, img_out_n;
+
+ stbi_io_callbacks io;
+ void *io_user_data;
+
+ int read_from_callbacks;
+ int buflen;
+ stbi_uc buffer_start[128];
+
+ stbi_uc *img_buffer, *img_buffer_end;
+ stbi_uc *img_buffer_original, *img_buffer_original_end;
+} stbi__context;
+
+
+static void stbi__refill_buffer(stbi__context *s);
+
+// initialize a memory-decode context
+static void stbi__start_mem(stbi__context *s, stbi_uc const *buffer, int len)
+{
+ s->io.read = NULL;
+ s->read_from_callbacks = 0;
+ s->img_buffer = s->img_buffer_original = (stbi_uc *) buffer;
+ s->img_buffer_end = s->img_buffer_original_end = (stbi_uc *) buffer+len;
+}
+
+// initialize a callback-based context
+static void stbi__start_callbacks(stbi__context *s, stbi_io_callbacks *c, void *user)
+{
+ s->io = *c;
+ s->io_user_data = user;
+ s->buflen = sizeof(s->buffer_start);
+ s->read_from_callbacks = 1;
+ s->img_buffer_original = s->buffer_start;
+ stbi__refill_buffer(s);
+ s->img_buffer_original_end = s->img_buffer_end;
+}
+
+#ifndef STBI_NO_STDIO
+
+static int stbi__stdio_read(void *user, char *data, int size)
+{
+ return (int) fread(data,1,size,(FILE*) user);
+}
+
+static void stbi__stdio_skip(void *user, int n)
+{
+ fseek((FILE*) user, n, SEEK_CUR);
+}
+
+static int stbi__stdio_eof(void *user)
+{
+ return feof((FILE*) user);
+}
+
+static stbi_io_callbacks stbi__stdio_callbacks =
+{
+ stbi__stdio_read,
+ stbi__stdio_skip,
+ stbi__stdio_eof,
+};
+
+static void stbi__start_file(stbi__context *s, FILE *f)
+{
+ stbi__start_callbacks(s, &stbi__stdio_callbacks, (void *) f);
+}
+
+//static void stop_file(stbi__context *s) { }
+
+#endif // !STBI_NO_STDIO
+
+static void stbi__rewind(stbi__context *s)
+{
+ // conceptually rewind SHOULD rewind to the beginning of the stream,
+ // but we just rewind to the beginning of the initial buffer, because
+ // we only use it after doing 'test', which only ever looks at at most 92 bytes
+ s->img_buffer = s->img_buffer_original;
+ s->img_buffer_end = s->img_buffer_original_end;
+}
+
+enum
+{
+ STBI_ORDER_RGB,
+ STBI_ORDER_BGR
+};
+
+typedef struct
+{
+ int bits_per_channel;
+ int num_channels;
+ int channel_order;
+} stbi__result_info;
+
+#ifndef STBI_NO_JPEG
+static int stbi__jpeg_test(stbi__context *s);
+static void *stbi__jpeg_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__jpeg_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_PNG
+static int stbi__png_test(stbi__context *s);
+static void *stbi__png_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__png_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_BMP
+static int stbi__bmp_test(stbi__context *s);
+static void *stbi__bmp_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__bmp_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_TGA
+static int stbi__tga_test(stbi__context *s);
+static void *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__tga_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_PSD
+static int stbi__psd_test(stbi__context *s);
+static void *stbi__psd_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri, int bpc);
+static int stbi__psd_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_HDR
+static int stbi__hdr_test(stbi__context *s);
+static float *stbi__hdr_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__hdr_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_PIC
+static int stbi__pic_test(stbi__context *s);
+static void *stbi__pic_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__pic_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_GIF
+static int stbi__gif_test(stbi__context *s);
+static void *stbi__gif_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__gif_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_PNM
+static int stbi__pnm_test(stbi__context *s);
+static void *stbi__pnm_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__pnm_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+// this is not threadsafe
+static const char *stbi__g_failure_reason;
+
+STBIDEF const char *stbi_failure_reason(void)
+{
+ return stbi__g_failure_reason;
+}
+
+static int stbi__err(const char *str)
+{
+ stbi__g_failure_reason = str;
+ return 0;
+}
+
+static void *stbi__malloc(size_t size)
+{
+ return STBI_MALLOC(size);
+}
+
+// stb_image uses ints pervasively, including for offset calculations.
+// therefore the largest decoded image size we can support with the
+// current code, even on 64-bit targets, is INT_MAX. this is not a
+// significant limitation for the intended use case.
+//
+// we do, however, need to make sure our size calculations don't
+// overflow. hence a few helper functions for size calculations that
+// multiply integers together, making sure that they're non-negative
+// and no overflow occurs.
+
+// return 1 if the sum is valid, 0 on overflow.
+// negative terms are considered invalid.
+static int stbi__addsizes_valid(int a, int b)
+{
+ if (b < 0) return 0;
+ // now 0 <= b <= INT_MAX, hence also
+ // 0 <= INT_MAX - b <= INTMAX.
+ // And "a + b <= INT_MAX" (which might overflow) is the
+ // same as a <= INT_MAX - b (no overflow)
+ return a <= INT_MAX - b;
+}
+
+// returns 1 if the product is valid, 0 on overflow.
+// negative factors are considered invalid.
+static int stbi__mul2sizes_valid(int a, int b)
+{
+ if (a < 0 || b < 0) return 0;
+ if (b == 0) return 1; // mul-by-0 is always safe
+ // portable way to check for no overflows in a*b
+ return a <= INT_MAX/b;
+}
+
+// returns 1 if "a*b + add" has no negative terms/factors and doesn't overflow
+static int stbi__mad2sizes_valid(int a, int b, int add)
+{
+ return stbi__mul2sizes_valid(a, b) && stbi__addsizes_valid(a*b, add);
+}
+
+// returns 1 if "a*b*c + add" has no negative terms/factors and doesn't overflow
+static int stbi__mad3sizes_valid(int a, int b, int c, int add)
+{
+ return stbi__mul2sizes_valid(a, b) && stbi__mul2sizes_valid(a*b, c) &&
+ stbi__addsizes_valid(a*b*c, add);
+}
+
+// returns 1 if "a*b*c*d + add" has no negative terms/factors and doesn't overflow
+static int stbi__mad4sizes_valid(int a, int b, int c, int d, int add)
+{
+ return stbi__mul2sizes_valid(a, b) && stbi__mul2sizes_valid(a*b, c) &&
+ stbi__mul2sizes_valid(a*b*c, d) && stbi__addsizes_valid(a*b*c*d, add);
+}
+
+// mallocs with size overflow checking
+static void *stbi__malloc_mad2(int a, int b, int add)
+{
+ if (!stbi__mad2sizes_valid(a, b, add)) return NULL;
+ return stbi__malloc(a*b + add);
+}
+
+static void *stbi__malloc_mad3(int a, int b, int c, int add)
+{
+ if (!stbi__mad3sizes_valid(a, b, c, add)) return NULL;
+ return stbi__malloc(a*b*c + add);
+}
+
+static void *stbi__malloc_mad4(int a, int b, int c, int d, int add)
+{
+ if (!stbi__mad4sizes_valid(a, b, c, d, add)) return NULL;
+ return stbi__malloc(a*b*c*d + add);
+}
+
+// stbi__err - error
+// stbi__errpf - error returning pointer to float
+// stbi__errpuc - error returning pointer to unsigned char
+
+#ifdef STBI_NO_FAILURE_STRINGS
+ #define stbi__err(x,y) 0
+#elif defined(STBI_FAILURE_USERMSG)
+ #define stbi__err(x,y) stbi__err(y)
+#else
+ #define stbi__err(x,y) stbi__err(x)
+#endif
+
+#define stbi__errpf(x,y) ((float *)(size_t) (stbi__err(x,y)?NULL:NULL))
+#define stbi__errpuc(x,y) ((unsigned char *)(size_t) (stbi__err(x,y)?NULL:NULL))
+
+STBIDEF void stbi_image_free(void *retval_from_stbi_load)
+{
+ STBI_FREE(retval_from_stbi_load);
+}
+
+#ifndef STBI_NO_LINEAR
+static float *stbi__ldr_to_hdr(stbi_uc *data, int x, int y, int comp);
+#endif
+
+#ifndef STBI_NO_HDR
+static stbi_uc *stbi__hdr_to_ldr(float *data, int x, int y, int comp);
+#endif
+
+static int stbi__vertically_flip_on_load = 0;
+
+STBIDEF void stbi_set_flip_vertically_on_load(int flag_true_if_should_flip)
+{
+ stbi__vertically_flip_on_load = flag_true_if_should_flip;
+}
+
+static void *stbi__load_main(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri, int bpc)
+{
+ memset(ri, 0, sizeof(*ri)); // make sure it's initialized if we add new fields
+ ri->bits_per_channel = 8; // default is 8 so most paths don't have to be changed
+ ri->channel_order = STBI_ORDER_RGB; // all current input & output are this, but this is here so we can add BGR order
+ ri->num_channels = 0;
+
+ #ifndef STBI_NO_JPEG
+ if (stbi__jpeg_test(s)) return stbi__jpeg_load(s,x,y,comp,req_comp, ri);
+ #endif
+ #ifndef STBI_NO_PNG
+ if (stbi__png_test(s)) return stbi__png_load(s,x,y,comp,req_comp, ri);
+ #endif
+ #ifndef STBI_NO_BMP
+ if (stbi__bmp_test(s)) return stbi__bmp_load(s,x,y,comp,req_comp, ri);
+ #endif
+ #ifndef STBI_NO_GIF
+ if (stbi__gif_test(s)) return stbi__gif_load(s,x,y,comp,req_comp, ri);
+ #endif
+ #ifndef STBI_NO_PSD
+ if (stbi__psd_test(s)) return stbi__psd_load(s,x,y,comp,req_comp, ri, bpc);
+ #endif
+ #ifndef STBI_NO_PIC
+ if (stbi__pic_test(s)) return stbi__pic_load(s,x,y,comp,req_comp, ri);
+ #endif
+ #ifndef STBI_NO_PNM
+ if (stbi__pnm_test(s)) return stbi__pnm_load(s,x,y,comp,req_comp, ri);
+ #endif
+
+ #ifndef STBI_NO_HDR
+ if (stbi__hdr_test(s)) {
+ float *hdr = stbi__hdr_load(s, x,y,comp,req_comp, ri);
+ return stbi__hdr_to_ldr(hdr, *x, *y, req_comp ? req_comp : *comp);
+ }
+ #endif
+
+ #ifndef STBI_NO_TGA
+ // test tga last because it's a crappy test!
+ if (stbi__tga_test(s))
+ return stbi__tga_load(s,x,y,comp,req_comp, ri);
+ #endif
+
+ return stbi__errpuc("unknown image type", "Image not of any known type, or corrupt");
+}
+
+static stbi_uc *stbi__convert_16_to_8(stbi__uint16 *orig, int w, int h, int channels)
+{
+ int i;
+ int img_len = w * h * channels;
+ stbi_uc *reduced;
+
+ reduced = (stbi_uc *) stbi__malloc(img_len);
+ if (reduced == NULL) return stbi__errpuc("outofmem", "Out of memory");
+
+ for (i = 0; i < img_len; ++i)
+ reduced[i] = (stbi_uc)((orig[i] >> 8) & 0xFF); // top half of each byte is sufficient approx of 16->8 bit scaling
+
+ STBI_FREE(orig);
+ return reduced;
+}
+
+static stbi__uint16 *stbi__convert_8_to_16(stbi_uc *orig, int w, int h, int channels)
+{
+ int i;
+ int img_len = w * h * channels;
+ stbi__uint16 *enlarged;
+
+ enlarged = (stbi__uint16 *) stbi__malloc(img_len*2);
+ if (enlarged == NULL) return (stbi__uint16 *) stbi__errpuc("outofmem", "Out of memory");
+
+ for (i = 0; i < img_len; ++i)
+ enlarged[i] = (stbi__uint16)((orig[i] << 8) + orig[i]); // replicate to high and low byte, maps 0->0, 255->0xffff
+
+ STBI_FREE(orig);
+ return enlarged;
+}
+
+static unsigned char *stbi__load_and_postprocess_8bit(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__result_info ri;
+ void *result = stbi__load_main(s, x, y, comp, req_comp, &ri, 8);
+
+ if (result == NULL)
+ return NULL;
+
+ if (ri.bits_per_channel != 8) {
+ STBI_ASSERT(ri.bits_per_channel == 16);
+ result = stbi__convert_16_to_8((stbi__uint16 *) result, *x, *y, req_comp == 0 ? *comp : req_comp);
+ ri.bits_per_channel = 8;
+ }
+
+ // @TODO: move stbi__convert_format to here
+
+ if (stbi__vertically_flip_on_load) {
+ int w = *x, h = *y;
+ int channels = req_comp ? req_comp : *comp;
+ int row,col,z;
+ stbi_uc *image = (stbi_uc *) result;
+
+ // @OPTIMIZE: use a bigger temp buffer and memcpy multiple pixels at once
+ for (row = 0; row < (h>>1); row++) {
+ for (col = 0; col < w; col++) {
+ for (z = 0; z < channels; z++) {
+ stbi_uc temp = image[(row * w + col) * channels + z];
+ image[(row * w + col) * channels + z] = image[((h - row - 1) * w + col) * channels + z];
+ image[((h - row - 1) * w + col) * channels + z] = temp;
+ }
+ }
+ }
+ }
+
+ return (unsigned char *) result;
+}
+
+static stbi__uint16 *stbi__load_and_postprocess_16bit(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__result_info ri;
+ void *result = stbi__load_main(s, x, y, comp, req_comp, &ri, 16);
+
+ if (result == NULL)
+ return NULL;
+
+ if (ri.bits_per_channel != 16) {
+ STBI_ASSERT(ri.bits_per_channel == 8);
+ result = stbi__convert_8_to_16((stbi_uc *) result, *x, *y, req_comp == 0 ? *comp : req_comp);
+ ri.bits_per_channel = 16;
+ }
+
+ // @TODO: move stbi__convert_format16 to here
+ // @TODO: special case RGB-to-Y (and RGBA-to-YA) for 8-bit-to-16-bit case to keep more precision
+
+ if (stbi__vertically_flip_on_load) {
+ int w = *x, h = *y;
+ int channels = req_comp ? req_comp : *comp;
+ int row,col,z;
+ stbi__uint16 *image = (stbi__uint16 *) result;
+
+ // @OPTIMIZE: use a bigger temp buffer and memcpy multiple pixels at once
+ for (row = 0; row < (h>>1); row++) {
+ for (col = 0; col < w; col++) {
+ for (z = 0; z < channels; z++) {
+ stbi__uint16 temp = image[(row * w + col) * channels + z];
+ image[(row * w + col) * channels + z] = image[((h - row - 1) * w + col) * channels + z];
+ image[((h - row - 1) * w + col) * channels + z] = temp;
+ }
+ }
+ }
+ }
+
+ return (stbi__uint16 *) result;
+}
+
+#ifndef STBI_NO_HDR
+static void stbi__float_postprocess(float *result, int *x, int *y, int *comp, int req_comp)
+{
+ if (stbi__vertically_flip_on_load && result != NULL) {
+ int w = *x, h = *y;
+ int depth = req_comp ? req_comp : *comp;
+ int row,col,z;
+ float temp;
+
+ // @OPTIMIZE: use a bigger temp buffer and memcpy multiple pixels at once
+ for (row = 0; row < (h>>1); row++) {
+ for (col = 0; col < w; col++) {
+ for (z = 0; z < depth; z++) {
+ temp = result[(row * w + col) * depth + z];
+ result[(row * w + col) * depth + z] = result[((h - row - 1) * w + col) * depth + z];
+ result[((h - row - 1) * w + col) * depth + z] = temp;
+ }
+ }
+ }
+ }
+}
+#endif
+
+#ifndef STBI_NO_STDIO
+
+static FILE *stbi__fopen(char const *filename, char const *mode)
+{
+ FILE *f;
+#if defined(_MSC_VER) && _MSC_VER >= 1400
+ if (0 != fopen_s(&f, filename, mode))
+ f=0;
+#else
+ f = fopen(filename, mode);
+#endif
+ return f;
+}
+
+
+STBIDEF stbi_uc *stbi_load(char const *filename, int *x, int *y, int *comp, int req_comp)
+{
+ FILE *f = stbi__fopen(filename, "rb");
+ unsigned char *result;
+ if (!f) return stbi__errpuc("can't fopen", "Unable to open file");
+ result = stbi_load_from_file(f,x,y,comp,req_comp);
+ fclose(f);
+ return result;
+}
+
+STBIDEF stbi_uc *stbi_load_from_file(FILE *f, int *x, int *y, int *comp, int req_comp)
+{
+ unsigned char *result;
+ stbi__context s;
+ stbi__start_file(&s,f);
+ result = stbi__load_and_postprocess_8bit(&s,x,y,comp,req_comp);
+ if (result) {
+ // need to 'unget' all the characters in the IO buffer
+ fseek(f, - (int) (s.img_buffer_end - s.img_buffer), SEEK_CUR);
+ }
+ return result;
+}
+
+STBIDEF stbi__uint16 *stbi_load_from_file_16(FILE *f, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__uint16 *result;
+ stbi__context s;
+ stbi__start_file(&s,f);
+ result = stbi__load_and_postprocess_16bit(&s,x,y,comp,req_comp);
+ if (result) {
+ // need to 'unget' all the characters in the IO buffer
+ fseek(f, - (int) (s.img_buffer_end - s.img_buffer), SEEK_CUR);
+ }
+ return result;
+}
+
+STBIDEF stbi_us *stbi_load_16(char const *filename, int *x, int *y, int *comp, int req_comp)
+{
+ FILE *f = stbi__fopen(filename, "rb");
+ stbi__uint16 *result;
+ if (!f) return (stbi_us *) stbi__errpuc("can't fopen", "Unable to open file");
+ result = stbi_load_from_file_16(f,x,y,comp,req_comp);
+ fclose(f);
+ return result;
+}
+
+
+#endif //!STBI_NO_STDIO
+
+STBIDEF stbi_uc *stbi_load_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__context s;
+ stbi__start_mem(&s,buffer,len);
+ return stbi__load_and_postprocess_8bit(&s,x,y,comp,req_comp);
+}
+
+STBIDEF stbi_uc *stbi_load_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__context s;
+ stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user);
+ return stbi__load_and_postprocess_8bit(&s,x,y,comp,req_comp);
+}
+
+#ifndef STBI_NO_LINEAR
+static float *stbi__loadf_main(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+{
+ unsigned char *data;
+ #ifndef STBI_NO_HDR
+ if (stbi__hdr_test(s)) {
+ stbi__result_info ri;
+ float *hdr_data = stbi__hdr_load(s,x,y,comp,req_comp, &ri);
+ if (hdr_data)
+ stbi__float_postprocess(hdr_data,x,y,comp,req_comp);
+ return hdr_data;
+ }
+ #endif
+ data = stbi__load_and_postprocess_8bit(s, x, y, comp, req_comp);
+ if (data)
+ return stbi__ldr_to_hdr(data, *x, *y, req_comp ? req_comp : *comp);
+ return stbi__errpf("unknown image type", "Image not of any known type, or corrupt");
+}
+
+STBIDEF float *stbi_loadf_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__context s;
+ stbi__start_mem(&s,buffer,len);
+ return stbi__loadf_main(&s,x,y,comp,req_comp);
+}
+
+STBIDEF float *stbi_loadf_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__context s;
+ stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user);
+ return stbi__loadf_main(&s,x,y,comp,req_comp);
+}
+
+#ifndef STBI_NO_STDIO
+STBIDEF float *stbi_loadf(char const *filename, int *x, int *y, int *comp, int req_comp)
+{
+ float *result;
+ FILE *f = stbi__fopen(filename, "rb");
+ if (!f) return stbi__errpf("can't fopen", "Unable to open file");
+ result = stbi_loadf_from_file(f,x,y,comp,req_comp);
+ fclose(f);
+ return result;
+}
+
+STBIDEF float *stbi_loadf_from_file(FILE *f, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__context s;
+ stbi__start_file(&s,f);
+ return stbi__loadf_main(&s,x,y,comp,req_comp);
+}
+#endif // !STBI_NO_STDIO
+
+#endif // !STBI_NO_LINEAR
+
+// these is-hdr-or-not is defined independent of whether STBI_NO_LINEAR is
+// defined, for API simplicity; if STBI_NO_LINEAR is defined, it always
+// reports false!
+
+STBIDEF int stbi_is_hdr_from_memory(stbi_uc const *buffer, int len)
+{
+ #ifndef STBI_NO_HDR
+ stbi__context s;
+ stbi__start_mem(&s,buffer,len);
+ return stbi__hdr_test(&s);
+ #else
+ STBI_NOTUSED(buffer);
+ STBI_NOTUSED(len);
+ return 0;
+ #endif
+}
+
+#ifndef STBI_NO_STDIO
+STBIDEF int stbi_is_hdr (char const *filename)
+{
+ FILE *f = stbi__fopen(filename, "rb");
+ int result=0;
+ if (f) {
+ result = stbi_is_hdr_from_file(f);
+ fclose(f);
+ }
+ return result;
+}
+
+STBIDEF int stbi_is_hdr_from_file(FILE *f)
+{
+ #ifndef STBI_NO_HDR
+ stbi__context s;
+ stbi__start_file(&s,f);
+ return stbi__hdr_test(&s);
+ #else
+ STBI_NOTUSED(f);
+ return 0;
+ #endif
+}
+#endif // !STBI_NO_STDIO
+
+STBIDEF int stbi_is_hdr_from_callbacks(stbi_io_callbacks const *clbk, void *user)
+{
+ #ifndef STBI_NO_HDR
+ stbi__context s;
+ stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user);
+ return stbi__hdr_test(&s);
+ #else
+ STBI_NOTUSED(clbk);
+ STBI_NOTUSED(user);
+ return 0;
+ #endif
+}
+
+#ifndef STBI_NO_LINEAR
+static float stbi__l2h_gamma=2.2f, stbi__l2h_scale=1.0f;
+
+STBIDEF void stbi_ldr_to_hdr_gamma(float gamma) { stbi__l2h_gamma = gamma; }
+STBIDEF void stbi_ldr_to_hdr_scale(float scale) { stbi__l2h_scale = scale; }
+#endif
+
+static float stbi__h2l_gamma_i=1.0f/2.2f, stbi__h2l_scale_i=1.0f;
+
+STBIDEF void stbi_hdr_to_ldr_gamma(float gamma) { stbi__h2l_gamma_i = 1/gamma; }
+STBIDEF void stbi_hdr_to_ldr_scale(float scale) { stbi__h2l_scale_i = 1/scale; }
+
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// Common code used by all image loaders
+//
+
+enum
+{
+ STBI__SCAN_load=0,
+ STBI__SCAN_type,
+ STBI__SCAN_header
+};
+
+static void stbi__refill_buffer(stbi__context *s)
+{
+ int n = (s->io.read)(s->io_user_data,(char*)s->buffer_start,s->buflen);
+ if (n == 0) {
+ // at end of file, treat same as if from memory, but need to handle case
+ // where s->img_buffer isn't pointing to safe memory, e.g. 0-byte file
+ s->read_from_callbacks = 0;
+ s->img_buffer = s->buffer_start;
+ s->img_buffer_end = s->buffer_start+1;
+ *s->img_buffer = 0;
+ } else {
+ s->img_buffer = s->buffer_start;
+ s->img_buffer_end = s->buffer_start + n;
+ }
+}
+
+stbi_inline static stbi_uc stbi__get8(stbi__context *s)
+{
+ if (s->img_buffer < s->img_buffer_end)
+ return *s->img_buffer++;
+ if (s->read_from_callbacks) {
+ stbi__refill_buffer(s);
+ return *s->img_buffer++;
+ }
+ return 0;
+}
+
+stbi_inline static int stbi__at_eof(stbi__context *s)
+{
+ if (s->io.read) {
+ if (!(s->io.eof)(s->io_user_data)) return 0;
+ // if feof() is true, check if buffer = end
+ // special case: we've only got the special 0 character at the end
+ if (s->read_from_callbacks == 0) return 1;
+ }
+
+ return s->img_buffer >= s->img_buffer_end;
+}
+
+static void stbi__skip(stbi__context *s, int n)
+{
+ if (n < 0) {
+ s->img_buffer = s->img_buffer_end;
+ return;
+ }
+ if (s->io.read) {
+ int blen = (int) (s->img_buffer_end - s->img_buffer);
+ if (blen < n) {
+ s->img_buffer = s->img_buffer_end;
+ (s->io.skip)(s->io_user_data, n - blen);
+ return;
+ }
+ }
+ s->img_buffer += n;
+}
+
+static int stbi__getn(stbi__context *s, stbi_uc *buffer, int n)
+{
+ if (s->io.read) {
+ int blen = (int) (s->img_buffer_end - s->img_buffer);
+ if (blen < n) {
+ int res, count;
+
+ memcpy(buffer, s->img_buffer, blen);
+
+ count = (s->io.read)(s->io_user_data, (char*) buffer + blen, n - blen);
+ res = (count == (n-blen));
+ s->img_buffer = s->img_buffer_end;
+ return res;
+ }
+ }
+
+ if (s->img_buffer+n <= s->img_buffer_end) {
+ memcpy(buffer, s->img_buffer, n);
+ s->img_buffer += n;
+ return 1;
+ } else
+ return 0;
+}
+
+static int stbi__get16be(stbi__context *s)
+{
+ int z = stbi__get8(s);
+ return (z << 8) + stbi__get8(s);
+}
+
+static stbi__uint32 stbi__get32be(stbi__context *s)
+{
+ stbi__uint32 z = stbi__get16be(s);
+ return (z << 16) + stbi__get16be(s);
+}
+
+#if defined(STBI_NO_BMP) && defined(STBI_NO_TGA) && defined(STBI_NO_GIF)
+// nothing
+#else
+static int stbi__get16le(stbi__context *s)
+{
+ int z = stbi__get8(s);
+ return z + (stbi__get8(s) << 8);
+}
+#endif
+
+#ifndef STBI_NO_BMP
+static stbi__uint32 stbi__get32le(stbi__context *s)
+{
+ stbi__uint32 z = stbi__get16le(s);
+ return z + (stbi__get16le(s) << 16);
+}
+#endif
+
+#define STBI__BYTECAST(x) ((stbi_uc) ((x) & 255)) // truncate int to byte without warnings
+
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// generic converter from built-in img_n to req_comp
+// individual types do this automatically as much as possible (e.g. jpeg
+// does all cases internally since it needs to colorspace convert anyway,
+// and it never has alpha, so very few cases ). png can automatically
+// interleave an alpha=255 channel, but falls back to this for other cases
+//
+// assume data buffer is malloced, so malloc a new one and free that one
+// only failure mode is malloc failing
+
+static stbi_uc stbi__compute_y(int r, int g, int b)
+{
+ return (stbi_uc) (((r*77) + (g*150) + (29*b)) >> 8);
+}
+
+static unsigned char *stbi__convert_format(unsigned char *data, int img_n, int req_comp, unsigned int x, unsigned int y)
+{
+ int i,j;
+ unsigned char *good;
+
+ if (req_comp == img_n) return data;
+ STBI_ASSERT(req_comp >= 1 && req_comp <= 4);
+
+ good = (unsigned char *) stbi__malloc_mad3(req_comp, x, y, 0);
+ if (good == NULL) {
+ STBI_FREE(data);
+ return stbi__errpuc("outofmem", "Out of memory");
+ }
+
+ for (j=0; j < (int) y; ++j) {
+ unsigned char *src = data + j * x * img_n ;
+ unsigned char *dest = good + j * x * req_comp;
+
+ #define STBI__COMBO(a,b) ((a)*8+(b))
+ #define STBI__CASE(a,b) case STBI__COMBO(a,b): for(i=x-1; i >= 0; --i, src += a, dest += b)
+ // convert source image with img_n components to one with req_comp components;
+ // avoid switch per pixel, so use switch per scanline and massive macros
+ switch (STBI__COMBO(img_n, req_comp)) {
+ STBI__CASE(1,2) { dest[0]=src[0], dest[1]=255; } break;
+ STBI__CASE(1,3) { dest[0]=dest[1]=dest[2]=src[0]; } break;
+ STBI__CASE(1,4) { dest[0]=dest[1]=dest[2]=src[0], dest[3]=255; } break;
+ STBI__CASE(2,1) { dest[0]=src[0]; } break;
+ STBI__CASE(2,3) { dest[0]=dest[1]=dest[2]=src[0]; } break;
+ STBI__CASE(2,4) { dest[0]=dest[1]=dest[2]=src[0], dest[3]=src[1]; } break;
+ STBI__CASE(3,4) { dest[0]=src[0],dest[1]=src[1],dest[2]=src[2],dest[3]=255; } break;
+ STBI__CASE(3,1) { dest[0]=stbi__compute_y(src[0],src[1],src[2]); } break;
+ STBI__CASE(3,2) { dest[0]=stbi__compute_y(src[0],src[1],src[2]), dest[1] = 255; } break;
+ STBI__CASE(4,1) { dest[0]=stbi__compute_y(src[0],src[1],src[2]); } break;
+ STBI__CASE(4,2) { dest[0]=stbi__compute_y(src[0],src[1],src[2]), dest[1] = src[3]; } break;
+ STBI__CASE(4,3) { dest[0]=src[0],dest[1]=src[1],dest[2]=src[2]; } break;
+ default: STBI_ASSERT(0);
+ }
+ #undef STBI__CASE
+ }
+
+ STBI_FREE(data);
+ return good;
+}
+
+static stbi__uint16 stbi__compute_y_16(int r, int g, int b)
+{
+ return (stbi__uint16) (((r*77) + (g*150) + (29*b)) >> 8);
+}
+
+static stbi__uint16 *stbi__convert_format16(stbi__uint16 *data, int img_n, int req_comp, unsigned int x, unsigned int y)
+{
+ int i,j;
+ stbi__uint16 *good;
+
+ if (req_comp == img_n) return data;
+ STBI_ASSERT(req_comp >= 1 && req_comp <= 4);
+
+ good = (stbi__uint16 *) stbi__malloc(req_comp * x * y * 2);
+ if (good == NULL) {
+ STBI_FREE(data);
+ return (stbi__uint16 *) stbi__errpuc("outofmem", "Out of memory");
+ }
+
+ for (j=0; j < (int) y; ++j) {
+ stbi__uint16 *src = data + j * x * img_n ;
+ stbi__uint16 *dest = good + j * x * req_comp;
+
+ #define STBI__COMBO(a,b) ((a)*8+(b))
+ #define STBI__CASE(a,b) case STBI__COMBO(a,b): for(i=x-1; i >= 0; --i, src += a, dest += b)
+ // convert source image with img_n components to one with req_comp components;
+ // avoid switch per pixel, so use switch per scanline and massive macros
+ switch (STBI__COMBO(img_n, req_comp)) {
+ STBI__CASE(1,2) { dest[0]=src[0], dest[1]=0xffff; } break;
+ STBI__CASE(1,3) { dest[0]=dest[1]=dest[2]=src[0]; } break;
+ STBI__CASE(1,4) { dest[0]=dest[1]=dest[2]=src[0], dest[3]=0xffff; } break;
+ STBI__CASE(2,1) { dest[0]=src[0]; } break;
+ STBI__CASE(2,3) { dest[0]=dest[1]=dest[2]=src[0]; } break;
+ STBI__CASE(2,4) { dest[0]=dest[1]=dest[2]=src[0], dest[3]=src[1]; } break;
+ STBI__CASE(3,4) { dest[0]=src[0],dest[1]=src[1],dest[2]=src[2],dest[3]=0xffff; } break;
+ STBI__CASE(3,1) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]); } break;
+ STBI__CASE(3,2) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]), dest[1] = 0xffff; } break;
+ STBI__CASE(4,1) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]); } break;
+ STBI__CASE(4,2) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]), dest[1] = src[3]; } break;
+ STBI__CASE(4,3) { dest[0]=src[0],dest[1]=src[1],dest[2]=src[2]; } break;
+ default: STBI_ASSERT(0);
+ }
+ #undef STBI__CASE
+ }
+
+ STBI_FREE(data);
+ return good;
+}
+
+#ifndef STBI_NO_LINEAR
+static float *stbi__ldr_to_hdr(stbi_uc *data, int x, int y, int comp)
+{
+ int i,k,n;
+ float *output;
+ if (!data) return NULL;
+ output = (float *) stbi__malloc_mad4(x, y, comp, sizeof(float), 0);
+ if (output == NULL) { STBI_FREE(data); return stbi__errpf("outofmem", "Out of memory"); }
+ // compute number of non-alpha components
+ if (comp & 1) n = comp; else n = comp-1;
+ for (i=0; i < x*y; ++i) {
+ for (k=0; k < n; ++k) {
+ output[i*comp + k] = (float) (pow(data[i*comp+k]/255.0f, stbi__l2h_gamma) * stbi__l2h_scale);
+ }
+ if (k < comp) output[i*comp + k] = data[i*comp+k]/255.0f;
+ }
+ STBI_FREE(data);
+ return output;
+}
+#endif
+
+#ifndef STBI_NO_HDR
+#define stbi__float2int(x) ((int) (x))
+static stbi_uc *stbi__hdr_to_ldr(float *data, int x, int y, int comp)
+{
+ int i,k,n;
+ stbi_uc *output;
+ if (!data) return NULL;
+ output = (stbi_uc *) stbi__malloc_mad3(x, y, comp, 0);
+ if (output == NULL) { STBI_FREE(data); return stbi__errpuc("outofmem", "Out of memory"); }
+ // compute number of non-alpha components
+ if (comp & 1) n = comp; else n = comp-1;
+ for (i=0; i < x*y; ++i) {
+ for (k=0; k < n; ++k) {
+ float z = (float) pow(data[i*comp+k]*stbi__h2l_scale_i, stbi__h2l_gamma_i) * 255 + 0.5f;
+ if (z < 0) z = 0;
+ if (z > 255) z = 255;
+ output[i*comp + k] = (stbi_uc) stbi__float2int(z);
+ }
+ if (k < comp) {
+ float z = data[i*comp+k] * 255 + 0.5f;
+ if (z < 0) z = 0;
+ if (z > 255) z = 255;
+ output[i*comp + k] = (stbi_uc) stbi__float2int(z);
+ }
+ }
+ STBI_FREE(data);
+ return output;
+}
+#endif
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// "baseline" JPEG/JFIF decoder
+//
+// simple implementation
+// - doesn't support delayed output of y-dimension
+// - simple interface (only one output format: 8-bit interleaved RGB)
+// - doesn't try to recover corrupt jpegs
+// - doesn't allow partial loading, loading multiple at once
+// - still fast on x86 (copying globals into locals doesn't help x86)
+// - allocates lots of intermediate memory (full size of all components)
+// - non-interleaved case requires this anyway
+// - allows good upsampling (see next)
+// high-quality
+// - upsampled channels are bilinearly interpolated, even across blocks
+// - quality integer IDCT derived from IJG's 'slow'
+// performance
+// - fast huffman; reasonable integer IDCT
+// - some SIMD kernels for common paths on targets with SSE2/NEON
+// - uses a lot of intermediate memory, could cache poorly
+
+#ifndef STBI_NO_JPEG
+
+// huffman decoding acceleration
+#define FAST_BITS 9 // larger handles more cases; smaller stomps less cache
+
+typedef struct
+{
+ stbi_uc fast[1 << FAST_BITS];
+ // weirdly, repacking this into AoS is a 10% speed loss, instead of a win
+ stbi__uint16 code[256];
+ stbi_uc values[256];
+ stbi_uc size[257];
+ unsigned int maxcode[18];
+ int delta[17]; // old 'firstsymbol' - old 'firstcode'
+} stbi__huffman;
+
+typedef struct
+{
+ stbi__context *s;
+ stbi__huffman huff_dc[4];
+ stbi__huffman huff_ac[4];
+ stbi__uint16 dequant[4][64];
+ stbi__int16 fast_ac[4][1 << FAST_BITS];
+
+// sizes for components, interleaved MCUs
+ int img_h_max, img_v_max;
+ int img_mcu_x, img_mcu_y;
+ int img_mcu_w, img_mcu_h;
+
+// definition of jpeg image component
+ struct
+ {
+ int id;
+ int h,v;
+ int tq;
+ int hd,ha;
+ int dc_pred;
+
+ int x,y,w2,h2;
+ stbi_uc *data;
+ void *raw_data, *raw_coeff;
+ stbi_uc *linebuf;
+ short *coeff; // progressive only
+ int coeff_w, coeff_h; // number of 8x8 coefficient blocks
+ } img_comp[4];
+
+ stbi__uint32 code_buffer; // jpeg entropy-coded buffer
+ int code_bits; // number of valid bits
+ unsigned char marker; // marker seen while filling entropy buffer
+ int nomore; // flag if we saw a marker so must stop
+
+ int progressive;
+ int spec_start;
+ int spec_end;
+ int succ_high;
+ int succ_low;
+ int eob_run;
+ int jfif;
+ int app14_color_transform; // Adobe APP14 tag
+ int rgb;
+
+ int scan_n, order[4];
+ int restart_interval, todo;
+
+// kernels
+ void (*idct_block_kernel)(stbi_uc *out, int out_stride, short data[64]);
+ void (*YCbCr_to_RGB_kernel)(stbi_uc *out, const stbi_uc *y, const stbi_uc *pcb, const stbi_uc *pcr, int count, int step);
+ stbi_uc *(*resample_row_hv_2_kernel)(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs);
+} stbi__jpeg;
+
+static int stbi__build_huffman(stbi__huffman *h, int *count)
+{
+ int i,j,k=0,code;
+ // build size list for each symbol (from JPEG spec)
+ for (i=0; i < 16; ++i)
+ for (j=0; j < count[i]; ++j)
+ h->size[k++] = (stbi_uc) (i+1);
+ h->size[k] = 0;
+
+ // compute actual symbols (from jpeg spec)
+ code = 0;
+ k = 0;
+ for(j=1; j <= 16; ++j) {
+ // compute delta to add to code to compute symbol id
+ h->delta[j] = k - code;
+ if (h->size[k] == j) {
+ while (h->size[k] == j)
+ h->code[k++] = (stbi__uint16) (code++);
+ if (code-1 >= (1 << j)) return stbi__err("bad code lengths","Corrupt JPEG");
+ }
+ // compute largest code + 1 for this size, preshifted as needed later
+ h->maxcode[j] = code << (16-j);
+ code <<= 1;
+ }
+ h->maxcode[j] = 0xffffffff;
+
+ // build non-spec acceleration table; 255 is flag for not-accelerated
+ memset(h->fast, 255, 1 << FAST_BITS);
+ for (i=0; i < k; ++i) {
+ int s = h->size[i];
+ if (s <= FAST_BITS) {
+ int c = h->code[i] << (FAST_BITS-s);
+ int m = 1 << (FAST_BITS-s);
+ for (j=0; j < m; ++j) {
+ h->fast[c+j] = (stbi_uc) i;
+ }
+ }
+ }
+ return 1;
+}
+
+// build a table that decodes both magnitude and value of small ACs in
+// one go.
+static void stbi__build_fast_ac(stbi__int16 *fast_ac, stbi__huffman *h)
+{
+ int i;
+ for (i=0; i < (1 << FAST_BITS); ++i) {
+ stbi_uc fast = h->fast[i];
+ fast_ac[i] = 0;
+ if (fast < 255) {
+ int rs = h->values[fast];
+ int run = (rs >> 4) & 15;
+ int magbits = rs & 15;
+ int len = h->size[fast];
+
+ if (magbits && len + magbits <= FAST_BITS) {
+ // magnitude code followed by receive_extend code
+ int k = ((i << len) & ((1 << FAST_BITS) - 1)) >> (FAST_BITS - magbits);
+ int m = 1 << (magbits - 1);
+ if (k < m) k += (~0U << magbits) + 1;
+ // if the result is small enough, we can fit it in fast_ac table
+ if (k >= -128 && k <= 127)
+ fast_ac[i] = (stbi__int16) ((k << 8) + (run << 4) + (len + magbits));
+ }
+ }
+ }
+}
+
+static void stbi__grow_buffer_unsafe(stbi__jpeg *j)
+{
+ do {
+ int b = j->nomore ? 0 : stbi__get8(j->s);
+ if (b == 0xff) {
+ int c = stbi__get8(j->s);
+ while (c == 0xff) c = stbi__get8(j->s); // consume fill bytes
+ if (c != 0) {
+ j->marker = (unsigned char) c;
+ j->nomore = 1;
+ return;
+ }
+ }
+ j->code_buffer |= b << (24 - j->code_bits);
+ j->code_bits += 8;
+ } while (j->code_bits <= 24);
+}
+
+// (1 << n) - 1
+static stbi__uint32 stbi__bmask[17]={0,1,3,7,15,31,63,127,255,511,1023,2047,4095,8191,16383,32767,65535};
+
+// decode a jpeg huffman value from the bitstream
+stbi_inline static int stbi__jpeg_huff_decode(stbi__jpeg *j, stbi__huffman *h)
+{
+ unsigned int temp;
+ int c,k;
+
+ if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
+
+ // look at the top FAST_BITS and determine what symbol ID it is,
+ // if the code is <= FAST_BITS
+ c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1);
+ k = h->fast[c];
+ if (k < 255) {
+ int s = h->size[k];
+ if (s > j->code_bits)
+ return -1;
+ j->code_buffer <<= s;
+ j->code_bits -= s;
+ return h->values[k];
+ }
+
+ // naive test is to shift the code_buffer down so k bits are
+ // valid, then test against maxcode. To speed this up, we've
+ // preshifted maxcode left so that it has (16-k) 0s at the
+ // end; in other words, regardless of the number of bits, it
+ // wants to be compared against something shifted to have 16;
+ // that way we don't need to shift inside the loop.
+ temp = j->code_buffer >> 16;
+ for (k=FAST_BITS+1 ; ; ++k)
+ if (temp < h->maxcode[k])
+ break;
+ if (k == 17) {
+ // error! code not found
+ j->code_bits -= 16;
+ return -1;
+ }
+
+ if (k > j->code_bits)
+ return -1;
+
+ // convert the huffman code to the symbol id
+ c = ((j->code_buffer >> (32 - k)) & stbi__bmask[k]) + h->delta[k];
+ STBI_ASSERT((((j->code_buffer) >> (32 - h->size[c])) & stbi__bmask[h->size[c]]) == h->code[c]);
+
+ // convert the id to a symbol
+ j->code_bits -= k;
+ j->code_buffer <<= k;
+ return h->values[c];
+}
+
+// bias[n] = (-1<<n) + 1
+static int const stbi__jbias[16] = {0,-1,-3,-7,-15,-31,-63,-127,-255,-511,-1023,-2047,-4095,-8191,-16383,-32767};
+
+// combined JPEG 'receive' and JPEG 'extend', since baseline
+// always extends everything it receives.
+stbi_inline static int stbi__extend_receive(stbi__jpeg *j, int n)
+{
+ unsigned int k;
+ int sgn;
+ if (j->code_bits < n) stbi__grow_buffer_unsafe(j);
+
+ sgn = (stbi__int32)j->code_buffer >> 31; // sign bit is always in MSB
+ k = stbi_lrot(j->code_buffer, n);
+ STBI_ASSERT(n >= 0 && n < (int) (sizeof(stbi__bmask)/sizeof(*stbi__bmask)));
+ j->code_buffer = k & ~stbi__bmask[n];
+ k &= stbi__bmask[n];
+ j->code_bits -= n;
+ return k + (stbi__jbias[n] & ~sgn);
+}
+
+// get some unsigned bits
+stbi_inline static int stbi__jpeg_get_bits(stbi__jpeg *j, int n)
+{
+ unsigned int k;
+ if (j->code_bits < n) stbi__grow_buffer_unsafe(j);
+ k = stbi_lrot(j->code_buffer, n);
+ j->code_buffer = k & ~stbi__bmask[n];
+ k &= stbi__bmask[n];
+ j->code_bits -= n;
+ return k;
+}
+
+stbi_inline static int stbi__jpeg_get_bit(stbi__jpeg *j)
+{
+ unsigned int k;
+ if (j->code_bits < 1) stbi__grow_buffer_unsafe(j);
+ k = j->code_buffer;
+ j->code_buffer <<= 1;
+ --j->code_bits;
+ return k & 0x80000000;
+}
+
+// given a value that's at position X in the zigzag stream,
+// where does it appear in the 8x8 matrix coded as row-major?
+static stbi_uc stbi__jpeg_dezigzag[64+15] =
+{
+ 0, 1, 8, 16, 9, 2, 3, 10,
+ 17, 24, 32, 25, 18, 11, 4, 5,
+ 12, 19, 26, 33, 40, 48, 41, 34,
+ 27, 20, 13, 6, 7, 14, 21, 28,
+ 35, 42, 49, 56, 57, 50, 43, 36,
+ 29, 22, 15, 23, 30, 37, 44, 51,
+ 58, 59, 52, 45, 38, 31, 39, 46,
+ 53, 60, 61, 54, 47, 55, 62, 63,
+ // let corrupt input sample past end
+ 63, 63, 63, 63, 63, 63, 63, 63,
+ 63, 63, 63, 63, 63, 63, 63
+};
+
+// decode one 64-entry block--
+static int stbi__jpeg_decode_block(stbi__jpeg *j, short data[64], stbi__huffman *hdc, stbi__huffman *hac, stbi__int16 *fac, int b, stbi__uint16 *dequant)
+{
+ int diff,dc,k;
+ int t;
+
+ if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
+ t = stbi__jpeg_huff_decode(j, hdc);
+ if (t < 0) return stbi__err("bad huffman code","Corrupt JPEG");
+
+ // 0 all the ac values now so we can do it 32-bits at a time
+ memset(data,0,64*sizeof(data[0]));
+
+ diff = t ? stbi__extend_receive(j, t) : 0;
+ dc = j->img_comp[b].dc_pred + diff;
+ j->img_comp[b].dc_pred = dc;
+ data[0] = (short) (dc * dequant[0]);
+
+ // decode AC components, see JPEG spec
+ k = 1;
+ do {
+ unsigned int zig;
+ int c,r,s;
+ if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
+ c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1);
+ r = fac[c];
+ if (r) { // fast-AC path
+ k += (r >> 4) & 15; // run
+ s = r & 15; // combined length
+ j->code_buffer <<= s;
+ j->code_bits -= s;
+ // decode into unzigzag'd location
+ zig = stbi__jpeg_dezigzag[k++];
+ data[zig] = (short) ((r >> 8) * dequant[zig]);
+ } else {
+ int rs = stbi__jpeg_huff_decode(j, hac);
+ if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG");
+ s = rs & 15;
+ r = rs >> 4;
+ if (s == 0) {
+ if (rs != 0xf0) break; // end block
+ k += 16;
+ } else {
+ k += r;
+ // decode into unzigzag'd location
+ zig = stbi__jpeg_dezigzag[k++];
+ data[zig] = (short) (stbi__extend_receive(j,s) * dequant[zig]);
+ }
+ }
+ } while (k < 64);
+ return 1;
+}
+
+static int stbi__jpeg_decode_block_prog_dc(stbi__jpeg *j, short data[64], stbi__huffman *hdc, int b)
+{
+ int diff,dc;
+ int t;
+ if (j->spec_end != 0) return stbi__err("can't merge dc and ac", "Corrupt JPEG");
+
+ if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
+
+ if (j->succ_high == 0) {
+ // first scan for DC coefficient, must be first
+ memset(data,0,64*sizeof(data[0])); // 0 all the ac values now
+ t = stbi__jpeg_huff_decode(j, hdc);
+ diff = t ? stbi__extend_receive(j, t) : 0;
+
+ dc = j->img_comp[b].dc_pred + diff;
+ j->img_comp[b].dc_pred = dc;
+ data[0] = (short) (dc << j->succ_low);
+ } else {
+ // refinement scan for DC coefficient
+ if (stbi__jpeg_get_bit(j))
+ data[0] += (short) (1 << j->succ_low);
+ }
+ return 1;
+}
+
+// @OPTIMIZE: store non-zigzagged during the decode passes,
+// and only de-zigzag when dequantizing
+static int stbi__jpeg_decode_block_prog_ac(stbi__jpeg *j, short data[64], stbi__huffman *hac, stbi__int16 *fac)
+{
+ int k;
+ if (j->spec_start == 0) return stbi__err("can't merge dc and ac", "Corrupt JPEG");
+
+ if (j->succ_high == 0) {
+ int shift = j->succ_low;
+
+ if (j->eob_run) {
+ --j->eob_run;
+ return 1;
+ }
+
+ k = j->spec_start;
+ do {
+ unsigned int zig;
+ int c,r,s;
+ if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
+ c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1);
+ r = fac[c];
+ if (r) { // fast-AC path
+ k += (r >> 4) & 15; // run
+ s = r & 15; // combined length
+ j->code_buffer <<= s;
+ j->code_bits -= s;
+ zig = stbi__jpeg_dezigzag[k++];
+ data[zig] = (short) ((r >> 8) << shift);
+ } else {
+ int rs = stbi__jpeg_huff_decode(j, hac);
+ if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG");
+ s = rs & 15;
+ r = rs >> 4;
+ if (s == 0) {
+ if (r < 15) {
+ j->eob_run = (1 << r);
+ if (r)
+ j->eob_run += stbi__jpeg_get_bits(j, r);
+ --j->eob_run;
+ break;
+ }
+ k += 16;
+ } else {
+ k += r;
+ zig = stbi__jpeg_dezigzag[k++];
+ data[zig] = (short) (stbi__extend_receive(j,s) << shift);
+ }
+ }
+ } while (k <= j->spec_end);
+ } else {
+ // refinement scan for these AC coefficients
+
+ short bit = (short) (1 << j->succ_low);
+
+ if (j->eob_run) {
+ --j->eob_run;
+ for (k = j->spec_start; k <= j->spec_end; ++k) {
+ short *p = &data[stbi__jpeg_dezigzag[k]];
+ if (*p != 0)
+ if (stbi__jpeg_get_bit(j))
+ if ((*p & bit)==0) {
+ if (*p > 0)
+ *p += bit;
+ else
+ *p -= bit;
+ }
+ }
+ } else {
+ k = j->spec_start;
+ do {
+ int r,s;
+ int rs = stbi__jpeg_huff_decode(j, hac); // @OPTIMIZE see if we can use the fast path here, advance-by-r is so slow, eh
+ if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG");
+ s = rs & 15;
+ r = rs >> 4;
+ if (s == 0) {
+ if (r < 15) {
+ j->eob_run = (1 << r) - 1;
+ if (r)
+ j->eob_run += stbi__jpeg_get_bits(j, r);
+ r = 64; // force end of block
+ } else {
+ // r=15 s=0 should write 16 0s, so we just do
+ // a run of 15 0s and then write s (which is 0),
+ // so we don't have to do anything special here
+ }
+ } else {
+ if (s != 1) return stbi__err("bad huffman code", "Corrupt JPEG");
+ // sign bit
+ if (stbi__jpeg_get_bit(j))
+ s = bit;
+ else
+ s = -bit;
+ }
+
+ // advance by r
+ while (k <= j->spec_end) {
+ short *p = &data[stbi__jpeg_dezigzag[k++]];
+ if (*p != 0) {
+ if (stbi__jpeg_get_bit(j))
+ if ((*p & bit)==0) {
+ if (*p > 0)
+ *p += bit;
+ else
+ *p -= bit;
+ }
+ } else {
+ if (r == 0) {
+ *p = (short) s;
+ break;
+ }
+ --r;
+ }
+ }
+ } while (k <= j->spec_end);
+ }
+ }
+ return 1;
+}
+
+// take a -128..127 value and stbi__clamp it and convert to 0..255
+stbi_inline static stbi_uc stbi__clamp(int x)
+{
+ // trick to use a single test to catch both cases
+ if ((unsigned int) x > 255) {
+ if (x < 0) return 0;
+ if (x > 255) return 255;
+ }
+ return (stbi_uc) x;
+}
+
+#define stbi__f2f(x) ((int) (((x) * 4096 + 0.5)))
+#define stbi__fsh(x) ((x) << 12)
+
+// derived from jidctint -- DCT_ISLOW
+#define STBI__IDCT_1D(s0,s1,s2,s3,s4,s5,s6,s7) \
+ int t0,t1,t2,t3,p1,p2,p3,p4,p5,x0,x1,x2,x3; \
+ p2 = s2; \
+ p3 = s6; \
+ p1 = (p2+p3) * stbi__f2f(0.5411961f); \
+ t2 = p1 + p3*stbi__f2f(-1.847759065f); \
+ t3 = p1 + p2*stbi__f2f( 0.765366865f); \
+ p2 = s0; \
+ p3 = s4; \
+ t0 = stbi__fsh(p2+p3); \
+ t1 = stbi__fsh(p2-p3); \
+ x0 = t0+t3; \
+ x3 = t0-t3; \
+ x1 = t1+t2; \
+ x2 = t1-t2; \
+ t0 = s7; \
+ t1 = s5; \
+ t2 = s3; \
+ t3 = s1; \
+ p3 = t0+t2; \
+ p4 = t1+t3; \
+ p1 = t0+t3; \
+ p2 = t1+t2; \
+ p5 = (p3+p4)*stbi__f2f( 1.175875602f); \
+ t0 = t0*stbi__f2f( 0.298631336f); \
+ t1 = t1*stbi__f2f( 2.053119869f); \
+ t2 = t2*stbi__f2f( 3.072711026f); \
+ t3 = t3*stbi__f2f( 1.501321110f); \
+ p1 = p5 + p1*stbi__f2f(-0.899976223f); \
+ p2 = p5 + p2*stbi__f2f(-2.562915447f); \
+ p3 = p3*stbi__f2f(-1.961570560f); \
+ p4 = p4*stbi__f2f(-0.390180644f); \
+ t3 += p1+p4; \
+ t2 += p2+p3; \
+ t1 += p2+p4; \
+ t0 += p1+p3;
+
+static void stbi__idct_block(stbi_uc *out, int out_stride, short data[64])
+{
+ int i,val[64],*v=val;
+ stbi_uc *o;
+ short *d = data;
+
+ // columns
+ for (i=0; i < 8; ++i,++d, ++v) {
+ // if all zeroes, shortcut -- this avoids dequantizing 0s and IDCTing
+ if (d[ 8]==0 && d[16]==0 && d[24]==0 && d[32]==0
+ && d[40]==0 && d[48]==0 && d[56]==0) {
+ // no shortcut 0 seconds
+ // (1|2|3|4|5|6|7)==0 0 seconds
+ // all separate -0.047 seconds
+ // 1 && 2|3 && 4|5 && 6|7: -0.047 seconds
+ int dcterm = d[0] << 2;
+ v[0] = v[8] = v[16] = v[24] = v[32] = v[40] = v[48] = v[56] = dcterm;
+ } else {
+ STBI__IDCT_1D(d[ 0],d[ 8],d[16],d[24],d[32],d[40],d[48],d[56])
+ // constants scaled things up by 1<<12; let's bring them back
+ // down, but keep 2 extra bits of precision
+ x0 += 512; x1 += 512; x2 += 512; x3 += 512;
+ v[ 0] = (x0+t3) >> 10;
+ v[56] = (x0-t3) >> 10;
+ v[ 8] = (x1+t2) >> 10;
+ v[48] = (x1-t2) >> 10;
+ v[16] = (x2+t1) >> 10;
+ v[40] = (x2-t1) >> 10;
+ v[24] = (x3+t0) >> 10;
+ v[32] = (x3-t0) >> 10;
+ }
+ }
+
+ for (i=0, v=val, o=out; i < 8; ++i,v+=8,o+=out_stride) {
+ // no fast case since the first 1D IDCT spread components out
+ STBI__IDCT_1D(v[0],v[1],v[2],v[3],v[4],v[5],v[6],v[7])
+ // constants scaled things up by 1<<12, plus we had 1<<2 from first
+ // loop, plus horizontal and vertical each scale by sqrt(8) so together
+ // we've got an extra 1<<3, so 1<<17 total we need to remove.
+ // so we want to round that, which means adding 0.5 * 1<<17,
+ // aka 65536. Also, we'll end up with -128 to 127 that we want
+ // to encode as 0..255 by adding 128, so we'll add that before the shift
+ x0 += 65536 + (128<<17);
+ x1 += 65536 + (128<<17);
+ x2 += 65536 + (128<<17);
+ x3 += 65536 + (128<<17);
+ // tried computing the shifts into temps, or'ing the temps to see
+ // if any were out of range, but that was slower
+ o[0] = stbi__clamp((x0+t3) >> 17);
+ o[7] = stbi__clamp((x0-t3) >> 17);
+ o[1] = stbi__clamp((x1+t2) >> 17);
+ o[6] = stbi__clamp((x1-t2) >> 17);
+ o[2] = stbi__clamp((x2+t1) >> 17);
+ o[5] = stbi__clamp((x2-t1) >> 17);
+ o[3] = stbi__clamp((x3+t0) >> 17);
+ o[4] = stbi__clamp((x3-t0) >> 17);
+ }
+}
+
+#ifdef STBI_SSE2
+// sse2 integer IDCT. not the fastest possible implementation but it
+// produces bit-identical results to the generic C version so it's
+// fully "transparent".
+static void stbi__idct_simd(stbi_uc *out, int out_stride, short data[64])
+{
+ // This is constructed to match our regular (generic) integer IDCT exactly.
+ __m128i row0, row1, row2, row3, row4, row5, row6, row7;
+ __m128i tmp;
+
+ // dot product constant: even elems=x, odd elems=y
+ #define dct_const(x,y) _mm_setr_epi16((x),(y),(x),(y),(x),(y),(x),(y))
+
+ // out(0) = c0[even]*x + c0[odd]*y (c0, x, y 16-bit, out 32-bit)
+ // out(1) = c1[even]*x + c1[odd]*y
+ #define dct_rot(out0,out1, x,y,c0,c1) \
+ __m128i c0##lo = _mm_unpacklo_epi16((x),(y)); \
+ __m128i c0##hi = _mm_unpackhi_epi16((x),(y)); \
+ __m128i out0##_l = _mm_madd_epi16(c0##lo, c0); \
+ __m128i out0##_h = _mm_madd_epi16(c0##hi, c0); \
+ __m128i out1##_l = _mm_madd_epi16(c0##lo, c1); \
+ __m128i out1##_h = _mm_madd_epi16(c0##hi, c1)
+
+ // out = in << 12 (in 16-bit, out 32-bit)
+ #define dct_widen(out, in) \
+ __m128i out##_l = _mm_srai_epi32(_mm_unpacklo_epi16(_mm_setzero_si128(), (in)), 4); \
+ __m128i out##_h = _mm_srai_epi32(_mm_unpackhi_epi16(_mm_setzero_si128(), (in)), 4)
+
+ // wide add
+ #define dct_wadd(out, a, b) \
+ __m128i out##_l = _mm_add_epi32(a##_l, b##_l); \
+ __m128i out##_h = _mm_add_epi32(a##_h, b##_h)
+
+ // wide sub
+ #define dct_wsub(out, a, b) \
+ __m128i out##_l = _mm_sub_epi32(a##_l, b##_l); \
+ __m128i out##_h = _mm_sub_epi32(a##_h, b##_h)
+
+ // butterfly a/b, add bias, then shift by "s" and pack
+ #define dct_bfly32o(out0, out1, a,b,bias,s) \
+ { \
+ __m128i abiased_l = _mm_add_epi32(a##_l, bias); \
+ __m128i abiased_h = _mm_add_epi32(a##_h, bias); \
+ dct_wadd(sum, abiased, b); \
+ dct_wsub(dif, abiased, b); \
+ out0 = _mm_packs_epi32(_mm_srai_epi32(sum_l, s), _mm_srai_epi32(sum_h, s)); \
+ out1 = _mm_packs_epi32(_mm_srai_epi32(dif_l, s), _mm_srai_epi32(dif_h, s)); \
+ }
+
+ // 8-bit interleave step (for transposes)
+ #define dct_interleave8(a, b) \
+ tmp = a; \
+ a = _mm_unpacklo_epi8(a, b); \
+ b = _mm_unpackhi_epi8(tmp, b)
+
+ // 16-bit interleave step (for transposes)
+ #define dct_interleave16(a, b) \
+ tmp = a; \
+ a = _mm_unpacklo_epi16(a, b); \
+ b = _mm_unpackhi_epi16(tmp, b)
+
+ #define dct_pass(bias,shift) \
+ { \
+ /* even part */ \
+ dct_rot(t2e,t3e, row2,row6, rot0_0,rot0_1); \
+ __m128i sum04 = _mm_add_epi16(row0, row4); \
+ __m128i dif04 = _mm_sub_epi16(row0, row4); \
+ dct_widen(t0e, sum04); \
+ dct_widen(t1e, dif04); \
+ dct_wadd(x0, t0e, t3e); \
+ dct_wsub(x3, t0e, t3e); \
+ dct_wadd(x1, t1e, t2e); \
+ dct_wsub(x2, t1e, t2e); \
+ /* odd part */ \
+ dct_rot(y0o,y2o, row7,row3, rot2_0,rot2_1); \
+ dct_rot(y1o,y3o, row5,row1, rot3_0,rot3_1); \
+ __m128i sum17 = _mm_add_epi16(row1, row7); \
+ __m128i sum35 = _mm_add_epi16(row3, row5); \
+ dct_rot(y4o,y5o, sum17,sum35, rot1_0,rot1_1); \
+ dct_wadd(x4, y0o, y4o); \
+ dct_wadd(x5, y1o, y5o); \
+ dct_wadd(x6, y2o, y5o); \
+ dct_wadd(x7, y3o, y4o); \
+ dct_bfly32o(row0,row7, x0,x7,bias,shift); \
+ dct_bfly32o(row1,row6, x1,x6,bias,shift); \
+ dct_bfly32o(row2,row5, x2,x5,bias,shift); \
+ dct_bfly32o(row3,row4, x3,x4,bias,shift); \
+ }
+
+ __m128i rot0_0 = dct_const(stbi__f2f(0.5411961f), stbi__f2f(0.5411961f) + stbi__f2f(-1.847759065f));
+ __m128i rot0_1 = dct_const(stbi__f2f(0.5411961f) + stbi__f2f( 0.765366865f), stbi__f2f(0.5411961f));
+ __m128i rot1_0 = dct_const(stbi__f2f(1.175875602f) + stbi__f2f(-0.899976223f), stbi__f2f(1.175875602f));
+ __m128i rot1_1 = dct_const(stbi__f2f(1.175875602f), stbi__f2f(1.175875602f) + stbi__f2f(-2.562915447f));
+ __m128i rot2_0 = dct_const(stbi__f2f(-1.961570560f) + stbi__f2f( 0.298631336f), stbi__f2f(-1.961570560f));
+ __m128i rot2_1 = dct_const(stbi__f2f(-1.961570560f), stbi__f2f(-1.961570560f) + stbi__f2f( 3.072711026f));
+ __m128i rot3_0 = dct_const(stbi__f2f(-0.390180644f) + stbi__f2f( 2.053119869f), stbi__f2f(-0.390180644f));
+ __m128i rot3_1 = dct_const(stbi__f2f(-0.390180644f), stbi__f2f(-0.390180644f) + stbi__f2f( 1.501321110f));
+
+ // rounding biases in column/row passes, see stbi__idct_block for explanation.
+ __m128i bias_0 = _mm_set1_epi32(512);
+ __m128i bias_1 = _mm_set1_epi32(65536 + (128<<17));
+
+ // load
+ row0 = _mm_load_si128((const __m128i *) (data + 0*8));
+ row1 = _mm_load_si128((const __m128i *) (data + 1*8));
+ row2 = _mm_load_si128((const __m128i *) (data + 2*8));
+ row3 = _mm_load_si128((const __m128i *) (data + 3*8));
+ row4 = _mm_load_si128((const __m128i *) (data + 4*8));
+ row5 = _mm_load_si128((const __m128i *) (data + 5*8));
+ row6 = _mm_load_si128((const __m128i *) (data + 6*8));
+ row7 = _mm_load_si128((const __m128i *) (data + 7*8));
+
+ // column pass
+ dct_pass(bias_0, 10);
+
+ {
+ // 16bit 8x8 transpose pass 1
+ dct_interleave16(row0, row4);
+ dct_interleave16(row1, row5);
+ dct_interleave16(row2, row6);
+ dct_interleave16(row3, row7);
+
+ // transpose pass 2
+ dct_interleave16(row0, row2);
+ dct_interleave16(row1, row3);
+ dct_interleave16(row4, row6);
+ dct_interleave16(row5, row7);
+
+ // transpose pass 3
+ dct_interleave16(row0, row1);
+ dct_interleave16(row2, row3);
+ dct_interleave16(row4, row5);
+ dct_interleave16(row6, row7);
+ }
+
+ // row pass
+ dct_pass(bias_1, 17);
+
+ {
+ // pack
+ __m128i p0 = _mm_packus_epi16(row0, row1); // a0a1a2a3...a7b0b1b2b3...b7
+ __m128i p1 = _mm_packus_epi16(row2, row3);
+ __m128i p2 = _mm_packus_epi16(row4, row5);
+ __m128i p3 = _mm_packus_epi16(row6, row7);
+
+ // 8bit 8x8 transpose pass 1
+ dct_interleave8(p0, p2); // a0e0a1e1...
+ dct_interleave8(p1, p3); // c0g0c1g1...
+
+ // transpose pass 2
+ dct_interleave8(p0, p1); // a0c0e0g0...
+ dct_interleave8(p2, p3); // b0d0f0h0...
+
+ // transpose pass 3
+ dct_interleave8(p0, p2); // a0b0c0d0...
+ dct_interleave8(p1, p3); // a4b4c4d4...
+
+ // store
+ _mm_storel_epi64((__m128i *) out, p0); out += out_stride;
+ _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p0, 0x4e)); out += out_stride;
+ _mm_storel_epi64((__m128i *) out, p2); out += out_stride;
+ _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p2, 0x4e)); out += out_stride;
+ _mm_storel_epi64((__m128i *) out, p1); out += out_stride;
+ _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p1, 0x4e)); out += out_stride;
+ _mm_storel_epi64((__m128i *) out, p3); out += out_stride;
+ _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p3, 0x4e));
+ }
+
+#undef dct_const
+#undef dct_rot
+#undef dct_widen
+#undef dct_wadd
+#undef dct_wsub
+#undef dct_bfly32o
+#undef dct_interleave8
+#undef dct_interleave16
+#undef dct_pass
+}
+
+#endif // STBI_SSE2
+
+#ifdef STBI_NEON
+
+// NEON integer IDCT. should produce bit-identical
+// results to the generic C version.
+static void stbi__idct_simd(stbi_uc *out, int out_stride, short data[64])
+{
+ int16x8_t row0, row1, row2, row3, row4, row5, row6, row7;
+
+ int16x4_t rot0_0 = vdup_n_s16(stbi__f2f(0.5411961f));
+ int16x4_t rot0_1 = vdup_n_s16(stbi__f2f(-1.847759065f));
+ int16x4_t rot0_2 = vdup_n_s16(stbi__f2f( 0.765366865f));
+ int16x4_t rot1_0 = vdup_n_s16(stbi__f2f( 1.175875602f));
+ int16x4_t rot1_1 = vdup_n_s16(stbi__f2f(-0.899976223f));
+ int16x4_t rot1_2 = vdup_n_s16(stbi__f2f(-2.562915447f));
+ int16x4_t rot2_0 = vdup_n_s16(stbi__f2f(-1.961570560f));
+ int16x4_t rot2_1 = vdup_n_s16(stbi__f2f(-0.390180644f));
+ int16x4_t rot3_0 = vdup_n_s16(stbi__f2f( 0.298631336f));
+ int16x4_t rot3_1 = vdup_n_s16(stbi__f2f( 2.053119869f));
+ int16x4_t rot3_2 = vdup_n_s16(stbi__f2f( 3.072711026f));
+ int16x4_t rot3_3 = vdup_n_s16(stbi__f2f( 1.501321110f));
+
+#define dct_long_mul(out, inq, coeff) \
+ int32x4_t out##_l = vmull_s16(vget_low_s16(inq), coeff); \
+ int32x4_t out##_h = vmull_s16(vget_high_s16(inq), coeff)
+
+#define dct_long_mac(out, acc, inq, coeff) \
+ int32x4_t out##_l = vmlal_s16(acc##_l, vget_low_s16(inq), coeff); \
+ int32x4_t out##_h = vmlal_s16(acc##_h, vget_high_s16(inq), coeff)
+
+#define dct_widen(out, inq) \
+ int32x4_t out##_l = vshll_n_s16(vget_low_s16(inq), 12); \
+ int32x4_t out##_h = vshll_n_s16(vget_high_s16(inq), 12)
+
+// wide add
+#define dct_wadd(out, a, b) \
+ int32x4_t out##_l = vaddq_s32(a##_l, b##_l); \
+ int32x4_t out##_h = vaddq_s32(a##_h, b##_h)
+
+// wide sub
+#define dct_wsub(out, a, b) \
+ int32x4_t out##_l = vsubq_s32(a##_l, b##_l); \
+ int32x4_t out##_h = vsubq_s32(a##_h, b##_h)
+
+// butterfly a/b, then shift using "shiftop" by "s" and pack
+#define dct_bfly32o(out0,out1, a,b,shiftop,s) \
+ { \
+ dct_wadd(sum, a, b); \
+ dct_wsub(dif, a, b); \
+ out0 = vcombine_s16(shiftop(sum_l, s), shiftop(sum_h, s)); \
+ out1 = vcombine_s16(shiftop(dif_l, s), shiftop(dif_h, s)); \
+ }
+
+#define dct_pass(shiftop, shift) \
+ { \
+ /* even part */ \
+ int16x8_t sum26 = vaddq_s16(row2, row6); \
+ dct_long_mul(p1e, sum26, rot0_0); \
+ dct_long_mac(t2e, p1e, row6, rot0_1); \
+ dct_long_mac(t3e, p1e, row2, rot0_2); \
+ int16x8_t sum04 = vaddq_s16(row0, row4); \
+ int16x8_t dif04 = vsubq_s16(row0, row4); \
+ dct_widen(t0e, sum04); \
+ dct_widen(t1e, dif04); \
+ dct_wadd(x0, t0e, t3e); \
+ dct_wsub(x3, t0e, t3e); \
+ dct_wadd(x1, t1e, t2e); \
+ dct_wsub(x2, t1e, t2e); \
+ /* odd part */ \
+ int16x8_t sum15 = vaddq_s16(row1, row5); \
+ int16x8_t sum17 = vaddq_s16(row1, row7); \
+ int16x8_t sum35 = vaddq_s16(row3, row5); \
+ int16x8_t sum37 = vaddq_s16(row3, row7); \
+ int16x8_t sumodd = vaddq_s16(sum17, sum35); \
+ dct_long_mul(p5o, sumodd, rot1_0); \
+ dct_long_mac(p1o, p5o, sum17, rot1_1); \
+ dct_long_mac(p2o, p5o, sum35, rot1_2); \
+ dct_long_mul(p3o, sum37, rot2_0); \
+ dct_long_mul(p4o, sum15, rot2_1); \
+ dct_wadd(sump13o, p1o, p3o); \
+ dct_wadd(sump24o, p2o, p4o); \
+ dct_wadd(sump23o, p2o, p3o); \
+ dct_wadd(sump14o, p1o, p4o); \
+ dct_long_mac(x4, sump13o, row7, rot3_0); \
+ dct_long_mac(x5, sump24o, row5, rot3_1); \
+ dct_long_mac(x6, sump23o, row3, rot3_2); \
+ dct_long_mac(x7, sump14o, row1, rot3_3); \
+ dct_bfly32o(row0,row7, x0,x7,shiftop,shift); \
+ dct_bfly32o(row1,row6, x1,x6,shiftop,shift); \
+ dct_bfly32o(row2,row5, x2,x5,shiftop,shift); \
+ dct_bfly32o(row3,row4, x3,x4,shiftop,shift); \
+ }
+
+ // load
+ row0 = vld1q_s16(data + 0*8);
+ row1 = vld1q_s16(data + 1*8);
+ row2 = vld1q_s16(data + 2*8);
+ row3 = vld1q_s16(data + 3*8);
+ row4 = vld1q_s16(data + 4*8);
+ row5 = vld1q_s16(data + 5*8);
+ row6 = vld1q_s16(data + 6*8);
+ row7 = vld1q_s16(data + 7*8);
+
+ // add DC bias
+ row0 = vaddq_s16(row0, vsetq_lane_s16(1024, vdupq_n_s16(0), 0));
+
+ // column pass
+ dct_pass(vrshrn_n_s32, 10);
+
+ // 16bit 8x8 transpose
+ {
+// these three map to a single VTRN.16, VTRN.32, and VSWP, respectively.
+// whether compilers actually get this is another story, sadly.
+#define dct_trn16(x, y) { int16x8x2_t t = vtrnq_s16(x, y); x = t.val[0]; y = t.val[1]; }
+#define dct_trn32(x, y) { int32x4x2_t t = vtrnq_s32(vreinterpretq_s32_s16(x), vreinterpretq_s32_s16(y)); x = vreinterpretq_s16_s32(t.val[0]); y = vreinterpretq_s16_s32(t.val[1]); }
+#define dct_trn64(x, y) { int16x8_t x0 = x; int16x8_t y0 = y; x = vcombine_s16(vget_low_s16(x0), vget_low_s16(y0)); y = vcombine_s16(vget_high_s16(x0), vget_high_s16(y0)); }
+
+ // pass 1
+ dct_trn16(row0, row1); // a0b0a2b2a4b4a6b6
+ dct_trn16(row2, row3);
+ dct_trn16(row4, row5);
+ dct_trn16(row6, row7);
+
+ // pass 2
+ dct_trn32(row0, row2); // a0b0c0d0a4b4c4d4
+ dct_trn32(row1, row3);
+ dct_trn32(row4, row6);
+ dct_trn32(row5, row7);
+
+ // pass 3
+ dct_trn64(row0, row4); // a0b0c0d0e0f0g0h0
+ dct_trn64(row1, row5);
+ dct_trn64(row2, row6);
+ dct_trn64(row3, row7);
+
+#undef dct_trn16
+#undef dct_trn32
+#undef dct_trn64
+ }
+
+ // row pass
+ // vrshrn_n_s32 only supports shifts up to 16, we need
+ // 17. so do a non-rounding shift of 16 first then follow
+ // up with a rounding shift by 1.
+ dct_pass(vshrn_n_s32, 16);
+
+ {
+ // pack and round
+ uint8x8_t p0 = vqrshrun_n_s16(row0, 1);
+ uint8x8_t p1 = vqrshrun_n_s16(row1, 1);
+ uint8x8_t p2 = vqrshrun_n_s16(row2, 1);
+ uint8x8_t p3 = vqrshrun_n_s16(row3, 1);
+ uint8x8_t p4 = vqrshrun_n_s16(row4, 1);
+ uint8x8_t p5 = vqrshrun_n_s16(row5, 1);
+ uint8x8_t p6 = vqrshrun_n_s16(row6, 1);
+ uint8x8_t p7 = vqrshrun_n_s16(row7, 1);
+
+ // again, these can translate into one instruction, but often don't.
+#define dct_trn8_8(x, y) { uint8x8x2_t t = vtrn_u8(x, y); x = t.val[0]; y = t.val[1]; }
+#define dct_trn8_16(x, y) { uint16x4x2_t t = vtrn_u16(vreinterpret_u16_u8(x), vreinterpret_u16_u8(y)); x = vreinterpret_u8_u16(t.val[0]); y = vreinterpret_u8_u16(t.val[1]); }
+#define dct_trn8_32(x, y) { uint32x2x2_t t = vtrn_u32(vreinterpret_u32_u8(x), vreinterpret_u32_u8(y)); x = vreinterpret_u8_u32(t.val[0]); y = vreinterpret_u8_u32(t.val[1]); }
+
+ // sadly can't use interleaved stores here since we only write
+ // 8 bytes to each scan line!
+
+ // 8x8 8-bit transpose pass 1
+ dct_trn8_8(p0, p1);
+ dct_trn8_8(p2, p3);
+ dct_trn8_8(p4, p5);
+ dct_trn8_8(p6, p7);
+
+ // pass 2
+ dct_trn8_16(p0, p2);
+ dct_trn8_16(p1, p3);
+ dct_trn8_16(p4, p6);
+ dct_trn8_16(p5, p7);
+
+ // pass 3
+ dct_trn8_32(p0, p4);
+ dct_trn8_32(p1, p5);
+ dct_trn8_32(p2, p6);
+ dct_trn8_32(p3, p7);
+
+ // store
+ vst1_u8(out, p0); out += out_stride;
+ vst1_u8(out, p1); out += out_stride;
+ vst1_u8(out, p2); out += out_stride;
+ vst1_u8(out, p3); out += out_stride;
+ vst1_u8(out, p4); out += out_stride;
+ vst1_u8(out, p5); out += out_stride;
+ vst1_u8(out, p6); out += out_stride;
+ vst1_u8(out, p7);
+
+#undef dct_trn8_8
+#undef dct_trn8_16
+#undef dct_trn8_32
+ }
+
+#undef dct_long_mul
+#undef dct_long_mac
+#undef dct_widen
+#undef dct_wadd
+#undef dct_wsub
+#undef dct_bfly32o
+#undef dct_pass
+}
+
+#endif // STBI_NEON
+
+#define STBI__MARKER_none 0xff
+// if there's a pending marker from the entropy stream, return that
+// otherwise, fetch from the stream and get a marker. if there's no
+// marker, return 0xff, which is never a valid marker value
+static stbi_uc stbi__get_marker(stbi__jpeg *j)
+{
+ stbi_uc x;
+ if (j->marker != STBI__MARKER_none) { x = j->marker; j->marker = STBI__MARKER_none; return x; }
+ x = stbi__get8(j->s);
+ if (x != 0xff) return STBI__MARKER_none;
+ while (x == 0xff)
+ x = stbi__get8(j->s); // consume repeated 0xff fill bytes
+ return x;
+}
+
+// in each scan, we'll have scan_n components, and the order
+// of the components is specified by order[]
+#define STBI__RESTART(x) ((x) >= 0xd0 && (x) <= 0xd7)
+
+// after a restart interval, stbi__jpeg_reset the entropy decoder and
+// the dc prediction
+static void stbi__jpeg_reset(stbi__jpeg *j)
+{
+ j->code_bits = 0;
+ j->code_buffer = 0;
+ j->nomore = 0;
+ j->img_comp[0].dc_pred = j->img_comp[1].dc_pred = j->img_comp[2].dc_pred = j->img_comp[3].dc_pred = 0;
+ j->marker = STBI__MARKER_none;
+ j->todo = j->restart_interval ? j->restart_interval : 0x7fffffff;
+ j->eob_run = 0;
+ // no more than 1<<31 MCUs if no restart_interal? that's plenty safe,
+ // since we don't even allow 1<<30 pixels
+}
+
+static int stbi__parse_entropy_coded_data(stbi__jpeg *z)
+{
+ stbi__jpeg_reset(z);
+ if (!z->progressive) {
+ if (z->scan_n == 1) {
+ int i,j;
+ STBI_SIMD_ALIGN(short, data[64]);
+ int n = z->order[0];
+ // non-interleaved data, we just need to process one block at a time,
+ // in trivial scanline order
+ // number of blocks to do just depends on how many actual "pixels" this
+ // component has, independent of interleaved MCU blocking and such
+ int w = (z->img_comp[n].x+7) >> 3;
+ int h = (z->img_comp[n].y+7) >> 3;
+ for (j=0; j < h; ++j) {
+ for (i=0; i < w; ++i) {
+ int ha = z->img_comp[n].ha;
+ if (!stbi__jpeg_decode_block(z, data, z->huff_dc+z->img_comp[n].hd, z->huff_ac+ha, z->fast_ac[ha], n, z->dequant[z->img_comp[n].tq])) return 0;
+ z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*j*8+i*8, z->img_comp[n].w2, data);
+ // every data block is an MCU, so countdown the restart interval
+ if (--z->todo <= 0) {
+ if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
+ // if it's NOT a restart, then just bail, so we get corrupt data
+ // rather than no data
+ if (!STBI__RESTART(z->marker)) return 1;
+ stbi__jpeg_reset(z);
+ }
+ }
+ }
+ return 1;
+ } else { // interleaved
+ int i,j,k,x,y;
+ STBI_SIMD_ALIGN(short, data[64]);
+ for (j=0; j < z->img_mcu_y; ++j) {
+ for (i=0; i < z->img_mcu_x; ++i) {
+ // scan an interleaved mcu... process scan_n components in order
+ for (k=0; k < z->scan_n; ++k) {
+ int n = z->order[k];
+ // scan out an mcu's worth of this component; that's just determined
+ // by the basic H and V specified for the component
+ for (y=0; y < z->img_comp[n].v; ++y) {
+ for (x=0; x < z->img_comp[n].h; ++x) {
+ int x2 = (i*z->img_comp[n].h + x)*8;
+ int y2 = (j*z->img_comp[n].v + y)*8;
+ int ha = z->img_comp[n].ha;
+ if (!stbi__jpeg_decode_block(z, data, z->huff_dc+z->img_comp[n].hd, z->huff_ac+ha, z->fast_ac[ha], n, z->dequant[z->img_comp[n].tq])) return 0;
+ z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*y2+x2, z->img_comp[n].w2, data);
+ }
+ }
+ }
+ // after all interleaved components, that's an interleaved MCU,
+ // so now count down the restart interval
+ if (--z->todo <= 0) {
+ if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
+ if (!STBI__RESTART(z->marker)) return 1;
+ stbi__jpeg_reset(z);
+ }
+ }
+ }
+ return 1;
+ }
+ } else {
+ if (z->scan_n == 1) {
+ int i,j;
+ int n = z->order[0];
+ // non-interleaved data, we just need to process one block at a time,
+ // in trivial scanline order
+ // number of blocks to do just depends on how many actual "pixels" this
+ // component has, independent of interleaved MCU blocking and such
+ int w = (z->img_comp[n].x+7) >> 3;
+ int h = (z->img_comp[n].y+7) >> 3;
+ for (j=0; j < h; ++j) {
+ for (i=0; i < w; ++i) {
+ short *data = z->img_comp[n].coeff + 64 * (i + j * z->img_comp[n].coeff_w);
+ if (z->spec_start == 0) {
+ if (!stbi__jpeg_decode_block_prog_dc(z, data, &z->huff_dc[z->img_comp[n].hd], n))
+ return 0;
+ } else {
+ int ha = z->img_comp[n].ha;
+ if (!stbi__jpeg_decode_block_prog_ac(z, data, &z->huff_ac[ha], z->fast_ac[ha]))
+ return 0;
+ }
+ // every data block is an MCU, so countdown the restart interval
+ if (--z->todo <= 0) {
+ if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
+ if (!STBI__RESTART(z->marker)) return 1;
+ stbi__jpeg_reset(z);
+ }
+ }
+ }
+ return 1;
+ } else { // interleaved
+ int i,j,k,x,y;
+ for (j=0; j < z->img_mcu_y; ++j) {
+ for (i=0; i < z->img_mcu_x; ++i) {
+ // scan an interleaved mcu... process scan_n components in order
+ for (k=0; k < z->scan_n; ++k) {
+ int n = z->order[k];
+ // scan out an mcu's worth of this component; that's just determined
+ // by the basic H and V specified for the component
+ for (y=0; y < z->img_comp[n].v; ++y) {
+ for (x=0; x < z->img_comp[n].h; ++x) {
+ int x2 = (i*z->img_comp[n].h + x);
+ int y2 = (j*z->img_comp[n].v + y);
+ short *data = z->img_comp[n].coeff + 64 * (x2 + y2 * z->img_comp[n].coeff_w);
+ if (!stbi__jpeg_decode_block_prog_dc(z, data, &z->huff_dc[z->img_comp[n].hd], n))
+ return 0;
+ }
+ }
+ }
+ // after all interleaved components, that's an interleaved MCU,
+ // so now count down the restart interval
+ if (--z->todo <= 0) {
+ if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
+ if (!STBI__RESTART(z->marker)) return 1;
+ stbi__jpeg_reset(z);
+ }
+ }
+ }
+ return 1;
+ }
+ }
+}
+
+static void stbi__jpeg_dequantize(short *data, stbi__uint16 *dequant)
+{
+ int i;
+ for (i=0; i < 64; ++i)
+ data[i] *= dequant[i];
+}
+
+static void stbi__jpeg_finish(stbi__jpeg *z)
+{
+ if (z->progressive) {
+ // dequantize and idct the data
+ int i,j,n;
+ for (n=0; n < z->s->img_n; ++n) {
+ int w = (z->img_comp[n].x+7) >> 3;
+ int h = (z->img_comp[n].y+7) >> 3;
+ for (j=0; j < h; ++j) {
+ for (i=0; i < w; ++i) {
+ short *data = z->img_comp[n].coeff + 64 * (i + j * z->img_comp[n].coeff_w);
+ stbi__jpeg_dequantize(data, z->dequant[z->img_comp[n].tq]);
+ z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*j*8+i*8, z->img_comp[n].w2, data);
+ }
+ }
+ }
+ }
+}
+
+static int stbi__process_marker(stbi__jpeg *z, int m)
+{
+ int L;
+ switch (m) {
+ case STBI__MARKER_none: // no marker found
+ return stbi__err("expected marker","Corrupt JPEG");
+
+ case 0xDD: // DRI - specify restart interval
+ if (stbi__get16be(z->s) != 4) return stbi__err("bad DRI len","Corrupt JPEG");
+ z->restart_interval = stbi__get16be(z->s);
+ return 1;
+
+ case 0xDB: // DQT - define quantization table
+ L = stbi__get16be(z->s)-2;
+ while (L > 0) {
+ int q = stbi__get8(z->s);
+ int p = q >> 4, sixteen = (p != 0);
+ int t = q & 15,i;
+ if (p != 0 && p != 1) return stbi__err("bad DQT type","Corrupt JPEG");
+ if (t > 3) return stbi__err("bad DQT table","Corrupt JPEG");
+
+ for (i=0; i < 64; ++i)
+ z->dequant[t][stbi__jpeg_dezigzag[i]] = sixteen ? stbi__get16be(z->s) : stbi__get8(z->s);
+ L -= (sixteen ? 129 : 65);
+ }
+ return L==0;
+
+ case 0xC4: // DHT - define huffman table
+ L = stbi__get16be(z->s)-2;
+ while (L > 0) {
+ stbi_uc *v;
+ int sizes[16],i,n=0;
+ int q = stbi__get8(z->s);
+ int tc = q >> 4;
+ int th = q & 15;
+ if (tc > 1 || th > 3) return stbi__err("bad DHT header","Corrupt JPEG");
+ for (i=0; i < 16; ++i) {
+ sizes[i] = stbi__get8(z->s);
+ n += sizes[i];
+ }
+ L -= 17;
+ if (tc == 0) {
+ if (!stbi__build_huffman(z->huff_dc+th, sizes)) return 0;
+ v = z->huff_dc[th].values;
+ } else {
+ if (!stbi__build_huffman(z->huff_ac+th, sizes)) return 0;
+ v = z->huff_ac[th].values;
+ }
+ for (i=0; i < n; ++i)
+ v[i] = stbi__get8(z->s);
+ if (tc != 0)
+ stbi__build_fast_ac(z->fast_ac[th], z->huff_ac + th);
+ L -= n;
+ }
+ return L==0;
+ }
+
+ // check for comment block or APP blocks
+ if ((m >= 0xE0 && m <= 0xEF) || m == 0xFE) {
+ L = stbi__get16be(z->s);
+ if (L < 2) {
+ if (m == 0xFE)
+ return stbi__err("bad COM len","Corrupt JPEG");
+ else
+ return stbi__err("bad APP len","Corrupt JPEG");
+ }
+ L -= 2;
+
+ if (m == 0xE0 && L >= 5) { // JFIF APP0 segment
+ static const unsigned char tag[5] = {'J','F','I','F','\0'};
+ int ok = 1;
+ int i;
+ for (i=0; i < 5; ++i)
+ if (stbi__get8(z->s) != tag[i])
+ ok = 0;
+ L -= 5;
+ if (ok)
+ z->jfif = 1;
+ } else if (m == 0xEE && L >= 12) { // Adobe APP14 segment
+ static const unsigned char tag[6] = {'A','d','o','b','e','\0'};
+ int ok = 1;
+ int i;
+ for (i=0; i < 6; ++i)
+ if (stbi__get8(z->s) != tag[i])
+ ok = 0;
+ L -= 6;
+ if (ok) {
+ stbi__get8(z->s); // version
+ stbi__get16be(z->s); // flags0
+ stbi__get16be(z->s); // flags1
+ z->app14_color_transform = stbi__get8(z->s); // color transform
+ L -= 6;
+ }
+ }
+
+ stbi__skip(z->s, L);
+ return 1;
+ }
+
+ return stbi__err("unknown marker","Corrupt JPEG");
+}
+
+// after we see SOS
+static int stbi__process_scan_header(stbi__jpeg *z)
+{
+ int i;
+ int Ls = stbi__get16be(z->s);
+ z->scan_n = stbi__get8(z->s);
+ if (z->scan_n < 1 || z->scan_n > 4 || z->scan_n > (int) z->s->img_n) return stbi__err("bad SOS component count","Corrupt JPEG");
+ if (Ls != 6+2*z->scan_n) return stbi__err("bad SOS len","Corrupt JPEG");
+ for (i=0; i < z->scan_n; ++i) {
+ int id = stbi__get8(z->s), which;
+ int q = stbi__get8(z->s);
+ for (which = 0; which < z->s->img_n; ++which)
+ if (z->img_comp[which].id == id)
+ break;
+ if (which == z->s->img_n) return 0; // no match
+ z->img_comp[which].hd = q >> 4; if (z->img_comp[which].hd > 3) return stbi__err("bad DC huff","Corrupt JPEG");
+ z->img_comp[which].ha = q & 15; if (z->img_comp[which].ha > 3) return stbi__err("bad AC huff","Corrupt JPEG");
+ z->order[i] = which;
+ }
+
+ {
+ int aa;
+ z->spec_start = stbi__get8(z->s);
+ z->spec_end = stbi__get8(z->s); // should be 63, but might be 0
+ aa = stbi__get8(z->s);
+ z->succ_high = (aa >> 4);
+ z->succ_low = (aa & 15);
+ if (z->progressive) {
+ if (z->spec_start > 63 || z->spec_end > 63 || z->spec_start > z->spec_end || z->succ_high > 13 || z->succ_low > 13)
+ return stbi__err("bad SOS", "Corrupt JPEG");
+ } else {
+ if (z->spec_start != 0) return stbi__err("bad SOS","Corrupt JPEG");
+ if (z->succ_high != 0 || z->succ_low != 0) return stbi__err("bad SOS","Corrupt JPEG");
+ z->spec_end = 63;
+ }
+ }
+
+ return 1;
+}
+
+static int stbi__free_jpeg_components(stbi__jpeg *z, int ncomp, int why)
+{
+ int i;
+ for (i=0; i < ncomp; ++i) {
+ if (z->img_comp[i].raw_data) {
+ STBI_FREE(z->img_comp[i].raw_data);
+ z->img_comp[i].raw_data = NULL;
+ z->img_comp[i].data = NULL;
+ }
+ if (z->img_comp[i].raw_coeff) {
+ STBI_FREE(z->img_comp[i].raw_coeff);
+ z->img_comp[i].raw_coeff = 0;
+ z->img_comp[i].coeff = 0;
+ }
+ if (z->img_comp[i].linebuf) {
+ STBI_FREE(z->img_comp[i].linebuf);
+ z->img_comp[i].linebuf = NULL;
+ }
+ }
+ return why;
+}
+
+static int stbi__process_frame_header(stbi__jpeg *z, int scan)
+{
+ stbi__context *s = z->s;
+ int Lf,p,i,q, h_max=1,v_max=1,c;
+ Lf = stbi__get16be(s); if (Lf < 11) return stbi__err("bad SOF len","Corrupt JPEG"); // JPEG
+ p = stbi__get8(s); if (p != 8) return stbi__err("only 8-bit","JPEG format not supported: 8-bit only"); // JPEG baseline
+ s->img_y = stbi__get16be(s); if (s->img_y == 0) return stbi__err("no header height", "JPEG format not supported: delayed height"); // Legal, but we don't handle it--but neither does IJG
+ s->img_x = stbi__get16be(s); if (s->img_x == 0) return stbi__err("0 width","Corrupt JPEG"); // JPEG requires
+ c = stbi__get8(s);
+ if (c != 3 && c != 1 && c != 4) return stbi__err("bad component count","Corrupt JPEG");
+ s->img_n = c;
+ for (i=0; i < c; ++i) {
+ z->img_comp[i].data = NULL;
+ z->img_comp[i].linebuf = NULL;
+ }
+
+ if (Lf != 8+3*s->img_n) return stbi__err("bad SOF len","Corrupt JPEG");
+
+ z->rgb = 0;
+ for (i=0; i < s->img_n; ++i) {
+ static unsigned char rgb[3] = { 'R', 'G', 'B' };
+ z->img_comp[i].id = stbi__get8(s);
+ if (s->img_n == 3 && z->img_comp[i].id == rgb[i])
+ ++z->rgb;
+ q = stbi__get8(s);
+ z->img_comp[i].h = (q >> 4); if (!z->img_comp[i].h || z->img_comp[i].h > 4) return stbi__err("bad H","Corrupt JPEG");
+ z->img_comp[i].v = q & 15; if (!z->img_comp[i].v || z->img_comp[i].v > 4) return stbi__err("bad V","Corrupt JPEG");
+ z->img_comp[i].tq = stbi__get8(s); if (z->img_comp[i].tq > 3) return stbi__err("bad TQ","Corrupt JPEG");
+ }
+
+ if (scan != STBI__SCAN_load) return 1;
+
+ if (!stbi__mad3sizes_valid(s->img_x, s->img_y, s->img_n, 0)) return stbi__err("too large", "Image too large to decode");
+
+ for (i=0; i < s->img_n; ++i) {
+ if (z->img_comp[i].h > h_max) h_max = z->img_comp[i].h;
+ if (z->img_comp[i].v > v_max) v_max = z->img_comp[i].v;
+ }
+
+ // compute interleaved mcu info
+ z->img_h_max = h_max;
+ z->img_v_max = v_max;
+ z->img_mcu_w = h_max * 8;
+ z->img_mcu_h = v_max * 8;
+ // these sizes can't be more than 17 bits
+ z->img_mcu_x = (s->img_x + z->img_mcu_w-1) / z->img_mcu_w;
+ z->img_mcu_y = (s->img_y + z->img_mcu_h-1) / z->img_mcu_h;
+
+ for (i=0; i < s->img_n; ++i) {
+ // number of effective pixels (e.g. for non-interleaved MCU)
+ z->img_comp[i].x = (s->img_x * z->img_comp[i].h + h_max-1) / h_max;
+ z->img_comp[i].y = (s->img_y * z->img_comp[i].v + v_max-1) / v_max;
+ // to simplify generation, we'll allocate enough memory to decode
+ // the bogus oversized data from using interleaved MCUs and their
+ // big blocks (e.g. a 16x16 iMCU on an image of width 33); we won't
+ // discard the extra data until colorspace conversion
+ //
+ // img_mcu_x, img_mcu_y: <=17 bits; comp[i].h and .v are <=4 (checked earlier)
+ // so these muls can't overflow with 32-bit ints (which we require)
+ z->img_comp[i].w2 = z->img_mcu_x * z->img_comp[i].h * 8;
+ z->img_comp[i].h2 = z->img_mcu_y * z->img_comp[i].v * 8;
+ z->img_comp[i].coeff = 0;
+ z->img_comp[i].raw_coeff = 0;
+ z->img_comp[i].linebuf = NULL;
+ z->img_comp[i].raw_data = stbi__malloc_mad2(z->img_comp[i].w2, z->img_comp[i].h2, 15);
+ if (z->img_comp[i].raw_data == NULL)
+ return stbi__free_jpeg_components(z, i+1, stbi__err("outofmem", "Out of memory"));
+ // align blocks for idct using mmx/sse
+ z->img_comp[i].data = (stbi_uc*) (((size_t) z->img_comp[i].raw_data + 15) & ~15);
+ if (z->progressive) {
+ // w2, h2 are multiples of 8 (see above)
+ z->img_comp[i].coeff_w = z->img_comp[i].w2 / 8;
+ z->img_comp[i].coeff_h = z->img_comp[i].h2 / 8;
+ z->img_comp[i].raw_coeff = stbi__malloc_mad3(z->img_comp[i].w2, z->img_comp[i].h2, sizeof(short), 15);
+ if (z->img_comp[i].raw_coeff == NULL)
+ return stbi__free_jpeg_components(z, i+1, stbi__err("outofmem", "Out of memory"));
+ z->img_comp[i].coeff = (short*) (((size_t) z->img_comp[i].raw_coeff + 15) & ~15);
+ }
+ }
+
+ return 1;
+}
+
+// use comparisons since in some cases we handle more than one case (e.g. SOF)
+#define stbi__DNL(x) ((x) == 0xdc)
+#define stbi__SOI(x) ((x) == 0xd8)
+#define stbi__EOI(x) ((x) == 0xd9)
+#define stbi__SOF(x) ((x) == 0xc0 || (x) == 0xc1 || (x) == 0xc2)
+#define stbi__SOS(x) ((x) == 0xda)
+
+#define stbi__SOF_progressive(x) ((x) == 0xc2)
+
+static int stbi__decode_jpeg_header(stbi__jpeg *z, int scan)
+{
+ int m;
+ z->jfif = 0;
+ z->app14_color_transform = -1; // valid values are 0,1,2
+ z->marker = STBI__MARKER_none; // initialize cached marker to empty
+ m = stbi__get_marker(z);
+ if (!stbi__SOI(m)) return stbi__err("no SOI","Corrupt JPEG");
+ if (scan == STBI__SCAN_type) return 1;
+ m = stbi__get_marker(z);
+ while (!stbi__SOF(m)) {
+ if (!stbi__process_marker(z,m)) return 0;
+ m = stbi__get_marker(z);
+ while (m == STBI__MARKER_none) {
+ // some files have extra padding after their blocks, so ok, we'll scan
+ if (stbi__at_eof(z->s)) return stbi__err("no SOF", "Corrupt JPEG");
+ m = stbi__get_marker(z);
+ }
+ }
+ z->progressive = stbi__SOF_progressive(m);
+ if (!stbi__process_frame_header(z, scan)) return 0;
+ return 1;
+}
+
+// decode image to YCbCr format
+static int stbi__decode_jpeg_image(stbi__jpeg *j)
+{
+ int m;
+ for (m = 0; m < 4; m++) {
+ j->img_comp[m].raw_data = NULL;
+ j->img_comp[m].raw_coeff = NULL;
+ }
+ j->restart_interval = 0;
+ if (!stbi__decode_jpeg_header(j, STBI__SCAN_load)) return 0;
+ m = stbi__get_marker(j);
+ while (!stbi__EOI(m)) {
+ if (stbi__SOS(m)) {
+ if (!stbi__process_scan_header(j)) return 0;
+ if (!stbi__parse_entropy_coded_data(j)) return 0;
+ if (j->marker == STBI__MARKER_none ) {
+ // handle 0s at the end of image data from IP Kamera 9060
+ while (!stbi__at_eof(j->s)) {
+ int x = stbi__get8(j->s);
+ if (x == 255) {
+ j->marker = stbi__get8(j->s);
+ break;
+ }
+ }
+ // if we reach eof without hitting a marker, stbi__get_marker() below will fail and we'll eventually return 0
+ }
+ } else if (stbi__DNL(m)) {
+ int Ld = stbi__get16be(j->s);
+ stbi__uint32 NL = stbi__get16be(j->s);
+ if (Ld != 4) stbi__err("bad DNL len", "Corrupt JPEG");
+ if (NL != j->s->img_y) stbi__err("bad DNL height", "Corrupt JPEG");
+ } else {
+ if (!stbi__process_marker(j, m)) return 0;
+ }
+ m = stbi__get_marker(j);
+ }
+ if (j->progressive)
+ stbi__jpeg_finish(j);
+ return 1;
+}
+
+// static jfif-centered resampling (across block boundaries)
+
+typedef stbi_uc *(*resample_row_func)(stbi_uc *out, stbi_uc *in0, stbi_uc *in1,
+ int w, int hs);
+
+#define stbi__div4(x) ((stbi_uc) ((x) >> 2))
+
+static stbi_uc *resample_row_1(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+{
+ STBI_NOTUSED(out);
+ STBI_NOTUSED(in_far);
+ STBI_NOTUSED(w);
+ STBI_NOTUSED(hs);
+ return in_near;
+}
+
+static stbi_uc* stbi__resample_row_v_2(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+{
+ // need to generate two samples vertically for every one in input
+ int i;
+ STBI_NOTUSED(hs);
+ for (i=0; i < w; ++i)
+ out[i] = stbi__div4(3*in_near[i] + in_far[i] + 2);
+ return out;
+}
+
+static stbi_uc* stbi__resample_row_h_2(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+{
+ // need to generate two samples horizontally for every one in input
+ int i;
+ stbi_uc *input = in_near;
+
+ if (w == 1) {
+ // if only one sample, can't do any interpolation
+ out[0] = out[1] = input[0];
+ return out;
+ }
+
+ out[0] = input[0];
+ out[1] = stbi__div4(input[0]*3 + input[1] + 2);
+ for (i=1; i < w-1; ++i) {
+ int n = 3*input[i]+2;
+ out[i*2+0] = stbi__div4(n+input[i-1]);
+ out[i*2+1] = stbi__div4(n+input[i+1]);
+ }
+ out[i*2+0] = stbi__div4(input[w-2]*3 + input[w-1] + 2);
+ out[i*2+1] = input[w-1];
+
+ STBI_NOTUSED(in_far);
+ STBI_NOTUSED(hs);
+
+ return out;
+}
+
+#define stbi__div16(x) ((stbi_uc) ((x) >> 4))
+
+static stbi_uc *stbi__resample_row_hv_2(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+{
+ // need to generate 2x2 samples for every one in input
+ int i,t0,t1;
+ if (w == 1) {
+ out[0] = out[1] = stbi__div4(3*in_near[0] + in_far[0] + 2);
+ return out;
+ }
+
+ t1 = 3*in_near[0] + in_far[0];
+ out[0] = stbi__div4(t1+2);
+ for (i=1; i < w; ++i) {
+ t0 = t1;
+ t1 = 3*in_near[i]+in_far[i];
+ out[i*2-1] = stbi__div16(3*t0 + t1 + 8);
+ out[i*2 ] = stbi__div16(3*t1 + t0 + 8);
+ }
+ out[w*2-1] = stbi__div4(t1+2);
+
+ STBI_NOTUSED(hs);
+
+ return out;
+}
+
+#if defined(STBI_SSE2) || defined(STBI_NEON)
+static stbi_uc *stbi__resample_row_hv_2_simd(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+{
+ // need to generate 2x2 samples for every one in input
+ int i=0,t0,t1;
+
+ if (w == 1) {
+ out[0] = out[1] = stbi__div4(3*in_near[0] + in_far[0] + 2);
+ return out;
+ }
+
+ t1 = 3*in_near[0] + in_far[0];
+ // process groups of 8 pixels for as long as we can.
+ // note we can't handle the last pixel in a row in this loop
+ // because we need to handle the filter boundary conditions.
+ for (; i < ((w-1) & ~7); i += 8) {
+#if defined(STBI_SSE2)
+ // load and perform the vertical filtering pass
+ // this uses 3*x + y = 4*x + (y - x)
+ __m128i zero = _mm_setzero_si128();
+ __m128i farb = _mm_loadl_epi64((__m128i *) (in_far + i));
+ __m128i nearb = _mm_loadl_epi64((__m128i *) (in_near + i));
+ __m128i farw = _mm_unpacklo_epi8(farb, zero);
+ __m128i nearw = _mm_unpacklo_epi8(nearb, zero);
+ __m128i diff = _mm_sub_epi16(farw, nearw);
+ __m128i nears = _mm_slli_epi16(nearw, 2);
+ __m128i curr = _mm_add_epi16(nears, diff); // current row
+
+ // horizontal filter works the same based on shifted vers of current
+ // row. "prev" is current row shifted right by 1 pixel; we need to
+ // insert the previous pixel value (from t1).
+ // "next" is current row shifted left by 1 pixel, with first pixel
+ // of next block of 8 pixels added in.
+ __m128i prv0 = _mm_slli_si128(curr, 2);
+ __m128i nxt0 = _mm_srli_si128(curr, 2);
+ __m128i prev = _mm_insert_epi16(prv0, t1, 0);
+ __m128i next = _mm_insert_epi16(nxt0, 3*in_near[i+8] + in_far[i+8], 7);
+
+ // horizontal filter, polyphase implementation since it's convenient:
+ // even pixels = 3*cur + prev = cur*4 + (prev - cur)
+ // odd pixels = 3*cur + next = cur*4 + (next - cur)
+ // note the shared term.
+ __m128i bias = _mm_set1_epi16(8);
+ __m128i curs = _mm_slli_epi16(curr, 2);
+ __m128i prvd = _mm_sub_epi16(prev, curr);
+ __m128i nxtd = _mm_sub_epi16(next, curr);
+ __m128i curb = _mm_add_epi16(curs, bias);
+ __m128i even = _mm_add_epi16(prvd, curb);
+ __m128i odd = _mm_add_epi16(nxtd, curb);
+
+ // interleave even and odd pixels, then undo scaling.
+ __m128i int0 = _mm_unpacklo_epi16(even, odd);
+ __m128i int1 = _mm_unpackhi_epi16(even, odd);
+ __m128i de0 = _mm_srli_epi16(int0, 4);
+ __m128i de1 = _mm_srli_epi16(int1, 4);
+
+ // pack and write output
+ __m128i outv = _mm_packus_epi16(de0, de1);
+ _mm_storeu_si128((__m128i *) (out + i*2), outv);
+#elif defined(STBI_NEON)
+ // load and perform the vertical filtering pass
+ // this uses 3*x + y = 4*x + (y - x)
+ uint8x8_t farb = vld1_u8(in_far + i);
+ uint8x8_t nearb = vld1_u8(in_near + i);
+ int16x8_t diff = vreinterpretq_s16_u16(vsubl_u8(farb, nearb));
+ int16x8_t nears = vreinterpretq_s16_u16(vshll_n_u8(nearb, 2));
+ int16x8_t curr = vaddq_s16(nears, diff); // current row
+
+ // horizontal filter works the same based on shifted vers of current
+ // row. "prev" is current row shifted right by 1 pixel; we need to
+ // insert the previous pixel value (from t1).
+ // "next" is current row shifted left by 1 pixel, with first pixel
+ // of next block of 8 pixels added in.
+ int16x8_t prv0 = vextq_s16(curr, curr, 7);
+ int16x8_t nxt0 = vextq_s16(curr, curr, 1);
+ int16x8_t prev = vsetq_lane_s16(t1, prv0, 0);
+ int16x8_t next = vsetq_lane_s16(3*in_near[i+8] + in_far[i+8], nxt0, 7);
+
+ // horizontal filter, polyphase implementation since it's convenient:
+ // even pixels = 3*cur + prev = cur*4 + (prev - cur)
+ // odd pixels = 3*cur + next = cur*4 + (next - cur)
+ // note the shared term.
+ int16x8_t curs = vshlq_n_s16(curr, 2);
+ int16x8_t prvd = vsubq_s16(prev, curr);
+ int16x8_t nxtd = vsubq_s16(next, curr);
+ int16x8_t even = vaddq_s16(curs, prvd);
+ int16x8_t odd = vaddq_s16(curs, nxtd);
+
+ // undo scaling and round, then store with even/odd phases interleaved
+ uint8x8x2_t o;
+ o.val[0] = vqrshrun_n_s16(even, 4);
+ o.val[1] = vqrshrun_n_s16(odd, 4);
+ vst2_u8(out + i*2, o);
+#endif
+
+ // "previous" value for next iter
+ t1 = 3*in_near[i+7] + in_far[i+7];
+ }
+
+ t0 = t1;
+ t1 = 3*in_near[i] + in_far[i];
+ out[i*2] = stbi__div16(3*t1 + t0 + 8);
+
+ for (++i; i < w; ++i) {
+ t0 = t1;
+ t1 = 3*in_near[i]+in_far[i];
+ out[i*2-1] = stbi__div16(3*t0 + t1 + 8);
+ out[i*2 ] = stbi__div16(3*t1 + t0 + 8);
+ }
+ out[w*2-1] = stbi__div4(t1+2);
+
+ STBI_NOTUSED(hs);
+
+ return out;
+}
+#endif
+
+static stbi_uc *stbi__resample_row_generic(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+{
+ // resample with nearest-neighbor
+ int i,j;
+ STBI_NOTUSED(in_far);
+ for (i=0; i < w; ++i)
+ for (j=0; j < hs; ++j)
+ out[i*hs+j] = in_near[i];
+ return out;
+}
+
+// this is a reduced-precision calculation of YCbCr-to-RGB introduced
+// to make sure the code produces the same results in both SIMD and scalar
+#define stbi__float2fixed(x) (((int) ((x) * 4096.0f + 0.5f)) << 8)
+static void stbi__YCbCr_to_RGB_row(stbi_uc *out, const stbi_uc *y, const stbi_uc *pcb, const stbi_uc *pcr, int count, int step)
+{
+ int i;
+ for (i=0; i < count; ++i) {
+ int y_fixed = (y[i] << 20) + (1<<19); // rounding
+ int r,g,b;
+ int cr = pcr[i] - 128;
+ int cb = pcb[i] - 128;
+ r = y_fixed + cr* stbi__float2fixed(1.40200f);
+ g = y_fixed + (cr*-stbi__float2fixed(0.71414f)) + ((cb*-stbi__float2fixed(0.34414f)) & 0xffff0000);
+ b = y_fixed + cb* stbi__float2fixed(1.77200f);
+ r >>= 20;
+ g >>= 20;
+ b >>= 20;
+ if ((unsigned) r > 255) { if (r < 0) r = 0; else r = 255; }
+ if ((unsigned) g > 255) { if (g < 0) g = 0; else g = 255; }
+ if ((unsigned) b > 255) { if (b < 0) b = 0; else b = 255; }
+ out[0] = (stbi_uc)r;
+ out[1] = (stbi_uc)g;
+ out[2] = (stbi_uc)b;
+ out[3] = 255;
+ out += step;
+ }
+}
+
+#if defined(STBI_SSE2) || defined(STBI_NEON)
+static void stbi__YCbCr_to_RGB_simd(stbi_uc *out, stbi_uc const *y, stbi_uc const *pcb, stbi_uc const *pcr, int count, int step)
+{
+ int i = 0;
+
+#ifdef STBI_SSE2
+ // step == 3 is pretty ugly on the final interleave, and i'm not convinced
+ // it's useful in practice (you wouldn't use it for textures, for example).
+ // so just accelerate step == 4 case.
+ if (step == 4) {
+ // this is a fairly straightforward implementation and not super-optimized.
+ __m128i signflip = _mm_set1_epi8(-0x80);
+ __m128i cr_const0 = _mm_set1_epi16( (short) ( 1.40200f*4096.0f+0.5f));
+ __m128i cr_const1 = _mm_set1_epi16( - (short) ( 0.71414f*4096.0f+0.5f));
+ __m128i cb_const0 = _mm_set1_epi16( - (short) ( 0.34414f*4096.0f+0.5f));
+ __m128i cb_const1 = _mm_set1_epi16( (short) ( 1.77200f*4096.0f+0.5f));
+ __m128i y_bias = _mm_set1_epi8((char) (unsigned char) 128);
+ __m128i xw = _mm_set1_epi16(255); // alpha channel
+
+ for (; i+7 < count; i += 8) {
+ // load
+ __m128i y_bytes = _mm_loadl_epi64((__m128i *) (y+i));
+ __m128i cr_bytes = _mm_loadl_epi64((__m128i *) (pcr+i));
+ __m128i cb_bytes = _mm_loadl_epi64((__m128i *) (pcb+i));
+ __m128i cr_biased = _mm_xor_si128(cr_bytes, signflip); // -128
+ __m128i cb_biased = _mm_xor_si128(cb_bytes, signflip); // -128
+
+ // unpack to short (and left-shift cr, cb by 8)
+ __m128i yw = _mm_unpacklo_epi8(y_bias, y_bytes);
+ __m128i crw = _mm_unpacklo_epi8(_mm_setzero_si128(), cr_biased);
+ __m128i cbw = _mm_unpacklo_epi8(_mm_setzero_si128(), cb_biased);
+
+ // color transform
+ __m128i yws = _mm_srli_epi16(yw, 4);
+ __m128i cr0 = _mm_mulhi_epi16(cr_const0, crw);
+ __m128i cb0 = _mm_mulhi_epi16(cb_const0, cbw);
+ __m128i cb1 = _mm_mulhi_epi16(cbw, cb_const1);
+ __m128i cr1 = _mm_mulhi_epi16(crw, cr_const1);
+ __m128i rws = _mm_add_epi16(cr0, yws);
+ __m128i gwt = _mm_add_epi16(cb0, yws);
+ __m128i bws = _mm_add_epi16(yws, cb1);
+ __m128i gws = _mm_add_epi16(gwt, cr1);
+
+ // descale
+ __m128i rw = _mm_srai_epi16(rws, 4);
+ __m128i bw = _mm_srai_epi16(bws, 4);
+ __m128i gw = _mm_srai_epi16(gws, 4);
+
+ // back to byte, set up for transpose
+ __m128i brb = _mm_packus_epi16(rw, bw);
+ __m128i gxb = _mm_packus_epi16(gw, xw);
+
+ // transpose to interleave channels
+ __m128i t0 = _mm_unpacklo_epi8(brb, gxb);
+ __m128i t1 = _mm_unpackhi_epi8(brb, gxb);
+ __m128i o0 = _mm_unpacklo_epi16(t0, t1);
+ __m128i o1 = _mm_unpackhi_epi16(t0, t1);
+
+ // store
+ _mm_storeu_si128((__m128i *) (out + 0), o0);
+ _mm_storeu_si128((__m128i *) (out + 16), o1);
+ out += 32;
+ }
+ }
+#endif
+
+#ifdef STBI_NEON
+ // in this version, step=3 support would be easy to add. but is there demand?
+ if (step == 4) {
+ // this is a fairly straightforward implementation and not super-optimized.
+ uint8x8_t signflip = vdup_n_u8(0x80);
+ int16x8_t cr_const0 = vdupq_n_s16( (short) ( 1.40200f*4096.0f+0.5f));
+ int16x8_t cr_const1 = vdupq_n_s16( - (short) ( 0.71414f*4096.0f+0.5f));
+ int16x8_t cb_const0 = vdupq_n_s16( - (short) ( 0.34414f*4096.0f+0.5f));
+ int16x8_t cb_const1 = vdupq_n_s16( (short) ( 1.77200f*4096.0f+0.5f));
+
+ for (; i+7 < count; i += 8) {
+ // load
+ uint8x8_t y_bytes = vld1_u8(y + i);
+ uint8x8_t cr_bytes = vld1_u8(pcr + i);
+ uint8x8_t cb_bytes = vld1_u8(pcb + i);
+ int8x8_t cr_biased = vreinterpret_s8_u8(vsub_u8(cr_bytes, signflip));
+ int8x8_t cb_biased = vreinterpret_s8_u8(vsub_u8(cb_bytes, signflip));
+
+ // expand to s16
+ int16x8_t yws = vreinterpretq_s16_u16(vshll_n_u8(y_bytes, 4));
+ int16x8_t crw = vshll_n_s8(cr_biased, 7);
+ int16x8_t cbw = vshll_n_s8(cb_biased, 7);
+
+ // color transform
+ int16x8_t cr0 = vqdmulhq_s16(crw, cr_const0);
+ int16x8_t cb0 = vqdmulhq_s16(cbw, cb_const0);
+ int16x8_t cr1 = vqdmulhq_s16(crw, cr_const1);
+ int16x8_t cb1 = vqdmulhq_s16(cbw, cb_const1);
+ int16x8_t rws = vaddq_s16(yws, cr0);
+ int16x8_t gws = vaddq_s16(vaddq_s16(yws, cb0), cr1);
+ int16x8_t bws = vaddq_s16(yws, cb1);
+
+ // undo scaling, round, convert to byte
+ uint8x8x4_t o;
+ o.val[0] = vqrshrun_n_s16(rws, 4);
+ o.val[1] = vqrshrun_n_s16(gws, 4);
+ o.val[2] = vqrshrun_n_s16(bws, 4);
+ o.val[3] = vdup_n_u8(255);
+
+ // store, interleaving r/g/b/a
+ vst4_u8(out, o);
+ out += 8*4;
+ }
+ }
+#endif
+
+ for (; i < count; ++i) {
+ int y_fixed = (y[i] << 20) + (1<<19); // rounding
+ int r,g,b;
+ int cr = pcr[i] - 128;
+ int cb = pcb[i] - 128;
+ r = y_fixed + cr* stbi__float2fixed(1.40200f);
+ g = y_fixed + cr*-stbi__float2fixed(0.71414f) + ((cb*-stbi__float2fixed(0.34414f)) & 0xffff0000);
+ b = y_fixed + cb* stbi__float2fixed(1.77200f);
+ r >>= 20;
+ g >>= 20;
+ b >>= 20;
+ if ((unsigned) r > 255) { if (r < 0) r = 0; else r = 255; }
+ if ((unsigned) g > 255) { if (g < 0) g = 0; else g = 255; }
+ if ((unsigned) b > 255) { if (b < 0) b = 0; else b = 255; }
+ out[0] = (stbi_uc)r;
+ out[1] = (stbi_uc)g;
+ out[2] = (stbi_uc)b;
+ out[3] = 255;
+ out += step;
+ }
+}
+#endif
+
+// set up the kernels
+static void stbi__setup_jpeg(stbi__jpeg *j)
+{
+ j->idct_block_kernel = stbi__idct_block;
+ j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_row;
+ j->resample_row_hv_2_kernel = stbi__resample_row_hv_2;
+
+#ifdef STBI_SSE2
+ if (stbi__sse2_available()) {
+ j->idct_block_kernel = stbi__idct_simd;
+ j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_simd;
+ j->resample_row_hv_2_kernel = stbi__resample_row_hv_2_simd;
+ }
+#endif
+
+#ifdef STBI_NEON
+ j->idct_block_kernel = stbi__idct_simd;
+ j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_simd;
+ j->resample_row_hv_2_kernel = stbi__resample_row_hv_2_simd;
+#endif
+}
+
+// clean up the temporary component buffers
+static void stbi__cleanup_jpeg(stbi__jpeg *j)
+{
+ stbi__free_jpeg_components(j, j->s->img_n, 0);
+}
+
+typedef struct
+{
+ resample_row_func resample;
+ stbi_uc *line0,*line1;
+ int hs,vs; // expansion factor in each axis
+ int w_lores; // horizontal pixels pre-expansion
+ int ystep; // how far through vertical expansion we are
+ int ypos; // which pre-expansion row we're on
+} stbi__resample;
+
+// fast 0..255 * 0..255 => 0..255 rounded multiplication
+static stbi_uc stbi__blinn_8x8(stbi_uc x, stbi_uc y)
+{
+ unsigned int t = x*y + 128;
+ return (stbi_uc) ((t + (t >>8)) >> 8);
+}
+
+static stbi_uc *load_jpeg_image(stbi__jpeg *z, int *out_x, int *out_y, int *comp, int req_comp)
+{
+ int n, decode_n, is_rgb;
+ z->s->img_n = 0; // make stbi__cleanup_jpeg safe
+
+ // validate req_comp
+ if (req_comp < 0 || req_comp > 4) return stbi__errpuc("bad req_comp", "Internal error");
+
+ // load a jpeg image from whichever source, but leave in YCbCr format
+ if (!stbi__decode_jpeg_image(z)) { stbi__cleanup_jpeg(z); return NULL; }
+
+ // determine actual number of components to generate
+ n = req_comp ? req_comp : z->s->img_n >= 3 ? 3 : 1;
+
+ is_rgb = z->s->img_n == 3 && (z->rgb == 3 || (z->app14_color_transform == 0 && !z->jfif));
+
+ if (z->s->img_n == 3 && n < 3 && !is_rgb)
+ decode_n = 1;
+ else
+ decode_n = z->s->img_n;
+
+ // resample and color-convert
+ {
+ int k;
+ unsigned int i,j;
+ stbi_uc *output;
+ stbi_uc *coutput[4];
+
+ stbi__resample res_comp[4];
+
+ for (k=0; k < decode_n; ++k) {
+ stbi__resample *r = &res_comp[k];
+
+ // allocate line buffer big enough for upsampling off the edges
+ // with upsample factor of 4
+ z->img_comp[k].linebuf = (stbi_uc *) stbi__malloc(z->s->img_x + 3);
+ if (!z->img_comp[k].linebuf) { stbi__cleanup_jpeg(z); return stbi__errpuc("outofmem", "Out of memory"); }
+
+ r->hs = z->img_h_max / z->img_comp[k].h;
+ r->vs = z->img_v_max / z->img_comp[k].v;
+ r->ystep = r->vs >> 1;
+ r->w_lores = (z->s->img_x + r->hs-1) / r->hs;
+ r->ypos = 0;
+ r->line0 = r->line1 = z->img_comp[k].data;
+
+ if (r->hs == 1 && r->vs == 1) r->resample = resample_row_1;
+ else if (r->hs == 1 && r->vs == 2) r->resample = stbi__resample_row_v_2;
+ else if (r->hs == 2 && r->vs == 1) r->resample = stbi__resample_row_h_2;
+ else if (r->hs == 2 && r->vs == 2) r->resample = z->resample_row_hv_2_kernel;
+ else r->resample = stbi__resample_row_generic;
+ }
+
+ // can't error after this so, this is safe
+ output = (stbi_uc *) stbi__malloc_mad3(n, z->s->img_x, z->s->img_y, 1);
+ if (!output) { stbi__cleanup_jpeg(z); return stbi__errpuc("outofmem", "Out of memory"); }
+
+ // now go ahead and resample
+ for (j=0; j < z->s->img_y; ++j) {
+ stbi_uc *out = output + n * z->s->img_x * j;
+ for (k=0; k < decode_n; ++k) {
+ stbi__resample *r = &res_comp[k];
+ int y_bot = r->ystep >= (r->vs >> 1);
+ coutput[k] = r->resample(z->img_comp[k].linebuf,
+ y_bot ? r->line1 : r->line0,
+ y_bot ? r->line0 : r->line1,
+ r->w_lores, r->hs);
+ if (++r->ystep >= r->vs) {
+ r->ystep = 0;
+ r->line0 = r->line1;
+ if (++r->ypos < z->img_comp[k].y)
+ r->line1 += z->img_comp[k].w2;
+ }
+ }
+ if (n >= 3) {
+ stbi_uc *y = coutput[0];
+ if (z->s->img_n == 3) {
+ if (is_rgb) {
+ for (i=0; i < z->s->img_x; ++i) {
+ out[0] = y[i];
+ out[1] = coutput[1][i];
+ out[2] = coutput[2][i];
+ out[3] = 255;
+ out += n;
+ }
+ } else {
+ z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n);
+ }
+ } else if (z->s->img_n == 4) {
+ if (z->app14_color_transform == 0) { // CMYK
+ for (i=0; i < z->s->img_x; ++i) {
+ stbi_uc k = coutput[3][i];
+ out[0] = stbi__blinn_8x8(coutput[0][i], k);
+ out[1] = stbi__blinn_8x8(coutput[1][i], k);
+ out[2] = stbi__blinn_8x8(coutput[2][i], k);
+ out[3] = 255;
+ out += n;
+ }
+ } else if (z->app14_color_transform == 2) { // YCCK
+ z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n);
+ for (i=0; i < z->s->img_x; ++i) {
+ stbi_uc k = coutput[3][i];
+ out[0] = stbi__blinn_8x8(255 - out[0], k);
+ out[1] = stbi__blinn_8x8(255 - out[1], k);
+ out[2] = stbi__blinn_8x8(255 - out[2], k);
+ out += n;
+ }
+ } else { // YCbCr + alpha? Ignore the fourth channel for now
+ z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n);
+ }
+ } else
+ for (i=0; i < z->s->img_x; ++i) {
+ out[0] = out[1] = out[2] = y[i];
+ out[3] = 255; // not used if n==3
+ out += n;
+ }
+ } else {
+ if (is_rgb) {
+ if (n == 1)
+ for (i=0; i < z->s->img_x; ++i)
+ *out++ = stbi__compute_y(coutput[0][i], coutput[1][i], coutput[2][i]);
+ else {
+ for (i=0; i < z->s->img_x; ++i, out += 2) {
+ out[0] = stbi__compute_y(coutput[0][i], coutput[1][i], coutput[2][i]);
+ out[1] = 255;
+ }
+ }
+ } else if (z->s->img_n == 4 && z->app14_color_transform == 0) {
+ for (i=0; i < z->s->img_x; ++i) {
+ stbi_uc k = coutput[3][i];
+ stbi_uc r = stbi__blinn_8x8(coutput[0][i], k);
+ stbi_uc g = stbi__blinn_8x8(coutput[1][i], k);
+ stbi_uc b = stbi__blinn_8x8(coutput[2][i], k);
+ out[0] = stbi__compute_y(r, g, b);
+ out[1] = 255;
+ out += n;
+ }
+ } else if (z->s->img_n == 4 && z->app14_color_transform == 2) {
+ for (i=0; i < z->s->img_x; ++i) {
+ out[0] = stbi__blinn_8x8(255 - coutput[0][i], coutput[3][i]);
+ out[1] = 255;
+ out += n;
+ }
+ } else {
+ stbi_uc *y = coutput[0];
+ if (n == 1)
+ for (i=0; i < z->s->img_x; ++i) out[i] = y[i];
+ else
+ for (i=0; i < z->s->img_x; ++i) *out++ = y[i], *out++ = 255;
+ }
+ }
+ }
+ stbi__cleanup_jpeg(z);
+ *out_x = z->s->img_x;
+ *out_y = z->s->img_y;
+ if (comp) *comp = z->s->img_n >= 3 ? 3 : 1; // report original components, not output
+ return output;
+ }
+}
+
+static void *stbi__jpeg_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+ unsigned char* result;
+ stbi__jpeg* j = (stbi__jpeg*) stbi__malloc(sizeof(stbi__jpeg));
+ STBI_NOTUSED(ri);
+ j->s = s;
+ stbi__setup_jpeg(j);
+ result = load_jpeg_image(j, x,y,comp,req_comp);
+ STBI_FREE(j);
+ return result;
+}
+
+static int stbi__jpeg_test(stbi__context *s)
+{
+ int r;
+ stbi__jpeg* j = (stbi__jpeg*)stbi__malloc(sizeof(stbi__jpeg));
+ j->s = s;
+ stbi__setup_jpeg(j);
+ r = stbi__decode_jpeg_header(j, STBI__SCAN_type);
+ stbi__rewind(s);
+ STBI_FREE(j);
+ return r;
+}
+
+static int stbi__jpeg_info_raw(stbi__jpeg *j, int *x, int *y, int *comp)
+{
+ if (!stbi__decode_jpeg_header(j, STBI__SCAN_header)) {
+ stbi__rewind( j->s );
+ return 0;
+ }
+ if (x) *x = j->s->img_x;
+ if (y) *y = j->s->img_y;
+ if (comp) *comp = j->s->img_n >= 3 ? 3 : 1;
+ return 1;
+}
+
+static int stbi__jpeg_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ int result;
+ stbi__jpeg* j = (stbi__jpeg*) (stbi__malloc(sizeof(stbi__jpeg)));
+ j->s = s;
+ result = stbi__jpeg_info_raw(j, x, y, comp);
+ STBI_FREE(j);
+ return result;
+}
+#endif
+
+// public domain zlib decode v0.2 Sean Barrett 2006-11-18
+// simple implementation
+// - all input must be provided in an upfront buffer
+// - all output is written to a single output buffer (can malloc/realloc)
+// performance
+// - fast huffman
+
+#ifndef STBI_NO_ZLIB
+
+// fast-way is faster to check than jpeg huffman, but slow way is slower
+#define STBI__ZFAST_BITS 9 // accelerate all cases in default tables
+#define STBI__ZFAST_MASK ((1 << STBI__ZFAST_BITS) - 1)
+
+// zlib-style huffman encoding
+// (jpegs packs from left, zlib from right, so can't share code)
+typedef struct
+{
+ stbi__uint16 fast[1 << STBI__ZFAST_BITS];
+ stbi__uint16 firstcode[16];
+ int maxcode[17];
+ stbi__uint16 firstsymbol[16];
+ stbi_uc size[288];
+ stbi__uint16 value[288];
+} stbi__zhuffman;
+
+stbi_inline static int stbi__bitreverse16(int n)
+{
+ n = ((n & 0xAAAA) >> 1) | ((n & 0x5555) << 1);
+ n = ((n & 0xCCCC) >> 2) | ((n & 0x3333) << 2);
+ n = ((n & 0xF0F0) >> 4) | ((n & 0x0F0F) << 4);
+ n = ((n & 0xFF00) >> 8) | ((n & 0x00FF) << 8);
+ return n;
+}
+
+stbi_inline static int stbi__bit_reverse(int v, int bits)
+{
+ STBI_ASSERT(bits <= 16);
+ // to bit reverse n bits, reverse 16 and shift
+ // e.g. 11 bits, bit reverse and shift away 5
+ return stbi__bitreverse16(v) >> (16-bits);
+}
+
+static int stbi__zbuild_huffman(stbi__zhuffman *z, const stbi_uc *sizelist, int num)
+{
+ int i,k=0;
+ int code, next_code[16], sizes[17];
+
+ // DEFLATE spec for generating codes
+ memset(sizes, 0, sizeof(sizes));
+ memset(z->fast, 0, sizeof(z->fast));
+ for (i=0; i < num; ++i)
+ ++sizes[sizelist[i]];
+ sizes[0] = 0;
+ for (i=1; i < 16; ++i)
+ if (sizes[i] > (1 << i))
+ return stbi__err("bad sizes", "Corrupt PNG");
+ code = 0;
+ for (i=1; i < 16; ++i) {
+ next_code[i] = code;
+ z->firstcode[i] = (stbi__uint16) code;
+ z->firstsymbol[i] = (stbi__uint16) k;
+ code = (code + sizes[i]);
+ if (sizes[i])
+ if (code-1 >= (1 << i)) return stbi__err("bad codelengths","Corrupt PNG");
+ z->maxcode[i] = code << (16-i); // preshift for inner loop
+ code <<= 1;
+ k += sizes[i];
+ }
+ z->maxcode[16] = 0x10000; // sentinel
+ for (i=0; i < num; ++i) {
+ int s = sizelist[i];
+ if (s) {
+ int c = next_code[s] - z->firstcode[s] + z->firstsymbol[s];
+ stbi__uint16 fastv = (stbi__uint16) ((s << 9) | i);
+ z->size [c] = (stbi_uc ) s;
+ z->value[c] = (stbi__uint16) i;
+ if (s <= STBI__ZFAST_BITS) {
+ int j = stbi__bit_reverse(next_code[s],s);
+ while (j < (1 << STBI__ZFAST_BITS)) {
+ z->fast[j] = fastv;
+ j += (1 << s);
+ }
+ }
+ ++next_code[s];
+ }
+ }
+ return 1;
+}
+
+// zlib-from-memory implementation for PNG reading
+// because PNG allows splitting the zlib stream arbitrarily,
+// and it's annoying structurally to have PNG call ZLIB call PNG,
+// we require PNG read all the IDATs and combine them into a single
+// memory buffer
+
+typedef struct
+{
+ stbi_uc *zbuffer, *zbuffer_end;
+ int num_bits;
+ stbi__uint32 code_buffer;
+
+ char *zout;
+ char *zout_start;
+ char *zout_end;
+ int z_expandable;
+
+ stbi__zhuffman z_length, z_distance;
+} stbi__zbuf;
+
+stbi_inline static stbi_uc stbi__zget8(stbi__zbuf *z)
+{
+ if (z->zbuffer >= z->zbuffer_end) return 0;
+ return *z->zbuffer++;
+}
+
+static void stbi__fill_bits(stbi__zbuf *z)
+{
+ do {
+ STBI_ASSERT(z->code_buffer < (1U << z->num_bits));
+ z->code_buffer |= (unsigned int) stbi__zget8(z) << z->num_bits;
+ z->num_bits += 8;
+ } while (z->num_bits <= 24);
+}
+
+stbi_inline static unsigned int stbi__zreceive(stbi__zbuf *z, int n)
+{
+ unsigned int k;
+ if (z->num_bits < n) stbi__fill_bits(z);
+ k = z->code_buffer & ((1 << n) - 1);
+ z->code_buffer >>= n;
+ z->num_bits -= n;
+ return k;
+}
+
+static int stbi__zhuffman_decode_slowpath(stbi__zbuf *a, stbi__zhuffman *z)
+{
+ int b,s,k;
+ // not resolved by fast table, so compute it the slow way
+ // use jpeg approach, which requires MSbits at top
+ k = stbi__bit_reverse(a->code_buffer, 16);
+ for (s=STBI__ZFAST_BITS+1; ; ++s)
+ if (k < z->maxcode[s])
+ break;
+ if (s == 16) return -1; // invalid code!
+ // code size is s, so:
+ b = (k >> (16-s)) - z->firstcode[s] + z->firstsymbol[s];
+ STBI_ASSERT(z->size[b] == s);
+ a->code_buffer >>= s;
+ a->num_bits -= s;
+ return z->value[b];
+}
+
+stbi_inline static int stbi__zhuffman_decode(stbi__zbuf *a, stbi__zhuffman *z)
+{
+ int b,s;
+ if (a->num_bits < 16) stbi__fill_bits(a);
+ b = z->fast[a->code_buffer & STBI__ZFAST_MASK];
+ if (b) {
+ s = b >> 9;
+ a->code_buffer >>= s;
+ a->num_bits -= s;
+ return b & 511;
+ }
+ return stbi__zhuffman_decode_slowpath(a, z);
+}
+
+static int stbi__zexpand(stbi__zbuf *z, char *zout, int n) // need to make room for n bytes
+{
+ char *q;
+ int cur, limit, old_limit;
+ z->zout = zout;
+ if (!z->z_expandable) return stbi__err("output buffer limit","Corrupt PNG");
+ cur = (int) (z->zout - z->zout_start);
+ limit = old_limit = (int) (z->zout_end - z->zout_start);
+ while (cur + n > limit)
+ limit *= 2;
+ q = (char *) STBI_REALLOC_SIZED(z->zout_start, old_limit, limit);
+ STBI_NOTUSED(old_limit);
+ if (q == NULL) return stbi__err("outofmem", "Out of memory");
+ z->zout_start = q;
+ z->zout = q + cur;
+ z->zout_end = q + limit;
+ return 1;
+}
+
+static int stbi__zlength_base[31] = {
+ 3,4,5,6,7,8,9,10,11,13,
+ 15,17,19,23,27,31,35,43,51,59,
+ 67,83,99,115,131,163,195,227,258,0,0 };
+
+static int stbi__zlength_extra[31]=
+{ 0,0,0,0,0,0,0,0,1,1,1,1,2,2,2,2,3,3,3,3,4,4,4,4,5,5,5,5,0,0,0 };
+
+static int stbi__zdist_base[32] = { 1,2,3,4,5,7,9,13,17,25,33,49,65,97,129,193,
+257,385,513,769,1025,1537,2049,3073,4097,6145,8193,12289,16385,24577,0,0};
+
+static int stbi__zdist_extra[32] =
+{ 0,0,0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,8,8,9,9,10,10,11,11,12,12,13,13};
+
+static int stbi__parse_huffman_block(stbi__zbuf *a)
+{
+ char *zout = a->zout;
+ for(;;) {
+ int z = stbi__zhuffman_decode(a, &a->z_length);
+ if (z < 256) {
+ if (z < 0) return stbi__err("bad huffman code","Corrupt PNG"); // error in huffman codes
+ if (zout >= a->zout_end) {
+ if (!stbi__zexpand(a, zout, 1)) return 0;
+ zout = a->zout;
+ }
+ *zout++ = (char) z;
+ } else {
+ stbi_uc *p;
+ int len,dist;
+ if (z == 256) {
+ a->zout = zout;
+ return 1;
+ }
+ z -= 257;
+ len = stbi__zlength_base[z];
+ if (stbi__zlength_extra[z]) len += stbi__zreceive(a, stbi__zlength_extra[z]);
+ z = stbi__zhuffman_decode(a, &a->z_distance);
+ if (z < 0) return stbi__err("bad huffman code","Corrupt PNG");
+ dist = stbi__zdist_base[z];
+ if (stbi__zdist_extra[z]) dist += stbi__zreceive(a, stbi__zdist_extra[z]);
+ if (zout - a->zout_start < dist) return stbi__err("bad dist","Corrupt PNG");
+ if (zout + len > a->zout_end) {
+ if (!stbi__zexpand(a, zout, len)) return 0;
+ zout = a->zout;
+ }
+ p = (stbi_uc *) (zout - dist);
+ if (dist == 1) { // run of one byte; common in images.
+ stbi_uc v = *p;
+ if (len) { do *zout++ = v; while (--len); }
+ } else {
+ if (len) { do *zout++ = *p++; while (--len); }
+ }
+ }
+ }
+}
+
+static int stbi__compute_huffman_codes(stbi__zbuf *a)
+{
+ static stbi_uc length_dezigzag[19] = { 16,17,18,0,8,7,9,6,10,5,11,4,12,3,13,2,14,1,15 };
+ stbi__zhuffman z_codelength;
+ stbi_uc lencodes[286+32+137];//padding for maximum single op
+ stbi_uc codelength_sizes[19];
+ int i,n;
+
+ int hlit = stbi__zreceive(a,5) + 257;
+ int hdist = stbi__zreceive(a,5) + 1;
+ int hclen = stbi__zreceive(a,4) + 4;
+ int ntot = hlit + hdist;
+
+ memset(codelength_sizes, 0, sizeof(codelength_sizes));
+ for (i=0; i < hclen; ++i) {
+ int s = stbi__zreceive(a,3);
+ codelength_sizes[length_dezigzag[i]] = (stbi_uc) s;
+ }
+ if (!stbi__zbuild_huffman(&z_codelength, codelength_sizes, 19)) return 0;
+
+ n = 0;
+ while (n < ntot) {
+ int c = stbi__zhuffman_decode(a, &z_codelength);
+ if (c < 0 || c >= 19) return stbi__err("bad codelengths", "Corrupt PNG");
+ if (c < 16)
+ lencodes[n++] = (stbi_uc) c;
+ else {
+ stbi_uc fill = 0;
+ if (c == 16) {
+ c = stbi__zreceive(a,2)+3;
+ if (n == 0) return stbi__err("bad codelengths", "Corrupt PNG");
+ fill = lencodes[n-1];
+ } else if (c == 17)
+ c = stbi__zreceive(a,3)+3;
+ else {
+ STBI_ASSERT(c == 18);
+ c = stbi__zreceive(a,7)+11;
+ }
+ if (ntot - n < c) return stbi__err("bad codelengths", "Corrupt PNG");
+ memset(lencodes+n, fill, c);
+ n += c;
+ }
+ }
+ if (n != ntot) return stbi__err("bad codelengths","Corrupt PNG");
+ if (!stbi__zbuild_huffman(&a->z_length, lencodes, hlit)) return 0;
+ if (!stbi__zbuild_huffman(&a->z_distance, lencodes+hlit, hdist)) return 0;
+ return 1;
+}
+
+static int stbi__parse_uncompressed_block(stbi__zbuf *a)
+{
+ stbi_uc header[4];
+ int len,nlen,k;
+ if (a->num_bits & 7)
+ stbi__zreceive(a, a->num_bits & 7); // discard
+ // drain the bit-packed data into header
+ k = 0;
+ while (a->num_bits > 0) {
+ header[k++] = (stbi_uc) (a->code_buffer & 255); // suppress MSVC run-time check
+ a->code_buffer >>= 8;
+ a->num_bits -= 8;
+ }
+ STBI_ASSERT(a->num_bits == 0);
+ // now fill header the normal way
+ while (k < 4)
+ header[k++] = stbi__zget8(a);
+ len = header[1] * 256 + header[0];
+ nlen = header[3] * 256 + header[2];
+ if (nlen != (len ^ 0xffff)) return stbi__err("zlib corrupt","Corrupt PNG");
+ if (a->zbuffer + len > a->zbuffer_end) return stbi__err("read past buffer","Corrupt PNG");
+ if (a->zout + len > a->zout_end)
+ if (!stbi__zexpand(a, a->zout, len)) return 0;
+ memcpy(a->zout, a->zbuffer, len);
+ a->zbuffer += len;
+ a->zout += len;
+ return 1;
+}
+
+static int stbi__parse_zlib_header(stbi__zbuf *a)
+{
+ int cmf = stbi__zget8(a);
+ int cm = cmf & 15;
+ /* int cinfo = cmf >> 4; */
+ int flg = stbi__zget8(a);
+ if ((cmf*256+flg) % 31 != 0) return stbi__err("bad zlib header","Corrupt PNG"); // zlib spec
+ if (flg & 32) return stbi__err("no preset dict","Corrupt PNG"); // preset dictionary not allowed in png
+ if (cm != 8) return stbi__err("bad compression","Corrupt PNG"); // DEFLATE required for png
+ // window = 1 << (8 + cinfo)... but who cares, we fully buffer output
+ return 1;
+}
+
+static const stbi_uc stbi__zdefault_length[288] =
+{
+ 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
+ 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
+ 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
+ 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
+ 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,
+ 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,
+ 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,
+ 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,
+ 7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7, 7,7,7,7,7,7,7,7,8,8,8,8,8,8,8,8
+};
+static const stbi_uc stbi__zdefault_distance[32] =
+{
+ 5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5
+};
+/*
+Init algorithm:
+{
+ int i; // use <= to match clearly with spec
+ for (i=0; i <= 143; ++i) stbi__zdefault_length[i] = 8;
+ for ( ; i <= 255; ++i) stbi__zdefault_length[i] = 9;
+ for ( ; i <= 279; ++i) stbi__zdefault_length[i] = 7;
+ for ( ; i <= 287; ++i) stbi__zdefault_length[i] = 8;
+
+ for (i=0; i <= 31; ++i) stbi__zdefault_distance[i] = 5;
+}
+*/
+
+static int stbi__parse_zlib(stbi__zbuf *a, int parse_header)
+{
+ int final, type;
+ if (parse_header)
+ if (!stbi__parse_zlib_header(a)) return 0;
+ a->num_bits = 0;
+ a->code_buffer = 0;
+ do {
+ final = stbi__zreceive(a,1);
+ type = stbi__zreceive(a,2);
+ if (type == 0) {
+ if (!stbi__parse_uncompressed_block(a)) return 0;
+ } else if (type == 3) {
+ return 0;
+ } else {
+ if (type == 1) {
+ // use fixed code lengths
+ if (!stbi__zbuild_huffman(&a->z_length , stbi__zdefault_length , 288)) return 0;
+ if (!stbi__zbuild_huffman(&a->z_distance, stbi__zdefault_distance, 32)) return 0;
+ } else {
+ if (!stbi__compute_huffman_codes(a)) return 0;
+ }
+ if (!stbi__parse_huffman_block(a)) return 0;
+ }
+ } while (!final);
+ return 1;
+}
+
+static int stbi__do_zlib(stbi__zbuf *a, char *obuf, int olen, int exp, int parse_header)
+{
+ a->zout_start = obuf;
+ a->zout = obuf;
+ a->zout_end = obuf + olen;
+ a->z_expandable = exp;
+
+ return stbi__parse_zlib(a, parse_header);
+}
+
+STBIDEF char *stbi_zlib_decode_malloc_guesssize(const char *buffer, int len, int initial_size, int *outlen)
+{
+ stbi__zbuf a;
+ char *p = (char *) stbi__malloc(initial_size);
+ if (p == NULL) return NULL;
+ a.zbuffer = (stbi_uc *) buffer;
+ a.zbuffer_end = (stbi_uc *) buffer + len;
+ if (stbi__do_zlib(&a, p, initial_size, 1, 1)) {
+ if (outlen) *outlen = (int) (a.zout - a.zout_start);
+ return a.zout_start;
+ } else {
+ STBI_FREE(a.zout_start);
+ return NULL;
+ }
+}
+
+STBIDEF char *stbi_zlib_decode_malloc(char const *buffer, int len, int *outlen)
+{
+ return stbi_zlib_decode_malloc_guesssize(buffer, len, 16384, outlen);
+}
+
+STBIDEF char *stbi_zlib_decode_malloc_guesssize_headerflag(const char *buffer, int len, int initial_size, int *outlen, int parse_header)
+{
+ stbi__zbuf a;
+ char *p = (char *) stbi__malloc(initial_size);
+ if (p == NULL) return NULL;
+ a.zbuffer = (stbi_uc *) buffer;
+ a.zbuffer_end = (stbi_uc *) buffer + len;
+ if (stbi__do_zlib(&a, p, initial_size, 1, parse_header)) {
+ if (outlen) *outlen = (int) (a.zout - a.zout_start);
+ return a.zout_start;
+ } else {
+ STBI_FREE(a.zout_start);
+ return NULL;
+ }
+}
+
+STBIDEF int stbi_zlib_decode_buffer(char *obuffer, int olen, char const *ibuffer, int ilen)
+{
+ stbi__zbuf a;
+ a.zbuffer = (stbi_uc *) ibuffer;
+ a.zbuffer_end = (stbi_uc *) ibuffer + ilen;
+ if (stbi__do_zlib(&a, obuffer, olen, 0, 1))
+ return (int) (a.zout - a.zout_start);
+ else
+ return -1;
+}
+
+STBIDEF char *stbi_zlib_decode_noheader_malloc(char const *buffer, int len, int *outlen)
+{
+ stbi__zbuf a;
+ char *p = (char *) stbi__malloc(16384);
+ if (p == NULL) return NULL;
+ a.zbuffer = (stbi_uc *) buffer;
+ a.zbuffer_end = (stbi_uc *) buffer+len;
+ if (stbi__do_zlib(&a, p, 16384, 1, 0)) {
+ if (outlen) *outlen = (int) (a.zout - a.zout_start);
+ return a.zout_start;
+ } else {
+ STBI_FREE(a.zout_start);
+ return NULL;
+ }
+}
+
+STBIDEF int stbi_zlib_decode_noheader_buffer(char *obuffer, int olen, const char *ibuffer, int ilen)
+{
+ stbi__zbuf a;
+ a.zbuffer = (stbi_uc *) ibuffer;
+ a.zbuffer_end = (stbi_uc *) ibuffer + ilen;
+ if (stbi__do_zlib(&a, obuffer, olen, 0, 0))
+ return (int) (a.zout - a.zout_start);
+ else
+ return -1;
+}
+#endif
+
+// public domain "baseline" PNG decoder v0.10 Sean Barrett 2006-11-18
+// simple implementation
+// - only 8-bit samples
+// - no CRC checking
+// - allocates lots of intermediate memory
+// - avoids problem of streaming data between subsystems
+// - avoids explicit window management
+// performance
+// - uses stb_zlib, a PD zlib implementation with fast huffman decoding
+
+#ifndef STBI_NO_PNG
+typedef struct
+{
+ stbi__uint32 length;
+ stbi__uint32 type;
+} stbi__pngchunk;
+
+static stbi__pngchunk stbi__get_chunk_header(stbi__context *s)
+{
+ stbi__pngchunk c;
+ c.length = stbi__get32be(s);
+ c.type = stbi__get32be(s);
+ return c;
+}
+
+static int stbi__check_png_header(stbi__context *s)
+{
+ static stbi_uc png_sig[8] = { 137,80,78,71,13,10,26,10 };
+ int i;
+ for (i=0; i < 8; ++i)
+ if (stbi__get8(s) != png_sig[i]) return stbi__err("bad png sig","Not a PNG");
+ return 1;
+}
+
+typedef struct
+{
+ stbi__context *s;
+ stbi_uc *idata, *expanded, *out;
+ int depth;
+} stbi__png;
+
+
+enum {
+ STBI__F_none=0,
+ STBI__F_sub=1,
+ STBI__F_up=2,
+ STBI__F_avg=3,
+ STBI__F_paeth=4,
+ // synthetic filters used for first scanline to avoid needing a dummy row of 0s
+ STBI__F_avg_first,
+ STBI__F_paeth_first
+};
+
+static stbi_uc first_row_filter[5] =
+{
+ STBI__F_none,
+ STBI__F_sub,
+ STBI__F_none,
+ STBI__F_avg_first,
+ STBI__F_paeth_first
+};
+
+static int stbi__paeth(int a, int b, int c)
+{
+ int p = a + b - c;
+ int pa = abs(p-a);
+ int pb = abs(p-b);
+ int pc = abs(p-c);
+ if (pa <= pb && pa <= pc) return a;
+ if (pb <= pc) return b;
+ return c;
+}
+
+static stbi_uc stbi__depth_scale_table[9] = { 0, 0xff, 0x55, 0, 0x11, 0,0,0, 0x01 };
+
+// create the png data from post-deflated data
+static int stbi__create_png_image_raw(stbi__png *a, stbi_uc *raw, stbi__uint32 raw_len, int out_n, stbi__uint32 x, stbi__uint32 y, int depth, int color)
+{
+ int bytes = (depth == 16? 2 : 1);
+ stbi__context *s = a->s;
+ stbi__uint32 i,j,stride = x*out_n*bytes;
+ stbi__uint32 img_len, img_width_bytes;
+ int k;
+ int img_n = s->img_n; // copy it into a local for later
+
+ int output_bytes = out_n*bytes;
+ int filter_bytes = img_n*bytes;
+ int width = x;
+
+ STBI_ASSERT(out_n == s->img_n || out_n == s->img_n+1);
+ a->out = (stbi_uc *) stbi__malloc_mad3(x, y, output_bytes, 0); // extra bytes to write off the end into
+ if (!a->out) return stbi__err("outofmem", "Out of memory");
+
+ img_width_bytes = (((img_n * x * depth) + 7) >> 3);
+ img_len = (img_width_bytes + 1) * y;
+ if (s->img_x == x && s->img_y == y) {
+ if (raw_len != img_len) return stbi__err("not enough pixels","Corrupt PNG");
+ } else { // interlaced:
+ if (raw_len < img_len) return stbi__err("not enough pixels","Corrupt PNG");
+ }
+
+ for (j=0; j < y; ++j) {
+ stbi_uc *cur = a->out + stride*j;
+ stbi_uc *prior;
+ int filter = *raw++;
+
+ if (filter > 4)
+ return stbi__err("invalid filter","Corrupt PNG");
+
+ if (depth < 8) {
+ STBI_ASSERT(img_width_bytes <= x);
+ cur += x*out_n - img_width_bytes; // store output to the rightmost img_len bytes, so we can decode in place
+ filter_bytes = 1;
+ width = img_width_bytes;
+ }
+ prior = cur - stride; // bugfix: need to compute this after 'cur +=' computation above
+
+ // if first row, use special filter that doesn't sample previous row
+ if (j == 0) filter = first_row_filter[filter];
+
+ // handle first byte explicitly
+ for (k=0; k < filter_bytes; ++k) {
+ switch (filter) {
+ case STBI__F_none : cur[k] = raw[k]; break;
+ case STBI__F_sub : cur[k] = raw[k]; break;
+ case STBI__F_up : cur[k] = STBI__BYTECAST(raw[k] + prior[k]); break;
+ case STBI__F_avg : cur[k] = STBI__BYTECAST(raw[k] + (prior[k]>>1)); break;
+ case STBI__F_paeth : cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(0,prior[k],0)); break;
+ case STBI__F_avg_first : cur[k] = raw[k]; break;
+ case STBI__F_paeth_first: cur[k] = raw[k]; break;
+ }
+ }
+
+ if (depth == 8) {
+ if (img_n != out_n)
+ cur[img_n] = 255; // first pixel
+ raw += img_n;
+ cur += out_n;
+ prior += out_n;
+ } else if (depth == 16) {
+ if (img_n != out_n) {
+ cur[filter_bytes] = 255; // first pixel top byte
+ cur[filter_bytes+1] = 255; // first pixel bottom byte
+ }
+ raw += filter_bytes;
+ cur += output_bytes;
+ prior += output_bytes;
+ } else {
+ raw += 1;
+ cur += 1;
+ prior += 1;
+ }
+
+ // this is a little gross, so that we don't switch per-pixel or per-component
+ if (depth < 8 || img_n == out_n) {
+ int nk = (width - 1)*filter_bytes;
+ #define STBI__CASE(f) \
+ case f: \
+ for (k=0; k < nk; ++k)
+ switch (filter) {
+ // "none" filter turns into a memcpy here; make that explicit.
+ case STBI__F_none: memcpy(cur, raw, nk); break;
+ STBI__CASE(STBI__F_sub) { cur[k] = STBI__BYTECAST(raw[k] + cur[k-filter_bytes]); } break;
+ STBI__CASE(STBI__F_up) { cur[k] = STBI__BYTECAST(raw[k] + prior[k]); } break;
+ STBI__CASE(STBI__F_avg) { cur[k] = STBI__BYTECAST(raw[k] + ((prior[k] + cur[k-filter_bytes])>>1)); } break;
+ STBI__CASE(STBI__F_paeth) { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k-filter_bytes],prior[k],prior[k-filter_bytes])); } break;
+ STBI__CASE(STBI__F_avg_first) { cur[k] = STBI__BYTECAST(raw[k] + (cur[k-filter_bytes] >> 1)); } break;
+ STBI__CASE(STBI__F_paeth_first) { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k-filter_bytes],0,0)); } break;
+ }
+ #undef STBI__CASE
+ raw += nk;
+ } else {
+ STBI_ASSERT(img_n+1 == out_n);
+ #define STBI__CASE(f) \
+ case f: \
+ for (i=x-1; i >= 1; --i, cur[filter_bytes]=255,raw+=filter_bytes,cur+=output_bytes,prior+=output_bytes) \
+ for (k=0; k < filter_bytes; ++k)
+ switch (filter) {
+ STBI__CASE(STBI__F_none) { cur[k] = raw[k]; } break;
+ STBI__CASE(STBI__F_sub) { cur[k] = STBI__BYTECAST(raw[k] + cur[k- output_bytes]); } break;
+ STBI__CASE(STBI__F_up) { cur[k] = STBI__BYTECAST(raw[k] + prior[k]); } break;
+ STBI__CASE(STBI__F_avg) { cur[k] = STBI__BYTECAST(raw[k] + ((prior[k] + cur[k- output_bytes])>>1)); } break;
+ STBI__CASE(STBI__F_paeth) { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k- output_bytes],prior[k],prior[k- output_bytes])); } break;
+ STBI__CASE(STBI__F_avg_first) { cur[k] = STBI__BYTECAST(raw[k] + (cur[k- output_bytes] >> 1)); } break;
+ STBI__CASE(STBI__F_paeth_first) { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k- output_bytes],0,0)); } break;
+ }
+ #undef STBI__CASE
+
+ // the loop above sets the high byte of the pixels' alpha, but for
+ // 16 bit png files we also need the low byte set. we'll do that here.
+ if (depth == 16) {
+ cur = a->out + stride*j; // start at the beginning of the row again
+ for (i=0; i < x; ++i,cur+=output_bytes) {
+ cur[filter_bytes+1] = 255;
+ }
+ }
+ }
+ }
+
+ // we make a separate pass to expand bits to pixels; for performance,
+ // this could run two scanlines behind the above code, so it won't
+ // intefere with filtering but will still be in the cache.
+ if (depth < 8) {
+ for (j=0; j < y; ++j) {
+ stbi_uc *cur = a->out + stride*j;
+ stbi_uc *in = a->out + stride*j + x*out_n - img_width_bytes;
+ // unpack 1/2/4-bit into a 8-bit buffer. allows us to keep the common 8-bit path optimal at minimal cost for 1/2/4-bit
+ // png guarante byte alignment, if width is not multiple of 8/4/2 we'll decode dummy trailing data that will be skipped in the later loop
+ stbi_uc scale = (color == 0) ? stbi__depth_scale_table[depth] : 1; // scale grayscale values to 0..255 range
+
+ // note that the final byte might overshoot and write more data than desired.
+ // we can allocate enough data that this never writes out of memory, but it
+ // could also overwrite the next scanline. can it overwrite non-empty data
+ // on the next scanline? yes, consider 1-pixel-wide scanlines with 1-bit-per-pixel.
+ // so we need to explicitly clamp the final ones
+
+ if (depth == 4) {
+ for (k=x*img_n; k >= 2; k-=2, ++in) {
+ *cur++ = scale * ((*in >> 4) );
+ *cur++ = scale * ((*in ) & 0x0f);
+ }
+ if (k > 0) *cur++ = scale * ((*in >> 4) );
+ } else if (depth == 2) {
+ for (k=x*img_n; k >= 4; k-=4, ++in) {
+ *cur++ = scale * ((*in >> 6) );
+ *cur++ = scale * ((*in >> 4) & 0x03);
+ *cur++ = scale * ((*in >> 2) & 0x03);
+ *cur++ = scale * ((*in ) & 0x03);
+ }
+ if (k > 0) *cur++ = scale * ((*in >> 6) );
+ if (k > 1) *cur++ = scale * ((*in >> 4) & 0x03);
+ if (k > 2) *cur++ = scale * ((*in >> 2) & 0x03);
+ } else if (depth == 1) {
+ for (k=x*img_n; k >= 8; k-=8, ++in) {
+ *cur++ = scale * ((*in >> 7) );
+ *cur++ = scale * ((*in >> 6) & 0x01);
+ *cur++ = scale * ((*in >> 5) & 0x01);
+ *cur++ = scale * ((*in >> 4) & 0x01);
+ *cur++ = scale * ((*in >> 3) & 0x01);
+ *cur++ = scale * ((*in >> 2) & 0x01);
+ *cur++ = scale * ((*in >> 1) & 0x01);
+ *cur++ = scale * ((*in ) & 0x01);
+ }
+ if (k > 0) *cur++ = scale * ((*in >> 7) );
+ if (k > 1) *cur++ = scale * ((*in >> 6) & 0x01);
+ if (k > 2) *cur++ = scale * ((*in >> 5) & 0x01);
+ if (k > 3) *cur++ = scale * ((*in >> 4) & 0x01);
+ if (k > 4) *cur++ = scale * ((*in >> 3) & 0x01);
+ if (k > 5) *cur++ = scale * ((*in >> 2) & 0x01);
+ if (k > 6) *cur++ = scale * ((*in >> 1) & 0x01);
+ }
+ if (img_n != out_n) {
+ int q;
+ // insert alpha = 255
+ cur = a->out + stride*j;
+ if (img_n == 1) {
+ for (q=x-1; q >= 0; --q) {
+ cur[q*2+1] = 255;
+ cur[q*2+0] = cur[q];
+ }
+ } else {
+ STBI_ASSERT(img_n == 3);
+ for (q=x-1; q >= 0; --q) {
+ cur[q*4+3] = 255;
+ cur[q*4+2] = cur[q*3+2];
+ cur[q*4+1] = cur[q*3+1];
+ cur[q*4+0] = cur[q*3+0];
+ }
+ }
+ }
+ }
+ } else if (depth == 16) {
+ // force the image data from big-endian to platform-native.
+ // this is done in a separate pass due to the decoding relying
+ // on the data being untouched, but could probably be done
+ // per-line during decode if care is taken.
+ stbi_uc *cur = a->out;
+ stbi__uint16 *cur16 = (stbi__uint16*)cur;
+
+ for(i=0; i < x*y*out_n; ++i,cur16++,cur+=2) {
+ *cur16 = (cur[0] << 8) | cur[1];
+ }
+ }
+
+ return 1;
+}
+
+static int stbi__create_png_image(stbi__png *a, stbi_uc *image_data, stbi__uint32 image_data_len, int out_n, int depth, int color, int interlaced)
+{
+ int bytes = (depth == 16 ? 2 : 1);
+ int out_bytes = out_n * bytes;
+ stbi_uc *final;
+ int p;
+ if (!interlaced)
+ return stbi__create_png_image_raw(a, image_data, image_data_len, out_n, a->s->img_x, a->s->img_y, depth, color);
+
+ // de-interlacing
+ final = (stbi_uc *) stbi__malloc_mad3(a->s->img_x, a->s->img_y, out_bytes, 0);
+ for (p=0; p < 7; ++p) {
+ int xorig[] = { 0,4,0,2,0,1,0 };
+ int yorig[] = { 0,0,4,0,2,0,1 };
+ int xspc[] = { 8,8,4,4,2,2,1 };
+ int yspc[] = { 8,8,8,4,4,2,2 };
+ int i,j,x,y;
+ // pass1_x[4] = 0, pass1_x[5] = 1, pass1_x[12] = 1
+ x = (a->s->img_x - xorig[p] + xspc[p]-1) / xspc[p];
+ y = (a->s->img_y - yorig[p] + yspc[p]-1) / yspc[p];
+ if (x && y) {
+ stbi__uint32 img_len = ((((a->s->img_n * x * depth) + 7) >> 3) + 1) * y;
+ if (!stbi__create_png_image_raw(a, image_data, image_data_len, out_n, x, y, depth, color)) {
+ STBI_FREE(final);
+ return 0;
+ }
+ for (j=0; j < y; ++j) {
+ for (i=0; i < x; ++i) {
+ int out_y = j*yspc[p]+yorig[p];
+ int out_x = i*xspc[p]+xorig[p];
+ memcpy(final + out_y*a->s->img_x*out_bytes + out_x*out_bytes,
+ a->out + (j*x+i)*out_bytes, out_bytes);
+ }
+ }
+ STBI_FREE(a->out);
+ image_data += img_len;
+ image_data_len -= img_len;
+ }
+ }
+ a->out = final;
+
+ return 1;
+}
+
+static int stbi__compute_transparency(stbi__png *z, stbi_uc tc[3], int out_n)
+{
+ stbi__context *s = z->s;
+ stbi__uint32 i, pixel_count = s->img_x * s->img_y;
+ stbi_uc *p = z->out;
+
+ // compute color-based transparency, assuming we've
+ // already got 255 as the alpha value in the output
+ STBI_ASSERT(out_n == 2 || out_n == 4);
+
+ if (out_n == 2) {
+ for (i=0; i < pixel_count; ++i) {
+ p[1] = (p[0] == tc[0] ? 0 : 255);
+ p += 2;
+ }
+ } else {
+ for (i=0; i < pixel_count; ++i) {
+ if (p[0] == tc[0] && p[1] == tc[1] && p[2] == tc[2])
+ p[3] = 0;
+ p += 4;
+ }
+ }
+ return 1;
+}
+
+static int stbi__compute_transparency16(stbi__png *z, stbi__uint16 tc[3], int out_n)
+{
+ stbi__context *s = z->s;
+ stbi__uint32 i, pixel_count = s->img_x * s->img_y;
+ stbi__uint16 *p = (stbi__uint16*) z->out;
+
+ // compute color-based transparency, assuming we've
+ // already got 65535 as the alpha value in the output
+ STBI_ASSERT(out_n == 2 || out_n == 4);
+
+ if (out_n == 2) {
+ for (i = 0; i < pixel_count; ++i) {
+ p[1] = (p[0] == tc[0] ? 0 : 65535);
+ p += 2;
+ }
+ } else {
+ for (i = 0; i < pixel_count; ++i) {
+ if (p[0] == tc[0] && p[1] == tc[1] && p[2] == tc[2])
+ p[3] = 0;
+ p += 4;
+ }
+ }
+ return 1;
+}
+
+static int stbi__expand_png_palette(stbi__png *a, stbi_uc *palette, int len, int pal_img_n)
+{
+ stbi__uint32 i, pixel_count = a->s->img_x * a->s->img_y;
+ stbi_uc *p, *temp_out, *orig = a->out;
+
+ p = (stbi_uc *) stbi__malloc_mad2(pixel_count, pal_img_n, 0);
+ if (p == NULL) return stbi__err("outofmem", "Out of memory");
+
+ // between here and free(out) below, exitting would leak
+ temp_out = p;
+
+ if (pal_img_n == 3) {
+ for (i=0; i < pixel_count; ++i) {
+ int n = orig[i]*4;
+ p[0] = palette[n ];
+ p[1] = palette[n+1];
+ p[2] = palette[n+2];
+ p += 3;
+ }
+ } else {
+ for (i=0; i < pixel_count; ++i) {
+ int n = orig[i]*4;
+ p[0] = palette[n ];
+ p[1] = palette[n+1];
+ p[2] = palette[n+2];
+ p[3] = palette[n+3];
+ p += 4;
+ }
+ }
+ STBI_FREE(a->out);
+ a->out = temp_out;
+
+ STBI_NOTUSED(len);
+
+ return 1;
+}
+
+static int stbi__unpremultiply_on_load = 0;
+static int stbi__de_iphone_flag = 0;
+
+STBIDEF void stbi_set_unpremultiply_on_load(int flag_true_if_should_unpremultiply)
+{
+ stbi__unpremultiply_on_load = flag_true_if_should_unpremultiply;
+}
+
+STBIDEF void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert)
+{
+ stbi__de_iphone_flag = flag_true_if_should_convert;
+}
+
+static void stbi__de_iphone(stbi__png *z)
+{
+ stbi__context *s = z->s;
+ stbi__uint32 i, pixel_count = s->img_x * s->img_y;
+ stbi_uc *p = z->out;
+
+ if (s->img_out_n == 3) { // convert bgr to rgb
+ for (i=0; i < pixel_count; ++i) {
+ stbi_uc t = p[0];
+ p[0] = p[2];
+ p[2] = t;
+ p += 3;
+ }
+ } else {
+ STBI_ASSERT(s->img_out_n == 4);
+ if (stbi__unpremultiply_on_load) {
+ // convert bgr to rgb and unpremultiply
+ for (i=0; i < pixel_count; ++i) {
+ stbi_uc a = p[3];
+ stbi_uc t = p[0];
+ if (a) {
+ p[0] = p[2] * 255 / a;
+ p[1] = p[1] * 255 / a;
+ p[2] = t * 255 / a;
+ } else {
+ p[0] = p[2];
+ p[2] = t;
+ }
+ p += 4;
+ }
+ } else {
+ // convert bgr to rgb
+ for (i=0; i < pixel_count; ++i) {
+ stbi_uc t = p[0];
+ p[0] = p[2];
+ p[2] = t;
+ p += 4;
+ }
+ }
+ }
+}
+
+#define STBI__PNG_TYPE(a,b,c,d) (((a) << 24) + ((b) << 16) + ((c) << 8) + (d))
+
+static int stbi__parse_png_file(stbi__png *z, int scan, int req_comp)
+{
+ stbi_uc palette[1024], pal_img_n=0;
+ stbi_uc has_trans=0, tc[3];
+ stbi__uint16 tc16[3];
+ stbi__uint32 ioff=0, idata_limit=0, i, pal_len=0;
+ int first=1,k,interlace=0, color=0, is_iphone=0;
+ stbi__context *s = z->s;
+
+ z->expanded = NULL;
+ z->idata = NULL;
+ z->out = NULL;
+
+ if (!stbi__check_png_header(s)) return 0;
+
+ if (scan == STBI__SCAN_type) return 1;
+
+ for (;;) {
+ stbi__pngchunk c = stbi__get_chunk_header(s);
+ switch (c.type) {
+ case STBI__PNG_TYPE('C','g','B','I'):
+ is_iphone = 1;
+ stbi__skip(s, c.length);
+ break;
+ case STBI__PNG_TYPE('I','H','D','R'): {
+ int comp,filter;
+ if (!first) return stbi__err("multiple IHDR","Corrupt PNG");
+ first = 0;
+ if (c.length != 13) return stbi__err("bad IHDR len","Corrupt PNG");
+ s->img_x = stbi__get32be(s); if (s->img_x > (1 << 24)) return stbi__err("too large","Very large image (corrupt?)");
+ s->img_y = stbi__get32be(s); if (s->img_y > (1 << 24)) return stbi__err("too large","Very large image (corrupt?)");
+ z->depth = stbi__get8(s); if (z->depth != 1 && z->depth != 2 && z->depth != 4 && z->depth != 8 && z->depth != 16) return stbi__err("1/2/4/8/16-bit only","PNG not supported: 1/2/4/8/16-bit only");
+ color = stbi__get8(s); if (color > 6) return stbi__err("bad ctype","Corrupt PNG");
+ if (color == 3 && z->depth == 16) return stbi__err("bad ctype","Corrupt PNG");
+ if (color == 3) pal_img_n = 3; else if (color & 1) return stbi__err("bad ctype","Corrupt PNG");
+ comp = stbi__get8(s); if (comp) return stbi__err("bad comp method","Corrupt PNG");
+ filter= stbi__get8(s); if (filter) return stbi__err("bad filter method","Corrupt PNG");
+ interlace = stbi__get8(s); if (interlace>1) return stbi__err("bad interlace method","Corrupt PNG");
+ if (!s->img_x || !s->img_y) return stbi__err("0-pixel image","Corrupt PNG");
+ if (!pal_img_n) {
+ s->img_n = (color & 2 ? 3 : 1) + (color & 4 ? 1 : 0);
+ if ((1 << 30) / s->img_x / s->img_n < s->img_y) return stbi__err("too large", "Image too large to decode");
+ if (scan == STBI__SCAN_header) return 1;
+ } else {
+ // if paletted, then pal_n is our final components, and
+ // img_n is # components to decompress/filter.
+ s->img_n = 1;
+ if ((1 << 30) / s->img_x / 4 < s->img_y) return stbi__err("too large","Corrupt PNG");
+ // if SCAN_header, have to scan to see if we have a tRNS
+ }
+ break;
+ }
+
+ case STBI__PNG_TYPE('P','L','T','E'): {
+ if (first) return stbi__err("first not IHDR", "Corrupt PNG");
+ if (c.length > 256*3) return stbi__err("invalid PLTE","Corrupt PNG");
+ pal_len = c.length / 3;
+ if (pal_len * 3 != c.length) return stbi__err("invalid PLTE","Corrupt PNG");
+ for (i=0; i < pal_len; ++i) {
+ palette[i*4+0] = stbi__get8(s);
+ palette[i*4+1] = stbi__get8(s);
+ palette[i*4+2] = stbi__get8(s);
+ palette[i*4+3] = 255;
+ }
+ break;
+ }
+
+ case STBI__PNG_TYPE('t','R','N','S'): {
+ if (first) return stbi__err("first not IHDR", "Corrupt PNG");
+ if (z->idata) return stbi__err("tRNS after IDAT","Corrupt PNG");
+ if (pal_img_n) {
+ if (scan == STBI__SCAN_header) { s->img_n = 4; return 1; }
+ if (pal_len == 0) return stbi__err("tRNS before PLTE","Corrupt PNG");
+ if (c.length > pal_len) return stbi__err("bad tRNS len","Corrupt PNG");
+ pal_img_n = 4;
+ for (i=0; i < c.length; ++i)
+ palette[i*4+3] = stbi__get8(s);
+ } else {
+ if (!(s->img_n & 1)) return stbi__err("tRNS with alpha","Corrupt PNG");
+ if (c.length != (stbi__uint32) s->img_n*2) return stbi__err("bad tRNS len","Corrupt PNG");
+ has_trans = 1;
+ if (z->depth == 16) {
+ for (k = 0; k < s->img_n; ++k) tc16[k] = (stbi__uint16)stbi__get16be(s); // copy the values as-is
+ } else {
+ for (k = 0; k < s->img_n; ++k) tc[k] = (stbi_uc)(stbi__get16be(s) & 255) * stbi__depth_scale_table[z->depth]; // non 8-bit images will be larger
+ }
+ }
+ break;
+ }
+
+ case STBI__PNG_TYPE('I','D','A','T'): {
+ if (first) return stbi__err("first not IHDR", "Corrupt PNG");
+ if (pal_img_n && !pal_len) return stbi__err("no PLTE","Corrupt PNG");
+ if (scan == STBI__SCAN_header) { s->img_n = pal_img_n; return 1; }
+ if ((int)(ioff + c.length) < (int)ioff) return 0;
+ if (ioff + c.length > idata_limit) {
+ stbi__uint32 idata_limit_old = idata_limit;
+ stbi_uc *p;
+ if (idata_limit == 0) idata_limit = c.length > 4096 ? c.length : 4096;
+ while (ioff + c.length > idata_limit)
+ idata_limit *= 2;
+ STBI_NOTUSED(idata_limit_old);
+ p = (stbi_uc *) STBI_REALLOC_SIZED(z->idata, idata_limit_old, idata_limit); if (p == NULL) return stbi__err("outofmem", "Out of memory");
+ z->idata = p;
+ }
+ if (!stbi__getn(s, z->idata+ioff,c.length)) return stbi__err("outofdata","Corrupt PNG");
+ ioff += c.length;
+ break;
+ }
+
+ case STBI__PNG_TYPE('I','E','N','D'): {
+ stbi__uint32 raw_len, bpl;
+ if (first) return stbi__err("first not IHDR", "Corrupt PNG");
+ if (scan != STBI__SCAN_load) return 1;
+ if (z->idata == NULL) return stbi__err("no IDAT","Corrupt PNG");
+ // initial guess for decoded data size to avoid unnecessary reallocs
+ bpl = (s->img_x * z->depth + 7) / 8; // bytes per line, per component
+ raw_len = bpl * s->img_y * s->img_n /* pixels */ + s->img_y /* filter mode per row */;
+ z->expanded = (stbi_uc *) stbi_zlib_decode_malloc_guesssize_headerflag((char *) z->idata, ioff, raw_len, (int *) &raw_len, !is_iphone);
+ if (z->expanded == NULL) return 0; // zlib should set error
+ STBI_FREE(z->idata); z->idata = NULL;
+ if ((req_comp == s->img_n+1 && req_comp != 3 && !pal_img_n) || has_trans)
+ s->img_out_n = s->img_n+1;
+ else
+ s->img_out_n = s->img_n;
+ if (!stbi__create_png_image(z, z->expanded, raw_len, s->img_out_n, z->depth, color, interlace)) return 0;
+ if (has_trans) {
+ if (z->depth == 16) {
+ if (!stbi__compute_transparency16(z, tc16, s->img_out_n)) return 0;
+ } else {
+ if (!stbi__compute_transparency(z, tc, s->img_out_n)) return 0;
+ }
+ }
+ if (is_iphone && stbi__de_iphone_flag && s->img_out_n > 2)
+ stbi__de_iphone(z);
+ if (pal_img_n) {
+ // pal_img_n == 3 or 4
+ s->img_n = pal_img_n; // record the actual colors we had
+ s->img_out_n = pal_img_n;
+ if (req_comp >= 3) s->img_out_n = req_comp;
+ if (!stbi__expand_png_palette(z, palette, pal_len, s->img_out_n))
+ return 0;
+ }
+ STBI_FREE(z->expanded); z->expanded = NULL;
+ return 1;
+ }
+
+ default:
+ // if critical, fail
+ if (first) return stbi__err("first not IHDR", "Corrupt PNG");
+ if ((c.type & (1 << 29)) == 0) {
+ #ifndef STBI_NO_FAILURE_STRINGS
+ // not threadsafe
+ static char invalid_chunk[] = "XXXX PNG chunk not known";
+ invalid_chunk[0] = STBI__BYTECAST(c.type >> 24);
+ invalid_chunk[1] = STBI__BYTECAST(c.type >> 16);
+ invalid_chunk[2] = STBI__BYTECAST(c.type >> 8);
+ invalid_chunk[3] = STBI__BYTECAST(c.type >> 0);
+ #endif
+ return stbi__err(invalid_chunk, "PNG not supported: unknown PNG chunk type");
+ }
+ stbi__skip(s, c.length);
+ break;
+ }
+ // end of PNG chunk, read and skip CRC
+ stbi__get32be(s);
+ }
+}
+
+static void *stbi__do_png(stbi__png *p, int *x, int *y, int *n, int req_comp, stbi__result_info *ri)
+{
+ void *result=NULL;
+ if (req_comp < 0 || req_comp > 4) return stbi__errpuc("bad req_comp", "Internal error");
+ if (stbi__parse_png_file(p, STBI__SCAN_load, req_comp)) {
+ if (p->depth < 8)
+ ri->bits_per_channel = 8;
+ else
+ ri->bits_per_channel = p->depth;
+ result = p->out;
+ p->out = NULL;
+ if (req_comp && req_comp != p->s->img_out_n) {
+ if (ri->bits_per_channel == 8)
+ result = stbi__convert_format((unsigned char *) result, p->s->img_out_n, req_comp, p->s->img_x, p->s->img_y);
+ else
+ result = stbi__convert_format16((stbi__uint16 *) result, p->s->img_out_n, req_comp, p->s->img_x, p->s->img_y);
+ p->s->img_out_n = req_comp;
+ if (result == NULL) return result;
+ }
+ *x = p->s->img_x;
+ *y = p->s->img_y;
+ if (n) *n = p->s->img_n;
+ }
+ STBI_FREE(p->out); p->out = NULL;
+ STBI_FREE(p->expanded); p->expanded = NULL;
+ STBI_FREE(p->idata); p->idata = NULL;
+
+ return result;
+}
+
+static void *stbi__png_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+ stbi__png p;
+ p.s = s;
+ return stbi__do_png(&p, x,y,comp,req_comp, ri);
+}
+
+static int stbi__png_test(stbi__context *s)
+{
+ int r;
+ r = stbi__check_png_header(s);
+ stbi__rewind(s);
+ return r;
+}
+
+static int stbi__png_info_raw(stbi__png *p, int *x, int *y, int *comp)
+{
+ if (!stbi__parse_png_file(p, STBI__SCAN_header, 0)) {
+ stbi__rewind( p->s );
+ return 0;
+ }
+ if (x) *x = p->s->img_x;
+ if (y) *y = p->s->img_y;
+ if (comp) *comp = p->s->img_n;
+ return 1;
+}
+
+static int stbi__png_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ stbi__png p;
+ p.s = s;
+ return stbi__png_info_raw(&p, x, y, comp);
+}
+#endif
+
+// Microsoft/Windows BMP image
+
+#ifndef STBI_NO_BMP
+static int stbi__bmp_test_raw(stbi__context *s)
+{
+ int r;
+ int sz;
+ if (stbi__get8(s) != 'B') return 0;
+ if (stbi__get8(s) != 'M') return 0;
+ stbi__get32le(s); // discard filesize
+ stbi__get16le(s); // discard reserved
+ stbi__get16le(s); // discard reserved
+ stbi__get32le(s); // discard data offset
+ sz = stbi__get32le(s);
+ r = (sz == 12 || sz == 40 || sz == 56 || sz == 108 || sz == 124);
+ return r;
+}
+
+static int stbi__bmp_test(stbi__context *s)
+{
+ int r = stbi__bmp_test_raw(s);
+ stbi__rewind(s);
+ return r;
+}
+
+
+// returns 0..31 for the highest set bit
+static int stbi__high_bit(unsigned int z)
+{
+ int n=0;
+ if (z == 0) return -1;
+ if (z >= 0x10000) n += 16, z >>= 16;
+ if (z >= 0x00100) n += 8, z >>= 8;
+ if (z >= 0x00010) n += 4, z >>= 4;
+ if (z >= 0x00004) n += 2, z >>= 2;
+ if (z >= 0x00002) n += 1, z >>= 1;
+ return n;
+}
+
+static int stbi__bitcount(unsigned int a)
+{
+ a = (a & 0x55555555) + ((a >> 1) & 0x55555555); // max 2
+ a = (a & 0x33333333) + ((a >> 2) & 0x33333333); // max 4
+ a = (a + (a >> 4)) & 0x0f0f0f0f; // max 8 per 4, now 8 bits
+ a = (a + (a >> 8)); // max 16 per 8 bits
+ a = (a + (a >> 16)); // max 32 per 8 bits
+ return a & 0xff;
+}
+
+static int stbi__shiftsigned(int v, int shift, int bits)
+{
+ int result;
+ int z=0;
+
+ if (shift < 0) v <<= -shift;
+ else v >>= shift;
+ result = v;
+
+ z = bits;
+ while (z < 8) {
+ result += v >> z;
+ z += bits;
+ }
+ return result;
+}
+
+typedef struct
+{
+ int bpp, offset, hsz;
+ unsigned int mr,mg,mb,ma, all_a;
+} stbi__bmp_data;
+
+static void *stbi__bmp_parse_header(stbi__context *s, stbi__bmp_data *info)
+{
+ int hsz;
+ if (stbi__get8(s) != 'B' || stbi__get8(s) != 'M') return stbi__errpuc("not BMP", "Corrupt BMP");
+ stbi__get32le(s); // discard filesize
+ stbi__get16le(s); // discard reserved
+ stbi__get16le(s); // discard reserved
+ info->offset = stbi__get32le(s);
+ info->hsz = hsz = stbi__get32le(s);
+ info->mr = info->mg = info->mb = info->ma = 0;
+
+ if (hsz != 12 && hsz != 40 && hsz != 56 && hsz != 108 && hsz != 124) return stbi__errpuc("unknown BMP", "BMP type not supported: unknown");
+ if (hsz == 12) {
+ s->img_x = stbi__get16le(s);
+ s->img_y = stbi__get16le(s);
+ } else {
+ s->img_x = stbi__get32le(s);
+ s->img_y = stbi__get32le(s);
+ }
+ if (stbi__get16le(s) != 1) return stbi__errpuc("bad BMP", "bad BMP");
+ info->bpp = stbi__get16le(s);
+ if (info->bpp == 1) return stbi__errpuc("monochrome", "BMP type not supported: 1-bit");
+ if (hsz != 12) {
+ int compress = stbi__get32le(s);
+ if (compress == 1 || compress == 2) return stbi__errpuc("BMP RLE", "BMP type not supported: RLE");
+ stbi__get32le(s); // discard sizeof
+ stbi__get32le(s); // discard hres
+ stbi__get32le(s); // discard vres
+ stbi__get32le(s); // discard colorsused
+ stbi__get32le(s); // discard max important
+ if (hsz == 40 || hsz == 56) {
+ if (hsz == 56) {
+ stbi__get32le(s);
+ stbi__get32le(s);
+ stbi__get32le(s);
+ stbi__get32le(s);
+ }
+ if (info->bpp == 16 || info->bpp == 32) {
+ if (compress == 0) {
+ if (info->bpp == 32) {
+ info->mr = 0xffu << 16;
+ info->mg = 0xffu << 8;
+ info->mb = 0xffu << 0;
+ info->ma = 0xffu << 24;
+ info->all_a = 0; // if all_a is 0 at end, then we loaded alpha channel but it was all 0
+ } else {
+ info->mr = 31u << 10;
+ info->mg = 31u << 5;
+ info->mb = 31u << 0;
+ }
+ } else if (compress == 3) {
+ info->mr = stbi__get32le(s);
+ info->mg = stbi__get32le(s);
+ info->mb = stbi__get32le(s);
+ // not documented, but generated by photoshop and handled by mspaint
+ if (info->mr == info->mg && info->mg == info->mb) {
+ // ?!?!?
+ return stbi__errpuc("bad BMP", "bad BMP");
+ }
+ } else
+ return stbi__errpuc("bad BMP", "bad BMP");
+ }
+ } else {
+ int i;
+ if (hsz != 108 && hsz != 124)
+ return stbi__errpuc("bad BMP", "bad BMP");
+ info->mr = stbi__get32le(s);
+ info->mg = stbi__get32le(s);
+ info->mb = stbi__get32le(s);
+ info->ma = stbi__get32le(s);
+ stbi__get32le(s); // discard color space
+ for (i=0; i < 12; ++i)
+ stbi__get32le(s); // discard color space parameters
+ if (hsz == 124) {
+ stbi__get32le(s); // discard rendering intent
+ stbi__get32le(s); // discard offset of profile data
+ stbi__get32le(s); // discard size of profile data
+ stbi__get32le(s); // discard reserved
+ }
+ }
+ }
+ return (void *) 1;
+}
+
+
+static void *stbi__bmp_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+ stbi_uc *out;
+ unsigned int mr=0,mg=0,mb=0,ma=0, all_a;
+ stbi_uc pal[256][4];
+ int psize=0,i,j,width;
+ int flip_vertically, pad, target;
+ stbi__bmp_data info;
+ STBI_NOTUSED(ri);
+
+ info.all_a = 255;
+ if (stbi__bmp_parse_header(s, &info) == NULL)
+ return NULL; // error code already set
+
+ flip_vertically = ((int) s->img_y) > 0;
+ s->img_y = abs((int) s->img_y);
+
+ mr = info.mr;
+ mg = info.mg;
+ mb = info.mb;
+ ma = info.ma;
+ all_a = info.all_a;
+
+ if (info.hsz == 12) {
+ if (info.bpp < 24)
+ psize = (info.offset - 14 - 24) / 3;
+ } else {
+ if (info.bpp < 16)
+ psize = (info.offset - 14 - info.hsz) >> 2;
+ }
+
+ s->img_n = ma ? 4 : 3;
+ if (req_comp && req_comp >= 3) // we can directly decode 3 or 4
+ target = req_comp;
+ else
+ target = s->img_n; // if they want monochrome, we'll post-convert
+
+ // sanity-check size
+ if (!stbi__mad3sizes_valid(target, s->img_x, s->img_y, 0))
+ return stbi__errpuc("too large", "Corrupt BMP");
+
+ out = (stbi_uc *) stbi__malloc_mad3(target, s->img_x, s->img_y, 0);
+ if (!out) return stbi__errpuc("outofmem", "Out of memory");
+ if (info.bpp < 16) {
+ int z=0;
+ if (psize == 0 || psize > 256) { STBI_FREE(out); return stbi__errpuc("invalid", "Corrupt BMP"); }
+ for (i=0; i < psize; ++i) {
+ pal[i][2] = stbi__get8(s);
+ pal[i][1] = stbi__get8(s);
+ pal[i][0] = stbi__get8(s);
+ if (info.hsz != 12) stbi__get8(s);
+ pal[i][3] = 255;
+ }
+ stbi__skip(s, info.offset - 14 - info.hsz - psize * (info.hsz == 12 ? 3 : 4));
+ if (info.bpp == 4) width = (s->img_x + 1) >> 1;
+ else if (info.bpp == 8) width = s->img_x;
+ else { STBI_FREE(out); return stbi__errpuc("bad bpp", "Corrupt BMP"); }
+ pad = (-width)&3;
+ for (j=0; j < (int) s->img_y; ++j) {
+ for (i=0; i < (int) s->img_x; i += 2) {
+ int v=stbi__get8(s),v2=0;
+ if (info.bpp == 4) {
+ v2 = v & 15;
+ v >>= 4;
+ }
+ out[z++] = pal[v][0];
+ out[z++] = pal[v][1];
+ out[z++] = pal[v][2];
+ if (target == 4) out[z++] = 255;
+ if (i+1 == (int) s->img_x) break;
+ v = (info.bpp == 8) ? stbi__get8(s) : v2;
+ out[z++] = pal[v][0];
+ out[z++] = pal[v][1];
+ out[z++] = pal[v][2];
+ if (target == 4) out[z++] = 255;
+ }
+ stbi__skip(s, pad);
+ }
+ } else {
+ int rshift=0,gshift=0,bshift=0,ashift=0,rcount=0,gcount=0,bcount=0,acount=0;
+ int z = 0;
+ int easy=0;
+ stbi__skip(s, info.offset - 14 - info.hsz);
+ if (info.bpp == 24) width = 3 * s->img_x;
+ else if (info.bpp == 16) width = 2*s->img_x;
+ else /* bpp = 32 and pad = 0 */ width=0;
+ pad = (-width) & 3;
+ if (info.bpp == 24) {
+ easy = 1;
+ } else if (info.bpp == 32) {
+ if (mb == 0xff && mg == 0xff00 && mr == 0x00ff0000 && ma == 0xff000000)
+ easy = 2;
+ }
+ if (!easy) {
+ if (!mr || !mg || !mb) { STBI_FREE(out); return stbi__errpuc("bad masks", "Corrupt BMP"); }
+ // right shift amt to put high bit in position #7
+ rshift = stbi__high_bit(mr)-7; rcount = stbi__bitcount(mr);
+ gshift = stbi__high_bit(mg)-7; gcount = stbi__bitcount(mg);
+ bshift = stbi__high_bit(mb)-7; bcount = stbi__bitcount(mb);
+ ashift = stbi__high_bit(ma)-7; acount = stbi__bitcount(ma);
+ }
+ for (j=0; j < (int) s->img_y; ++j) {
+ if (easy) {
+ for (i=0; i < (int) s->img_x; ++i) {
+ unsigned char a;
+ out[z+2] = stbi__get8(s);
+ out[z+1] = stbi__get8(s);
+ out[z+0] = stbi__get8(s);
+ z += 3;
+ a = (easy == 2 ? stbi__get8(s) : 255);
+ all_a |= a;
+ if (target == 4) out[z++] = a;
+ }
+ } else {
+ int bpp = info.bpp;
+ for (i=0; i < (int) s->img_x; ++i) {
+ stbi__uint32 v = (bpp == 16 ? (stbi__uint32) stbi__get16le(s) : stbi__get32le(s));
+ int a;
+ out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mr, rshift, rcount));
+ out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mg, gshift, gcount));
+ out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mb, bshift, bcount));
+ a = (ma ? stbi__shiftsigned(v & ma, ashift, acount) : 255);
+ all_a |= a;
+ if (target == 4) out[z++] = STBI__BYTECAST(a);
+ }
+ }
+ stbi__skip(s, pad);
+ }
+ }
+
+ // if alpha channel is all 0s, replace with all 255s
+ if (target == 4 && all_a == 0)
+ for (i=4*s->img_x*s->img_y-1; i >= 0; i -= 4)
+ out[i] = 255;
+
+ if (flip_vertically) {
+ stbi_uc t;
+ for (j=0; j < (int) s->img_y>>1; ++j) {
+ stbi_uc *p1 = out + j *s->img_x*target;
+ stbi_uc *p2 = out + (s->img_y-1-j)*s->img_x*target;
+ for (i=0; i < (int) s->img_x*target; ++i) {
+ t = p1[i], p1[i] = p2[i], p2[i] = t;
+ }
+ }
+ }
+
+ if (req_comp && req_comp != target) {
+ out = stbi__convert_format(out, target, req_comp, s->img_x, s->img_y);
+ if (out == NULL) return out; // stbi__convert_format frees input on failure
+ }
+
+ *x = s->img_x;
+ *y = s->img_y;
+ if (comp) *comp = s->img_n;
+ return out;
+}
+#endif
+
+// Targa Truevision - TGA
+// by Jonathan Dummer
+#ifndef STBI_NO_TGA
+// returns STBI_rgb or whatever, 0 on error
+static int stbi__tga_get_comp(int bits_per_pixel, int is_grey, int* is_rgb16)
+{
+ // only RGB or RGBA (incl. 16bit) or grey allowed
+ if(is_rgb16) *is_rgb16 = 0;
+ switch(bits_per_pixel) {
+ case 8: return STBI_grey;
+ case 16: if(is_grey) return STBI_grey_alpha;
+ // else: fall-through
+ case 15: if(is_rgb16) *is_rgb16 = 1;
+ return STBI_rgb;
+ case 24: // fall-through
+ case 32: return bits_per_pixel/8;
+ default: return 0;
+ }
+}
+
+static int stbi__tga_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ int tga_w, tga_h, tga_comp, tga_image_type, tga_bits_per_pixel, tga_colormap_bpp;
+ int sz, tga_colormap_type;
+ stbi__get8(s); // discard Offset
+ tga_colormap_type = stbi__get8(s); // colormap type
+ if( tga_colormap_type > 1 ) {
+ stbi__rewind(s);
+ return 0; // only RGB or indexed allowed
+ }
+ tga_image_type = stbi__get8(s); // image type
+ if ( tga_colormap_type == 1 ) { // colormapped (paletted) image
+ if (tga_image_type != 1 && tga_image_type != 9) {
+ stbi__rewind(s);
+ return 0;
+ }
+ stbi__skip(s,4); // skip index of first colormap entry and number of entries
+ sz = stbi__get8(s); // check bits per palette color entry
+ if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) {
+ stbi__rewind(s);
+ return 0;
+ }
+ stbi__skip(s,4); // skip image x and y origin
+ tga_colormap_bpp = sz;
+ } else { // "normal" image w/o colormap - only RGB or grey allowed, +/- RLE
+ if ( (tga_image_type != 2) && (tga_image_type != 3) && (tga_image_type != 10) && (tga_image_type != 11) ) {
+ stbi__rewind(s);
+ return 0; // only RGB or grey allowed, +/- RLE
+ }
+ stbi__skip(s,9); // skip colormap specification and image x/y origin
+ tga_colormap_bpp = 0;
+ }
+ tga_w = stbi__get16le(s);
+ if( tga_w < 1 ) {
+ stbi__rewind(s);
+ return 0; // test width
+ }
+ tga_h = stbi__get16le(s);
+ if( tga_h < 1 ) {
+ stbi__rewind(s);
+ return 0; // test height
+ }
+ tga_bits_per_pixel = stbi__get8(s); // bits per pixel
+ stbi__get8(s); // ignore alpha bits
+ if (tga_colormap_bpp != 0) {
+ if((tga_bits_per_pixel != 8) && (tga_bits_per_pixel != 16)) {
+ // when using a colormap, tga_bits_per_pixel is the size of the indexes
+ // I don't think anything but 8 or 16bit indexes makes sense
+ stbi__rewind(s);
+ return 0;
+ }
+ tga_comp = stbi__tga_get_comp(tga_colormap_bpp, 0, NULL);
+ } else {
+ tga_comp = stbi__tga_get_comp(tga_bits_per_pixel, (tga_image_type == 3) || (tga_image_type == 11), NULL);
+ }
+ if(!tga_comp) {
+ stbi__rewind(s);
+ return 0;
+ }
+ if (x) *x = tga_w;
+ if (y) *y = tga_h;
+ if (comp) *comp = tga_comp;
+ return 1; // seems to have passed everything
+}
+
+static int stbi__tga_test(stbi__context *s)
+{
+ int res = 0;
+ int sz, tga_color_type;
+ stbi__get8(s); // discard Offset
+ tga_color_type = stbi__get8(s); // color type
+ if ( tga_color_type > 1 ) goto errorEnd; // only RGB or indexed allowed
+ sz = stbi__get8(s); // image type
+ if ( tga_color_type == 1 ) { // colormapped (paletted) image
+ if (sz != 1 && sz != 9) goto errorEnd; // colortype 1 demands image type 1 or 9
+ stbi__skip(s,4); // skip index of first colormap entry and number of entries
+ sz = stbi__get8(s); // check bits per palette color entry
+ if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) goto errorEnd;
+ stbi__skip(s,4); // skip image x and y origin
+ } else { // "normal" image w/o colormap
+ if ( (sz != 2) && (sz != 3) && (sz != 10) && (sz != 11) ) goto errorEnd; // only RGB or grey allowed, +/- RLE
+ stbi__skip(s,9); // skip colormap specification and image x/y origin
+ }
+ if ( stbi__get16le(s) < 1 ) goto errorEnd; // test width
+ if ( stbi__get16le(s) < 1 ) goto errorEnd; // test height
+ sz = stbi__get8(s); // bits per pixel
+ if ( (tga_color_type == 1) && (sz != 8) && (sz != 16) ) goto errorEnd; // for colormapped images, bpp is size of an index
+ if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) goto errorEnd;
+
+ res = 1; // if we got this far, everything's good and we can return 1 instead of 0
+
+errorEnd:
+ stbi__rewind(s);
+ return res;
+}
+
+// read 16bit value and convert to 24bit RGB
+static void stbi__tga_read_rgb16(stbi__context *s, stbi_uc* out)
+{
+ stbi__uint16 px = (stbi__uint16)stbi__get16le(s);
+ stbi__uint16 fiveBitMask = 31;
+ // we have 3 channels with 5bits each
+ int r = (px >> 10) & fiveBitMask;
+ int g = (px >> 5) & fiveBitMask;
+ int b = px & fiveBitMask;
+ // Note that this saves the data in RGB(A) order, so it doesn't need to be swapped later
+ out[0] = (stbi_uc)((r * 255)/31);
+ out[1] = (stbi_uc)((g * 255)/31);
+ out[2] = (stbi_uc)((b * 255)/31);
+
+ // some people claim that the most significant bit might be used for alpha
+ // (possibly if an alpha-bit is set in the "image descriptor byte")
+ // but that only made 16bit test images completely translucent..
+ // so let's treat all 15 and 16bit TGAs as RGB with no alpha.
+}
+
+static void *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+ // read in the TGA header stuff
+ int tga_offset = stbi__get8(s);
+ int tga_indexed = stbi__get8(s);
+ int tga_image_type = stbi__get8(s);
+ int tga_is_RLE = 0;
+ int tga_palette_start = stbi__get16le(s);
+ int tga_palette_len = stbi__get16le(s);
+ int tga_palette_bits = stbi__get8(s);
+ int tga_x_origin = stbi__get16le(s);
+ int tga_y_origin = stbi__get16le(s);
+ int tga_width = stbi__get16le(s);
+ int tga_height = stbi__get16le(s);
+ int tga_bits_per_pixel = stbi__get8(s);
+ int tga_comp, tga_rgb16=0;
+ int tga_inverted = stbi__get8(s);
+ // int tga_alpha_bits = tga_inverted & 15; // the 4 lowest bits - unused (useless?)
+ // image data
+ unsigned char *tga_data;
+ unsigned char *tga_palette = NULL;
+ int i, j;
+ unsigned char raw_data[4] = {0};
+ int RLE_count = 0;
+ int RLE_repeating = 0;
+ int read_next_pixel = 1;
+ STBI_NOTUSED(ri);
+
+ // do a tiny bit of precessing
+ if ( tga_image_type >= 8 )
+ {
+ tga_image_type -= 8;
+ tga_is_RLE = 1;
+ }
+ tga_inverted = 1 - ((tga_inverted >> 5) & 1);
+
+ // If I'm paletted, then I'll use the number of bits from the palette
+ if ( tga_indexed ) tga_comp = stbi__tga_get_comp(tga_palette_bits, 0, &tga_rgb16);
+ else tga_comp = stbi__tga_get_comp(tga_bits_per_pixel, (tga_image_type == 3), &tga_rgb16);
+
+ if(!tga_comp) // shouldn't really happen, stbi__tga_test() should have ensured basic consistency
+ return stbi__errpuc("bad format", "Can't find out TGA pixelformat");
+
+ // tga info
+ *x = tga_width;
+ *y = tga_height;
+ if (comp) *comp = tga_comp;
+
+ if (!stbi__mad3sizes_valid(tga_width, tga_height, tga_comp, 0))
+ return stbi__errpuc("too large", "Corrupt TGA");
+
+ tga_data = (unsigned char*)stbi__malloc_mad3(tga_width, tga_height, tga_comp, 0);
+ if (!tga_data) return stbi__errpuc("outofmem", "Out of memory");
+
+ // skip to the data's starting position (offset usually = 0)
+ stbi__skip(s, tga_offset );
+
+ if ( !tga_indexed && !tga_is_RLE && !tga_rgb16 ) {
+ for (i=0; i < tga_height; ++i) {
+ int row = tga_inverted ? tga_height -i - 1 : i;
+ stbi_uc *tga_row = tga_data + row*tga_width*tga_comp;
+ stbi__getn(s, tga_row, tga_width * tga_comp);
+ }
+ } else {
+ // do I need to load a palette?
+ if ( tga_indexed)
+ {
+ // any data to skip? (offset usually = 0)
+ stbi__skip(s, tga_palette_start );
+ // load the palette
+ tga_palette = (unsigned char*)stbi__malloc_mad2(tga_palette_len, tga_comp, 0);
+ if (!tga_palette) {
+ STBI_FREE(tga_data);
+ return stbi__errpuc("outofmem", "Out of memory");
+ }
+ if (tga_rgb16) {
+ stbi_uc *pal_entry = tga_palette;
+ STBI_ASSERT(tga_comp == STBI_rgb);
+ for (i=0; i < tga_palette_len; ++i) {
+ stbi__tga_read_rgb16(s, pal_entry);
+ pal_entry += tga_comp;
+ }
+ } else if (!stbi__getn(s, tga_palette, tga_palette_len * tga_comp)) {
+ STBI_FREE(tga_data);
+ STBI_FREE(tga_palette);
+ return stbi__errpuc("bad palette", "Corrupt TGA");
+ }
+ }
+ // load the data
+ for (i=0; i < tga_width * tga_height; ++i)
+ {
+ // if I'm in RLE mode, do I need to get a RLE stbi__pngchunk?
+ if ( tga_is_RLE )
+ {
+ if ( RLE_count == 0 )
+ {
+ // yep, get the next byte as a RLE command
+ int RLE_cmd = stbi__get8(s);
+ RLE_count = 1 + (RLE_cmd & 127);
+ RLE_repeating = RLE_cmd >> 7;
+ read_next_pixel = 1;
+ } else if ( !RLE_repeating )
+ {
+ read_next_pixel = 1;
+ }
+ } else
+ {
+ read_next_pixel = 1;
+ }
+ // OK, if I need to read a pixel, do it now
+ if ( read_next_pixel )
+ {
+ // load however much data we did have
+ if ( tga_indexed )
+ {
+ // read in index, then perform the lookup
+ int pal_idx = (tga_bits_per_pixel == 8) ? stbi__get8(s) : stbi__get16le(s);
+ if ( pal_idx >= tga_palette_len ) {
+ // invalid index
+ pal_idx = 0;
+ }
+ pal_idx *= tga_comp;
+ for (j = 0; j < tga_comp; ++j) {
+ raw_data[j] = tga_palette[pal_idx+j];
+ }
+ } else if(tga_rgb16) {
+ STBI_ASSERT(tga_comp == STBI_rgb);
+ stbi__tga_read_rgb16(s, raw_data);
+ } else {
+ // read in the data raw
+ for (j = 0; j < tga_comp; ++j) {
+ raw_data[j] = stbi__get8(s);
+ }
+ }
+ // clear the reading flag for the next pixel
+ read_next_pixel = 0;
+ } // end of reading a pixel
+
+ // copy data
+ for (j = 0; j < tga_comp; ++j)
+ tga_data[i*tga_comp+j] = raw_data[j];
+
+ // in case we're in RLE mode, keep counting down
+ --RLE_count;
+ }
+ // do I need to invert the image?
+ if ( tga_inverted )
+ {
+ for (j = 0; j*2 < tga_height; ++j)
+ {
+ int index1 = j * tga_width * tga_comp;
+ int index2 = (tga_height - 1 - j) * tga_width * tga_comp;
+ for (i = tga_width * tga_comp; i > 0; --i)
+ {
+ unsigned char temp = tga_data[index1];
+ tga_data[index1] = tga_data[index2];
+ tga_data[index2] = temp;
+ ++index1;
+ ++index2;
+ }
+ }
+ }
+ // clear my palette, if I had one
+ if ( tga_palette != NULL )
+ {
+ STBI_FREE( tga_palette );
+ }
+ }
+
+ // swap RGB - if the source data was RGB16, it already is in the right order
+ if (tga_comp >= 3 && !tga_rgb16)
+ {
+ unsigned char* tga_pixel = tga_data;
+ for (i=0; i < tga_width * tga_height; ++i)
+ {
+ unsigned char temp = tga_pixel[0];
+ tga_pixel[0] = tga_pixel[2];
+ tga_pixel[2] = temp;
+ tga_pixel += tga_comp;
+ }
+ }
+
+ // convert to target component count
+ if (req_comp && req_comp != tga_comp)
+ tga_data = stbi__convert_format(tga_data, tga_comp, req_comp, tga_width, tga_height);
+
+ // the things I do to get rid of an error message, and yet keep
+ // Microsoft's C compilers happy... [8^(
+ tga_palette_start = tga_palette_len = tga_palette_bits =
+ tga_x_origin = tga_y_origin = 0;
+ // OK, done
+ return tga_data;
+}
+#endif
+
+// *************************************************************************************************
+// Photoshop PSD loader -- PD by Thatcher Ulrich, integration by Nicolas Schulz, tweaked by STB
+
+#ifndef STBI_NO_PSD
+static int stbi__psd_test(stbi__context *s)
+{
+ int r = (stbi__get32be(s) == 0x38425053);
+ stbi__rewind(s);
+ return r;
+}
+
+static int stbi__psd_decode_rle(stbi__context *s, stbi_uc *p, int pixelCount)
+{
+ int count, nleft, len;
+
+ count = 0;
+ while ((nleft = pixelCount - count) > 0) {
+ len = stbi__get8(s);
+ if (len == 128) {
+ // No-op.
+ } else if (len < 128) {
+ // Copy next len+1 bytes literally.
+ len++;
+ if (len > nleft) return 0; // corrupt data
+ count += len;
+ while (len) {
+ *p = stbi__get8(s);
+ p += 4;
+ len--;
+ }
+ } else if (len > 128) {
+ stbi_uc val;
+ // Next -len+1 bytes in the dest are replicated from next source byte.
+ // (Interpret len as a negative 8-bit int.)
+ len = 257 - len;
+ if (len > nleft) return 0; // corrupt data
+ val = stbi__get8(s);
+ count += len;
+ while (len) {
+ *p = val;
+ p += 4;
+ len--;
+ }
+ }
+ }
+
+ return 1;
+}
+
+static void *stbi__psd_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri, int bpc)
+{
+ int pixelCount;
+ int channelCount, compression;
+ int channel, i;
+ int bitdepth;
+ int w,h;
+ stbi_uc *out;
+ STBI_NOTUSED(ri);
+
+ // Check identifier
+ if (stbi__get32be(s) != 0x38425053) // "8BPS"
+ return stbi__errpuc("not PSD", "Corrupt PSD image");
+
+ // Check file type version.
+ if (stbi__get16be(s) != 1)
+ return stbi__errpuc("wrong version", "Unsupported version of PSD image");
+
+ // Skip 6 reserved bytes.
+ stbi__skip(s, 6 );
+
+ // Read the number of channels (R, G, B, A, etc).
+ channelCount = stbi__get16be(s);
+ if (channelCount < 0 || channelCount > 16)
+ return stbi__errpuc("wrong channel count", "Unsupported number of channels in PSD image");
+
+ // Read the rows and columns of the image.
+ h = stbi__get32be(s);
+ w = stbi__get32be(s);
+
+ // Make sure the depth is 8 bits.
+ bitdepth = stbi__get16be(s);
+ if (bitdepth != 8 && bitdepth != 16)
+ return stbi__errpuc("unsupported bit depth", "PSD bit depth is not 8 or 16 bit");
+
+ // Make sure the color mode is RGB.
+ // Valid options are:
+ // 0: Bitmap
+ // 1: Grayscale
+ // 2: Indexed color
+ // 3: RGB color
+ // 4: CMYK color
+ // 7: Multichannel
+ // 8: Duotone
+ // 9: Lab color
+ if (stbi__get16be(s) != 3)
+ return stbi__errpuc("wrong color format", "PSD is not in RGB color format");
+
+ // Skip the Mode Data. (It's the palette for indexed color; other info for other modes.)
+ stbi__skip(s,stbi__get32be(s) );
+
+ // Skip the image resources. (resolution, pen tool paths, etc)
+ stbi__skip(s, stbi__get32be(s) );
+
+ // Skip the reserved data.
+ stbi__skip(s, stbi__get32be(s) );
+
+ // Find out if the data is compressed.
+ // Known values:
+ // 0: no compression
+ // 1: RLE compressed
+ compression = stbi__get16be(s);
+ if (compression > 1)
+ return stbi__errpuc("bad compression", "PSD has an unknown compression format");
+
+ // Check size
+ if (!stbi__mad3sizes_valid(4, w, h, 0))
+ return stbi__errpuc("too large", "Corrupt PSD");
+
+ // Create the destination image.
+
+ if (!compression && bitdepth == 16 && bpc == 16) {
+ out = (stbi_uc *) stbi__malloc_mad3(8, w, h, 0);
+ ri->bits_per_channel = 16;
+ } else
+ out = (stbi_uc *) stbi__malloc(4 * w*h);
+
+ if (!out) return stbi__errpuc("outofmem", "Out of memory");
+ pixelCount = w*h;
+
+ // Initialize the data to zero.
+ //memset( out, 0, pixelCount * 4 );
+
+ // Finally, the image data.
+ if (compression) {
+ // RLE as used by .PSD and .TIFF
+ // Loop until you get the number of unpacked bytes you are expecting:
+ // Read the next source byte into n.
+ // If n is between 0 and 127 inclusive, copy the next n+1 bytes literally.
+ // Else if n is between -127 and -1 inclusive, copy the next byte -n+1 times.
+ // Else if n is 128, noop.
+ // Endloop
+
+ // The RLE-compressed data is preceeded by a 2-byte data count for each row in the data,
+ // which we're going to just skip.
+ stbi__skip(s, h * channelCount * 2 );
+
+ // Read the RLE data by channel.
+ for (channel = 0; channel < 4; channel++) {
+ stbi_uc *p;
+
+ p = out+channel;
+ if (channel >= channelCount) {
+ // Fill this channel with default data.
+ for (i = 0; i < pixelCount; i++, p += 4)
+ *p = (channel == 3 ? 255 : 0);
+ } else {
+ // Read the RLE data.
+ if (!stbi__psd_decode_rle(s, p, pixelCount)) {
+ STBI_FREE(out);
+ return stbi__errpuc("corrupt", "bad RLE data");
+ }
+ }
+ }
+
+ } else {
+ // We're at the raw image data. It's each channel in order (Red, Green, Blue, Alpha, ...)
+ // where each channel consists of an 8-bit (or 16-bit) value for each pixel in the image.
+
+ // Read the data by channel.
+ for (channel = 0; channel < 4; channel++) {
+ if (channel >= channelCount) {
+ // Fill this channel with default data.
+ if (bitdepth == 16 && bpc == 16) {
+ stbi__uint16 *q = ((stbi__uint16 *) out) + channel;
+ stbi__uint16 val = channel == 3 ? 65535 : 0;
+ for (i = 0; i < pixelCount; i++, q += 4)
+ *q = val;
+ } else {
+ stbi_uc *p = out+channel;
+ stbi_uc val = channel == 3 ? 255 : 0;
+ for (i = 0; i < pixelCount; i++, p += 4)
+ *p = val;
+ }
+ } else {
+ if (ri->bits_per_channel == 16) { // output bpc
+ stbi__uint16 *q = ((stbi__uint16 *) out) + channel;
+ for (i = 0; i < pixelCount; i++, q += 4)
+ *q = (stbi__uint16) stbi__get16be(s);
+ } else {
+ stbi_uc *p = out+channel;
+ if (bitdepth == 16) { // input bpc
+ for (i = 0; i < pixelCount; i++, p += 4)
+ *p = (stbi_uc) (stbi__get16be(s) >> 8);
+ } else {
+ for (i = 0; i < pixelCount; i++, p += 4)
+ *p = stbi__get8(s);
+ }
+ }
+ }
+ }
+ }
+
+ // remove weird white matte from PSD
+ if (channelCount >= 4) {
+ if (ri->bits_per_channel == 16) {
+ for (i=0; i < w*h; ++i) {
+ stbi__uint16 *pixel = (stbi__uint16 *) out + 4*i;
+ if (pixel[3] != 0 && pixel[3] != 65535) {
+ float a = pixel[3] / 65535.0f;
+ float ra = 1.0f / a;
+ float inv_a = 65535.0f * (1 - ra);
+ pixel[0] = (stbi__uint16) (pixel[0]*ra + inv_a);
+ pixel[1] = (stbi__uint16) (pixel[1]*ra + inv_a);
+ pixel[2] = (stbi__uint16) (pixel[2]*ra + inv_a);
+ }
+ }
+ } else {
+ for (i=0; i < w*h; ++i) {
+ unsigned char *pixel = out + 4*i;
+ if (pixel[3] != 0 && pixel[3] != 255) {
+ float a = pixel[3] / 255.0f;
+ float ra = 1.0f / a;
+ float inv_a = 255.0f * (1 - ra);
+ pixel[0] = (unsigned char) (pixel[0]*ra + inv_a);
+ pixel[1] = (unsigned char) (pixel[1]*ra + inv_a);
+ pixel[2] = (unsigned char) (pixel[2]*ra + inv_a);
+ }
+ }
+ }
+ }
+
+ // convert to desired output format
+ if (req_comp && req_comp != 4) {
+ if (ri->bits_per_channel == 16)
+ out = (stbi_uc *) stbi__convert_format16((stbi__uint16 *) out, 4, req_comp, w, h);
+ else
+ out = stbi__convert_format(out, 4, req_comp, w, h);
+ if (out == NULL) return out; // stbi__convert_format frees input on failure
+ }
+
+ if (comp) *comp = 4;
+ *y = h;
+ *x = w;
+
+ return out;
+}
+#endif
+
+// *************************************************************************************************
+// Softimage PIC loader
+// by Tom Seddon
+//
+// See http://softimage.wiki.softimage.com/index.php/INFO:_PIC_file_format
+// See http://ozviz.wasp.uwa.edu.au/~pbourke/dataformats/softimagepic/
+
+#ifndef STBI_NO_PIC
+static int stbi__pic_is4(stbi__context *s,const char *str)
+{
+ int i;
+ for (i=0; i<4; ++i)
+ if (stbi__get8(s) != (stbi_uc)str[i])
+ return 0;
+
+ return 1;
+}
+
+static int stbi__pic_test_core(stbi__context *s)
+{
+ int i;
+
+ if (!stbi__pic_is4(s,"\x53\x80\xF6\x34"))
+ return 0;
+
+ for(i=0;i<84;++i)
+ stbi__get8(s);
+
+ if (!stbi__pic_is4(s,"PICT"))
+ return 0;
+
+ return 1;
+}
+
+typedef struct
+{
+ stbi_uc size,type,channel;
+} stbi__pic_packet;
+
+static stbi_uc *stbi__readval(stbi__context *s, int channel, stbi_uc *dest)
+{
+ int mask=0x80, i;
+
+ for (i=0; i<4; ++i, mask>>=1) {
+ if (channel & mask) {
+ if (stbi__at_eof(s)) return stbi__errpuc("bad file","PIC file too short");
+ dest[i]=stbi__get8(s);
+ }
+ }
+
+ return dest;
+}
+
+static void stbi__copyval(int channel,stbi_uc *dest,const stbi_uc *src)
+{
+ int mask=0x80,i;
+
+ for (i=0;i<4; ++i, mask>>=1)
+ if (channel&mask)
+ dest[i]=src[i];
+}
+
+static stbi_uc *stbi__pic_load_core(stbi__context *s,int width,int height,int *comp, stbi_uc *result)
+{
+ int act_comp=0,num_packets=0,y,chained;
+ stbi__pic_packet packets[10];
+
+ // this will (should...) cater for even some bizarre stuff like having data
+ // for the same channel in multiple packets.
+ do {
+ stbi__pic_packet *packet;
+
+ if (num_packets==sizeof(packets)/sizeof(packets[0]))
+ return stbi__errpuc("bad format","too many packets");
+
+ packet = &packets[num_packets++];
+
+ chained = stbi__get8(s);
+ packet->size = stbi__get8(s);
+ packet->type = stbi__get8(s);
+ packet->channel = stbi__get8(s);
+
+ act_comp |= packet->channel;
+
+ if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (reading packets)");
+ if (packet->size != 8) return stbi__errpuc("bad format","packet isn't 8bpp");
+ } while (chained);
+
+ *comp = (act_comp & 0x10 ? 4 : 3); // has alpha channel?
+
+ for(y=0; y<height; ++y) {
+ int packet_idx;
+
+ for(packet_idx=0; packet_idx < num_packets; ++packet_idx) {
+ stbi__pic_packet *packet = &packets[packet_idx];
+ stbi_uc *dest = result+y*width*4;
+
+ switch (packet->type) {
+ default:
+ return stbi__errpuc("bad format","packet has bad compression type");
+
+ case 0: {//uncompressed
+ int x;
+
+ for(x=0;x<width;++x, dest+=4)
+ if (!stbi__readval(s,packet->channel,dest))
+ return 0;
+ break;
+ }
+
+ case 1://Pure RLE
+ {
+ int left=width, i;
+
+ while (left>0) {
+ stbi_uc count,value[4];
+
+ count=stbi__get8(s);
+ if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (pure read count)");
+
+ if (count > left)
+ count = (stbi_uc) left;
+
+ if (!stbi__readval(s,packet->channel,value)) return 0;
+
+ for(i=0; i<count; ++i,dest+=4)
+ stbi__copyval(packet->channel,dest,value);
+ left -= count;
+ }
+ }
+ break;
+
+ case 2: {//Mixed RLE
+ int left=width;
+ while (left>0) {
+ int count = stbi__get8(s), i;
+ if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (mixed read count)");
+
+ if (count >= 128) { // Repeated
+ stbi_uc value[4];
+
+ if (count==128)
+ count = stbi__get16be(s);
+ else
+ count -= 127;
+ if (count > left)
+ return stbi__errpuc("bad file","scanline overrun");
+
+ if (!stbi__readval(s,packet->channel,value))
+ return 0;
+
+ for(i=0;i<count;++i, dest += 4)
+ stbi__copyval(packet->channel,dest,value);
+ } else { // Raw
+ ++count;
+ if (count>left) return stbi__errpuc("bad file","scanline overrun");
+
+ for(i=0;i<count;++i, dest+=4)
+ if (!stbi__readval(s,packet->channel,dest))
+ return 0;
+ }
+ left-=count;
+ }
+ break;
+ }
+ }
+ }
+ }
+
+ return result;
+}
+
+static void *stbi__pic_load(stbi__context *s,int *px,int *py,int *comp,int req_comp, stbi__result_info *ri)
+{
+ stbi_uc *result;
+ int i, x,y, internal_comp;
+ STBI_NOTUSED(ri);
+
+ if (!comp) comp = &internal_comp;
+
+ for (i=0; i<92; ++i)
+ stbi__get8(s);
+
+ x = stbi__get16be(s);
+ y = stbi__get16be(s);
+ if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (pic header)");
+ if (!stbi__mad3sizes_valid(x, y, 4, 0)) return stbi__errpuc("too large", "PIC image too large to decode");
+
+ stbi__get32be(s); //skip `ratio'
+ stbi__get16be(s); //skip `fields'
+ stbi__get16be(s); //skip `pad'
+
+ // intermediate buffer is RGBA
+ result = (stbi_uc *) stbi__malloc_mad3(x, y, 4, 0);
+ memset(result, 0xff, x*y*4);
+
+ if (!stbi__pic_load_core(s,x,y,comp, result)) {
+ STBI_FREE(result);
+ result=0;
+ }
+ *px = x;
+ *py = y;
+ if (req_comp == 0) req_comp = *comp;
+ result=stbi__convert_format(result,4,req_comp,x,y);
+
+ return result;
+}
+
+static int stbi__pic_test(stbi__context *s)
+{
+ int r = stbi__pic_test_core(s);
+ stbi__rewind(s);
+ return r;
+}
+#endif
+
+// *************************************************************************************************
+// GIF loader -- public domain by Jean-Marc Lienher -- simplified/shrunk by stb
+
+#ifndef STBI_NO_GIF
+typedef struct
+{
+ stbi__int16 prefix;
+ stbi_uc first;
+ stbi_uc suffix;
+} stbi__gif_lzw;
+
+typedef struct
+{
+ int w,h;
+ stbi_uc *out, *old_out; // output buffer (always 4 components)
+ int flags, bgindex, ratio, transparent, eflags, delay;
+ stbi_uc pal[256][4];
+ stbi_uc lpal[256][4];
+ stbi__gif_lzw codes[4096];
+ stbi_uc *color_table;
+ int parse, step;
+ int lflags;
+ int start_x, start_y;
+ int max_x, max_y;
+ int cur_x, cur_y;
+ int line_size;
+} stbi__gif;
+
+static int stbi__gif_test_raw(stbi__context *s)
+{
+ int sz;
+ if (stbi__get8(s) != 'G' || stbi__get8(s) != 'I' || stbi__get8(s) != 'F' || stbi__get8(s) != '8') return 0;
+ sz = stbi__get8(s);
+ if (sz != '9' && sz != '7') return 0;
+ if (stbi__get8(s) != 'a') return 0;
+ return 1;
+}
+
+static int stbi__gif_test(stbi__context *s)
+{
+ int r = stbi__gif_test_raw(s);
+ stbi__rewind(s);
+ return r;
+}
+
+static void stbi__gif_parse_colortable(stbi__context *s, stbi_uc pal[256][4], int num_entries, int transp)
+{
+ int i;
+ for (i=0; i < num_entries; ++i) {
+ pal[i][2] = stbi__get8(s);
+ pal[i][1] = stbi__get8(s);
+ pal[i][0] = stbi__get8(s);
+ pal[i][3] = transp == i ? 0 : 255;
+ }
+}
+
+static int stbi__gif_header(stbi__context *s, stbi__gif *g, int *comp, int is_info)
+{
+ stbi_uc version;
+ if (stbi__get8(s) != 'G' || stbi__get8(s) != 'I' || stbi__get8(s) != 'F' || stbi__get8(s) != '8')
+ return stbi__err("not GIF", "Corrupt GIF");
+
+ version = stbi__get8(s);
+ if (version != '7' && version != '9') return stbi__err("not GIF", "Corrupt GIF");
+ if (stbi__get8(s) != 'a') return stbi__err("not GIF", "Corrupt GIF");
+
+ stbi__g_failure_reason = "";
+ g->w = stbi__get16le(s);
+ g->h = stbi__get16le(s);
+ g->flags = stbi__get8(s);
+ g->bgindex = stbi__get8(s);
+ g->ratio = stbi__get8(s);
+ g->transparent = -1;
+
+ if (comp != 0) *comp = 4; // can't actually tell whether it's 3 or 4 until we parse the comments
+
+ if (is_info) return 1;
+
+ if (g->flags & 0x80)
+ stbi__gif_parse_colortable(s,g->pal, 2 << (g->flags & 7), -1);
+
+ return 1;
+}
+
+static int stbi__gif_info_raw(stbi__context *s, int *x, int *y, int *comp)
+{
+ stbi__gif* g = (stbi__gif*) stbi__malloc(sizeof(stbi__gif));
+ if (!stbi__gif_header(s, g, comp, 1)) {
+ STBI_FREE(g);
+ stbi__rewind( s );
+ return 0;
+ }
+ if (x) *x = g->w;
+ if (y) *y = g->h;
+ STBI_FREE(g);
+ return 1;
+}
+
+static void stbi__out_gif_code(stbi__gif *g, stbi__uint16 code)
+{
+ stbi_uc *p, *c;
+
+ // recurse to decode the prefixes, since the linked-list is backwards,
+ // and working backwards through an interleaved image would be nasty
+ if (g->codes[code].prefix >= 0)
+ stbi__out_gif_code(g, g->codes[code].prefix);
+
+ if (g->cur_y >= g->max_y) return;
+
+ p = &g->out[g->cur_x + g->cur_y];
+ c = &g->color_table[g->codes[code].suffix * 4];
+
+ if (c[3] >= 128) {
+ p[0] = c[2];
+ p[1] = c[1];
+ p[2] = c[0];
+ p[3] = c[3];
+ }
+ g->cur_x += 4;
+
+ if (g->cur_x >= g->max_x) {
+ g->cur_x = g->start_x;
+ g->cur_y += g->step;
+
+ while (g->cur_y >= g->max_y && g->parse > 0) {
+ g->step = (1 << g->parse) * g->line_size;
+ g->cur_y = g->start_y + (g->step >> 1);
+ --g->parse;
+ }
+ }
+}
+
+static stbi_uc *stbi__process_gif_raster(stbi__context *s, stbi__gif *g)
+{
+ stbi_uc lzw_cs;
+ stbi__int32 len, init_code;
+ stbi__uint32 first;
+ stbi__int32 codesize, codemask, avail, oldcode, bits, valid_bits, clear;
+ stbi__gif_lzw *p;
+
+ lzw_cs = stbi__get8(s);
+ if (lzw_cs > 12) return NULL;
+ clear = 1 << lzw_cs;
+ first = 1;
+ codesize = lzw_cs + 1;
+ codemask = (1 << codesize) - 1;
+ bits = 0;
+ valid_bits = 0;
+ for (init_code = 0; init_code < clear; init_code++) {
+ g->codes[init_code].prefix = -1;
+ g->codes[init_code].first = (stbi_uc) init_code;
+ g->codes[init_code].suffix = (stbi_uc) init_code;
+ }
+
+ // support no starting clear code
+ avail = clear+2;
+ oldcode = -1;
+
+ len = 0;
+ for(;;) {
+ if (valid_bits < codesize) {
+ if (len == 0) {
+ len = stbi__get8(s); // start new block
+ if (len == 0)
+ return g->out;
+ }
+ --len;
+ bits |= (stbi__int32) stbi__get8(s) << valid_bits;
+ valid_bits += 8;
+ } else {
+ stbi__int32 code = bits & codemask;
+ bits >>= codesize;
+ valid_bits -= codesize;
+ // @OPTIMIZE: is there some way we can accelerate the non-clear path?
+ if (code == clear) { // clear code
+ codesize = lzw_cs + 1;
+ codemask = (1 << codesize) - 1;
+ avail = clear + 2;
+ oldcode = -1;
+ first = 0;
+ } else if (code == clear + 1) { // end of stream code
+ stbi__skip(s, len);
+ while ((len = stbi__get8(s)) > 0)
+ stbi__skip(s,len);
+ return g->out;
+ } else if (code <= avail) {
+ if (first) return stbi__errpuc("no clear code", "Corrupt GIF");
+
+ if (oldcode >= 0) {
+ p = &g->codes[avail++];
+ if (avail > 4096) return stbi__errpuc("too many codes", "Corrupt GIF");
+ p->prefix = (stbi__int16) oldcode;
+ p->first = g->codes[oldcode].first;
+ p->suffix = (code == avail) ? p->first : g->codes[code].first;
+ } else if (code == avail)
+ return stbi__errpuc("illegal code in raster", "Corrupt GIF");
+
+ stbi__out_gif_code(g, (stbi__uint16) code);
+
+ if ((avail & codemask) == 0 && avail <= 0x0FFF) {
+ codesize++;
+ codemask = (1 << codesize) - 1;
+ }
+
+ oldcode = code;
+ } else {
+ return stbi__errpuc("illegal code in raster", "Corrupt GIF");
+ }
+ }
+ }
+}
+
+static void stbi__fill_gif_background(stbi__gif *g, int x0, int y0, int x1, int y1)
+{
+ int x, y;
+ stbi_uc *c = g->pal[g->bgindex];
+ for (y = y0; y < y1; y += 4 * g->w) {
+ for (x = x0; x < x1; x += 4) {
+ stbi_uc *p = &g->out[y + x];
+ p[0] = c[2];
+ p[1] = c[1];
+ p[2] = c[0];
+ p[3] = 0;
+ }
+ }
+}
+
+// this function is designed to support animated gifs, although stb_image doesn't support it
+static stbi_uc *stbi__gif_load_next(stbi__context *s, stbi__gif *g, int *comp, int req_comp)
+{
+ int i;
+ stbi_uc *prev_out = 0;
+
+ if (g->out == 0 && !stbi__gif_header(s, g, comp,0))
+ return 0; // stbi__g_failure_reason set by stbi__gif_header
+
+ if (!stbi__mad3sizes_valid(g->w, g->h, 4, 0))
+ return stbi__errpuc("too large", "GIF too large");
+
+ prev_out = g->out;
+ g->out = (stbi_uc *) stbi__malloc_mad3(4, g->w, g->h, 0);
+ if (g->out == 0) return stbi__errpuc("outofmem", "Out of memory");
+
+ switch ((g->eflags & 0x1C) >> 2) {
+ case 0: // unspecified (also always used on 1st frame)
+ stbi__fill_gif_background(g, 0, 0, 4 * g->w, 4 * g->w * g->h);
+ break;
+ case 1: // do not dispose
+ if (prev_out) memcpy(g->out, prev_out, 4 * g->w * g->h);
+ g->old_out = prev_out;
+ break;
+ case 2: // dispose to background
+ if (prev_out) memcpy(g->out, prev_out, 4 * g->w * g->h);
+ stbi__fill_gif_background(g, g->start_x, g->start_y, g->max_x, g->max_y);
+ break;
+ case 3: // dispose to previous
+ if (g->old_out) {
+ for (i = g->start_y; i < g->max_y; i += 4 * g->w)
+ memcpy(&g->out[i + g->start_x], &g->old_out[i + g->start_x], g->max_x - g->start_x);
+ }
+ break;
+ }
+
+ for (;;) {
+ switch (stbi__get8(s)) {
+ case 0x2C: /* Image Descriptor */
+ {
+ int prev_trans = -1;
+ stbi__int32 x, y, w, h;
+ stbi_uc *o;
+
+ x = stbi__get16le(s);
+ y = stbi__get16le(s);
+ w = stbi__get16le(s);
+ h = stbi__get16le(s);
+ if (((x + w) > (g->w)) || ((y + h) > (g->h)))
+ return stbi__errpuc("bad Image Descriptor", "Corrupt GIF");
+
+ g->line_size = g->w * 4;
+ g->start_x = x * 4;
+ g->start_y = y * g->line_size;
+ g->max_x = g->start_x + w * 4;
+ g->max_y = g->start_y + h * g->line_size;
+ g->cur_x = g->start_x;
+ g->cur_y = g->start_y;
+
+ g->lflags = stbi__get8(s);
+
+ if (g->lflags & 0x40) {
+ g->step = 8 * g->line_size; // first interlaced spacing
+ g->parse = 3;
+ } else {
+ g->step = g->line_size;
+ g->parse = 0;
+ }
+
+ if (g->lflags & 0x80) {
+ stbi__gif_parse_colortable(s,g->lpal, 2 << (g->lflags & 7), g->eflags & 0x01 ? g->transparent : -1);
+ g->color_table = (stbi_uc *) g->lpal;
+ } else if (g->flags & 0x80) {
+ if (g->transparent >= 0 && (g->eflags & 0x01)) {
+ prev_trans = g->pal[g->transparent][3];
+ g->pal[g->transparent][3] = 0;
+ }
+ g->color_table = (stbi_uc *) g->pal;
+ } else
+ return stbi__errpuc("missing color table", "Corrupt GIF");
+
+ o = stbi__process_gif_raster(s, g);
+ if (o == NULL) return NULL;
+
+ if (prev_trans != -1)
+ g->pal[g->transparent][3] = (stbi_uc) prev_trans;
+
+ return o;
+ }
+
+ case 0x21: // Comment Extension.
+ {
+ int len;
+ if (stbi__get8(s) == 0xF9) { // Graphic Control Extension.
+ len = stbi__get8(s);
+ if (len == 4) {
+ g->eflags = stbi__get8(s);
+ g->delay = stbi__get16le(s);
+ g->transparent = stbi__get8(s);
+ } else {
+ stbi__skip(s, len);
+ break;
+ }
+ }
+ while ((len = stbi__get8(s)) != 0)
+ stbi__skip(s, len);
+ break;
+ }
+
+ case 0x3B: // gif stream termination code
+ return (stbi_uc *) s; // using '1' causes warning on some compilers
+
+ default:
+ return stbi__errpuc("unknown code", "Corrupt GIF");
+ }
+ }
+
+ STBI_NOTUSED(req_comp);
+}
+
+static void *stbi__gif_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+ stbi_uc *u = 0;
+ stbi__gif* g = (stbi__gif*) stbi__malloc(sizeof(stbi__gif));
+ memset(g, 0, sizeof(*g));
+ STBI_NOTUSED(ri);
+
+ u = stbi__gif_load_next(s, g, comp, req_comp);
+ if (u == (stbi_uc *) s) u = 0; // end of animated gif marker
+ if (u) {
+ *x = g->w;
+ *y = g->h;
+ if (req_comp && req_comp != 4)
+ u = stbi__convert_format(u, 4, req_comp, g->w, g->h);
+ }
+ else if (g->out)
+ STBI_FREE(g->out);
+ STBI_FREE(g);
+ return u;
+}
+
+static int stbi__gif_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ return stbi__gif_info_raw(s,x,y,comp);
+}
+#endif
+
+// *************************************************************************************************
+// Radiance RGBE HDR loader
+// originally by Nicolas Schulz
+#ifndef STBI_NO_HDR
+static int stbi__hdr_test_core(stbi__context *s, const char *signature)
+{
+ int i;
+ for (i=0; signature[i]; ++i)
+ if (stbi__get8(s) != signature[i])
+ return 0;
+ stbi__rewind(s);
+ return 1;
+}
+
+static int stbi__hdr_test(stbi__context* s)
+{
+ int r = stbi__hdr_test_core(s, "#?RADIANCE\n");
+ stbi__rewind(s);
+ if(!r) {
+ r = stbi__hdr_test_core(s, "#?RGBE\n");
+ stbi__rewind(s);
+ }
+ return r;
+}
+
+#define STBI__HDR_BUFLEN 1024
+static char *stbi__hdr_gettoken(stbi__context *z, char *buffer)
+{
+ int len=0;
+ char c = '\0';
+
+ c = (char) stbi__get8(z);
+
+ while (!stbi__at_eof(z) && c != '\n') {
+ buffer[len++] = c;
+ if (len == STBI__HDR_BUFLEN-1) {
+ // flush to end of line
+ while (!stbi__at_eof(z) && stbi__get8(z) != '\n')
+ ;
+ break;
+ }
+ c = (char) stbi__get8(z);
+ }
+
+ buffer[len] = 0;
+ return buffer;
+}
+
+static void stbi__hdr_convert(float *output, stbi_uc *input, int req_comp)
+{
+ if ( input[3] != 0 ) {
+ float f1;
+ // Exponent
+ f1 = (float) ldexp(1.0f, input[3] - (int)(128 + 8));
+ if (req_comp <= 2)
+ output[0] = (input[0] + input[1] + input[2]) * f1 / 3;
+ else {
+ output[0] = input[0] * f1;
+ output[1] = input[1] * f1;
+ output[2] = input[2] * f1;
+ }
+ if (req_comp == 2) output[1] = 1;
+ if (req_comp == 4) output[3] = 1;
+ } else {
+ switch (req_comp) {
+ case 4: output[3] = 1; /* fallthrough */
+ case 3: output[0] = output[1] = output[2] = 0;
+ break;
+ case 2: output[1] = 1; /* fallthrough */
+ case 1: output[0] = 0;
+ break;
+ }
+ }
+}
+
+static float *stbi__hdr_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+ char buffer[STBI__HDR_BUFLEN];
+ char *token;
+ int valid = 0;
+ int width, height;
+ stbi_uc *scanline;
+ float *hdr_data;
+ int len;
+ unsigned char count, value;
+ int i, j, k, c1,c2, z;
+ const char *headerToken;
+ STBI_NOTUSED(ri);
+
+ // Check identifier
+ headerToken = stbi__hdr_gettoken(s,buffer);
+ if (strcmp(headerToken, "#?RADIANCE") != 0 && strcmp(headerToken, "#?RGBE") != 0)
+ return stbi__errpf("not HDR", "Corrupt HDR image");
+
+ // Parse header
+ for(;;) {
+ token = stbi__hdr_gettoken(s,buffer);
+ if (token[0] == 0) break;
+ if (strcmp(token, "FORMAT=32-bit_rle_rgbe") == 0) valid = 1;
+ }
+
+ if (!valid) return stbi__errpf("unsupported format", "Unsupported HDR format");
+
+ // Parse width and height
+ // can't use sscanf() if we're not using stdio!
+ token = stbi__hdr_gettoken(s,buffer);
+ if (strncmp(token, "-Y ", 3)) return stbi__errpf("unsupported data layout", "Unsupported HDR format");
+ token += 3;
+ height = (int) strtol(token, &token, 10);
+ while (*token == ' ') ++token;
+ if (strncmp(token, "+X ", 3)) return stbi__errpf("unsupported data layout", "Unsupported HDR format");
+ token += 3;
+ width = (int) strtol(token, NULL, 10);
+
+ *x = width;
+ *y = height;
+
+ if (comp) *comp = 3;
+ if (req_comp == 0) req_comp = 3;
+
+ if (!stbi__mad4sizes_valid(width, height, req_comp, sizeof(float), 0))
+ return stbi__errpf("too large", "HDR image is too large");
+
+ // Read data
+ hdr_data = (float *) stbi__malloc_mad4(width, height, req_comp, sizeof(float), 0);
+ if (!hdr_data)
+ return stbi__errpf("outofmem", "Out of memory");
+
+ // Load image data
+ // image data is stored as some number of sca
+ if ( width < 8 || width >= 32768) {
+ // Read flat data
+ for (j=0; j < height; ++j) {
+ for (i=0; i < width; ++i) {
+ stbi_uc rgbe[4];
+ main_decode_loop:
+ stbi__getn(s, rgbe, 4);
+ stbi__hdr_convert(hdr_data + j * width * req_comp + i * req_comp, rgbe, req_comp);
+ }
+ }
+ } else {
+ // Read RLE-encoded data
+ scanline = NULL;
+
+ for (j = 0; j < height; ++j) {
+ c1 = stbi__get8(s);
+ c2 = stbi__get8(s);
+ len = stbi__get8(s);
+ if (c1 != 2 || c2 != 2 || (len & 0x80)) {
+ // not run-length encoded, so we have to actually use THIS data as a decoded
+ // pixel (note this can't be a valid pixel--one of RGB must be >= 128)
+ stbi_uc rgbe[4];
+ rgbe[0] = (stbi_uc) c1;
+ rgbe[1] = (stbi_uc) c2;
+ rgbe[2] = (stbi_uc) len;
+ rgbe[3] = (stbi_uc) stbi__get8(s);
+ stbi__hdr_convert(hdr_data, rgbe, req_comp);
+ i = 1;
+ j = 0;
+ STBI_FREE(scanline);
+ goto main_decode_loop; // yes, this makes no sense
+ }
+ len <<= 8;
+ len |= stbi__get8(s);
+ if (len != width) { STBI_FREE(hdr_data); STBI_FREE(scanline); return stbi__errpf("invalid decoded scanline length", "corrupt HDR"); }
+ if (scanline == NULL) {
+ scanline = (stbi_uc *) stbi__malloc_mad2(width, 4, 0);
+ if (!scanline) {
+ STBI_FREE(hdr_data);
+ return stbi__errpf("outofmem", "Out of memory");
+ }
+ }
+
+ for (k = 0; k < 4; ++k) {
+ int nleft;
+ i = 0;
+ while ((nleft = width - i) > 0) {
+ count = stbi__get8(s);
+ if (count > 128) {
+ // Run
+ value = stbi__get8(s);
+ count -= 128;
+ if (count > nleft) { STBI_FREE(hdr_data); STBI_FREE(scanline); return stbi__errpf("corrupt", "bad RLE data in HDR"); }
+ for (z = 0; z < count; ++z)
+ scanline[i++ * 4 + k] = value;
+ } else {
+ // Dump
+ if (count > nleft) { STBI_FREE(hdr_data); STBI_FREE(scanline); return stbi__errpf("corrupt", "bad RLE data in HDR"); }
+ for (z = 0; z < count; ++z)
+ scanline[i++ * 4 + k] = stbi__get8(s);
+ }
+ }
+ }
+ for (i=0; i < width; ++i)
+ stbi__hdr_convert(hdr_data+(j*width + i)*req_comp, scanline + i*4, req_comp);
+ }
+ if (scanline)
+ STBI_FREE(scanline);
+ }
+
+ return hdr_data;
+}
+
+static int stbi__hdr_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ char buffer[STBI__HDR_BUFLEN];
+ char *token;
+ int valid = 0;
+ int dummy;
+
+ if (!x) x = &dummy;
+ if (!y) y = &dummy;
+ if (!comp) comp = &dummy;
+
+ if (stbi__hdr_test(s) == 0) {
+ stbi__rewind( s );
+ return 0;
+ }
+
+ for(;;) {
+ token = stbi__hdr_gettoken(s,buffer);
+ if (token[0] == 0) break;
+ if (strcmp(token, "FORMAT=32-bit_rle_rgbe") == 0) valid = 1;
+ }
+
+ if (!valid) {
+ stbi__rewind( s );
+ return 0;
+ }
+ token = stbi__hdr_gettoken(s,buffer);
+ if (strncmp(token, "-Y ", 3)) {
+ stbi__rewind( s );
+ return 0;
+ }
+ token += 3;
+ *y = (int) strtol(token, &token, 10);
+ while (*token == ' ') ++token;
+ if (strncmp(token, "+X ", 3)) {
+ stbi__rewind( s );
+ return 0;
+ }
+ token += 3;
+ *x = (int) strtol(token, NULL, 10);
+ *comp = 3;
+ return 1;
+}
+#endif // STBI_NO_HDR
+
+#ifndef STBI_NO_BMP
+static int stbi__bmp_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ void *p;
+ stbi__bmp_data info;
+
+ info.all_a = 255;
+ p = stbi__bmp_parse_header(s, &info);
+ stbi__rewind( s );
+ if (p == NULL)
+ return 0;
+ if (x) *x = s->img_x;
+ if (y) *y = s->img_y;
+ if (comp) *comp = info.ma ? 4 : 3;
+ return 1;
+}
+#endif
+
+#ifndef STBI_NO_PSD
+static int stbi__psd_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ int channelCount, dummy;
+ if (!x) x = &dummy;
+ if (!y) y = &dummy;
+ if (!comp) comp = &dummy;
+ if (stbi__get32be(s) != 0x38425053) {
+ stbi__rewind( s );
+ return 0;
+ }
+ if (stbi__get16be(s) != 1) {
+ stbi__rewind( s );
+ return 0;
+ }
+ stbi__skip(s, 6);
+ channelCount = stbi__get16be(s);
+ if (channelCount < 0 || channelCount > 16) {
+ stbi__rewind( s );
+ return 0;
+ }
+ *y = stbi__get32be(s);
+ *x = stbi__get32be(s);
+ if (stbi__get16be(s) != 8) {
+ stbi__rewind( s );
+ return 0;
+ }
+ if (stbi__get16be(s) != 3) {
+ stbi__rewind( s );
+ return 0;
+ }
+ *comp = 4;
+ return 1;
+}
+#endif
+
+#ifndef STBI_NO_PIC
+static int stbi__pic_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ int act_comp=0,num_packets=0,chained,dummy;
+ stbi__pic_packet packets[10];
+
+ if (!x) x = &dummy;
+ if (!y) y = &dummy;
+ if (!comp) comp = &dummy;
+
+ if (!stbi__pic_is4(s,"\x53\x80\xF6\x34")) {
+ stbi__rewind(s);
+ return 0;
+ }
+
+ stbi__skip(s, 88);
+
+ *x = stbi__get16be(s);
+ *y = stbi__get16be(s);
+ if (stbi__at_eof(s)) {
+ stbi__rewind( s);
+ return 0;
+ }
+ if ( (*x) != 0 && (1 << 28) / (*x) < (*y)) {
+ stbi__rewind( s );
+ return 0;
+ }
+
+ stbi__skip(s, 8);
+
+ do {
+ stbi__pic_packet *packet;
+
+ if (num_packets==sizeof(packets)/sizeof(packets[0]))
+ return 0;
+
+ packet = &packets[num_packets++];
+ chained = stbi__get8(s);
+ packet->size = stbi__get8(s);
+ packet->type = stbi__get8(s);
+ packet->channel = stbi__get8(s);
+ act_comp |= packet->channel;
+
+ if (stbi__at_eof(s)) {
+ stbi__rewind( s );
+ return 0;
+ }
+ if (packet->size != 8) {
+ stbi__rewind( s );
+ return 0;
+ }
+ } while (chained);
+
+ *comp = (act_comp & 0x10 ? 4 : 3);
+
+ return 1;
+}
+#endif
+
+// *************************************************************************************************
+// Portable Gray Map and Portable Pixel Map loader
+// by Ken Miller
+//
+// PGM: http://netpbm.sourceforge.net/doc/pgm.html
+// PPM: http://netpbm.sourceforge.net/doc/ppm.html
+//
+// Known limitations:
+// Does not support comments in the header section
+// Does not support ASCII image data (formats P2 and P3)
+// Does not support 16-bit-per-channel
+
+#ifndef STBI_NO_PNM
+
+static int stbi__pnm_test(stbi__context *s)
+{
+ char p, t;
+ p = (char) stbi__get8(s);
+ t = (char) stbi__get8(s);
+ if (p != 'P' || (t != '5' && t != '6')) {
+ stbi__rewind( s );
+ return 0;
+ }
+ return 1;
+}
+
+static void *stbi__pnm_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+ stbi_uc *out;
+ STBI_NOTUSED(ri);
+
+ if (!stbi__pnm_info(s, (int *)&s->img_x, (int *)&s->img_y, (int *)&s->img_n))
+ return 0;
+
+ *x = s->img_x;
+ *y = s->img_y;
+ if (comp) *comp = s->img_n;
+
+ if (!stbi__mad3sizes_valid(s->img_n, s->img_x, s->img_y, 0))
+ return stbi__errpuc("too large", "PNM too large");
+
+ out = (stbi_uc *) stbi__malloc_mad3(s->img_n, s->img_x, s->img_y, 0);
+ if (!out) return stbi__errpuc("outofmem", "Out of memory");
+ stbi__getn(s, out, s->img_n * s->img_x * s->img_y);
+
+ if (req_comp && req_comp != s->img_n) {
+ out = stbi__convert_format(out, s->img_n, req_comp, s->img_x, s->img_y);
+ if (out == NULL) return out; // stbi__convert_format frees input on failure
+ }
+ return out;
+}
+
+static int stbi__pnm_isspace(char c)
+{
+ return c == ' ' || c == '\t' || c == '\n' || c == '\v' || c == '\f' || c == '\r';
+}
+
+static void stbi__pnm_skip_whitespace(stbi__context *s, char *c)
+{
+ for (;;) {
+ while (!stbi__at_eof(s) && stbi__pnm_isspace(*c))
+ *c = (char) stbi__get8(s);
+
+ if (stbi__at_eof(s) || *c != '#')
+ break;
+
+ while (!stbi__at_eof(s) && *c != '\n' && *c != '\r' )
+ *c = (char) stbi__get8(s);
+ }
+}
+
+static int stbi__pnm_isdigit(char c)
+{
+ return c >= '0' && c <= '9';
+}
+
+static int stbi__pnm_getinteger(stbi__context *s, char *c)
+{
+ int value = 0;
+
+ while (!stbi__at_eof(s) && stbi__pnm_isdigit(*c)) {
+ value = value*10 + (*c - '0');
+ *c = (char) stbi__get8(s);
+ }
+
+ return value;
+}
+
+static int stbi__pnm_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ int maxv, dummy;
+ char c, p, t;
+
+ if (!x) x = &dummy;
+ if (!y) y = &dummy;
+ if (!comp) comp = &dummy;
+
+ stbi__rewind(s);
+
+ // Get identifier
+ p = (char) stbi__get8(s);
+ t = (char) stbi__get8(s);
+ if (p != 'P' || (t != '5' && t != '6')) {
+ stbi__rewind(s);
+ return 0;
+ }
+
+ *comp = (t == '6') ? 3 : 1; // '5' is 1-component .pgm; '6' is 3-component .ppm
+
+ c = (char) stbi__get8(s);
+ stbi__pnm_skip_whitespace(s, &c);
+
+ *x = stbi__pnm_getinteger(s, &c); // read width
+ stbi__pnm_skip_whitespace(s, &c);
+
+ *y = stbi__pnm_getinteger(s, &c); // read height
+ stbi__pnm_skip_whitespace(s, &c);
+
+ maxv = stbi__pnm_getinteger(s, &c); // read max value
+
+ if (maxv > 255)
+ return stbi__err("max value > 255", "PPM image not 8-bit");
+ else
+ return 1;
+}
+#endif
+
+static int stbi__info_main(stbi__context *s, int *x, int *y, int *comp)
+{
+ #ifndef STBI_NO_JPEG
+ if (stbi__jpeg_info(s, x, y, comp)) return 1;
+ #endif
+
+ #ifndef STBI_NO_PNG
+ if (stbi__png_info(s, x, y, comp)) return 1;
+ #endif
+
+ #ifndef STBI_NO_GIF
+ if (stbi__gif_info(s, x, y, comp)) return 1;
+ #endif
+
+ #ifndef STBI_NO_BMP
+ if (stbi__bmp_info(s, x, y, comp)) return 1;
+ #endif
+
+ #ifndef STBI_NO_PSD
+ if (stbi__psd_info(s, x, y, comp)) return 1;
+ #endif
+
+ #ifndef STBI_NO_PIC
+ if (stbi__pic_info(s, x, y, comp)) return 1;
+ #endif
+
+ #ifndef STBI_NO_PNM
+ if (stbi__pnm_info(s, x, y, comp)) return 1;
+ #endif
+
+ #ifndef STBI_NO_HDR
+ if (stbi__hdr_info(s, x, y, comp)) return 1;
+ #endif
+
+ // test tga last because it's a crappy test!
+ #ifndef STBI_NO_TGA
+ if (stbi__tga_info(s, x, y, comp))
+ return 1;
+ #endif
+ return stbi__err("unknown image type", "Image not of any known type, or corrupt");
+}
+
+#ifndef STBI_NO_STDIO
+STBIDEF int stbi_info(char const *filename, int *x, int *y, int *comp)
+{
+ FILE *f = stbi__fopen(filename, "rb");
+ int result;
+ if (!f) return stbi__err("can't fopen", "Unable to open file");
+ result = stbi_info_from_file(f, x, y, comp);
+ fclose(f);
+ return result;
+}
+
+STBIDEF int stbi_info_from_file(FILE *f, int *x, int *y, int *comp)
+{
+ int r;
+ stbi__context s;
+ long pos = ftell(f);
+ stbi__start_file(&s, f);
+ r = stbi__info_main(&s,x,y,comp);
+ fseek(f,pos,SEEK_SET);
+ return r;
+}
+#endif // !STBI_NO_STDIO
+
+STBIDEF int stbi_info_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp)
+{
+ stbi__context s;
+ stbi__start_mem(&s,buffer,len);
+ return stbi__info_main(&s,x,y,comp);
+}
+
+STBIDEF int stbi_info_from_callbacks(stbi_io_callbacks const *c, void *user, int *x, int *y, int *comp)
+{
+ stbi__context s;
+ stbi__start_callbacks(&s, (stbi_io_callbacks *) c, user);
+ return stbi__info_main(&s,x,y,comp);
+}
+
+#endif // STB_IMAGE_IMPLEMENTATION
+
+/*
+ revision history:
+ 2.15 (2017-03-18) fix png-1,2,4 bug; now all Imagenet JPGs decode;
+ warning fixes; disable run-time SSE detection on gcc;
+ uniform handling of optional "return" values;
+ thread-safe initialization of zlib tables
+ 2.14 (2017-03-03) remove deprecated STBI_JPEG_OLD; fixes for Imagenet JPGs
+ 2.13 (2016-11-29) add 16-bit API, only supported for PNG right now
+ 2.12 (2016-04-02) fix typo in 2.11 PSD fix that caused crashes
+ 2.11 (2016-04-02) allocate large structures on the stack
+ remove white matting for transparent PSD
+ fix reported channel count for PNG & BMP
+ re-enable SSE2 in non-gcc 64-bit
+ support RGB-formatted JPEG
+ read 16-bit PNGs (only as 8-bit)
+ 2.10 (2016-01-22) avoid warning introduced in 2.09 by STBI_REALLOC_SIZED
+ 2.09 (2016-01-16) allow comments in PNM files
+ 16-bit-per-pixel TGA (not bit-per-component)
+ info() for TGA could break due to .hdr handling
+ info() for BMP to shares code instead of sloppy parse
+ can use STBI_REALLOC_SIZED if allocator doesn't support realloc
+ code cleanup
+ 2.08 (2015-09-13) fix to 2.07 cleanup, reading RGB PSD as RGBA
+ 2.07 (2015-09-13) fix compiler warnings
+ partial animated GIF support
+ limited 16-bpc PSD support
+ #ifdef unused functions
+ bug with < 92 byte PIC,PNM,HDR,TGA
+ 2.06 (2015-04-19) fix bug where PSD returns wrong '*comp' value
+ 2.05 (2015-04-19) fix bug in progressive JPEG handling, fix warning
+ 2.04 (2015-04-15) try to re-enable SIMD on MinGW 64-bit
+ 2.03 (2015-04-12) extra corruption checking (mmozeiko)
+ stbi_set_flip_vertically_on_load (nguillemot)
+ fix NEON support; fix mingw support
+ 2.02 (2015-01-19) fix incorrect assert, fix warning
+ 2.01 (2015-01-17) fix various warnings; suppress SIMD on gcc 32-bit without -msse2
+ 2.00b (2014-12-25) fix STBI_MALLOC in progressive JPEG
+ 2.00 (2014-12-25) optimize JPG, including x86 SSE2 & NEON SIMD (ryg)
+ progressive JPEG (stb)
+ PGM/PPM support (Ken Miller)
+ STBI_MALLOC,STBI_REALLOC,STBI_FREE
+ GIF bugfix -- seemingly never worked
+ STBI_NO_*, STBI_ONLY_*
+ 1.48 (2014-12-14) fix incorrectly-named assert()
+ 1.47 (2014-12-14) 1/2/4-bit PNG support, both direct and paletted (Omar Cornut & stb)
+ optimize PNG (ryg)
+ fix bug in interlaced PNG with user-specified channel count (stb)
+ 1.46 (2014-08-26)
+ fix broken tRNS chunk (colorkey-style transparency) in non-paletted PNG
+ 1.45 (2014-08-16)
+ fix MSVC-ARM internal compiler error by wrapping malloc
+ 1.44 (2014-08-07)
+ various warning fixes from Ronny Chevalier
+ 1.43 (2014-07-15)
+ fix MSVC-only compiler problem in code changed in 1.42
+ 1.42 (2014-07-09)
+ don't define _CRT_SECURE_NO_WARNINGS (affects user code)
+ fixes to stbi__cleanup_jpeg path
+ added STBI_ASSERT to avoid requiring assert.h
+ 1.41 (2014-06-25)
+ fix search&replace from 1.36 that messed up comments/error messages
+ 1.40 (2014-06-22)
+ fix gcc struct-initialization warning
+ 1.39 (2014-06-15)
+ fix to TGA optimization when req_comp != number of components in TGA;
+ fix to GIF loading because BMP wasn't rewinding (whoops, no GIFs in my test suite)
+ add support for BMP version 5 (more ignored fields)
+ 1.38 (2014-06-06)
+ suppress MSVC warnings on integer casts truncating values
+ fix accidental rename of 'skip' field of I/O
+ 1.37 (2014-06-04)
+ remove duplicate typedef
+ 1.36 (2014-06-03)
+ convert to header file single-file library
+ if de-iphone isn't set, load iphone images color-swapped instead of returning NULL
+ 1.35 (2014-05-27)
+ various warnings
+ fix broken STBI_SIMD path
+ fix bug where stbi_load_from_file no longer left file pointer in correct place
+ fix broken non-easy path for 32-bit BMP (possibly never used)
+ TGA optimization by Arseny Kapoulkine
+ 1.34 (unknown)
+ use STBI_NOTUSED in stbi__resample_row_generic(), fix one more leak in tga failure case
+ 1.33 (2011-07-14)
+ make stbi_is_hdr work in STBI_NO_HDR (as specified), minor compiler-friendly improvements
+ 1.32 (2011-07-13)
+ support for "info" function for all supported filetypes (SpartanJ)
+ 1.31 (2011-06-20)
+ a few more leak fixes, bug in PNG handling (SpartanJ)
+ 1.30 (2011-06-11)
+ added ability to load files via callbacks to accomidate custom input streams (Ben Wenger)
+ removed deprecated format-specific test/load functions
+ removed support for installable file formats (stbi_loader) -- would have been broken for IO callbacks anyway
+ error cases in bmp and tga give messages and don't leak (Raymond Barbiero, grisha)
+ fix inefficiency in decoding 32-bit BMP (David Woo)
+ 1.29 (2010-08-16)
+ various warning fixes from Aurelien Pocheville
+ 1.28 (2010-08-01)
+ fix bug in GIF palette transparency (SpartanJ)
+ 1.27 (2010-08-01)
+ cast-to-stbi_uc to fix warnings
+ 1.26 (2010-07-24)
+ fix bug in file buffering for PNG reported by SpartanJ
+ 1.25 (2010-07-17)
+ refix trans_data warning (Won Chun)
+ 1.24 (2010-07-12)
+ perf improvements reading from files on platforms with lock-heavy fgetc()
+ minor perf improvements for jpeg
+ deprecated type-specific functions so we'll get feedback if they're needed
+ attempt to fix trans_data warning (Won Chun)
+ 1.23 fixed bug in iPhone support
+ 1.22 (2010-07-10)
+ removed image *writing* support
+ stbi_info support from Jetro Lauha
+ GIF support from Jean-Marc Lienher
+ iPhone PNG-extensions from James Brown
+ warning-fixes from Nicolas Schulz and Janez Zemva (i.stbi__err. Janez (U+017D)emva)
+ 1.21 fix use of 'stbi_uc' in header (reported by jon blow)
+ 1.20 added support for Softimage PIC, by Tom Seddon
+ 1.19 bug in interlaced PNG corruption check (found by ryg)
+ 1.18 (2008-08-02)
+ fix a threading bug (local mutable static)
+ 1.17 support interlaced PNG
+ 1.16 major bugfix - stbi__convert_format converted one too many pixels
+ 1.15 initialize some fields for thread safety
+ 1.14 fix threadsafe conversion bug
+ header-file-only version (#define STBI_HEADER_FILE_ONLY before including)
+ 1.13 threadsafe
+ 1.12 const qualifiers in the API
+ 1.11 Support installable IDCT, colorspace conversion routines
+ 1.10 Fixes for 64-bit (don't use "unsigned long")
+ optimized upsampling by Fabian "ryg" Giesen
+ 1.09 Fix format-conversion for PSD code (bad global variables!)
+ 1.08 Thatcher Ulrich's PSD code integrated by Nicolas Schulz
+ 1.07 attempt to fix C++ warning/errors again
+ 1.06 attempt to fix C++ warning/errors again
+ 1.05 fix TGA loading to return correct *comp and use good luminance calc
+ 1.04 default float alpha is 1, not 255; use 'void *' for stbi_image_free
+ 1.03 bugfixes to STBI_NO_STDIO, STBI_NO_HDR
+ 1.02 support for (subset of) HDR files, float interface for preferred access to them
+ 1.01 fix bug: possible bug in handling right-side up bmps... not sure
+ fix bug: the stbi__bmp_load() and stbi__tga_load() functions didn't work at all
+ 1.00 interface to zlib that skips zlib header
+ 0.99 correct handling of alpha in palette
+ 0.98 TGA loader by lonesock; dynamically add loaders (untested)
+ 0.97 jpeg errors on too large a file; also catch another malloc failure
+ 0.96 fix detection of invalid v value - particleman@mollyrocket forum
+ 0.95 during header scan, seek to markers in case of padding
+ 0.94 STBI_NO_STDIO to disable stdio usage; rename all #defines the same
+ 0.93 handle jpegtran output; verbose errors
+ 0.92 read 4,8,16,24,32-bit BMP files of several formats
+ 0.91 output 24-bit Windows 3.0 BMP files
+ 0.90 fix a few more warnings; bump version number to approach 1.0
+ 0.61 bugfixes due to Marc LeBlanc, Christopher Lloyd
+ 0.60 fix compiling as c++
+ 0.59 fix warnings: merge Dave Moore's -Wall fixes
+ 0.58 fix bug: zlib uncompressed mode len/nlen was wrong endian
+ 0.57 fix bug: jpg last huffman symbol before marker was >9 bits but less than 16 available
+ 0.56 fix bug: zlib uncompressed mode len vs. nlen
+ 0.55 fix bug: restart_interval not initialized to 0
+ 0.54 allow NULL for 'int *comp'
+ 0.53 fix bug in png 3->4; speedup png decoding
+ 0.52 png handles req_comp=3,4 directly; minor cleanup; jpeg comments
+ 0.51 obey req_comp requests, 1-component jpegs return as 1-component,
+ on 'test' only check type, not whether we support this variant
+ 0.50 (2006-11-19)
+ first released version
+*/
+
+
+/*
+------------------------------------------------------------------------------
+This software is available under 2 licenses -- choose whichever you prefer.
+------------------------------------------------------------------------------
+ALTERNATIVE A - MIT License
+Copyright (c) 2017 Sean Barrett
+Permission is hereby granted, free of charge, to any person obtaining a copy of
+this software and associated documentation files (the "Software"), to deal in
+the Software without restriction, including without limitation the rights to
+use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
+of the Software, and to permit persons to whom the Software is furnished to do
+so, subject to the following conditions:
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+------------------------------------------------------------------------------
+ALTERNATIVE B - Public Domain (www.unlicense.org)
+This is free and unencumbered software released into the public domain.
+Anyone is free to copy, modify, publish, use, compile, sell, or distribute this
+software, either in source code form or as a compiled binary, for any purpose,
+commercial or non-commercial, and by any means.
+In jurisdictions that recognize copyright laws, the author or authors of this
+software dedicate any and all copyright interest in the software to the public
+domain. We make this dedication for the benefit of the public at large and to
+the detriment of our heirs and successors. We intend this dedication to be an
+overt act of relinquishment in perpetuity of all present and future rights to
+this software under copyright law.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+------------------------------------------------------------------------------
+*/
--- /dev/null
+/* stb_image_resize - v0.94 - public domain image resizing
+ by Jorge L Rodriguez (@VinoBS) - 2014
+ http://github.com/nothings/stb
+
+ Written with emphasis on usability, portability, and efficiency. (No
+ SIMD or threads, so it be easily outperformed by libs that use those.)
+ Only scaling and translation is supported, no rotations or shears.
+ Easy API downsamples w/Mitchell filter, upsamples w/cubic interpolation.
+
+ COMPILING & LINKING
+ In one C/C++ file that #includes this file, do this:
+ #define STB_IMAGE_RESIZE_IMPLEMENTATION
+ before the #include. That will create the implementation in that file.
+
+ QUICKSTART
+ stbir_resize_uint8( input_pixels , in_w , in_h , 0,
+ output_pixels, out_w, out_h, 0, num_channels)
+ stbir_resize_float(...)
+ stbir_resize_uint8_srgb( input_pixels , in_w , in_h , 0,
+ output_pixels, out_w, out_h, 0,
+ num_channels , alpha_chan , 0)
+ stbir_resize_uint8_srgb_edgemode(
+ input_pixels , in_w , in_h , 0,
+ output_pixels, out_w, out_h, 0,
+ num_channels , alpha_chan , 0, STBIR_EDGE_CLAMP)
+ // WRAP/REFLECT/ZERO
+
+ FULL API
+ See the "header file" section of the source for API documentation.
+
+ ADDITIONAL DOCUMENTATION
+
+ SRGB & FLOATING POINT REPRESENTATION
+ The sRGB functions presume IEEE floating point. If you do not have
+ IEEE floating point, define STBIR_NON_IEEE_FLOAT. This will use
+ a slower implementation.
+
+ MEMORY ALLOCATION
+ The resize functions here perform a single memory allocation using
+ malloc. To control the memory allocation, before the #include that
+ triggers the implementation, do:
+
+ #define STBIR_MALLOC(size,context) ...
+ #define STBIR_FREE(ptr,context) ...
+
+ Each resize function makes exactly one call to malloc/free, so to use
+ temp memory, store the temp memory in the context and return that.
+
+ ASSERT
+ Define STBIR_ASSERT(boolval) to override assert() and not use assert.h
+
+ OPTIMIZATION
+ Define STBIR_SATURATE_INT to compute clamp values in-range using
+ integer operations instead of float operations. This may be faster
+ on some platforms.
+
+ DEFAULT FILTERS
+ For functions which don't provide explicit control over what filters
+ to use, you can change the compile-time defaults with
+
+ #define STBIR_DEFAULT_FILTER_UPSAMPLE STBIR_FILTER_something
+ #define STBIR_DEFAULT_FILTER_DOWNSAMPLE STBIR_FILTER_something
+
+ See stbir_filter in the header-file section for the list of filters.
+
+ NEW FILTERS
+ A number of 1D filter kernels are used. For a list of
+ supported filters see the stbir_filter enum. To add a new filter,
+ write a filter function and add it to stbir__filter_info_table.
+
+ PROGRESS
+ For interactive use with slow resize operations, you can install
+ a progress-report callback:
+
+ #define STBIR_PROGRESS_REPORT(val) some_func(val)
+
+ The parameter val is a float which goes from 0 to 1 as progress is made.
+
+ For example:
+
+ static void my_progress_report(float progress);
+ #define STBIR_PROGRESS_REPORT(val) my_progress_report(val)
+
+ #define STB_IMAGE_RESIZE_IMPLEMENTATION
+ #include "stb_image_resize.h"
+
+ static void my_progress_report(float progress)
+ {
+ printf("Progress: %f%%\n", progress*100);
+ }
+
+ MAX CHANNELS
+ If your image has more than 64 channels, define STBIR_MAX_CHANNELS
+ to the max you'll have.
+
+ ALPHA CHANNEL
+ Most of the resizing functions provide the ability to control how
+ the alpha channel of an image is processed. The important things
+ to know about this:
+
+ 1. The best mathematically-behaved version of alpha to use is
+ called "premultiplied alpha", in which the other color channels
+ have had the alpha value multiplied in. If you use premultiplied
+ alpha, linear filtering (such as image resampling done by this
+ library, or performed in texture units on GPUs) does the "right
+ thing". While premultiplied alpha is standard in the movie CGI
+ industry, it is still uncommon in the videogame/real-time world.
+
+ If you linearly filter non-premultiplied alpha, strange effects
+ occur. (For example, the 50/50 average of 99% transparent bright green
+ and 1% transparent black produces 50% transparent dark green when
+ non-premultiplied, whereas premultiplied it produces 50%
+ transparent near-black. The former introduces green energy
+ that doesn't exist in the source image.)
+
+ 2. Artists should not edit premultiplied-alpha images; artists
+ want non-premultiplied alpha images. Thus, art tools generally output
+ non-premultiplied alpha images.
+
+ 3. You will get best results in most cases by converting images
+ to premultiplied alpha before processing them mathematically.
+
+ 4. If you pass the flag STBIR_FLAG_ALPHA_PREMULTIPLIED, the
+ resizer does not do anything special for the alpha channel;
+ it is resampled identically to other channels. This produces
+ the correct results for premultiplied-alpha images, but produces
+ less-than-ideal results for non-premultiplied-alpha images.
+
+ 5. If you do not pass the flag STBIR_FLAG_ALPHA_PREMULTIPLIED,
+ then the resizer weights the contribution of input pixels
+ based on their alpha values, or, equivalently, it multiplies
+ the alpha value into the color channels, resamples, then divides
+ by the resultant alpha value. Input pixels which have alpha=0 do
+ not contribute at all to output pixels unless _all_ of the input
+ pixels affecting that output pixel have alpha=0, in which case
+ the result for that pixel is the same as it would be without
+ STBIR_FLAG_ALPHA_PREMULTIPLIED. However, this is only true for
+ input images in integer formats. For input images in float format,
+ input pixels with alpha=0 have no effect, and output pixels
+ which have alpha=0 will be 0 in all channels. (For float images,
+ you can manually achieve the same result by adding a tiny epsilon
+ value to the alpha channel of every image, and then subtracting
+ or clamping it at the end.)
+
+ 6. You can suppress the behavior described in #5 and make
+ all-0-alpha pixels have 0 in all channels by #defining
+ STBIR_NO_ALPHA_EPSILON.
+
+ 7. You can separately control whether the alpha channel is
+ interpreted as linear or affected by the colorspace. By default
+ it is linear; you almost never want to apply the colorspace.
+ (For example, graphics hardware does not apply sRGB conversion
+ to the alpha channel.)
+
+ CONTRIBUTORS
+ Jorge L Rodriguez: Implementation
+ Sean Barrett: API design, optimizations
+ Aras Pranckevicius: bugfix
+
+ REVISIONS
+ 0.94 (2017-03-18) fixed warnings
+ 0.93 (2017-03-03) fixed bug with certain combinations of heights
+ 0.92 (2017-01-02) fix integer overflow on large (>2GB) images
+ 0.91 (2016-04-02) fix warnings; fix handling of subpixel regions
+ 0.90 (2014-09-17) first released version
+
+ LICENSE
+ See end of file for license information.
+
+ TODO
+ Don't decode all of the image data when only processing a partial tile
+ Don't use full-width decode buffers when only processing a partial tile
+ When processing wide images, break processing into tiles so data fits in L1 cache
+ Installable filters?
+ Resize that respects alpha test coverage
+ (Reference code: FloatImage::alphaTestCoverage and FloatImage::scaleAlphaToCoverage:
+ https://code.google.com/p/nvidia-texture-tools/source/browse/trunk/src/nvimage/FloatImage.cpp )
+*/
+
+#ifndef STBIR_INCLUDE_STB_IMAGE_RESIZE_H
+#define STBIR_INCLUDE_STB_IMAGE_RESIZE_H
+
+#ifdef _MSC_VER
+typedef unsigned char stbir_uint8;
+typedef unsigned short stbir_uint16;
+typedef unsigned int stbir_uint32;
+#else
+#include <stdint.h>
+typedef uint8_t stbir_uint8;
+typedef uint16_t stbir_uint16;
+typedef uint32_t stbir_uint32;
+#endif
+
+#ifdef STB_IMAGE_RESIZE_STATIC
+#define STBIRDEF static
+#else
+#ifdef __cplusplus
+#define STBIRDEF extern "C"
+#else
+#define STBIRDEF extern
+#endif
+#endif
+
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// Easy-to-use API:
+//
+// * "input pixels" points to an array of image data with 'num_channels' channels (e.g. RGB=3, RGBA=4)
+// * input_w is input image width (x-axis), input_h is input image height (y-axis)
+// * stride is the offset between successive rows of image data in memory, in bytes. you can
+// specify 0 to mean packed continuously in memory
+// * alpha channel is treated identically to other channels.
+// * colorspace is linear or sRGB as specified by function name
+// * returned result is 1 for success or 0 in case of an error.
+// #define STBIR_ASSERT() to trigger an assert on parameter validation errors.
+// * Memory required grows approximately linearly with input and output size, but with
+// discontinuities at input_w == output_w and input_h == output_h.
+// * These functions use a "default" resampling filter defined at compile time. To change the filter,
+// you can change the compile-time defaults by #defining STBIR_DEFAULT_FILTER_UPSAMPLE
+// and STBIR_DEFAULT_FILTER_DOWNSAMPLE, or you can use the medium-complexity API.
+
+STBIRDEF int stbir_resize_uint8( const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels);
+
+STBIRDEF int stbir_resize_float( const float *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ float *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels);
+
+
+// The following functions interpret image data as gamma-corrected sRGB.
+// Specify STBIR_ALPHA_CHANNEL_NONE if you have no alpha channel,
+// or otherwise provide the index of the alpha channel. Flags value
+// of 0 will probably do the right thing if you're not sure what
+// the flags mean.
+
+#define STBIR_ALPHA_CHANNEL_NONE -1
+
+// Set this flag if your texture has premultiplied alpha. Otherwise, stbir will
+// use alpha-weighted resampling (effectively premultiplying, resampling,
+// then unpremultiplying).
+#define STBIR_FLAG_ALPHA_PREMULTIPLIED (1 << 0)
+// The specified alpha channel should be handled as gamma-corrected value even
+// when doing sRGB operations.
+#define STBIR_FLAG_ALPHA_USES_COLORSPACE (1 << 1)
+
+STBIRDEF int stbir_resize_uint8_srgb(const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags);
+
+
+typedef enum
+{
+ STBIR_EDGE_CLAMP = 1,
+ STBIR_EDGE_REFLECT = 2,
+ STBIR_EDGE_WRAP = 3,
+ STBIR_EDGE_ZERO = 4,
+} stbir_edge;
+
+// This function adds the ability to specify how requests to sample off the edge of the image are handled.
+STBIRDEF int stbir_resize_uint8_srgb_edgemode(const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode);
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// Medium-complexity API
+//
+// This extends the easy-to-use API as follows:
+//
+// * Alpha-channel can be processed separately
+// * If alpha_channel is not STBIR_ALPHA_CHANNEL_NONE
+// * Alpha channel will not be gamma corrected (unless flags&STBIR_FLAG_GAMMA_CORRECT)
+// * Filters will be weighted by alpha channel (unless flags&STBIR_FLAG_ALPHA_PREMULTIPLIED)
+// * Filter can be selected explicitly
+// * uint16 image type
+// * sRGB colorspace available for all types
+// * context parameter for passing to STBIR_MALLOC
+
+typedef enum
+{
+ STBIR_FILTER_DEFAULT = 0, // use same filter type that easy-to-use API chooses
+ STBIR_FILTER_BOX = 1, // A trapezoid w/1-pixel wide ramps, same result as box for integer scale ratios
+ STBIR_FILTER_TRIANGLE = 2, // On upsampling, produces same results as bilinear texture filtering
+ STBIR_FILTER_CUBICBSPLINE = 3, // The cubic b-spline (aka Mitchell-Netrevalli with B=1,C=0), gaussian-esque
+ STBIR_FILTER_CATMULLROM = 4, // An interpolating cubic spline
+ STBIR_FILTER_MITCHELL = 5, // Mitchell-Netrevalli filter with B=1/3, C=1/3
+} stbir_filter;
+
+typedef enum
+{
+ STBIR_COLORSPACE_LINEAR,
+ STBIR_COLORSPACE_SRGB,
+
+ STBIR_MAX_COLORSPACES,
+} stbir_colorspace;
+
+// The following functions are all identical except for the type of the image data
+
+STBIRDEF int stbir_resize_uint8_generic( const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space,
+ void *alloc_context);
+
+STBIRDEF int stbir_resize_uint16_generic(const stbir_uint16 *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ stbir_uint16 *output_pixels , int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space,
+ void *alloc_context);
+
+STBIRDEF int stbir_resize_float_generic( const float *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ float *output_pixels , int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space,
+ void *alloc_context);
+
+
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// Full-complexity API
+//
+// This extends the medium API as follows:
+//
+// * uint32 image type
+// * not typesafe
+// * separate filter types for each axis
+// * separate edge modes for each axis
+// * can specify scale explicitly for subpixel correctness
+// * can specify image source tile using texture coordinates
+
+typedef enum
+{
+ STBIR_TYPE_UINT8 ,
+ STBIR_TYPE_UINT16,
+ STBIR_TYPE_UINT32,
+ STBIR_TYPE_FLOAT ,
+
+ STBIR_MAX_TYPES
+} stbir_datatype;
+
+STBIRDEF int stbir_resize( const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ stbir_datatype datatype,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical,
+ stbir_filter filter_horizontal, stbir_filter filter_vertical,
+ stbir_colorspace space, void *alloc_context);
+
+STBIRDEF int stbir_resize_subpixel(const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ stbir_datatype datatype,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical,
+ stbir_filter filter_horizontal, stbir_filter filter_vertical,
+ stbir_colorspace space, void *alloc_context,
+ float x_scale, float y_scale,
+ float x_offset, float y_offset);
+
+STBIRDEF int stbir_resize_region( const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ stbir_datatype datatype,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical,
+ stbir_filter filter_horizontal, stbir_filter filter_vertical,
+ stbir_colorspace space, void *alloc_context,
+ float s0, float t0, float s1, float t1);
+// (s0, t0) & (s1, t1) are the top-left and bottom right corner (uv addressing style: [0, 1]x[0, 1]) of a region of the input image to use.
+
+//
+//
+//// end header file /////////////////////////////////////////////////////
+#endif // STBIR_INCLUDE_STB_IMAGE_RESIZE_H
+
+
+
+
+
+#ifdef STB_IMAGE_RESIZE_IMPLEMENTATION
+
+#ifndef STBIR_ASSERT
+#include <assert.h>
+#define STBIR_ASSERT(x) assert(x)
+#endif
+
+// For memset
+#include <string.h>
+
+#include <math.h>
+
+#ifndef STBIR_MALLOC
+#include <stdlib.h>
+#define STBIR_MALLOC(size,c) malloc(size)
+#define STBIR_FREE(ptr,c) free(ptr)
+#endif
+
+#ifndef _MSC_VER
+#ifdef __cplusplus
+#define stbir__inline inline
+#else
+#define stbir__inline
+#endif
+#else
+#define stbir__inline __forceinline
+#endif
+
+
+// should produce compiler error if size is wrong
+typedef unsigned char stbir__validate_uint32[sizeof(stbir_uint32) == 4 ? 1 : -1];
+
+#ifdef _MSC_VER
+#define STBIR__NOTUSED(v) (void)(v)
+#else
+#define STBIR__NOTUSED(v) (void)sizeof(v)
+#endif
+
+#define STBIR__ARRAY_SIZE(a) (sizeof((a))/sizeof((a)[0]))
+
+#ifndef STBIR_DEFAULT_FILTER_UPSAMPLE
+#define STBIR_DEFAULT_FILTER_UPSAMPLE STBIR_FILTER_CATMULLROM
+#endif
+
+#ifndef STBIR_DEFAULT_FILTER_DOWNSAMPLE
+#define STBIR_DEFAULT_FILTER_DOWNSAMPLE STBIR_FILTER_MITCHELL
+#endif
+
+#ifndef STBIR_PROGRESS_REPORT
+#define STBIR_PROGRESS_REPORT(float_0_to_1)
+#endif
+
+#ifndef STBIR_MAX_CHANNELS
+#define STBIR_MAX_CHANNELS 64
+#endif
+
+#if STBIR_MAX_CHANNELS > 65536
+#error "Too many channels; STBIR_MAX_CHANNELS must be no more than 65536."
+// because we store the indices in 16-bit variables
+#endif
+
+// This value is added to alpha just before premultiplication to avoid
+// zeroing out color values. It is equivalent to 2^-80. If you don't want
+// that behavior (it may interfere if you have floating point images with
+// very small alpha values) then you can define STBIR_NO_ALPHA_EPSILON to
+// disable it.
+#ifndef STBIR_ALPHA_EPSILON
+#define STBIR_ALPHA_EPSILON ((float)1 / (1 << 20) / (1 << 20) / (1 << 20) / (1 << 20))
+#endif
+
+
+
+#ifdef _MSC_VER
+#define STBIR__UNUSED_PARAM(v) (void)(v)
+#else
+#define STBIR__UNUSED_PARAM(v) (void)sizeof(v)
+#endif
+
+// must match stbir_datatype
+static unsigned char stbir__type_size[] = {
+ 1, // STBIR_TYPE_UINT8
+ 2, // STBIR_TYPE_UINT16
+ 4, // STBIR_TYPE_UINT32
+ 4, // STBIR_TYPE_FLOAT
+};
+
+// Kernel function centered at 0
+typedef float (stbir__kernel_fn)(float x, float scale);
+typedef float (stbir__support_fn)(float scale);
+
+typedef struct
+{
+ stbir__kernel_fn* kernel;
+ stbir__support_fn* support;
+} stbir__filter_info;
+
+// When upsampling, the contributors are which source pixels contribute.
+// When downsampling, the contributors are which destination pixels are contributed to.
+typedef struct
+{
+ int n0; // First contributing pixel
+ int n1; // Last contributing pixel
+} stbir__contributors;
+
+typedef struct
+{
+ const void* input_data;
+ int input_w;
+ int input_h;
+ int input_stride_bytes;
+
+ void* output_data;
+ int output_w;
+ int output_h;
+ int output_stride_bytes;
+
+ float s0, t0, s1, t1;
+
+ float horizontal_shift; // Units: output pixels
+ float vertical_shift; // Units: output pixels
+ float horizontal_scale;
+ float vertical_scale;
+
+ int channels;
+ int alpha_channel;
+ stbir_uint32 flags;
+ stbir_datatype type;
+ stbir_filter horizontal_filter;
+ stbir_filter vertical_filter;
+ stbir_edge edge_horizontal;
+ stbir_edge edge_vertical;
+ stbir_colorspace colorspace;
+
+ stbir__contributors* horizontal_contributors;
+ float* horizontal_coefficients;
+
+ stbir__contributors* vertical_contributors;
+ float* vertical_coefficients;
+
+ int decode_buffer_pixels;
+ float* decode_buffer;
+
+ float* horizontal_buffer;
+
+ // cache these because ceil/floor are inexplicably showing up in profile
+ int horizontal_coefficient_width;
+ int vertical_coefficient_width;
+ int horizontal_filter_pixel_width;
+ int vertical_filter_pixel_width;
+ int horizontal_filter_pixel_margin;
+ int vertical_filter_pixel_margin;
+ int horizontal_num_contributors;
+ int vertical_num_contributors;
+
+ int ring_buffer_length_bytes; // The length of an individual entry in the ring buffer. The total number of ring buffers is stbir__get_filter_pixel_width(filter)
+ int ring_buffer_num_entries; // Total number of entries in the ring buffer.
+ int ring_buffer_first_scanline;
+ int ring_buffer_last_scanline;
+ int ring_buffer_begin_index; // first_scanline is at this index in the ring buffer
+ float* ring_buffer;
+
+ float* encode_buffer; // A temporary buffer to store floats so we don't lose precision while we do multiply-adds.
+
+ int horizontal_contributors_size;
+ int horizontal_coefficients_size;
+ int vertical_contributors_size;
+ int vertical_coefficients_size;
+ int decode_buffer_size;
+ int horizontal_buffer_size;
+ int ring_buffer_size;
+ int encode_buffer_size;
+} stbir__info;
+
+
+static const float stbir__max_uint8_as_float = 255.0f;
+static const float stbir__max_uint16_as_float = 65535.0f;
+static const double stbir__max_uint32_as_float = 4294967295.0;
+
+
+static stbir__inline int stbir__min(int a, int b)
+{
+ return a < b ? a : b;
+}
+
+static stbir__inline float stbir__saturate(float x)
+{
+ if (x < 0)
+ return 0;
+
+ if (x > 1)
+ return 1;
+
+ return x;
+}
+
+#ifdef STBIR_SATURATE_INT
+static stbir__inline stbir_uint8 stbir__saturate8(int x)
+{
+ if ((unsigned int) x <= 255)
+ return x;
+
+ if (x < 0)
+ return 0;
+
+ return 255;
+}
+
+static stbir__inline stbir_uint16 stbir__saturate16(int x)
+{
+ if ((unsigned int) x <= 65535)
+ return x;
+
+ if (x < 0)
+ return 0;
+
+ return 65535;
+}
+#endif
+
+static float stbir__srgb_uchar_to_linear_float[256] = {
+ 0.000000f, 0.000304f, 0.000607f, 0.000911f, 0.001214f, 0.001518f, 0.001821f, 0.002125f, 0.002428f, 0.002732f, 0.003035f,
+ 0.003347f, 0.003677f, 0.004025f, 0.004391f, 0.004777f, 0.005182f, 0.005605f, 0.006049f, 0.006512f, 0.006995f, 0.007499f,
+ 0.008023f, 0.008568f, 0.009134f, 0.009721f, 0.010330f, 0.010960f, 0.011612f, 0.012286f, 0.012983f, 0.013702f, 0.014444f,
+ 0.015209f, 0.015996f, 0.016807f, 0.017642f, 0.018500f, 0.019382f, 0.020289f, 0.021219f, 0.022174f, 0.023153f, 0.024158f,
+ 0.025187f, 0.026241f, 0.027321f, 0.028426f, 0.029557f, 0.030713f, 0.031896f, 0.033105f, 0.034340f, 0.035601f, 0.036889f,
+ 0.038204f, 0.039546f, 0.040915f, 0.042311f, 0.043735f, 0.045186f, 0.046665f, 0.048172f, 0.049707f, 0.051269f, 0.052861f,
+ 0.054480f, 0.056128f, 0.057805f, 0.059511f, 0.061246f, 0.063010f, 0.064803f, 0.066626f, 0.068478f, 0.070360f, 0.072272f,
+ 0.074214f, 0.076185f, 0.078187f, 0.080220f, 0.082283f, 0.084376f, 0.086500f, 0.088656f, 0.090842f, 0.093059f, 0.095307f,
+ 0.097587f, 0.099899f, 0.102242f, 0.104616f, 0.107023f, 0.109462f, 0.111932f, 0.114435f, 0.116971f, 0.119538f, 0.122139f,
+ 0.124772f, 0.127438f, 0.130136f, 0.132868f, 0.135633f, 0.138432f, 0.141263f, 0.144128f, 0.147027f, 0.149960f, 0.152926f,
+ 0.155926f, 0.158961f, 0.162029f, 0.165132f, 0.168269f, 0.171441f, 0.174647f, 0.177888f, 0.181164f, 0.184475f, 0.187821f,
+ 0.191202f, 0.194618f, 0.198069f, 0.201556f, 0.205079f, 0.208637f, 0.212231f, 0.215861f, 0.219526f, 0.223228f, 0.226966f,
+ 0.230740f, 0.234551f, 0.238398f, 0.242281f, 0.246201f, 0.250158f, 0.254152f, 0.258183f, 0.262251f, 0.266356f, 0.270498f,
+ 0.274677f, 0.278894f, 0.283149f, 0.287441f, 0.291771f, 0.296138f, 0.300544f, 0.304987f, 0.309469f, 0.313989f, 0.318547f,
+ 0.323143f, 0.327778f, 0.332452f, 0.337164f, 0.341914f, 0.346704f, 0.351533f, 0.356400f, 0.361307f, 0.366253f, 0.371238f,
+ 0.376262f, 0.381326f, 0.386430f, 0.391573f, 0.396755f, 0.401978f, 0.407240f, 0.412543f, 0.417885f, 0.423268f, 0.428691f,
+ 0.434154f, 0.439657f, 0.445201f, 0.450786f, 0.456411f, 0.462077f, 0.467784f, 0.473532f, 0.479320f, 0.485150f, 0.491021f,
+ 0.496933f, 0.502887f, 0.508881f, 0.514918f, 0.520996f, 0.527115f, 0.533276f, 0.539480f, 0.545725f, 0.552011f, 0.558340f,
+ 0.564712f, 0.571125f, 0.577581f, 0.584078f, 0.590619f, 0.597202f, 0.603827f, 0.610496f, 0.617207f, 0.623960f, 0.630757f,
+ 0.637597f, 0.644480f, 0.651406f, 0.658375f, 0.665387f, 0.672443f, 0.679543f, 0.686685f, 0.693872f, 0.701102f, 0.708376f,
+ 0.715694f, 0.723055f, 0.730461f, 0.737911f, 0.745404f, 0.752942f, 0.760525f, 0.768151f, 0.775822f, 0.783538f, 0.791298f,
+ 0.799103f, 0.806952f, 0.814847f, 0.822786f, 0.830770f, 0.838799f, 0.846873f, 0.854993f, 0.863157f, 0.871367f, 0.879622f,
+ 0.887923f, 0.896269f, 0.904661f, 0.913099f, 0.921582f, 0.930111f, 0.938686f, 0.947307f, 0.955974f, 0.964686f, 0.973445f,
+ 0.982251f, 0.991102f, 1.0f
+};
+
+static float stbir__srgb_to_linear(float f)
+{
+ if (f <= 0.04045f)
+ return f / 12.92f;
+ else
+ return (float)pow((f + 0.055f) / 1.055f, 2.4f);
+}
+
+static float stbir__linear_to_srgb(float f)
+{
+ if (f <= 0.0031308f)
+ return f * 12.92f;
+ else
+ return 1.055f * (float)pow(f, 1 / 2.4f) - 0.055f;
+}
+
+#ifndef STBIR_NON_IEEE_FLOAT
+// From https://gist.github.com/rygorous/2203834
+
+typedef union
+{
+ stbir_uint32 u;
+ float f;
+} stbir__FP32;
+
+static const stbir_uint32 fp32_to_srgb8_tab4[104] = {
+ 0x0073000d, 0x007a000d, 0x0080000d, 0x0087000d, 0x008d000d, 0x0094000d, 0x009a000d, 0x00a1000d,
+ 0x00a7001a, 0x00b4001a, 0x00c1001a, 0x00ce001a, 0x00da001a, 0x00e7001a, 0x00f4001a, 0x0101001a,
+ 0x010e0033, 0x01280033, 0x01410033, 0x015b0033, 0x01750033, 0x018f0033, 0x01a80033, 0x01c20033,
+ 0x01dc0067, 0x020f0067, 0x02430067, 0x02760067, 0x02aa0067, 0x02dd0067, 0x03110067, 0x03440067,
+ 0x037800ce, 0x03df00ce, 0x044600ce, 0x04ad00ce, 0x051400ce, 0x057b00c5, 0x05dd00bc, 0x063b00b5,
+ 0x06970158, 0x07420142, 0x07e30130, 0x087b0120, 0x090b0112, 0x09940106, 0x0a1700fc, 0x0a9500f2,
+ 0x0b0f01cb, 0x0bf401ae, 0x0ccb0195, 0x0d950180, 0x0e56016e, 0x0f0d015e, 0x0fbc0150, 0x10630143,
+ 0x11070264, 0x1238023e, 0x1357021d, 0x14660201, 0x156601e9, 0x165a01d3, 0x174401c0, 0x182401af,
+ 0x18fe0331, 0x1a9602fe, 0x1c1502d2, 0x1d7e02ad, 0x1ed4028d, 0x201a0270, 0x21520256, 0x227d0240,
+ 0x239f0443, 0x25c003fe, 0x27bf03c4, 0x29a10392, 0x2b6a0367, 0x2d1d0341, 0x2ebe031f, 0x304d0300,
+ 0x31d105b0, 0x34a80555, 0x37520507, 0x39d504c5, 0x3c37048b, 0x3e7c0458, 0x40a8042a, 0x42bd0401,
+ 0x44c20798, 0x488e071e, 0x4c1c06b6, 0x4f76065d, 0x52a50610, 0x55ac05cc, 0x5892058f, 0x5b590559,
+ 0x5e0c0a23, 0x631c0980, 0x67db08f6, 0x6c55087f, 0x70940818, 0x74a007bd, 0x787d076c, 0x7c330723,
+};
+
+static stbir_uint8 stbir__linear_to_srgb_uchar(float in)
+{
+ static const stbir__FP32 almostone = { 0x3f7fffff }; // 1-eps
+ static const stbir__FP32 minval = { (127-13) << 23 };
+ stbir_uint32 tab,bias,scale,t;
+ stbir__FP32 f;
+
+ // Clamp to [2^(-13), 1-eps]; these two values map to 0 and 1, respectively.
+ // The tests are carefully written so that NaNs map to 0, same as in the reference
+ // implementation.
+ if (!(in > minval.f)) // written this way to catch NaNs
+ in = minval.f;
+ if (in > almostone.f)
+ in = almostone.f;
+
+ // Do the table lookup and unpack bias, scale
+ f.f = in;
+ tab = fp32_to_srgb8_tab4[(f.u - minval.u) >> 20];
+ bias = (tab >> 16) << 9;
+ scale = tab & 0xffff;
+
+ // Grab next-highest mantissa bits and perform linear interpolation
+ t = (f.u >> 12) & 0xff;
+ return (unsigned char) ((bias + scale*t) >> 16);
+}
+
+#else
+// sRGB transition values, scaled by 1<<28
+static int stbir__srgb_offset_to_linear_scaled[256] =
+{
+ 0, 40738, 122216, 203693, 285170, 366648, 448125, 529603,
+ 611080, 692557, 774035, 855852, 942009, 1033024, 1128971, 1229926,
+ 1335959, 1447142, 1563542, 1685229, 1812268, 1944725, 2082664, 2226148,
+ 2375238, 2529996, 2690481, 2856753, 3028870, 3206888, 3390865, 3580856,
+ 3776916, 3979100, 4187460, 4402049, 4622919, 4850123, 5083710, 5323731,
+ 5570236, 5823273, 6082892, 6349140, 6622065, 6901714, 7188133, 7481369,
+ 7781466, 8088471, 8402427, 8723380, 9051372, 9386448, 9728650, 10078021,
+ 10434603, 10798439, 11169569, 11548036, 11933879, 12327139, 12727857, 13136073,
+ 13551826, 13975156, 14406100, 14844697, 15290987, 15745007, 16206795, 16676389,
+ 17153826, 17639142, 18132374, 18633560, 19142734, 19659934, 20185196, 20718552,
+ 21260042, 21809696, 22367554, 22933648, 23508010, 24090680, 24681686, 25281066,
+ 25888850, 26505076, 27129772, 27762974, 28404716, 29055026, 29713942, 30381490,
+ 31057708, 31742624, 32436272, 33138682, 33849884, 34569912, 35298800, 36036568,
+ 36783260, 37538896, 38303512, 39077136, 39859796, 40651528, 41452360, 42262316,
+ 43081432, 43909732, 44747252, 45594016, 46450052, 47315392, 48190064, 49074096,
+ 49967516, 50870356, 51782636, 52704392, 53635648, 54576432, 55526772, 56486700,
+ 57456236, 58435408, 59424248, 60422780, 61431036, 62449032, 63476804, 64514376,
+ 65561776, 66619028, 67686160, 68763192, 69850160, 70947088, 72053992, 73170912,
+ 74297864, 75434880, 76581976, 77739184, 78906536, 80084040, 81271736, 82469648,
+ 83677792, 84896192, 86124888, 87363888, 88613232, 89872928, 91143016, 92423512,
+ 93714432, 95015816, 96327688, 97650056, 98982952, 100326408, 101680440, 103045072,
+ 104420320, 105806224, 107202800, 108610064, 110028048, 111456776, 112896264, 114346544,
+ 115807632, 117279552, 118762328, 120255976, 121760536, 123276016, 124802440, 126339832,
+ 127888216, 129447616, 131018048, 132599544, 134192112, 135795792, 137410592, 139036528,
+ 140673648, 142321952, 143981456, 145652208, 147334208, 149027488, 150732064, 152447968,
+ 154175200, 155913792, 157663776, 159425168, 161197984, 162982240, 164777968, 166585184,
+ 168403904, 170234160, 172075968, 173929344, 175794320, 177670896, 179559120, 181458992,
+ 183370528, 185293776, 187228736, 189175424, 191133888, 193104112, 195086128, 197079968,
+ 199085648, 201103184, 203132592, 205173888, 207227120, 209292272, 211369392, 213458480,
+ 215559568, 217672656, 219797792, 221934976, 224084240, 226245600, 228419056, 230604656,
+ 232802400, 235012320, 237234432, 239468736, 241715280, 243974080, 246245120, 248528464,
+ 250824112, 253132064, 255452368, 257785040, 260130080, 262487520, 264857376, 267239664,
+};
+
+static stbir_uint8 stbir__linear_to_srgb_uchar(float f)
+{
+ int x = (int) (f * (1 << 28)); // has headroom so you don't need to clamp
+ int v = 0;
+ int i;
+
+ // Refine the guess with a short binary search.
+ i = v + 128; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+ i = v + 64; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+ i = v + 32; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+ i = v + 16; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+ i = v + 8; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+ i = v + 4; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+ i = v + 2; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+ i = v + 1; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+
+ return (stbir_uint8) v;
+}
+#endif
+
+static float stbir__filter_trapezoid(float x, float scale)
+{
+ float halfscale = scale / 2;
+ float t = 0.5f + halfscale;
+ STBIR_ASSERT(scale <= 1);
+
+ x = (float)fabs(x);
+
+ if (x >= t)
+ return 0;
+ else
+ {
+ float r = 0.5f - halfscale;
+ if (x <= r)
+ return 1;
+ else
+ return (t - x) / scale;
+ }
+}
+
+static float stbir__support_trapezoid(float scale)
+{
+ STBIR_ASSERT(scale <= 1);
+ return 0.5f + scale / 2;
+}
+
+static float stbir__filter_triangle(float x, float s)
+{
+ STBIR__UNUSED_PARAM(s);
+
+ x = (float)fabs(x);
+
+ if (x <= 1.0f)
+ return 1 - x;
+ else
+ return 0;
+}
+
+static float stbir__filter_cubic(float x, float s)
+{
+ STBIR__UNUSED_PARAM(s);
+
+ x = (float)fabs(x);
+
+ if (x < 1.0f)
+ return (4 + x*x*(3*x - 6))/6;
+ else if (x < 2.0f)
+ return (8 + x*(-12 + x*(6 - x)))/6;
+
+ return (0.0f);
+}
+
+static float stbir__filter_catmullrom(float x, float s)
+{
+ STBIR__UNUSED_PARAM(s);
+
+ x = (float)fabs(x);
+
+ if (x < 1.0f)
+ return 1 - x*x*(2.5f - 1.5f*x);
+ else if (x < 2.0f)
+ return 2 - x*(4 + x*(0.5f*x - 2.5f));
+
+ return (0.0f);
+}
+
+static float stbir__filter_mitchell(float x, float s)
+{
+ STBIR__UNUSED_PARAM(s);
+
+ x = (float)fabs(x);
+
+ if (x < 1.0f)
+ return (16 + x*x*(21 * x - 36))/18;
+ else if (x < 2.0f)
+ return (32 + x*(-60 + x*(36 - 7*x)))/18;
+
+ return (0.0f);
+}
+
+static float stbir__support_zero(float s)
+{
+ STBIR__UNUSED_PARAM(s);
+ return 0;
+}
+
+static float stbir__support_one(float s)
+{
+ STBIR__UNUSED_PARAM(s);
+ return 1;
+}
+
+static float stbir__support_two(float s)
+{
+ STBIR__UNUSED_PARAM(s);
+ return 2;
+}
+
+static stbir__filter_info stbir__filter_info_table[] = {
+ { NULL, stbir__support_zero },
+ { stbir__filter_trapezoid, stbir__support_trapezoid },
+ { stbir__filter_triangle, stbir__support_one },
+ { stbir__filter_cubic, stbir__support_two },
+ { stbir__filter_catmullrom, stbir__support_two },
+ { stbir__filter_mitchell, stbir__support_two },
+};
+
+stbir__inline static int stbir__use_upsampling(float ratio)
+{
+ return ratio > 1;
+}
+
+stbir__inline static int stbir__use_width_upsampling(stbir__info* stbir_info)
+{
+ return stbir__use_upsampling(stbir_info->horizontal_scale);
+}
+
+stbir__inline static int stbir__use_height_upsampling(stbir__info* stbir_info)
+{
+ return stbir__use_upsampling(stbir_info->vertical_scale);
+}
+
+// This is the maximum number of input samples that can affect an output sample
+// with the given filter
+static int stbir__get_filter_pixel_width(stbir_filter filter, float scale)
+{
+ STBIR_ASSERT(filter != 0);
+ STBIR_ASSERT(filter < STBIR__ARRAY_SIZE(stbir__filter_info_table));
+
+ if (stbir__use_upsampling(scale))
+ return (int)ceil(stbir__filter_info_table[filter].support(1/scale) * 2);
+ else
+ return (int)ceil(stbir__filter_info_table[filter].support(scale) * 2 / scale);
+}
+
+// This is how much to expand buffers to account for filters seeking outside
+// the image boundaries.
+static int stbir__get_filter_pixel_margin(stbir_filter filter, float scale)
+{
+ return stbir__get_filter_pixel_width(filter, scale) / 2;
+}
+
+static int stbir__get_coefficient_width(stbir_filter filter, float scale)
+{
+ if (stbir__use_upsampling(scale))
+ return (int)ceil(stbir__filter_info_table[filter].support(1 / scale) * 2);
+ else
+ return (int)ceil(stbir__filter_info_table[filter].support(scale) * 2);
+}
+
+static int stbir__get_contributors(float scale, stbir_filter filter, int input_size, int output_size)
+{
+ if (stbir__use_upsampling(scale))
+ return output_size;
+ else
+ return (input_size + stbir__get_filter_pixel_margin(filter, scale) * 2);
+}
+
+static int stbir__get_total_horizontal_coefficients(stbir__info* info)
+{
+ return info->horizontal_num_contributors
+ * stbir__get_coefficient_width (info->horizontal_filter, info->horizontal_scale);
+}
+
+static int stbir__get_total_vertical_coefficients(stbir__info* info)
+{
+ return info->vertical_num_contributors
+ * stbir__get_coefficient_width (info->vertical_filter, info->vertical_scale);
+}
+
+static stbir__contributors* stbir__get_contributor(stbir__contributors* contributors, int n)
+{
+ return &contributors[n];
+}
+
+// For perf reasons this code is duplicated in stbir__resample_horizontal_upsample/downsample,
+// if you change it here change it there too.
+static float* stbir__get_coefficient(float* coefficients, stbir_filter filter, float scale, int n, int c)
+{
+ int width = stbir__get_coefficient_width(filter, scale);
+ return &coefficients[width*n + c];
+}
+
+static int stbir__edge_wrap_slow(stbir_edge edge, int n, int max)
+{
+ switch (edge)
+ {
+ case STBIR_EDGE_ZERO:
+ return 0; // we'll decode the wrong pixel here, and then overwrite with 0s later
+
+ case STBIR_EDGE_CLAMP:
+ if (n < 0)
+ return 0;
+
+ if (n >= max)
+ return max - 1;
+
+ return n; // NOTREACHED
+
+ case STBIR_EDGE_REFLECT:
+ {
+ if (n < 0)
+ {
+ if (n < max)
+ return -n;
+ else
+ return max - 1;
+ }
+
+ if (n >= max)
+ {
+ int max2 = max * 2;
+ if (n >= max2)
+ return 0;
+ else
+ return max2 - n - 1;
+ }
+
+ return n; // NOTREACHED
+ }
+
+ case STBIR_EDGE_WRAP:
+ if (n >= 0)
+ return (n % max);
+ else
+ {
+ int m = (-n) % max;
+
+ if (m != 0)
+ m = max - m;
+
+ return (m);
+ }
+ return n; // NOTREACHED
+
+ default:
+ STBIR_ASSERT(!"Unimplemented edge type");
+ return 0;
+ }
+}
+
+stbir__inline static int stbir__edge_wrap(stbir_edge edge, int n, int max)
+{
+ // avoid per-pixel switch
+ if (n >= 0 && n < max)
+ return n;
+ return stbir__edge_wrap_slow(edge, n, max);
+}
+
+// What input pixels contribute to this output pixel?
+static void stbir__calculate_sample_range_upsample(int n, float out_filter_radius, float scale_ratio, float out_shift, int* in_first_pixel, int* in_last_pixel, float* in_center_of_out)
+{
+ float out_pixel_center = (float)n + 0.5f;
+ float out_pixel_influence_lowerbound = out_pixel_center - out_filter_radius;
+ float out_pixel_influence_upperbound = out_pixel_center + out_filter_radius;
+
+ float in_pixel_influence_lowerbound = (out_pixel_influence_lowerbound + out_shift) / scale_ratio;
+ float in_pixel_influence_upperbound = (out_pixel_influence_upperbound + out_shift) / scale_ratio;
+
+ *in_center_of_out = (out_pixel_center + out_shift) / scale_ratio;
+ *in_first_pixel = (int)(floor(in_pixel_influence_lowerbound + 0.5));
+ *in_last_pixel = (int)(floor(in_pixel_influence_upperbound - 0.5));
+}
+
+// What output pixels does this input pixel contribute to?
+static void stbir__calculate_sample_range_downsample(int n, float in_pixels_radius, float scale_ratio, float out_shift, int* out_first_pixel, int* out_last_pixel, float* out_center_of_in)
+{
+ float in_pixel_center = (float)n + 0.5f;
+ float in_pixel_influence_lowerbound = in_pixel_center - in_pixels_radius;
+ float in_pixel_influence_upperbound = in_pixel_center + in_pixels_radius;
+
+ float out_pixel_influence_lowerbound = in_pixel_influence_lowerbound * scale_ratio - out_shift;
+ float out_pixel_influence_upperbound = in_pixel_influence_upperbound * scale_ratio - out_shift;
+
+ *out_center_of_in = in_pixel_center * scale_ratio - out_shift;
+ *out_first_pixel = (int)(floor(out_pixel_influence_lowerbound + 0.5));
+ *out_last_pixel = (int)(floor(out_pixel_influence_upperbound - 0.5));
+}
+
+static void stbir__calculate_coefficients_upsample(stbir_filter filter, float scale, int in_first_pixel, int in_last_pixel, float in_center_of_out, stbir__contributors* contributor, float* coefficient_group)
+{
+ int i;
+ float total_filter = 0;
+ float filter_scale;
+
+ STBIR_ASSERT(in_last_pixel - in_first_pixel <= (int)ceil(stbir__filter_info_table[filter].support(1/scale) * 2)); // Taken directly from stbir__get_coefficient_width() which we can't call because we don't know if we're horizontal or vertical.
+
+ contributor->n0 = in_first_pixel;
+ contributor->n1 = in_last_pixel;
+
+ STBIR_ASSERT(contributor->n1 >= contributor->n0);
+
+ for (i = 0; i <= in_last_pixel - in_first_pixel; i++)
+ {
+ float in_pixel_center = (float)(i + in_first_pixel) + 0.5f;
+ coefficient_group[i] = stbir__filter_info_table[filter].kernel(in_center_of_out - in_pixel_center, 1 / scale);
+
+ // If the coefficient is zero, skip it. (Don't do the <0 check here, we want the influence of those outside pixels.)
+ if (i == 0 && !coefficient_group[i])
+ {
+ contributor->n0 = ++in_first_pixel;
+ i--;
+ continue;
+ }
+
+ total_filter += coefficient_group[i];
+ }
+
+ STBIR_ASSERT(stbir__filter_info_table[filter].kernel((float)(in_last_pixel + 1) + 0.5f - in_center_of_out, 1/scale) == 0);
+
+ STBIR_ASSERT(total_filter > 0.9);
+ STBIR_ASSERT(total_filter < 1.1f); // Make sure it's not way off.
+
+ // Make sure the sum of all coefficients is 1.
+ filter_scale = 1 / total_filter;
+
+ for (i = 0; i <= in_last_pixel - in_first_pixel; i++)
+ coefficient_group[i] *= filter_scale;
+
+ for (i = in_last_pixel - in_first_pixel; i >= 0; i--)
+ {
+ if (coefficient_group[i])
+ break;
+
+ // This line has no weight. We can skip it.
+ contributor->n1 = contributor->n0 + i - 1;
+ }
+}
+
+static void stbir__calculate_coefficients_downsample(stbir_filter filter, float scale_ratio, int out_first_pixel, int out_last_pixel, float out_center_of_in, stbir__contributors* contributor, float* coefficient_group)
+{
+ int i;
+
+ STBIR_ASSERT(out_last_pixel - out_first_pixel <= (int)ceil(stbir__filter_info_table[filter].support(scale_ratio) * 2)); // Taken directly from stbir__get_coefficient_width() which we can't call because we don't know if we're horizontal or vertical.
+
+ contributor->n0 = out_first_pixel;
+ contributor->n1 = out_last_pixel;
+
+ STBIR_ASSERT(contributor->n1 >= contributor->n0);
+
+ for (i = 0; i <= out_last_pixel - out_first_pixel; i++)
+ {
+ float out_pixel_center = (float)(i + out_first_pixel) + 0.5f;
+ float x = out_pixel_center - out_center_of_in;
+ coefficient_group[i] = stbir__filter_info_table[filter].kernel(x, scale_ratio) * scale_ratio;
+ }
+
+ STBIR_ASSERT(stbir__filter_info_table[filter].kernel((float)(out_last_pixel + 1) + 0.5f - out_center_of_in, scale_ratio) == 0);
+
+ for (i = out_last_pixel - out_first_pixel; i >= 0; i--)
+ {
+ if (coefficient_group[i])
+ break;
+
+ // This line has no weight. We can skip it.
+ contributor->n1 = contributor->n0 + i - 1;
+ }
+}
+
+static void stbir__normalize_downsample_coefficients(stbir__contributors* contributors, float* coefficients, stbir_filter filter, float scale_ratio, int input_size, int output_size)
+{
+ int num_contributors = stbir__get_contributors(scale_ratio, filter, input_size, output_size);
+ int num_coefficients = stbir__get_coefficient_width(filter, scale_ratio);
+ int i, j;
+ int skip;
+
+ for (i = 0; i < output_size; i++)
+ {
+ float scale;
+ float total = 0;
+
+ for (j = 0; j < num_contributors; j++)
+ {
+ if (i >= contributors[j].n0 && i <= contributors[j].n1)
+ {
+ float coefficient = *stbir__get_coefficient(coefficients, filter, scale_ratio, j, i - contributors[j].n0);
+ total += coefficient;
+ }
+ else if (i < contributors[j].n0)
+ break;
+ }
+
+ STBIR_ASSERT(total > 0.9f);
+ STBIR_ASSERT(total < 1.1f);
+
+ scale = 1 / total;
+
+ for (j = 0; j < num_contributors; j++)
+ {
+ if (i >= contributors[j].n0 && i <= contributors[j].n1)
+ *stbir__get_coefficient(coefficients, filter, scale_ratio, j, i - contributors[j].n0) *= scale;
+ else if (i < contributors[j].n0)
+ break;
+ }
+ }
+
+ // Optimize: Skip zero coefficients and contributions outside of image bounds.
+ // Do this after normalizing because normalization depends on the n0/n1 values.
+ for (j = 0; j < num_contributors; j++)
+ {
+ int range, max, width;
+
+ skip = 0;
+ while (*stbir__get_coefficient(coefficients, filter, scale_ratio, j, skip) == 0)
+ skip++;
+
+ contributors[j].n0 += skip;
+
+ while (contributors[j].n0 < 0)
+ {
+ contributors[j].n0++;
+ skip++;
+ }
+
+ range = contributors[j].n1 - contributors[j].n0 + 1;
+ max = stbir__min(num_coefficients, range);
+
+ width = stbir__get_coefficient_width(filter, scale_ratio);
+ for (i = 0; i < max; i++)
+ {
+ if (i + skip >= width)
+ break;
+
+ *stbir__get_coefficient(coefficients, filter, scale_ratio, j, i) = *stbir__get_coefficient(coefficients, filter, scale_ratio, j, i + skip);
+ }
+
+ continue;
+ }
+
+ // Using min to avoid writing into invalid pixels.
+ for (i = 0; i < num_contributors; i++)
+ contributors[i].n1 = stbir__min(contributors[i].n1, output_size - 1);
+}
+
+// Each scan line uses the same kernel values so we should calculate the kernel
+// values once and then we can use them for every scan line.
+static void stbir__calculate_filters(stbir__contributors* contributors, float* coefficients, stbir_filter filter, float scale_ratio, float shift, int input_size, int output_size)
+{
+ int n;
+ int total_contributors = stbir__get_contributors(scale_ratio, filter, input_size, output_size);
+
+ if (stbir__use_upsampling(scale_ratio))
+ {
+ float out_pixels_radius = stbir__filter_info_table[filter].support(1 / scale_ratio) * scale_ratio;
+
+ // Looping through out pixels
+ for (n = 0; n < total_contributors; n++)
+ {
+ float in_center_of_out; // Center of the current out pixel in the in pixel space
+ int in_first_pixel, in_last_pixel;
+
+ stbir__calculate_sample_range_upsample(n, out_pixels_radius, scale_ratio, shift, &in_first_pixel, &in_last_pixel, &in_center_of_out);
+
+ stbir__calculate_coefficients_upsample(filter, scale_ratio, in_first_pixel, in_last_pixel, in_center_of_out, stbir__get_contributor(contributors, n), stbir__get_coefficient(coefficients, filter, scale_ratio, n, 0));
+ }
+ }
+ else
+ {
+ float in_pixels_radius = stbir__filter_info_table[filter].support(scale_ratio) / scale_ratio;
+
+ // Looping through in pixels
+ for (n = 0; n < total_contributors; n++)
+ {
+ float out_center_of_in; // Center of the current out pixel in the in pixel space
+ int out_first_pixel, out_last_pixel;
+ int n_adjusted = n - stbir__get_filter_pixel_margin(filter, scale_ratio);
+
+ stbir__calculate_sample_range_downsample(n_adjusted, in_pixels_radius, scale_ratio, shift, &out_first_pixel, &out_last_pixel, &out_center_of_in);
+
+ stbir__calculate_coefficients_downsample(filter, scale_ratio, out_first_pixel, out_last_pixel, out_center_of_in, stbir__get_contributor(contributors, n), stbir__get_coefficient(coefficients, filter, scale_ratio, n, 0));
+ }
+
+ stbir__normalize_downsample_coefficients(contributors, coefficients, filter, scale_ratio, input_size, output_size);
+ }
+}
+
+static float* stbir__get_decode_buffer(stbir__info* stbir_info)
+{
+ // The 0 index of the decode buffer starts after the margin. This makes
+ // it okay to use negative indexes on the decode buffer.
+ return &stbir_info->decode_buffer[stbir_info->horizontal_filter_pixel_margin * stbir_info->channels];
+}
+
+#define STBIR__DECODE(type, colorspace) ((type) * (STBIR_MAX_COLORSPACES) + (colorspace))
+
+static void stbir__decode_scanline(stbir__info* stbir_info, int n)
+{
+ int c;
+ int channels = stbir_info->channels;
+ int alpha_channel = stbir_info->alpha_channel;
+ int type = stbir_info->type;
+ int colorspace = stbir_info->colorspace;
+ int input_w = stbir_info->input_w;
+ size_t input_stride_bytes = stbir_info->input_stride_bytes;
+ float* decode_buffer = stbir__get_decode_buffer(stbir_info);
+ stbir_edge edge_horizontal = stbir_info->edge_horizontal;
+ stbir_edge edge_vertical = stbir_info->edge_vertical;
+ size_t in_buffer_row_offset = stbir__edge_wrap(edge_vertical, n, stbir_info->input_h) * input_stride_bytes;
+ const void* input_data = (char *) stbir_info->input_data + in_buffer_row_offset;
+ int max_x = input_w + stbir_info->horizontal_filter_pixel_margin;
+ int decode = STBIR__DECODE(type, colorspace);
+
+ int x = -stbir_info->horizontal_filter_pixel_margin;
+
+ // special handling for STBIR_EDGE_ZERO because it needs to return an item that doesn't appear in the input,
+ // and we want to avoid paying overhead on every pixel if not STBIR_EDGE_ZERO
+ if (edge_vertical == STBIR_EDGE_ZERO && (n < 0 || n >= stbir_info->input_h))
+ {
+ for (; x < max_x; x++)
+ for (c = 0; c < channels; c++)
+ decode_buffer[x*channels + c] = 0;
+ return;
+ }
+
+ switch (decode)
+ {
+ case STBIR__DECODE(STBIR_TYPE_UINT8, STBIR_COLORSPACE_LINEAR):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = ((float)((const unsigned char*)input_data)[input_pixel_index + c]) / stbir__max_uint8_as_float;
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT8, STBIR_COLORSPACE_SRGB):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = stbir__srgb_uchar_to_linear_float[((const unsigned char*)input_data)[input_pixel_index + c]];
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ decode_buffer[decode_pixel_index + alpha_channel] = ((float)((const unsigned char*)input_data)[input_pixel_index + alpha_channel]) / stbir__max_uint8_as_float;
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT16, STBIR_COLORSPACE_LINEAR):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = ((float)((const unsigned short*)input_data)[input_pixel_index + c]) / stbir__max_uint16_as_float;
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT16, STBIR_COLORSPACE_SRGB):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = stbir__srgb_to_linear(((float)((const unsigned short*)input_data)[input_pixel_index + c]) / stbir__max_uint16_as_float);
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ decode_buffer[decode_pixel_index + alpha_channel] = ((float)((const unsigned short*)input_data)[input_pixel_index + alpha_channel]) / stbir__max_uint16_as_float;
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT32, STBIR_COLORSPACE_LINEAR):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = (float)(((double)((const unsigned int*)input_data)[input_pixel_index + c]) / stbir__max_uint32_as_float);
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT32, STBIR_COLORSPACE_SRGB):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = stbir__srgb_to_linear((float)(((double)((const unsigned int*)input_data)[input_pixel_index + c]) / stbir__max_uint32_as_float));
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ decode_buffer[decode_pixel_index + alpha_channel] = (float)(((double)((const unsigned int*)input_data)[input_pixel_index + alpha_channel]) / stbir__max_uint32_as_float);
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_FLOAT, STBIR_COLORSPACE_LINEAR):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = ((const float*)input_data)[input_pixel_index + c];
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_FLOAT, STBIR_COLORSPACE_SRGB):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = stbir__srgb_to_linear(((const float*)input_data)[input_pixel_index + c]);
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ decode_buffer[decode_pixel_index + alpha_channel] = ((const float*)input_data)[input_pixel_index + alpha_channel];
+ }
+
+ break;
+
+ default:
+ STBIR_ASSERT(!"Unknown type/colorspace/channels combination.");
+ break;
+ }
+
+ if (!(stbir_info->flags & STBIR_FLAG_ALPHA_PREMULTIPLIED))
+ {
+ for (x = -stbir_info->horizontal_filter_pixel_margin; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+
+ // If the alpha value is 0 it will clobber the color values. Make sure it's not.
+ float alpha = decode_buffer[decode_pixel_index + alpha_channel];
+#ifndef STBIR_NO_ALPHA_EPSILON
+ if (stbir_info->type != STBIR_TYPE_FLOAT) {
+ alpha += STBIR_ALPHA_EPSILON;
+ decode_buffer[decode_pixel_index + alpha_channel] = alpha;
+ }
+#endif
+ for (c = 0; c < channels; c++)
+ {
+ if (c == alpha_channel)
+ continue;
+
+ decode_buffer[decode_pixel_index + c] *= alpha;
+ }
+ }
+ }
+
+ if (edge_horizontal == STBIR_EDGE_ZERO)
+ {
+ for (x = -stbir_info->horizontal_filter_pixel_margin; x < 0; x++)
+ {
+ for (c = 0; c < channels; c++)
+ decode_buffer[x*channels + c] = 0;
+ }
+ for (x = input_w; x < max_x; x++)
+ {
+ for (c = 0; c < channels; c++)
+ decode_buffer[x*channels + c] = 0;
+ }
+ }
+}
+
+static float* stbir__get_ring_buffer_entry(float* ring_buffer, int index, int ring_buffer_length)
+{
+ return &ring_buffer[index * ring_buffer_length];
+}
+
+static float* stbir__add_empty_ring_buffer_entry(stbir__info* stbir_info, int n)
+{
+ int ring_buffer_index;
+ float* ring_buffer;
+
+ stbir_info->ring_buffer_last_scanline = n;
+
+ if (stbir_info->ring_buffer_begin_index < 0)
+ {
+ ring_buffer_index = stbir_info->ring_buffer_begin_index = 0;
+ stbir_info->ring_buffer_first_scanline = n;
+ }
+ else
+ {
+ ring_buffer_index = (stbir_info->ring_buffer_begin_index + (stbir_info->ring_buffer_last_scanline - stbir_info->ring_buffer_first_scanline)) % stbir_info->ring_buffer_num_entries;
+ STBIR_ASSERT(ring_buffer_index != stbir_info->ring_buffer_begin_index);
+ }
+
+ ring_buffer = stbir__get_ring_buffer_entry(stbir_info->ring_buffer, ring_buffer_index, stbir_info->ring_buffer_length_bytes / sizeof(float));
+ memset(ring_buffer, 0, stbir_info->ring_buffer_length_bytes);
+
+ return ring_buffer;
+}
+
+
+static void stbir__resample_horizontal_upsample(stbir__info* stbir_info, float* output_buffer)
+{
+ int x, k;
+ int output_w = stbir_info->output_w;
+ int channels = stbir_info->channels;
+ float* decode_buffer = stbir__get_decode_buffer(stbir_info);
+ stbir__contributors* horizontal_contributors = stbir_info->horizontal_contributors;
+ float* horizontal_coefficients = stbir_info->horizontal_coefficients;
+ int coefficient_width = stbir_info->horizontal_coefficient_width;
+
+ for (x = 0; x < output_w; x++)
+ {
+ int n0 = horizontal_contributors[x].n0;
+ int n1 = horizontal_contributors[x].n1;
+
+ int out_pixel_index = x * channels;
+ int coefficient_group = coefficient_width * x;
+ int coefficient_counter = 0;
+
+ STBIR_ASSERT(n1 >= n0);
+ STBIR_ASSERT(n0 >= -stbir_info->horizontal_filter_pixel_margin);
+ STBIR_ASSERT(n1 >= -stbir_info->horizontal_filter_pixel_margin);
+ STBIR_ASSERT(n0 < stbir_info->input_w + stbir_info->horizontal_filter_pixel_margin);
+ STBIR_ASSERT(n1 < stbir_info->input_w + stbir_info->horizontal_filter_pixel_margin);
+
+ switch (channels) {
+ case 1:
+ for (k = n0; k <= n1; k++)
+ {
+ int in_pixel_index = k * 1;
+ float coefficient = horizontal_coefficients[coefficient_group + coefficient_counter++];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ }
+ break;
+ case 2:
+ for (k = n0; k <= n1; k++)
+ {
+ int in_pixel_index = k * 2;
+ float coefficient = horizontal_coefficients[coefficient_group + coefficient_counter++];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
+ }
+ break;
+ case 3:
+ for (k = n0; k <= n1; k++)
+ {
+ int in_pixel_index = k * 3;
+ float coefficient = horizontal_coefficients[coefficient_group + coefficient_counter++];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
+ output_buffer[out_pixel_index + 2] += decode_buffer[in_pixel_index + 2] * coefficient;
+ }
+ break;
+ case 4:
+ for (k = n0; k <= n1; k++)
+ {
+ int in_pixel_index = k * 4;
+ float coefficient = horizontal_coefficients[coefficient_group + coefficient_counter++];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
+ output_buffer[out_pixel_index + 2] += decode_buffer[in_pixel_index + 2] * coefficient;
+ output_buffer[out_pixel_index + 3] += decode_buffer[in_pixel_index + 3] * coefficient;
+ }
+ break;
+ default:
+ for (k = n0; k <= n1; k++)
+ {
+ int in_pixel_index = k * channels;
+ float coefficient = horizontal_coefficients[coefficient_group + coefficient_counter++];
+ int c;
+ STBIR_ASSERT(coefficient != 0);
+ for (c = 0; c < channels; c++)
+ output_buffer[out_pixel_index + c] += decode_buffer[in_pixel_index + c] * coefficient;
+ }
+ break;
+ }
+ }
+}
+
+static void stbir__resample_horizontal_downsample(stbir__info* stbir_info, float* output_buffer)
+{
+ int x, k;
+ int input_w = stbir_info->input_w;
+ int channels = stbir_info->channels;
+ float* decode_buffer = stbir__get_decode_buffer(stbir_info);
+ stbir__contributors* horizontal_contributors = stbir_info->horizontal_contributors;
+ float* horizontal_coefficients = stbir_info->horizontal_coefficients;
+ int coefficient_width = stbir_info->horizontal_coefficient_width;
+ int filter_pixel_margin = stbir_info->horizontal_filter_pixel_margin;
+ int max_x = input_w + filter_pixel_margin * 2;
+
+ STBIR_ASSERT(!stbir__use_width_upsampling(stbir_info));
+
+ switch (channels) {
+ case 1:
+ for (x = 0; x < max_x; x++)
+ {
+ int n0 = horizontal_contributors[x].n0;
+ int n1 = horizontal_contributors[x].n1;
+
+ int in_x = x - filter_pixel_margin;
+ int in_pixel_index = in_x * 1;
+ int max_n = n1;
+ int coefficient_group = coefficient_width * x;
+
+ for (k = n0; k <= max_n; k++)
+ {
+ int out_pixel_index = k * 1;
+ float coefficient = horizontal_coefficients[coefficient_group + k - n0];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ }
+ }
+ break;
+
+ case 2:
+ for (x = 0; x < max_x; x++)
+ {
+ int n0 = horizontal_contributors[x].n0;
+ int n1 = horizontal_contributors[x].n1;
+
+ int in_x = x - filter_pixel_margin;
+ int in_pixel_index = in_x * 2;
+ int max_n = n1;
+ int coefficient_group = coefficient_width * x;
+
+ for (k = n0; k <= max_n; k++)
+ {
+ int out_pixel_index = k * 2;
+ float coefficient = horizontal_coefficients[coefficient_group + k - n0];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
+ }
+ }
+ break;
+
+ case 3:
+ for (x = 0; x < max_x; x++)
+ {
+ int n0 = horizontal_contributors[x].n0;
+ int n1 = horizontal_contributors[x].n1;
+
+ int in_x = x - filter_pixel_margin;
+ int in_pixel_index = in_x * 3;
+ int max_n = n1;
+ int coefficient_group = coefficient_width * x;
+
+ for (k = n0; k <= max_n; k++)
+ {
+ int out_pixel_index = k * 3;
+ float coefficient = horizontal_coefficients[coefficient_group + k - n0];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
+ output_buffer[out_pixel_index + 2] += decode_buffer[in_pixel_index + 2] * coefficient;
+ }
+ }
+ break;
+
+ case 4:
+ for (x = 0; x < max_x; x++)
+ {
+ int n0 = horizontal_contributors[x].n0;
+ int n1 = horizontal_contributors[x].n1;
+
+ int in_x = x - filter_pixel_margin;
+ int in_pixel_index = in_x * 4;
+ int max_n = n1;
+ int coefficient_group = coefficient_width * x;
+
+ for (k = n0; k <= max_n; k++)
+ {
+ int out_pixel_index = k * 4;
+ float coefficient = horizontal_coefficients[coefficient_group + k - n0];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
+ output_buffer[out_pixel_index + 2] += decode_buffer[in_pixel_index + 2] * coefficient;
+ output_buffer[out_pixel_index + 3] += decode_buffer[in_pixel_index + 3] * coefficient;
+ }
+ }
+ break;
+
+ default:
+ for (x = 0; x < max_x; x++)
+ {
+ int n0 = horizontal_contributors[x].n0;
+ int n1 = horizontal_contributors[x].n1;
+
+ int in_x = x - filter_pixel_margin;
+ int in_pixel_index = in_x * channels;
+ int max_n = n1;
+ int coefficient_group = coefficient_width * x;
+
+ for (k = n0; k <= max_n; k++)
+ {
+ int c;
+ int out_pixel_index = k * channels;
+ float coefficient = horizontal_coefficients[coefficient_group + k - n0];
+ STBIR_ASSERT(coefficient != 0);
+ for (c = 0; c < channels; c++)
+ output_buffer[out_pixel_index + c] += decode_buffer[in_pixel_index + c] * coefficient;
+ }
+ }
+ break;
+ }
+}
+
+static void stbir__decode_and_resample_upsample(stbir__info* stbir_info, int n)
+{
+ // Decode the nth scanline from the source image into the decode buffer.
+ stbir__decode_scanline(stbir_info, n);
+
+ // Now resample it into the ring buffer.
+ if (stbir__use_width_upsampling(stbir_info))
+ stbir__resample_horizontal_upsample(stbir_info, stbir__add_empty_ring_buffer_entry(stbir_info, n));
+ else
+ stbir__resample_horizontal_downsample(stbir_info, stbir__add_empty_ring_buffer_entry(stbir_info, n));
+
+ // Now it's sitting in the ring buffer ready to be used as source for the vertical sampling.
+}
+
+static void stbir__decode_and_resample_downsample(stbir__info* stbir_info, int n)
+{
+ // Decode the nth scanline from the source image into the decode buffer.
+ stbir__decode_scanline(stbir_info, n);
+
+ memset(stbir_info->horizontal_buffer, 0, stbir_info->output_w * stbir_info->channels * sizeof(float));
+
+ // Now resample it into the horizontal buffer.
+ if (stbir__use_width_upsampling(stbir_info))
+ stbir__resample_horizontal_upsample(stbir_info, stbir_info->horizontal_buffer);
+ else
+ stbir__resample_horizontal_downsample(stbir_info, stbir_info->horizontal_buffer);
+
+ // Now it's sitting in the horizontal buffer ready to be distributed into the ring buffers.
+}
+
+// Get the specified scan line from the ring buffer.
+static float* stbir__get_ring_buffer_scanline(int get_scanline, float* ring_buffer, int begin_index, int first_scanline, int ring_buffer_num_entries, int ring_buffer_length)
+{
+ int ring_buffer_index = (begin_index + (get_scanline - first_scanline)) % ring_buffer_num_entries;
+ return stbir__get_ring_buffer_entry(ring_buffer, ring_buffer_index, ring_buffer_length);
+}
+
+
+static void stbir__encode_scanline(stbir__info* stbir_info, int num_pixels, void *output_buffer, float *encode_buffer, int channels, int alpha_channel, int decode)
+{
+ int x;
+ int n;
+ int num_nonalpha;
+ stbir_uint16 nonalpha[STBIR_MAX_CHANNELS];
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_PREMULTIPLIED))
+ {
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ float alpha = encode_buffer[pixel_index + alpha_channel];
+ float reciprocal_alpha = alpha ? 1.0f / alpha : 0;
+
+ // unrolling this produced a 1% slowdown upscaling a large RGBA linear-space image on my machine - stb
+ for (n = 0; n < channels; n++)
+ if (n != alpha_channel)
+ encode_buffer[pixel_index + n] *= reciprocal_alpha;
+
+ // We added in a small epsilon to prevent the color channel from being deleted with zero alpha.
+ // Because we only add it for integer types, it will automatically be discarded on integer
+ // conversion, so we don't need to subtract it back out (which would be problematic for
+ // numeric precision reasons).
+ }
+ }
+
+ // build a table of all channels that need colorspace correction, so
+ // we don't perform colorspace correction on channels that don't need it.
+ for (x = 0, num_nonalpha = 0; x < channels; ++x)
+ {
+ if (x != alpha_channel || (stbir_info->flags & STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ {
+ nonalpha[num_nonalpha++] = (stbir_uint16)x;
+ }
+ }
+
+ #define STBIR__ROUND_INT(f) ((int) ((f)+0.5))
+ #define STBIR__ROUND_UINT(f) ((stbir_uint32) ((f)+0.5))
+
+ #ifdef STBIR__SATURATE_INT
+ #define STBIR__ENCODE_LINEAR8(f) stbir__saturate8 (STBIR__ROUND_INT((f) * stbir__max_uint8_as_float ))
+ #define STBIR__ENCODE_LINEAR16(f) stbir__saturate16(STBIR__ROUND_INT((f) * stbir__max_uint16_as_float))
+ #else
+ #define STBIR__ENCODE_LINEAR8(f) (unsigned char ) STBIR__ROUND_INT(stbir__saturate(f) * stbir__max_uint8_as_float )
+ #define STBIR__ENCODE_LINEAR16(f) (unsigned short) STBIR__ROUND_INT(stbir__saturate(f) * stbir__max_uint16_as_float)
+ #endif
+
+ switch (decode)
+ {
+ case STBIR__DECODE(STBIR_TYPE_UINT8, STBIR_COLORSPACE_LINEAR):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < channels; n++)
+ {
+ int index = pixel_index + n;
+ ((unsigned char*)output_buffer)[index] = STBIR__ENCODE_LINEAR8(encode_buffer[index]);
+ }
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT8, STBIR_COLORSPACE_SRGB):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < num_nonalpha; n++)
+ {
+ int index = pixel_index + nonalpha[n];
+ ((unsigned char*)output_buffer)[index] = stbir__linear_to_srgb_uchar(encode_buffer[index]);
+ }
+
+ if (!(stbir_info->flags & STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ ((unsigned char *)output_buffer)[pixel_index + alpha_channel] = STBIR__ENCODE_LINEAR8(encode_buffer[pixel_index+alpha_channel]);
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT16, STBIR_COLORSPACE_LINEAR):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < channels; n++)
+ {
+ int index = pixel_index + n;
+ ((unsigned short*)output_buffer)[index] = STBIR__ENCODE_LINEAR16(encode_buffer[index]);
+ }
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT16, STBIR_COLORSPACE_SRGB):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < num_nonalpha; n++)
+ {
+ int index = pixel_index + nonalpha[n];
+ ((unsigned short*)output_buffer)[index] = (unsigned short)STBIR__ROUND_INT(stbir__linear_to_srgb(stbir__saturate(encode_buffer[index])) * stbir__max_uint16_as_float);
+ }
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ ((unsigned short*)output_buffer)[pixel_index + alpha_channel] = STBIR__ENCODE_LINEAR16(encode_buffer[pixel_index + alpha_channel]);
+ }
+
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT32, STBIR_COLORSPACE_LINEAR):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < channels; n++)
+ {
+ int index = pixel_index + n;
+ ((unsigned int*)output_buffer)[index] = (unsigned int)STBIR__ROUND_UINT(((double)stbir__saturate(encode_buffer[index])) * stbir__max_uint32_as_float);
+ }
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT32, STBIR_COLORSPACE_SRGB):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < num_nonalpha; n++)
+ {
+ int index = pixel_index + nonalpha[n];
+ ((unsigned int*)output_buffer)[index] = (unsigned int)STBIR__ROUND_UINT(((double)stbir__linear_to_srgb(stbir__saturate(encode_buffer[index]))) * stbir__max_uint32_as_float);
+ }
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ ((unsigned int*)output_buffer)[pixel_index + alpha_channel] = (unsigned int)STBIR__ROUND_INT(((double)stbir__saturate(encode_buffer[pixel_index + alpha_channel])) * stbir__max_uint32_as_float);
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_FLOAT, STBIR_COLORSPACE_LINEAR):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < channels; n++)
+ {
+ int index = pixel_index + n;
+ ((float*)output_buffer)[index] = encode_buffer[index];
+ }
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_FLOAT, STBIR_COLORSPACE_SRGB):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < num_nonalpha; n++)
+ {
+ int index = pixel_index + nonalpha[n];
+ ((float*)output_buffer)[index] = stbir__linear_to_srgb(encode_buffer[index]);
+ }
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ ((float*)output_buffer)[pixel_index + alpha_channel] = encode_buffer[pixel_index + alpha_channel];
+ }
+ break;
+
+ default:
+ STBIR_ASSERT(!"Unknown type/colorspace/channels combination.");
+ break;
+ }
+}
+
+static void stbir__resample_vertical_upsample(stbir__info* stbir_info, int n)
+{
+ int x, k;
+ int output_w = stbir_info->output_w;
+ stbir__contributors* vertical_contributors = stbir_info->vertical_contributors;
+ float* vertical_coefficients = stbir_info->vertical_coefficients;
+ int channels = stbir_info->channels;
+ int alpha_channel = stbir_info->alpha_channel;
+ int type = stbir_info->type;
+ int colorspace = stbir_info->colorspace;
+ int ring_buffer_entries = stbir_info->ring_buffer_num_entries;
+ void* output_data = stbir_info->output_data;
+ float* encode_buffer = stbir_info->encode_buffer;
+ int decode = STBIR__DECODE(type, colorspace);
+ int coefficient_width = stbir_info->vertical_coefficient_width;
+ int coefficient_counter;
+ int contributor = n;
+
+ float* ring_buffer = stbir_info->ring_buffer;
+ int ring_buffer_begin_index = stbir_info->ring_buffer_begin_index;
+ int ring_buffer_first_scanline = stbir_info->ring_buffer_first_scanline;
+ int ring_buffer_length = stbir_info->ring_buffer_length_bytes/sizeof(float);
+
+ int n0,n1, output_row_start;
+ int coefficient_group = coefficient_width * contributor;
+
+ n0 = vertical_contributors[contributor].n0;
+ n1 = vertical_contributors[contributor].n1;
+
+ output_row_start = n * stbir_info->output_stride_bytes;
+
+ STBIR_ASSERT(stbir__use_height_upsampling(stbir_info));
+
+ memset(encode_buffer, 0, output_w * sizeof(float) * channels);
+
+ // I tried reblocking this for better cache usage of encode_buffer
+ // (using x_outer, k, x_inner), but it lost speed. -- stb
+
+ coefficient_counter = 0;
+ switch (channels) {
+ case 1:
+ for (k = n0; k <= n1; k++)
+ {
+ int coefficient_index = coefficient_counter++;
+ float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length);
+ float coefficient = vertical_coefficients[coefficient_group + coefficient_index];
+ for (x = 0; x < output_w; ++x)
+ {
+ int in_pixel_index = x * 1;
+ encode_buffer[in_pixel_index + 0] += ring_buffer_entry[in_pixel_index + 0] * coefficient;
+ }
+ }
+ break;
+ case 2:
+ for (k = n0; k <= n1; k++)
+ {
+ int coefficient_index = coefficient_counter++;
+ float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length);
+ float coefficient = vertical_coefficients[coefficient_group + coefficient_index];
+ for (x = 0; x < output_w; ++x)
+ {
+ int in_pixel_index = x * 2;
+ encode_buffer[in_pixel_index + 0] += ring_buffer_entry[in_pixel_index + 0] * coefficient;
+ encode_buffer[in_pixel_index + 1] += ring_buffer_entry[in_pixel_index + 1] * coefficient;
+ }
+ }
+ break;
+ case 3:
+ for (k = n0; k <= n1; k++)
+ {
+ int coefficient_index = coefficient_counter++;
+ float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length);
+ float coefficient = vertical_coefficients[coefficient_group + coefficient_index];
+ for (x = 0; x < output_w; ++x)
+ {
+ int in_pixel_index = x * 3;
+ encode_buffer[in_pixel_index + 0] += ring_buffer_entry[in_pixel_index + 0] * coefficient;
+ encode_buffer[in_pixel_index + 1] += ring_buffer_entry[in_pixel_index + 1] * coefficient;
+ encode_buffer[in_pixel_index + 2] += ring_buffer_entry[in_pixel_index + 2] * coefficient;
+ }
+ }
+ break;
+ case 4:
+ for (k = n0; k <= n1; k++)
+ {
+ int coefficient_index = coefficient_counter++;
+ float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length);
+ float coefficient = vertical_coefficients[coefficient_group + coefficient_index];
+ for (x = 0; x < output_w; ++x)
+ {
+ int in_pixel_index = x * 4;
+ encode_buffer[in_pixel_index + 0] += ring_buffer_entry[in_pixel_index + 0] * coefficient;
+ encode_buffer[in_pixel_index + 1] += ring_buffer_entry[in_pixel_index + 1] * coefficient;
+ encode_buffer[in_pixel_index + 2] += ring_buffer_entry[in_pixel_index + 2] * coefficient;
+ encode_buffer[in_pixel_index + 3] += ring_buffer_entry[in_pixel_index + 3] * coefficient;
+ }
+ }
+ break;
+ default:
+ for (k = n0; k <= n1; k++)
+ {
+ int coefficient_index = coefficient_counter++;
+ float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length);
+ float coefficient = vertical_coefficients[coefficient_group + coefficient_index];
+ for (x = 0; x < output_w; ++x)
+ {
+ int in_pixel_index = x * channels;
+ int c;
+ for (c = 0; c < channels; c++)
+ encode_buffer[in_pixel_index + c] += ring_buffer_entry[in_pixel_index + c] * coefficient;
+ }
+ }
+ break;
+ }
+ stbir__encode_scanline(stbir_info, output_w, (char *) output_data + output_row_start, encode_buffer, channels, alpha_channel, decode);
+}
+
+static void stbir__resample_vertical_downsample(stbir__info* stbir_info, int n)
+{
+ int x, k;
+ int output_w = stbir_info->output_w;
+ stbir__contributors* vertical_contributors = stbir_info->vertical_contributors;
+ float* vertical_coefficients = stbir_info->vertical_coefficients;
+ int channels = stbir_info->channels;
+ int ring_buffer_entries = stbir_info->ring_buffer_num_entries;
+ float* horizontal_buffer = stbir_info->horizontal_buffer;
+ int coefficient_width = stbir_info->vertical_coefficient_width;
+ int contributor = n + stbir_info->vertical_filter_pixel_margin;
+
+ float* ring_buffer = stbir_info->ring_buffer;
+ int ring_buffer_begin_index = stbir_info->ring_buffer_begin_index;
+ int ring_buffer_first_scanline = stbir_info->ring_buffer_first_scanline;
+ int ring_buffer_length = stbir_info->ring_buffer_length_bytes/sizeof(float);
+ int n0,n1;
+
+ n0 = vertical_contributors[contributor].n0;
+ n1 = vertical_contributors[contributor].n1;
+
+ STBIR_ASSERT(!stbir__use_height_upsampling(stbir_info));
+
+ for (k = n0; k <= n1; k++)
+ {
+ int coefficient_index = k - n0;
+ int coefficient_group = coefficient_width * contributor;
+ float coefficient = vertical_coefficients[coefficient_group + coefficient_index];
+
+ float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length);
+
+ switch (channels) {
+ case 1:
+ for (x = 0; x < output_w; x++)
+ {
+ int in_pixel_index = x * 1;
+ ring_buffer_entry[in_pixel_index + 0] += horizontal_buffer[in_pixel_index + 0] * coefficient;
+ }
+ break;
+ case 2:
+ for (x = 0; x < output_w; x++)
+ {
+ int in_pixel_index = x * 2;
+ ring_buffer_entry[in_pixel_index + 0] += horizontal_buffer[in_pixel_index + 0] * coefficient;
+ ring_buffer_entry[in_pixel_index + 1] += horizontal_buffer[in_pixel_index + 1] * coefficient;
+ }
+ break;
+ case 3:
+ for (x = 0; x < output_w; x++)
+ {
+ int in_pixel_index = x * 3;
+ ring_buffer_entry[in_pixel_index + 0] += horizontal_buffer[in_pixel_index + 0] * coefficient;
+ ring_buffer_entry[in_pixel_index + 1] += horizontal_buffer[in_pixel_index + 1] * coefficient;
+ ring_buffer_entry[in_pixel_index + 2] += horizontal_buffer[in_pixel_index + 2] * coefficient;
+ }
+ break;
+ case 4:
+ for (x = 0; x < output_w; x++)
+ {
+ int in_pixel_index = x * 4;
+ ring_buffer_entry[in_pixel_index + 0] += horizontal_buffer[in_pixel_index + 0] * coefficient;
+ ring_buffer_entry[in_pixel_index + 1] += horizontal_buffer[in_pixel_index + 1] * coefficient;
+ ring_buffer_entry[in_pixel_index + 2] += horizontal_buffer[in_pixel_index + 2] * coefficient;
+ ring_buffer_entry[in_pixel_index + 3] += horizontal_buffer[in_pixel_index + 3] * coefficient;
+ }
+ break;
+ default:
+ for (x = 0; x < output_w; x++)
+ {
+ int in_pixel_index = x * channels;
+
+ int c;
+ for (c = 0; c < channels; c++)
+ ring_buffer_entry[in_pixel_index + c] += horizontal_buffer[in_pixel_index + c] * coefficient;
+ }
+ break;
+ }
+ }
+}
+
+static void stbir__buffer_loop_upsample(stbir__info* stbir_info)
+{
+ int y;
+ float scale_ratio = stbir_info->vertical_scale;
+ float out_scanlines_radius = stbir__filter_info_table[stbir_info->vertical_filter].support(1/scale_ratio) * scale_ratio;
+
+ STBIR_ASSERT(stbir__use_height_upsampling(stbir_info));
+
+ for (y = 0; y < stbir_info->output_h; y++)
+ {
+ float in_center_of_out = 0; // Center of the current out scanline in the in scanline space
+ int in_first_scanline = 0, in_last_scanline = 0;
+
+ stbir__calculate_sample_range_upsample(y, out_scanlines_radius, scale_ratio, stbir_info->vertical_shift, &in_first_scanline, &in_last_scanline, &in_center_of_out);
+
+ STBIR_ASSERT(in_last_scanline - in_first_scanline + 1 <= stbir_info->ring_buffer_num_entries);
+
+ if (stbir_info->ring_buffer_begin_index >= 0)
+ {
+ // Get rid of whatever we don't need anymore.
+ while (in_first_scanline > stbir_info->ring_buffer_first_scanline)
+ {
+ if (stbir_info->ring_buffer_first_scanline == stbir_info->ring_buffer_last_scanline)
+ {
+ // We just popped the last scanline off the ring buffer.
+ // Reset it to the empty state.
+ stbir_info->ring_buffer_begin_index = -1;
+ stbir_info->ring_buffer_first_scanline = 0;
+ stbir_info->ring_buffer_last_scanline = 0;
+ break;
+ }
+ else
+ {
+ stbir_info->ring_buffer_first_scanline++;
+ stbir_info->ring_buffer_begin_index = (stbir_info->ring_buffer_begin_index + 1) % stbir_info->ring_buffer_num_entries;
+ }
+ }
+ }
+
+ // Load in new ones.
+ if (stbir_info->ring_buffer_begin_index < 0)
+ stbir__decode_and_resample_upsample(stbir_info, in_first_scanline);
+
+ while (in_last_scanline > stbir_info->ring_buffer_last_scanline)
+ stbir__decode_and_resample_upsample(stbir_info, stbir_info->ring_buffer_last_scanline + 1);
+
+ // Now all buffers should be ready to write a row of vertical sampling.
+ stbir__resample_vertical_upsample(stbir_info, y);
+
+ STBIR_PROGRESS_REPORT((float)y / stbir_info->output_h);
+ }
+}
+
+static void stbir__empty_ring_buffer(stbir__info* stbir_info, int first_necessary_scanline)
+{
+ int output_stride_bytes = stbir_info->output_stride_bytes;
+ int channels = stbir_info->channels;
+ int alpha_channel = stbir_info->alpha_channel;
+ int type = stbir_info->type;
+ int colorspace = stbir_info->colorspace;
+ int output_w = stbir_info->output_w;
+ void* output_data = stbir_info->output_data;
+ int decode = STBIR__DECODE(type, colorspace);
+
+ float* ring_buffer = stbir_info->ring_buffer;
+ int ring_buffer_length = stbir_info->ring_buffer_length_bytes/sizeof(float);
+
+ if (stbir_info->ring_buffer_begin_index >= 0)
+ {
+ // Get rid of whatever we don't need anymore.
+ while (first_necessary_scanline > stbir_info->ring_buffer_first_scanline)
+ {
+ if (stbir_info->ring_buffer_first_scanline >= 0 && stbir_info->ring_buffer_first_scanline < stbir_info->output_h)
+ {
+ int output_row_start = stbir_info->ring_buffer_first_scanline * output_stride_bytes;
+ float* ring_buffer_entry = stbir__get_ring_buffer_entry(ring_buffer, stbir_info->ring_buffer_begin_index, ring_buffer_length);
+ stbir__encode_scanline(stbir_info, output_w, (char *) output_data + output_row_start, ring_buffer_entry, channels, alpha_channel, decode);
+ STBIR_PROGRESS_REPORT((float)stbir_info->ring_buffer_first_scanline / stbir_info->output_h);
+ }
+
+ if (stbir_info->ring_buffer_first_scanline == stbir_info->ring_buffer_last_scanline)
+ {
+ // We just popped the last scanline off the ring buffer.
+ // Reset it to the empty state.
+ stbir_info->ring_buffer_begin_index = -1;
+ stbir_info->ring_buffer_first_scanline = 0;
+ stbir_info->ring_buffer_last_scanline = 0;
+ break;
+ }
+ else
+ {
+ stbir_info->ring_buffer_first_scanline++;
+ stbir_info->ring_buffer_begin_index = (stbir_info->ring_buffer_begin_index + 1) % stbir_info->ring_buffer_num_entries;
+ }
+ }
+ }
+}
+
+static void stbir__buffer_loop_downsample(stbir__info* stbir_info)
+{
+ int y;
+ float scale_ratio = stbir_info->vertical_scale;
+ int output_h = stbir_info->output_h;
+ float in_pixels_radius = stbir__filter_info_table[stbir_info->vertical_filter].support(scale_ratio) / scale_ratio;
+ int pixel_margin = stbir_info->vertical_filter_pixel_margin;
+ int max_y = stbir_info->input_h + pixel_margin;
+
+ STBIR_ASSERT(!stbir__use_height_upsampling(stbir_info));
+
+ for (y = -pixel_margin; y < max_y; y++)
+ {
+ float out_center_of_in; // Center of the current out scanline in the in scanline space
+ int out_first_scanline, out_last_scanline;
+
+ stbir__calculate_sample_range_downsample(y, in_pixels_radius, scale_ratio, stbir_info->vertical_shift, &out_first_scanline, &out_last_scanline, &out_center_of_in);
+
+ STBIR_ASSERT(out_last_scanline - out_first_scanline + 1 <= stbir_info->ring_buffer_num_entries);
+
+ if (out_last_scanline < 0 || out_first_scanline >= output_h)
+ continue;
+
+ stbir__empty_ring_buffer(stbir_info, out_first_scanline);
+
+ stbir__decode_and_resample_downsample(stbir_info, y);
+
+ // Load in new ones.
+ if (stbir_info->ring_buffer_begin_index < 0)
+ stbir__add_empty_ring_buffer_entry(stbir_info, out_first_scanline);
+
+ while (out_last_scanline > stbir_info->ring_buffer_last_scanline)
+ stbir__add_empty_ring_buffer_entry(stbir_info, stbir_info->ring_buffer_last_scanline + 1);
+
+ // Now the horizontal buffer is ready to write to all ring buffer rows.
+ stbir__resample_vertical_downsample(stbir_info, y);
+ }
+
+ stbir__empty_ring_buffer(stbir_info, stbir_info->output_h);
+}
+
+static void stbir__setup(stbir__info *info, int input_w, int input_h, int output_w, int output_h, int channels)
+{
+ info->input_w = input_w;
+ info->input_h = input_h;
+ info->output_w = output_w;
+ info->output_h = output_h;
+ info->channels = channels;
+}
+
+static void stbir__calculate_transform(stbir__info *info, float s0, float t0, float s1, float t1, float *transform)
+{
+ info->s0 = s0;
+ info->t0 = t0;
+ info->s1 = s1;
+ info->t1 = t1;
+
+ if (transform)
+ {
+ info->horizontal_scale = transform[0];
+ info->vertical_scale = transform[1];
+ info->horizontal_shift = transform[2];
+ info->vertical_shift = transform[3];
+ }
+ else
+ {
+ info->horizontal_scale = ((float)info->output_w / info->input_w) / (s1 - s0);
+ info->vertical_scale = ((float)info->output_h / info->input_h) / (t1 - t0);
+
+ info->horizontal_shift = s0 * info->output_w / (s1 - s0);
+ info->vertical_shift = t0 * info->output_h / (t1 - t0);
+ }
+}
+
+static void stbir__choose_filter(stbir__info *info, stbir_filter h_filter, stbir_filter v_filter)
+{
+ if (h_filter == 0)
+ h_filter = stbir__use_upsampling(info->horizontal_scale) ? STBIR_DEFAULT_FILTER_UPSAMPLE : STBIR_DEFAULT_FILTER_DOWNSAMPLE;
+ if (v_filter == 0)
+ v_filter = stbir__use_upsampling(info->vertical_scale) ? STBIR_DEFAULT_FILTER_UPSAMPLE : STBIR_DEFAULT_FILTER_DOWNSAMPLE;
+ info->horizontal_filter = h_filter;
+ info->vertical_filter = v_filter;
+}
+
+static stbir_uint32 stbir__calculate_memory(stbir__info *info)
+{
+ int pixel_margin = stbir__get_filter_pixel_margin(info->horizontal_filter, info->horizontal_scale);
+ int filter_height = stbir__get_filter_pixel_width(info->vertical_filter, info->vertical_scale);
+
+ info->horizontal_num_contributors = stbir__get_contributors(info->horizontal_scale, info->horizontal_filter, info->input_w, info->output_w);
+ info->vertical_num_contributors = stbir__get_contributors(info->vertical_scale , info->vertical_filter , info->input_h, info->output_h);
+
+ // One extra entry because floating point precision problems sometimes cause an extra to be necessary.
+ info->ring_buffer_num_entries = filter_height + 1;
+
+ info->horizontal_contributors_size = info->horizontal_num_contributors * sizeof(stbir__contributors);
+ info->horizontal_coefficients_size = stbir__get_total_horizontal_coefficients(info) * sizeof(float);
+ info->vertical_contributors_size = info->vertical_num_contributors * sizeof(stbir__contributors);
+ info->vertical_coefficients_size = stbir__get_total_vertical_coefficients(info) * sizeof(float);
+ info->decode_buffer_size = (info->input_w + pixel_margin * 2) * info->channels * sizeof(float);
+ info->horizontal_buffer_size = info->output_w * info->channels * sizeof(float);
+ info->ring_buffer_size = info->output_w * info->channels * info->ring_buffer_num_entries * sizeof(float);
+ info->encode_buffer_size = info->output_w * info->channels * sizeof(float);
+
+ STBIR_ASSERT(info->horizontal_filter != 0);
+ STBIR_ASSERT(info->horizontal_filter < STBIR__ARRAY_SIZE(stbir__filter_info_table)); // this now happens too late
+ STBIR_ASSERT(info->vertical_filter != 0);
+ STBIR_ASSERT(info->vertical_filter < STBIR__ARRAY_SIZE(stbir__filter_info_table)); // this now happens too late
+
+ if (stbir__use_height_upsampling(info))
+ // The horizontal buffer is for when we're downsampling the height and we
+ // can't output the result of sampling the decode buffer directly into the
+ // ring buffers.
+ info->horizontal_buffer_size = 0;
+ else
+ // The encode buffer is to retain precision in the height upsampling method
+ // and isn't used when height downsampling.
+ info->encode_buffer_size = 0;
+
+ return info->horizontal_contributors_size + info->horizontal_coefficients_size
+ + info->vertical_contributors_size + info->vertical_coefficients_size
+ + info->decode_buffer_size + info->horizontal_buffer_size
+ + info->ring_buffer_size + info->encode_buffer_size;
+}
+
+static int stbir__resize_allocated(stbir__info *info,
+ const void* input_data, int input_stride_in_bytes,
+ void* output_data, int output_stride_in_bytes,
+ int alpha_channel, stbir_uint32 flags, stbir_datatype type,
+ stbir_edge edge_horizontal, stbir_edge edge_vertical, stbir_colorspace colorspace,
+ void* tempmem, size_t tempmem_size_in_bytes)
+{
+ size_t memory_required = stbir__calculate_memory(info);
+
+ int width_stride_input = input_stride_in_bytes ? input_stride_in_bytes : info->channels * info->input_w * stbir__type_size[type];
+ int width_stride_output = output_stride_in_bytes ? output_stride_in_bytes : info->channels * info->output_w * stbir__type_size[type];
+
+#ifdef STBIR_DEBUG_OVERWRITE_TEST
+#define OVERWRITE_ARRAY_SIZE 8
+ unsigned char overwrite_output_before_pre[OVERWRITE_ARRAY_SIZE];
+ unsigned char overwrite_tempmem_before_pre[OVERWRITE_ARRAY_SIZE];
+ unsigned char overwrite_output_after_pre[OVERWRITE_ARRAY_SIZE];
+ unsigned char overwrite_tempmem_after_pre[OVERWRITE_ARRAY_SIZE];
+
+ size_t begin_forbidden = width_stride_output * (info->output_h - 1) + info->output_w * info->channels * stbir__type_size[type];
+ memcpy(overwrite_output_before_pre, &((unsigned char*)output_data)[-OVERWRITE_ARRAY_SIZE], OVERWRITE_ARRAY_SIZE);
+ memcpy(overwrite_output_after_pre, &((unsigned char*)output_data)[begin_forbidden], OVERWRITE_ARRAY_SIZE);
+ memcpy(overwrite_tempmem_before_pre, &((unsigned char*)tempmem)[-OVERWRITE_ARRAY_SIZE], OVERWRITE_ARRAY_SIZE);
+ memcpy(overwrite_tempmem_after_pre, &((unsigned char*)tempmem)[tempmem_size_in_bytes], OVERWRITE_ARRAY_SIZE);
+#endif
+
+ STBIR_ASSERT(info->channels >= 0);
+ STBIR_ASSERT(info->channels <= STBIR_MAX_CHANNELS);
+
+ if (info->channels < 0 || info->channels > STBIR_MAX_CHANNELS)
+ return 0;
+
+ STBIR_ASSERT(info->horizontal_filter < STBIR__ARRAY_SIZE(stbir__filter_info_table));
+ STBIR_ASSERT(info->vertical_filter < STBIR__ARRAY_SIZE(stbir__filter_info_table));
+
+ if (info->horizontal_filter >= STBIR__ARRAY_SIZE(stbir__filter_info_table))
+ return 0;
+ if (info->vertical_filter >= STBIR__ARRAY_SIZE(stbir__filter_info_table))
+ return 0;
+
+ if (alpha_channel < 0)
+ flags |= STBIR_FLAG_ALPHA_USES_COLORSPACE | STBIR_FLAG_ALPHA_PREMULTIPLIED;
+
+ if (!(flags&STBIR_FLAG_ALPHA_USES_COLORSPACE) || !(flags&STBIR_FLAG_ALPHA_PREMULTIPLIED))
+ STBIR_ASSERT(alpha_channel >= 0 && alpha_channel < info->channels);
+
+ if (alpha_channel >= info->channels)
+ return 0;
+
+ STBIR_ASSERT(tempmem);
+
+ if (!tempmem)
+ return 0;
+
+ STBIR_ASSERT(tempmem_size_in_bytes >= memory_required);
+
+ if (tempmem_size_in_bytes < memory_required)
+ return 0;
+
+ memset(tempmem, 0, tempmem_size_in_bytes);
+
+ info->input_data = input_data;
+ info->input_stride_bytes = width_stride_input;
+
+ info->output_data = output_data;
+ info->output_stride_bytes = width_stride_output;
+
+ info->alpha_channel = alpha_channel;
+ info->flags = flags;
+ info->type = type;
+ info->edge_horizontal = edge_horizontal;
+ info->edge_vertical = edge_vertical;
+ info->colorspace = colorspace;
+
+ info->horizontal_coefficient_width = stbir__get_coefficient_width (info->horizontal_filter, info->horizontal_scale);
+ info->vertical_coefficient_width = stbir__get_coefficient_width (info->vertical_filter , info->vertical_scale );
+ info->horizontal_filter_pixel_width = stbir__get_filter_pixel_width (info->horizontal_filter, info->horizontal_scale);
+ info->vertical_filter_pixel_width = stbir__get_filter_pixel_width (info->vertical_filter , info->vertical_scale );
+ info->horizontal_filter_pixel_margin = stbir__get_filter_pixel_margin(info->horizontal_filter, info->horizontal_scale);
+ info->vertical_filter_pixel_margin = stbir__get_filter_pixel_margin(info->vertical_filter , info->vertical_scale );
+
+ info->ring_buffer_length_bytes = info->output_w * info->channels * sizeof(float);
+ info->decode_buffer_pixels = info->input_w + info->horizontal_filter_pixel_margin * 2;
+
+#define STBIR__NEXT_MEMPTR(current, newtype) (newtype*)(((unsigned char*)current) + current##_size)
+
+ info->horizontal_contributors = (stbir__contributors *) tempmem;
+ info->horizontal_coefficients = STBIR__NEXT_MEMPTR(info->horizontal_contributors, float);
+ info->vertical_contributors = STBIR__NEXT_MEMPTR(info->horizontal_coefficients, stbir__contributors);
+ info->vertical_coefficients = STBIR__NEXT_MEMPTR(info->vertical_contributors, float);
+ info->decode_buffer = STBIR__NEXT_MEMPTR(info->vertical_coefficients, float);
+
+ if (stbir__use_height_upsampling(info))
+ {
+ info->horizontal_buffer = NULL;
+ info->ring_buffer = STBIR__NEXT_MEMPTR(info->decode_buffer, float);
+ info->encode_buffer = STBIR__NEXT_MEMPTR(info->ring_buffer, float);
+
+ STBIR_ASSERT((size_t)STBIR__NEXT_MEMPTR(info->encode_buffer, unsigned char) == (size_t)tempmem + tempmem_size_in_bytes);
+ }
+ else
+ {
+ info->horizontal_buffer = STBIR__NEXT_MEMPTR(info->decode_buffer, float);
+ info->ring_buffer = STBIR__NEXT_MEMPTR(info->horizontal_buffer, float);
+ info->encode_buffer = NULL;
+
+ STBIR_ASSERT((size_t)STBIR__NEXT_MEMPTR(info->ring_buffer, unsigned char) == (size_t)tempmem + tempmem_size_in_bytes);
+ }
+
+#undef STBIR__NEXT_MEMPTR
+
+ // This signals that the ring buffer is empty
+ info->ring_buffer_begin_index = -1;
+
+ stbir__calculate_filters(info->horizontal_contributors, info->horizontal_coefficients, info->horizontal_filter, info->horizontal_scale, info->horizontal_shift, info->input_w, info->output_w);
+ stbir__calculate_filters(info->vertical_contributors, info->vertical_coefficients, info->vertical_filter, info->vertical_scale, info->vertical_shift, info->input_h, info->output_h);
+
+ STBIR_PROGRESS_REPORT(0);
+
+ if (stbir__use_height_upsampling(info))
+ stbir__buffer_loop_upsample(info);
+ else
+ stbir__buffer_loop_downsample(info);
+
+ STBIR_PROGRESS_REPORT(1);
+
+#ifdef STBIR_DEBUG_OVERWRITE_TEST
+ STBIR_ASSERT(memcmp(overwrite_output_before_pre, &((unsigned char*)output_data)[-OVERWRITE_ARRAY_SIZE], OVERWRITE_ARRAY_SIZE) == 0);
+ STBIR_ASSERT(memcmp(overwrite_output_after_pre, &((unsigned char*)output_data)[begin_forbidden], OVERWRITE_ARRAY_SIZE) == 0);
+ STBIR_ASSERT(memcmp(overwrite_tempmem_before_pre, &((unsigned char*)tempmem)[-OVERWRITE_ARRAY_SIZE], OVERWRITE_ARRAY_SIZE) == 0);
+ STBIR_ASSERT(memcmp(overwrite_tempmem_after_pre, &((unsigned char*)tempmem)[tempmem_size_in_bytes], OVERWRITE_ARRAY_SIZE) == 0);
+#endif
+
+ return 1;
+}
+
+
+static int stbir__resize_arbitrary(
+ void *alloc_context,
+ const void* input_data, int input_w, int input_h, int input_stride_in_bytes,
+ void* output_data, int output_w, int output_h, int output_stride_in_bytes,
+ float s0, float t0, float s1, float t1, float *transform,
+ int channels, int alpha_channel, stbir_uint32 flags, stbir_datatype type,
+ stbir_filter h_filter, stbir_filter v_filter,
+ stbir_edge edge_horizontal, stbir_edge edge_vertical, stbir_colorspace colorspace)
+{
+ stbir__info info;
+ int result;
+ size_t memory_required;
+ void* extra_memory;
+
+ stbir__setup(&info, input_w, input_h, output_w, output_h, channels);
+ stbir__calculate_transform(&info, s0,t0,s1,t1,transform);
+ stbir__choose_filter(&info, h_filter, v_filter);
+ memory_required = stbir__calculate_memory(&info);
+ extra_memory = STBIR_MALLOC(memory_required, alloc_context);
+
+ if (!extra_memory)
+ return 0;
+
+ result = stbir__resize_allocated(&info, input_data, input_stride_in_bytes,
+ output_data, output_stride_in_bytes,
+ alpha_channel, flags, type,
+ edge_horizontal, edge_vertical,
+ colorspace, extra_memory, memory_required);
+
+ STBIR_FREE(extra_memory, alloc_context);
+
+ return result;
+}
+
+STBIRDEF int stbir_resize_uint8( const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels)
+{
+ return stbir__resize_arbitrary(NULL, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,-1,0, STBIR_TYPE_UINT8, STBIR_FILTER_DEFAULT, STBIR_FILTER_DEFAULT,
+ STBIR_EDGE_CLAMP, STBIR_EDGE_CLAMP, STBIR_COLORSPACE_LINEAR);
+}
+
+STBIRDEF int stbir_resize_float( const float *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ float *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels)
+{
+ return stbir__resize_arbitrary(NULL, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,-1,0, STBIR_TYPE_FLOAT, STBIR_FILTER_DEFAULT, STBIR_FILTER_DEFAULT,
+ STBIR_EDGE_CLAMP, STBIR_EDGE_CLAMP, STBIR_COLORSPACE_LINEAR);
+}
+
+STBIRDEF int stbir_resize_uint8_srgb(const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags)
+{
+ return stbir__resize_arbitrary(NULL, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,alpha_channel,flags, STBIR_TYPE_UINT8, STBIR_FILTER_DEFAULT, STBIR_FILTER_DEFAULT,
+ STBIR_EDGE_CLAMP, STBIR_EDGE_CLAMP, STBIR_COLORSPACE_SRGB);
+}
+
+STBIRDEF int stbir_resize_uint8_srgb_edgemode(const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode)
+{
+ return stbir__resize_arbitrary(NULL, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,alpha_channel,flags, STBIR_TYPE_UINT8, STBIR_FILTER_DEFAULT, STBIR_FILTER_DEFAULT,
+ edge_wrap_mode, edge_wrap_mode, STBIR_COLORSPACE_SRGB);
+}
+
+STBIRDEF int stbir_resize_uint8_generic( const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space,
+ void *alloc_context)
+{
+ return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,alpha_channel,flags, STBIR_TYPE_UINT8, filter, filter,
+ edge_wrap_mode, edge_wrap_mode, space);
+}
+
+STBIRDEF int stbir_resize_uint16_generic(const stbir_uint16 *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ stbir_uint16 *output_pixels , int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space,
+ void *alloc_context)
+{
+ return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,alpha_channel,flags, STBIR_TYPE_UINT16, filter, filter,
+ edge_wrap_mode, edge_wrap_mode, space);
+}
+
+
+STBIRDEF int stbir_resize_float_generic( const float *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ float *output_pixels , int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space,
+ void *alloc_context)
+{
+ return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,alpha_channel,flags, STBIR_TYPE_FLOAT, filter, filter,
+ edge_wrap_mode, edge_wrap_mode, space);
+}
+
+
+STBIRDEF int stbir_resize( const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ stbir_datatype datatype,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical,
+ stbir_filter filter_horizontal, stbir_filter filter_vertical,
+ stbir_colorspace space, void *alloc_context)
+{
+ return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,alpha_channel,flags, datatype, filter_horizontal, filter_vertical,
+ edge_mode_horizontal, edge_mode_vertical, space);
+}
+
+
+STBIRDEF int stbir_resize_subpixel(const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ stbir_datatype datatype,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical,
+ stbir_filter filter_horizontal, stbir_filter filter_vertical,
+ stbir_colorspace space, void *alloc_context,
+ float x_scale, float y_scale,
+ float x_offset, float y_offset)
+{
+ float transform[4];
+ transform[0] = x_scale;
+ transform[1] = y_scale;
+ transform[2] = x_offset;
+ transform[3] = y_offset;
+ return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,transform,num_channels,alpha_channel,flags, datatype, filter_horizontal, filter_vertical,
+ edge_mode_horizontal, edge_mode_vertical, space);
+}
+
+STBIRDEF int stbir_resize_region( const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ stbir_datatype datatype,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical,
+ stbir_filter filter_horizontal, stbir_filter filter_vertical,
+ stbir_colorspace space, void *alloc_context,
+ float s0, float t0, float s1, float t1)
+{
+ return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ s0,t0,s1,t1,NULL,num_channels,alpha_channel,flags, datatype, filter_horizontal, filter_vertical,
+ edge_mode_horizontal, edge_mode_vertical, space);
+}
+
+#endif // STB_IMAGE_RESIZE_IMPLEMENTATION
+
+/*
+------------------------------------------------------------------------------
+This software is available under 2 licenses -- choose whichever you prefer.
+------------------------------------------------------------------------------
+ALTERNATIVE A - MIT License
+Copyright (c) 2017 Sean Barrett
+Permission is hereby granted, free of charge, to any person obtaining a copy of
+this software and associated documentation files (the "Software"), to deal in
+the Software without restriction, including without limitation the rights to
+use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
+of the Software, and to permit persons to whom the Software is furnished to do
+so, subject to the following conditions:
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+------------------------------------------------------------------------------
+ALTERNATIVE B - Public Domain (www.unlicense.org)
+This is free and unencumbered software released into the public domain.
+Anyone is free to copy, modify, publish, use, compile, sell, or distribute this
+software, either in source code form or as a compiled binary, for any purpose,
+commercial or non-commercial, and by any means.
+In jurisdictions that recognize copyright laws, the author or authors of this
+software dedicate any and all copyright interest in the software to the public
+domain. We make this dedication for the benefit of the public at large and to
+the detriment of our heirs and successors. We intend this dedication to be an
+overt act of relinquishment in perpetuity of all present and future rights to
+this software under copyright law.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+------------------------------------------------------------------------------
+*/
--- /dev/null
+input: "data"
+input_shape {
+ dim: 1
+ dim: 3
+ dim: 224
+ dim: 224
+}
--- /dev/null
+#! /usr/bin/env python3
+
+# Copyright 2017 Intel Corporation.
+# The source code, information and material ("Material") contained herein is
+# owned by Intel Corporation or its suppliers or licensors, and title to such
+# Material remains with Intel Corporation or its suppliers or licensors.
+# The Material contains proprietary information of Intel or its suppliers and
+# licensors. The Material is protected by worldwide copyright laws and treaty
+# provisions.
+# No part of the Material may be used, copied, reproduced, modified, published,
+# uploaded, posted, transmitted, distributed or disclosed in any way without
+# Intel's prior express written permission. No license under any patent,
+# copyright or other intellectual property rights in the Material is granted to
+# or conferred upon you, either expressly, by implication, inducement, estoppel
+# or otherwise.
+# Any license under such intellectual property rights must be express and
+# approved by Intel in writing.
+
+from mvnc import mvncapi as mvnc
+import sys
+import numpy
+import cv2
+import time
+import csv
+import os
+import sys
+
+dim=(224,224)
+EXAMPLES_BASE_DIR='../../'
+
+# ***************************************************************
+# get labels
+# ***************************************************************
+labels_file=EXAMPLES_BASE_DIR+'data/ilsvrc12/synset_words.txt'
+labels=numpy.loadtxt(labels_file,str,delimiter='\t')
+
+# ***************************************************************
+# configure the NCS
+# ***************************************************************
+mvnc.SetGlobalOption(mvnc.GlobalOption.LOG_LEVEL, 2)
+
+# ***************************************************************
+# Get a list of ALL the sticks that are plugged in
+# ***************************************************************
+devices = mvnc.EnumerateDevices()
+if len(devices) == 0:
+ print('No devices found')
+ quit()
+
+# ***************************************************************
+# Pick the first stick to run the network
+# ***************************************************************
+device = mvnc.Device(devices[0])
+
+# ***************************************************************
+# Open the NCS
+# ***************************************************************
+device.OpenDevice()
+
+network_blob='graph'
+
+#Load blob
+with open(network_blob, mode='rb') as f:
+ blob = f.read()
+
+graph = device.AllocateGraph(blob)
+
+# ***************************************************************
+# Load the image
+# ***************************************************************
+ilsvrc_mean = numpy.load(EXAMPLES_BASE_DIR+'data/ilsvrc12/ilsvrc_2012_mean.npy').mean(1).mean(1) #loading the mean file
+img = cv2.imread(EXAMPLES_BASE_DIR+'data/images/nps_electric_guitar.png')
+img=cv2.resize(img,dim)
+img = img.astype(numpy.float32)
+img[:,:,0] = (img[:,:,0] - ilsvrc_mean[0])
+img[:,:,1] = (img[:,:,1] - ilsvrc_mean[1])
+img[:,:,2] = (img[:,:,2] - ilsvrc_mean[2])
+
+# ***************************************************************
+# Send the image to the NCS
+# ***************************************************************
+graph.LoadTensor(img.astype(numpy.float16), 'user object')
+
+# ***************************************************************
+# Get the result from the NCS
+# ***************************************************************
+output, userobj = graph.GetResult()
+
+# ***************************************************************
+# Print the results of the inference form the NCS
+# ***************************************************************
+order = output.argsort()[::-1][:6]
+print('\n------- predictions --------')
+for i in range(0,5):
+ print ('prediction ' + str(i) + ' (probability ' + str(output[order[i]]) + ') is ' + labels[order[i]] + ' label index is: ' + str(order[i]) )
+
+
+# ***************************************************************
+# Clean up the graph and the device
+# ***************************************************************
+graph.DeallocateGraph()
+device.CloseDevice()
+
+
+
+
--- /dev/null
+
+ifneq ($(findstring movidius, $(PYTHONPATH)), movidius)
+ export PYTHONPATH:=/opt/movidius/caffe/python:/opt/movidius/mvnc/python:$(PYTHONPATH)
+endif
+
+TOPTARGETS := all clean check compile profile run
+
+SUBDIRS := $(wildcard */.)
+
+$(TOPTARGETS): $(SUBDIRS)
+$(SUBDIRS):
+ $(MAKE) -C $@ $(MAKECMDGOALS)
+
+.PHONY: $(TOPTARGETS) $(SUBDIRS)
+
+.PHONY: help
+help:
+ @echo "Possible Make targets"
+ @echo " make help - shows this message"
+ @echo " make all - Makes all targets"
+ @echo " make clean - Removes all temp files from all directories"
+ @echo " make check - Runs check on all caffe/tensorflow models"
+ @echo " make profile - Runs profile on all caffe/tensorflow models"
+ @echo " make compile - Runs compile on all caffe/tensorflow models"
+ @echo " make run - Runs all caffe/tensorflow/apps"
--- /dev/null
+
+ifneq ($(findstring movidius, $(PYTHONPATH)), movidius)
+ export PYTHONPATH:=/opt/movidius/caffe/python:/opt/movidius/mvnc/python:$(PYTHONPATH)
+endif
+
+NCCOMPILE = mvNCCompile
+NCPROFILE = mvNCProfile
+NCCHECK = mvNCCheck
+
+
+PROTOTXT_FILENAME= deploy.prototxt
+GET_PROTOTXT = wget -P . https://github.com/DeepScale/SqueezeNet/raw/master/SqueezeNet_v1.0/${PROTOTXT_FILENAME}
+
+CAFFEMODEL_FILENAME = squeezenet_v1.0.caffemodel
+GET_CAFFEMODEL = wget -P . -N https://github.com/DeepScale/SqueezeNet/raw/master/SqueezeNet_v1.0/${CAFFEMODEL_FILENAME}
+
+
+.PHONY: all
+all: profile check compile cpp
+
+.PHONY: prereqs
+prereqs:
+ @echo "\nmaking prereqs"
+ (cd ../../data/ilsvrc12; make)
+ @sed -i 's/\r//' run.py
+ @chmod +x run.py
+ @if [ -e LICENSE.squeezenet ] ; \
+ then \
+ echo "LICENSE file already exists"; \
+ else \
+ echo "Downloading LICENSE file"; \
+ wget -P . -N https://github.com/DeepScale/SqueezeNet/raw/master/LICENSE; mv LICENSE LICENSE.squeezenet; \
+ fi ; \
+
+
+.PHONY: prototxt
+prototxt: prereqs
+ @echo "\nmaking prototxt"
+ @if [ -e ${PROTOTXT_FILENAME} ] ; \
+ then \
+ echo "Prototxt file already exists"; \
+ else \
+ echo "Downloading Prototxt file"; \
+ ${GET_PROTOTXT}; \
+ if [ -e ${PROTOTXT_FILENAME} ] ; \
+ then \
+ echo "modifying prototxt file."; \
+ sed -i '11s/10/1/' ${PROTOTXT_FILENAME}; sed -i '530s/pad/#pad/' ${PROTOTXT_FILENAME}; \
+ else \
+ echo "***\nError - Could not download prototxt file. Check network and proxy settings \n***\n"; \
+ exit 1; \
+ fi ; \
+ fi
+
+.PHONY: caffemodel
+caffemodel:
+ @echo "\nmaking caffemodel"
+ @if [ -e ${CAFFEMODEL_FILENAME} ] ; \
+ then \
+ echo "caffemodel file already exists"; \
+ else \
+ echo "Downloading caffemodel file"; \
+ ${GET_CAFFEMODEL}; \
+ if ! [ -e ${CAFFEMODEL_FILENAME} ] ; \
+ then \
+ echo "***\nError - Could not download caffemodel file. Check network and proxy settings \n***\n"; \
+ exit 1; \
+ fi ; \
+ fi
+
+.PHONY: profile
+profile: prototxt
+ @echo "\nmaking profile"
+ ${NCPROFILE} ${PROTOTXT_FILENAME} -s 12
+
+.PHONY: browse_profile
+browse_profile: profile
+ @echo "\nmaking browse_profile"
+ @if [ -e output_report.html ] ; \
+ then \
+ firefox output_report.html & \
+ else \
+ @echo "***\nError - output_report.html not found" ; \
+ fi ;
+
+.PHONY: compile
+compile: prototxt caffemodel
+ @echo "\nmaking compile"
+ ${NCCOMPILE} -w ${CAFFEMODEL_FILENAME} -s 12 ${PROTOTXT_FILENAME}
+
+.PHONY: check
+check: prototxt caffemodel
+ @echo "\nmaking check"
+#-${NCCHECK} -w ${CAFFEMODEL_FILENAME} -i ../../data/images/cat.jpg -s 12 -id 281 ${PROTOTXT_FILENAME} -S 255 -M ../../data/ilsvrc12/ilsvrc_2012_mean.npy
+ ${NCCHECK} -w ${CAFFEMODEL_FILENAME} -i ../../data/images/cat.jpg -s 12 -id 281 ${PROTOTXT_FILENAME} -S 255 -M 120
+
+.PHONY: run
+run: compile
+ @echo "\nmaking run"
+ ./run.py
+
+.PHONY: run_py
+run_py: compile
+ @echo "\nmaking run_py"
+ ./run.py
+
+.PHONY: cpp
+cpp: cpp/run.cpp cpp/fp16.c
+ @echo "\nmaking cpp"
+ g++ cpp/run.cpp cpp/fp16.c -o cpp/run_cpp -lmvnc
+
+.PHONY: run_cpp
+run_cpp: compile cpp
+ @echo "\nmaking run_cpp"
+ cd cpp; ./run_cpp; cd ..
+
+.PHONY: help
+help:
+ @echo "possible make targets: ";
+ @echo " make help - shows this message";
+ @echo " make all - makes the following: prototxt, profile, compile, check, cpp, run_py, run_cpp";
+ @echo " make prototxt - downloads and adds input shape to Caffe prototxt file";
+ @echo " make caffemodel - downloads the caffemodel for the network"
+ @echo " make compile - runs SDK compiler tool to compile the NCS graph file for the network";
+ @echo " make check - runs SDK checker tool to verify an NCS graph file";
+ @echo " make profile - runs the SDK profiler tool to profile the network creating output_report.html";
+ @echo " make browse_profile - runs the SDK profiler tool and brings up report in browser.";
+ @echo " make cpp - builds the run_cpp executable example";
+ @echo " make run_cpp - runs the run_cpp executable example program";
+ @echo " make run_py - runs the run.py python example program";
+ @echo " make clean - removes all created content"
+
+clean_caffe_model:
+ @echo "\nmaking clean_caffe_model"
+ rm -f ${PROTOTXT_FILENAME}
+ rm -f ${CAFFEMODEL_FILENAME}
+
+clean: clean_caffe_model
+ @echo "\nmaking clean"
+ rm -f graph
+ rm -f output.gv
+ rm -f output.gv.svg
+ rm -f output_report.html
+ rm -f output_expected.npy
+ rm -f zero_weights.caffemodel
+ rm -f output_result.npy
+ rm -f output_val.csv
+ rm -f cpp/run_cpp
+ rm -f LICENSE.squeezenet
--- /dev/null
+# Introduction
+The [SqueezeNet V1.0](https://github.com/DeepScale/SqueezeNet) network can be used for image classification. The provided Makefile does the following
+1. Downloads the Caffe prototxt file and makes any changes necessary to work with the Movidius Neural Compute SDK
+2. Downloads and generates the required ilsvrc12 data
+3. Downloads the .caffemodel file which was trained and provided by BVLC.
+3. Profiles, Compiles and Checks the network using the Neural Compute SDK.
+4. There is a python example (run.py) and a C++ example (cpp/run.cpp) which both do a single inference on an image as an example of how to use this network with the Neural Compute API thats provided in the Neural Compute SDK.
+
+# Makefile
+Provided Makefile describes various targets that help with the above mentioned tasks.
+
+## make help
+Shows makefile possible targets and brief descriptions.
+
+## make all
+Makes the following: prototxt, caffemodel, profile, compile, check, cpp, run, run_cpp.
+
+## make prototxt
+Downloads the Caffe prototxt file and makes a few changes necessary to work with the Movidius Neural Compute SDK.
+
+## make caffemodel
+Downloads the Caffe model file
+
+## make profile
+Runs the provided network on the NCS and generates per layer statistics that are helpful for understanding the performance of the network on the Neural Compute Stick. Output diplayed on terminal and the output_report.html file is also created. Demonstrates NCSDK tool: mvNCProfile
+
+## make browse_profile
+profiles the network similar to make profile and then brings up output_report.html in a browser. Demonstrates NCSDK tool: mvNCProfile
+
+## make compile
+Uses the network description and the trained weights files to generate a Movidius internal 'graph' format file. This file is later loaded on the Neural Compute Stick where the inferences on the network can be executed. Demonstrates NCSDK tool: mvNCCompile
+
+## make check
+Runs the network on Caffe on the CPU and compares results when run on the Neural Compute Stick. Consistency results are output to the terminal. Demonstrates the NCSDK tool: mvNCCheck.
+
+## make run_py
+Runs the provided run.py python script which sends a single image to the Neural Compute Stick and receives and displays the inference results.
+
+## make cpp
+Builds the C++ example program run_cpp which can be executed with make run_cpp.
+
+## make run_cpp
+Runs the provided run_cpp executable program that is built via make cpp. This program sends a single image to the Neural Compute Stick and receives and displays the inference results.
+
+## make clean
+Removes all the temporary and target files that are created by the Makefile.
--- /dev/null
+#! /bin/bash
+
+g++ run.cpp fp16.c -o run_cpp -lmvnc
+
+
+
--- /dev/null
+#include "fp16.h"
+
+// Copied from Numpy
+
+static unsigned half2float(unsigned short h)
+{
+ unsigned short h_exp, h_sig;
+ unsigned f_sgn, f_exp, f_sig;
+
+ h_exp = (h&0x7c00u);
+ f_sgn = ((unsigned)h&0x8000u) << 16;
+ switch (h_exp) {
+ case 0x0000u: /* 0 or subnormal */
+ h_sig = (h&0x03ffu);
+ /* Signed zero */
+ if (h_sig == 0) {
+ return f_sgn;
+ }
+ /* Subnormal */
+ h_sig <<= 1;
+ while ((h_sig&0x0400u) == 0) {
+ h_sig <<= 1;
+ h_exp++;
+ }
+ f_exp = ((unsigned)(127 - 15 - h_exp)) << 23;
+ f_sig = ((unsigned)(h_sig&0x03ffu)) << 13;
+ return f_sgn + f_exp + f_sig;
+ case 0x7c00u: /* inf or NaN */
+ /* All-ones exponent and a copy of the significand */
+ return f_sgn + 0x7f800000u + (((unsigned)(h&0x03ffu)) << 13);
+ default: /* normalized */
+ /* Just need to adjust the exponent and shift */
+ return f_sgn + (((unsigned)(h&0x7fffu) + 0x1c000u) << 13);
+ }
+}
+
+unsigned short float2half(unsigned f)
+{
+ unsigned f_exp, f_sig;
+ unsigned short h_sgn, h_exp, h_sig;
+
+ h_sgn = (unsigned short) ((f&0x80000000u) >> 16);
+ f_exp = (f&0x7f800000u);
+
+ /* Exponent overflow/NaN converts to signed inf/NaN */
+ if (f_exp >= 0x47800000u) {
+ if (f_exp == 0x7f800000u) {
+ /* Inf or NaN */
+ f_sig = (f&0x007fffffu);
+ if (f_sig != 0) {
+ /* NaN - propagate the flag in the significand... */
+ unsigned short ret = (unsigned short) (0x7c00u + (f_sig >> 13));
+ /* ...but make sure it stays a NaN */
+ if (ret == 0x7c00u) {
+ ret++;
+ }
+ return h_sgn + ret;
+ } else {
+ /* signed inf */
+ return (unsigned short) (h_sgn + 0x7c00u);
+ }
+ } else {
+ /* overflow to signed inf */
+#if NPY_HALF_GENERATE_OVERFLOW
+ npy_set_floatstatus_overflow();
+#endif
+ return (unsigned short) (h_sgn + 0x7c00u);
+ }
+ }
+
+ /* Exponent underflow converts to a subnormal half or signed zero */
+ if (f_exp <= 0x38000000u) {
+ /*
+ * Signed zeros, subnormal floats, and floats with small
+ * exponents all convert to signed zero halfs.
+ */
+ if (f_exp < 0x33000000u) {
+#if NPY_HALF_GENERATE_UNDERFLOW
+ /* If f != 0, it underflowed to 0 */
+ if ((f&0x7fffffff) != 0) {
+ npy_set_floatstatus_underflow();
+ }
+#endif
+ return h_sgn;
+ }
+ /* Make the subnormal significand */
+ f_exp >>= 23;
+ f_sig = (0x00800000u + (f&0x007fffffu));
+#if NPY_HALF_GENERATE_UNDERFLOW
+ /* If it's not exactly represented, it underflowed */
+ if ((f_sig&(((unsigned)1 << (126 - f_exp)) - 1)) != 0) {
+ npy_set_floatstatus_underflow();
+ }
+#endif
+ f_sig >>= (113 - f_exp);
+ /* Handle rounding by adding 1 to the bit beyond half precision */
+#if NPY_HALF_ROUND_TIES_TO_EVEN
+ /*
+ * If the last bit in the half significand is 0 (already even), and
+ * the remaining bit pattern is 1000...0, then we do not add one
+ * to the bit after the half significand. In all other cases, we do.
+ */
+ if ((f_sig&0x00003fffu) != 0x00001000u) {
+ f_sig += 0x00001000u;
+ }
+#else
+ f_sig += 0x00001000u;
+#endif
+ h_sig = (unsigned short) (f_sig >> 13);
+ /*
+ * If the rounding causes a bit to spill into h_exp, it will
+ * increment h_exp from zero to one and h_sig will be zero.
+ * This is the correct result.
+ */
+ return (unsigned short) (h_sgn + h_sig);
+ }
+
+ /* Regular case with no overflow or underflow */
+ h_exp = (unsigned short) ((f_exp - 0x38000000u) >> 13);
+ /* Handle rounding by adding 1 to the bit beyond half precision */
+ f_sig = (f&0x007fffffu);
+#if NPY_HALF_ROUND_TIES_TO_EVEN
+ /*
+ * If the last bit in the half significand is 0 (already even), and
+ * the remaining bit pattern is 1000...0, then we do not add one
+ * to the bit after the half significand. In all other cases, we do.
+ */
+ if ((f_sig&0x00003fffu) != 0x00001000u) {
+ f_sig += 0x00001000u;
+ }
+#else
+ f_sig += 0x00001000u;
+#endif
+ h_sig = (unsigned short) (f_sig >> 13);
+ /*
+ * If the rounding causes a bit to spill into h_exp, it will
+ * increment h_exp by one and h_sig will be zero. This is the
+ * correct result. h_exp may increment to 15, at greatest, in
+ * which case the result overflows to a signed inf.
+ */
+#if NPY_HALF_GENERATE_OVERFLOW
+ h_sig += h_exp;
+ if (h_sig == 0x7c00u) {
+ npy_set_floatstatus_overflow();
+ }
+ return h_sgn + h_sig;
+#else
+ return h_sgn + h_exp + h_sig;
+#endif
+}
+
+void floattofp16(unsigned char *dst, float *src, unsigned nelem)
+{
+ unsigned i;
+ unsigned short *_dst = (unsigned short *)dst;
+ unsigned *_src = (unsigned *)src;
+
+ for(i = 0; i < nelem; i++)
+ _dst[i] = float2half(_src[i]);
+}
+
+void fp16tofloat(float *dst, unsigned char *src, unsigned nelem)
+{
+ unsigned i;
+ unsigned *_dst = (unsigned *)dst;
+ unsigned short *_src = (unsigned short *)src;
+
+ for(i = 0; i < nelem; i++)
+ _dst[i] = half2float(_src[i]);
+}
+
--- /dev/null
+// Copied from Numpy
+
+
+static unsigned half2float(unsigned short h);
+unsigned short float2half(unsigned f);
+void floattofp16(unsigned char *dst, float *src, unsigned nelem);
+void fp16tofloat(float *dst, unsigned char *src, unsigned nelem);
+
+
--- /dev/null
+// Copyright 2017 Intel Corporation.
+// The source code, information and material ("Material") contained herein is
+// owned by Intel Corporation or its suppliers or licensors, and title to such
+// Material remains with Intel Corporation or its suppliers or licensors.
+// The Material contains proprietary information of Intel or its suppliers and
+// licensors. The Material is protected by worldwide copyright laws and treaty
+// provisions.
+// No part of the Material may be used, copied, reproduced, modified, published,
+// uploaded, posted, transmitted, distributed or disclosed in any way without
+// Intel's prior express written permission. No license under any patent,
+// copyright or other intellectual property rights in the Material is granted to
+// or conferred upon you, either expressly, by implication, inducement, estoppel
+// or otherwise.
+// Any license under such intellectual property rights must be express and
+// approved by Intel in writing.
+
+#include <stdio.h>
+#include <stdlib.h>
+
+#define STB_IMAGE_IMPLEMENTATION
+#include "stb_image.h"
+
+#define STB_IMAGE_RESIZE_IMPLEMENTATION
+#include "stb_image_resize.h"
+
+#include "fp16.h"
+#include <mvnc.h>
+
+
+// somewhat arbitrary buffer size for the device name
+#define NAME_SIZE 100
+
+// graph file name - assume we are running in this directory: ncsdk/examples/caffe/GoogLeNet/cpp
+#define GRAPH_FILE_NAME "../graph"
+
+// image file name - assume we are running in this directory: ncsdk/examples/caffe/GoogLeNet/cpp
+#define IMAGE_FILE_NAME "../../../data/images/nps_electric_guitar.png"
+
+
+// 16 bits. will use this to store half precision floats since C++ has no
+// built in support for it.
+typedef unsigned short half;
+
+// GoogleNet image dimensions, network mean values for each channel in BGR order.
+const int networkDim = 227;
+float networkMean[] = {0.40787054*255.0, 0.45752458*255.0, 0.48109378*255.0};
+
+// Load a graph file
+// caller must free the buffer returned.
+void *LoadFile(const char *path, unsigned int *length)
+{
+ FILE *fp;
+ char *buf;
+
+ fp = fopen(path, "rb");
+ if(fp == NULL)
+ return 0;
+ fseek(fp, 0, SEEK_END);
+ *length = ftell(fp);
+ rewind(fp);
+ if(!(buf = (char*) malloc(*length)))
+ {
+ fclose(fp);
+ return 0;
+ }
+ if(fread(buf, 1, *length, fp) != *length)
+ {
+ fclose(fp);
+ free(buf);
+ return 0;
+ }
+ fclose(fp);
+ return buf;
+}
+
+
+half *LoadImage(const char *path, int reqsize, float *mean)
+{
+ int width, height, cp, i;
+ unsigned char *img, *imgresized;
+ float *imgfp32;
+ half *imgfp16;
+
+ img = stbi_load(path, &width, &height, &cp, 3);
+ if(!img)
+ {
+ printf("The picture %s could not be loaded\n", path);
+ return 0;
+ }
+ imgresized = (unsigned char*) malloc(3*reqsize*reqsize);
+ if(!imgresized)
+ {
+ free(img);
+ perror("malloc");
+ return 0;
+ }
+ stbir_resize_uint8(img, width, height, 0, imgresized, reqsize, reqsize, 0, 3);
+ free(img);
+ imgfp32 = (float*) malloc(sizeof(*imgfp32) * reqsize * reqsize * 3);
+ if(!imgfp32)
+ {
+ free(imgresized);
+ perror("malloc");
+ return 0;
+ }
+ for(i = 0; i < reqsize * reqsize * 3; i++)
+ imgfp32[i] = imgresized[i];
+ free(imgresized);
+ imgfp16 = (half*) malloc(sizeof(*imgfp16) * reqsize * reqsize * 3);
+ if(!imgfp16)
+ {
+ free(imgfp32);
+ perror("malloc");
+ return 0;
+ }
+ for(i = 0; i < reqsize*reqsize; i++)
+ {
+ float blue, green, red;
+ blue = imgfp32[3*i+2];
+ green = imgfp32[3*i+1];
+ red = imgfp32[3*i+0];
+
+ imgfp32[3*i+0] = blue-mean[0];
+ imgfp32[3*i+1] = green-mean[1];
+ imgfp32[3*i+2] = red-mean[2];
+
+ // uncomment to see what values are getting passed to mvncLoadTensor() before conversion to half float
+ //printf("Blue: %f, Grean: %f, Red: %f \n", imgfp32[3*i+0], imgfp32[3*i+1], imgfp32[3*i+2]);
+ }
+ floattofp16((unsigned char *)imgfp16, imgfp32, 3*reqsize*reqsize);
+ free(imgfp32);
+ return imgfp16;
+}
+
+
+int main(int argc, char** argv)
+{
+ mvncStatus retCode;
+ void *deviceHandle;
+ char devName[NAME_SIZE];
+ retCode = mvncGetDeviceName(0, devName, NAME_SIZE);
+ if (retCode != MVNC_OK)
+ { // failed to get device name, maybe none plugged in.
+ printf("No NCS devices found\n");
+ exit(-1);
+ }
+
+ // Try to open the NCS device via the device name
+ retCode = mvncOpenDevice(devName, &deviceHandle);
+ if (retCode != MVNC_OK)
+ { // failed to open the device.
+ printf("Could not open NCS device\n");
+ exit(-1);
+ }
+
+ // deviceHandle is ready to use now.
+ // Pass it to other NC API calls as needed and close it when finished.
+ printf("Successfully opened NCS device!\n");
+
+ // Now read in a graph file
+ unsigned int graphFileLen;
+ void* graphFileBuf = LoadFile(GRAPH_FILE_NAME, &graphFileLen);
+
+ // allocate the graph
+ void* graphHandle;
+ retCode = mvncAllocateGraph(deviceHandle, &graphHandle, graphFileBuf, graphFileLen);
+ if (retCode != MVNC_OK)
+ { // error allocating graph
+ printf("Could not allocate graph for file: %s\n", GRAPH_FILE_NAME);
+ printf("Error from mvncAllocateGraph is: %d\n", retCode);
+ }
+ else
+ { // successfully allocated graph. Now graphHandle is ready to go.
+ // use graphHandle for other API calls and call mvncDeallocateGraph
+ // when done with it.
+ printf("Successfully allocated graph for %s\n", GRAPH_FILE_NAME);
+
+ // LoadImage will read image from disk, convert channels to floats
+ // subtract network mean for each value in each channel. Then, convert
+ // floats to half precision floats and return pointer to the buffer
+ // of half precision floats (Fp16s)
+ half* imageBufFp16 = LoadImage(IMAGE_FILE_NAME, networkDim, networkMean);
+
+ // calculate the length of the buffer that contains the half precision floats.
+ // 3 channels * width * height * sizeof a 16bit float
+ unsigned int lenBufFp16 = 3*networkDim*networkDim*sizeof(*imageBufFp16);
+
+ // start the inference with mvncLoadTensor()
+ retCode = mvncLoadTensor(graphHandle, imageBufFp16, lenBufFp16, NULL);
+ if (retCode != MVNC_OK)
+ { // error loading tensor
+ printf("Could not load tensor\n");
+ printf("Error from mvncLoadTensor is: %d\n", retCode);
+ }
+ else
+ { // the inference has been started, now call mvncGetResult() for the
+ // inference result
+ printf("Successfully loaded the tensor for image %s\n", IMAGE_FILE_NAME);
+
+ void* resultData16;
+ void* userParam;
+ unsigned int lenResultData;
+ retCode = mvncGetResult(graphHandle, &resultData16, &lenResultData, &userParam);
+ if (retCode == MVNC_OK)
+ { // Successfully got the result. The inference result is in the buffer pointed to by resultData
+ printf("Successfully got the inference result for image %s\n", IMAGE_FILE_NAME);
+ printf("resultData is %d bytes which is %d 16-bit floats.\n", lenResultData, lenResultData/(int)sizeof(half));
+
+ // convert half precision floats to full floats
+ int numResults = lenResultData / sizeof(half);
+ float* resultData32;
+ resultData32 = (float*)malloc(numResults * sizeof(*resultData32));
+ fp16tofloat(resultData32, (unsigned char*)resultData16, numResults);
+
+ float maxResult = 0.0;
+ int maxIndex = -1;
+ for (int index = 0; index < numResults; index++)
+ {
+ // printf("Category %d is: %f\n", index, resultData32[index]);
+ if (resultData32[index] > maxResult)
+ {
+ maxResult = resultData32[index];
+ maxIndex = index;
+ }
+ }
+ printf("Index of top result is: %d\n", maxIndex);
+ printf("Probability of top result is: %f\n", resultData32[maxIndex]);
+ }
+ }
+
+ retCode = mvncDeallocateGraph(graphHandle);
+ graphHandle = NULL;
+ }
+
+ free(graphFileBuf);
+ retCode = mvncCloseDevice(deviceHandle);
+ deviceHandle = NULL;
+}
--- /dev/null
+/* stb_image - v2.15 - public domain image loader - http://nothings.org/stb_image.h
+ no warranty implied; use at your own risk
+
+ Do this:
+ #define STB_IMAGE_IMPLEMENTATION
+ before you include this file in *one* C or C++ file to create the implementation.
+
+ // i.e. it should look like this:
+ #include ...
+ #include ...
+ #include ...
+ #define STB_IMAGE_IMPLEMENTATION
+ #include "stb_image.h"
+
+ You can #define STBI_ASSERT(x) before the #include to avoid using assert.h.
+ And #define STBI_MALLOC, STBI_REALLOC, and STBI_FREE to avoid using malloc,realloc,free
+
+
+ QUICK NOTES:
+ Primarily of interest to game developers and other people who can
+ avoid problematic images and only need the trivial interface
+
+ JPEG baseline & progressive (12 bpc/arithmetic not supported, same as stock IJG lib)
+ PNG 1/2/4/8/16-bit-per-channel
+
+ TGA (not sure what subset, if a subset)
+ BMP non-1bpp, non-RLE
+ PSD (composited view only, no extra channels, 8/16 bit-per-channel)
+
+ GIF (*comp always reports as 4-channel)
+ HDR (radiance rgbE format)
+ PIC (Softimage PIC)
+ PNM (PPM and PGM binary only)
+
+ Animated GIF still needs a proper API, but here's one way to do it:
+ http://gist.github.com/urraka/685d9a6340b26b830d49
+
+ - decode from memory or through FILE (define STBI_NO_STDIO to remove code)
+ - decode from arbitrary I/O callbacks
+ - SIMD acceleration on x86/x64 (SSE2) and ARM (NEON)
+
+ Full documentation under "DOCUMENTATION" below.
+
+
+LICENSE
+
+ See end of file for license information.
+
+RECENT REVISION HISTORY:
+
+ 2.15 (2017-03-18) fix png-1,2,4; all Imagenet JPGs; no runtime SSE detection on GCC
+ 2.14 (2017-03-03) remove deprecated STBI_JPEG_OLD; fixes for Imagenet JPGs
+ 2.13 (2016-12-04) experimental 16-bit API, only for PNG so far; fixes
+ 2.12 (2016-04-02) fix typo in 2.11 PSD fix that caused crashes
+ 2.11 (2016-04-02) 16-bit PNGS; enable SSE2 in non-gcc x64
+ RGB-format JPEG; remove white matting in PSD;
+ allocate large structures on the stack;
+ correct channel count for PNG & BMP
+ 2.10 (2016-01-22) avoid warning introduced in 2.09
+ 2.09 (2016-01-16) 16-bit TGA; comments in PNM files; STBI_REALLOC_SIZED
+ 2.08 (2015-09-13) fix to 2.07 cleanup, reading RGB PSD as RGBA
+ 2.07 (2015-09-13) partial animated GIF support
+ limited 16-bit PSD support
+ minor bugs, code cleanup, and compiler warnings
+
+ See end of file for full revision history.
+
+
+ ============================ Contributors =========================
+
+ Image formats Extensions, features
+ Sean Barrett (jpeg, png, bmp) Jetro Lauha (stbi_info)
+ Nicolas Schulz (hdr, psd) Martin "SpartanJ" Golini (stbi_info)
+ Jonathan Dummer (tga) James "moose2000" Brown (iPhone PNG)
+ Jean-Marc Lienher (gif) Ben "Disch" Wenger (io callbacks)
+ Tom Seddon (pic) Omar Cornut (1/2/4-bit PNG)
+ Thatcher Ulrich (psd) Nicolas Guillemot (vertical flip)
+ Ken Miller (pgm, ppm) Richard Mitton (16-bit PSD)
+ github:urraka (animated gif) Junggon Kim (PNM comments)
+ Daniel Gibson (16-bit TGA)
+ socks-the-fox (16-bit PNG)
+ Jeremy Sawicki (handle all ImageNet JPGs)
+ Optimizations & bugfixes
+ Fabian "ryg" Giesen
+ Arseny Kapoulkine
+
+ Bug & warning fixes
+ Marc LeBlanc David Woo Guillaume George Martins Mozeiko
+ Christpher Lloyd Jerry Jansson Joseph Thomson Phil Jordan
+ Dave Moore Roy Eltham Hayaki Saito Nathan Reed
+ Won Chun Luke Graham Johan Duparc Nick Verigakis
+ the Horde3D community Thomas Ruf Ronny Chevalier Baldur Karlsson
+ Janez Zemva John Bartholomew Michal Cichon github:rlyeh
+ Jonathan Blow Ken Hamada Tero Hanninen github:romigrou
+ Laurent Gomila Cort Stratton Sergio Gonzalez github:svdijk
+ Aruelien Pocheville Thibault Reuille Cass Everitt github:snagar
+ Ryamond Barbiero Paul Du Bois Engin Manap github:Zelex
+ Michaelangel007@github Philipp Wiesemann Dale Weiler github:grim210
+ Oriol Ferrer Mesia Josh Tobin Matthew Gregan github:sammyhw
+ Blazej Dariusz Roszkowski Gregory Mullen github:phprus
+
+*/
+
+#ifndef STBI_INCLUDE_STB_IMAGE_H
+#define STBI_INCLUDE_STB_IMAGE_H
+
+// DOCUMENTATION
+//
+// Limitations:
+// - no 16-bit-per-channel PNG
+// - no 12-bit-per-channel JPEG
+// - no JPEGs with arithmetic coding
+// - no 1-bit BMP
+// - GIF always returns *comp=4
+//
+// Basic usage (see HDR discussion below for HDR usage):
+// int x,y,n;
+// unsigned char *data = stbi_load(filename, &x, &y, &n, 0);
+// // ... process data if not NULL ...
+// // ... x = width, y = height, n = # 8-bit components per pixel ...
+// // ... replace '0' with '1'..'4' to force that many components per pixel
+// // ... but 'n' will always be the number that it would have been if you said 0
+// stbi_image_free(data)
+//
+// Standard parameters:
+// int *x -- outputs image width in pixels
+// int *y -- outputs image height in pixels
+// int *channels_in_file -- outputs # of image components in image file
+// int desired_channels -- if non-zero, # of image components requested in result
+//
+// The return value from an image loader is an 'unsigned char *' which points
+// to the pixel data, or NULL on an allocation failure or if the image is
+// corrupt or invalid. The pixel data consists of *y scanlines of *x pixels,
+// with each pixel consisting of N interleaved 8-bit components; the first
+// pixel pointed to is top-left-most in the image. There is no padding between
+// image scanlines or between pixels, regardless of format. The number of
+// components N is 'req_comp' if req_comp is non-zero, or *comp otherwise.
+// If req_comp is non-zero, *comp has the number of components that _would_
+// have been output otherwise. E.g. if you set req_comp to 4, you will always
+// get RGBA output, but you can check *comp to see if it's trivially opaque
+// because e.g. there were only 3 channels in the source image.
+//
+// An output image with N components has the following components interleaved
+// in this order in each pixel:
+//
+// N=#comp components
+// 1 grey
+// 2 grey, alpha
+// 3 red, green, blue
+// 4 red, green, blue, alpha
+//
+// If image loading fails for any reason, the return value will be NULL,
+// and *x, *y, *comp will be unchanged. The function stbi_failure_reason()
+// can be queried for an extremely brief, end-user unfriendly explanation
+// of why the load failed. Define STBI_NO_FAILURE_STRINGS to avoid
+// compiling these strings at all, and STBI_FAILURE_USERMSG to get slightly
+// more user-friendly ones.
+//
+// Paletted PNG, BMP, GIF, and PIC images are automatically depalettized.
+//
+// ===========================================================================
+//
+// Philosophy
+//
+// stb libraries are designed with the following priorities:
+//
+// 1. easy to use
+// 2. easy to maintain
+// 3. good performance
+//
+// Sometimes I let "good performance" creep up in priority over "easy to maintain",
+// and for best performance I may provide less-easy-to-use APIs that give higher
+// performance, in addition to the easy to use ones. Nevertheless, it's important
+// to keep in mind that from the standpoint of you, a client of this library,
+// all you care about is #1 and #3, and stb libraries DO NOT emphasize #3 above all.
+//
+// Some secondary priorities arise directly from the first two, some of which
+// make more explicit reasons why performance can't be emphasized.
+//
+// - Portable ("ease of use")
+// - Small source code footprint ("easy to maintain")
+// - No dependencies ("ease of use")
+//
+// ===========================================================================
+//
+// I/O callbacks
+//
+// I/O callbacks allow you to read from arbitrary sources, like packaged
+// files or some other source. Data read from callbacks are processed
+// through a small internal buffer (currently 128 bytes) to try to reduce
+// overhead.
+//
+// The three functions you must define are "read" (reads some bytes of data),
+// "skip" (skips some bytes of data), "eof" (reports if the stream is at the end).
+//
+// ===========================================================================
+//
+// SIMD support
+//
+// The JPEG decoder will try to automatically use SIMD kernels on x86 when
+// supported by the compiler. For ARM Neon support, you must explicitly
+// request it.
+//
+// (The old do-it-yourself SIMD API is no longer supported in the current
+// code.)
+//
+// On x86, SSE2 will automatically be used when available based on a run-time
+// test; if not, the generic C versions are used as a fall-back. On ARM targets,
+// the typical path is to have separate builds for NEON and non-NEON devices
+// (at least this is true for iOS and Android). Therefore, the NEON support is
+// toggled by a build flag: define STBI_NEON to get NEON loops.
+//
+// If for some reason you do not want to use any of SIMD code, or if
+// you have issues compiling it, you can disable it entirely by
+// defining STBI_NO_SIMD.
+//
+// ===========================================================================
+//
+// HDR image support (disable by defining STBI_NO_HDR)
+//
+// stb_image now supports loading HDR images in general, and currently
+// the Radiance .HDR file format, although the support is provided
+// generically. You can still load any file through the existing interface;
+// if you attempt to load an HDR file, it will be automatically remapped to
+// LDR, assuming gamma 2.2 and an arbitrary scale factor defaulting to 1;
+// both of these constants can be reconfigured through this interface:
+//
+// stbi_hdr_to_ldr_gamma(2.2f);
+// stbi_hdr_to_ldr_scale(1.0f);
+//
+// (note, do not use _inverse_ constants; stbi_image will invert them
+// appropriately).
+//
+// Additionally, there is a new, parallel interface for loading files as
+// (linear) floats to preserve the full dynamic range:
+//
+// float *data = stbi_loadf(filename, &x, &y, &n, 0);
+//
+// If you load LDR images through this interface, those images will
+// be promoted to floating point values, run through the inverse of
+// constants corresponding to the above:
+//
+// stbi_ldr_to_hdr_scale(1.0f);
+// stbi_ldr_to_hdr_gamma(2.2f);
+//
+// Finally, given a filename (or an open file or memory block--see header
+// file for details) containing image data, you can query for the "most
+// appropriate" interface to use (that is, whether the image is HDR or
+// not), using:
+//
+// stbi_is_hdr(char *filename);
+//
+// ===========================================================================
+//
+// iPhone PNG support:
+//
+// By default we convert iphone-formatted PNGs back to RGB, even though
+// they are internally encoded differently. You can disable this conversion
+// by by calling stbi_convert_iphone_png_to_rgb(0), in which case
+// you will always just get the native iphone "format" through (which
+// is BGR stored in RGB).
+//
+// Call stbi_set_unpremultiply_on_load(1) as well to force a divide per
+// pixel to remove any premultiplied alpha *only* if the image file explicitly
+// says there's premultiplied data (currently only happens in iPhone images,
+// and only if iPhone convert-to-rgb processing is on).
+//
+// ===========================================================================
+//
+// ADDITIONAL CONFIGURATION
+//
+// - You can suppress implementation of any of the decoders to reduce
+// your code footprint by #defining one or more of the following
+// symbols before creating the implementation.
+//
+// STBI_NO_JPEG
+// STBI_NO_PNG
+// STBI_NO_BMP
+// STBI_NO_PSD
+// STBI_NO_TGA
+// STBI_NO_GIF
+// STBI_NO_HDR
+// STBI_NO_PIC
+// STBI_NO_PNM (.ppm and .pgm)
+//
+// - You can request *only* certain decoders and suppress all other ones
+// (this will be more forward-compatible, as addition of new decoders
+// doesn't require you to disable them explicitly):
+//
+// STBI_ONLY_JPEG
+// STBI_ONLY_PNG
+// STBI_ONLY_BMP
+// STBI_ONLY_PSD
+// STBI_ONLY_TGA
+// STBI_ONLY_GIF
+// STBI_ONLY_HDR
+// STBI_ONLY_PIC
+// STBI_ONLY_PNM (.ppm and .pgm)
+//
+// - If you use STBI_NO_PNG (or _ONLY_ without PNG), and you still
+// want the zlib decoder to be available, #define STBI_SUPPORT_ZLIB
+//
+
+
+#ifndef STBI_NO_STDIO
+#include <stdio.h>
+#endif // STBI_NO_STDIO
+
+#define STBI_VERSION 1
+
+enum
+{
+ STBI_default = 0, // only used for req_comp
+
+ STBI_grey = 1,
+ STBI_grey_alpha = 2,
+ STBI_rgb = 3,
+ STBI_rgb_alpha = 4
+};
+
+typedef unsigned char stbi_uc;
+typedef unsigned short stbi_us;
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifdef STB_IMAGE_STATIC
+#define STBIDEF static
+#else
+#define STBIDEF extern
+#endif
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// PRIMARY API - works on images of any type
+//
+
+//
+// load image by filename, open file, or memory buffer
+//
+
+typedef struct
+{
+ int (*read) (void *user,char *data,int size); // fill 'data' with 'size' bytes. return number of bytes actually read
+ void (*skip) (void *user,int n); // skip the next 'n' bytes, or 'unget' the last -n bytes if negative
+ int (*eof) (void *user); // returns nonzero if we are at end of file/data
+} stbi_io_callbacks;
+
+////////////////////////////////////
+//
+// 8-bits-per-channel interface
+//
+
+STBIDEF stbi_uc *stbi_load (char const *filename, int *x, int *y, int *channels_in_file, int desired_channels);
+STBIDEF stbi_uc *stbi_load_from_memory (stbi_uc const *buffer, int len , int *x, int *y, int *channels_in_file, int desired_channels);
+STBIDEF stbi_uc *stbi_load_from_callbacks(stbi_io_callbacks const *clbk , void *user, int *x, int *y, int *channels_in_file, int desired_channels);
+
+#ifndef STBI_NO_STDIO
+STBIDEF stbi_uc *stbi_load_from_file (FILE *f, int *x, int *y, int *channels_in_file, int desired_channels);
+// for stbi_load_from_file, file pointer is left pointing immediately after image
+#endif
+
+////////////////////////////////////
+//
+// 16-bits-per-channel interface
+//
+
+STBIDEF stbi_us *stbi_load_16(char const *filename, int *x, int *y, int *channels_in_file, int desired_channels);
+#ifndef STBI_NO_STDIO
+STBIDEF stbi_us *stbi_load_from_file_16(FILE *f, int *x, int *y, int *channels_in_file, int desired_channels);
+#endif
+// @TODO the other variants
+
+////////////////////////////////////
+//
+// float-per-channel interface
+//
+#ifndef STBI_NO_LINEAR
+ STBIDEF float *stbi_loadf (char const *filename, int *x, int *y, int *channels_in_file, int desired_channels);
+ STBIDEF float *stbi_loadf_from_memory (stbi_uc const *buffer, int len, int *x, int *y, int *channels_in_file, int desired_channels);
+ STBIDEF float *stbi_loadf_from_callbacks (stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *channels_in_file, int desired_channels);
+
+ #ifndef STBI_NO_STDIO
+ STBIDEF float *stbi_loadf_from_file (FILE *f, int *x, int *y, int *channels_in_file, int desired_channels);
+ #endif
+#endif
+
+#ifndef STBI_NO_HDR
+ STBIDEF void stbi_hdr_to_ldr_gamma(float gamma);
+ STBIDEF void stbi_hdr_to_ldr_scale(float scale);
+#endif // STBI_NO_HDR
+
+#ifndef STBI_NO_LINEAR
+ STBIDEF void stbi_ldr_to_hdr_gamma(float gamma);
+ STBIDEF void stbi_ldr_to_hdr_scale(float scale);
+#endif // STBI_NO_LINEAR
+
+// stbi_is_hdr is always defined, but always returns false if STBI_NO_HDR
+STBIDEF int stbi_is_hdr_from_callbacks(stbi_io_callbacks const *clbk, void *user);
+STBIDEF int stbi_is_hdr_from_memory(stbi_uc const *buffer, int len);
+#ifndef STBI_NO_STDIO
+STBIDEF int stbi_is_hdr (char const *filename);
+STBIDEF int stbi_is_hdr_from_file(FILE *f);
+#endif // STBI_NO_STDIO
+
+
+// get a VERY brief reason for failure
+// NOT THREADSAFE
+STBIDEF const char *stbi_failure_reason (void);
+
+// free the loaded image -- this is just free()
+STBIDEF void stbi_image_free (void *retval_from_stbi_load);
+
+// get image dimensions & components without fully decoding
+STBIDEF int stbi_info_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp);
+STBIDEF int stbi_info_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp);
+
+#ifndef STBI_NO_STDIO
+STBIDEF int stbi_info (char const *filename, int *x, int *y, int *comp);
+STBIDEF int stbi_info_from_file (FILE *f, int *x, int *y, int *comp);
+
+#endif
+
+
+
+// for image formats that explicitly notate that they have premultiplied alpha,
+// we just return the colors as stored in the file. set this flag to force
+// unpremultiplication. results are undefined if the unpremultiply overflow.
+STBIDEF void stbi_set_unpremultiply_on_load(int flag_true_if_should_unpremultiply);
+
+// indicate whether we should process iphone images back to canonical format,
+// or just pass them through "as-is"
+STBIDEF void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert);
+
+// flip the image vertically, so the first pixel in the output array is the bottom left
+STBIDEF void stbi_set_flip_vertically_on_load(int flag_true_if_should_flip);
+
+// ZLIB client - used by PNG, available for other purposes
+
+STBIDEF char *stbi_zlib_decode_malloc_guesssize(const char *buffer, int len, int initial_size, int *outlen);
+STBIDEF char *stbi_zlib_decode_malloc_guesssize_headerflag(const char *buffer, int len, int initial_size, int *outlen, int parse_header);
+STBIDEF char *stbi_zlib_decode_malloc(const char *buffer, int len, int *outlen);
+STBIDEF int stbi_zlib_decode_buffer(char *obuffer, int olen, const char *ibuffer, int ilen);
+
+STBIDEF char *stbi_zlib_decode_noheader_malloc(const char *buffer, int len, int *outlen);
+STBIDEF int stbi_zlib_decode_noheader_buffer(char *obuffer, int olen, const char *ibuffer, int ilen);
+
+
+#ifdef __cplusplus
+}
+#endif
+
+//
+//
+//// end header file /////////////////////////////////////////////////////
+#endif // STBI_INCLUDE_STB_IMAGE_H
+
+#ifdef STB_IMAGE_IMPLEMENTATION
+
+#if defined(STBI_ONLY_JPEG) || defined(STBI_ONLY_PNG) || defined(STBI_ONLY_BMP) \
+ || defined(STBI_ONLY_TGA) || defined(STBI_ONLY_GIF) || defined(STBI_ONLY_PSD) \
+ || defined(STBI_ONLY_HDR) || defined(STBI_ONLY_PIC) || defined(STBI_ONLY_PNM) \
+ || defined(STBI_ONLY_ZLIB)
+ #ifndef STBI_ONLY_JPEG
+ #define STBI_NO_JPEG
+ #endif
+ #ifndef STBI_ONLY_PNG
+ #define STBI_NO_PNG
+ #endif
+ #ifndef STBI_ONLY_BMP
+ #define STBI_NO_BMP
+ #endif
+ #ifndef STBI_ONLY_PSD
+ #define STBI_NO_PSD
+ #endif
+ #ifndef STBI_ONLY_TGA
+ #define STBI_NO_TGA
+ #endif
+ #ifndef STBI_ONLY_GIF
+ #define STBI_NO_GIF
+ #endif
+ #ifndef STBI_ONLY_HDR
+ #define STBI_NO_HDR
+ #endif
+ #ifndef STBI_ONLY_PIC
+ #define STBI_NO_PIC
+ #endif
+ #ifndef STBI_ONLY_PNM
+ #define STBI_NO_PNM
+ #endif
+#endif
+
+#if defined(STBI_NO_PNG) && !defined(STBI_SUPPORT_ZLIB) && !defined(STBI_NO_ZLIB)
+#define STBI_NO_ZLIB
+#endif
+
+
+#include <stdarg.h>
+#include <stddef.h> // ptrdiff_t on osx
+#include <stdlib.h>
+#include <string.h>
+#include <limits.h>
+
+#if !defined(STBI_NO_LINEAR) || !defined(STBI_NO_HDR)
+#include <math.h> // ldexp
+#endif
+
+#ifndef STBI_NO_STDIO
+#include <stdio.h>
+#endif
+
+#ifndef STBI_ASSERT
+#include <assert.h>
+#define STBI_ASSERT(x) assert(x)
+#endif
+
+
+#ifndef _MSC_VER
+ #ifdef __cplusplus
+ #define stbi_inline inline
+ #else
+ #define stbi_inline
+ #endif
+#else
+ #define stbi_inline __forceinline
+#endif
+
+
+#ifdef _MSC_VER
+typedef unsigned short stbi__uint16;
+typedef signed short stbi__int16;
+typedef unsigned int stbi__uint32;
+typedef signed int stbi__int32;
+#else
+#include <stdint.h>
+typedef uint16_t stbi__uint16;
+typedef int16_t stbi__int16;
+typedef uint32_t stbi__uint32;
+typedef int32_t stbi__int32;
+#endif
+
+// should produce compiler error if size is wrong
+typedef unsigned char validate_uint32[sizeof(stbi__uint32)==4 ? 1 : -1];
+
+#ifdef _MSC_VER
+#define STBI_NOTUSED(v) (void)(v)
+#else
+#define STBI_NOTUSED(v) (void)sizeof(v)
+#endif
+
+#ifdef _MSC_VER
+#define STBI_HAS_LROTL
+#endif
+
+#ifdef STBI_HAS_LROTL
+ #define stbi_lrot(x,y) _lrotl(x,y)
+#else
+ #define stbi_lrot(x,y) (((x) << (y)) | ((x) >> (32 - (y))))
+#endif
+
+#if defined(STBI_MALLOC) && defined(STBI_FREE) && (defined(STBI_REALLOC) || defined(STBI_REALLOC_SIZED))
+// ok
+#elif !defined(STBI_MALLOC) && !defined(STBI_FREE) && !defined(STBI_REALLOC) && !defined(STBI_REALLOC_SIZED)
+// ok
+#else
+#error "Must define all or none of STBI_MALLOC, STBI_FREE, and STBI_REALLOC (or STBI_REALLOC_SIZED)."
+#endif
+
+#ifndef STBI_MALLOC
+#define STBI_MALLOC(sz) malloc(sz)
+#define STBI_REALLOC(p,newsz) realloc(p,newsz)
+#define STBI_FREE(p) free(p)
+#endif
+
+#ifndef STBI_REALLOC_SIZED
+#define STBI_REALLOC_SIZED(p,oldsz,newsz) STBI_REALLOC(p,newsz)
+#endif
+
+// x86/x64 detection
+#if defined(__x86_64__) || defined(_M_X64)
+#define STBI__X64_TARGET
+#elif defined(__i386) || defined(_M_IX86)
+#define STBI__X86_TARGET
+#endif
+
+#if defined(__GNUC__) && defined(STBI__X86_TARGET) && !defined(__SSE2__) && !defined(STBI_NO_SIMD)
+// gcc doesn't support sse2 intrinsics unless you compile with -msse2,
+// which in turn means it gets to use SSE2 everywhere. This is unfortunate,
+// but previous attempts to provide the SSE2 functions with runtime
+// detection caused numerous issues. The way architecture extensions are
+// exposed in GCC/Clang is, sadly, not really suited for one-file libs.
+// New behavior: if compiled with -msse2, we use SSE2 without any
+// detection; if not, we don't use it at all.
+#define STBI_NO_SIMD
+#endif
+
+#if defined(__MINGW32__) && defined(STBI__X86_TARGET) && !defined(STBI_MINGW_ENABLE_SSE2) && !defined(STBI_NO_SIMD)
+// Note that __MINGW32__ doesn't actually mean 32-bit, so we have to avoid STBI__X64_TARGET
+//
+// 32-bit MinGW wants ESP to be 16-byte aligned, but this is not in the
+// Windows ABI and VC++ as well as Windows DLLs don't maintain that invariant.
+// As a result, enabling SSE2 on 32-bit MinGW is dangerous when not
+// simultaneously enabling "-mstackrealign".
+//
+// See https://github.com/nothings/stb/issues/81 for more information.
+//
+// So default to no SSE2 on 32-bit MinGW. If you've read this far and added
+// -mstackrealign to your build settings, feel free to #define STBI_MINGW_ENABLE_SSE2.
+#define STBI_NO_SIMD
+#endif
+
+#if !defined(STBI_NO_SIMD) && (defined(STBI__X86_TARGET) || defined(STBI__X64_TARGET))
+#define STBI_SSE2
+#include <emmintrin.h>
+
+#ifdef _MSC_VER
+
+#if _MSC_VER >= 1400 // not VC6
+#include <intrin.h> // __cpuid
+static int stbi__cpuid3(void)
+{
+ int info[4];
+ __cpuid(info,1);
+ return info[3];
+}
+#else
+static int stbi__cpuid3(void)
+{
+ int res;
+ __asm {
+ mov eax,1
+ cpuid
+ mov res,edx
+ }
+ return res;
+}
+#endif
+
+#define STBI_SIMD_ALIGN(type, name) __declspec(align(16)) type name
+
+static int stbi__sse2_available()
+{
+ int info3 = stbi__cpuid3();
+ return ((info3 >> 26) & 1) != 0;
+}
+#else // assume GCC-style if not VC++
+#define STBI_SIMD_ALIGN(type, name) type name __attribute__((aligned(16)))
+
+static int stbi__sse2_available()
+{
+ // If we're even attempting to compile this on GCC/Clang, that means
+ // -msse2 is on, which means the compiler is allowed to use SSE2
+ // instructions at will, and so are we.
+ return 1;
+}
+#endif
+#endif
+
+// ARM NEON
+#if defined(STBI_NO_SIMD) && defined(STBI_NEON)
+#undef STBI_NEON
+#endif
+
+#ifdef STBI_NEON
+#include <arm_neon.h>
+// assume GCC or Clang on ARM targets
+#define STBI_SIMD_ALIGN(type, name) type name __attribute__((aligned(16)))
+#endif
+
+#ifndef STBI_SIMD_ALIGN
+#define STBI_SIMD_ALIGN(type, name) type name
+#endif
+
+///////////////////////////////////////////////
+//
+// stbi__context struct and start_xxx functions
+
+// stbi__context structure is our basic context used by all images, so it
+// contains all the IO context, plus some basic image information
+typedef struct
+{
+ stbi__uint32 img_x, img_y;
+ int img_n, img_out_n;
+
+ stbi_io_callbacks io;
+ void *io_user_data;
+
+ int read_from_callbacks;
+ int buflen;
+ stbi_uc buffer_start[128];
+
+ stbi_uc *img_buffer, *img_buffer_end;
+ stbi_uc *img_buffer_original, *img_buffer_original_end;
+} stbi__context;
+
+
+static void stbi__refill_buffer(stbi__context *s);
+
+// initialize a memory-decode context
+static void stbi__start_mem(stbi__context *s, stbi_uc const *buffer, int len)
+{
+ s->io.read = NULL;
+ s->read_from_callbacks = 0;
+ s->img_buffer = s->img_buffer_original = (stbi_uc *) buffer;
+ s->img_buffer_end = s->img_buffer_original_end = (stbi_uc *) buffer+len;
+}
+
+// initialize a callback-based context
+static void stbi__start_callbacks(stbi__context *s, stbi_io_callbacks *c, void *user)
+{
+ s->io = *c;
+ s->io_user_data = user;
+ s->buflen = sizeof(s->buffer_start);
+ s->read_from_callbacks = 1;
+ s->img_buffer_original = s->buffer_start;
+ stbi__refill_buffer(s);
+ s->img_buffer_original_end = s->img_buffer_end;
+}
+
+#ifndef STBI_NO_STDIO
+
+static int stbi__stdio_read(void *user, char *data, int size)
+{
+ return (int) fread(data,1,size,(FILE*) user);
+}
+
+static void stbi__stdio_skip(void *user, int n)
+{
+ fseek((FILE*) user, n, SEEK_CUR);
+}
+
+static int stbi__stdio_eof(void *user)
+{
+ return feof((FILE*) user);
+}
+
+static stbi_io_callbacks stbi__stdio_callbacks =
+{
+ stbi__stdio_read,
+ stbi__stdio_skip,
+ stbi__stdio_eof,
+};
+
+static void stbi__start_file(stbi__context *s, FILE *f)
+{
+ stbi__start_callbacks(s, &stbi__stdio_callbacks, (void *) f);
+}
+
+//static void stop_file(stbi__context *s) { }
+
+#endif // !STBI_NO_STDIO
+
+static void stbi__rewind(stbi__context *s)
+{
+ // conceptually rewind SHOULD rewind to the beginning of the stream,
+ // but we just rewind to the beginning of the initial buffer, because
+ // we only use it after doing 'test', which only ever looks at at most 92 bytes
+ s->img_buffer = s->img_buffer_original;
+ s->img_buffer_end = s->img_buffer_original_end;
+}
+
+enum
+{
+ STBI_ORDER_RGB,
+ STBI_ORDER_BGR
+};
+
+typedef struct
+{
+ int bits_per_channel;
+ int num_channels;
+ int channel_order;
+} stbi__result_info;
+
+#ifndef STBI_NO_JPEG
+static int stbi__jpeg_test(stbi__context *s);
+static void *stbi__jpeg_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__jpeg_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_PNG
+static int stbi__png_test(stbi__context *s);
+static void *stbi__png_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__png_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_BMP
+static int stbi__bmp_test(stbi__context *s);
+static void *stbi__bmp_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__bmp_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_TGA
+static int stbi__tga_test(stbi__context *s);
+static void *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__tga_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_PSD
+static int stbi__psd_test(stbi__context *s);
+static void *stbi__psd_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri, int bpc);
+static int stbi__psd_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_HDR
+static int stbi__hdr_test(stbi__context *s);
+static float *stbi__hdr_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__hdr_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_PIC
+static int stbi__pic_test(stbi__context *s);
+static void *stbi__pic_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__pic_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_GIF
+static int stbi__gif_test(stbi__context *s);
+static void *stbi__gif_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__gif_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_PNM
+static int stbi__pnm_test(stbi__context *s);
+static void *stbi__pnm_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int stbi__pnm_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+// this is not threadsafe
+static const char *stbi__g_failure_reason;
+
+STBIDEF const char *stbi_failure_reason(void)
+{
+ return stbi__g_failure_reason;
+}
+
+static int stbi__err(const char *str)
+{
+ stbi__g_failure_reason = str;
+ return 0;
+}
+
+static void *stbi__malloc(size_t size)
+{
+ return STBI_MALLOC(size);
+}
+
+// stb_image uses ints pervasively, including for offset calculations.
+// therefore the largest decoded image size we can support with the
+// current code, even on 64-bit targets, is INT_MAX. this is not a
+// significant limitation for the intended use case.
+//
+// we do, however, need to make sure our size calculations don't
+// overflow. hence a few helper functions for size calculations that
+// multiply integers together, making sure that they're non-negative
+// and no overflow occurs.
+
+// return 1 if the sum is valid, 0 on overflow.
+// negative terms are considered invalid.
+static int stbi__addsizes_valid(int a, int b)
+{
+ if (b < 0) return 0;
+ // now 0 <= b <= INT_MAX, hence also
+ // 0 <= INT_MAX - b <= INTMAX.
+ // And "a + b <= INT_MAX" (which might overflow) is the
+ // same as a <= INT_MAX - b (no overflow)
+ return a <= INT_MAX - b;
+}
+
+// returns 1 if the product is valid, 0 on overflow.
+// negative factors are considered invalid.
+static int stbi__mul2sizes_valid(int a, int b)
+{
+ if (a < 0 || b < 0) return 0;
+ if (b == 0) return 1; // mul-by-0 is always safe
+ // portable way to check for no overflows in a*b
+ return a <= INT_MAX/b;
+}
+
+// returns 1 if "a*b + add" has no negative terms/factors and doesn't overflow
+static int stbi__mad2sizes_valid(int a, int b, int add)
+{
+ return stbi__mul2sizes_valid(a, b) && stbi__addsizes_valid(a*b, add);
+}
+
+// returns 1 if "a*b*c + add" has no negative terms/factors and doesn't overflow
+static int stbi__mad3sizes_valid(int a, int b, int c, int add)
+{
+ return stbi__mul2sizes_valid(a, b) && stbi__mul2sizes_valid(a*b, c) &&
+ stbi__addsizes_valid(a*b*c, add);
+}
+
+// returns 1 if "a*b*c*d + add" has no negative terms/factors and doesn't overflow
+static int stbi__mad4sizes_valid(int a, int b, int c, int d, int add)
+{
+ return stbi__mul2sizes_valid(a, b) && stbi__mul2sizes_valid(a*b, c) &&
+ stbi__mul2sizes_valid(a*b*c, d) && stbi__addsizes_valid(a*b*c*d, add);
+}
+
+// mallocs with size overflow checking
+static void *stbi__malloc_mad2(int a, int b, int add)
+{
+ if (!stbi__mad2sizes_valid(a, b, add)) return NULL;
+ return stbi__malloc(a*b + add);
+}
+
+static void *stbi__malloc_mad3(int a, int b, int c, int add)
+{
+ if (!stbi__mad3sizes_valid(a, b, c, add)) return NULL;
+ return stbi__malloc(a*b*c + add);
+}
+
+static void *stbi__malloc_mad4(int a, int b, int c, int d, int add)
+{
+ if (!stbi__mad4sizes_valid(a, b, c, d, add)) return NULL;
+ return stbi__malloc(a*b*c*d + add);
+}
+
+// stbi__err - error
+// stbi__errpf - error returning pointer to float
+// stbi__errpuc - error returning pointer to unsigned char
+
+#ifdef STBI_NO_FAILURE_STRINGS
+ #define stbi__err(x,y) 0
+#elif defined(STBI_FAILURE_USERMSG)
+ #define stbi__err(x,y) stbi__err(y)
+#else
+ #define stbi__err(x,y) stbi__err(x)
+#endif
+
+#define stbi__errpf(x,y) ((float *)(size_t) (stbi__err(x,y)?NULL:NULL))
+#define stbi__errpuc(x,y) ((unsigned char *)(size_t) (stbi__err(x,y)?NULL:NULL))
+
+STBIDEF void stbi_image_free(void *retval_from_stbi_load)
+{
+ STBI_FREE(retval_from_stbi_load);
+}
+
+#ifndef STBI_NO_LINEAR
+static float *stbi__ldr_to_hdr(stbi_uc *data, int x, int y, int comp);
+#endif
+
+#ifndef STBI_NO_HDR
+static stbi_uc *stbi__hdr_to_ldr(float *data, int x, int y, int comp);
+#endif
+
+static int stbi__vertically_flip_on_load = 0;
+
+STBIDEF void stbi_set_flip_vertically_on_load(int flag_true_if_should_flip)
+{
+ stbi__vertically_flip_on_load = flag_true_if_should_flip;
+}
+
+static void *stbi__load_main(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri, int bpc)
+{
+ memset(ri, 0, sizeof(*ri)); // make sure it's initialized if we add new fields
+ ri->bits_per_channel = 8; // default is 8 so most paths don't have to be changed
+ ri->channel_order = STBI_ORDER_RGB; // all current input & output are this, but this is here so we can add BGR order
+ ri->num_channels = 0;
+
+ #ifndef STBI_NO_JPEG
+ if (stbi__jpeg_test(s)) return stbi__jpeg_load(s,x,y,comp,req_comp, ri);
+ #endif
+ #ifndef STBI_NO_PNG
+ if (stbi__png_test(s)) return stbi__png_load(s,x,y,comp,req_comp, ri);
+ #endif
+ #ifndef STBI_NO_BMP
+ if (stbi__bmp_test(s)) return stbi__bmp_load(s,x,y,comp,req_comp, ri);
+ #endif
+ #ifndef STBI_NO_GIF
+ if (stbi__gif_test(s)) return stbi__gif_load(s,x,y,comp,req_comp, ri);
+ #endif
+ #ifndef STBI_NO_PSD
+ if (stbi__psd_test(s)) return stbi__psd_load(s,x,y,comp,req_comp, ri, bpc);
+ #endif
+ #ifndef STBI_NO_PIC
+ if (stbi__pic_test(s)) return stbi__pic_load(s,x,y,comp,req_comp, ri);
+ #endif
+ #ifndef STBI_NO_PNM
+ if (stbi__pnm_test(s)) return stbi__pnm_load(s,x,y,comp,req_comp, ri);
+ #endif
+
+ #ifndef STBI_NO_HDR
+ if (stbi__hdr_test(s)) {
+ float *hdr = stbi__hdr_load(s, x,y,comp,req_comp, ri);
+ return stbi__hdr_to_ldr(hdr, *x, *y, req_comp ? req_comp : *comp);
+ }
+ #endif
+
+ #ifndef STBI_NO_TGA
+ // test tga last because it's a crappy test!
+ if (stbi__tga_test(s))
+ return stbi__tga_load(s,x,y,comp,req_comp, ri);
+ #endif
+
+ return stbi__errpuc("unknown image type", "Image not of any known type, or corrupt");
+}
+
+static stbi_uc *stbi__convert_16_to_8(stbi__uint16 *orig, int w, int h, int channels)
+{
+ int i;
+ int img_len = w * h * channels;
+ stbi_uc *reduced;
+
+ reduced = (stbi_uc *) stbi__malloc(img_len);
+ if (reduced == NULL) return stbi__errpuc("outofmem", "Out of memory");
+
+ for (i = 0; i < img_len; ++i)
+ reduced[i] = (stbi_uc)((orig[i] >> 8) & 0xFF); // top half of each byte is sufficient approx of 16->8 bit scaling
+
+ STBI_FREE(orig);
+ return reduced;
+}
+
+static stbi__uint16 *stbi__convert_8_to_16(stbi_uc *orig, int w, int h, int channels)
+{
+ int i;
+ int img_len = w * h * channels;
+ stbi__uint16 *enlarged;
+
+ enlarged = (stbi__uint16 *) stbi__malloc(img_len*2);
+ if (enlarged == NULL) return (stbi__uint16 *) stbi__errpuc("outofmem", "Out of memory");
+
+ for (i = 0; i < img_len; ++i)
+ enlarged[i] = (stbi__uint16)((orig[i] << 8) + orig[i]); // replicate to high and low byte, maps 0->0, 255->0xffff
+
+ STBI_FREE(orig);
+ return enlarged;
+}
+
+static unsigned char *stbi__load_and_postprocess_8bit(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__result_info ri;
+ void *result = stbi__load_main(s, x, y, comp, req_comp, &ri, 8);
+
+ if (result == NULL)
+ return NULL;
+
+ if (ri.bits_per_channel != 8) {
+ STBI_ASSERT(ri.bits_per_channel == 16);
+ result = stbi__convert_16_to_8((stbi__uint16 *) result, *x, *y, req_comp == 0 ? *comp : req_comp);
+ ri.bits_per_channel = 8;
+ }
+
+ // @TODO: move stbi__convert_format to here
+
+ if (stbi__vertically_flip_on_load) {
+ int w = *x, h = *y;
+ int channels = req_comp ? req_comp : *comp;
+ int row,col,z;
+ stbi_uc *image = (stbi_uc *) result;
+
+ // @OPTIMIZE: use a bigger temp buffer and memcpy multiple pixels at once
+ for (row = 0; row < (h>>1); row++) {
+ for (col = 0; col < w; col++) {
+ for (z = 0; z < channels; z++) {
+ stbi_uc temp = image[(row * w + col) * channels + z];
+ image[(row * w + col) * channels + z] = image[((h - row - 1) * w + col) * channels + z];
+ image[((h - row - 1) * w + col) * channels + z] = temp;
+ }
+ }
+ }
+ }
+
+ return (unsigned char *) result;
+}
+
+static stbi__uint16 *stbi__load_and_postprocess_16bit(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__result_info ri;
+ void *result = stbi__load_main(s, x, y, comp, req_comp, &ri, 16);
+
+ if (result == NULL)
+ return NULL;
+
+ if (ri.bits_per_channel != 16) {
+ STBI_ASSERT(ri.bits_per_channel == 8);
+ result = stbi__convert_8_to_16((stbi_uc *) result, *x, *y, req_comp == 0 ? *comp : req_comp);
+ ri.bits_per_channel = 16;
+ }
+
+ // @TODO: move stbi__convert_format16 to here
+ // @TODO: special case RGB-to-Y (and RGBA-to-YA) for 8-bit-to-16-bit case to keep more precision
+
+ if (stbi__vertically_flip_on_load) {
+ int w = *x, h = *y;
+ int channels = req_comp ? req_comp : *comp;
+ int row,col,z;
+ stbi__uint16 *image = (stbi__uint16 *) result;
+
+ // @OPTIMIZE: use a bigger temp buffer and memcpy multiple pixels at once
+ for (row = 0; row < (h>>1); row++) {
+ for (col = 0; col < w; col++) {
+ for (z = 0; z < channels; z++) {
+ stbi__uint16 temp = image[(row * w + col) * channels + z];
+ image[(row * w + col) * channels + z] = image[((h - row - 1) * w + col) * channels + z];
+ image[((h - row - 1) * w + col) * channels + z] = temp;
+ }
+ }
+ }
+ }
+
+ return (stbi__uint16 *) result;
+}
+
+#ifndef STBI_NO_HDR
+static void stbi__float_postprocess(float *result, int *x, int *y, int *comp, int req_comp)
+{
+ if (stbi__vertically_flip_on_load && result != NULL) {
+ int w = *x, h = *y;
+ int depth = req_comp ? req_comp : *comp;
+ int row,col,z;
+ float temp;
+
+ // @OPTIMIZE: use a bigger temp buffer and memcpy multiple pixels at once
+ for (row = 0; row < (h>>1); row++) {
+ for (col = 0; col < w; col++) {
+ for (z = 0; z < depth; z++) {
+ temp = result[(row * w + col) * depth + z];
+ result[(row * w + col) * depth + z] = result[((h - row - 1) * w + col) * depth + z];
+ result[((h - row - 1) * w + col) * depth + z] = temp;
+ }
+ }
+ }
+ }
+}
+#endif
+
+#ifndef STBI_NO_STDIO
+
+static FILE *stbi__fopen(char const *filename, char const *mode)
+{
+ FILE *f;
+#if defined(_MSC_VER) && _MSC_VER >= 1400
+ if (0 != fopen_s(&f, filename, mode))
+ f=0;
+#else
+ f = fopen(filename, mode);
+#endif
+ return f;
+}
+
+
+STBIDEF stbi_uc *stbi_load(char const *filename, int *x, int *y, int *comp, int req_comp)
+{
+ FILE *f = stbi__fopen(filename, "rb");
+ unsigned char *result;
+ if (!f) return stbi__errpuc("can't fopen", "Unable to open file");
+ result = stbi_load_from_file(f,x,y,comp,req_comp);
+ fclose(f);
+ return result;
+}
+
+STBIDEF stbi_uc *stbi_load_from_file(FILE *f, int *x, int *y, int *comp, int req_comp)
+{
+ unsigned char *result;
+ stbi__context s;
+ stbi__start_file(&s,f);
+ result = stbi__load_and_postprocess_8bit(&s,x,y,comp,req_comp);
+ if (result) {
+ // need to 'unget' all the characters in the IO buffer
+ fseek(f, - (int) (s.img_buffer_end - s.img_buffer), SEEK_CUR);
+ }
+ return result;
+}
+
+STBIDEF stbi__uint16 *stbi_load_from_file_16(FILE *f, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__uint16 *result;
+ stbi__context s;
+ stbi__start_file(&s,f);
+ result = stbi__load_and_postprocess_16bit(&s,x,y,comp,req_comp);
+ if (result) {
+ // need to 'unget' all the characters in the IO buffer
+ fseek(f, - (int) (s.img_buffer_end - s.img_buffer), SEEK_CUR);
+ }
+ return result;
+}
+
+STBIDEF stbi_us *stbi_load_16(char const *filename, int *x, int *y, int *comp, int req_comp)
+{
+ FILE *f = stbi__fopen(filename, "rb");
+ stbi__uint16 *result;
+ if (!f) return (stbi_us *) stbi__errpuc("can't fopen", "Unable to open file");
+ result = stbi_load_from_file_16(f,x,y,comp,req_comp);
+ fclose(f);
+ return result;
+}
+
+
+#endif //!STBI_NO_STDIO
+
+STBIDEF stbi_uc *stbi_load_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__context s;
+ stbi__start_mem(&s,buffer,len);
+ return stbi__load_and_postprocess_8bit(&s,x,y,comp,req_comp);
+}
+
+STBIDEF stbi_uc *stbi_load_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__context s;
+ stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user);
+ return stbi__load_and_postprocess_8bit(&s,x,y,comp,req_comp);
+}
+
+#ifndef STBI_NO_LINEAR
+static float *stbi__loadf_main(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+{
+ unsigned char *data;
+ #ifndef STBI_NO_HDR
+ if (stbi__hdr_test(s)) {
+ stbi__result_info ri;
+ float *hdr_data = stbi__hdr_load(s,x,y,comp,req_comp, &ri);
+ if (hdr_data)
+ stbi__float_postprocess(hdr_data,x,y,comp,req_comp);
+ return hdr_data;
+ }
+ #endif
+ data = stbi__load_and_postprocess_8bit(s, x, y, comp, req_comp);
+ if (data)
+ return stbi__ldr_to_hdr(data, *x, *y, req_comp ? req_comp : *comp);
+ return stbi__errpf("unknown image type", "Image not of any known type, or corrupt");
+}
+
+STBIDEF float *stbi_loadf_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__context s;
+ stbi__start_mem(&s,buffer,len);
+ return stbi__loadf_main(&s,x,y,comp,req_comp);
+}
+
+STBIDEF float *stbi_loadf_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__context s;
+ stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user);
+ return stbi__loadf_main(&s,x,y,comp,req_comp);
+}
+
+#ifndef STBI_NO_STDIO
+STBIDEF float *stbi_loadf(char const *filename, int *x, int *y, int *comp, int req_comp)
+{
+ float *result;
+ FILE *f = stbi__fopen(filename, "rb");
+ if (!f) return stbi__errpf("can't fopen", "Unable to open file");
+ result = stbi_loadf_from_file(f,x,y,comp,req_comp);
+ fclose(f);
+ return result;
+}
+
+STBIDEF float *stbi_loadf_from_file(FILE *f, int *x, int *y, int *comp, int req_comp)
+{
+ stbi__context s;
+ stbi__start_file(&s,f);
+ return stbi__loadf_main(&s,x,y,comp,req_comp);
+}
+#endif // !STBI_NO_STDIO
+
+#endif // !STBI_NO_LINEAR
+
+// these is-hdr-or-not is defined independent of whether STBI_NO_LINEAR is
+// defined, for API simplicity; if STBI_NO_LINEAR is defined, it always
+// reports false!
+
+STBIDEF int stbi_is_hdr_from_memory(stbi_uc const *buffer, int len)
+{
+ #ifndef STBI_NO_HDR
+ stbi__context s;
+ stbi__start_mem(&s,buffer,len);
+ return stbi__hdr_test(&s);
+ #else
+ STBI_NOTUSED(buffer);
+ STBI_NOTUSED(len);
+ return 0;
+ #endif
+}
+
+#ifndef STBI_NO_STDIO
+STBIDEF int stbi_is_hdr (char const *filename)
+{
+ FILE *f = stbi__fopen(filename, "rb");
+ int result=0;
+ if (f) {
+ result = stbi_is_hdr_from_file(f);
+ fclose(f);
+ }
+ return result;
+}
+
+STBIDEF int stbi_is_hdr_from_file(FILE *f)
+{
+ #ifndef STBI_NO_HDR
+ stbi__context s;
+ stbi__start_file(&s,f);
+ return stbi__hdr_test(&s);
+ #else
+ STBI_NOTUSED(f);
+ return 0;
+ #endif
+}
+#endif // !STBI_NO_STDIO
+
+STBIDEF int stbi_is_hdr_from_callbacks(stbi_io_callbacks const *clbk, void *user)
+{
+ #ifndef STBI_NO_HDR
+ stbi__context s;
+ stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user);
+ return stbi__hdr_test(&s);
+ #else
+ STBI_NOTUSED(clbk);
+ STBI_NOTUSED(user);
+ return 0;
+ #endif
+}
+
+#ifndef STBI_NO_LINEAR
+static float stbi__l2h_gamma=2.2f, stbi__l2h_scale=1.0f;
+
+STBIDEF void stbi_ldr_to_hdr_gamma(float gamma) { stbi__l2h_gamma = gamma; }
+STBIDEF void stbi_ldr_to_hdr_scale(float scale) { stbi__l2h_scale = scale; }
+#endif
+
+static float stbi__h2l_gamma_i=1.0f/2.2f, stbi__h2l_scale_i=1.0f;
+
+STBIDEF void stbi_hdr_to_ldr_gamma(float gamma) { stbi__h2l_gamma_i = 1/gamma; }
+STBIDEF void stbi_hdr_to_ldr_scale(float scale) { stbi__h2l_scale_i = 1/scale; }
+
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// Common code used by all image loaders
+//
+
+enum
+{
+ STBI__SCAN_load=0,
+ STBI__SCAN_type,
+ STBI__SCAN_header
+};
+
+static void stbi__refill_buffer(stbi__context *s)
+{
+ int n = (s->io.read)(s->io_user_data,(char*)s->buffer_start,s->buflen);
+ if (n == 0) {
+ // at end of file, treat same as if from memory, but need to handle case
+ // where s->img_buffer isn't pointing to safe memory, e.g. 0-byte file
+ s->read_from_callbacks = 0;
+ s->img_buffer = s->buffer_start;
+ s->img_buffer_end = s->buffer_start+1;
+ *s->img_buffer = 0;
+ } else {
+ s->img_buffer = s->buffer_start;
+ s->img_buffer_end = s->buffer_start + n;
+ }
+}
+
+stbi_inline static stbi_uc stbi__get8(stbi__context *s)
+{
+ if (s->img_buffer < s->img_buffer_end)
+ return *s->img_buffer++;
+ if (s->read_from_callbacks) {
+ stbi__refill_buffer(s);
+ return *s->img_buffer++;
+ }
+ return 0;
+}
+
+stbi_inline static int stbi__at_eof(stbi__context *s)
+{
+ if (s->io.read) {
+ if (!(s->io.eof)(s->io_user_data)) return 0;
+ // if feof() is true, check if buffer = end
+ // special case: we've only got the special 0 character at the end
+ if (s->read_from_callbacks == 0) return 1;
+ }
+
+ return s->img_buffer >= s->img_buffer_end;
+}
+
+static void stbi__skip(stbi__context *s, int n)
+{
+ if (n < 0) {
+ s->img_buffer = s->img_buffer_end;
+ return;
+ }
+ if (s->io.read) {
+ int blen = (int) (s->img_buffer_end - s->img_buffer);
+ if (blen < n) {
+ s->img_buffer = s->img_buffer_end;
+ (s->io.skip)(s->io_user_data, n - blen);
+ return;
+ }
+ }
+ s->img_buffer += n;
+}
+
+static int stbi__getn(stbi__context *s, stbi_uc *buffer, int n)
+{
+ if (s->io.read) {
+ int blen = (int) (s->img_buffer_end - s->img_buffer);
+ if (blen < n) {
+ int res, count;
+
+ memcpy(buffer, s->img_buffer, blen);
+
+ count = (s->io.read)(s->io_user_data, (char*) buffer + blen, n - blen);
+ res = (count == (n-blen));
+ s->img_buffer = s->img_buffer_end;
+ return res;
+ }
+ }
+
+ if (s->img_buffer+n <= s->img_buffer_end) {
+ memcpy(buffer, s->img_buffer, n);
+ s->img_buffer += n;
+ return 1;
+ } else
+ return 0;
+}
+
+static int stbi__get16be(stbi__context *s)
+{
+ int z = stbi__get8(s);
+ return (z << 8) + stbi__get8(s);
+}
+
+static stbi__uint32 stbi__get32be(stbi__context *s)
+{
+ stbi__uint32 z = stbi__get16be(s);
+ return (z << 16) + stbi__get16be(s);
+}
+
+#if defined(STBI_NO_BMP) && defined(STBI_NO_TGA) && defined(STBI_NO_GIF)
+// nothing
+#else
+static int stbi__get16le(stbi__context *s)
+{
+ int z = stbi__get8(s);
+ return z + (stbi__get8(s) << 8);
+}
+#endif
+
+#ifndef STBI_NO_BMP
+static stbi__uint32 stbi__get32le(stbi__context *s)
+{
+ stbi__uint32 z = stbi__get16le(s);
+ return z + (stbi__get16le(s) << 16);
+}
+#endif
+
+#define STBI__BYTECAST(x) ((stbi_uc) ((x) & 255)) // truncate int to byte without warnings
+
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// generic converter from built-in img_n to req_comp
+// individual types do this automatically as much as possible (e.g. jpeg
+// does all cases internally since it needs to colorspace convert anyway,
+// and it never has alpha, so very few cases ). png can automatically
+// interleave an alpha=255 channel, but falls back to this for other cases
+//
+// assume data buffer is malloced, so malloc a new one and free that one
+// only failure mode is malloc failing
+
+static stbi_uc stbi__compute_y(int r, int g, int b)
+{
+ return (stbi_uc) (((r*77) + (g*150) + (29*b)) >> 8);
+}
+
+static unsigned char *stbi__convert_format(unsigned char *data, int img_n, int req_comp, unsigned int x, unsigned int y)
+{
+ int i,j;
+ unsigned char *good;
+
+ if (req_comp == img_n) return data;
+ STBI_ASSERT(req_comp >= 1 && req_comp <= 4);
+
+ good = (unsigned char *) stbi__malloc_mad3(req_comp, x, y, 0);
+ if (good == NULL) {
+ STBI_FREE(data);
+ return stbi__errpuc("outofmem", "Out of memory");
+ }
+
+ for (j=0; j < (int) y; ++j) {
+ unsigned char *src = data + j * x * img_n ;
+ unsigned char *dest = good + j * x * req_comp;
+
+ #define STBI__COMBO(a,b) ((a)*8+(b))
+ #define STBI__CASE(a,b) case STBI__COMBO(a,b): for(i=x-1; i >= 0; --i, src += a, dest += b)
+ // convert source image with img_n components to one with req_comp components;
+ // avoid switch per pixel, so use switch per scanline and massive macros
+ switch (STBI__COMBO(img_n, req_comp)) {
+ STBI__CASE(1,2) { dest[0]=src[0], dest[1]=255; } break;
+ STBI__CASE(1,3) { dest[0]=dest[1]=dest[2]=src[0]; } break;
+ STBI__CASE(1,4) { dest[0]=dest[1]=dest[2]=src[0], dest[3]=255; } break;
+ STBI__CASE(2,1) { dest[0]=src[0]; } break;
+ STBI__CASE(2,3) { dest[0]=dest[1]=dest[2]=src[0]; } break;
+ STBI__CASE(2,4) { dest[0]=dest[1]=dest[2]=src[0], dest[3]=src[1]; } break;
+ STBI__CASE(3,4) { dest[0]=src[0],dest[1]=src[1],dest[2]=src[2],dest[3]=255; } break;
+ STBI__CASE(3,1) { dest[0]=stbi__compute_y(src[0],src[1],src[2]); } break;
+ STBI__CASE(3,2) { dest[0]=stbi__compute_y(src[0],src[1],src[2]), dest[1] = 255; } break;
+ STBI__CASE(4,1) { dest[0]=stbi__compute_y(src[0],src[1],src[2]); } break;
+ STBI__CASE(4,2) { dest[0]=stbi__compute_y(src[0],src[1],src[2]), dest[1] = src[3]; } break;
+ STBI__CASE(4,3) { dest[0]=src[0],dest[1]=src[1],dest[2]=src[2]; } break;
+ default: STBI_ASSERT(0);
+ }
+ #undef STBI__CASE
+ }
+
+ STBI_FREE(data);
+ return good;
+}
+
+static stbi__uint16 stbi__compute_y_16(int r, int g, int b)
+{
+ return (stbi__uint16) (((r*77) + (g*150) + (29*b)) >> 8);
+}
+
+static stbi__uint16 *stbi__convert_format16(stbi__uint16 *data, int img_n, int req_comp, unsigned int x, unsigned int y)
+{
+ int i,j;
+ stbi__uint16 *good;
+
+ if (req_comp == img_n) return data;
+ STBI_ASSERT(req_comp >= 1 && req_comp <= 4);
+
+ good = (stbi__uint16 *) stbi__malloc(req_comp * x * y * 2);
+ if (good == NULL) {
+ STBI_FREE(data);
+ return (stbi__uint16 *) stbi__errpuc("outofmem", "Out of memory");
+ }
+
+ for (j=0; j < (int) y; ++j) {
+ stbi__uint16 *src = data + j * x * img_n ;
+ stbi__uint16 *dest = good + j * x * req_comp;
+
+ #define STBI__COMBO(a,b) ((a)*8+(b))
+ #define STBI__CASE(a,b) case STBI__COMBO(a,b): for(i=x-1; i >= 0; --i, src += a, dest += b)
+ // convert source image with img_n components to one with req_comp components;
+ // avoid switch per pixel, so use switch per scanline and massive macros
+ switch (STBI__COMBO(img_n, req_comp)) {
+ STBI__CASE(1,2) { dest[0]=src[0], dest[1]=0xffff; } break;
+ STBI__CASE(1,3) { dest[0]=dest[1]=dest[2]=src[0]; } break;
+ STBI__CASE(1,4) { dest[0]=dest[1]=dest[2]=src[0], dest[3]=0xffff; } break;
+ STBI__CASE(2,1) { dest[0]=src[0]; } break;
+ STBI__CASE(2,3) { dest[0]=dest[1]=dest[2]=src[0]; } break;
+ STBI__CASE(2,4) { dest[0]=dest[1]=dest[2]=src[0], dest[3]=src[1]; } break;
+ STBI__CASE(3,4) { dest[0]=src[0],dest[1]=src[1],dest[2]=src[2],dest[3]=0xffff; } break;
+ STBI__CASE(3,1) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]); } break;
+ STBI__CASE(3,2) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]), dest[1] = 0xffff; } break;
+ STBI__CASE(4,1) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]); } break;
+ STBI__CASE(4,2) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]), dest[1] = src[3]; } break;
+ STBI__CASE(4,3) { dest[0]=src[0],dest[1]=src[1],dest[2]=src[2]; } break;
+ default: STBI_ASSERT(0);
+ }
+ #undef STBI__CASE
+ }
+
+ STBI_FREE(data);
+ return good;
+}
+
+#ifndef STBI_NO_LINEAR
+static float *stbi__ldr_to_hdr(stbi_uc *data, int x, int y, int comp)
+{
+ int i,k,n;
+ float *output;
+ if (!data) return NULL;
+ output = (float *) stbi__malloc_mad4(x, y, comp, sizeof(float), 0);
+ if (output == NULL) { STBI_FREE(data); return stbi__errpf("outofmem", "Out of memory"); }
+ // compute number of non-alpha components
+ if (comp & 1) n = comp; else n = comp-1;
+ for (i=0; i < x*y; ++i) {
+ for (k=0; k < n; ++k) {
+ output[i*comp + k] = (float) (pow(data[i*comp+k]/255.0f, stbi__l2h_gamma) * stbi__l2h_scale);
+ }
+ if (k < comp) output[i*comp + k] = data[i*comp+k]/255.0f;
+ }
+ STBI_FREE(data);
+ return output;
+}
+#endif
+
+#ifndef STBI_NO_HDR
+#define stbi__float2int(x) ((int) (x))
+static stbi_uc *stbi__hdr_to_ldr(float *data, int x, int y, int comp)
+{
+ int i,k,n;
+ stbi_uc *output;
+ if (!data) return NULL;
+ output = (stbi_uc *) stbi__malloc_mad3(x, y, comp, 0);
+ if (output == NULL) { STBI_FREE(data); return stbi__errpuc("outofmem", "Out of memory"); }
+ // compute number of non-alpha components
+ if (comp & 1) n = comp; else n = comp-1;
+ for (i=0; i < x*y; ++i) {
+ for (k=0; k < n; ++k) {
+ float z = (float) pow(data[i*comp+k]*stbi__h2l_scale_i, stbi__h2l_gamma_i) * 255 + 0.5f;
+ if (z < 0) z = 0;
+ if (z > 255) z = 255;
+ output[i*comp + k] = (stbi_uc) stbi__float2int(z);
+ }
+ if (k < comp) {
+ float z = data[i*comp+k] * 255 + 0.5f;
+ if (z < 0) z = 0;
+ if (z > 255) z = 255;
+ output[i*comp + k] = (stbi_uc) stbi__float2int(z);
+ }
+ }
+ STBI_FREE(data);
+ return output;
+}
+#endif
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// "baseline" JPEG/JFIF decoder
+//
+// simple implementation
+// - doesn't support delayed output of y-dimension
+// - simple interface (only one output format: 8-bit interleaved RGB)
+// - doesn't try to recover corrupt jpegs
+// - doesn't allow partial loading, loading multiple at once
+// - still fast on x86 (copying globals into locals doesn't help x86)
+// - allocates lots of intermediate memory (full size of all components)
+// - non-interleaved case requires this anyway
+// - allows good upsampling (see next)
+// high-quality
+// - upsampled channels are bilinearly interpolated, even across blocks
+// - quality integer IDCT derived from IJG's 'slow'
+// performance
+// - fast huffman; reasonable integer IDCT
+// - some SIMD kernels for common paths on targets with SSE2/NEON
+// - uses a lot of intermediate memory, could cache poorly
+
+#ifndef STBI_NO_JPEG
+
+// huffman decoding acceleration
+#define FAST_BITS 9 // larger handles more cases; smaller stomps less cache
+
+typedef struct
+{
+ stbi_uc fast[1 << FAST_BITS];
+ // weirdly, repacking this into AoS is a 10% speed loss, instead of a win
+ stbi__uint16 code[256];
+ stbi_uc values[256];
+ stbi_uc size[257];
+ unsigned int maxcode[18];
+ int delta[17]; // old 'firstsymbol' - old 'firstcode'
+} stbi__huffman;
+
+typedef struct
+{
+ stbi__context *s;
+ stbi__huffman huff_dc[4];
+ stbi__huffman huff_ac[4];
+ stbi__uint16 dequant[4][64];
+ stbi__int16 fast_ac[4][1 << FAST_BITS];
+
+// sizes for components, interleaved MCUs
+ int img_h_max, img_v_max;
+ int img_mcu_x, img_mcu_y;
+ int img_mcu_w, img_mcu_h;
+
+// definition of jpeg image component
+ struct
+ {
+ int id;
+ int h,v;
+ int tq;
+ int hd,ha;
+ int dc_pred;
+
+ int x,y,w2,h2;
+ stbi_uc *data;
+ void *raw_data, *raw_coeff;
+ stbi_uc *linebuf;
+ short *coeff; // progressive only
+ int coeff_w, coeff_h; // number of 8x8 coefficient blocks
+ } img_comp[4];
+
+ stbi__uint32 code_buffer; // jpeg entropy-coded buffer
+ int code_bits; // number of valid bits
+ unsigned char marker; // marker seen while filling entropy buffer
+ int nomore; // flag if we saw a marker so must stop
+
+ int progressive;
+ int spec_start;
+ int spec_end;
+ int succ_high;
+ int succ_low;
+ int eob_run;
+ int jfif;
+ int app14_color_transform; // Adobe APP14 tag
+ int rgb;
+
+ int scan_n, order[4];
+ int restart_interval, todo;
+
+// kernels
+ void (*idct_block_kernel)(stbi_uc *out, int out_stride, short data[64]);
+ void (*YCbCr_to_RGB_kernel)(stbi_uc *out, const stbi_uc *y, const stbi_uc *pcb, const stbi_uc *pcr, int count, int step);
+ stbi_uc *(*resample_row_hv_2_kernel)(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs);
+} stbi__jpeg;
+
+static int stbi__build_huffman(stbi__huffman *h, int *count)
+{
+ int i,j,k=0,code;
+ // build size list for each symbol (from JPEG spec)
+ for (i=0; i < 16; ++i)
+ for (j=0; j < count[i]; ++j)
+ h->size[k++] = (stbi_uc) (i+1);
+ h->size[k] = 0;
+
+ // compute actual symbols (from jpeg spec)
+ code = 0;
+ k = 0;
+ for(j=1; j <= 16; ++j) {
+ // compute delta to add to code to compute symbol id
+ h->delta[j] = k - code;
+ if (h->size[k] == j) {
+ while (h->size[k] == j)
+ h->code[k++] = (stbi__uint16) (code++);
+ if (code-1 >= (1 << j)) return stbi__err("bad code lengths","Corrupt JPEG");
+ }
+ // compute largest code + 1 for this size, preshifted as needed later
+ h->maxcode[j] = code << (16-j);
+ code <<= 1;
+ }
+ h->maxcode[j] = 0xffffffff;
+
+ // build non-spec acceleration table; 255 is flag for not-accelerated
+ memset(h->fast, 255, 1 << FAST_BITS);
+ for (i=0; i < k; ++i) {
+ int s = h->size[i];
+ if (s <= FAST_BITS) {
+ int c = h->code[i] << (FAST_BITS-s);
+ int m = 1 << (FAST_BITS-s);
+ for (j=0; j < m; ++j) {
+ h->fast[c+j] = (stbi_uc) i;
+ }
+ }
+ }
+ return 1;
+}
+
+// build a table that decodes both magnitude and value of small ACs in
+// one go.
+static void stbi__build_fast_ac(stbi__int16 *fast_ac, stbi__huffman *h)
+{
+ int i;
+ for (i=0; i < (1 << FAST_BITS); ++i) {
+ stbi_uc fast = h->fast[i];
+ fast_ac[i] = 0;
+ if (fast < 255) {
+ int rs = h->values[fast];
+ int run = (rs >> 4) & 15;
+ int magbits = rs & 15;
+ int len = h->size[fast];
+
+ if (magbits && len + magbits <= FAST_BITS) {
+ // magnitude code followed by receive_extend code
+ int k = ((i << len) & ((1 << FAST_BITS) - 1)) >> (FAST_BITS - magbits);
+ int m = 1 << (magbits - 1);
+ if (k < m) k += (~0U << magbits) + 1;
+ // if the result is small enough, we can fit it in fast_ac table
+ if (k >= -128 && k <= 127)
+ fast_ac[i] = (stbi__int16) ((k << 8) + (run << 4) + (len + magbits));
+ }
+ }
+ }
+}
+
+static void stbi__grow_buffer_unsafe(stbi__jpeg *j)
+{
+ do {
+ int b = j->nomore ? 0 : stbi__get8(j->s);
+ if (b == 0xff) {
+ int c = stbi__get8(j->s);
+ while (c == 0xff) c = stbi__get8(j->s); // consume fill bytes
+ if (c != 0) {
+ j->marker = (unsigned char) c;
+ j->nomore = 1;
+ return;
+ }
+ }
+ j->code_buffer |= b << (24 - j->code_bits);
+ j->code_bits += 8;
+ } while (j->code_bits <= 24);
+}
+
+// (1 << n) - 1
+static stbi__uint32 stbi__bmask[17]={0,1,3,7,15,31,63,127,255,511,1023,2047,4095,8191,16383,32767,65535};
+
+// decode a jpeg huffman value from the bitstream
+stbi_inline static int stbi__jpeg_huff_decode(stbi__jpeg *j, stbi__huffman *h)
+{
+ unsigned int temp;
+ int c,k;
+
+ if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
+
+ // look at the top FAST_BITS and determine what symbol ID it is,
+ // if the code is <= FAST_BITS
+ c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1);
+ k = h->fast[c];
+ if (k < 255) {
+ int s = h->size[k];
+ if (s > j->code_bits)
+ return -1;
+ j->code_buffer <<= s;
+ j->code_bits -= s;
+ return h->values[k];
+ }
+
+ // naive test is to shift the code_buffer down so k bits are
+ // valid, then test against maxcode. To speed this up, we've
+ // preshifted maxcode left so that it has (16-k) 0s at the
+ // end; in other words, regardless of the number of bits, it
+ // wants to be compared against something shifted to have 16;
+ // that way we don't need to shift inside the loop.
+ temp = j->code_buffer >> 16;
+ for (k=FAST_BITS+1 ; ; ++k)
+ if (temp < h->maxcode[k])
+ break;
+ if (k == 17) {
+ // error! code not found
+ j->code_bits -= 16;
+ return -1;
+ }
+
+ if (k > j->code_bits)
+ return -1;
+
+ // convert the huffman code to the symbol id
+ c = ((j->code_buffer >> (32 - k)) & stbi__bmask[k]) + h->delta[k];
+ STBI_ASSERT((((j->code_buffer) >> (32 - h->size[c])) & stbi__bmask[h->size[c]]) == h->code[c]);
+
+ // convert the id to a symbol
+ j->code_bits -= k;
+ j->code_buffer <<= k;
+ return h->values[c];
+}
+
+// bias[n] = (-1<<n) + 1
+static int const stbi__jbias[16] = {0,-1,-3,-7,-15,-31,-63,-127,-255,-511,-1023,-2047,-4095,-8191,-16383,-32767};
+
+// combined JPEG 'receive' and JPEG 'extend', since baseline
+// always extends everything it receives.
+stbi_inline static int stbi__extend_receive(stbi__jpeg *j, int n)
+{
+ unsigned int k;
+ int sgn;
+ if (j->code_bits < n) stbi__grow_buffer_unsafe(j);
+
+ sgn = (stbi__int32)j->code_buffer >> 31; // sign bit is always in MSB
+ k = stbi_lrot(j->code_buffer, n);
+ STBI_ASSERT(n >= 0 && n < (int) (sizeof(stbi__bmask)/sizeof(*stbi__bmask)));
+ j->code_buffer = k & ~stbi__bmask[n];
+ k &= stbi__bmask[n];
+ j->code_bits -= n;
+ return k + (stbi__jbias[n] & ~sgn);
+}
+
+// get some unsigned bits
+stbi_inline static int stbi__jpeg_get_bits(stbi__jpeg *j, int n)
+{
+ unsigned int k;
+ if (j->code_bits < n) stbi__grow_buffer_unsafe(j);
+ k = stbi_lrot(j->code_buffer, n);
+ j->code_buffer = k & ~stbi__bmask[n];
+ k &= stbi__bmask[n];
+ j->code_bits -= n;
+ return k;
+}
+
+stbi_inline static int stbi__jpeg_get_bit(stbi__jpeg *j)
+{
+ unsigned int k;
+ if (j->code_bits < 1) stbi__grow_buffer_unsafe(j);
+ k = j->code_buffer;
+ j->code_buffer <<= 1;
+ --j->code_bits;
+ return k & 0x80000000;
+}
+
+// given a value that's at position X in the zigzag stream,
+// where does it appear in the 8x8 matrix coded as row-major?
+static stbi_uc stbi__jpeg_dezigzag[64+15] =
+{
+ 0, 1, 8, 16, 9, 2, 3, 10,
+ 17, 24, 32, 25, 18, 11, 4, 5,
+ 12, 19, 26, 33, 40, 48, 41, 34,
+ 27, 20, 13, 6, 7, 14, 21, 28,
+ 35, 42, 49, 56, 57, 50, 43, 36,
+ 29, 22, 15, 23, 30, 37, 44, 51,
+ 58, 59, 52, 45, 38, 31, 39, 46,
+ 53, 60, 61, 54, 47, 55, 62, 63,
+ // let corrupt input sample past end
+ 63, 63, 63, 63, 63, 63, 63, 63,
+ 63, 63, 63, 63, 63, 63, 63
+};
+
+// decode one 64-entry block--
+static int stbi__jpeg_decode_block(stbi__jpeg *j, short data[64], stbi__huffman *hdc, stbi__huffman *hac, stbi__int16 *fac, int b, stbi__uint16 *dequant)
+{
+ int diff,dc,k;
+ int t;
+
+ if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
+ t = stbi__jpeg_huff_decode(j, hdc);
+ if (t < 0) return stbi__err("bad huffman code","Corrupt JPEG");
+
+ // 0 all the ac values now so we can do it 32-bits at a time
+ memset(data,0,64*sizeof(data[0]));
+
+ diff = t ? stbi__extend_receive(j, t) : 0;
+ dc = j->img_comp[b].dc_pred + diff;
+ j->img_comp[b].dc_pred = dc;
+ data[0] = (short) (dc * dequant[0]);
+
+ // decode AC components, see JPEG spec
+ k = 1;
+ do {
+ unsigned int zig;
+ int c,r,s;
+ if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
+ c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1);
+ r = fac[c];
+ if (r) { // fast-AC path
+ k += (r >> 4) & 15; // run
+ s = r & 15; // combined length
+ j->code_buffer <<= s;
+ j->code_bits -= s;
+ // decode into unzigzag'd location
+ zig = stbi__jpeg_dezigzag[k++];
+ data[zig] = (short) ((r >> 8) * dequant[zig]);
+ } else {
+ int rs = stbi__jpeg_huff_decode(j, hac);
+ if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG");
+ s = rs & 15;
+ r = rs >> 4;
+ if (s == 0) {
+ if (rs != 0xf0) break; // end block
+ k += 16;
+ } else {
+ k += r;
+ // decode into unzigzag'd location
+ zig = stbi__jpeg_dezigzag[k++];
+ data[zig] = (short) (stbi__extend_receive(j,s) * dequant[zig]);
+ }
+ }
+ } while (k < 64);
+ return 1;
+}
+
+static int stbi__jpeg_decode_block_prog_dc(stbi__jpeg *j, short data[64], stbi__huffman *hdc, int b)
+{
+ int diff,dc;
+ int t;
+ if (j->spec_end != 0) return stbi__err("can't merge dc and ac", "Corrupt JPEG");
+
+ if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
+
+ if (j->succ_high == 0) {
+ // first scan for DC coefficient, must be first
+ memset(data,0,64*sizeof(data[0])); // 0 all the ac values now
+ t = stbi__jpeg_huff_decode(j, hdc);
+ diff = t ? stbi__extend_receive(j, t) : 0;
+
+ dc = j->img_comp[b].dc_pred + diff;
+ j->img_comp[b].dc_pred = dc;
+ data[0] = (short) (dc << j->succ_low);
+ } else {
+ // refinement scan for DC coefficient
+ if (stbi__jpeg_get_bit(j))
+ data[0] += (short) (1 << j->succ_low);
+ }
+ return 1;
+}
+
+// @OPTIMIZE: store non-zigzagged during the decode passes,
+// and only de-zigzag when dequantizing
+static int stbi__jpeg_decode_block_prog_ac(stbi__jpeg *j, short data[64], stbi__huffman *hac, stbi__int16 *fac)
+{
+ int k;
+ if (j->spec_start == 0) return stbi__err("can't merge dc and ac", "Corrupt JPEG");
+
+ if (j->succ_high == 0) {
+ int shift = j->succ_low;
+
+ if (j->eob_run) {
+ --j->eob_run;
+ return 1;
+ }
+
+ k = j->spec_start;
+ do {
+ unsigned int zig;
+ int c,r,s;
+ if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
+ c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1);
+ r = fac[c];
+ if (r) { // fast-AC path
+ k += (r >> 4) & 15; // run
+ s = r & 15; // combined length
+ j->code_buffer <<= s;
+ j->code_bits -= s;
+ zig = stbi__jpeg_dezigzag[k++];
+ data[zig] = (short) ((r >> 8) << shift);
+ } else {
+ int rs = stbi__jpeg_huff_decode(j, hac);
+ if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG");
+ s = rs & 15;
+ r = rs >> 4;
+ if (s == 0) {
+ if (r < 15) {
+ j->eob_run = (1 << r);
+ if (r)
+ j->eob_run += stbi__jpeg_get_bits(j, r);
+ --j->eob_run;
+ break;
+ }
+ k += 16;
+ } else {
+ k += r;
+ zig = stbi__jpeg_dezigzag[k++];
+ data[zig] = (short) (stbi__extend_receive(j,s) << shift);
+ }
+ }
+ } while (k <= j->spec_end);
+ } else {
+ // refinement scan for these AC coefficients
+
+ short bit = (short) (1 << j->succ_low);
+
+ if (j->eob_run) {
+ --j->eob_run;
+ for (k = j->spec_start; k <= j->spec_end; ++k) {
+ short *p = &data[stbi__jpeg_dezigzag[k]];
+ if (*p != 0)
+ if (stbi__jpeg_get_bit(j))
+ if ((*p & bit)==0) {
+ if (*p > 0)
+ *p += bit;
+ else
+ *p -= bit;
+ }
+ }
+ } else {
+ k = j->spec_start;
+ do {
+ int r,s;
+ int rs = stbi__jpeg_huff_decode(j, hac); // @OPTIMIZE see if we can use the fast path here, advance-by-r is so slow, eh
+ if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG");
+ s = rs & 15;
+ r = rs >> 4;
+ if (s == 0) {
+ if (r < 15) {
+ j->eob_run = (1 << r) - 1;
+ if (r)
+ j->eob_run += stbi__jpeg_get_bits(j, r);
+ r = 64; // force end of block
+ } else {
+ // r=15 s=0 should write 16 0s, so we just do
+ // a run of 15 0s and then write s (which is 0),
+ // so we don't have to do anything special here
+ }
+ } else {
+ if (s != 1) return stbi__err("bad huffman code", "Corrupt JPEG");
+ // sign bit
+ if (stbi__jpeg_get_bit(j))
+ s = bit;
+ else
+ s = -bit;
+ }
+
+ // advance by r
+ while (k <= j->spec_end) {
+ short *p = &data[stbi__jpeg_dezigzag[k++]];
+ if (*p != 0) {
+ if (stbi__jpeg_get_bit(j))
+ if ((*p & bit)==0) {
+ if (*p > 0)
+ *p += bit;
+ else
+ *p -= bit;
+ }
+ } else {
+ if (r == 0) {
+ *p = (short) s;
+ break;
+ }
+ --r;
+ }
+ }
+ } while (k <= j->spec_end);
+ }
+ }
+ return 1;
+}
+
+// take a -128..127 value and stbi__clamp it and convert to 0..255
+stbi_inline static stbi_uc stbi__clamp(int x)
+{
+ // trick to use a single test to catch both cases
+ if ((unsigned int) x > 255) {
+ if (x < 0) return 0;
+ if (x > 255) return 255;
+ }
+ return (stbi_uc) x;
+}
+
+#define stbi__f2f(x) ((int) (((x) * 4096 + 0.5)))
+#define stbi__fsh(x) ((x) << 12)
+
+// derived from jidctint -- DCT_ISLOW
+#define STBI__IDCT_1D(s0,s1,s2,s3,s4,s5,s6,s7) \
+ int t0,t1,t2,t3,p1,p2,p3,p4,p5,x0,x1,x2,x3; \
+ p2 = s2; \
+ p3 = s6; \
+ p1 = (p2+p3) * stbi__f2f(0.5411961f); \
+ t2 = p1 + p3*stbi__f2f(-1.847759065f); \
+ t3 = p1 + p2*stbi__f2f( 0.765366865f); \
+ p2 = s0; \
+ p3 = s4; \
+ t0 = stbi__fsh(p2+p3); \
+ t1 = stbi__fsh(p2-p3); \
+ x0 = t0+t3; \
+ x3 = t0-t3; \
+ x1 = t1+t2; \
+ x2 = t1-t2; \
+ t0 = s7; \
+ t1 = s5; \
+ t2 = s3; \
+ t3 = s1; \
+ p3 = t0+t2; \
+ p4 = t1+t3; \
+ p1 = t0+t3; \
+ p2 = t1+t2; \
+ p5 = (p3+p4)*stbi__f2f( 1.175875602f); \
+ t0 = t0*stbi__f2f( 0.298631336f); \
+ t1 = t1*stbi__f2f( 2.053119869f); \
+ t2 = t2*stbi__f2f( 3.072711026f); \
+ t3 = t3*stbi__f2f( 1.501321110f); \
+ p1 = p5 + p1*stbi__f2f(-0.899976223f); \
+ p2 = p5 + p2*stbi__f2f(-2.562915447f); \
+ p3 = p3*stbi__f2f(-1.961570560f); \
+ p4 = p4*stbi__f2f(-0.390180644f); \
+ t3 += p1+p4; \
+ t2 += p2+p3; \
+ t1 += p2+p4; \
+ t0 += p1+p3;
+
+static void stbi__idct_block(stbi_uc *out, int out_stride, short data[64])
+{
+ int i,val[64],*v=val;
+ stbi_uc *o;
+ short *d = data;
+
+ // columns
+ for (i=0; i < 8; ++i,++d, ++v) {
+ // if all zeroes, shortcut -- this avoids dequantizing 0s and IDCTing
+ if (d[ 8]==0 && d[16]==0 && d[24]==0 && d[32]==0
+ && d[40]==0 && d[48]==0 && d[56]==0) {
+ // no shortcut 0 seconds
+ // (1|2|3|4|5|6|7)==0 0 seconds
+ // all separate -0.047 seconds
+ // 1 && 2|3 && 4|5 && 6|7: -0.047 seconds
+ int dcterm = d[0] << 2;
+ v[0] = v[8] = v[16] = v[24] = v[32] = v[40] = v[48] = v[56] = dcterm;
+ } else {
+ STBI__IDCT_1D(d[ 0],d[ 8],d[16],d[24],d[32],d[40],d[48],d[56])
+ // constants scaled things up by 1<<12; let's bring them back
+ // down, but keep 2 extra bits of precision
+ x0 += 512; x1 += 512; x2 += 512; x3 += 512;
+ v[ 0] = (x0+t3) >> 10;
+ v[56] = (x0-t3) >> 10;
+ v[ 8] = (x1+t2) >> 10;
+ v[48] = (x1-t2) >> 10;
+ v[16] = (x2+t1) >> 10;
+ v[40] = (x2-t1) >> 10;
+ v[24] = (x3+t0) >> 10;
+ v[32] = (x3-t0) >> 10;
+ }
+ }
+
+ for (i=0, v=val, o=out; i < 8; ++i,v+=8,o+=out_stride) {
+ // no fast case since the first 1D IDCT spread components out
+ STBI__IDCT_1D(v[0],v[1],v[2],v[3],v[4],v[5],v[6],v[7])
+ // constants scaled things up by 1<<12, plus we had 1<<2 from first
+ // loop, plus horizontal and vertical each scale by sqrt(8) so together
+ // we've got an extra 1<<3, so 1<<17 total we need to remove.
+ // so we want to round that, which means adding 0.5 * 1<<17,
+ // aka 65536. Also, we'll end up with -128 to 127 that we want
+ // to encode as 0..255 by adding 128, so we'll add that before the shift
+ x0 += 65536 + (128<<17);
+ x1 += 65536 + (128<<17);
+ x2 += 65536 + (128<<17);
+ x3 += 65536 + (128<<17);
+ // tried computing the shifts into temps, or'ing the temps to see
+ // if any were out of range, but that was slower
+ o[0] = stbi__clamp((x0+t3) >> 17);
+ o[7] = stbi__clamp((x0-t3) >> 17);
+ o[1] = stbi__clamp((x1+t2) >> 17);
+ o[6] = stbi__clamp((x1-t2) >> 17);
+ o[2] = stbi__clamp((x2+t1) >> 17);
+ o[5] = stbi__clamp((x2-t1) >> 17);
+ o[3] = stbi__clamp((x3+t0) >> 17);
+ o[4] = stbi__clamp((x3-t0) >> 17);
+ }
+}
+
+#ifdef STBI_SSE2
+// sse2 integer IDCT. not the fastest possible implementation but it
+// produces bit-identical results to the generic C version so it's
+// fully "transparent".
+static void stbi__idct_simd(stbi_uc *out, int out_stride, short data[64])
+{
+ // This is constructed to match our regular (generic) integer IDCT exactly.
+ __m128i row0, row1, row2, row3, row4, row5, row6, row7;
+ __m128i tmp;
+
+ // dot product constant: even elems=x, odd elems=y
+ #define dct_const(x,y) _mm_setr_epi16((x),(y),(x),(y),(x),(y),(x),(y))
+
+ // out(0) = c0[even]*x + c0[odd]*y (c0, x, y 16-bit, out 32-bit)
+ // out(1) = c1[even]*x + c1[odd]*y
+ #define dct_rot(out0,out1, x,y,c0,c1) \
+ __m128i c0##lo = _mm_unpacklo_epi16((x),(y)); \
+ __m128i c0##hi = _mm_unpackhi_epi16((x),(y)); \
+ __m128i out0##_l = _mm_madd_epi16(c0##lo, c0); \
+ __m128i out0##_h = _mm_madd_epi16(c0##hi, c0); \
+ __m128i out1##_l = _mm_madd_epi16(c0##lo, c1); \
+ __m128i out1##_h = _mm_madd_epi16(c0##hi, c1)
+
+ // out = in << 12 (in 16-bit, out 32-bit)
+ #define dct_widen(out, in) \
+ __m128i out##_l = _mm_srai_epi32(_mm_unpacklo_epi16(_mm_setzero_si128(), (in)), 4); \
+ __m128i out##_h = _mm_srai_epi32(_mm_unpackhi_epi16(_mm_setzero_si128(), (in)), 4)
+
+ // wide add
+ #define dct_wadd(out, a, b) \
+ __m128i out##_l = _mm_add_epi32(a##_l, b##_l); \
+ __m128i out##_h = _mm_add_epi32(a##_h, b##_h)
+
+ // wide sub
+ #define dct_wsub(out, a, b) \
+ __m128i out##_l = _mm_sub_epi32(a##_l, b##_l); \
+ __m128i out##_h = _mm_sub_epi32(a##_h, b##_h)
+
+ // butterfly a/b, add bias, then shift by "s" and pack
+ #define dct_bfly32o(out0, out1, a,b,bias,s) \
+ { \
+ __m128i abiased_l = _mm_add_epi32(a##_l, bias); \
+ __m128i abiased_h = _mm_add_epi32(a##_h, bias); \
+ dct_wadd(sum, abiased, b); \
+ dct_wsub(dif, abiased, b); \
+ out0 = _mm_packs_epi32(_mm_srai_epi32(sum_l, s), _mm_srai_epi32(sum_h, s)); \
+ out1 = _mm_packs_epi32(_mm_srai_epi32(dif_l, s), _mm_srai_epi32(dif_h, s)); \
+ }
+
+ // 8-bit interleave step (for transposes)
+ #define dct_interleave8(a, b) \
+ tmp = a; \
+ a = _mm_unpacklo_epi8(a, b); \
+ b = _mm_unpackhi_epi8(tmp, b)
+
+ // 16-bit interleave step (for transposes)
+ #define dct_interleave16(a, b) \
+ tmp = a; \
+ a = _mm_unpacklo_epi16(a, b); \
+ b = _mm_unpackhi_epi16(tmp, b)
+
+ #define dct_pass(bias,shift) \
+ { \
+ /* even part */ \
+ dct_rot(t2e,t3e, row2,row6, rot0_0,rot0_1); \
+ __m128i sum04 = _mm_add_epi16(row0, row4); \
+ __m128i dif04 = _mm_sub_epi16(row0, row4); \
+ dct_widen(t0e, sum04); \
+ dct_widen(t1e, dif04); \
+ dct_wadd(x0, t0e, t3e); \
+ dct_wsub(x3, t0e, t3e); \
+ dct_wadd(x1, t1e, t2e); \
+ dct_wsub(x2, t1e, t2e); \
+ /* odd part */ \
+ dct_rot(y0o,y2o, row7,row3, rot2_0,rot2_1); \
+ dct_rot(y1o,y3o, row5,row1, rot3_0,rot3_1); \
+ __m128i sum17 = _mm_add_epi16(row1, row7); \
+ __m128i sum35 = _mm_add_epi16(row3, row5); \
+ dct_rot(y4o,y5o, sum17,sum35, rot1_0,rot1_1); \
+ dct_wadd(x4, y0o, y4o); \
+ dct_wadd(x5, y1o, y5o); \
+ dct_wadd(x6, y2o, y5o); \
+ dct_wadd(x7, y3o, y4o); \
+ dct_bfly32o(row0,row7, x0,x7,bias,shift); \
+ dct_bfly32o(row1,row6, x1,x6,bias,shift); \
+ dct_bfly32o(row2,row5, x2,x5,bias,shift); \
+ dct_bfly32o(row3,row4, x3,x4,bias,shift); \
+ }
+
+ __m128i rot0_0 = dct_const(stbi__f2f(0.5411961f), stbi__f2f(0.5411961f) + stbi__f2f(-1.847759065f));
+ __m128i rot0_1 = dct_const(stbi__f2f(0.5411961f) + stbi__f2f( 0.765366865f), stbi__f2f(0.5411961f));
+ __m128i rot1_0 = dct_const(stbi__f2f(1.175875602f) + stbi__f2f(-0.899976223f), stbi__f2f(1.175875602f));
+ __m128i rot1_1 = dct_const(stbi__f2f(1.175875602f), stbi__f2f(1.175875602f) + stbi__f2f(-2.562915447f));
+ __m128i rot2_0 = dct_const(stbi__f2f(-1.961570560f) + stbi__f2f( 0.298631336f), stbi__f2f(-1.961570560f));
+ __m128i rot2_1 = dct_const(stbi__f2f(-1.961570560f), stbi__f2f(-1.961570560f) + stbi__f2f( 3.072711026f));
+ __m128i rot3_0 = dct_const(stbi__f2f(-0.390180644f) + stbi__f2f( 2.053119869f), stbi__f2f(-0.390180644f));
+ __m128i rot3_1 = dct_const(stbi__f2f(-0.390180644f), stbi__f2f(-0.390180644f) + stbi__f2f( 1.501321110f));
+
+ // rounding biases in column/row passes, see stbi__idct_block for explanation.
+ __m128i bias_0 = _mm_set1_epi32(512);
+ __m128i bias_1 = _mm_set1_epi32(65536 + (128<<17));
+
+ // load
+ row0 = _mm_load_si128((const __m128i *) (data + 0*8));
+ row1 = _mm_load_si128((const __m128i *) (data + 1*8));
+ row2 = _mm_load_si128((const __m128i *) (data + 2*8));
+ row3 = _mm_load_si128((const __m128i *) (data + 3*8));
+ row4 = _mm_load_si128((const __m128i *) (data + 4*8));
+ row5 = _mm_load_si128((const __m128i *) (data + 5*8));
+ row6 = _mm_load_si128((const __m128i *) (data + 6*8));
+ row7 = _mm_load_si128((const __m128i *) (data + 7*8));
+
+ // column pass
+ dct_pass(bias_0, 10);
+
+ {
+ // 16bit 8x8 transpose pass 1
+ dct_interleave16(row0, row4);
+ dct_interleave16(row1, row5);
+ dct_interleave16(row2, row6);
+ dct_interleave16(row3, row7);
+
+ // transpose pass 2
+ dct_interleave16(row0, row2);
+ dct_interleave16(row1, row3);
+ dct_interleave16(row4, row6);
+ dct_interleave16(row5, row7);
+
+ // transpose pass 3
+ dct_interleave16(row0, row1);
+ dct_interleave16(row2, row3);
+ dct_interleave16(row4, row5);
+ dct_interleave16(row6, row7);
+ }
+
+ // row pass
+ dct_pass(bias_1, 17);
+
+ {
+ // pack
+ __m128i p0 = _mm_packus_epi16(row0, row1); // a0a1a2a3...a7b0b1b2b3...b7
+ __m128i p1 = _mm_packus_epi16(row2, row3);
+ __m128i p2 = _mm_packus_epi16(row4, row5);
+ __m128i p3 = _mm_packus_epi16(row6, row7);
+
+ // 8bit 8x8 transpose pass 1
+ dct_interleave8(p0, p2); // a0e0a1e1...
+ dct_interleave8(p1, p3); // c0g0c1g1...
+
+ // transpose pass 2
+ dct_interleave8(p0, p1); // a0c0e0g0...
+ dct_interleave8(p2, p3); // b0d0f0h0...
+
+ // transpose pass 3
+ dct_interleave8(p0, p2); // a0b0c0d0...
+ dct_interleave8(p1, p3); // a4b4c4d4...
+
+ // store
+ _mm_storel_epi64((__m128i *) out, p0); out += out_stride;
+ _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p0, 0x4e)); out += out_stride;
+ _mm_storel_epi64((__m128i *) out, p2); out += out_stride;
+ _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p2, 0x4e)); out += out_stride;
+ _mm_storel_epi64((__m128i *) out, p1); out += out_stride;
+ _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p1, 0x4e)); out += out_stride;
+ _mm_storel_epi64((__m128i *) out, p3); out += out_stride;
+ _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p3, 0x4e));
+ }
+
+#undef dct_const
+#undef dct_rot
+#undef dct_widen
+#undef dct_wadd
+#undef dct_wsub
+#undef dct_bfly32o
+#undef dct_interleave8
+#undef dct_interleave16
+#undef dct_pass
+}
+
+#endif // STBI_SSE2
+
+#ifdef STBI_NEON
+
+// NEON integer IDCT. should produce bit-identical
+// results to the generic C version.
+static void stbi__idct_simd(stbi_uc *out, int out_stride, short data[64])
+{
+ int16x8_t row0, row1, row2, row3, row4, row5, row6, row7;
+
+ int16x4_t rot0_0 = vdup_n_s16(stbi__f2f(0.5411961f));
+ int16x4_t rot0_1 = vdup_n_s16(stbi__f2f(-1.847759065f));
+ int16x4_t rot0_2 = vdup_n_s16(stbi__f2f( 0.765366865f));
+ int16x4_t rot1_0 = vdup_n_s16(stbi__f2f( 1.175875602f));
+ int16x4_t rot1_1 = vdup_n_s16(stbi__f2f(-0.899976223f));
+ int16x4_t rot1_2 = vdup_n_s16(stbi__f2f(-2.562915447f));
+ int16x4_t rot2_0 = vdup_n_s16(stbi__f2f(-1.961570560f));
+ int16x4_t rot2_1 = vdup_n_s16(stbi__f2f(-0.390180644f));
+ int16x4_t rot3_0 = vdup_n_s16(stbi__f2f( 0.298631336f));
+ int16x4_t rot3_1 = vdup_n_s16(stbi__f2f( 2.053119869f));
+ int16x4_t rot3_2 = vdup_n_s16(stbi__f2f( 3.072711026f));
+ int16x4_t rot3_3 = vdup_n_s16(stbi__f2f( 1.501321110f));
+
+#define dct_long_mul(out, inq, coeff) \
+ int32x4_t out##_l = vmull_s16(vget_low_s16(inq), coeff); \
+ int32x4_t out##_h = vmull_s16(vget_high_s16(inq), coeff)
+
+#define dct_long_mac(out, acc, inq, coeff) \
+ int32x4_t out##_l = vmlal_s16(acc##_l, vget_low_s16(inq), coeff); \
+ int32x4_t out##_h = vmlal_s16(acc##_h, vget_high_s16(inq), coeff)
+
+#define dct_widen(out, inq) \
+ int32x4_t out##_l = vshll_n_s16(vget_low_s16(inq), 12); \
+ int32x4_t out##_h = vshll_n_s16(vget_high_s16(inq), 12)
+
+// wide add
+#define dct_wadd(out, a, b) \
+ int32x4_t out##_l = vaddq_s32(a##_l, b##_l); \
+ int32x4_t out##_h = vaddq_s32(a##_h, b##_h)
+
+// wide sub
+#define dct_wsub(out, a, b) \
+ int32x4_t out##_l = vsubq_s32(a##_l, b##_l); \
+ int32x4_t out##_h = vsubq_s32(a##_h, b##_h)
+
+// butterfly a/b, then shift using "shiftop" by "s" and pack
+#define dct_bfly32o(out0,out1, a,b,shiftop,s) \
+ { \
+ dct_wadd(sum, a, b); \
+ dct_wsub(dif, a, b); \
+ out0 = vcombine_s16(shiftop(sum_l, s), shiftop(sum_h, s)); \
+ out1 = vcombine_s16(shiftop(dif_l, s), shiftop(dif_h, s)); \
+ }
+
+#define dct_pass(shiftop, shift) \
+ { \
+ /* even part */ \
+ int16x8_t sum26 = vaddq_s16(row2, row6); \
+ dct_long_mul(p1e, sum26, rot0_0); \
+ dct_long_mac(t2e, p1e, row6, rot0_1); \
+ dct_long_mac(t3e, p1e, row2, rot0_2); \
+ int16x8_t sum04 = vaddq_s16(row0, row4); \
+ int16x8_t dif04 = vsubq_s16(row0, row4); \
+ dct_widen(t0e, sum04); \
+ dct_widen(t1e, dif04); \
+ dct_wadd(x0, t0e, t3e); \
+ dct_wsub(x3, t0e, t3e); \
+ dct_wadd(x1, t1e, t2e); \
+ dct_wsub(x2, t1e, t2e); \
+ /* odd part */ \
+ int16x8_t sum15 = vaddq_s16(row1, row5); \
+ int16x8_t sum17 = vaddq_s16(row1, row7); \
+ int16x8_t sum35 = vaddq_s16(row3, row5); \
+ int16x8_t sum37 = vaddq_s16(row3, row7); \
+ int16x8_t sumodd = vaddq_s16(sum17, sum35); \
+ dct_long_mul(p5o, sumodd, rot1_0); \
+ dct_long_mac(p1o, p5o, sum17, rot1_1); \
+ dct_long_mac(p2o, p5o, sum35, rot1_2); \
+ dct_long_mul(p3o, sum37, rot2_0); \
+ dct_long_mul(p4o, sum15, rot2_1); \
+ dct_wadd(sump13o, p1o, p3o); \
+ dct_wadd(sump24o, p2o, p4o); \
+ dct_wadd(sump23o, p2o, p3o); \
+ dct_wadd(sump14o, p1o, p4o); \
+ dct_long_mac(x4, sump13o, row7, rot3_0); \
+ dct_long_mac(x5, sump24o, row5, rot3_1); \
+ dct_long_mac(x6, sump23o, row3, rot3_2); \
+ dct_long_mac(x7, sump14o, row1, rot3_3); \
+ dct_bfly32o(row0,row7, x0,x7,shiftop,shift); \
+ dct_bfly32o(row1,row6, x1,x6,shiftop,shift); \
+ dct_bfly32o(row2,row5, x2,x5,shiftop,shift); \
+ dct_bfly32o(row3,row4, x3,x4,shiftop,shift); \
+ }
+
+ // load
+ row0 = vld1q_s16(data + 0*8);
+ row1 = vld1q_s16(data + 1*8);
+ row2 = vld1q_s16(data + 2*8);
+ row3 = vld1q_s16(data + 3*8);
+ row4 = vld1q_s16(data + 4*8);
+ row5 = vld1q_s16(data + 5*8);
+ row6 = vld1q_s16(data + 6*8);
+ row7 = vld1q_s16(data + 7*8);
+
+ // add DC bias
+ row0 = vaddq_s16(row0, vsetq_lane_s16(1024, vdupq_n_s16(0), 0));
+
+ // column pass
+ dct_pass(vrshrn_n_s32, 10);
+
+ // 16bit 8x8 transpose
+ {
+// these three map to a single VTRN.16, VTRN.32, and VSWP, respectively.
+// whether compilers actually get this is another story, sadly.
+#define dct_trn16(x, y) { int16x8x2_t t = vtrnq_s16(x, y); x = t.val[0]; y = t.val[1]; }
+#define dct_trn32(x, y) { int32x4x2_t t = vtrnq_s32(vreinterpretq_s32_s16(x), vreinterpretq_s32_s16(y)); x = vreinterpretq_s16_s32(t.val[0]); y = vreinterpretq_s16_s32(t.val[1]); }
+#define dct_trn64(x, y) { int16x8_t x0 = x; int16x8_t y0 = y; x = vcombine_s16(vget_low_s16(x0), vget_low_s16(y0)); y = vcombine_s16(vget_high_s16(x0), vget_high_s16(y0)); }
+
+ // pass 1
+ dct_trn16(row0, row1); // a0b0a2b2a4b4a6b6
+ dct_trn16(row2, row3);
+ dct_trn16(row4, row5);
+ dct_trn16(row6, row7);
+
+ // pass 2
+ dct_trn32(row0, row2); // a0b0c0d0a4b4c4d4
+ dct_trn32(row1, row3);
+ dct_trn32(row4, row6);
+ dct_trn32(row5, row7);
+
+ // pass 3
+ dct_trn64(row0, row4); // a0b0c0d0e0f0g0h0
+ dct_trn64(row1, row5);
+ dct_trn64(row2, row6);
+ dct_trn64(row3, row7);
+
+#undef dct_trn16
+#undef dct_trn32
+#undef dct_trn64
+ }
+
+ // row pass
+ // vrshrn_n_s32 only supports shifts up to 16, we need
+ // 17. so do a non-rounding shift of 16 first then follow
+ // up with a rounding shift by 1.
+ dct_pass(vshrn_n_s32, 16);
+
+ {
+ // pack and round
+ uint8x8_t p0 = vqrshrun_n_s16(row0, 1);
+ uint8x8_t p1 = vqrshrun_n_s16(row1, 1);
+ uint8x8_t p2 = vqrshrun_n_s16(row2, 1);
+ uint8x8_t p3 = vqrshrun_n_s16(row3, 1);
+ uint8x8_t p4 = vqrshrun_n_s16(row4, 1);
+ uint8x8_t p5 = vqrshrun_n_s16(row5, 1);
+ uint8x8_t p6 = vqrshrun_n_s16(row6, 1);
+ uint8x8_t p7 = vqrshrun_n_s16(row7, 1);
+
+ // again, these can translate into one instruction, but often don't.
+#define dct_trn8_8(x, y) { uint8x8x2_t t = vtrn_u8(x, y); x = t.val[0]; y = t.val[1]; }
+#define dct_trn8_16(x, y) { uint16x4x2_t t = vtrn_u16(vreinterpret_u16_u8(x), vreinterpret_u16_u8(y)); x = vreinterpret_u8_u16(t.val[0]); y = vreinterpret_u8_u16(t.val[1]); }
+#define dct_trn8_32(x, y) { uint32x2x2_t t = vtrn_u32(vreinterpret_u32_u8(x), vreinterpret_u32_u8(y)); x = vreinterpret_u8_u32(t.val[0]); y = vreinterpret_u8_u32(t.val[1]); }
+
+ // sadly can't use interleaved stores here since we only write
+ // 8 bytes to each scan line!
+
+ // 8x8 8-bit transpose pass 1
+ dct_trn8_8(p0, p1);
+ dct_trn8_8(p2, p3);
+ dct_trn8_8(p4, p5);
+ dct_trn8_8(p6, p7);
+
+ // pass 2
+ dct_trn8_16(p0, p2);
+ dct_trn8_16(p1, p3);
+ dct_trn8_16(p4, p6);
+ dct_trn8_16(p5, p7);
+
+ // pass 3
+ dct_trn8_32(p0, p4);
+ dct_trn8_32(p1, p5);
+ dct_trn8_32(p2, p6);
+ dct_trn8_32(p3, p7);
+
+ // store
+ vst1_u8(out, p0); out += out_stride;
+ vst1_u8(out, p1); out += out_stride;
+ vst1_u8(out, p2); out += out_stride;
+ vst1_u8(out, p3); out += out_stride;
+ vst1_u8(out, p4); out += out_stride;
+ vst1_u8(out, p5); out += out_stride;
+ vst1_u8(out, p6); out += out_stride;
+ vst1_u8(out, p7);
+
+#undef dct_trn8_8
+#undef dct_trn8_16
+#undef dct_trn8_32
+ }
+
+#undef dct_long_mul
+#undef dct_long_mac
+#undef dct_widen
+#undef dct_wadd
+#undef dct_wsub
+#undef dct_bfly32o
+#undef dct_pass
+}
+
+#endif // STBI_NEON
+
+#define STBI__MARKER_none 0xff
+// if there's a pending marker from the entropy stream, return that
+// otherwise, fetch from the stream and get a marker. if there's no
+// marker, return 0xff, which is never a valid marker value
+static stbi_uc stbi__get_marker(stbi__jpeg *j)
+{
+ stbi_uc x;
+ if (j->marker != STBI__MARKER_none) { x = j->marker; j->marker = STBI__MARKER_none; return x; }
+ x = stbi__get8(j->s);
+ if (x != 0xff) return STBI__MARKER_none;
+ while (x == 0xff)
+ x = stbi__get8(j->s); // consume repeated 0xff fill bytes
+ return x;
+}
+
+// in each scan, we'll have scan_n components, and the order
+// of the components is specified by order[]
+#define STBI__RESTART(x) ((x) >= 0xd0 && (x) <= 0xd7)
+
+// after a restart interval, stbi__jpeg_reset the entropy decoder and
+// the dc prediction
+static void stbi__jpeg_reset(stbi__jpeg *j)
+{
+ j->code_bits = 0;
+ j->code_buffer = 0;
+ j->nomore = 0;
+ j->img_comp[0].dc_pred = j->img_comp[1].dc_pred = j->img_comp[2].dc_pred = j->img_comp[3].dc_pred = 0;
+ j->marker = STBI__MARKER_none;
+ j->todo = j->restart_interval ? j->restart_interval : 0x7fffffff;
+ j->eob_run = 0;
+ // no more than 1<<31 MCUs if no restart_interal? that's plenty safe,
+ // since we don't even allow 1<<30 pixels
+}
+
+static int stbi__parse_entropy_coded_data(stbi__jpeg *z)
+{
+ stbi__jpeg_reset(z);
+ if (!z->progressive) {
+ if (z->scan_n == 1) {
+ int i,j;
+ STBI_SIMD_ALIGN(short, data[64]);
+ int n = z->order[0];
+ // non-interleaved data, we just need to process one block at a time,
+ // in trivial scanline order
+ // number of blocks to do just depends on how many actual "pixels" this
+ // component has, independent of interleaved MCU blocking and such
+ int w = (z->img_comp[n].x+7) >> 3;
+ int h = (z->img_comp[n].y+7) >> 3;
+ for (j=0; j < h; ++j) {
+ for (i=0; i < w; ++i) {
+ int ha = z->img_comp[n].ha;
+ if (!stbi__jpeg_decode_block(z, data, z->huff_dc+z->img_comp[n].hd, z->huff_ac+ha, z->fast_ac[ha], n, z->dequant[z->img_comp[n].tq])) return 0;
+ z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*j*8+i*8, z->img_comp[n].w2, data);
+ // every data block is an MCU, so countdown the restart interval
+ if (--z->todo <= 0) {
+ if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
+ // if it's NOT a restart, then just bail, so we get corrupt data
+ // rather than no data
+ if (!STBI__RESTART(z->marker)) return 1;
+ stbi__jpeg_reset(z);
+ }
+ }
+ }
+ return 1;
+ } else { // interleaved
+ int i,j,k,x,y;
+ STBI_SIMD_ALIGN(short, data[64]);
+ for (j=0; j < z->img_mcu_y; ++j) {
+ for (i=0; i < z->img_mcu_x; ++i) {
+ // scan an interleaved mcu... process scan_n components in order
+ for (k=0; k < z->scan_n; ++k) {
+ int n = z->order[k];
+ // scan out an mcu's worth of this component; that's just determined
+ // by the basic H and V specified for the component
+ for (y=0; y < z->img_comp[n].v; ++y) {
+ for (x=0; x < z->img_comp[n].h; ++x) {
+ int x2 = (i*z->img_comp[n].h + x)*8;
+ int y2 = (j*z->img_comp[n].v + y)*8;
+ int ha = z->img_comp[n].ha;
+ if (!stbi__jpeg_decode_block(z, data, z->huff_dc+z->img_comp[n].hd, z->huff_ac+ha, z->fast_ac[ha], n, z->dequant[z->img_comp[n].tq])) return 0;
+ z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*y2+x2, z->img_comp[n].w2, data);
+ }
+ }
+ }
+ // after all interleaved components, that's an interleaved MCU,
+ // so now count down the restart interval
+ if (--z->todo <= 0) {
+ if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
+ if (!STBI__RESTART(z->marker)) return 1;
+ stbi__jpeg_reset(z);
+ }
+ }
+ }
+ return 1;
+ }
+ } else {
+ if (z->scan_n == 1) {
+ int i,j;
+ int n = z->order[0];
+ // non-interleaved data, we just need to process one block at a time,
+ // in trivial scanline order
+ // number of blocks to do just depends on how many actual "pixels" this
+ // component has, independent of interleaved MCU blocking and such
+ int w = (z->img_comp[n].x+7) >> 3;
+ int h = (z->img_comp[n].y+7) >> 3;
+ for (j=0; j < h; ++j) {
+ for (i=0; i < w; ++i) {
+ short *data = z->img_comp[n].coeff + 64 * (i + j * z->img_comp[n].coeff_w);
+ if (z->spec_start == 0) {
+ if (!stbi__jpeg_decode_block_prog_dc(z, data, &z->huff_dc[z->img_comp[n].hd], n))
+ return 0;
+ } else {
+ int ha = z->img_comp[n].ha;
+ if (!stbi__jpeg_decode_block_prog_ac(z, data, &z->huff_ac[ha], z->fast_ac[ha]))
+ return 0;
+ }
+ // every data block is an MCU, so countdown the restart interval
+ if (--z->todo <= 0) {
+ if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
+ if (!STBI__RESTART(z->marker)) return 1;
+ stbi__jpeg_reset(z);
+ }
+ }
+ }
+ return 1;
+ } else { // interleaved
+ int i,j,k,x,y;
+ for (j=0; j < z->img_mcu_y; ++j) {
+ for (i=0; i < z->img_mcu_x; ++i) {
+ // scan an interleaved mcu... process scan_n components in order
+ for (k=0; k < z->scan_n; ++k) {
+ int n = z->order[k];
+ // scan out an mcu's worth of this component; that's just determined
+ // by the basic H and V specified for the component
+ for (y=0; y < z->img_comp[n].v; ++y) {
+ for (x=0; x < z->img_comp[n].h; ++x) {
+ int x2 = (i*z->img_comp[n].h + x);
+ int y2 = (j*z->img_comp[n].v + y);
+ short *data = z->img_comp[n].coeff + 64 * (x2 + y2 * z->img_comp[n].coeff_w);
+ if (!stbi__jpeg_decode_block_prog_dc(z, data, &z->huff_dc[z->img_comp[n].hd], n))
+ return 0;
+ }
+ }
+ }
+ // after all interleaved components, that's an interleaved MCU,
+ // so now count down the restart interval
+ if (--z->todo <= 0) {
+ if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
+ if (!STBI__RESTART(z->marker)) return 1;
+ stbi__jpeg_reset(z);
+ }
+ }
+ }
+ return 1;
+ }
+ }
+}
+
+static void stbi__jpeg_dequantize(short *data, stbi__uint16 *dequant)
+{
+ int i;
+ for (i=0; i < 64; ++i)
+ data[i] *= dequant[i];
+}
+
+static void stbi__jpeg_finish(stbi__jpeg *z)
+{
+ if (z->progressive) {
+ // dequantize and idct the data
+ int i,j,n;
+ for (n=0; n < z->s->img_n; ++n) {
+ int w = (z->img_comp[n].x+7) >> 3;
+ int h = (z->img_comp[n].y+7) >> 3;
+ for (j=0; j < h; ++j) {
+ for (i=0; i < w; ++i) {
+ short *data = z->img_comp[n].coeff + 64 * (i + j * z->img_comp[n].coeff_w);
+ stbi__jpeg_dequantize(data, z->dequant[z->img_comp[n].tq]);
+ z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*j*8+i*8, z->img_comp[n].w2, data);
+ }
+ }
+ }
+ }
+}
+
+static int stbi__process_marker(stbi__jpeg *z, int m)
+{
+ int L;
+ switch (m) {
+ case STBI__MARKER_none: // no marker found
+ return stbi__err("expected marker","Corrupt JPEG");
+
+ case 0xDD: // DRI - specify restart interval
+ if (stbi__get16be(z->s) != 4) return stbi__err("bad DRI len","Corrupt JPEG");
+ z->restart_interval = stbi__get16be(z->s);
+ return 1;
+
+ case 0xDB: // DQT - define quantization table
+ L = stbi__get16be(z->s)-2;
+ while (L > 0) {
+ int q = stbi__get8(z->s);
+ int p = q >> 4, sixteen = (p != 0);
+ int t = q & 15,i;
+ if (p != 0 && p != 1) return stbi__err("bad DQT type","Corrupt JPEG");
+ if (t > 3) return stbi__err("bad DQT table","Corrupt JPEG");
+
+ for (i=0; i < 64; ++i)
+ z->dequant[t][stbi__jpeg_dezigzag[i]] = sixteen ? stbi__get16be(z->s) : stbi__get8(z->s);
+ L -= (sixteen ? 129 : 65);
+ }
+ return L==0;
+
+ case 0xC4: // DHT - define huffman table
+ L = stbi__get16be(z->s)-2;
+ while (L > 0) {
+ stbi_uc *v;
+ int sizes[16],i,n=0;
+ int q = stbi__get8(z->s);
+ int tc = q >> 4;
+ int th = q & 15;
+ if (tc > 1 || th > 3) return stbi__err("bad DHT header","Corrupt JPEG");
+ for (i=0; i < 16; ++i) {
+ sizes[i] = stbi__get8(z->s);
+ n += sizes[i];
+ }
+ L -= 17;
+ if (tc == 0) {
+ if (!stbi__build_huffman(z->huff_dc+th, sizes)) return 0;
+ v = z->huff_dc[th].values;
+ } else {
+ if (!stbi__build_huffman(z->huff_ac+th, sizes)) return 0;
+ v = z->huff_ac[th].values;
+ }
+ for (i=0; i < n; ++i)
+ v[i] = stbi__get8(z->s);
+ if (tc != 0)
+ stbi__build_fast_ac(z->fast_ac[th], z->huff_ac + th);
+ L -= n;
+ }
+ return L==0;
+ }
+
+ // check for comment block or APP blocks
+ if ((m >= 0xE0 && m <= 0xEF) || m == 0xFE) {
+ L = stbi__get16be(z->s);
+ if (L < 2) {
+ if (m == 0xFE)
+ return stbi__err("bad COM len","Corrupt JPEG");
+ else
+ return stbi__err("bad APP len","Corrupt JPEG");
+ }
+ L -= 2;
+
+ if (m == 0xE0 && L >= 5) { // JFIF APP0 segment
+ static const unsigned char tag[5] = {'J','F','I','F','\0'};
+ int ok = 1;
+ int i;
+ for (i=0; i < 5; ++i)
+ if (stbi__get8(z->s) != tag[i])
+ ok = 0;
+ L -= 5;
+ if (ok)
+ z->jfif = 1;
+ } else if (m == 0xEE && L >= 12) { // Adobe APP14 segment
+ static const unsigned char tag[6] = {'A','d','o','b','e','\0'};
+ int ok = 1;
+ int i;
+ for (i=0; i < 6; ++i)
+ if (stbi__get8(z->s) != tag[i])
+ ok = 0;
+ L -= 6;
+ if (ok) {
+ stbi__get8(z->s); // version
+ stbi__get16be(z->s); // flags0
+ stbi__get16be(z->s); // flags1
+ z->app14_color_transform = stbi__get8(z->s); // color transform
+ L -= 6;
+ }
+ }
+
+ stbi__skip(z->s, L);
+ return 1;
+ }
+
+ return stbi__err("unknown marker","Corrupt JPEG");
+}
+
+// after we see SOS
+static int stbi__process_scan_header(stbi__jpeg *z)
+{
+ int i;
+ int Ls = stbi__get16be(z->s);
+ z->scan_n = stbi__get8(z->s);
+ if (z->scan_n < 1 || z->scan_n > 4 || z->scan_n > (int) z->s->img_n) return stbi__err("bad SOS component count","Corrupt JPEG");
+ if (Ls != 6+2*z->scan_n) return stbi__err("bad SOS len","Corrupt JPEG");
+ for (i=0; i < z->scan_n; ++i) {
+ int id = stbi__get8(z->s), which;
+ int q = stbi__get8(z->s);
+ for (which = 0; which < z->s->img_n; ++which)
+ if (z->img_comp[which].id == id)
+ break;
+ if (which == z->s->img_n) return 0; // no match
+ z->img_comp[which].hd = q >> 4; if (z->img_comp[which].hd > 3) return stbi__err("bad DC huff","Corrupt JPEG");
+ z->img_comp[which].ha = q & 15; if (z->img_comp[which].ha > 3) return stbi__err("bad AC huff","Corrupt JPEG");
+ z->order[i] = which;
+ }
+
+ {
+ int aa;
+ z->spec_start = stbi__get8(z->s);
+ z->spec_end = stbi__get8(z->s); // should be 63, but might be 0
+ aa = stbi__get8(z->s);
+ z->succ_high = (aa >> 4);
+ z->succ_low = (aa & 15);
+ if (z->progressive) {
+ if (z->spec_start > 63 || z->spec_end > 63 || z->spec_start > z->spec_end || z->succ_high > 13 || z->succ_low > 13)
+ return stbi__err("bad SOS", "Corrupt JPEG");
+ } else {
+ if (z->spec_start != 0) return stbi__err("bad SOS","Corrupt JPEG");
+ if (z->succ_high != 0 || z->succ_low != 0) return stbi__err("bad SOS","Corrupt JPEG");
+ z->spec_end = 63;
+ }
+ }
+
+ return 1;
+}
+
+static int stbi__free_jpeg_components(stbi__jpeg *z, int ncomp, int why)
+{
+ int i;
+ for (i=0; i < ncomp; ++i) {
+ if (z->img_comp[i].raw_data) {
+ STBI_FREE(z->img_comp[i].raw_data);
+ z->img_comp[i].raw_data = NULL;
+ z->img_comp[i].data = NULL;
+ }
+ if (z->img_comp[i].raw_coeff) {
+ STBI_FREE(z->img_comp[i].raw_coeff);
+ z->img_comp[i].raw_coeff = 0;
+ z->img_comp[i].coeff = 0;
+ }
+ if (z->img_comp[i].linebuf) {
+ STBI_FREE(z->img_comp[i].linebuf);
+ z->img_comp[i].linebuf = NULL;
+ }
+ }
+ return why;
+}
+
+static int stbi__process_frame_header(stbi__jpeg *z, int scan)
+{
+ stbi__context *s = z->s;
+ int Lf,p,i,q, h_max=1,v_max=1,c;
+ Lf = stbi__get16be(s); if (Lf < 11) return stbi__err("bad SOF len","Corrupt JPEG"); // JPEG
+ p = stbi__get8(s); if (p != 8) return stbi__err("only 8-bit","JPEG format not supported: 8-bit only"); // JPEG baseline
+ s->img_y = stbi__get16be(s); if (s->img_y == 0) return stbi__err("no header height", "JPEG format not supported: delayed height"); // Legal, but we don't handle it--but neither does IJG
+ s->img_x = stbi__get16be(s); if (s->img_x == 0) return stbi__err("0 width","Corrupt JPEG"); // JPEG requires
+ c = stbi__get8(s);
+ if (c != 3 && c != 1 && c != 4) return stbi__err("bad component count","Corrupt JPEG");
+ s->img_n = c;
+ for (i=0; i < c; ++i) {
+ z->img_comp[i].data = NULL;
+ z->img_comp[i].linebuf = NULL;
+ }
+
+ if (Lf != 8+3*s->img_n) return stbi__err("bad SOF len","Corrupt JPEG");
+
+ z->rgb = 0;
+ for (i=0; i < s->img_n; ++i) {
+ static unsigned char rgb[3] = { 'R', 'G', 'B' };
+ z->img_comp[i].id = stbi__get8(s);
+ if (s->img_n == 3 && z->img_comp[i].id == rgb[i])
+ ++z->rgb;
+ q = stbi__get8(s);
+ z->img_comp[i].h = (q >> 4); if (!z->img_comp[i].h || z->img_comp[i].h > 4) return stbi__err("bad H","Corrupt JPEG");
+ z->img_comp[i].v = q & 15; if (!z->img_comp[i].v || z->img_comp[i].v > 4) return stbi__err("bad V","Corrupt JPEG");
+ z->img_comp[i].tq = stbi__get8(s); if (z->img_comp[i].tq > 3) return stbi__err("bad TQ","Corrupt JPEG");
+ }
+
+ if (scan != STBI__SCAN_load) return 1;
+
+ if (!stbi__mad3sizes_valid(s->img_x, s->img_y, s->img_n, 0)) return stbi__err("too large", "Image too large to decode");
+
+ for (i=0; i < s->img_n; ++i) {
+ if (z->img_comp[i].h > h_max) h_max = z->img_comp[i].h;
+ if (z->img_comp[i].v > v_max) v_max = z->img_comp[i].v;
+ }
+
+ // compute interleaved mcu info
+ z->img_h_max = h_max;
+ z->img_v_max = v_max;
+ z->img_mcu_w = h_max * 8;
+ z->img_mcu_h = v_max * 8;
+ // these sizes can't be more than 17 bits
+ z->img_mcu_x = (s->img_x + z->img_mcu_w-1) / z->img_mcu_w;
+ z->img_mcu_y = (s->img_y + z->img_mcu_h-1) / z->img_mcu_h;
+
+ for (i=0; i < s->img_n; ++i) {
+ // number of effective pixels (e.g. for non-interleaved MCU)
+ z->img_comp[i].x = (s->img_x * z->img_comp[i].h + h_max-1) / h_max;
+ z->img_comp[i].y = (s->img_y * z->img_comp[i].v + v_max-1) / v_max;
+ // to simplify generation, we'll allocate enough memory to decode
+ // the bogus oversized data from using interleaved MCUs and their
+ // big blocks (e.g. a 16x16 iMCU on an image of width 33); we won't
+ // discard the extra data until colorspace conversion
+ //
+ // img_mcu_x, img_mcu_y: <=17 bits; comp[i].h and .v are <=4 (checked earlier)
+ // so these muls can't overflow with 32-bit ints (which we require)
+ z->img_comp[i].w2 = z->img_mcu_x * z->img_comp[i].h * 8;
+ z->img_comp[i].h2 = z->img_mcu_y * z->img_comp[i].v * 8;
+ z->img_comp[i].coeff = 0;
+ z->img_comp[i].raw_coeff = 0;
+ z->img_comp[i].linebuf = NULL;
+ z->img_comp[i].raw_data = stbi__malloc_mad2(z->img_comp[i].w2, z->img_comp[i].h2, 15);
+ if (z->img_comp[i].raw_data == NULL)
+ return stbi__free_jpeg_components(z, i+1, stbi__err("outofmem", "Out of memory"));
+ // align blocks for idct using mmx/sse
+ z->img_comp[i].data = (stbi_uc*) (((size_t) z->img_comp[i].raw_data + 15) & ~15);
+ if (z->progressive) {
+ // w2, h2 are multiples of 8 (see above)
+ z->img_comp[i].coeff_w = z->img_comp[i].w2 / 8;
+ z->img_comp[i].coeff_h = z->img_comp[i].h2 / 8;
+ z->img_comp[i].raw_coeff = stbi__malloc_mad3(z->img_comp[i].w2, z->img_comp[i].h2, sizeof(short), 15);
+ if (z->img_comp[i].raw_coeff == NULL)
+ return stbi__free_jpeg_components(z, i+1, stbi__err("outofmem", "Out of memory"));
+ z->img_comp[i].coeff = (short*) (((size_t) z->img_comp[i].raw_coeff + 15) & ~15);
+ }
+ }
+
+ return 1;
+}
+
+// use comparisons since in some cases we handle more than one case (e.g. SOF)
+#define stbi__DNL(x) ((x) == 0xdc)
+#define stbi__SOI(x) ((x) == 0xd8)
+#define stbi__EOI(x) ((x) == 0xd9)
+#define stbi__SOF(x) ((x) == 0xc0 || (x) == 0xc1 || (x) == 0xc2)
+#define stbi__SOS(x) ((x) == 0xda)
+
+#define stbi__SOF_progressive(x) ((x) == 0xc2)
+
+static int stbi__decode_jpeg_header(stbi__jpeg *z, int scan)
+{
+ int m;
+ z->jfif = 0;
+ z->app14_color_transform = -1; // valid values are 0,1,2
+ z->marker = STBI__MARKER_none; // initialize cached marker to empty
+ m = stbi__get_marker(z);
+ if (!stbi__SOI(m)) return stbi__err("no SOI","Corrupt JPEG");
+ if (scan == STBI__SCAN_type) return 1;
+ m = stbi__get_marker(z);
+ while (!stbi__SOF(m)) {
+ if (!stbi__process_marker(z,m)) return 0;
+ m = stbi__get_marker(z);
+ while (m == STBI__MARKER_none) {
+ // some files have extra padding after their blocks, so ok, we'll scan
+ if (stbi__at_eof(z->s)) return stbi__err("no SOF", "Corrupt JPEG");
+ m = stbi__get_marker(z);
+ }
+ }
+ z->progressive = stbi__SOF_progressive(m);
+ if (!stbi__process_frame_header(z, scan)) return 0;
+ return 1;
+}
+
+// decode image to YCbCr format
+static int stbi__decode_jpeg_image(stbi__jpeg *j)
+{
+ int m;
+ for (m = 0; m < 4; m++) {
+ j->img_comp[m].raw_data = NULL;
+ j->img_comp[m].raw_coeff = NULL;
+ }
+ j->restart_interval = 0;
+ if (!stbi__decode_jpeg_header(j, STBI__SCAN_load)) return 0;
+ m = stbi__get_marker(j);
+ while (!stbi__EOI(m)) {
+ if (stbi__SOS(m)) {
+ if (!stbi__process_scan_header(j)) return 0;
+ if (!stbi__parse_entropy_coded_data(j)) return 0;
+ if (j->marker == STBI__MARKER_none ) {
+ // handle 0s at the end of image data from IP Kamera 9060
+ while (!stbi__at_eof(j->s)) {
+ int x = stbi__get8(j->s);
+ if (x == 255) {
+ j->marker = stbi__get8(j->s);
+ break;
+ }
+ }
+ // if we reach eof without hitting a marker, stbi__get_marker() below will fail and we'll eventually return 0
+ }
+ } else if (stbi__DNL(m)) {
+ int Ld = stbi__get16be(j->s);
+ stbi__uint32 NL = stbi__get16be(j->s);
+ if (Ld != 4) stbi__err("bad DNL len", "Corrupt JPEG");
+ if (NL != j->s->img_y) stbi__err("bad DNL height", "Corrupt JPEG");
+ } else {
+ if (!stbi__process_marker(j, m)) return 0;
+ }
+ m = stbi__get_marker(j);
+ }
+ if (j->progressive)
+ stbi__jpeg_finish(j);
+ return 1;
+}
+
+// static jfif-centered resampling (across block boundaries)
+
+typedef stbi_uc *(*resample_row_func)(stbi_uc *out, stbi_uc *in0, stbi_uc *in1,
+ int w, int hs);
+
+#define stbi__div4(x) ((stbi_uc) ((x) >> 2))
+
+static stbi_uc *resample_row_1(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+{
+ STBI_NOTUSED(out);
+ STBI_NOTUSED(in_far);
+ STBI_NOTUSED(w);
+ STBI_NOTUSED(hs);
+ return in_near;
+}
+
+static stbi_uc* stbi__resample_row_v_2(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+{
+ // need to generate two samples vertically for every one in input
+ int i;
+ STBI_NOTUSED(hs);
+ for (i=0; i < w; ++i)
+ out[i] = stbi__div4(3*in_near[i] + in_far[i] + 2);
+ return out;
+}
+
+static stbi_uc* stbi__resample_row_h_2(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+{
+ // need to generate two samples horizontally for every one in input
+ int i;
+ stbi_uc *input = in_near;
+
+ if (w == 1) {
+ // if only one sample, can't do any interpolation
+ out[0] = out[1] = input[0];
+ return out;
+ }
+
+ out[0] = input[0];
+ out[1] = stbi__div4(input[0]*3 + input[1] + 2);
+ for (i=1; i < w-1; ++i) {
+ int n = 3*input[i]+2;
+ out[i*2+0] = stbi__div4(n+input[i-1]);
+ out[i*2+1] = stbi__div4(n+input[i+1]);
+ }
+ out[i*2+0] = stbi__div4(input[w-2]*3 + input[w-1] + 2);
+ out[i*2+1] = input[w-1];
+
+ STBI_NOTUSED(in_far);
+ STBI_NOTUSED(hs);
+
+ return out;
+}
+
+#define stbi__div16(x) ((stbi_uc) ((x) >> 4))
+
+static stbi_uc *stbi__resample_row_hv_2(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+{
+ // need to generate 2x2 samples for every one in input
+ int i,t0,t1;
+ if (w == 1) {
+ out[0] = out[1] = stbi__div4(3*in_near[0] + in_far[0] + 2);
+ return out;
+ }
+
+ t1 = 3*in_near[0] + in_far[0];
+ out[0] = stbi__div4(t1+2);
+ for (i=1; i < w; ++i) {
+ t0 = t1;
+ t1 = 3*in_near[i]+in_far[i];
+ out[i*2-1] = stbi__div16(3*t0 + t1 + 8);
+ out[i*2 ] = stbi__div16(3*t1 + t0 + 8);
+ }
+ out[w*2-1] = stbi__div4(t1+2);
+
+ STBI_NOTUSED(hs);
+
+ return out;
+}
+
+#if defined(STBI_SSE2) || defined(STBI_NEON)
+static stbi_uc *stbi__resample_row_hv_2_simd(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+{
+ // need to generate 2x2 samples for every one in input
+ int i=0,t0,t1;
+
+ if (w == 1) {
+ out[0] = out[1] = stbi__div4(3*in_near[0] + in_far[0] + 2);
+ return out;
+ }
+
+ t1 = 3*in_near[0] + in_far[0];
+ // process groups of 8 pixels for as long as we can.
+ // note we can't handle the last pixel in a row in this loop
+ // because we need to handle the filter boundary conditions.
+ for (; i < ((w-1) & ~7); i += 8) {
+#if defined(STBI_SSE2)
+ // load and perform the vertical filtering pass
+ // this uses 3*x + y = 4*x + (y - x)
+ __m128i zero = _mm_setzero_si128();
+ __m128i farb = _mm_loadl_epi64((__m128i *) (in_far + i));
+ __m128i nearb = _mm_loadl_epi64((__m128i *) (in_near + i));
+ __m128i farw = _mm_unpacklo_epi8(farb, zero);
+ __m128i nearw = _mm_unpacklo_epi8(nearb, zero);
+ __m128i diff = _mm_sub_epi16(farw, nearw);
+ __m128i nears = _mm_slli_epi16(nearw, 2);
+ __m128i curr = _mm_add_epi16(nears, diff); // current row
+
+ // horizontal filter works the same based on shifted vers of current
+ // row. "prev" is current row shifted right by 1 pixel; we need to
+ // insert the previous pixel value (from t1).
+ // "next" is current row shifted left by 1 pixel, with first pixel
+ // of next block of 8 pixels added in.
+ __m128i prv0 = _mm_slli_si128(curr, 2);
+ __m128i nxt0 = _mm_srli_si128(curr, 2);
+ __m128i prev = _mm_insert_epi16(prv0, t1, 0);
+ __m128i next = _mm_insert_epi16(nxt0, 3*in_near[i+8] + in_far[i+8], 7);
+
+ // horizontal filter, polyphase implementation since it's convenient:
+ // even pixels = 3*cur + prev = cur*4 + (prev - cur)
+ // odd pixels = 3*cur + next = cur*4 + (next - cur)
+ // note the shared term.
+ __m128i bias = _mm_set1_epi16(8);
+ __m128i curs = _mm_slli_epi16(curr, 2);
+ __m128i prvd = _mm_sub_epi16(prev, curr);
+ __m128i nxtd = _mm_sub_epi16(next, curr);
+ __m128i curb = _mm_add_epi16(curs, bias);
+ __m128i even = _mm_add_epi16(prvd, curb);
+ __m128i odd = _mm_add_epi16(nxtd, curb);
+
+ // interleave even and odd pixels, then undo scaling.
+ __m128i int0 = _mm_unpacklo_epi16(even, odd);
+ __m128i int1 = _mm_unpackhi_epi16(even, odd);
+ __m128i de0 = _mm_srli_epi16(int0, 4);
+ __m128i de1 = _mm_srli_epi16(int1, 4);
+
+ // pack and write output
+ __m128i outv = _mm_packus_epi16(de0, de1);
+ _mm_storeu_si128((__m128i *) (out + i*2), outv);
+#elif defined(STBI_NEON)
+ // load and perform the vertical filtering pass
+ // this uses 3*x + y = 4*x + (y - x)
+ uint8x8_t farb = vld1_u8(in_far + i);
+ uint8x8_t nearb = vld1_u8(in_near + i);
+ int16x8_t diff = vreinterpretq_s16_u16(vsubl_u8(farb, nearb));
+ int16x8_t nears = vreinterpretq_s16_u16(vshll_n_u8(nearb, 2));
+ int16x8_t curr = vaddq_s16(nears, diff); // current row
+
+ // horizontal filter works the same based on shifted vers of current
+ // row. "prev" is current row shifted right by 1 pixel; we need to
+ // insert the previous pixel value (from t1).
+ // "next" is current row shifted left by 1 pixel, with first pixel
+ // of next block of 8 pixels added in.
+ int16x8_t prv0 = vextq_s16(curr, curr, 7);
+ int16x8_t nxt0 = vextq_s16(curr, curr, 1);
+ int16x8_t prev = vsetq_lane_s16(t1, prv0, 0);
+ int16x8_t next = vsetq_lane_s16(3*in_near[i+8] + in_far[i+8], nxt0, 7);
+
+ // horizontal filter, polyphase implementation since it's convenient:
+ // even pixels = 3*cur + prev = cur*4 + (prev - cur)
+ // odd pixels = 3*cur + next = cur*4 + (next - cur)
+ // note the shared term.
+ int16x8_t curs = vshlq_n_s16(curr, 2);
+ int16x8_t prvd = vsubq_s16(prev, curr);
+ int16x8_t nxtd = vsubq_s16(next, curr);
+ int16x8_t even = vaddq_s16(curs, prvd);
+ int16x8_t odd = vaddq_s16(curs, nxtd);
+
+ // undo scaling and round, then store with even/odd phases interleaved
+ uint8x8x2_t o;
+ o.val[0] = vqrshrun_n_s16(even, 4);
+ o.val[1] = vqrshrun_n_s16(odd, 4);
+ vst2_u8(out + i*2, o);
+#endif
+
+ // "previous" value for next iter
+ t1 = 3*in_near[i+7] + in_far[i+7];
+ }
+
+ t0 = t1;
+ t1 = 3*in_near[i] + in_far[i];
+ out[i*2] = stbi__div16(3*t1 + t0 + 8);
+
+ for (++i; i < w; ++i) {
+ t0 = t1;
+ t1 = 3*in_near[i]+in_far[i];
+ out[i*2-1] = stbi__div16(3*t0 + t1 + 8);
+ out[i*2 ] = stbi__div16(3*t1 + t0 + 8);
+ }
+ out[w*2-1] = stbi__div4(t1+2);
+
+ STBI_NOTUSED(hs);
+
+ return out;
+}
+#endif
+
+static stbi_uc *stbi__resample_row_generic(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+{
+ // resample with nearest-neighbor
+ int i,j;
+ STBI_NOTUSED(in_far);
+ for (i=0; i < w; ++i)
+ for (j=0; j < hs; ++j)
+ out[i*hs+j] = in_near[i];
+ return out;
+}
+
+// this is a reduced-precision calculation of YCbCr-to-RGB introduced
+// to make sure the code produces the same results in both SIMD and scalar
+#define stbi__float2fixed(x) (((int) ((x) * 4096.0f + 0.5f)) << 8)
+static void stbi__YCbCr_to_RGB_row(stbi_uc *out, const stbi_uc *y, const stbi_uc *pcb, const stbi_uc *pcr, int count, int step)
+{
+ int i;
+ for (i=0; i < count; ++i) {
+ int y_fixed = (y[i] << 20) + (1<<19); // rounding
+ int r,g,b;
+ int cr = pcr[i] - 128;
+ int cb = pcb[i] - 128;
+ r = y_fixed + cr* stbi__float2fixed(1.40200f);
+ g = y_fixed + (cr*-stbi__float2fixed(0.71414f)) + ((cb*-stbi__float2fixed(0.34414f)) & 0xffff0000);
+ b = y_fixed + cb* stbi__float2fixed(1.77200f);
+ r >>= 20;
+ g >>= 20;
+ b >>= 20;
+ if ((unsigned) r > 255) { if (r < 0) r = 0; else r = 255; }
+ if ((unsigned) g > 255) { if (g < 0) g = 0; else g = 255; }
+ if ((unsigned) b > 255) { if (b < 0) b = 0; else b = 255; }
+ out[0] = (stbi_uc)r;
+ out[1] = (stbi_uc)g;
+ out[2] = (stbi_uc)b;
+ out[3] = 255;
+ out += step;
+ }
+}
+
+#if defined(STBI_SSE2) || defined(STBI_NEON)
+static void stbi__YCbCr_to_RGB_simd(stbi_uc *out, stbi_uc const *y, stbi_uc const *pcb, stbi_uc const *pcr, int count, int step)
+{
+ int i = 0;
+
+#ifdef STBI_SSE2
+ // step == 3 is pretty ugly on the final interleave, and i'm not convinced
+ // it's useful in practice (you wouldn't use it for textures, for example).
+ // so just accelerate step == 4 case.
+ if (step == 4) {
+ // this is a fairly straightforward implementation and not super-optimized.
+ __m128i signflip = _mm_set1_epi8(-0x80);
+ __m128i cr_const0 = _mm_set1_epi16( (short) ( 1.40200f*4096.0f+0.5f));
+ __m128i cr_const1 = _mm_set1_epi16( - (short) ( 0.71414f*4096.0f+0.5f));
+ __m128i cb_const0 = _mm_set1_epi16( - (short) ( 0.34414f*4096.0f+0.5f));
+ __m128i cb_const1 = _mm_set1_epi16( (short) ( 1.77200f*4096.0f+0.5f));
+ __m128i y_bias = _mm_set1_epi8((char) (unsigned char) 128);
+ __m128i xw = _mm_set1_epi16(255); // alpha channel
+
+ for (; i+7 < count; i += 8) {
+ // load
+ __m128i y_bytes = _mm_loadl_epi64((__m128i *) (y+i));
+ __m128i cr_bytes = _mm_loadl_epi64((__m128i *) (pcr+i));
+ __m128i cb_bytes = _mm_loadl_epi64((__m128i *) (pcb+i));
+ __m128i cr_biased = _mm_xor_si128(cr_bytes, signflip); // -128
+ __m128i cb_biased = _mm_xor_si128(cb_bytes, signflip); // -128
+
+ // unpack to short (and left-shift cr, cb by 8)
+ __m128i yw = _mm_unpacklo_epi8(y_bias, y_bytes);
+ __m128i crw = _mm_unpacklo_epi8(_mm_setzero_si128(), cr_biased);
+ __m128i cbw = _mm_unpacklo_epi8(_mm_setzero_si128(), cb_biased);
+
+ // color transform
+ __m128i yws = _mm_srli_epi16(yw, 4);
+ __m128i cr0 = _mm_mulhi_epi16(cr_const0, crw);
+ __m128i cb0 = _mm_mulhi_epi16(cb_const0, cbw);
+ __m128i cb1 = _mm_mulhi_epi16(cbw, cb_const1);
+ __m128i cr1 = _mm_mulhi_epi16(crw, cr_const1);
+ __m128i rws = _mm_add_epi16(cr0, yws);
+ __m128i gwt = _mm_add_epi16(cb0, yws);
+ __m128i bws = _mm_add_epi16(yws, cb1);
+ __m128i gws = _mm_add_epi16(gwt, cr1);
+
+ // descale
+ __m128i rw = _mm_srai_epi16(rws, 4);
+ __m128i bw = _mm_srai_epi16(bws, 4);
+ __m128i gw = _mm_srai_epi16(gws, 4);
+
+ // back to byte, set up for transpose
+ __m128i brb = _mm_packus_epi16(rw, bw);
+ __m128i gxb = _mm_packus_epi16(gw, xw);
+
+ // transpose to interleave channels
+ __m128i t0 = _mm_unpacklo_epi8(brb, gxb);
+ __m128i t1 = _mm_unpackhi_epi8(brb, gxb);
+ __m128i o0 = _mm_unpacklo_epi16(t0, t1);
+ __m128i o1 = _mm_unpackhi_epi16(t0, t1);
+
+ // store
+ _mm_storeu_si128((__m128i *) (out + 0), o0);
+ _mm_storeu_si128((__m128i *) (out + 16), o1);
+ out += 32;
+ }
+ }
+#endif
+
+#ifdef STBI_NEON
+ // in this version, step=3 support would be easy to add. but is there demand?
+ if (step == 4) {
+ // this is a fairly straightforward implementation and not super-optimized.
+ uint8x8_t signflip = vdup_n_u8(0x80);
+ int16x8_t cr_const0 = vdupq_n_s16( (short) ( 1.40200f*4096.0f+0.5f));
+ int16x8_t cr_const1 = vdupq_n_s16( - (short) ( 0.71414f*4096.0f+0.5f));
+ int16x8_t cb_const0 = vdupq_n_s16( - (short) ( 0.34414f*4096.0f+0.5f));
+ int16x8_t cb_const1 = vdupq_n_s16( (short) ( 1.77200f*4096.0f+0.5f));
+
+ for (; i+7 < count; i += 8) {
+ // load
+ uint8x8_t y_bytes = vld1_u8(y + i);
+ uint8x8_t cr_bytes = vld1_u8(pcr + i);
+ uint8x8_t cb_bytes = vld1_u8(pcb + i);
+ int8x8_t cr_biased = vreinterpret_s8_u8(vsub_u8(cr_bytes, signflip));
+ int8x8_t cb_biased = vreinterpret_s8_u8(vsub_u8(cb_bytes, signflip));
+
+ // expand to s16
+ int16x8_t yws = vreinterpretq_s16_u16(vshll_n_u8(y_bytes, 4));
+ int16x8_t crw = vshll_n_s8(cr_biased, 7);
+ int16x8_t cbw = vshll_n_s8(cb_biased, 7);
+
+ // color transform
+ int16x8_t cr0 = vqdmulhq_s16(crw, cr_const0);
+ int16x8_t cb0 = vqdmulhq_s16(cbw, cb_const0);
+ int16x8_t cr1 = vqdmulhq_s16(crw, cr_const1);
+ int16x8_t cb1 = vqdmulhq_s16(cbw, cb_const1);
+ int16x8_t rws = vaddq_s16(yws, cr0);
+ int16x8_t gws = vaddq_s16(vaddq_s16(yws, cb0), cr1);
+ int16x8_t bws = vaddq_s16(yws, cb1);
+
+ // undo scaling, round, convert to byte
+ uint8x8x4_t o;
+ o.val[0] = vqrshrun_n_s16(rws, 4);
+ o.val[1] = vqrshrun_n_s16(gws, 4);
+ o.val[2] = vqrshrun_n_s16(bws, 4);
+ o.val[3] = vdup_n_u8(255);
+
+ // store, interleaving r/g/b/a
+ vst4_u8(out, o);
+ out += 8*4;
+ }
+ }
+#endif
+
+ for (; i < count; ++i) {
+ int y_fixed = (y[i] << 20) + (1<<19); // rounding
+ int r,g,b;
+ int cr = pcr[i] - 128;
+ int cb = pcb[i] - 128;
+ r = y_fixed + cr* stbi__float2fixed(1.40200f);
+ g = y_fixed + cr*-stbi__float2fixed(0.71414f) + ((cb*-stbi__float2fixed(0.34414f)) & 0xffff0000);
+ b = y_fixed + cb* stbi__float2fixed(1.77200f);
+ r >>= 20;
+ g >>= 20;
+ b >>= 20;
+ if ((unsigned) r > 255) { if (r < 0) r = 0; else r = 255; }
+ if ((unsigned) g > 255) { if (g < 0) g = 0; else g = 255; }
+ if ((unsigned) b > 255) { if (b < 0) b = 0; else b = 255; }
+ out[0] = (stbi_uc)r;
+ out[1] = (stbi_uc)g;
+ out[2] = (stbi_uc)b;
+ out[3] = 255;
+ out += step;
+ }
+}
+#endif
+
+// set up the kernels
+static void stbi__setup_jpeg(stbi__jpeg *j)
+{
+ j->idct_block_kernel = stbi__idct_block;
+ j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_row;
+ j->resample_row_hv_2_kernel = stbi__resample_row_hv_2;
+
+#ifdef STBI_SSE2
+ if (stbi__sse2_available()) {
+ j->idct_block_kernel = stbi__idct_simd;
+ j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_simd;
+ j->resample_row_hv_2_kernel = stbi__resample_row_hv_2_simd;
+ }
+#endif
+
+#ifdef STBI_NEON
+ j->idct_block_kernel = stbi__idct_simd;
+ j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_simd;
+ j->resample_row_hv_2_kernel = stbi__resample_row_hv_2_simd;
+#endif
+}
+
+// clean up the temporary component buffers
+static void stbi__cleanup_jpeg(stbi__jpeg *j)
+{
+ stbi__free_jpeg_components(j, j->s->img_n, 0);
+}
+
+typedef struct
+{
+ resample_row_func resample;
+ stbi_uc *line0,*line1;
+ int hs,vs; // expansion factor in each axis
+ int w_lores; // horizontal pixels pre-expansion
+ int ystep; // how far through vertical expansion we are
+ int ypos; // which pre-expansion row we're on
+} stbi__resample;
+
+// fast 0..255 * 0..255 => 0..255 rounded multiplication
+static stbi_uc stbi__blinn_8x8(stbi_uc x, stbi_uc y)
+{
+ unsigned int t = x*y + 128;
+ return (stbi_uc) ((t + (t >>8)) >> 8);
+}
+
+static stbi_uc *load_jpeg_image(stbi__jpeg *z, int *out_x, int *out_y, int *comp, int req_comp)
+{
+ int n, decode_n, is_rgb;
+ z->s->img_n = 0; // make stbi__cleanup_jpeg safe
+
+ // validate req_comp
+ if (req_comp < 0 || req_comp > 4) return stbi__errpuc("bad req_comp", "Internal error");
+
+ // load a jpeg image from whichever source, but leave in YCbCr format
+ if (!stbi__decode_jpeg_image(z)) { stbi__cleanup_jpeg(z); return NULL; }
+
+ // determine actual number of components to generate
+ n = req_comp ? req_comp : z->s->img_n >= 3 ? 3 : 1;
+
+ is_rgb = z->s->img_n == 3 && (z->rgb == 3 || (z->app14_color_transform == 0 && !z->jfif));
+
+ if (z->s->img_n == 3 && n < 3 && !is_rgb)
+ decode_n = 1;
+ else
+ decode_n = z->s->img_n;
+
+ // resample and color-convert
+ {
+ int k;
+ unsigned int i,j;
+ stbi_uc *output;
+ stbi_uc *coutput[4];
+
+ stbi__resample res_comp[4];
+
+ for (k=0; k < decode_n; ++k) {
+ stbi__resample *r = &res_comp[k];
+
+ // allocate line buffer big enough for upsampling off the edges
+ // with upsample factor of 4
+ z->img_comp[k].linebuf = (stbi_uc *) stbi__malloc(z->s->img_x + 3);
+ if (!z->img_comp[k].linebuf) { stbi__cleanup_jpeg(z); return stbi__errpuc("outofmem", "Out of memory"); }
+
+ r->hs = z->img_h_max / z->img_comp[k].h;
+ r->vs = z->img_v_max / z->img_comp[k].v;
+ r->ystep = r->vs >> 1;
+ r->w_lores = (z->s->img_x + r->hs-1) / r->hs;
+ r->ypos = 0;
+ r->line0 = r->line1 = z->img_comp[k].data;
+
+ if (r->hs == 1 && r->vs == 1) r->resample = resample_row_1;
+ else if (r->hs == 1 && r->vs == 2) r->resample = stbi__resample_row_v_2;
+ else if (r->hs == 2 && r->vs == 1) r->resample = stbi__resample_row_h_2;
+ else if (r->hs == 2 && r->vs == 2) r->resample = z->resample_row_hv_2_kernel;
+ else r->resample = stbi__resample_row_generic;
+ }
+
+ // can't error after this so, this is safe
+ output = (stbi_uc *) stbi__malloc_mad3(n, z->s->img_x, z->s->img_y, 1);
+ if (!output) { stbi__cleanup_jpeg(z); return stbi__errpuc("outofmem", "Out of memory"); }
+
+ // now go ahead and resample
+ for (j=0; j < z->s->img_y; ++j) {
+ stbi_uc *out = output + n * z->s->img_x * j;
+ for (k=0; k < decode_n; ++k) {
+ stbi__resample *r = &res_comp[k];
+ int y_bot = r->ystep >= (r->vs >> 1);
+ coutput[k] = r->resample(z->img_comp[k].linebuf,
+ y_bot ? r->line1 : r->line0,
+ y_bot ? r->line0 : r->line1,
+ r->w_lores, r->hs);
+ if (++r->ystep >= r->vs) {
+ r->ystep = 0;
+ r->line0 = r->line1;
+ if (++r->ypos < z->img_comp[k].y)
+ r->line1 += z->img_comp[k].w2;
+ }
+ }
+ if (n >= 3) {
+ stbi_uc *y = coutput[0];
+ if (z->s->img_n == 3) {
+ if (is_rgb) {
+ for (i=0; i < z->s->img_x; ++i) {
+ out[0] = y[i];
+ out[1] = coutput[1][i];
+ out[2] = coutput[2][i];
+ out[3] = 255;
+ out += n;
+ }
+ } else {
+ z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n);
+ }
+ } else if (z->s->img_n == 4) {
+ if (z->app14_color_transform == 0) { // CMYK
+ for (i=0; i < z->s->img_x; ++i) {
+ stbi_uc k = coutput[3][i];
+ out[0] = stbi__blinn_8x8(coutput[0][i], k);
+ out[1] = stbi__blinn_8x8(coutput[1][i], k);
+ out[2] = stbi__blinn_8x8(coutput[2][i], k);
+ out[3] = 255;
+ out += n;
+ }
+ } else if (z->app14_color_transform == 2) { // YCCK
+ z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n);
+ for (i=0; i < z->s->img_x; ++i) {
+ stbi_uc k = coutput[3][i];
+ out[0] = stbi__blinn_8x8(255 - out[0], k);
+ out[1] = stbi__blinn_8x8(255 - out[1], k);
+ out[2] = stbi__blinn_8x8(255 - out[2], k);
+ out += n;
+ }
+ } else { // YCbCr + alpha? Ignore the fourth channel for now
+ z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n);
+ }
+ } else
+ for (i=0; i < z->s->img_x; ++i) {
+ out[0] = out[1] = out[2] = y[i];
+ out[3] = 255; // not used if n==3
+ out += n;
+ }
+ } else {
+ if (is_rgb) {
+ if (n == 1)
+ for (i=0; i < z->s->img_x; ++i)
+ *out++ = stbi__compute_y(coutput[0][i], coutput[1][i], coutput[2][i]);
+ else {
+ for (i=0; i < z->s->img_x; ++i, out += 2) {
+ out[0] = stbi__compute_y(coutput[0][i], coutput[1][i], coutput[2][i]);
+ out[1] = 255;
+ }
+ }
+ } else if (z->s->img_n == 4 && z->app14_color_transform == 0) {
+ for (i=0; i < z->s->img_x; ++i) {
+ stbi_uc k = coutput[3][i];
+ stbi_uc r = stbi__blinn_8x8(coutput[0][i], k);
+ stbi_uc g = stbi__blinn_8x8(coutput[1][i], k);
+ stbi_uc b = stbi__blinn_8x8(coutput[2][i], k);
+ out[0] = stbi__compute_y(r, g, b);
+ out[1] = 255;
+ out += n;
+ }
+ } else if (z->s->img_n == 4 && z->app14_color_transform == 2) {
+ for (i=0; i < z->s->img_x; ++i) {
+ out[0] = stbi__blinn_8x8(255 - coutput[0][i], coutput[3][i]);
+ out[1] = 255;
+ out += n;
+ }
+ } else {
+ stbi_uc *y = coutput[0];
+ if (n == 1)
+ for (i=0; i < z->s->img_x; ++i) out[i] = y[i];
+ else
+ for (i=0; i < z->s->img_x; ++i) *out++ = y[i], *out++ = 255;
+ }
+ }
+ }
+ stbi__cleanup_jpeg(z);
+ *out_x = z->s->img_x;
+ *out_y = z->s->img_y;
+ if (comp) *comp = z->s->img_n >= 3 ? 3 : 1; // report original components, not output
+ return output;
+ }
+}
+
+static void *stbi__jpeg_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+ unsigned char* result;
+ stbi__jpeg* j = (stbi__jpeg*) stbi__malloc(sizeof(stbi__jpeg));
+ STBI_NOTUSED(ri);
+ j->s = s;
+ stbi__setup_jpeg(j);
+ result = load_jpeg_image(j, x,y,comp,req_comp);
+ STBI_FREE(j);
+ return result;
+}
+
+static int stbi__jpeg_test(stbi__context *s)
+{
+ int r;
+ stbi__jpeg* j = (stbi__jpeg*)stbi__malloc(sizeof(stbi__jpeg));
+ j->s = s;
+ stbi__setup_jpeg(j);
+ r = stbi__decode_jpeg_header(j, STBI__SCAN_type);
+ stbi__rewind(s);
+ STBI_FREE(j);
+ return r;
+}
+
+static int stbi__jpeg_info_raw(stbi__jpeg *j, int *x, int *y, int *comp)
+{
+ if (!stbi__decode_jpeg_header(j, STBI__SCAN_header)) {
+ stbi__rewind( j->s );
+ return 0;
+ }
+ if (x) *x = j->s->img_x;
+ if (y) *y = j->s->img_y;
+ if (comp) *comp = j->s->img_n >= 3 ? 3 : 1;
+ return 1;
+}
+
+static int stbi__jpeg_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ int result;
+ stbi__jpeg* j = (stbi__jpeg*) (stbi__malloc(sizeof(stbi__jpeg)));
+ j->s = s;
+ result = stbi__jpeg_info_raw(j, x, y, comp);
+ STBI_FREE(j);
+ return result;
+}
+#endif
+
+// public domain zlib decode v0.2 Sean Barrett 2006-11-18
+// simple implementation
+// - all input must be provided in an upfront buffer
+// - all output is written to a single output buffer (can malloc/realloc)
+// performance
+// - fast huffman
+
+#ifndef STBI_NO_ZLIB
+
+// fast-way is faster to check than jpeg huffman, but slow way is slower
+#define STBI__ZFAST_BITS 9 // accelerate all cases in default tables
+#define STBI__ZFAST_MASK ((1 << STBI__ZFAST_BITS) - 1)
+
+// zlib-style huffman encoding
+// (jpegs packs from left, zlib from right, so can't share code)
+typedef struct
+{
+ stbi__uint16 fast[1 << STBI__ZFAST_BITS];
+ stbi__uint16 firstcode[16];
+ int maxcode[17];
+ stbi__uint16 firstsymbol[16];
+ stbi_uc size[288];
+ stbi__uint16 value[288];
+} stbi__zhuffman;
+
+stbi_inline static int stbi__bitreverse16(int n)
+{
+ n = ((n & 0xAAAA) >> 1) | ((n & 0x5555) << 1);
+ n = ((n & 0xCCCC) >> 2) | ((n & 0x3333) << 2);
+ n = ((n & 0xF0F0) >> 4) | ((n & 0x0F0F) << 4);
+ n = ((n & 0xFF00) >> 8) | ((n & 0x00FF) << 8);
+ return n;
+}
+
+stbi_inline static int stbi__bit_reverse(int v, int bits)
+{
+ STBI_ASSERT(bits <= 16);
+ // to bit reverse n bits, reverse 16 and shift
+ // e.g. 11 bits, bit reverse and shift away 5
+ return stbi__bitreverse16(v) >> (16-bits);
+}
+
+static int stbi__zbuild_huffman(stbi__zhuffman *z, const stbi_uc *sizelist, int num)
+{
+ int i,k=0;
+ int code, next_code[16], sizes[17];
+
+ // DEFLATE spec for generating codes
+ memset(sizes, 0, sizeof(sizes));
+ memset(z->fast, 0, sizeof(z->fast));
+ for (i=0; i < num; ++i)
+ ++sizes[sizelist[i]];
+ sizes[0] = 0;
+ for (i=1; i < 16; ++i)
+ if (sizes[i] > (1 << i))
+ return stbi__err("bad sizes", "Corrupt PNG");
+ code = 0;
+ for (i=1; i < 16; ++i) {
+ next_code[i] = code;
+ z->firstcode[i] = (stbi__uint16) code;
+ z->firstsymbol[i] = (stbi__uint16) k;
+ code = (code + sizes[i]);
+ if (sizes[i])
+ if (code-1 >= (1 << i)) return stbi__err("bad codelengths","Corrupt PNG");
+ z->maxcode[i] = code << (16-i); // preshift for inner loop
+ code <<= 1;
+ k += sizes[i];
+ }
+ z->maxcode[16] = 0x10000; // sentinel
+ for (i=0; i < num; ++i) {
+ int s = sizelist[i];
+ if (s) {
+ int c = next_code[s] - z->firstcode[s] + z->firstsymbol[s];
+ stbi__uint16 fastv = (stbi__uint16) ((s << 9) | i);
+ z->size [c] = (stbi_uc ) s;
+ z->value[c] = (stbi__uint16) i;
+ if (s <= STBI__ZFAST_BITS) {
+ int j = stbi__bit_reverse(next_code[s],s);
+ while (j < (1 << STBI__ZFAST_BITS)) {
+ z->fast[j] = fastv;
+ j += (1 << s);
+ }
+ }
+ ++next_code[s];
+ }
+ }
+ return 1;
+}
+
+// zlib-from-memory implementation for PNG reading
+// because PNG allows splitting the zlib stream arbitrarily,
+// and it's annoying structurally to have PNG call ZLIB call PNG,
+// we require PNG read all the IDATs and combine them into a single
+// memory buffer
+
+typedef struct
+{
+ stbi_uc *zbuffer, *zbuffer_end;
+ int num_bits;
+ stbi__uint32 code_buffer;
+
+ char *zout;
+ char *zout_start;
+ char *zout_end;
+ int z_expandable;
+
+ stbi__zhuffman z_length, z_distance;
+} stbi__zbuf;
+
+stbi_inline static stbi_uc stbi__zget8(stbi__zbuf *z)
+{
+ if (z->zbuffer >= z->zbuffer_end) return 0;
+ return *z->zbuffer++;
+}
+
+static void stbi__fill_bits(stbi__zbuf *z)
+{
+ do {
+ STBI_ASSERT(z->code_buffer < (1U << z->num_bits));
+ z->code_buffer |= (unsigned int) stbi__zget8(z) << z->num_bits;
+ z->num_bits += 8;
+ } while (z->num_bits <= 24);
+}
+
+stbi_inline static unsigned int stbi__zreceive(stbi__zbuf *z, int n)
+{
+ unsigned int k;
+ if (z->num_bits < n) stbi__fill_bits(z);
+ k = z->code_buffer & ((1 << n) - 1);
+ z->code_buffer >>= n;
+ z->num_bits -= n;
+ return k;
+}
+
+static int stbi__zhuffman_decode_slowpath(stbi__zbuf *a, stbi__zhuffman *z)
+{
+ int b,s,k;
+ // not resolved by fast table, so compute it the slow way
+ // use jpeg approach, which requires MSbits at top
+ k = stbi__bit_reverse(a->code_buffer, 16);
+ for (s=STBI__ZFAST_BITS+1; ; ++s)
+ if (k < z->maxcode[s])
+ break;
+ if (s == 16) return -1; // invalid code!
+ // code size is s, so:
+ b = (k >> (16-s)) - z->firstcode[s] + z->firstsymbol[s];
+ STBI_ASSERT(z->size[b] == s);
+ a->code_buffer >>= s;
+ a->num_bits -= s;
+ return z->value[b];
+}
+
+stbi_inline static int stbi__zhuffman_decode(stbi__zbuf *a, stbi__zhuffman *z)
+{
+ int b,s;
+ if (a->num_bits < 16) stbi__fill_bits(a);
+ b = z->fast[a->code_buffer & STBI__ZFAST_MASK];
+ if (b) {
+ s = b >> 9;
+ a->code_buffer >>= s;
+ a->num_bits -= s;
+ return b & 511;
+ }
+ return stbi__zhuffman_decode_slowpath(a, z);
+}
+
+static int stbi__zexpand(stbi__zbuf *z, char *zout, int n) // need to make room for n bytes
+{
+ char *q;
+ int cur, limit, old_limit;
+ z->zout = zout;
+ if (!z->z_expandable) return stbi__err("output buffer limit","Corrupt PNG");
+ cur = (int) (z->zout - z->zout_start);
+ limit = old_limit = (int) (z->zout_end - z->zout_start);
+ while (cur + n > limit)
+ limit *= 2;
+ q = (char *) STBI_REALLOC_SIZED(z->zout_start, old_limit, limit);
+ STBI_NOTUSED(old_limit);
+ if (q == NULL) return stbi__err("outofmem", "Out of memory");
+ z->zout_start = q;
+ z->zout = q + cur;
+ z->zout_end = q + limit;
+ return 1;
+}
+
+static int stbi__zlength_base[31] = {
+ 3,4,5,6,7,8,9,10,11,13,
+ 15,17,19,23,27,31,35,43,51,59,
+ 67,83,99,115,131,163,195,227,258,0,0 };
+
+static int stbi__zlength_extra[31]=
+{ 0,0,0,0,0,0,0,0,1,1,1,1,2,2,2,2,3,3,3,3,4,4,4,4,5,5,5,5,0,0,0 };
+
+static int stbi__zdist_base[32] = { 1,2,3,4,5,7,9,13,17,25,33,49,65,97,129,193,
+257,385,513,769,1025,1537,2049,3073,4097,6145,8193,12289,16385,24577,0,0};
+
+static int stbi__zdist_extra[32] =
+{ 0,0,0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,8,8,9,9,10,10,11,11,12,12,13,13};
+
+static int stbi__parse_huffman_block(stbi__zbuf *a)
+{
+ char *zout = a->zout;
+ for(;;) {
+ int z = stbi__zhuffman_decode(a, &a->z_length);
+ if (z < 256) {
+ if (z < 0) return stbi__err("bad huffman code","Corrupt PNG"); // error in huffman codes
+ if (zout >= a->zout_end) {
+ if (!stbi__zexpand(a, zout, 1)) return 0;
+ zout = a->zout;
+ }
+ *zout++ = (char) z;
+ } else {
+ stbi_uc *p;
+ int len,dist;
+ if (z == 256) {
+ a->zout = zout;
+ return 1;
+ }
+ z -= 257;
+ len = stbi__zlength_base[z];
+ if (stbi__zlength_extra[z]) len += stbi__zreceive(a, stbi__zlength_extra[z]);
+ z = stbi__zhuffman_decode(a, &a->z_distance);
+ if (z < 0) return stbi__err("bad huffman code","Corrupt PNG");
+ dist = stbi__zdist_base[z];
+ if (stbi__zdist_extra[z]) dist += stbi__zreceive(a, stbi__zdist_extra[z]);
+ if (zout - a->zout_start < dist) return stbi__err("bad dist","Corrupt PNG");
+ if (zout + len > a->zout_end) {
+ if (!stbi__zexpand(a, zout, len)) return 0;
+ zout = a->zout;
+ }
+ p = (stbi_uc *) (zout - dist);
+ if (dist == 1) { // run of one byte; common in images.
+ stbi_uc v = *p;
+ if (len) { do *zout++ = v; while (--len); }
+ } else {
+ if (len) { do *zout++ = *p++; while (--len); }
+ }
+ }
+ }
+}
+
+static int stbi__compute_huffman_codes(stbi__zbuf *a)
+{
+ static stbi_uc length_dezigzag[19] = { 16,17,18,0,8,7,9,6,10,5,11,4,12,3,13,2,14,1,15 };
+ stbi__zhuffman z_codelength;
+ stbi_uc lencodes[286+32+137];//padding for maximum single op
+ stbi_uc codelength_sizes[19];
+ int i,n;
+
+ int hlit = stbi__zreceive(a,5) + 257;
+ int hdist = stbi__zreceive(a,5) + 1;
+ int hclen = stbi__zreceive(a,4) + 4;
+ int ntot = hlit + hdist;
+
+ memset(codelength_sizes, 0, sizeof(codelength_sizes));
+ for (i=0; i < hclen; ++i) {
+ int s = stbi__zreceive(a,3);
+ codelength_sizes[length_dezigzag[i]] = (stbi_uc) s;
+ }
+ if (!stbi__zbuild_huffman(&z_codelength, codelength_sizes, 19)) return 0;
+
+ n = 0;
+ while (n < ntot) {
+ int c = stbi__zhuffman_decode(a, &z_codelength);
+ if (c < 0 || c >= 19) return stbi__err("bad codelengths", "Corrupt PNG");
+ if (c < 16)
+ lencodes[n++] = (stbi_uc) c;
+ else {
+ stbi_uc fill = 0;
+ if (c == 16) {
+ c = stbi__zreceive(a,2)+3;
+ if (n == 0) return stbi__err("bad codelengths", "Corrupt PNG");
+ fill = lencodes[n-1];
+ } else if (c == 17)
+ c = stbi__zreceive(a,3)+3;
+ else {
+ STBI_ASSERT(c == 18);
+ c = stbi__zreceive(a,7)+11;
+ }
+ if (ntot - n < c) return stbi__err("bad codelengths", "Corrupt PNG");
+ memset(lencodes+n, fill, c);
+ n += c;
+ }
+ }
+ if (n != ntot) return stbi__err("bad codelengths","Corrupt PNG");
+ if (!stbi__zbuild_huffman(&a->z_length, lencodes, hlit)) return 0;
+ if (!stbi__zbuild_huffman(&a->z_distance, lencodes+hlit, hdist)) return 0;
+ return 1;
+}
+
+static int stbi__parse_uncompressed_block(stbi__zbuf *a)
+{
+ stbi_uc header[4];
+ int len,nlen,k;
+ if (a->num_bits & 7)
+ stbi__zreceive(a, a->num_bits & 7); // discard
+ // drain the bit-packed data into header
+ k = 0;
+ while (a->num_bits > 0) {
+ header[k++] = (stbi_uc) (a->code_buffer & 255); // suppress MSVC run-time check
+ a->code_buffer >>= 8;
+ a->num_bits -= 8;
+ }
+ STBI_ASSERT(a->num_bits == 0);
+ // now fill header the normal way
+ while (k < 4)
+ header[k++] = stbi__zget8(a);
+ len = header[1] * 256 + header[0];
+ nlen = header[3] * 256 + header[2];
+ if (nlen != (len ^ 0xffff)) return stbi__err("zlib corrupt","Corrupt PNG");
+ if (a->zbuffer + len > a->zbuffer_end) return stbi__err("read past buffer","Corrupt PNG");
+ if (a->zout + len > a->zout_end)
+ if (!stbi__zexpand(a, a->zout, len)) return 0;
+ memcpy(a->zout, a->zbuffer, len);
+ a->zbuffer += len;
+ a->zout += len;
+ return 1;
+}
+
+static int stbi__parse_zlib_header(stbi__zbuf *a)
+{
+ int cmf = stbi__zget8(a);
+ int cm = cmf & 15;
+ /* int cinfo = cmf >> 4; */
+ int flg = stbi__zget8(a);
+ if ((cmf*256+flg) % 31 != 0) return stbi__err("bad zlib header","Corrupt PNG"); // zlib spec
+ if (flg & 32) return stbi__err("no preset dict","Corrupt PNG"); // preset dictionary not allowed in png
+ if (cm != 8) return stbi__err("bad compression","Corrupt PNG"); // DEFLATE required for png
+ // window = 1 << (8 + cinfo)... but who cares, we fully buffer output
+ return 1;
+}
+
+static const stbi_uc stbi__zdefault_length[288] =
+{
+ 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
+ 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
+ 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
+ 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
+ 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,
+ 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,
+ 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,
+ 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,
+ 7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7, 7,7,7,7,7,7,7,7,8,8,8,8,8,8,8,8
+};
+static const stbi_uc stbi__zdefault_distance[32] =
+{
+ 5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5
+};
+/*
+Init algorithm:
+{
+ int i; // use <= to match clearly with spec
+ for (i=0; i <= 143; ++i) stbi__zdefault_length[i] = 8;
+ for ( ; i <= 255; ++i) stbi__zdefault_length[i] = 9;
+ for ( ; i <= 279; ++i) stbi__zdefault_length[i] = 7;
+ for ( ; i <= 287; ++i) stbi__zdefault_length[i] = 8;
+
+ for (i=0; i <= 31; ++i) stbi__zdefault_distance[i] = 5;
+}
+*/
+
+static int stbi__parse_zlib(stbi__zbuf *a, int parse_header)
+{
+ int final, type;
+ if (parse_header)
+ if (!stbi__parse_zlib_header(a)) return 0;
+ a->num_bits = 0;
+ a->code_buffer = 0;
+ do {
+ final = stbi__zreceive(a,1);
+ type = stbi__zreceive(a,2);
+ if (type == 0) {
+ if (!stbi__parse_uncompressed_block(a)) return 0;
+ } else if (type == 3) {
+ return 0;
+ } else {
+ if (type == 1) {
+ // use fixed code lengths
+ if (!stbi__zbuild_huffman(&a->z_length , stbi__zdefault_length , 288)) return 0;
+ if (!stbi__zbuild_huffman(&a->z_distance, stbi__zdefault_distance, 32)) return 0;
+ } else {
+ if (!stbi__compute_huffman_codes(a)) return 0;
+ }
+ if (!stbi__parse_huffman_block(a)) return 0;
+ }
+ } while (!final);
+ return 1;
+}
+
+static int stbi__do_zlib(stbi__zbuf *a, char *obuf, int olen, int exp, int parse_header)
+{
+ a->zout_start = obuf;
+ a->zout = obuf;
+ a->zout_end = obuf + olen;
+ a->z_expandable = exp;
+
+ return stbi__parse_zlib(a, parse_header);
+}
+
+STBIDEF char *stbi_zlib_decode_malloc_guesssize(const char *buffer, int len, int initial_size, int *outlen)
+{
+ stbi__zbuf a;
+ char *p = (char *) stbi__malloc(initial_size);
+ if (p == NULL) return NULL;
+ a.zbuffer = (stbi_uc *) buffer;
+ a.zbuffer_end = (stbi_uc *) buffer + len;
+ if (stbi__do_zlib(&a, p, initial_size, 1, 1)) {
+ if (outlen) *outlen = (int) (a.zout - a.zout_start);
+ return a.zout_start;
+ } else {
+ STBI_FREE(a.zout_start);
+ return NULL;
+ }
+}
+
+STBIDEF char *stbi_zlib_decode_malloc(char const *buffer, int len, int *outlen)
+{
+ return stbi_zlib_decode_malloc_guesssize(buffer, len, 16384, outlen);
+}
+
+STBIDEF char *stbi_zlib_decode_malloc_guesssize_headerflag(const char *buffer, int len, int initial_size, int *outlen, int parse_header)
+{
+ stbi__zbuf a;
+ char *p = (char *) stbi__malloc(initial_size);
+ if (p == NULL) return NULL;
+ a.zbuffer = (stbi_uc *) buffer;
+ a.zbuffer_end = (stbi_uc *) buffer + len;
+ if (stbi__do_zlib(&a, p, initial_size, 1, parse_header)) {
+ if (outlen) *outlen = (int) (a.zout - a.zout_start);
+ return a.zout_start;
+ } else {
+ STBI_FREE(a.zout_start);
+ return NULL;
+ }
+}
+
+STBIDEF int stbi_zlib_decode_buffer(char *obuffer, int olen, char const *ibuffer, int ilen)
+{
+ stbi__zbuf a;
+ a.zbuffer = (stbi_uc *) ibuffer;
+ a.zbuffer_end = (stbi_uc *) ibuffer + ilen;
+ if (stbi__do_zlib(&a, obuffer, olen, 0, 1))
+ return (int) (a.zout - a.zout_start);
+ else
+ return -1;
+}
+
+STBIDEF char *stbi_zlib_decode_noheader_malloc(char const *buffer, int len, int *outlen)
+{
+ stbi__zbuf a;
+ char *p = (char *) stbi__malloc(16384);
+ if (p == NULL) return NULL;
+ a.zbuffer = (stbi_uc *) buffer;
+ a.zbuffer_end = (stbi_uc *) buffer+len;
+ if (stbi__do_zlib(&a, p, 16384, 1, 0)) {
+ if (outlen) *outlen = (int) (a.zout - a.zout_start);
+ return a.zout_start;
+ } else {
+ STBI_FREE(a.zout_start);
+ return NULL;
+ }
+}
+
+STBIDEF int stbi_zlib_decode_noheader_buffer(char *obuffer, int olen, const char *ibuffer, int ilen)
+{
+ stbi__zbuf a;
+ a.zbuffer = (stbi_uc *) ibuffer;
+ a.zbuffer_end = (stbi_uc *) ibuffer + ilen;
+ if (stbi__do_zlib(&a, obuffer, olen, 0, 0))
+ return (int) (a.zout - a.zout_start);
+ else
+ return -1;
+}
+#endif
+
+// public domain "baseline" PNG decoder v0.10 Sean Barrett 2006-11-18
+// simple implementation
+// - only 8-bit samples
+// - no CRC checking
+// - allocates lots of intermediate memory
+// - avoids problem of streaming data between subsystems
+// - avoids explicit window management
+// performance
+// - uses stb_zlib, a PD zlib implementation with fast huffman decoding
+
+#ifndef STBI_NO_PNG
+typedef struct
+{
+ stbi__uint32 length;
+ stbi__uint32 type;
+} stbi__pngchunk;
+
+static stbi__pngchunk stbi__get_chunk_header(stbi__context *s)
+{
+ stbi__pngchunk c;
+ c.length = stbi__get32be(s);
+ c.type = stbi__get32be(s);
+ return c;
+}
+
+static int stbi__check_png_header(stbi__context *s)
+{
+ static stbi_uc png_sig[8] = { 137,80,78,71,13,10,26,10 };
+ int i;
+ for (i=0; i < 8; ++i)
+ if (stbi__get8(s) != png_sig[i]) return stbi__err("bad png sig","Not a PNG");
+ return 1;
+}
+
+typedef struct
+{
+ stbi__context *s;
+ stbi_uc *idata, *expanded, *out;
+ int depth;
+} stbi__png;
+
+
+enum {
+ STBI__F_none=0,
+ STBI__F_sub=1,
+ STBI__F_up=2,
+ STBI__F_avg=3,
+ STBI__F_paeth=4,
+ // synthetic filters used for first scanline to avoid needing a dummy row of 0s
+ STBI__F_avg_first,
+ STBI__F_paeth_first
+};
+
+static stbi_uc first_row_filter[5] =
+{
+ STBI__F_none,
+ STBI__F_sub,
+ STBI__F_none,
+ STBI__F_avg_first,
+ STBI__F_paeth_first
+};
+
+static int stbi__paeth(int a, int b, int c)
+{
+ int p = a + b - c;
+ int pa = abs(p-a);
+ int pb = abs(p-b);
+ int pc = abs(p-c);
+ if (pa <= pb && pa <= pc) return a;
+ if (pb <= pc) return b;
+ return c;
+}
+
+static stbi_uc stbi__depth_scale_table[9] = { 0, 0xff, 0x55, 0, 0x11, 0,0,0, 0x01 };
+
+// create the png data from post-deflated data
+static int stbi__create_png_image_raw(stbi__png *a, stbi_uc *raw, stbi__uint32 raw_len, int out_n, stbi__uint32 x, stbi__uint32 y, int depth, int color)
+{
+ int bytes = (depth == 16? 2 : 1);
+ stbi__context *s = a->s;
+ stbi__uint32 i,j,stride = x*out_n*bytes;
+ stbi__uint32 img_len, img_width_bytes;
+ int k;
+ int img_n = s->img_n; // copy it into a local for later
+
+ int output_bytes = out_n*bytes;
+ int filter_bytes = img_n*bytes;
+ int width = x;
+
+ STBI_ASSERT(out_n == s->img_n || out_n == s->img_n+1);
+ a->out = (stbi_uc *) stbi__malloc_mad3(x, y, output_bytes, 0); // extra bytes to write off the end into
+ if (!a->out) return stbi__err("outofmem", "Out of memory");
+
+ img_width_bytes = (((img_n * x * depth) + 7) >> 3);
+ img_len = (img_width_bytes + 1) * y;
+ if (s->img_x == x && s->img_y == y) {
+ if (raw_len != img_len) return stbi__err("not enough pixels","Corrupt PNG");
+ } else { // interlaced:
+ if (raw_len < img_len) return stbi__err("not enough pixels","Corrupt PNG");
+ }
+
+ for (j=0; j < y; ++j) {
+ stbi_uc *cur = a->out + stride*j;
+ stbi_uc *prior;
+ int filter = *raw++;
+
+ if (filter > 4)
+ return stbi__err("invalid filter","Corrupt PNG");
+
+ if (depth < 8) {
+ STBI_ASSERT(img_width_bytes <= x);
+ cur += x*out_n - img_width_bytes; // store output to the rightmost img_len bytes, so we can decode in place
+ filter_bytes = 1;
+ width = img_width_bytes;
+ }
+ prior = cur - stride; // bugfix: need to compute this after 'cur +=' computation above
+
+ // if first row, use special filter that doesn't sample previous row
+ if (j == 0) filter = first_row_filter[filter];
+
+ // handle first byte explicitly
+ for (k=0; k < filter_bytes; ++k) {
+ switch (filter) {
+ case STBI__F_none : cur[k] = raw[k]; break;
+ case STBI__F_sub : cur[k] = raw[k]; break;
+ case STBI__F_up : cur[k] = STBI__BYTECAST(raw[k] + prior[k]); break;
+ case STBI__F_avg : cur[k] = STBI__BYTECAST(raw[k] + (prior[k]>>1)); break;
+ case STBI__F_paeth : cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(0,prior[k],0)); break;
+ case STBI__F_avg_first : cur[k] = raw[k]; break;
+ case STBI__F_paeth_first: cur[k] = raw[k]; break;
+ }
+ }
+
+ if (depth == 8) {
+ if (img_n != out_n)
+ cur[img_n] = 255; // first pixel
+ raw += img_n;
+ cur += out_n;
+ prior += out_n;
+ } else if (depth == 16) {
+ if (img_n != out_n) {
+ cur[filter_bytes] = 255; // first pixel top byte
+ cur[filter_bytes+1] = 255; // first pixel bottom byte
+ }
+ raw += filter_bytes;
+ cur += output_bytes;
+ prior += output_bytes;
+ } else {
+ raw += 1;
+ cur += 1;
+ prior += 1;
+ }
+
+ // this is a little gross, so that we don't switch per-pixel or per-component
+ if (depth < 8 || img_n == out_n) {
+ int nk = (width - 1)*filter_bytes;
+ #define STBI__CASE(f) \
+ case f: \
+ for (k=0; k < nk; ++k)
+ switch (filter) {
+ // "none" filter turns into a memcpy here; make that explicit.
+ case STBI__F_none: memcpy(cur, raw, nk); break;
+ STBI__CASE(STBI__F_sub) { cur[k] = STBI__BYTECAST(raw[k] + cur[k-filter_bytes]); } break;
+ STBI__CASE(STBI__F_up) { cur[k] = STBI__BYTECAST(raw[k] + prior[k]); } break;
+ STBI__CASE(STBI__F_avg) { cur[k] = STBI__BYTECAST(raw[k] + ((prior[k] + cur[k-filter_bytes])>>1)); } break;
+ STBI__CASE(STBI__F_paeth) { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k-filter_bytes],prior[k],prior[k-filter_bytes])); } break;
+ STBI__CASE(STBI__F_avg_first) { cur[k] = STBI__BYTECAST(raw[k] + (cur[k-filter_bytes] >> 1)); } break;
+ STBI__CASE(STBI__F_paeth_first) { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k-filter_bytes],0,0)); } break;
+ }
+ #undef STBI__CASE
+ raw += nk;
+ } else {
+ STBI_ASSERT(img_n+1 == out_n);
+ #define STBI__CASE(f) \
+ case f: \
+ for (i=x-1; i >= 1; --i, cur[filter_bytes]=255,raw+=filter_bytes,cur+=output_bytes,prior+=output_bytes) \
+ for (k=0; k < filter_bytes; ++k)
+ switch (filter) {
+ STBI__CASE(STBI__F_none) { cur[k] = raw[k]; } break;
+ STBI__CASE(STBI__F_sub) { cur[k] = STBI__BYTECAST(raw[k] + cur[k- output_bytes]); } break;
+ STBI__CASE(STBI__F_up) { cur[k] = STBI__BYTECAST(raw[k] + prior[k]); } break;
+ STBI__CASE(STBI__F_avg) { cur[k] = STBI__BYTECAST(raw[k] + ((prior[k] + cur[k- output_bytes])>>1)); } break;
+ STBI__CASE(STBI__F_paeth) { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k- output_bytes],prior[k],prior[k- output_bytes])); } break;
+ STBI__CASE(STBI__F_avg_first) { cur[k] = STBI__BYTECAST(raw[k] + (cur[k- output_bytes] >> 1)); } break;
+ STBI__CASE(STBI__F_paeth_first) { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k- output_bytes],0,0)); } break;
+ }
+ #undef STBI__CASE
+
+ // the loop above sets the high byte of the pixels' alpha, but for
+ // 16 bit png files we also need the low byte set. we'll do that here.
+ if (depth == 16) {
+ cur = a->out + stride*j; // start at the beginning of the row again
+ for (i=0; i < x; ++i,cur+=output_bytes) {
+ cur[filter_bytes+1] = 255;
+ }
+ }
+ }
+ }
+
+ // we make a separate pass to expand bits to pixels; for performance,
+ // this could run two scanlines behind the above code, so it won't
+ // intefere with filtering but will still be in the cache.
+ if (depth < 8) {
+ for (j=0; j < y; ++j) {
+ stbi_uc *cur = a->out + stride*j;
+ stbi_uc *in = a->out + stride*j + x*out_n - img_width_bytes;
+ // unpack 1/2/4-bit into a 8-bit buffer. allows us to keep the common 8-bit path optimal at minimal cost for 1/2/4-bit
+ // png guarante byte alignment, if width is not multiple of 8/4/2 we'll decode dummy trailing data that will be skipped in the later loop
+ stbi_uc scale = (color == 0) ? stbi__depth_scale_table[depth] : 1; // scale grayscale values to 0..255 range
+
+ // note that the final byte might overshoot and write more data than desired.
+ // we can allocate enough data that this never writes out of memory, but it
+ // could also overwrite the next scanline. can it overwrite non-empty data
+ // on the next scanline? yes, consider 1-pixel-wide scanlines with 1-bit-per-pixel.
+ // so we need to explicitly clamp the final ones
+
+ if (depth == 4) {
+ for (k=x*img_n; k >= 2; k-=2, ++in) {
+ *cur++ = scale * ((*in >> 4) );
+ *cur++ = scale * ((*in ) & 0x0f);
+ }
+ if (k > 0) *cur++ = scale * ((*in >> 4) );
+ } else if (depth == 2) {
+ for (k=x*img_n; k >= 4; k-=4, ++in) {
+ *cur++ = scale * ((*in >> 6) );
+ *cur++ = scale * ((*in >> 4) & 0x03);
+ *cur++ = scale * ((*in >> 2) & 0x03);
+ *cur++ = scale * ((*in ) & 0x03);
+ }
+ if (k > 0) *cur++ = scale * ((*in >> 6) );
+ if (k > 1) *cur++ = scale * ((*in >> 4) & 0x03);
+ if (k > 2) *cur++ = scale * ((*in >> 2) & 0x03);
+ } else if (depth == 1) {
+ for (k=x*img_n; k >= 8; k-=8, ++in) {
+ *cur++ = scale * ((*in >> 7) );
+ *cur++ = scale * ((*in >> 6) & 0x01);
+ *cur++ = scale * ((*in >> 5) & 0x01);
+ *cur++ = scale * ((*in >> 4) & 0x01);
+ *cur++ = scale * ((*in >> 3) & 0x01);
+ *cur++ = scale * ((*in >> 2) & 0x01);
+ *cur++ = scale * ((*in >> 1) & 0x01);
+ *cur++ = scale * ((*in ) & 0x01);
+ }
+ if (k > 0) *cur++ = scale * ((*in >> 7) );
+ if (k > 1) *cur++ = scale * ((*in >> 6) & 0x01);
+ if (k > 2) *cur++ = scale * ((*in >> 5) & 0x01);
+ if (k > 3) *cur++ = scale * ((*in >> 4) & 0x01);
+ if (k > 4) *cur++ = scale * ((*in >> 3) & 0x01);
+ if (k > 5) *cur++ = scale * ((*in >> 2) & 0x01);
+ if (k > 6) *cur++ = scale * ((*in >> 1) & 0x01);
+ }
+ if (img_n != out_n) {
+ int q;
+ // insert alpha = 255
+ cur = a->out + stride*j;
+ if (img_n == 1) {
+ for (q=x-1; q >= 0; --q) {
+ cur[q*2+1] = 255;
+ cur[q*2+0] = cur[q];
+ }
+ } else {
+ STBI_ASSERT(img_n == 3);
+ for (q=x-1; q >= 0; --q) {
+ cur[q*4+3] = 255;
+ cur[q*4+2] = cur[q*3+2];
+ cur[q*4+1] = cur[q*3+1];
+ cur[q*4+0] = cur[q*3+0];
+ }
+ }
+ }
+ }
+ } else if (depth == 16) {
+ // force the image data from big-endian to platform-native.
+ // this is done in a separate pass due to the decoding relying
+ // on the data being untouched, but could probably be done
+ // per-line during decode if care is taken.
+ stbi_uc *cur = a->out;
+ stbi__uint16 *cur16 = (stbi__uint16*)cur;
+
+ for(i=0; i < x*y*out_n; ++i,cur16++,cur+=2) {
+ *cur16 = (cur[0] << 8) | cur[1];
+ }
+ }
+
+ return 1;
+}
+
+static int stbi__create_png_image(stbi__png *a, stbi_uc *image_data, stbi__uint32 image_data_len, int out_n, int depth, int color, int interlaced)
+{
+ int bytes = (depth == 16 ? 2 : 1);
+ int out_bytes = out_n * bytes;
+ stbi_uc *final;
+ int p;
+ if (!interlaced)
+ return stbi__create_png_image_raw(a, image_data, image_data_len, out_n, a->s->img_x, a->s->img_y, depth, color);
+
+ // de-interlacing
+ final = (stbi_uc *) stbi__malloc_mad3(a->s->img_x, a->s->img_y, out_bytes, 0);
+ for (p=0; p < 7; ++p) {
+ int xorig[] = { 0,4,0,2,0,1,0 };
+ int yorig[] = { 0,0,4,0,2,0,1 };
+ int xspc[] = { 8,8,4,4,2,2,1 };
+ int yspc[] = { 8,8,8,4,4,2,2 };
+ int i,j,x,y;
+ // pass1_x[4] = 0, pass1_x[5] = 1, pass1_x[12] = 1
+ x = (a->s->img_x - xorig[p] + xspc[p]-1) / xspc[p];
+ y = (a->s->img_y - yorig[p] + yspc[p]-1) / yspc[p];
+ if (x && y) {
+ stbi__uint32 img_len = ((((a->s->img_n * x * depth) + 7) >> 3) + 1) * y;
+ if (!stbi__create_png_image_raw(a, image_data, image_data_len, out_n, x, y, depth, color)) {
+ STBI_FREE(final);
+ return 0;
+ }
+ for (j=0; j < y; ++j) {
+ for (i=0; i < x; ++i) {
+ int out_y = j*yspc[p]+yorig[p];
+ int out_x = i*xspc[p]+xorig[p];
+ memcpy(final + out_y*a->s->img_x*out_bytes + out_x*out_bytes,
+ a->out + (j*x+i)*out_bytes, out_bytes);
+ }
+ }
+ STBI_FREE(a->out);
+ image_data += img_len;
+ image_data_len -= img_len;
+ }
+ }
+ a->out = final;
+
+ return 1;
+}
+
+static int stbi__compute_transparency(stbi__png *z, stbi_uc tc[3], int out_n)
+{
+ stbi__context *s = z->s;
+ stbi__uint32 i, pixel_count = s->img_x * s->img_y;
+ stbi_uc *p = z->out;
+
+ // compute color-based transparency, assuming we've
+ // already got 255 as the alpha value in the output
+ STBI_ASSERT(out_n == 2 || out_n == 4);
+
+ if (out_n == 2) {
+ for (i=0; i < pixel_count; ++i) {
+ p[1] = (p[0] == tc[0] ? 0 : 255);
+ p += 2;
+ }
+ } else {
+ for (i=0; i < pixel_count; ++i) {
+ if (p[0] == tc[0] && p[1] == tc[1] && p[2] == tc[2])
+ p[3] = 0;
+ p += 4;
+ }
+ }
+ return 1;
+}
+
+static int stbi__compute_transparency16(stbi__png *z, stbi__uint16 tc[3], int out_n)
+{
+ stbi__context *s = z->s;
+ stbi__uint32 i, pixel_count = s->img_x * s->img_y;
+ stbi__uint16 *p = (stbi__uint16*) z->out;
+
+ // compute color-based transparency, assuming we've
+ // already got 65535 as the alpha value in the output
+ STBI_ASSERT(out_n == 2 || out_n == 4);
+
+ if (out_n == 2) {
+ for (i = 0; i < pixel_count; ++i) {
+ p[1] = (p[0] == tc[0] ? 0 : 65535);
+ p += 2;
+ }
+ } else {
+ for (i = 0; i < pixel_count; ++i) {
+ if (p[0] == tc[0] && p[1] == tc[1] && p[2] == tc[2])
+ p[3] = 0;
+ p += 4;
+ }
+ }
+ return 1;
+}
+
+static int stbi__expand_png_palette(stbi__png *a, stbi_uc *palette, int len, int pal_img_n)
+{
+ stbi__uint32 i, pixel_count = a->s->img_x * a->s->img_y;
+ stbi_uc *p, *temp_out, *orig = a->out;
+
+ p = (stbi_uc *) stbi__malloc_mad2(pixel_count, pal_img_n, 0);
+ if (p == NULL) return stbi__err("outofmem", "Out of memory");
+
+ // between here and free(out) below, exitting would leak
+ temp_out = p;
+
+ if (pal_img_n == 3) {
+ for (i=0; i < pixel_count; ++i) {
+ int n = orig[i]*4;
+ p[0] = palette[n ];
+ p[1] = palette[n+1];
+ p[2] = palette[n+2];
+ p += 3;
+ }
+ } else {
+ for (i=0; i < pixel_count; ++i) {
+ int n = orig[i]*4;
+ p[0] = palette[n ];
+ p[1] = palette[n+1];
+ p[2] = palette[n+2];
+ p[3] = palette[n+3];
+ p += 4;
+ }
+ }
+ STBI_FREE(a->out);
+ a->out = temp_out;
+
+ STBI_NOTUSED(len);
+
+ return 1;
+}
+
+static int stbi__unpremultiply_on_load = 0;
+static int stbi__de_iphone_flag = 0;
+
+STBIDEF void stbi_set_unpremultiply_on_load(int flag_true_if_should_unpremultiply)
+{
+ stbi__unpremultiply_on_load = flag_true_if_should_unpremultiply;
+}
+
+STBIDEF void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert)
+{
+ stbi__de_iphone_flag = flag_true_if_should_convert;
+}
+
+static void stbi__de_iphone(stbi__png *z)
+{
+ stbi__context *s = z->s;
+ stbi__uint32 i, pixel_count = s->img_x * s->img_y;
+ stbi_uc *p = z->out;
+
+ if (s->img_out_n == 3) { // convert bgr to rgb
+ for (i=0; i < pixel_count; ++i) {
+ stbi_uc t = p[0];
+ p[0] = p[2];
+ p[2] = t;
+ p += 3;
+ }
+ } else {
+ STBI_ASSERT(s->img_out_n == 4);
+ if (stbi__unpremultiply_on_load) {
+ // convert bgr to rgb and unpremultiply
+ for (i=0; i < pixel_count; ++i) {
+ stbi_uc a = p[3];
+ stbi_uc t = p[0];
+ if (a) {
+ p[0] = p[2] * 255 / a;
+ p[1] = p[1] * 255 / a;
+ p[2] = t * 255 / a;
+ } else {
+ p[0] = p[2];
+ p[2] = t;
+ }
+ p += 4;
+ }
+ } else {
+ // convert bgr to rgb
+ for (i=0; i < pixel_count; ++i) {
+ stbi_uc t = p[0];
+ p[0] = p[2];
+ p[2] = t;
+ p += 4;
+ }
+ }
+ }
+}
+
+#define STBI__PNG_TYPE(a,b,c,d) (((a) << 24) + ((b) << 16) + ((c) << 8) + (d))
+
+static int stbi__parse_png_file(stbi__png *z, int scan, int req_comp)
+{
+ stbi_uc palette[1024], pal_img_n=0;
+ stbi_uc has_trans=0, tc[3];
+ stbi__uint16 tc16[3];
+ stbi__uint32 ioff=0, idata_limit=0, i, pal_len=0;
+ int first=1,k,interlace=0, color=0, is_iphone=0;
+ stbi__context *s = z->s;
+
+ z->expanded = NULL;
+ z->idata = NULL;
+ z->out = NULL;
+
+ if (!stbi__check_png_header(s)) return 0;
+
+ if (scan == STBI__SCAN_type) return 1;
+
+ for (;;) {
+ stbi__pngchunk c = stbi__get_chunk_header(s);
+ switch (c.type) {
+ case STBI__PNG_TYPE('C','g','B','I'):
+ is_iphone = 1;
+ stbi__skip(s, c.length);
+ break;
+ case STBI__PNG_TYPE('I','H','D','R'): {
+ int comp,filter;
+ if (!first) return stbi__err("multiple IHDR","Corrupt PNG");
+ first = 0;
+ if (c.length != 13) return stbi__err("bad IHDR len","Corrupt PNG");
+ s->img_x = stbi__get32be(s); if (s->img_x > (1 << 24)) return stbi__err("too large","Very large image (corrupt?)");
+ s->img_y = stbi__get32be(s); if (s->img_y > (1 << 24)) return stbi__err("too large","Very large image (corrupt?)");
+ z->depth = stbi__get8(s); if (z->depth != 1 && z->depth != 2 && z->depth != 4 && z->depth != 8 && z->depth != 16) return stbi__err("1/2/4/8/16-bit only","PNG not supported: 1/2/4/8/16-bit only");
+ color = stbi__get8(s); if (color > 6) return stbi__err("bad ctype","Corrupt PNG");
+ if (color == 3 && z->depth == 16) return stbi__err("bad ctype","Corrupt PNG");
+ if (color == 3) pal_img_n = 3; else if (color & 1) return stbi__err("bad ctype","Corrupt PNG");
+ comp = stbi__get8(s); if (comp) return stbi__err("bad comp method","Corrupt PNG");
+ filter= stbi__get8(s); if (filter) return stbi__err("bad filter method","Corrupt PNG");
+ interlace = stbi__get8(s); if (interlace>1) return stbi__err("bad interlace method","Corrupt PNG");
+ if (!s->img_x || !s->img_y) return stbi__err("0-pixel image","Corrupt PNG");
+ if (!pal_img_n) {
+ s->img_n = (color & 2 ? 3 : 1) + (color & 4 ? 1 : 0);
+ if ((1 << 30) / s->img_x / s->img_n < s->img_y) return stbi__err("too large", "Image too large to decode");
+ if (scan == STBI__SCAN_header) return 1;
+ } else {
+ // if paletted, then pal_n is our final components, and
+ // img_n is # components to decompress/filter.
+ s->img_n = 1;
+ if ((1 << 30) / s->img_x / 4 < s->img_y) return stbi__err("too large","Corrupt PNG");
+ // if SCAN_header, have to scan to see if we have a tRNS
+ }
+ break;
+ }
+
+ case STBI__PNG_TYPE('P','L','T','E'): {
+ if (first) return stbi__err("first not IHDR", "Corrupt PNG");
+ if (c.length > 256*3) return stbi__err("invalid PLTE","Corrupt PNG");
+ pal_len = c.length / 3;
+ if (pal_len * 3 != c.length) return stbi__err("invalid PLTE","Corrupt PNG");
+ for (i=0; i < pal_len; ++i) {
+ palette[i*4+0] = stbi__get8(s);
+ palette[i*4+1] = stbi__get8(s);
+ palette[i*4+2] = stbi__get8(s);
+ palette[i*4+3] = 255;
+ }
+ break;
+ }
+
+ case STBI__PNG_TYPE('t','R','N','S'): {
+ if (first) return stbi__err("first not IHDR", "Corrupt PNG");
+ if (z->idata) return stbi__err("tRNS after IDAT","Corrupt PNG");
+ if (pal_img_n) {
+ if (scan == STBI__SCAN_header) { s->img_n = 4; return 1; }
+ if (pal_len == 0) return stbi__err("tRNS before PLTE","Corrupt PNG");
+ if (c.length > pal_len) return stbi__err("bad tRNS len","Corrupt PNG");
+ pal_img_n = 4;
+ for (i=0; i < c.length; ++i)
+ palette[i*4+3] = stbi__get8(s);
+ } else {
+ if (!(s->img_n & 1)) return stbi__err("tRNS with alpha","Corrupt PNG");
+ if (c.length != (stbi__uint32) s->img_n*2) return stbi__err("bad tRNS len","Corrupt PNG");
+ has_trans = 1;
+ if (z->depth == 16) {
+ for (k = 0; k < s->img_n; ++k) tc16[k] = (stbi__uint16)stbi__get16be(s); // copy the values as-is
+ } else {
+ for (k = 0; k < s->img_n; ++k) tc[k] = (stbi_uc)(stbi__get16be(s) & 255) * stbi__depth_scale_table[z->depth]; // non 8-bit images will be larger
+ }
+ }
+ break;
+ }
+
+ case STBI__PNG_TYPE('I','D','A','T'): {
+ if (first) return stbi__err("first not IHDR", "Corrupt PNG");
+ if (pal_img_n && !pal_len) return stbi__err("no PLTE","Corrupt PNG");
+ if (scan == STBI__SCAN_header) { s->img_n = pal_img_n; return 1; }
+ if ((int)(ioff + c.length) < (int)ioff) return 0;
+ if (ioff + c.length > idata_limit) {
+ stbi__uint32 idata_limit_old = idata_limit;
+ stbi_uc *p;
+ if (idata_limit == 0) idata_limit = c.length > 4096 ? c.length : 4096;
+ while (ioff + c.length > idata_limit)
+ idata_limit *= 2;
+ STBI_NOTUSED(idata_limit_old);
+ p = (stbi_uc *) STBI_REALLOC_SIZED(z->idata, idata_limit_old, idata_limit); if (p == NULL) return stbi__err("outofmem", "Out of memory");
+ z->idata = p;
+ }
+ if (!stbi__getn(s, z->idata+ioff,c.length)) return stbi__err("outofdata","Corrupt PNG");
+ ioff += c.length;
+ break;
+ }
+
+ case STBI__PNG_TYPE('I','E','N','D'): {
+ stbi__uint32 raw_len, bpl;
+ if (first) return stbi__err("first not IHDR", "Corrupt PNG");
+ if (scan != STBI__SCAN_load) return 1;
+ if (z->idata == NULL) return stbi__err("no IDAT","Corrupt PNG");
+ // initial guess for decoded data size to avoid unnecessary reallocs
+ bpl = (s->img_x * z->depth + 7) / 8; // bytes per line, per component
+ raw_len = bpl * s->img_y * s->img_n /* pixels */ + s->img_y /* filter mode per row */;
+ z->expanded = (stbi_uc *) stbi_zlib_decode_malloc_guesssize_headerflag((char *) z->idata, ioff, raw_len, (int *) &raw_len, !is_iphone);
+ if (z->expanded == NULL) return 0; // zlib should set error
+ STBI_FREE(z->idata); z->idata = NULL;
+ if ((req_comp == s->img_n+1 && req_comp != 3 && !pal_img_n) || has_trans)
+ s->img_out_n = s->img_n+1;
+ else
+ s->img_out_n = s->img_n;
+ if (!stbi__create_png_image(z, z->expanded, raw_len, s->img_out_n, z->depth, color, interlace)) return 0;
+ if (has_trans) {
+ if (z->depth == 16) {
+ if (!stbi__compute_transparency16(z, tc16, s->img_out_n)) return 0;
+ } else {
+ if (!stbi__compute_transparency(z, tc, s->img_out_n)) return 0;
+ }
+ }
+ if (is_iphone && stbi__de_iphone_flag && s->img_out_n > 2)
+ stbi__de_iphone(z);
+ if (pal_img_n) {
+ // pal_img_n == 3 or 4
+ s->img_n = pal_img_n; // record the actual colors we had
+ s->img_out_n = pal_img_n;
+ if (req_comp >= 3) s->img_out_n = req_comp;
+ if (!stbi__expand_png_palette(z, palette, pal_len, s->img_out_n))
+ return 0;
+ }
+ STBI_FREE(z->expanded); z->expanded = NULL;
+ return 1;
+ }
+
+ default:
+ // if critical, fail
+ if (first) return stbi__err("first not IHDR", "Corrupt PNG");
+ if ((c.type & (1 << 29)) == 0) {
+ #ifndef STBI_NO_FAILURE_STRINGS
+ // not threadsafe
+ static char invalid_chunk[] = "XXXX PNG chunk not known";
+ invalid_chunk[0] = STBI__BYTECAST(c.type >> 24);
+ invalid_chunk[1] = STBI__BYTECAST(c.type >> 16);
+ invalid_chunk[2] = STBI__BYTECAST(c.type >> 8);
+ invalid_chunk[3] = STBI__BYTECAST(c.type >> 0);
+ #endif
+ return stbi__err(invalid_chunk, "PNG not supported: unknown PNG chunk type");
+ }
+ stbi__skip(s, c.length);
+ break;
+ }
+ // end of PNG chunk, read and skip CRC
+ stbi__get32be(s);
+ }
+}
+
+static void *stbi__do_png(stbi__png *p, int *x, int *y, int *n, int req_comp, stbi__result_info *ri)
+{
+ void *result=NULL;
+ if (req_comp < 0 || req_comp > 4) return stbi__errpuc("bad req_comp", "Internal error");
+ if (stbi__parse_png_file(p, STBI__SCAN_load, req_comp)) {
+ if (p->depth < 8)
+ ri->bits_per_channel = 8;
+ else
+ ri->bits_per_channel = p->depth;
+ result = p->out;
+ p->out = NULL;
+ if (req_comp && req_comp != p->s->img_out_n) {
+ if (ri->bits_per_channel == 8)
+ result = stbi__convert_format((unsigned char *) result, p->s->img_out_n, req_comp, p->s->img_x, p->s->img_y);
+ else
+ result = stbi__convert_format16((stbi__uint16 *) result, p->s->img_out_n, req_comp, p->s->img_x, p->s->img_y);
+ p->s->img_out_n = req_comp;
+ if (result == NULL) return result;
+ }
+ *x = p->s->img_x;
+ *y = p->s->img_y;
+ if (n) *n = p->s->img_n;
+ }
+ STBI_FREE(p->out); p->out = NULL;
+ STBI_FREE(p->expanded); p->expanded = NULL;
+ STBI_FREE(p->idata); p->idata = NULL;
+
+ return result;
+}
+
+static void *stbi__png_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+ stbi__png p;
+ p.s = s;
+ return stbi__do_png(&p, x,y,comp,req_comp, ri);
+}
+
+static int stbi__png_test(stbi__context *s)
+{
+ int r;
+ r = stbi__check_png_header(s);
+ stbi__rewind(s);
+ return r;
+}
+
+static int stbi__png_info_raw(stbi__png *p, int *x, int *y, int *comp)
+{
+ if (!stbi__parse_png_file(p, STBI__SCAN_header, 0)) {
+ stbi__rewind( p->s );
+ return 0;
+ }
+ if (x) *x = p->s->img_x;
+ if (y) *y = p->s->img_y;
+ if (comp) *comp = p->s->img_n;
+ return 1;
+}
+
+static int stbi__png_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ stbi__png p;
+ p.s = s;
+ return stbi__png_info_raw(&p, x, y, comp);
+}
+#endif
+
+// Microsoft/Windows BMP image
+
+#ifndef STBI_NO_BMP
+static int stbi__bmp_test_raw(stbi__context *s)
+{
+ int r;
+ int sz;
+ if (stbi__get8(s) != 'B') return 0;
+ if (stbi__get8(s) != 'M') return 0;
+ stbi__get32le(s); // discard filesize
+ stbi__get16le(s); // discard reserved
+ stbi__get16le(s); // discard reserved
+ stbi__get32le(s); // discard data offset
+ sz = stbi__get32le(s);
+ r = (sz == 12 || sz == 40 || sz == 56 || sz == 108 || sz == 124);
+ return r;
+}
+
+static int stbi__bmp_test(stbi__context *s)
+{
+ int r = stbi__bmp_test_raw(s);
+ stbi__rewind(s);
+ return r;
+}
+
+
+// returns 0..31 for the highest set bit
+static int stbi__high_bit(unsigned int z)
+{
+ int n=0;
+ if (z == 0) return -1;
+ if (z >= 0x10000) n += 16, z >>= 16;
+ if (z >= 0x00100) n += 8, z >>= 8;
+ if (z >= 0x00010) n += 4, z >>= 4;
+ if (z >= 0x00004) n += 2, z >>= 2;
+ if (z >= 0x00002) n += 1, z >>= 1;
+ return n;
+}
+
+static int stbi__bitcount(unsigned int a)
+{
+ a = (a & 0x55555555) + ((a >> 1) & 0x55555555); // max 2
+ a = (a & 0x33333333) + ((a >> 2) & 0x33333333); // max 4
+ a = (a + (a >> 4)) & 0x0f0f0f0f; // max 8 per 4, now 8 bits
+ a = (a + (a >> 8)); // max 16 per 8 bits
+ a = (a + (a >> 16)); // max 32 per 8 bits
+ return a & 0xff;
+}
+
+static int stbi__shiftsigned(int v, int shift, int bits)
+{
+ int result;
+ int z=0;
+
+ if (shift < 0) v <<= -shift;
+ else v >>= shift;
+ result = v;
+
+ z = bits;
+ while (z < 8) {
+ result += v >> z;
+ z += bits;
+ }
+ return result;
+}
+
+typedef struct
+{
+ int bpp, offset, hsz;
+ unsigned int mr,mg,mb,ma, all_a;
+} stbi__bmp_data;
+
+static void *stbi__bmp_parse_header(stbi__context *s, stbi__bmp_data *info)
+{
+ int hsz;
+ if (stbi__get8(s) != 'B' || stbi__get8(s) != 'M') return stbi__errpuc("not BMP", "Corrupt BMP");
+ stbi__get32le(s); // discard filesize
+ stbi__get16le(s); // discard reserved
+ stbi__get16le(s); // discard reserved
+ info->offset = stbi__get32le(s);
+ info->hsz = hsz = stbi__get32le(s);
+ info->mr = info->mg = info->mb = info->ma = 0;
+
+ if (hsz != 12 && hsz != 40 && hsz != 56 && hsz != 108 && hsz != 124) return stbi__errpuc("unknown BMP", "BMP type not supported: unknown");
+ if (hsz == 12) {
+ s->img_x = stbi__get16le(s);
+ s->img_y = stbi__get16le(s);
+ } else {
+ s->img_x = stbi__get32le(s);
+ s->img_y = stbi__get32le(s);
+ }
+ if (stbi__get16le(s) != 1) return stbi__errpuc("bad BMP", "bad BMP");
+ info->bpp = stbi__get16le(s);
+ if (info->bpp == 1) return stbi__errpuc("monochrome", "BMP type not supported: 1-bit");
+ if (hsz != 12) {
+ int compress = stbi__get32le(s);
+ if (compress == 1 || compress == 2) return stbi__errpuc("BMP RLE", "BMP type not supported: RLE");
+ stbi__get32le(s); // discard sizeof
+ stbi__get32le(s); // discard hres
+ stbi__get32le(s); // discard vres
+ stbi__get32le(s); // discard colorsused
+ stbi__get32le(s); // discard max important
+ if (hsz == 40 || hsz == 56) {
+ if (hsz == 56) {
+ stbi__get32le(s);
+ stbi__get32le(s);
+ stbi__get32le(s);
+ stbi__get32le(s);
+ }
+ if (info->bpp == 16 || info->bpp == 32) {
+ if (compress == 0) {
+ if (info->bpp == 32) {
+ info->mr = 0xffu << 16;
+ info->mg = 0xffu << 8;
+ info->mb = 0xffu << 0;
+ info->ma = 0xffu << 24;
+ info->all_a = 0; // if all_a is 0 at end, then we loaded alpha channel but it was all 0
+ } else {
+ info->mr = 31u << 10;
+ info->mg = 31u << 5;
+ info->mb = 31u << 0;
+ }
+ } else if (compress == 3) {
+ info->mr = stbi__get32le(s);
+ info->mg = stbi__get32le(s);
+ info->mb = stbi__get32le(s);
+ // not documented, but generated by photoshop and handled by mspaint
+ if (info->mr == info->mg && info->mg == info->mb) {
+ // ?!?!?
+ return stbi__errpuc("bad BMP", "bad BMP");
+ }
+ } else
+ return stbi__errpuc("bad BMP", "bad BMP");
+ }
+ } else {
+ int i;
+ if (hsz != 108 && hsz != 124)
+ return stbi__errpuc("bad BMP", "bad BMP");
+ info->mr = stbi__get32le(s);
+ info->mg = stbi__get32le(s);
+ info->mb = stbi__get32le(s);
+ info->ma = stbi__get32le(s);
+ stbi__get32le(s); // discard color space
+ for (i=0; i < 12; ++i)
+ stbi__get32le(s); // discard color space parameters
+ if (hsz == 124) {
+ stbi__get32le(s); // discard rendering intent
+ stbi__get32le(s); // discard offset of profile data
+ stbi__get32le(s); // discard size of profile data
+ stbi__get32le(s); // discard reserved
+ }
+ }
+ }
+ return (void *) 1;
+}
+
+
+static void *stbi__bmp_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+ stbi_uc *out;
+ unsigned int mr=0,mg=0,mb=0,ma=0, all_a;
+ stbi_uc pal[256][4];
+ int psize=0,i,j,width;
+ int flip_vertically, pad, target;
+ stbi__bmp_data info;
+ STBI_NOTUSED(ri);
+
+ info.all_a = 255;
+ if (stbi__bmp_parse_header(s, &info) == NULL)
+ return NULL; // error code already set
+
+ flip_vertically = ((int) s->img_y) > 0;
+ s->img_y = abs((int) s->img_y);
+
+ mr = info.mr;
+ mg = info.mg;
+ mb = info.mb;
+ ma = info.ma;
+ all_a = info.all_a;
+
+ if (info.hsz == 12) {
+ if (info.bpp < 24)
+ psize = (info.offset - 14 - 24) / 3;
+ } else {
+ if (info.bpp < 16)
+ psize = (info.offset - 14 - info.hsz) >> 2;
+ }
+
+ s->img_n = ma ? 4 : 3;
+ if (req_comp && req_comp >= 3) // we can directly decode 3 or 4
+ target = req_comp;
+ else
+ target = s->img_n; // if they want monochrome, we'll post-convert
+
+ // sanity-check size
+ if (!stbi__mad3sizes_valid(target, s->img_x, s->img_y, 0))
+ return stbi__errpuc("too large", "Corrupt BMP");
+
+ out = (stbi_uc *) stbi__malloc_mad3(target, s->img_x, s->img_y, 0);
+ if (!out) return stbi__errpuc("outofmem", "Out of memory");
+ if (info.bpp < 16) {
+ int z=0;
+ if (psize == 0 || psize > 256) { STBI_FREE(out); return stbi__errpuc("invalid", "Corrupt BMP"); }
+ for (i=0; i < psize; ++i) {
+ pal[i][2] = stbi__get8(s);
+ pal[i][1] = stbi__get8(s);
+ pal[i][0] = stbi__get8(s);
+ if (info.hsz != 12) stbi__get8(s);
+ pal[i][3] = 255;
+ }
+ stbi__skip(s, info.offset - 14 - info.hsz - psize * (info.hsz == 12 ? 3 : 4));
+ if (info.bpp == 4) width = (s->img_x + 1) >> 1;
+ else if (info.bpp == 8) width = s->img_x;
+ else { STBI_FREE(out); return stbi__errpuc("bad bpp", "Corrupt BMP"); }
+ pad = (-width)&3;
+ for (j=0; j < (int) s->img_y; ++j) {
+ for (i=0; i < (int) s->img_x; i += 2) {
+ int v=stbi__get8(s),v2=0;
+ if (info.bpp == 4) {
+ v2 = v & 15;
+ v >>= 4;
+ }
+ out[z++] = pal[v][0];
+ out[z++] = pal[v][1];
+ out[z++] = pal[v][2];
+ if (target == 4) out[z++] = 255;
+ if (i+1 == (int) s->img_x) break;
+ v = (info.bpp == 8) ? stbi__get8(s) : v2;
+ out[z++] = pal[v][0];
+ out[z++] = pal[v][1];
+ out[z++] = pal[v][2];
+ if (target == 4) out[z++] = 255;
+ }
+ stbi__skip(s, pad);
+ }
+ } else {
+ int rshift=0,gshift=0,bshift=0,ashift=0,rcount=0,gcount=0,bcount=0,acount=0;
+ int z = 0;
+ int easy=0;
+ stbi__skip(s, info.offset - 14 - info.hsz);
+ if (info.bpp == 24) width = 3 * s->img_x;
+ else if (info.bpp == 16) width = 2*s->img_x;
+ else /* bpp = 32 and pad = 0 */ width=0;
+ pad = (-width) & 3;
+ if (info.bpp == 24) {
+ easy = 1;
+ } else if (info.bpp == 32) {
+ if (mb == 0xff && mg == 0xff00 && mr == 0x00ff0000 && ma == 0xff000000)
+ easy = 2;
+ }
+ if (!easy) {
+ if (!mr || !mg || !mb) { STBI_FREE(out); return stbi__errpuc("bad masks", "Corrupt BMP"); }
+ // right shift amt to put high bit in position #7
+ rshift = stbi__high_bit(mr)-7; rcount = stbi__bitcount(mr);
+ gshift = stbi__high_bit(mg)-7; gcount = stbi__bitcount(mg);
+ bshift = stbi__high_bit(mb)-7; bcount = stbi__bitcount(mb);
+ ashift = stbi__high_bit(ma)-7; acount = stbi__bitcount(ma);
+ }
+ for (j=0; j < (int) s->img_y; ++j) {
+ if (easy) {
+ for (i=0; i < (int) s->img_x; ++i) {
+ unsigned char a;
+ out[z+2] = stbi__get8(s);
+ out[z+1] = stbi__get8(s);
+ out[z+0] = stbi__get8(s);
+ z += 3;
+ a = (easy == 2 ? stbi__get8(s) : 255);
+ all_a |= a;
+ if (target == 4) out[z++] = a;
+ }
+ } else {
+ int bpp = info.bpp;
+ for (i=0; i < (int) s->img_x; ++i) {
+ stbi__uint32 v = (bpp == 16 ? (stbi__uint32) stbi__get16le(s) : stbi__get32le(s));
+ int a;
+ out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mr, rshift, rcount));
+ out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mg, gshift, gcount));
+ out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mb, bshift, bcount));
+ a = (ma ? stbi__shiftsigned(v & ma, ashift, acount) : 255);
+ all_a |= a;
+ if (target == 4) out[z++] = STBI__BYTECAST(a);
+ }
+ }
+ stbi__skip(s, pad);
+ }
+ }
+
+ // if alpha channel is all 0s, replace with all 255s
+ if (target == 4 && all_a == 0)
+ for (i=4*s->img_x*s->img_y-1; i >= 0; i -= 4)
+ out[i] = 255;
+
+ if (flip_vertically) {
+ stbi_uc t;
+ for (j=0; j < (int) s->img_y>>1; ++j) {
+ stbi_uc *p1 = out + j *s->img_x*target;
+ stbi_uc *p2 = out + (s->img_y-1-j)*s->img_x*target;
+ for (i=0; i < (int) s->img_x*target; ++i) {
+ t = p1[i], p1[i] = p2[i], p2[i] = t;
+ }
+ }
+ }
+
+ if (req_comp && req_comp != target) {
+ out = stbi__convert_format(out, target, req_comp, s->img_x, s->img_y);
+ if (out == NULL) return out; // stbi__convert_format frees input on failure
+ }
+
+ *x = s->img_x;
+ *y = s->img_y;
+ if (comp) *comp = s->img_n;
+ return out;
+}
+#endif
+
+// Targa Truevision - TGA
+// by Jonathan Dummer
+#ifndef STBI_NO_TGA
+// returns STBI_rgb or whatever, 0 on error
+static int stbi__tga_get_comp(int bits_per_pixel, int is_grey, int* is_rgb16)
+{
+ // only RGB or RGBA (incl. 16bit) or grey allowed
+ if(is_rgb16) *is_rgb16 = 0;
+ switch(bits_per_pixel) {
+ case 8: return STBI_grey;
+ case 16: if(is_grey) return STBI_grey_alpha;
+ // else: fall-through
+ case 15: if(is_rgb16) *is_rgb16 = 1;
+ return STBI_rgb;
+ case 24: // fall-through
+ case 32: return bits_per_pixel/8;
+ default: return 0;
+ }
+}
+
+static int stbi__tga_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ int tga_w, tga_h, tga_comp, tga_image_type, tga_bits_per_pixel, tga_colormap_bpp;
+ int sz, tga_colormap_type;
+ stbi__get8(s); // discard Offset
+ tga_colormap_type = stbi__get8(s); // colormap type
+ if( tga_colormap_type > 1 ) {
+ stbi__rewind(s);
+ return 0; // only RGB or indexed allowed
+ }
+ tga_image_type = stbi__get8(s); // image type
+ if ( tga_colormap_type == 1 ) { // colormapped (paletted) image
+ if (tga_image_type != 1 && tga_image_type != 9) {
+ stbi__rewind(s);
+ return 0;
+ }
+ stbi__skip(s,4); // skip index of first colormap entry and number of entries
+ sz = stbi__get8(s); // check bits per palette color entry
+ if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) {
+ stbi__rewind(s);
+ return 0;
+ }
+ stbi__skip(s,4); // skip image x and y origin
+ tga_colormap_bpp = sz;
+ } else { // "normal" image w/o colormap - only RGB or grey allowed, +/- RLE
+ if ( (tga_image_type != 2) && (tga_image_type != 3) && (tga_image_type != 10) && (tga_image_type != 11) ) {
+ stbi__rewind(s);
+ return 0; // only RGB or grey allowed, +/- RLE
+ }
+ stbi__skip(s,9); // skip colormap specification and image x/y origin
+ tga_colormap_bpp = 0;
+ }
+ tga_w = stbi__get16le(s);
+ if( tga_w < 1 ) {
+ stbi__rewind(s);
+ return 0; // test width
+ }
+ tga_h = stbi__get16le(s);
+ if( tga_h < 1 ) {
+ stbi__rewind(s);
+ return 0; // test height
+ }
+ tga_bits_per_pixel = stbi__get8(s); // bits per pixel
+ stbi__get8(s); // ignore alpha bits
+ if (tga_colormap_bpp != 0) {
+ if((tga_bits_per_pixel != 8) && (tga_bits_per_pixel != 16)) {
+ // when using a colormap, tga_bits_per_pixel is the size of the indexes
+ // I don't think anything but 8 or 16bit indexes makes sense
+ stbi__rewind(s);
+ return 0;
+ }
+ tga_comp = stbi__tga_get_comp(tga_colormap_bpp, 0, NULL);
+ } else {
+ tga_comp = stbi__tga_get_comp(tga_bits_per_pixel, (tga_image_type == 3) || (tga_image_type == 11), NULL);
+ }
+ if(!tga_comp) {
+ stbi__rewind(s);
+ return 0;
+ }
+ if (x) *x = tga_w;
+ if (y) *y = tga_h;
+ if (comp) *comp = tga_comp;
+ return 1; // seems to have passed everything
+}
+
+static int stbi__tga_test(stbi__context *s)
+{
+ int res = 0;
+ int sz, tga_color_type;
+ stbi__get8(s); // discard Offset
+ tga_color_type = stbi__get8(s); // color type
+ if ( tga_color_type > 1 ) goto errorEnd; // only RGB or indexed allowed
+ sz = stbi__get8(s); // image type
+ if ( tga_color_type == 1 ) { // colormapped (paletted) image
+ if (sz != 1 && sz != 9) goto errorEnd; // colortype 1 demands image type 1 or 9
+ stbi__skip(s,4); // skip index of first colormap entry and number of entries
+ sz = stbi__get8(s); // check bits per palette color entry
+ if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) goto errorEnd;
+ stbi__skip(s,4); // skip image x and y origin
+ } else { // "normal" image w/o colormap
+ if ( (sz != 2) && (sz != 3) && (sz != 10) && (sz != 11) ) goto errorEnd; // only RGB or grey allowed, +/- RLE
+ stbi__skip(s,9); // skip colormap specification and image x/y origin
+ }
+ if ( stbi__get16le(s) < 1 ) goto errorEnd; // test width
+ if ( stbi__get16le(s) < 1 ) goto errorEnd; // test height
+ sz = stbi__get8(s); // bits per pixel
+ if ( (tga_color_type == 1) && (sz != 8) && (sz != 16) ) goto errorEnd; // for colormapped images, bpp is size of an index
+ if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) goto errorEnd;
+
+ res = 1; // if we got this far, everything's good and we can return 1 instead of 0
+
+errorEnd:
+ stbi__rewind(s);
+ return res;
+}
+
+// read 16bit value and convert to 24bit RGB
+static void stbi__tga_read_rgb16(stbi__context *s, stbi_uc* out)
+{
+ stbi__uint16 px = (stbi__uint16)stbi__get16le(s);
+ stbi__uint16 fiveBitMask = 31;
+ // we have 3 channels with 5bits each
+ int r = (px >> 10) & fiveBitMask;
+ int g = (px >> 5) & fiveBitMask;
+ int b = px & fiveBitMask;
+ // Note that this saves the data in RGB(A) order, so it doesn't need to be swapped later
+ out[0] = (stbi_uc)((r * 255)/31);
+ out[1] = (stbi_uc)((g * 255)/31);
+ out[2] = (stbi_uc)((b * 255)/31);
+
+ // some people claim that the most significant bit might be used for alpha
+ // (possibly if an alpha-bit is set in the "image descriptor byte")
+ // but that only made 16bit test images completely translucent..
+ // so let's treat all 15 and 16bit TGAs as RGB with no alpha.
+}
+
+static void *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+ // read in the TGA header stuff
+ int tga_offset = stbi__get8(s);
+ int tga_indexed = stbi__get8(s);
+ int tga_image_type = stbi__get8(s);
+ int tga_is_RLE = 0;
+ int tga_palette_start = stbi__get16le(s);
+ int tga_palette_len = stbi__get16le(s);
+ int tga_palette_bits = stbi__get8(s);
+ int tga_x_origin = stbi__get16le(s);
+ int tga_y_origin = stbi__get16le(s);
+ int tga_width = stbi__get16le(s);
+ int tga_height = stbi__get16le(s);
+ int tga_bits_per_pixel = stbi__get8(s);
+ int tga_comp, tga_rgb16=0;
+ int tga_inverted = stbi__get8(s);
+ // int tga_alpha_bits = tga_inverted & 15; // the 4 lowest bits - unused (useless?)
+ // image data
+ unsigned char *tga_data;
+ unsigned char *tga_palette = NULL;
+ int i, j;
+ unsigned char raw_data[4] = {0};
+ int RLE_count = 0;
+ int RLE_repeating = 0;
+ int read_next_pixel = 1;
+ STBI_NOTUSED(ri);
+
+ // do a tiny bit of precessing
+ if ( tga_image_type >= 8 )
+ {
+ tga_image_type -= 8;
+ tga_is_RLE = 1;
+ }
+ tga_inverted = 1 - ((tga_inverted >> 5) & 1);
+
+ // If I'm paletted, then I'll use the number of bits from the palette
+ if ( tga_indexed ) tga_comp = stbi__tga_get_comp(tga_palette_bits, 0, &tga_rgb16);
+ else tga_comp = stbi__tga_get_comp(tga_bits_per_pixel, (tga_image_type == 3), &tga_rgb16);
+
+ if(!tga_comp) // shouldn't really happen, stbi__tga_test() should have ensured basic consistency
+ return stbi__errpuc("bad format", "Can't find out TGA pixelformat");
+
+ // tga info
+ *x = tga_width;
+ *y = tga_height;
+ if (comp) *comp = tga_comp;
+
+ if (!stbi__mad3sizes_valid(tga_width, tga_height, tga_comp, 0))
+ return stbi__errpuc("too large", "Corrupt TGA");
+
+ tga_data = (unsigned char*)stbi__malloc_mad3(tga_width, tga_height, tga_comp, 0);
+ if (!tga_data) return stbi__errpuc("outofmem", "Out of memory");
+
+ // skip to the data's starting position (offset usually = 0)
+ stbi__skip(s, tga_offset );
+
+ if ( !tga_indexed && !tga_is_RLE && !tga_rgb16 ) {
+ for (i=0; i < tga_height; ++i) {
+ int row = tga_inverted ? tga_height -i - 1 : i;
+ stbi_uc *tga_row = tga_data + row*tga_width*tga_comp;
+ stbi__getn(s, tga_row, tga_width * tga_comp);
+ }
+ } else {
+ // do I need to load a palette?
+ if ( tga_indexed)
+ {
+ // any data to skip? (offset usually = 0)
+ stbi__skip(s, tga_palette_start );
+ // load the palette
+ tga_palette = (unsigned char*)stbi__malloc_mad2(tga_palette_len, tga_comp, 0);
+ if (!tga_palette) {
+ STBI_FREE(tga_data);
+ return stbi__errpuc("outofmem", "Out of memory");
+ }
+ if (tga_rgb16) {
+ stbi_uc *pal_entry = tga_palette;
+ STBI_ASSERT(tga_comp == STBI_rgb);
+ for (i=0; i < tga_palette_len; ++i) {
+ stbi__tga_read_rgb16(s, pal_entry);
+ pal_entry += tga_comp;
+ }
+ } else if (!stbi__getn(s, tga_palette, tga_palette_len * tga_comp)) {
+ STBI_FREE(tga_data);
+ STBI_FREE(tga_palette);
+ return stbi__errpuc("bad palette", "Corrupt TGA");
+ }
+ }
+ // load the data
+ for (i=0; i < tga_width * tga_height; ++i)
+ {
+ // if I'm in RLE mode, do I need to get a RLE stbi__pngchunk?
+ if ( tga_is_RLE )
+ {
+ if ( RLE_count == 0 )
+ {
+ // yep, get the next byte as a RLE command
+ int RLE_cmd = stbi__get8(s);
+ RLE_count = 1 + (RLE_cmd & 127);
+ RLE_repeating = RLE_cmd >> 7;
+ read_next_pixel = 1;
+ } else if ( !RLE_repeating )
+ {
+ read_next_pixel = 1;
+ }
+ } else
+ {
+ read_next_pixel = 1;
+ }
+ // OK, if I need to read a pixel, do it now
+ if ( read_next_pixel )
+ {
+ // load however much data we did have
+ if ( tga_indexed )
+ {
+ // read in index, then perform the lookup
+ int pal_idx = (tga_bits_per_pixel == 8) ? stbi__get8(s) : stbi__get16le(s);
+ if ( pal_idx >= tga_palette_len ) {
+ // invalid index
+ pal_idx = 0;
+ }
+ pal_idx *= tga_comp;
+ for (j = 0; j < tga_comp; ++j) {
+ raw_data[j] = tga_palette[pal_idx+j];
+ }
+ } else if(tga_rgb16) {
+ STBI_ASSERT(tga_comp == STBI_rgb);
+ stbi__tga_read_rgb16(s, raw_data);
+ } else {
+ // read in the data raw
+ for (j = 0; j < tga_comp; ++j) {
+ raw_data[j] = stbi__get8(s);
+ }
+ }
+ // clear the reading flag for the next pixel
+ read_next_pixel = 0;
+ } // end of reading a pixel
+
+ // copy data
+ for (j = 0; j < tga_comp; ++j)
+ tga_data[i*tga_comp+j] = raw_data[j];
+
+ // in case we're in RLE mode, keep counting down
+ --RLE_count;
+ }
+ // do I need to invert the image?
+ if ( tga_inverted )
+ {
+ for (j = 0; j*2 < tga_height; ++j)
+ {
+ int index1 = j * tga_width * tga_comp;
+ int index2 = (tga_height - 1 - j) * tga_width * tga_comp;
+ for (i = tga_width * tga_comp; i > 0; --i)
+ {
+ unsigned char temp = tga_data[index1];
+ tga_data[index1] = tga_data[index2];
+ tga_data[index2] = temp;
+ ++index1;
+ ++index2;
+ }
+ }
+ }
+ // clear my palette, if I had one
+ if ( tga_palette != NULL )
+ {
+ STBI_FREE( tga_palette );
+ }
+ }
+
+ // swap RGB - if the source data was RGB16, it already is in the right order
+ if (tga_comp >= 3 && !tga_rgb16)
+ {
+ unsigned char* tga_pixel = tga_data;
+ for (i=0; i < tga_width * tga_height; ++i)
+ {
+ unsigned char temp = tga_pixel[0];
+ tga_pixel[0] = tga_pixel[2];
+ tga_pixel[2] = temp;
+ tga_pixel += tga_comp;
+ }
+ }
+
+ // convert to target component count
+ if (req_comp && req_comp != tga_comp)
+ tga_data = stbi__convert_format(tga_data, tga_comp, req_comp, tga_width, tga_height);
+
+ // the things I do to get rid of an error message, and yet keep
+ // Microsoft's C compilers happy... [8^(
+ tga_palette_start = tga_palette_len = tga_palette_bits =
+ tga_x_origin = tga_y_origin = 0;
+ // OK, done
+ return tga_data;
+}
+#endif
+
+// *************************************************************************************************
+// Photoshop PSD loader -- PD by Thatcher Ulrich, integration by Nicolas Schulz, tweaked by STB
+
+#ifndef STBI_NO_PSD
+static int stbi__psd_test(stbi__context *s)
+{
+ int r = (stbi__get32be(s) == 0x38425053);
+ stbi__rewind(s);
+ return r;
+}
+
+static int stbi__psd_decode_rle(stbi__context *s, stbi_uc *p, int pixelCount)
+{
+ int count, nleft, len;
+
+ count = 0;
+ while ((nleft = pixelCount - count) > 0) {
+ len = stbi__get8(s);
+ if (len == 128) {
+ // No-op.
+ } else if (len < 128) {
+ // Copy next len+1 bytes literally.
+ len++;
+ if (len > nleft) return 0; // corrupt data
+ count += len;
+ while (len) {
+ *p = stbi__get8(s);
+ p += 4;
+ len--;
+ }
+ } else if (len > 128) {
+ stbi_uc val;
+ // Next -len+1 bytes in the dest are replicated from next source byte.
+ // (Interpret len as a negative 8-bit int.)
+ len = 257 - len;
+ if (len > nleft) return 0; // corrupt data
+ val = stbi__get8(s);
+ count += len;
+ while (len) {
+ *p = val;
+ p += 4;
+ len--;
+ }
+ }
+ }
+
+ return 1;
+}
+
+static void *stbi__psd_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri, int bpc)
+{
+ int pixelCount;
+ int channelCount, compression;
+ int channel, i;
+ int bitdepth;
+ int w,h;
+ stbi_uc *out;
+ STBI_NOTUSED(ri);
+
+ // Check identifier
+ if (stbi__get32be(s) != 0x38425053) // "8BPS"
+ return stbi__errpuc("not PSD", "Corrupt PSD image");
+
+ // Check file type version.
+ if (stbi__get16be(s) != 1)
+ return stbi__errpuc("wrong version", "Unsupported version of PSD image");
+
+ // Skip 6 reserved bytes.
+ stbi__skip(s, 6 );
+
+ // Read the number of channels (R, G, B, A, etc).
+ channelCount = stbi__get16be(s);
+ if (channelCount < 0 || channelCount > 16)
+ return stbi__errpuc("wrong channel count", "Unsupported number of channels in PSD image");
+
+ // Read the rows and columns of the image.
+ h = stbi__get32be(s);
+ w = stbi__get32be(s);
+
+ // Make sure the depth is 8 bits.
+ bitdepth = stbi__get16be(s);
+ if (bitdepth != 8 && bitdepth != 16)
+ return stbi__errpuc("unsupported bit depth", "PSD bit depth is not 8 or 16 bit");
+
+ // Make sure the color mode is RGB.
+ // Valid options are:
+ // 0: Bitmap
+ // 1: Grayscale
+ // 2: Indexed color
+ // 3: RGB color
+ // 4: CMYK color
+ // 7: Multichannel
+ // 8: Duotone
+ // 9: Lab color
+ if (stbi__get16be(s) != 3)
+ return stbi__errpuc("wrong color format", "PSD is not in RGB color format");
+
+ // Skip the Mode Data. (It's the palette for indexed color; other info for other modes.)
+ stbi__skip(s,stbi__get32be(s) );
+
+ // Skip the image resources. (resolution, pen tool paths, etc)
+ stbi__skip(s, stbi__get32be(s) );
+
+ // Skip the reserved data.
+ stbi__skip(s, stbi__get32be(s) );
+
+ // Find out if the data is compressed.
+ // Known values:
+ // 0: no compression
+ // 1: RLE compressed
+ compression = stbi__get16be(s);
+ if (compression > 1)
+ return stbi__errpuc("bad compression", "PSD has an unknown compression format");
+
+ // Check size
+ if (!stbi__mad3sizes_valid(4, w, h, 0))
+ return stbi__errpuc("too large", "Corrupt PSD");
+
+ // Create the destination image.
+
+ if (!compression && bitdepth == 16 && bpc == 16) {
+ out = (stbi_uc *) stbi__malloc_mad3(8, w, h, 0);
+ ri->bits_per_channel = 16;
+ } else
+ out = (stbi_uc *) stbi__malloc(4 * w*h);
+
+ if (!out) return stbi__errpuc("outofmem", "Out of memory");
+ pixelCount = w*h;
+
+ // Initialize the data to zero.
+ //memset( out, 0, pixelCount * 4 );
+
+ // Finally, the image data.
+ if (compression) {
+ // RLE as used by .PSD and .TIFF
+ // Loop until you get the number of unpacked bytes you are expecting:
+ // Read the next source byte into n.
+ // If n is between 0 and 127 inclusive, copy the next n+1 bytes literally.
+ // Else if n is between -127 and -1 inclusive, copy the next byte -n+1 times.
+ // Else if n is 128, noop.
+ // Endloop
+
+ // The RLE-compressed data is preceeded by a 2-byte data count for each row in the data,
+ // which we're going to just skip.
+ stbi__skip(s, h * channelCount * 2 );
+
+ // Read the RLE data by channel.
+ for (channel = 0; channel < 4; channel++) {
+ stbi_uc *p;
+
+ p = out+channel;
+ if (channel >= channelCount) {
+ // Fill this channel with default data.
+ for (i = 0; i < pixelCount; i++, p += 4)
+ *p = (channel == 3 ? 255 : 0);
+ } else {
+ // Read the RLE data.
+ if (!stbi__psd_decode_rle(s, p, pixelCount)) {
+ STBI_FREE(out);
+ return stbi__errpuc("corrupt", "bad RLE data");
+ }
+ }
+ }
+
+ } else {
+ // We're at the raw image data. It's each channel in order (Red, Green, Blue, Alpha, ...)
+ // where each channel consists of an 8-bit (or 16-bit) value for each pixel in the image.
+
+ // Read the data by channel.
+ for (channel = 0; channel < 4; channel++) {
+ if (channel >= channelCount) {
+ // Fill this channel with default data.
+ if (bitdepth == 16 && bpc == 16) {
+ stbi__uint16 *q = ((stbi__uint16 *) out) + channel;
+ stbi__uint16 val = channel == 3 ? 65535 : 0;
+ for (i = 0; i < pixelCount; i++, q += 4)
+ *q = val;
+ } else {
+ stbi_uc *p = out+channel;
+ stbi_uc val = channel == 3 ? 255 : 0;
+ for (i = 0; i < pixelCount; i++, p += 4)
+ *p = val;
+ }
+ } else {
+ if (ri->bits_per_channel == 16) { // output bpc
+ stbi__uint16 *q = ((stbi__uint16 *) out) + channel;
+ for (i = 0; i < pixelCount; i++, q += 4)
+ *q = (stbi__uint16) stbi__get16be(s);
+ } else {
+ stbi_uc *p = out+channel;
+ if (bitdepth == 16) { // input bpc
+ for (i = 0; i < pixelCount; i++, p += 4)
+ *p = (stbi_uc) (stbi__get16be(s) >> 8);
+ } else {
+ for (i = 0; i < pixelCount; i++, p += 4)
+ *p = stbi__get8(s);
+ }
+ }
+ }
+ }
+ }
+
+ // remove weird white matte from PSD
+ if (channelCount >= 4) {
+ if (ri->bits_per_channel == 16) {
+ for (i=0; i < w*h; ++i) {
+ stbi__uint16 *pixel = (stbi__uint16 *) out + 4*i;
+ if (pixel[3] != 0 && pixel[3] != 65535) {
+ float a = pixel[3] / 65535.0f;
+ float ra = 1.0f / a;
+ float inv_a = 65535.0f * (1 - ra);
+ pixel[0] = (stbi__uint16) (pixel[0]*ra + inv_a);
+ pixel[1] = (stbi__uint16) (pixel[1]*ra + inv_a);
+ pixel[2] = (stbi__uint16) (pixel[2]*ra + inv_a);
+ }
+ }
+ } else {
+ for (i=0; i < w*h; ++i) {
+ unsigned char *pixel = out + 4*i;
+ if (pixel[3] != 0 && pixel[3] != 255) {
+ float a = pixel[3] / 255.0f;
+ float ra = 1.0f / a;
+ float inv_a = 255.0f * (1 - ra);
+ pixel[0] = (unsigned char) (pixel[0]*ra + inv_a);
+ pixel[1] = (unsigned char) (pixel[1]*ra + inv_a);
+ pixel[2] = (unsigned char) (pixel[2]*ra + inv_a);
+ }
+ }
+ }
+ }
+
+ // convert to desired output format
+ if (req_comp && req_comp != 4) {
+ if (ri->bits_per_channel == 16)
+ out = (stbi_uc *) stbi__convert_format16((stbi__uint16 *) out, 4, req_comp, w, h);
+ else
+ out = stbi__convert_format(out, 4, req_comp, w, h);
+ if (out == NULL) return out; // stbi__convert_format frees input on failure
+ }
+
+ if (comp) *comp = 4;
+ *y = h;
+ *x = w;
+
+ return out;
+}
+#endif
+
+// *************************************************************************************************
+// Softimage PIC loader
+// by Tom Seddon
+//
+// See http://softimage.wiki.softimage.com/index.php/INFO:_PIC_file_format
+// See http://ozviz.wasp.uwa.edu.au/~pbourke/dataformats/softimagepic/
+
+#ifndef STBI_NO_PIC
+static int stbi__pic_is4(stbi__context *s,const char *str)
+{
+ int i;
+ for (i=0; i<4; ++i)
+ if (stbi__get8(s) != (stbi_uc)str[i])
+ return 0;
+
+ return 1;
+}
+
+static int stbi__pic_test_core(stbi__context *s)
+{
+ int i;
+
+ if (!stbi__pic_is4(s,"\x53\x80\xF6\x34"))
+ return 0;
+
+ for(i=0;i<84;++i)
+ stbi__get8(s);
+
+ if (!stbi__pic_is4(s,"PICT"))
+ return 0;
+
+ return 1;
+}
+
+typedef struct
+{
+ stbi_uc size,type,channel;
+} stbi__pic_packet;
+
+static stbi_uc *stbi__readval(stbi__context *s, int channel, stbi_uc *dest)
+{
+ int mask=0x80, i;
+
+ for (i=0; i<4; ++i, mask>>=1) {
+ if (channel & mask) {
+ if (stbi__at_eof(s)) return stbi__errpuc("bad file","PIC file too short");
+ dest[i]=stbi__get8(s);
+ }
+ }
+
+ return dest;
+}
+
+static void stbi__copyval(int channel,stbi_uc *dest,const stbi_uc *src)
+{
+ int mask=0x80,i;
+
+ for (i=0;i<4; ++i, mask>>=1)
+ if (channel&mask)
+ dest[i]=src[i];
+}
+
+static stbi_uc *stbi__pic_load_core(stbi__context *s,int width,int height,int *comp, stbi_uc *result)
+{
+ int act_comp=0,num_packets=0,y,chained;
+ stbi__pic_packet packets[10];
+
+ // this will (should...) cater for even some bizarre stuff like having data
+ // for the same channel in multiple packets.
+ do {
+ stbi__pic_packet *packet;
+
+ if (num_packets==sizeof(packets)/sizeof(packets[0]))
+ return stbi__errpuc("bad format","too many packets");
+
+ packet = &packets[num_packets++];
+
+ chained = stbi__get8(s);
+ packet->size = stbi__get8(s);
+ packet->type = stbi__get8(s);
+ packet->channel = stbi__get8(s);
+
+ act_comp |= packet->channel;
+
+ if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (reading packets)");
+ if (packet->size != 8) return stbi__errpuc("bad format","packet isn't 8bpp");
+ } while (chained);
+
+ *comp = (act_comp & 0x10 ? 4 : 3); // has alpha channel?
+
+ for(y=0; y<height; ++y) {
+ int packet_idx;
+
+ for(packet_idx=0; packet_idx < num_packets; ++packet_idx) {
+ stbi__pic_packet *packet = &packets[packet_idx];
+ stbi_uc *dest = result+y*width*4;
+
+ switch (packet->type) {
+ default:
+ return stbi__errpuc("bad format","packet has bad compression type");
+
+ case 0: {//uncompressed
+ int x;
+
+ for(x=0;x<width;++x, dest+=4)
+ if (!stbi__readval(s,packet->channel,dest))
+ return 0;
+ break;
+ }
+
+ case 1://Pure RLE
+ {
+ int left=width, i;
+
+ while (left>0) {
+ stbi_uc count,value[4];
+
+ count=stbi__get8(s);
+ if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (pure read count)");
+
+ if (count > left)
+ count = (stbi_uc) left;
+
+ if (!stbi__readval(s,packet->channel,value)) return 0;
+
+ for(i=0; i<count; ++i,dest+=4)
+ stbi__copyval(packet->channel,dest,value);
+ left -= count;
+ }
+ }
+ break;
+
+ case 2: {//Mixed RLE
+ int left=width;
+ while (left>0) {
+ int count = stbi__get8(s), i;
+ if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (mixed read count)");
+
+ if (count >= 128) { // Repeated
+ stbi_uc value[4];
+
+ if (count==128)
+ count = stbi__get16be(s);
+ else
+ count -= 127;
+ if (count > left)
+ return stbi__errpuc("bad file","scanline overrun");
+
+ if (!stbi__readval(s,packet->channel,value))
+ return 0;
+
+ for(i=0;i<count;++i, dest += 4)
+ stbi__copyval(packet->channel,dest,value);
+ } else { // Raw
+ ++count;
+ if (count>left) return stbi__errpuc("bad file","scanline overrun");
+
+ for(i=0;i<count;++i, dest+=4)
+ if (!stbi__readval(s,packet->channel,dest))
+ return 0;
+ }
+ left-=count;
+ }
+ break;
+ }
+ }
+ }
+ }
+
+ return result;
+}
+
+static void *stbi__pic_load(stbi__context *s,int *px,int *py,int *comp,int req_comp, stbi__result_info *ri)
+{
+ stbi_uc *result;
+ int i, x,y, internal_comp;
+ STBI_NOTUSED(ri);
+
+ if (!comp) comp = &internal_comp;
+
+ for (i=0; i<92; ++i)
+ stbi__get8(s);
+
+ x = stbi__get16be(s);
+ y = stbi__get16be(s);
+ if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (pic header)");
+ if (!stbi__mad3sizes_valid(x, y, 4, 0)) return stbi__errpuc("too large", "PIC image too large to decode");
+
+ stbi__get32be(s); //skip `ratio'
+ stbi__get16be(s); //skip `fields'
+ stbi__get16be(s); //skip `pad'
+
+ // intermediate buffer is RGBA
+ result = (stbi_uc *) stbi__malloc_mad3(x, y, 4, 0);
+ memset(result, 0xff, x*y*4);
+
+ if (!stbi__pic_load_core(s,x,y,comp, result)) {
+ STBI_FREE(result);
+ result=0;
+ }
+ *px = x;
+ *py = y;
+ if (req_comp == 0) req_comp = *comp;
+ result=stbi__convert_format(result,4,req_comp,x,y);
+
+ return result;
+}
+
+static int stbi__pic_test(stbi__context *s)
+{
+ int r = stbi__pic_test_core(s);
+ stbi__rewind(s);
+ return r;
+}
+#endif
+
+// *************************************************************************************************
+// GIF loader -- public domain by Jean-Marc Lienher -- simplified/shrunk by stb
+
+#ifndef STBI_NO_GIF
+typedef struct
+{
+ stbi__int16 prefix;
+ stbi_uc first;
+ stbi_uc suffix;
+} stbi__gif_lzw;
+
+typedef struct
+{
+ int w,h;
+ stbi_uc *out, *old_out; // output buffer (always 4 components)
+ int flags, bgindex, ratio, transparent, eflags, delay;
+ stbi_uc pal[256][4];
+ stbi_uc lpal[256][4];
+ stbi__gif_lzw codes[4096];
+ stbi_uc *color_table;
+ int parse, step;
+ int lflags;
+ int start_x, start_y;
+ int max_x, max_y;
+ int cur_x, cur_y;
+ int line_size;
+} stbi__gif;
+
+static int stbi__gif_test_raw(stbi__context *s)
+{
+ int sz;
+ if (stbi__get8(s) != 'G' || stbi__get8(s) != 'I' || stbi__get8(s) != 'F' || stbi__get8(s) != '8') return 0;
+ sz = stbi__get8(s);
+ if (sz != '9' && sz != '7') return 0;
+ if (stbi__get8(s) != 'a') return 0;
+ return 1;
+}
+
+static int stbi__gif_test(stbi__context *s)
+{
+ int r = stbi__gif_test_raw(s);
+ stbi__rewind(s);
+ return r;
+}
+
+static void stbi__gif_parse_colortable(stbi__context *s, stbi_uc pal[256][4], int num_entries, int transp)
+{
+ int i;
+ for (i=0; i < num_entries; ++i) {
+ pal[i][2] = stbi__get8(s);
+ pal[i][1] = stbi__get8(s);
+ pal[i][0] = stbi__get8(s);
+ pal[i][3] = transp == i ? 0 : 255;
+ }
+}
+
+static int stbi__gif_header(stbi__context *s, stbi__gif *g, int *comp, int is_info)
+{
+ stbi_uc version;
+ if (stbi__get8(s) != 'G' || stbi__get8(s) != 'I' || stbi__get8(s) != 'F' || stbi__get8(s) != '8')
+ return stbi__err("not GIF", "Corrupt GIF");
+
+ version = stbi__get8(s);
+ if (version != '7' && version != '9') return stbi__err("not GIF", "Corrupt GIF");
+ if (stbi__get8(s) != 'a') return stbi__err("not GIF", "Corrupt GIF");
+
+ stbi__g_failure_reason = "";
+ g->w = stbi__get16le(s);
+ g->h = stbi__get16le(s);
+ g->flags = stbi__get8(s);
+ g->bgindex = stbi__get8(s);
+ g->ratio = stbi__get8(s);
+ g->transparent = -1;
+
+ if (comp != 0) *comp = 4; // can't actually tell whether it's 3 or 4 until we parse the comments
+
+ if (is_info) return 1;
+
+ if (g->flags & 0x80)
+ stbi__gif_parse_colortable(s,g->pal, 2 << (g->flags & 7), -1);
+
+ return 1;
+}
+
+static int stbi__gif_info_raw(stbi__context *s, int *x, int *y, int *comp)
+{
+ stbi__gif* g = (stbi__gif*) stbi__malloc(sizeof(stbi__gif));
+ if (!stbi__gif_header(s, g, comp, 1)) {
+ STBI_FREE(g);
+ stbi__rewind( s );
+ return 0;
+ }
+ if (x) *x = g->w;
+ if (y) *y = g->h;
+ STBI_FREE(g);
+ return 1;
+}
+
+static void stbi__out_gif_code(stbi__gif *g, stbi__uint16 code)
+{
+ stbi_uc *p, *c;
+
+ // recurse to decode the prefixes, since the linked-list is backwards,
+ // and working backwards through an interleaved image would be nasty
+ if (g->codes[code].prefix >= 0)
+ stbi__out_gif_code(g, g->codes[code].prefix);
+
+ if (g->cur_y >= g->max_y) return;
+
+ p = &g->out[g->cur_x + g->cur_y];
+ c = &g->color_table[g->codes[code].suffix * 4];
+
+ if (c[3] >= 128) {
+ p[0] = c[2];
+ p[1] = c[1];
+ p[2] = c[0];
+ p[3] = c[3];
+ }
+ g->cur_x += 4;
+
+ if (g->cur_x >= g->max_x) {
+ g->cur_x = g->start_x;
+ g->cur_y += g->step;
+
+ while (g->cur_y >= g->max_y && g->parse > 0) {
+ g->step = (1 << g->parse) * g->line_size;
+ g->cur_y = g->start_y + (g->step >> 1);
+ --g->parse;
+ }
+ }
+}
+
+static stbi_uc *stbi__process_gif_raster(stbi__context *s, stbi__gif *g)
+{
+ stbi_uc lzw_cs;
+ stbi__int32 len, init_code;
+ stbi__uint32 first;
+ stbi__int32 codesize, codemask, avail, oldcode, bits, valid_bits, clear;
+ stbi__gif_lzw *p;
+
+ lzw_cs = stbi__get8(s);
+ if (lzw_cs > 12) return NULL;
+ clear = 1 << lzw_cs;
+ first = 1;
+ codesize = lzw_cs + 1;
+ codemask = (1 << codesize) - 1;
+ bits = 0;
+ valid_bits = 0;
+ for (init_code = 0; init_code < clear; init_code++) {
+ g->codes[init_code].prefix = -1;
+ g->codes[init_code].first = (stbi_uc) init_code;
+ g->codes[init_code].suffix = (stbi_uc) init_code;
+ }
+
+ // support no starting clear code
+ avail = clear+2;
+ oldcode = -1;
+
+ len = 0;
+ for(;;) {
+ if (valid_bits < codesize) {
+ if (len == 0) {
+ len = stbi__get8(s); // start new block
+ if (len == 0)
+ return g->out;
+ }
+ --len;
+ bits |= (stbi__int32) stbi__get8(s) << valid_bits;
+ valid_bits += 8;
+ } else {
+ stbi__int32 code = bits & codemask;
+ bits >>= codesize;
+ valid_bits -= codesize;
+ // @OPTIMIZE: is there some way we can accelerate the non-clear path?
+ if (code == clear) { // clear code
+ codesize = lzw_cs + 1;
+ codemask = (1 << codesize) - 1;
+ avail = clear + 2;
+ oldcode = -1;
+ first = 0;
+ } else if (code == clear + 1) { // end of stream code
+ stbi__skip(s, len);
+ while ((len = stbi__get8(s)) > 0)
+ stbi__skip(s,len);
+ return g->out;
+ } else if (code <= avail) {
+ if (first) return stbi__errpuc("no clear code", "Corrupt GIF");
+
+ if (oldcode >= 0) {
+ p = &g->codes[avail++];
+ if (avail > 4096) return stbi__errpuc("too many codes", "Corrupt GIF");
+ p->prefix = (stbi__int16) oldcode;
+ p->first = g->codes[oldcode].first;
+ p->suffix = (code == avail) ? p->first : g->codes[code].first;
+ } else if (code == avail)
+ return stbi__errpuc("illegal code in raster", "Corrupt GIF");
+
+ stbi__out_gif_code(g, (stbi__uint16) code);
+
+ if ((avail & codemask) == 0 && avail <= 0x0FFF) {
+ codesize++;
+ codemask = (1 << codesize) - 1;
+ }
+
+ oldcode = code;
+ } else {
+ return stbi__errpuc("illegal code in raster", "Corrupt GIF");
+ }
+ }
+ }
+}
+
+static void stbi__fill_gif_background(stbi__gif *g, int x0, int y0, int x1, int y1)
+{
+ int x, y;
+ stbi_uc *c = g->pal[g->bgindex];
+ for (y = y0; y < y1; y += 4 * g->w) {
+ for (x = x0; x < x1; x += 4) {
+ stbi_uc *p = &g->out[y + x];
+ p[0] = c[2];
+ p[1] = c[1];
+ p[2] = c[0];
+ p[3] = 0;
+ }
+ }
+}
+
+// this function is designed to support animated gifs, although stb_image doesn't support it
+static stbi_uc *stbi__gif_load_next(stbi__context *s, stbi__gif *g, int *comp, int req_comp)
+{
+ int i;
+ stbi_uc *prev_out = 0;
+
+ if (g->out == 0 && !stbi__gif_header(s, g, comp,0))
+ return 0; // stbi__g_failure_reason set by stbi__gif_header
+
+ if (!stbi__mad3sizes_valid(g->w, g->h, 4, 0))
+ return stbi__errpuc("too large", "GIF too large");
+
+ prev_out = g->out;
+ g->out = (stbi_uc *) stbi__malloc_mad3(4, g->w, g->h, 0);
+ if (g->out == 0) return stbi__errpuc("outofmem", "Out of memory");
+
+ switch ((g->eflags & 0x1C) >> 2) {
+ case 0: // unspecified (also always used on 1st frame)
+ stbi__fill_gif_background(g, 0, 0, 4 * g->w, 4 * g->w * g->h);
+ break;
+ case 1: // do not dispose
+ if (prev_out) memcpy(g->out, prev_out, 4 * g->w * g->h);
+ g->old_out = prev_out;
+ break;
+ case 2: // dispose to background
+ if (prev_out) memcpy(g->out, prev_out, 4 * g->w * g->h);
+ stbi__fill_gif_background(g, g->start_x, g->start_y, g->max_x, g->max_y);
+ break;
+ case 3: // dispose to previous
+ if (g->old_out) {
+ for (i = g->start_y; i < g->max_y; i += 4 * g->w)
+ memcpy(&g->out[i + g->start_x], &g->old_out[i + g->start_x], g->max_x - g->start_x);
+ }
+ break;
+ }
+
+ for (;;) {
+ switch (stbi__get8(s)) {
+ case 0x2C: /* Image Descriptor */
+ {
+ int prev_trans = -1;
+ stbi__int32 x, y, w, h;
+ stbi_uc *o;
+
+ x = stbi__get16le(s);
+ y = stbi__get16le(s);
+ w = stbi__get16le(s);
+ h = stbi__get16le(s);
+ if (((x + w) > (g->w)) || ((y + h) > (g->h)))
+ return stbi__errpuc("bad Image Descriptor", "Corrupt GIF");
+
+ g->line_size = g->w * 4;
+ g->start_x = x * 4;
+ g->start_y = y * g->line_size;
+ g->max_x = g->start_x + w * 4;
+ g->max_y = g->start_y + h * g->line_size;
+ g->cur_x = g->start_x;
+ g->cur_y = g->start_y;
+
+ g->lflags = stbi__get8(s);
+
+ if (g->lflags & 0x40) {
+ g->step = 8 * g->line_size; // first interlaced spacing
+ g->parse = 3;
+ } else {
+ g->step = g->line_size;
+ g->parse = 0;
+ }
+
+ if (g->lflags & 0x80) {
+ stbi__gif_parse_colortable(s,g->lpal, 2 << (g->lflags & 7), g->eflags & 0x01 ? g->transparent : -1);
+ g->color_table = (stbi_uc *) g->lpal;
+ } else if (g->flags & 0x80) {
+ if (g->transparent >= 0 && (g->eflags & 0x01)) {
+ prev_trans = g->pal[g->transparent][3];
+ g->pal[g->transparent][3] = 0;
+ }
+ g->color_table = (stbi_uc *) g->pal;
+ } else
+ return stbi__errpuc("missing color table", "Corrupt GIF");
+
+ o = stbi__process_gif_raster(s, g);
+ if (o == NULL) return NULL;
+
+ if (prev_trans != -1)
+ g->pal[g->transparent][3] = (stbi_uc) prev_trans;
+
+ return o;
+ }
+
+ case 0x21: // Comment Extension.
+ {
+ int len;
+ if (stbi__get8(s) == 0xF9) { // Graphic Control Extension.
+ len = stbi__get8(s);
+ if (len == 4) {
+ g->eflags = stbi__get8(s);
+ g->delay = stbi__get16le(s);
+ g->transparent = stbi__get8(s);
+ } else {
+ stbi__skip(s, len);
+ break;
+ }
+ }
+ while ((len = stbi__get8(s)) != 0)
+ stbi__skip(s, len);
+ break;
+ }
+
+ case 0x3B: // gif stream termination code
+ return (stbi_uc *) s; // using '1' causes warning on some compilers
+
+ default:
+ return stbi__errpuc("unknown code", "Corrupt GIF");
+ }
+ }
+
+ STBI_NOTUSED(req_comp);
+}
+
+static void *stbi__gif_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+ stbi_uc *u = 0;
+ stbi__gif* g = (stbi__gif*) stbi__malloc(sizeof(stbi__gif));
+ memset(g, 0, sizeof(*g));
+ STBI_NOTUSED(ri);
+
+ u = stbi__gif_load_next(s, g, comp, req_comp);
+ if (u == (stbi_uc *) s) u = 0; // end of animated gif marker
+ if (u) {
+ *x = g->w;
+ *y = g->h;
+ if (req_comp && req_comp != 4)
+ u = stbi__convert_format(u, 4, req_comp, g->w, g->h);
+ }
+ else if (g->out)
+ STBI_FREE(g->out);
+ STBI_FREE(g);
+ return u;
+}
+
+static int stbi__gif_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ return stbi__gif_info_raw(s,x,y,comp);
+}
+#endif
+
+// *************************************************************************************************
+// Radiance RGBE HDR loader
+// originally by Nicolas Schulz
+#ifndef STBI_NO_HDR
+static int stbi__hdr_test_core(stbi__context *s, const char *signature)
+{
+ int i;
+ for (i=0; signature[i]; ++i)
+ if (stbi__get8(s) != signature[i])
+ return 0;
+ stbi__rewind(s);
+ return 1;
+}
+
+static int stbi__hdr_test(stbi__context* s)
+{
+ int r = stbi__hdr_test_core(s, "#?RADIANCE\n");
+ stbi__rewind(s);
+ if(!r) {
+ r = stbi__hdr_test_core(s, "#?RGBE\n");
+ stbi__rewind(s);
+ }
+ return r;
+}
+
+#define STBI__HDR_BUFLEN 1024
+static char *stbi__hdr_gettoken(stbi__context *z, char *buffer)
+{
+ int len=0;
+ char c = '\0';
+
+ c = (char) stbi__get8(z);
+
+ while (!stbi__at_eof(z) && c != '\n') {
+ buffer[len++] = c;
+ if (len == STBI__HDR_BUFLEN-1) {
+ // flush to end of line
+ while (!stbi__at_eof(z) && stbi__get8(z) != '\n')
+ ;
+ break;
+ }
+ c = (char) stbi__get8(z);
+ }
+
+ buffer[len] = 0;
+ return buffer;
+}
+
+static void stbi__hdr_convert(float *output, stbi_uc *input, int req_comp)
+{
+ if ( input[3] != 0 ) {
+ float f1;
+ // Exponent
+ f1 = (float) ldexp(1.0f, input[3] - (int)(128 + 8));
+ if (req_comp <= 2)
+ output[0] = (input[0] + input[1] + input[2]) * f1 / 3;
+ else {
+ output[0] = input[0] * f1;
+ output[1] = input[1] * f1;
+ output[2] = input[2] * f1;
+ }
+ if (req_comp == 2) output[1] = 1;
+ if (req_comp == 4) output[3] = 1;
+ } else {
+ switch (req_comp) {
+ case 4: output[3] = 1; /* fallthrough */
+ case 3: output[0] = output[1] = output[2] = 0;
+ break;
+ case 2: output[1] = 1; /* fallthrough */
+ case 1: output[0] = 0;
+ break;
+ }
+ }
+}
+
+static float *stbi__hdr_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+ char buffer[STBI__HDR_BUFLEN];
+ char *token;
+ int valid = 0;
+ int width, height;
+ stbi_uc *scanline;
+ float *hdr_data;
+ int len;
+ unsigned char count, value;
+ int i, j, k, c1,c2, z;
+ const char *headerToken;
+ STBI_NOTUSED(ri);
+
+ // Check identifier
+ headerToken = stbi__hdr_gettoken(s,buffer);
+ if (strcmp(headerToken, "#?RADIANCE") != 0 && strcmp(headerToken, "#?RGBE") != 0)
+ return stbi__errpf("not HDR", "Corrupt HDR image");
+
+ // Parse header
+ for(;;) {
+ token = stbi__hdr_gettoken(s,buffer);
+ if (token[0] == 0) break;
+ if (strcmp(token, "FORMAT=32-bit_rle_rgbe") == 0) valid = 1;
+ }
+
+ if (!valid) return stbi__errpf("unsupported format", "Unsupported HDR format");
+
+ // Parse width and height
+ // can't use sscanf() if we're not using stdio!
+ token = stbi__hdr_gettoken(s,buffer);
+ if (strncmp(token, "-Y ", 3)) return stbi__errpf("unsupported data layout", "Unsupported HDR format");
+ token += 3;
+ height = (int) strtol(token, &token, 10);
+ while (*token == ' ') ++token;
+ if (strncmp(token, "+X ", 3)) return stbi__errpf("unsupported data layout", "Unsupported HDR format");
+ token += 3;
+ width = (int) strtol(token, NULL, 10);
+
+ *x = width;
+ *y = height;
+
+ if (comp) *comp = 3;
+ if (req_comp == 0) req_comp = 3;
+
+ if (!stbi__mad4sizes_valid(width, height, req_comp, sizeof(float), 0))
+ return stbi__errpf("too large", "HDR image is too large");
+
+ // Read data
+ hdr_data = (float *) stbi__malloc_mad4(width, height, req_comp, sizeof(float), 0);
+ if (!hdr_data)
+ return stbi__errpf("outofmem", "Out of memory");
+
+ // Load image data
+ // image data is stored as some number of sca
+ if ( width < 8 || width >= 32768) {
+ // Read flat data
+ for (j=0; j < height; ++j) {
+ for (i=0; i < width; ++i) {
+ stbi_uc rgbe[4];
+ main_decode_loop:
+ stbi__getn(s, rgbe, 4);
+ stbi__hdr_convert(hdr_data + j * width * req_comp + i * req_comp, rgbe, req_comp);
+ }
+ }
+ } else {
+ // Read RLE-encoded data
+ scanline = NULL;
+
+ for (j = 0; j < height; ++j) {
+ c1 = stbi__get8(s);
+ c2 = stbi__get8(s);
+ len = stbi__get8(s);
+ if (c1 != 2 || c2 != 2 || (len & 0x80)) {
+ // not run-length encoded, so we have to actually use THIS data as a decoded
+ // pixel (note this can't be a valid pixel--one of RGB must be >= 128)
+ stbi_uc rgbe[4];
+ rgbe[0] = (stbi_uc) c1;
+ rgbe[1] = (stbi_uc) c2;
+ rgbe[2] = (stbi_uc) len;
+ rgbe[3] = (stbi_uc) stbi__get8(s);
+ stbi__hdr_convert(hdr_data, rgbe, req_comp);
+ i = 1;
+ j = 0;
+ STBI_FREE(scanline);
+ goto main_decode_loop; // yes, this makes no sense
+ }
+ len <<= 8;
+ len |= stbi__get8(s);
+ if (len != width) { STBI_FREE(hdr_data); STBI_FREE(scanline); return stbi__errpf("invalid decoded scanline length", "corrupt HDR"); }
+ if (scanline == NULL) {
+ scanline = (stbi_uc *) stbi__malloc_mad2(width, 4, 0);
+ if (!scanline) {
+ STBI_FREE(hdr_data);
+ return stbi__errpf("outofmem", "Out of memory");
+ }
+ }
+
+ for (k = 0; k < 4; ++k) {
+ int nleft;
+ i = 0;
+ while ((nleft = width - i) > 0) {
+ count = stbi__get8(s);
+ if (count > 128) {
+ // Run
+ value = stbi__get8(s);
+ count -= 128;
+ if (count > nleft) { STBI_FREE(hdr_data); STBI_FREE(scanline); return stbi__errpf("corrupt", "bad RLE data in HDR"); }
+ for (z = 0; z < count; ++z)
+ scanline[i++ * 4 + k] = value;
+ } else {
+ // Dump
+ if (count > nleft) { STBI_FREE(hdr_data); STBI_FREE(scanline); return stbi__errpf("corrupt", "bad RLE data in HDR"); }
+ for (z = 0; z < count; ++z)
+ scanline[i++ * 4 + k] = stbi__get8(s);
+ }
+ }
+ }
+ for (i=0; i < width; ++i)
+ stbi__hdr_convert(hdr_data+(j*width + i)*req_comp, scanline + i*4, req_comp);
+ }
+ if (scanline)
+ STBI_FREE(scanline);
+ }
+
+ return hdr_data;
+}
+
+static int stbi__hdr_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ char buffer[STBI__HDR_BUFLEN];
+ char *token;
+ int valid = 0;
+ int dummy;
+
+ if (!x) x = &dummy;
+ if (!y) y = &dummy;
+ if (!comp) comp = &dummy;
+
+ if (stbi__hdr_test(s) == 0) {
+ stbi__rewind( s );
+ return 0;
+ }
+
+ for(;;) {
+ token = stbi__hdr_gettoken(s,buffer);
+ if (token[0] == 0) break;
+ if (strcmp(token, "FORMAT=32-bit_rle_rgbe") == 0) valid = 1;
+ }
+
+ if (!valid) {
+ stbi__rewind( s );
+ return 0;
+ }
+ token = stbi__hdr_gettoken(s,buffer);
+ if (strncmp(token, "-Y ", 3)) {
+ stbi__rewind( s );
+ return 0;
+ }
+ token += 3;
+ *y = (int) strtol(token, &token, 10);
+ while (*token == ' ') ++token;
+ if (strncmp(token, "+X ", 3)) {
+ stbi__rewind( s );
+ return 0;
+ }
+ token += 3;
+ *x = (int) strtol(token, NULL, 10);
+ *comp = 3;
+ return 1;
+}
+#endif // STBI_NO_HDR
+
+#ifndef STBI_NO_BMP
+static int stbi__bmp_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ void *p;
+ stbi__bmp_data info;
+
+ info.all_a = 255;
+ p = stbi__bmp_parse_header(s, &info);
+ stbi__rewind( s );
+ if (p == NULL)
+ return 0;
+ if (x) *x = s->img_x;
+ if (y) *y = s->img_y;
+ if (comp) *comp = info.ma ? 4 : 3;
+ return 1;
+}
+#endif
+
+#ifndef STBI_NO_PSD
+static int stbi__psd_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ int channelCount, dummy;
+ if (!x) x = &dummy;
+ if (!y) y = &dummy;
+ if (!comp) comp = &dummy;
+ if (stbi__get32be(s) != 0x38425053) {
+ stbi__rewind( s );
+ return 0;
+ }
+ if (stbi__get16be(s) != 1) {
+ stbi__rewind( s );
+ return 0;
+ }
+ stbi__skip(s, 6);
+ channelCount = stbi__get16be(s);
+ if (channelCount < 0 || channelCount > 16) {
+ stbi__rewind( s );
+ return 0;
+ }
+ *y = stbi__get32be(s);
+ *x = stbi__get32be(s);
+ if (stbi__get16be(s) != 8) {
+ stbi__rewind( s );
+ return 0;
+ }
+ if (stbi__get16be(s) != 3) {
+ stbi__rewind( s );
+ return 0;
+ }
+ *comp = 4;
+ return 1;
+}
+#endif
+
+#ifndef STBI_NO_PIC
+static int stbi__pic_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ int act_comp=0,num_packets=0,chained,dummy;
+ stbi__pic_packet packets[10];
+
+ if (!x) x = &dummy;
+ if (!y) y = &dummy;
+ if (!comp) comp = &dummy;
+
+ if (!stbi__pic_is4(s,"\x53\x80\xF6\x34")) {
+ stbi__rewind(s);
+ return 0;
+ }
+
+ stbi__skip(s, 88);
+
+ *x = stbi__get16be(s);
+ *y = stbi__get16be(s);
+ if (stbi__at_eof(s)) {
+ stbi__rewind( s);
+ return 0;
+ }
+ if ( (*x) != 0 && (1 << 28) / (*x) < (*y)) {
+ stbi__rewind( s );
+ return 0;
+ }
+
+ stbi__skip(s, 8);
+
+ do {
+ stbi__pic_packet *packet;
+
+ if (num_packets==sizeof(packets)/sizeof(packets[0]))
+ return 0;
+
+ packet = &packets[num_packets++];
+ chained = stbi__get8(s);
+ packet->size = stbi__get8(s);
+ packet->type = stbi__get8(s);
+ packet->channel = stbi__get8(s);
+ act_comp |= packet->channel;
+
+ if (stbi__at_eof(s)) {
+ stbi__rewind( s );
+ return 0;
+ }
+ if (packet->size != 8) {
+ stbi__rewind( s );
+ return 0;
+ }
+ } while (chained);
+
+ *comp = (act_comp & 0x10 ? 4 : 3);
+
+ return 1;
+}
+#endif
+
+// *************************************************************************************************
+// Portable Gray Map and Portable Pixel Map loader
+// by Ken Miller
+//
+// PGM: http://netpbm.sourceforge.net/doc/pgm.html
+// PPM: http://netpbm.sourceforge.net/doc/ppm.html
+//
+// Known limitations:
+// Does not support comments in the header section
+// Does not support ASCII image data (formats P2 and P3)
+// Does not support 16-bit-per-channel
+
+#ifndef STBI_NO_PNM
+
+static int stbi__pnm_test(stbi__context *s)
+{
+ char p, t;
+ p = (char) stbi__get8(s);
+ t = (char) stbi__get8(s);
+ if (p != 'P' || (t != '5' && t != '6')) {
+ stbi__rewind( s );
+ return 0;
+ }
+ return 1;
+}
+
+static void *stbi__pnm_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+ stbi_uc *out;
+ STBI_NOTUSED(ri);
+
+ if (!stbi__pnm_info(s, (int *)&s->img_x, (int *)&s->img_y, (int *)&s->img_n))
+ return 0;
+
+ *x = s->img_x;
+ *y = s->img_y;
+ if (comp) *comp = s->img_n;
+
+ if (!stbi__mad3sizes_valid(s->img_n, s->img_x, s->img_y, 0))
+ return stbi__errpuc("too large", "PNM too large");
+
+ out = (stbi_uc *) stbi__malloc_mad3(s->img_n, s->img_x, s->img_y, 0);
+ if (!out) return stbi__errpuc("outofmem", "Out of memory");
+ stbi__getn(s, out, s->img_n * s->img_x * s->img_y);
+
+ if (req_comp && req_comp != s->img_n) {
+ out = stbi__convert_format(out, s->img_n, req_comp, s->img_x, s->img_y);
+ if (out == NULL) return out; // stbi__convert_format frees input on failure
+ }
+ return out;
+}
+
+static int stbi__pnm_isspace(char c)
+{
+ return c == ' ' || c == '\t' || c == '\n' || c == '\v' || c == '\f' || c == '\r';
+}
+
+static void stbi__pnm_skip_whitespace(stbi__context *s, char *c)
+{
+ for (;;) {
+ while (!stbi__at_eof(s) && stbi__pnm_isspace(*c))
+ *c = (char) stbi__get8(s);
+
+ if (stbi__at_eof(s) || *c != '#')
+ break;
+
+ while (!stbi__at_eof(s) && *c != '\n' && *c != '\r' )
+ *c = (char) stbi__get8(s);
+ }
+}
+
+static int stbi__pnm_isdigit(char c)
+{
+ return c >= '0' && c <= '9';
+}
+
+static int stbi__pnm_getinteger(stbi__context *s, char *c)
+{
+ int value = 0;
+
+ while (!stbi__at_eof(s) && stbi__pnm_isdigit(*c)) {
+ value = value*10 + (*c - '0');
+ *c = (char) stbi__get8(s);
+ }
+
+ return value;
+}
+
+static int stbi__pnm_info(stbi__context *s, int *x, int *y, int *comp)
+{
+ int maxv, dummy;
+ char c, p, t;
+
+ if (!x) x = &dummy;
+ if (!y) y = &dummy;
+ if (!comp) comp = &dummy;
+
+ stbi__rewind(s);
+
+ // Get identifier
+ p = (char) stbi__get8(s);
+ t = (char) stbi__get8(s);
+ if (p != 'P' || (t != '5' && t != '6')) {
+ stbi__rewind(s);
+ return 0;
+ }
+
+ *comp = (t == '6') ? 3 : 1; // '5' is 1-component .pgm; '6' is 3-component .ppm
+
+ c = (char) stbi__get8(s);
+ stbi__pnm_skip_whitespace(s, &c);
+
+ *x = stbi__pnm_getinteger(s, &c); // read width
+ stbi__pnm_skip_whitespace(s, &c);
+
+ *y = stbi__pnm_getinteger(s, &c); // read height
+ stbi__pnm_skip_whitespace(s, &c);
+
+ maxv = stbi__pnm_getinteger(s, &c); // read max value
+
+ if (maxv > 255)
+ return stbi__err("max value > 255", "PPM image not 8-bit");
+ else
+ return 1;
+}
+#endif
+
+static int stbi__info_main(stbi__context *s, int *x, int *y, int *comp)
+{
+ #ifndef STBI_NO_JPEG
+ if (stbi__jpeg_info(s, x, y, comp)) return 1;
+ #endif
+
+ #ifndef STBI_NO_PNG
+ if (stbi__png_info(s, x, y, comp)) return 1;
+ #endif
+
+ #ifndef STBI_NO_GIF
+ if (stbi__gif_info(s, x, y, comp)) return 1;
+ #endif
+
+ #ifndef STBI_NO_BMP
+ if (stbi__bmp_info(s, x, y, comp)) return 1;
+ #endif
+
+ #ifndef STBI_NO_PSD
+ if (stbi__psd_info(s, x, y, comp)) return 1;
+ #endif
+
+ #ifndef STBI_NO_PIC
+ if (stbi__pic_info(s, x, y, comp)) return 1;
+ #endif
+
+ #ifndef STBI_NO_PNM
+ if (stbi__pnm_info(s, x, y, comp)) return 1;
+ #endif
+
+ #ifndef STBI_NO_HDR
+ if (stbi__hdr_info(s, x, y, comp)) return 1;
+ #endif
+
+ // test tga last because it's a crappy test!
+ #ifndef STBI_NO_TGA
+ if (stbi__tga_info(s, x, y, comp))
+ return 1;
+ #endif
+ return stbi__err("unknown image type", "Image not of any known type, or corrupt");
+}
+
+#ifndef STBI_NO_STDIO
+STBIDEF int stbi_info(char const *filename, int *x, int *y, int *comp)
+{
+ FILE *f = stbi__fopen(filename, "rb");
+ int result;
+ if (!f) return stbi__err("can't fopen", "Unable to open file");
+ result = stbi_info_from_file(f, x, y, comp);
+ fclose(f);
+ return result;
+}
+
+STBIDEF int stbi_info_from_file(FILE *f, int *x, int *y, int *comp)
+{
+ int r;
+ stbi__context s;
+ long pos = ftell(f);
+ stbi__start_file(&s, f);
+ r = stbi__info_main(&s,x,y,comp);
+ fseek(f,pos,SEEK_SET);
+ return r;
+}
+#endif // !STBI_NO_STDIO
+
+STBIDEF int stbi_info_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp)
+{
+ stbi__context s;
+ stbi__start_mem(&s,buffer,len);
+ return stbi__info_main(&s,x,y,comp);
+}
+
+STBIDEF int stbi_info_from_callbacks(stbi_io_callbacks const *c, void *user, int *x, int *y, int *comp)
+{
+ stbi__context s;
+ stbi__start_callbacks(&s, (stbi_io_callbacks *) c, user);
+ return stbi__info_main(&s,x,y,comp);
+}
+
+#endif // STB_IMAGE_IMPLEMENTATION
+
+/*
+ revision history:
+ 2.15 (2017-03-18) fix png-1,2,4 bug; now all Imagenet JPGs decode;
+ warning fixes; disable run-time SSE detection on gcc;
+ uniform handling of optional "return" values;
+ thread-safe initialization of zlib tables
+ 2.14 (2017-03-03) remove deprecated STBI_JPEG_OLD; fixes for Imagenet JPGs
+ 2.13 (2016-11-29) add 16-bit API, only supported for PNG right now
+ 2.12 (2016-04-02) fix typo in 2.11 PSD fix that caused crashes
+ 2.11 (2016-04-02) allocate large structures on the stack
+ remove white matting for transparent PSD
+ fix reported channel count for PNG & BMP
+ re-enable SSE2 in non-gcc 64-bit
+ support RGB-formatted JPEG
+ read 16-bit PNGs (only as 8-bit)
+ 2.10 (2016-01-22) avoid warning introduced in 2.09 by STBI_REALLOC_SIZED
+ 2.09 (2016-01-16) allow comments in PNM files
+ 16-bit-per-pixel TGA (not bit-per-component)
+ info() for TGA could break due to .hdr handling
+ info() for BMP to shares code instead of sloppy parse
+ can use STBI_REALLOC_SIZED if allocator doesn't support realloc
+ code cleanup
+ 2.08 (2015-09-13) fix to 2.07 cleanup, reading RGB PSD as RGBA
+ 2.07 (2015-09-13) fix compiler warnings
+ partial animated GIF support
+ limited 16-bpc PSD support
+ #ifdef unused functions
+ bug with < 92 byte PIC,PNM,HDR,TGA
+ 2.06 (2015-04-19) fix bug where PSD returns wrong '*comp' value
+ 2.05 (2015-04-19) fix bug in progressive JPEG handling, fix warning
+ 2.04 (2015-04-15) try to re-enable SIMD on MinGW 64-bit
+ 2.03 (2015-04-12) extra corruption checking (mmozeiko)
+ stbi_set_flip_vertically_on_load (nguillemot)
+ fix NEON support; fix mingw support
+ 2.02 (2015-01-19) fix incorrect assert, fix warning
+ 2.01 (2015-01-17) fix various warnings; suppress SIMD on gcc 32-bit without -msse2
+ 2.00b (2014-12-25) fix STBI_MALLOC in progressive JPEG
+ 2.00 (2014-12-25) optimize JPG, including x86 SSE2 & NEON SIMD (ryg)
+ progressive JPEG (stb)
+ PGM/PPM support (Ken Miller)
+ STBI_MALLOC,STBI_REALLOC,STBI_FREE
+ GIF bugfix -- seemingly never worked
+ STBI_NO_*, STBI_ONLY_*
+ 1.48 (2014-12-14) fix incorrectly-named assert()
+ 1.47 (2014-12-14) 1/2/4-bit PNG support, both direct and paletted (Omar Cornut & stb)
+ optimize PNG (ryg)
+ fix bug in interlaced PNG with user-specified channel count (stb)
+ 1.46 (2014-08-26)
+ fix broken tRNS chunk (colorkey-style transparency) in non-paletted PNG
+ 1.45 (2014-08-16)
+ fix MSVC-ARM internal compiler error by wrapping malloc
+ 1.44 (2014-08-07)
+ various warning fixes from Ronny Chevalier
+ 1.43 (2014-07-15)
+ fix MSVC-only compiler problem in code changed in 1.42
+ 1.42 (2014-07-09)
+ don't define _CRT_SECURE_NO_WARNINGS (affects user code)
+ fixes to stbi__cleanup_jpeg path
+ added STBI_ASSERT to avoid requiring assert.h
+ 1.41 (2014-06-25)
+ fix search&replace from 1.36 that messed up comments/error messages
+ 1.40 (2014-06-22)
+ fix gcc struct-initialization warning
+ 1.39 (2014-06-15)
+ fix to TGA optimization when req_comp != number of components in TGA;
+ fix to GIF loading because BMP wasn't rewinding (whoops, no GIFs in my test suite)
+ add support for BMP version 5 (more ignored fields)
+ 1.38 (2014-06-06)
+ suppress MSVC warnings on integer casts truncating values
+ fix accidental rename of 'skip' field of I/O
+ 1.37 (2014-06-04)
+ remove duplicate typedef
+ 1.36 (2014-06-03)
+ convert to header file single-file library
+ if de-iphone isn't set, load iphone images color-swapped instead of returning NULL
+ 1.35 (2014-05-27)
+ various warnings
+ fix broken STBI_SIMD path
+ fix bug where stbi_load_from_file no longer left file pointer in correct place
+ fix broken non-easy path for 32-bit BMP (possibly never used)
+ TGA optimization by Arseny Kapoulkine
+ 1.34 (unknown)
+ use STBI_NOTUSED in stbi__resample_row_generic(), fix one more leak in tga failure case
+ 1.33 (2011-07-14)
+ make stbi_is_hdr work in STBI_NO_HDR (as specified), minor compiler-friendly improvements
+ 1.32 (2011-07-13)
+ support for "info" function for all supported filetypes (SpartanJ)
+ 1.31 (2011-06-20)
+ a few more leak fixes, bug in PNG handling (SpartanJ)
+ 1.30 (2011-06-11)
+ added ability to load files via callbacks to accomidate custom input streams (Ben Wenger)
+ removed deprecated format-specific test/load functions
+ removed support for installable file formats (stbi_loader) -- would have been broken for IO callbacks anyway
+ error cases in bmp and tga give messages and don't leak (Raymond Barbiero, grisha)
+ fix inefficiency in decoding 32-bit BMP (David Woo)
+ 1.29 (2010-08-16)
+ various warning fixes from Aurelien Pocheville
+ 1.28 (2010-08-01)
+ fix bug in GIF palette transparency (SpartanJ)
+ 1.27 (2010-08-01)
+ cast-to-stbi_uc to fix warnings
+ 1.26 (2010-07-24)
+ fix bug in file buffering for PNG reported by SpartanJ
+ 1.25 (2010-07-17)
+ refix trans_data warning (Won Chun)
+ 1.24 (2010-07-12)
+ perf improvements reading from files on platforms with lock-heavy fgetc()
+ minor perf improvements for jpeg
+ deprecated type-specific functions so we'll get feedback if they're needed
+ attempt to fix trans_data warning (Won Chun)
+ 1.23 fixed bug in iPhone support
+ 1.22 (2010-07-10)
+ removed image *writing* support
+ stbi_info support from Jetro Lauha
+ GIF support from Jean-Marc Lienher
+ iPhone PNG-extensions from James Brown
+ warning-fixes from Nicolas Schulz and Janez Zemva (i.stbi__err. Janez (U+017D)emva)
+ 1.21 fix use of 'stbi_uc' in header (reported by jon blow)
+ 1.20 added support for Softimage PIC, by Tom Seddon
+ 1.19 bug in interlaced PNG corruption check (found by ryg)
+ 1.18 (2008-08-02)
+ fix a threading bug (local mutable static)
+ 1.17 support interlaced PNG
+ 1.16 major bugfix - stbi__convert_format converted one too many pixels
+ 1.15 initialize some fields for thread safety
+ 1.14 fix threadsafe conversion bug
+ header-file-only version (#define STBI_HEADER_FILE_ONLY before including)
+ 1.13 threadsafe
+ 1.12 const qualifiers in the API
+ 1.11 Support installable IDCT, colorspace conversion routines
+ 1.10 Fixes for 64-bit (don't use "unsigned long")
+ optimized upsampling by Fabian "ryg" Giesen
+ 1.09 Fix format-conversion for PSD code (bad global variables!)
+ 1.08 Thatcher Ulrich's PSD code integrated by Nicolas Schulz
+ 1.07 attempt to fix C++ warning/errors again
+ 1.06 attempt to fix C++ warning/errors again
+ 1.05 fix TGA loading to return correct *comp and use good luminance calc
+ 1.04 default float alpha is 1, not 255; use 'void *' for stbi_image_free
+ 1.03 bugfixes to STBI_NO_STDIO, STBI_NO_HDR
+ 1.02 support for (subset of) HDR files, float interface for preferred access to them
+ 1.01 fix bug: possible bug in handling right-side up bmps... not sure
+ fix bug: the stbi__bmp_load() and stbi__tga_load() functions didn't work at all
+ 1.00 interface to zlib that skips zlib header
+ 0.99 correct handling of alpha in palette
+ 0.98 TGA loader by lonesock; dynamically add loaders (untested)
+ 0.97 jpeg errors on too large a file; also catch another malloc failure
+ 0.96 fix detection of invalid v value - particleman@mollyrocket forum
+ 0.95 during header scan, seek to markers in case of padding
+ 0.94 STBI_NO_STDIO to disable stdio usage; rename all #defines the same
+ 0.93 handle jpegtran output; verbose errors
+ 0.92 read 4,8,16,24,32-bit BMP files of several formats
+ 0.91 output 24-bit Windows 3.0 BMP files
+ 0.90 fix a few more warnings; bump version number to approach 1.0
+ 0.61 bugfixes due to Marc LeBlanc, Christopher Lloyd
+ 0.60 fix compiling as c++
+ 0.59 fix warnings: merge Dave Moore's -Wall fixes
+ 0.58 fix bug: zlib uncompressed mode len/nlen was wrong endian
+ 0.57 fix bug: jpg last huffman symbol before marker was >9 bits but less than 16 available
+ 0.56 fix bug: zlib uncompressed mode len vs. nlen
+ 0.55 fix bug: restart_interval not initialized to 0
+ 0.54 allow NULL for 'int *comp'
+ 0.53 fix bug in png 3->4; speedup png decoding
+ 0.52 png handles req_comp=3,4 directly; minor cleanup; jpeg comments
+ 0.51 obey req_comp requests, 1-component jpegs return as 1-component,
+ on 'test' only check type, not whether we support this variant
+ 0.50 (2006-11-19)
+ first released version
+*/
+
+
+/*
+------------------------------------------------------------------------------
+This software is available under 2 licenses -- choose whichever you prefer.
+------------------------------------------------------------------------------
+ALTERNATIVE A - MIT License
+Copyright (c) 2017 Sean Barrett
+Permission is hereby granted, free of charge, to any person obtaining a copy of
+this software and associated documentation files (the "Software"), to deal in
+the Software without restriction, including without limitation the rights to
+use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
+of the Software, and to permit persons to whom the Software is furnished to do
+so, subject to the following conditions:
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+------------------------------------------------------------------------------
+ALTERNATIVE B - Public Domain (www.unlicense.org)
+This is free and unencumbered software released into the public domain.
+Anyone is free to copy, modify, publish, use, compile, sell, or distribute this
+software, either in source code form or as a compiled binary, for any purpose,
+commercial or non-commercial, and by any means.
+In jurisdictions that recognize copyright laws, the author or authors of this
+software dedicate any and all copyright interest in the software to the public
+domain. We make this dedication for the benefit of the public at large and to
+the detriment of our heirs and successors. We intend this dedication to be an
+overt act of relinquishment in perpetuity of all present and future rights to
+this software under copyright law.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+------------------------------------------------------------------------------
+*/
--- /dev/null
+/* stb_image_resize - v0.94 - public domain image resizing
+ by Jorge L Rodriguez (@VinoBS) - 2014
+ http://github.com/nothings/stb
+
+ Written with emphasis on usability, portability, and efficiency. (No
+ SIMD or threads, so it be easily outperformed by libs that use those.)
+ Only scaling and translation is supported, no rotations or shears.
+ Easy API downsamples w/Mitchell filter, upsamples w/cubic interpolation.
+
+ COMPILING & LINKING
+ In one C/C++ file that #includes this file, do this:
+ #define STB_IMAGE_RESIZE_IMPLEMENTATION
+ before the #include. That will create the implementation in that file.
+
+ QUICKSTART
+ stbir_resize_uint8( input_pixels , in_w , in_h , 0,
+ output_pixels, out_w, out_h, 0, num_channels)
+ stbir_resize_float(...)
+ stbir_resize_uint8_srgb( input_pixels , in_w , in_h , 0,
+ output_pixels, out_w, out_h, 0,
+ num_channels , alpha_chan , 0)
+ stbir_resize_uint8_srgb_edgemode(
+ input_pixels , in_w , in_h , 0,
+ output_pixels, out_w, out_h, 0,
+ num_channels , alpha_chan , 0, STBIR_EDGE_CLAMP)
+ // WRAP/REFLECT/ZERO
+
+ FULL API
+ See the "header file" section of the source for API documentation.
+
+ ADDITIONAL DOCUMENTATION
+
+ SRGB & FLOATING POINT REPRESENTATION
+ The sRGB functions presume IEEE floating point. If you do not have
+ IEEE floating point, define STBIR_NON_IEEE_FLOAT. This will use
+ a slower implementation.
+
+ MEMORY ALLOCATION
+ The resize functions here perform a single memory allocation using
+ malloc. To control the memory allocation, before the #include that
+ triggers the implementation, do:
+
+ #define STBIR_MALLOC(size,context) ...
+ #define STBIR_FREE(ptr,context) ...
+
+ Each resize function makes exactly one call to malloc/free, so to use
+ temp memory, store the temp memory in the context and return that.
+
+ ASSERT
+ Define STBIR_ASSERT(boolval) to override assert() and not use assert.h
+
+ OPTIMIZATION
+ Define STBIR_SATURATE_INT to compute clamp values in-range using
+ integer operations instead of float operations. This may be faster
+ on some platforms.
+
+ DEFAULT FILTERS
+ For functions which don't provide explicit control over what filters
+ to use, you can change the compile-time defaults with
+
+ #define STBIR_DEFAULT_FILTER_UPSAMPLE STBIR_FILTER_something
+ #define STBIR_DEFAULT_FILTER_DOWNSAMPLE STBIR_FILTER_something
+
+ See stbir_filter in the header-file section for the list of filters.
+
+ NEW FILTERS
+ A number of 1D filter kernels are used. For a list of
+ supported filters see the stbir_filter enum. To add a new filter,
+ write a filter function and add it to stbir__filter_info_table.
+
+ PROGRESS
+ For interactive use with slow resize operations, you can install
+ a progress-report callback:
+
+ #define STBIR_PROGRESS_REPORT(val) some_func(val)
+
+ The parameter val is a float which goes from 0 to 1 as progress is made.
+
+ For example:
+
+ static void my_progress_report(float progress);
+ #define STBIR_PROGRESS_REPORT(val) my_progress_report(val)
+
+ #define STB_IMAGE_RESIZE_IMPLEMENTATION
+ #include "stb_image_resize.h"
+
+ static void my_progress_report(float progress)
+ {
+ printf("Progress: %f%%\n", progress*100);
+ }
+
+ MAX CHANNELS
+ If your image has more than 64 channels, define STBIR_MAX_CHANNELS
+ to the max you'll have.
+
+ ALPHA CHANNEL
+ Most of the resizing functions provide the ability to control how
+ the alpha channel of an image is processed. The important things
+ to know about this:
+
+ 1. The best mathematically-behaved version of alpha to use is
+ called "premultiplied alpha", in which the other color channels
+ have had the alpha value multiplied in. If you use premultiplied
+ alpha, linear filtering (such as image resampling done by this
+ library, or performed in texture units on GPUs) does the "right
+ thing". While premultiplied alpha is standard in the movie CGI
+ industry, it is still uncommon in the videogame/real-time world.
+
+ If you linearly filter non-premultiplied alpha, strange effects
+ occur. (For example, the 50/50 average of 99% transparent bright green
+ and 1% transparent black produces 50% transparent dark green when
+ non-premultiplied, whereas premultiplied it produces 50%
+ transparent near-black. The former introduces green energy
+ that doesn't exist in the source image.)
+
+ 2. Artists should not edit premultiplied-alpha images; artists
+ want non-premultiplied alpha images. Thus, art tools generally output
+ non-premultiplied alpha images.
+
+ 3. You will get best results in most cases by converting images
+ to premultiplied alpha before processing them mathematically.
+
+ 4. If you pass the flag STBIR_FLAG_ALPHA_PREMULTIPLIED, the
+ resizer does not do anything special for the alpha channel;
+ it is resampled identically to other channels. This produces
+ the correct results for premultiplied-alpha images, but produces
+ less-than-ideal results for non-premultiplied-alpha images.
+
+ 5. If you do not pass the flag STBIR_FLAG_ALPHA_PREMULTIPLIED,
+ then the resizer weights the contribution of input pixels
+ based on their alpha values, or, equivalently, it multiplies
+ the alpha value into the color channels, resamples, then divides
+ by the resultant alpha value. Input pixels which have alpha=0 do
+ not contribute at all to output pixels unless _all_ of the input
+ pixels affecting that output pixel have alpha=0, in which case
+ the result for that pixel is the same as it would be without
+ STBIR_FLAG_ALPHA_PREMULTIPLIED. However, this is only true for
+ input images in integer formats. For input images in float format,
+ input pixels with alpha=0 have no effect, and output pixels
+ which have alpha=0 will be 0 in all channels. (For float images,
+ you can manually achieve the same result by adding a tiny epsilon
+ value to the alpha channel of every image, and then subtracting
+ or clamping it at the end.)
+
+ 6. You can suppress the behavior described in #5 and make
+ all-0-alpha pixels have 0 in all channels by #defining
+ STBIR_NO_ALPHA_EPSILON.
+
+ 7. You can separately control whether the alpha channel is
+ interpreted as linear or affected by the colorspace. By default
+ it is linear; you almost never want to apply the colorspace.
+ (For example, graphics hardware does not apply sRGB conversion
+ to the alpha channel.)
+
+ CONTRIBUTORS
+ Jorge L Rodriguez: Implementation
+ Sean Barrett: API design, optimizations
+ Aras Pranckevicius: bugfix
+
+ REVISIONS
+ 0.94 (2017-03-18) fixed warnings
+ 0.93 (2017-03-03) fixed bug with certain combinations of heights
+ 0.92 (2017-01-02) fix integer overflow on large (>2GB) images
+ 0.91 (2016-04-02) fix warnings; fix handling of subpixel regions
+ 0.90 (2014-09-17) first released version
+
+ LICENSE
+ See end of file for license information.
+
+ TODO
+ Don't decode all of the image data when only processing a partial tile
+ Don't use full-width decode buffers when only processing a partial tile
+ When processing wide images, break processing into tiles so data fits in L1 cache
+ Installable filters?
+ Resize that respects alpha test coverage
+ (Reference code: FloatImage::alphaTestCoverage and FloatImage::scaleAlphaToCoverage:
+ https://code.google.com/p/nvidia-texture-tools/source/browse/trunk/src/nvimage/FloatImage.cpp )
+*/
+
+#ifndef STBIR_INCLUDE_STB_IMAGE_RESIZE_H
+#define STBIR_INCLUDE_STB_IMAGE_RESIZE_H
+
+#ifdef _MSC_VER
+typedef unsigned char stbir_uint8;
+typedef unsigned short stbir_uint16;
+typedef unsigned int stbir_uint32;
+#else
+#include <stdint.h>
+typedef uint8_t stbir_uint8;
+typedef uint16_t stbir_uint16;
+typedef uint32_t stbir_uint32;
+#endif
+
+#ifdef STB_IMAGE_RESIZE_STATIC
+#define STBIRDEF static
+#else
+#ifdef __cplusplus
+#define STBIRDEF extern "C"
+#else
+#define STBIRDEF extern
+#endif
+#endif
+
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// Easy-to-use API:
+//
+// * "input pixels" points to an array of image data with 'num_channels' channels (e.g. RGB=3, RGBA=4)
+// * input_w is input image width (x-axis), input_h is input image height (y-axis)
+// * stride is the offset between successive rows of image data in memory, in bytes. you can
+// specify 0 to mean packed continuously in memory
+// * alpha channel is treated identically to other channels.
+// * colorspace is linear or sRGB as specified by function name
+// * returned result is 1 for success or 0 in case of an error.
+// #define STBIR_ASSERT() to trigger an assert on parameter validation errors.
+// * Memory required grows approximately linearly with input and output size, but with
+// discontinuities at input_w == output_w and input_h == output_h.
+// * These functions use a "default" resampling filter defined at compile time. To change the filter,
+// you can change the compile-time defaults by #defining STBIR_DEFAULT_FILTER_UPSAMPLE
+// and STBIR_DEFAULT_FILTER_DOWNSAMPLE, or you can use the medium-complexity API.
+
+STBIRDEF int stbir_resize_uint8( const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels);
+
+STBIRDEF int stbir_resize_float( const float *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ float *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels);
+
+
+// The following functions interpret image data as gamma-corrected sRGB.
+// Specify STBIR_ALPHA_CHANNEL_NONE if you have no alpha channel,
+// or otherwise provide the index of the alpha channel. Flags value
+// of 0 will probably do the right thing if you're not sure what
+// the flags mean.
+
+#define STBIR_ALPHA_CHANNEL_NONE -1
+
+// Set this flag if your texture has premultiplied alpha. Otherwise, stbir will
+// use alpha-weighted resampling (effectively premultiplying, resampling,
+// then unpremultiplying).
+#define STBIR_FLAG_ALPHA_PREMULTIPLIED (1 << 0)
+// The specified alpha channel should be handled as gamma-corrected value even
+// when doing sRGB operations.
+#define STBIR_FLAG_ALPHA_USES_COLORSPACE (1 << 1)
+
+STBIRDEF int stbir_resize_uint8_srgb(const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags);
+
+
+typedef enum
+{
+ STBIR_EDGE_CLAMP = 1,
+ STBIR_EDGE_REFLECT = 2,
+ STBIR_EDGE_WRAP = 3,
+ STBIR_EDGE_ZERO = 4,
+} stbir_edge;
+
+// This function adds the ability to specify how requests to sample off the edge of the image are handled.
+STBIRDEF int stbir_resize_uint8_srgb_edgemode(const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode);
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// Medium-complexity API
+//
+// This extends the easy-to-use API as follows:
+//
+// * Alpha-channel can be processed separately
+// * If alpha_channel is not STBIR_ALPHA_CHANNEL_NONE
+// * Alpha channel will not be gamma corrected (unless flags&STBIR_FLAG_GAMMA_CORRECT)
+// * Filters will be weighted by alpha channel (unless flags&STBIR_FLAG_ALPHA_PREMULTIPLIED)
+// * Filter can be selected explicitly
+// * uint16 image type
+// * sRGB colorspace available for all types
+// * context parameter for passing to STBIR_MALLOC
+
+typedef enum
+{
+ STBIR_FILTER_DEFAULT = 0, // use same filter type that easy-to-use API chooses
+ STBIR_FILTER_BOX = 1, // A trapezoid w/1-pixel wide ramps, same result as box for integer scale ratios
+ STBIR_FILTER_TRIANGLE = 2, // On upsampling, produces same results as bilinear texture filtering
+ STBIR_FILTER_CUBICBSPLINE = 3, // The cubic b-spline (aka Mitchell-Netrevalli with B=1,C=0), gaussian-esque
+ STBIR_FILTER_CATMULLROM = 4, // An interpolating cubic spline
+ STBIR_FILTER_MITCHELL = 5, // Mitchell-Netrevalli filter with B=1/3, C=1/3
+} stbir_filter;
+
+typedef enum
+{
+ STBIR_COLORSPACE_LINEAR,
+ STBIR_COLORSPACE_SRGB,
+
+ STBIR_MAX_COLORSPACES,
+} stbir_colorspace;
+
+// The following functions are all identical except for the type of the image data
+
+STBIRDEF int stbir_resize_uint8_generic( const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space,
+ void *alloc_context);
+
+STBIRDEF int stbir_resize_uint16_generic(const stbir_uint16 *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ stbir_uint16 *output_pixels , int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space,
+ void *alloc_context);
+
+STBIRDEF int stbir_resize_float_generic( const float *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ float *output_pixels , int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space,
+ void *alloc_context);
+
+
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// Full-complexity API
+//
+// This extends the medium API as follows:
+//
+// * uint32 image type
+// * not typesafe
+// * separate filter types for each axis
+// * separate edge modes for each axis
+// * can specify scale explicitly for subpixel correctness
+// * can specify image source tile using texture coordinates
+
+typedef enum
+{
+ STBIR_TYPE_UINT8 ,
+ STBIR_TYPE_UINT16,
+ STBIR_TYPE_UINT32,
+ STBIR_TYPE_FLOAT ,
+
+ STBIR_MAX_TYPES
+} stbir_datatype;
+
+STBIRDEF int stbir_resize( const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ stbir_datatype datatype,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical,
+ stbir_filter filter_horizontal, stbir_filter filter_vertical,
+ stbir_colorspace space, void *alloc_context);
+
+STBIRDEF int stbir_resize_subpixel(const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ stbir_datatype datatype,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical,
+ stbir_filter filter_horizontal, stbir_filter filter_vertical,
+ stbir_colorspace space, void *alloc_context,
+ float x_scale, float y_scale,
+ float x_offset, float y_offset);
+
+STBIRDEF int stbir_resize_region( const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ stbir_datatype datatype,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical,
+ stbir_filter filter_horizontal, stbir_filter filter_vertical,
+ stbir_colorspace space, void *alloc_context,
+ float s0, float t0, float s1, float t1);
+// (s0, t0) & (s1, t1) are the top-left and bottom right corner (uv addressing style: [0, 1]x[0, 1]) of a region of the input image to use.
+
+//
+//
+//// end header file /////////////////////////////////////////////////////
+#endif // STBIR_INCLUDE_STB_IMAGE_RESIZE_H
+
+
+
+
+
+#ifdef STB_IMAGE_RESIZE_IMPLEMENTATION
+
+#ifndef STBIR_ASSERT
+#include <assert.h>
+#define STBIR_ASSERT(x) assert(x)
+#endif
+
+// For memset
+#include <string.h>
+
+#include <math.h>
+
+#ifndef STBIR_MALLOC
+#include <stdlib.h>
+#define STBIR_MALLOC(size,c) malloc(size)
+#define STBIR_FREE(ptr,c) free(ptr)
+#endif
+
+#ifndef _MSC_VER
+#ifdef __cplusplus
+#define stbir__inline inline
+#else
+#define stbir__inline
+#endif
+#else
+#define stbir__inline __forceinline
+#endif
+
+
+// should produce compiler error if size is wrong
+typedef unsigned char stbir__validate_uint32[sizeof(stbir_uint32) == 4 ? 1 : -1];
+
+#ifdef _MSC_VER
+#define STBIR__NOTUSED(v) (void)(v)
+#else
+#define STBIR__NOTUSED(v) (void)sizeof(v)
+#endif
+
+#define STBIR__ARRAY_SIZE(a) (sizeof((a))/sizeof((a)[0]))
+
+#ifndef STBIR_DEFAULT_FILTER_UPSAMPLE
+#define STBIR_DEFAULT_FILTER_UPSAMPLE STBIR_FILTER_CATMULLROM
+#endif
+
+#ifndef STBIR_DEFAULT_FILTER_DOWNSAMPLE
+#define STBIR_DEFAULT_FILTER_DOWNSAMPLE STBIR_FILTER_MITCHELL
+#endif
+
+#ifndef STBIR_PROGRESS_REPORT
+#define STBIR_PROGRESS_REPORT(float_0_to_1)
+#endif
+
+#ifndef STBIR_MAX_CHANNELS
+#define STBIR_MAX_CHANNELS 64
+#endif
+
+#if STBIR_MAX_CHANNELS > 65536
+#error "Too many channels; STBIR_MAX_CHANNELS must be no more than 65536."
+// because we store the indices in 16-bit variables
+#endif
+
+// This value is added to alpha just before premultiplication to avoid
+// zeroing out color values. It is equivalent to 2^-80. If you don't want
+// that behavior (it may interfere if you have floating point images with
+// very small alpha values) then you can define STBIR_NO_ALPHA_EPSILON to
+// disable it.
+#ifndef STBIR_ALPHA_EPSILON
+#define STBIR_ALPHA_EPSILON ((float)1 / (1 << 20) / (1 << 20) / (1 << 20) / (1 << 20))
+#endif
+
+
+
+#ifdef _MSC_VER
+#define STBIR__UNUSED_PARAM(v) (void)(v)
+#else
+#define STBIR__UNUSED_PARAM(v) (void)sizeof(v)
+#endif
+
+// must match stbir_datatype
+static unsigned char stbir__type_size[] = {
+ 1, // STBIR_TYPE_UINT8
+ 2, // STBIR_TYPE_UINT16
+ 4, // STBIR_TYPE_UINT32
+ 4, // STBIR_TYPE_FLOAT
+};
+
+// Kernel function centered at 0
+typedef float (stbir__kernel_fn)(float x, float scale);
+typedef float (stbir__support_fn)(float scale);
+
+typedef struct
+{
+ stbir__kernel_fn* kernel;
+ stbir__support_fn* support;
+} stbir__filter_info;
+
+// When upsampling, the contributors are which source pixels contribute.
+// When downsampling, the contributors are which destination pixels are contributed to.
+typedef struct
+{
+ int n0; // First contributing pixel
+ int n1; // Last contributing pixel
+} stbir__contributors;
+
+typedef struct
+{
+ const void* input_data;
+ int input_w;
+ int input_h;
+ int input_stride_bytes;
+
+ void* output_data;
+ int output_w;
+ int output_h;
+ int output_stride_bytes;
+
+ float s0, t0, s1, t1;
+
+ float horizontal_shift; // Units: output pixels
+ float vertical_shift; // Units: output pixels
+ float horizontal_scale;
+ float vertical_scale;
+
+ int channels;
+ int alpha_channel;
+ stbir_uint32 flags;
+ stbir_datatype type;
+ stbir_filter horizontal_filter;
+ stbir_filter vertical_filter;
+ stbir_edge edge_horizontal;
+ stbir_edge edge_vertical;
+ stbir_colorspace colorspace;
+
+ stbir__contributors* horizontal_contributors;
+ float* horizontal_coefficients;
+
+ stbir__contributors* vertical_contributors;
+ float* vertical_coefficients;
+
+ int decode_buffer_pixels;
+ float* decode_buffer;
+
+ float* horizontal_buffer;
+
+ // cache these because ceil/floor are inexplicably showing up in profile
+ int horizontal_coefficient_width;
+ int vertical_coefficient_width;
+ int horizontal_filter_pixel_width;
+ int vertical_filter_pixel_width;
+ int horizontal_filter_pixel_margin;
+ int vertical_filter_pixel_margin;
+ int horizontal_num_contributors;
+ int vertical_num_contributors;
+
+ int ring_buffer_length_bytes; // The length of an individual entry in the ring buffer. The total number of ring buffers is stbir__get_filter_pixel_width(filter)
+ int ring_buffer_num_entries; // Total number of entries in the ring buffer.
+ int ring_buffer_first_scanline;
+ int ring_buffer_last_scanline;
+ int ring_buffer_begin_index; // first_scanline is at this index in the ring buffer
+ float* ring_buffer;
+
+ float* encode_buffer; // A temporary buffer to store floats so we don't lose precision while we do multiply-adds.
+
+ int horizontal_contributors_size;
+ int horizontal_coefficients_size;
+ int vertical_contributors_size;
+ int vertical_coefficients_size;
+ int decode_buffer_size;
+ int horizontal_buffer_size;
+ int ring_buffer_size;
+ int encode_buffer_size;
+} stbir__info;
+
+
+static const float stbir__max_uint8_as_float = 255.0f;
+static const float stbir__max_uint16_as_float = 65535.0f;
+static const double stbir__max_uint32_as_float = 4294967295.0;
+
+
+static stbir__inline int stbir__min(int a, int b)
+{
+ return a < b ? a : b;
+}
+
+static stbir__inline float stbir__saturate(float x)
+{
+ if (x < 0)
+ return 0;
+
+ if (x > 1)
+ return 1;
+
+ return x;
+}
+
+#ifdef STBIR_SATURATE_INT
+static stbir__inline stbir_uint8 stbir__saturate8(int x)
+{
+ if ((unsigned int) x <= 255)
+ return x;
+
+ if (x < 0)
+ return 0;
+
+ return 255;
+}
+
+static stbir__inline stbir_uint16 stbir__saturate16(int x)
+{
+ if ((unsigned int) x <= 65535)
+ return x;
+
+ if (x < 0)
+ return 0;
+
+ return 65535;
+}
+#endif
+
+static float stbir__srgb_uchar_to_linear_float[256] = {
+ 0.000000f, 0.000304f, 0.000607f, 0.000911f, 0.001214f, 0.001518f, 0.001821f, 0.002125f, 0.002428f, 0.002732f, 0.003035f,
+ 0.003347f, 0.003677f, 0.004025f, 0.004391f, 0.004777f, 0.005182f, 0.005605f, 0.006049f, 0.006512f, 0.006995f, 0.007499f,
+ 0.008023f, 0.008568f, 0.009134f, 0.009721f, 0.010330f, 0.010960f, 0.011612f, 0.012286f, 0.012983f, 0.013702f, 0.014444f,
+ 0.015209f, 0.015996f, 0.016807f, 0.017642f, 0.018500f, 0.019382f, 0.020289f, 0.021219f, 0.022174f, 0.023153f, 0.024158f,
+ 0.025187f, 0.026241f, 0.027321f, 0.028426f, 0.029557f, 0.030713f, 0.031896f, 0.033105f, 0.034340f, 0.035601f, 0.036889f,
+ 0.038204f, 0.039546f, 0.040915f, 0.042311f, 0.043735f, 0.045186f, 0.046665f, 0.048172f, 0.049707f, 0.051269f, 0.052861f,
+ 0.054480f, 0.056128f, 0.057805f, 0.059511f, 0.061246f, 0.063010f, 0.064803f, 0.066626f, 0.068478f, 0.070360f, 0.072272f,
+ 0.074214f, 0.076185f, 0.078187f, 0.080220f, 0.082283f, 0.084376f, 0.086500f, 0.088656f, 0.090842f, 0.093059f, 0.095307f,
+ 0.097587f, 0.099899f, 0.102242f, 0.104616f, 0.107023f, 0.109462f, 0.111932f, 0.114435f, 0.116971f, 0.119538f, 0.122139f,
+ 0.124772f, 0.127438f, 0.130136f, 0.132868f, 0.135633f, 0.138432f, 0.141263f, 0.144128f, 0.147027f, 0.149960f, 0.152926f,
+ 0.155926f, 0.158961f, 0.162029f, 0.165132f, 0.168269f, 0.171441f, 0.174647f, 0.177888f, 0.181164f, 0.184475f, 0.187821f,
+ 0.191202f, 0.194618f, 0.198069f, 0.201556f, 0.205079f, 0.208637f, 0.212231f, 0.215861f, 0.219526f, 0.223228f, 0.226966f,
+ 0.230740f, 0.234551f, 0.238398f, 0.242281f, 0.246201f, 0.250158f, 0.254152f, 0.258183f, 0.262251f, 0.266356f, 0.270498f,
+ 0.274677f, 0.278894f, 0.283149f, 0.287441f, 0.291771f, 0.296138f, 0.300544f, 0.304987f, 0.309469f, 0.313989f, 0.318547f,
+ 0.323143f, 0.327778f, 0.332452f, 0.337164f, 0.341914f, 0.346704f, 0.351533f, 0.356400f, 0.361307f, 0.366253f, 0.371238f,
+ 0.376262f, 0.381326f, 0.386430f, 0.391573f, 0.396755f, 0.401978f, 0.407240f, 0.412543f, 0.417885f, 0.423268f, 0.428691f,
+ 0.434154f, 0.439657f, 0.445201f, 0.450786f, 0.456411f, 0.462077f, 0.467784f, 0.473532f, 0.479320f, 0.485150f, 0.491021f,
+ 0.496933f, 0.502887f, 0.508881f, 0.514918f, 0.520996f, 0.527115f, 0.533276f, 0.539480f, 0.545725f, 0.552011f, 0.558340f,
+ 0.564712f, 0.571125f, 0.577581f, 0.584078f, 0.590619f, 0.597202f, 0.603827f, 0.610496f, 0.617207f, 0.623960f, 0.630757f,
+ 0.637597f, 0.644480f, 0.651406f, 0.658375f, 0.665387f, 0.672443f, 0.679543f, 0.686685f, 0.693872f, 0.701102f, 0.708376f,
+ 0.715694f, 0.723055f, 0.730461f, 0.737911f, 0.745404f, 0.752942f, 0.760525f, 0.768151f, 0.775822f, 0.783538f, 0.791298f,
+ 0.799103f, 0.806952f, 0.814847f, 0.822786f, 0.830770f, 0.838799f, 0.846873f, 0.854993f, 0.863157f, 0.871367f, 0.879622f,
+ 0.887923f, 0.896269f, 0.904661f, 0.913099f, 0.921582f, 0.930111f, 0.938686f, 0.947307f, 0.955974f, 0.964686f, 0.973445f,
+ 0.982251f, 0.991102f, 1.0f
+};
+
+static float stbir__srgb_to_linear(float f)
+{
+ if (f <= 0.04045f)
+ return f / 12.92f;
+ else
+ return (float)pow((f + 0.055f) / 1.055f, 2.4f);
+}
+
+static float stbir__linear_to_srgb(float f)
+{
+ if (f <= 0.0031308f)
+ return f * 12.92f;
+ else
+ return 1.055f * (float)pow(f, 1 / 2.4f) - 0.055f;
+}
+
+#ifndef STBIR_NON_IEEE_FLOAT
+// From https://gist.github.com/rygorous/2203834
+
+typedef union
+{
+ stbir_uint32 u;
+ float f;
+} stbir__FP32;
+
+static const stbir_uint32 fp32_to_srgb8_tab4[104] = {
+ 0x0073000d, 0x007a000d, 0x0080000d, 0x0087000d, 0x008d000d, 0x0094000d, 0x009a000d, 0x00a1000d,
+ 0x00a7001a, 0x00b4001a, 0x00c1001a, 0x00ce001a, 0x00da001a, 0x00e7001a, 0x00f4001a, 0x0101001a,
+ 0x010e0033, 0x01280033, 0x01410033, 0x015b0033, 0x01750033, 0x018f0033, 0x01a80033, 0x01c20033,
+ 0x01dc0067, 0x020f0067, 0x02430067, 0x02760067, 0x02aa0067, 0x02dd0067, 0x03110067, 0x03440067,
+ 0x037800ce, 0x03df00ce, 0x044600ce, 0x04ad00ce, 0x051400ce, 0x057b00c5, 0x05dd00bc, 0x063b00b5,
+ 0x06970158, 0x07420142, 0x07e30130, 0x087b0120, 0x090b0112, 0x09940106, 0x0a1700fc, 0x0a9500f2,
+ 0x0b0f01cb, 0x0bf401ae, 0x0ccb0195, 0x0d950180, 0x0e56016e, 0x0f0d015e, 0x0fbc0150, 0x10630143,
+ 0x11070264, 0x1238023e, 0x1357021d, 0x14660201, 0x156601e9, 0x165a01d3, 0x174401c0, 0x182401af,
+ 0x18fe0331, 0x1a9602fe, 0x1c1502d2, 0x1d7e02ad, 0x1ed4028d, 0x201a0270, 0x21520256, 0x227d0240,
+ 0x239f0443, 0x25c003fe, 0x27bf03c4, 0x29a10392, 0x2b6a0367, 0x2d1d0341, 0x2ebe031f, 0x304d0300,
+ 0x31d105b0, 0x34a80555, 0x37520507, 0x39d504c5, 0x3c37048b, 0x3e7c0458, 0x40a8042a, 0x42bd0401,
+ 0x44c20798, 0x488e071e, 0x4c1c06b6, 0x4f76065d, 0x52a50610, 0x55ac05cc, 0x5892058f, 0x5b590559,
+ 0x5e0c0a23, 0x631c0980, 0x67db08f6, 0x6c55087f, 0x70940818, 0x74a007bd, 0x787d076c, 0x7c330723,
+};
+
+static stbir_uint8 stbir__linear_to_srgb_uchar(float in)
+{
+ static const stbir__FP32 almostone = { 0x3f7fffff }; // 1-eps
+ static const stbir__FP32 minval = { (127-13) << 23 };
+ stbir_uint32 tab,bias,scale,t;
+ stbir__FP32 f;
+
+ // Clamp to [2^(-13), 1-eps]; these two values map to 0 and 1, respectively.
+ // The tests are carefully written so that NaNs map to 0, same as in the reference
+ // implementation.
+ if (!(in > minval.f)) // written this way to catch NaNs
+ in = minval.f;
+ if (in > almostone.f)
+ in = almostone.f;
+
+ // Do the table lookup and unpack bias, scale
+ f.f = in;
+ tab = fp32_to_srgb8_tab4[(f.u - minval.u) >> 20];
+ bias = (tab >> 16) << 9;
+ scale = tab & 0xffff;
+
+ // Grab next-highest mantissa bits and perform linear interpolation
+ t = (f.u >> 12) & 0xff;
+ return (unsigned char) ((bias + scale*t) >> 16);
+}
+
+#else
+// sRGB transition values, scaled by 1<<28
+static int stbir__srgb_offset_to_linear_scaled[256] =
+{
+ 0, 40738, 122216, 203693, 285170, 366648, 448125, 529603,
+ 611080, 692557, 774035, 855852, 942009, 1033024, 1128971, 1229926,
+ 1335959, 1447142, 1563542, 1685229, 1812268, 1944725, 2082664, 2226148,
+ 2375238, 2529996, 2690481, 2856753, 3028870, 3206888, 3390865, 3580856,
+ 3776916, 3979100, 4187460, 4402049, 4622919, 4850123, 5083710, 5323731,
+ 5570236, 5823273, 6082892, 6349140, 6622065, 6901714, 7188133, 7481369,
+ 7781466, 8088471, 8402427, 8723380, 9051372, 9386448, 9728650, 10078021,
+ 10434603, 10798439, 11169569, 11548036, 11933879, 12327139, 12727857, 13136073,
+ 13551826, 13975156, 14406100, 14844697, 15290987, 15745007, 16206795, 16676389,
+ 17153826, 17639142, 18132374, 18633560, 19142734, 19659934, 20185196, 20718552,
+ 21260042, 21809696, 22367554, 22933648, 23508010, 24090680, 24681686, 25281066,
+ 25888850, 26505076, 27129772, 27762974, 28404716, 29055026, 29713942, 30381490,
+ 31057708, 31742624, 32436272, 33138682, 33849884, 34569912, 35298800, 36036568,
+ 36783260, 37538896, 38303512, 39077136, 39859796, 40651528, 41452360, 42262316,
+ 43081432, 43909732, 44747252, 45594016, 46450052, 47315392, 48190064, 49074096,
+ 49967516, 50870356, 51782636, 52704392, 53635648, 54576432, 55526772, 56486700,
+ 57456236, 58435408, 59424248, 60422780, 61431036, 62449032, 63476804, 64514376,
+ 65561776, 66619028, 67686160, 68763192, 69850160, 70947088, 72053992, 73170912,
+ 74297864, 75434880, 76581976, 77739184, 78906536, 80084040, 81271736, 82469648,
+ 83677792, 84896192, 86124888, 87363888, 88613232, 89872928, 91143016, 92423512,
+ 93714432, 95015816, 96327688, 97650056, 98982952, 100326408, 101680440, 103045072,
+ 104420320, 105806224, 107202800, 108610064, 110028048, 111456776, 112896264, 114346544,
+ 115807632, 117279552, 118762328, 120255976, 121760536, 123276016, 124802440, 126339832,
+ 127888216, 129447616, 131018048, 132599544, 134192112, 135795792, 137410592, 139036528,
+ 140673648, 142321952, 143981456, 145652208, 147334208, 149027488, 150732064, 152447968,
+ 154175200, 155913792, 157663776, 159425168, 161197984, 162982240, 164777968, 166585184,
+ 168403904, 170234160, 172075968, 173929344, 175794320, 177670896, 179559120, 181458992,
+ 183370528, 185293776, 187228736, 189175424, 191133888, 193104112, 195086128, 197079968,
+ 199085648, 201103184, 203132592, 205173888, 207227120, 209292272, 211369392, 213458480,
+ 215559568, 217672656, 219797792, 221934976, 224084240, 226245600, 228419056, 230604656,
+ 232802400, 235012320, 237234432, 239468736, 241715280, 243974080, 246245120, 248528464,
+ 250824112, 253132064, 255452368, 257785040, 260130080, 262487520, 264857376, 267239664,
+};
+
+static stbir_uint8 stbir__linear_to_srgb_uchar(float f)
+{
+ int x = (int) (f * (1 << 28)); // has headroom so you don't need to clamp
+ int v = 0;
+ int i;
+
+ // Refine the guess with a short binary search.
+ i = v + 128; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+ i = v + 64; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+ i = v + 32; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+ i = v + 16; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+ i = v + 8; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+ i = v + 4; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+ i = v + 2; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+ i = v + 1; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+
+ return (stbir_uint8) v;
+}
+#endif
+
+static float stbir__filter_trapezoid(float x, float scale)
+{
+ float halfscale = scale / 2;
+ float t = 0.5f + halfscale;
+ STBIR_ASSERT(scale <= 1);
+
+ x = (float)fabs(x);
+
+ if (x >= t)
+ return 0;
+ else
+ {
+ float r = 0.5f - halfscale;
+ if (x <= r)
+ return 1;
+ else
+ return (t - x) / scale;
+ }
+}
+
+static float stbir__support_trapezoid(float scale)
+{
+ STBIR_ASSERT(scale <= 1);
+ return 0.5f + scale / 2;
+}
+
+static float stbir__filter_triangle(float x, float s)
+{
+ STBIR__UNUSED_PARAM(s);
+
+ x = (float)fabs(x);
+
+ if (x <= 1.0f)
+ return 1 - x;
+ else
+ return 0;
+}
+
+static float stbir__filter_cubic(float x, float s)
+{
+ STBIR__UNUSED_PARAM(s);
+
+ x = (float)fabs(x);
+
+ if (x < 1.0f)
+ return (4 + x*x*(3*x - 6))/6;
+ else if (x < 2.0f)
+ return (8 + x*(-12 + x*(6 - x)))/6;
+
+ return (0.0f);
+}
+
+static float stbir__filter_catmullrom(float x, float s)
+{
+ STBIR__UNUSED_PARAM(s);
+
+ x = (float)fabs(x);
+
+ if (x < 1.0f)
+ return 1 - x*x*(2.5f - 1.5f*x);
+ else if (x < 2.0f)
+ return 2 - x*(4 + x*(0.5f*x - 2.5f));
+
+ return (0.0f);
+}
+
+static float stbir__filter_mitchell(float x, float s)
+{
+ STBIR__UNUSED_PARAM(s);
+
+ x = (float)fabs(x);
+
+ if (x < 1.0f)
+ return (16 + x*x*(21 * x - 36))/18;
+ else if (x < 2.0f)
+ return (32 + x*(-60 + x*(36 - 7*x)))/18;
+
+ return (0.0f);
+}
+
+static float stbir__support_zero(float s)
+{
+ STBIR__UNUSED_PARAM(s);
+ return 0;
+}
+
+static float stbir__support_one(float s)
+{
+ STBIR__UNUSED_PARAM(s);
+ return 1;
+}
+
+static float stbir__support_two(float s)
+{
+ STBIR__UNUSED_PARAM(s);
+ return 2;
+}
+
+static stbir__filter_info stbir__filter_info_table[] = {
+ { NULL, stbir__support_zero },
+ { stbir__filter_trapezoid, stbir__support_trapezoid },
+ { stbir__filter_triangle, stbir__support_one },
+ { stbir__filter_cubic, stbir__support_two },
+ { stbir__filter_catmullrom, stbir__support_two },
+ { stbir__filter_mitchell, stbir__support_two },
+};
+
+stbir__inline static int stbir__use_upsampling(float ratio)
+{
+ return ratio > 1;
+}
+
+stbir__inline static int stbir__use_width_upsampling(stbir__info* stbir_info)
+{
+ return stbir__use_upsampling(stbir_info->horizontal_scale);
+}
+
+stbir__inline static int stbir__use_height_upsampling(stbir__info* stbir_info)
+{
+ return stbir__use_upsampling(stbir_info->vertical_scale);
+}
+
+// This is the maximum number of input samples that can affect an output sample
+// with the given filter
+static int stbir__get_filter_pixel_width(stbir_filter filter, float scale)
+{
+ STBIR_ASSERT(filter != 0);
+ STBIR_ASSERT(filter < STBIR__ARRAY_SIZE(stbir__filter_info_table));
+
+ if (stbir__use_upsampling(scale))
+ return (int)ceil(stbir__filter_info_table[filter].support(1/scale) * 2);
+ else
+ return (int)ceil(stbir__filter_info_table[filter].support(scale) * 2 / scale);
+}
+
+// This is how much to expand buffers to account for filters seeking outside
+// the image boundaries.
+static int stbir__get_filter_pixel_margin(stbir_filter filter, float scale)
+{
+ return stbir__get_filter_pixel_width(filter, scale) / 2;
+}
+
+static int stbir__get_coefficient_width(stbir_filter filter, float scale)
+{
+ if (stbir__use_upsampling(scale))
+ return (int)ceil(stbir__filter_info_table[filter].support(1 / scale) * 2);
+ else
+ return (int)ceil(stbir__filter_info_table[filter].support(scale) * 2);
+}
+
+static int stbir__get_contributors(float scale, stbir_filter filter, int input_size, int output_size)
+{
+ if (stbir__use_upsampling(scale))
+ return output_size;
+ else
+ return (input_size + stbir__get_filter_pixel_margin(filter, scale) * 2);
+}
+
+static int stbir__get_total_horizontal_coefficients(stbir__info* info)
+{
+ return info->horizontal_num_contributors
+ * stbir__get_coefficient_width (info->horizontal_filter, info->horizontal_scale);
+}
+
+static int stbir__get_total_vertical_coefficients(stbir__info* info)
+{
+ return info->vertical_num_contributors
+ * stbir__get_coefficient_width (info->vertical_filter, info->vertical_scale);
+}
+
+static stbir__contributors* stbir__get_contributor(stbir__contributors* contributors, int n)
+{
+ return &contributors[n];
+}
+
+// For perf reasons this code is duplicated in stbir__resample_horizontal_upsample/downsample,
+// if you change it here change it there too.
+static float* stbir__get_coefficient(float* coefficients, stbir_filter filter, float scale, int n, int c)
+{
+ int width = stbir__get_coefficient_width(filter, scale);
+ return &coefficients[width*n + c];
+}
+
+static int stbir__edge_wrap_slow(stbir_edge edge, int n, int max)
+{
+ switch (edge)
+ {
+ case STBIR_EDGE_ZERO:
+ return 0; // we'll decode the wrong pixel here, and then overwrite with 0s later
+
+ case STBIR_EDGE_CLAMP:
+ if (n < 0)
+ return 0;
+
+ if (n >= max)
+ return max - 1;
+
+ return n; // NOTREACHED
+
+ case STBIR_EDGE_REFLECT:
+ {
+ if (n < 0)
+ {
+ if (n < max)
+ return -n;
+ else
+ return max - 1;
+ }
+
+ if (n >= max)
+ {
+ int max2 = max * 2;
+ if (n >= max2)
+ return 0;
+ else
+ return max2 - n - 1;
+ }
+
+ return n; // NOTREACHED
+ }
+
+ case STBIR_EDGE_WRAP:
+ if (n >= 0)
+ return (n % max);
+ else
+ {
+ int m = (-n) % max;
+
+ if (m != 0)
+ m = max - m;
+
+ return (m);
+ }
+ return n; // NOTREACHED
+
+ default:
+ STBIR_ASSERT(!"Unimplemented edge type");
+ return 0;
+ }
+}
+
+stbir__inline static int stbir__edge_wrap(stbir_edge edge, int n, int max)
+{
+ // avoid per-pixel switch
+ if (n >= 0 && n < max)
+ return n;
+ return stbir__edge_wrap_slow(edge, n, max);
+}
+
+// What input pixels contribute to this output pixel?
+static void stbir__calculate_sample_range_upsample(int n, float out_filter_radius, float scale_ratio, float out_shift, int* in_first_pixel, int* in_last_pixel, float* in_center_of_out)
+{
+ float out_pixel_center = (float)n + 0.5f;
+ float out_pixel_influence_lowerbound = out_pixel_center - out_filter_radius;
+ float out_pixel_influence_upperbound = out_pixel_center + out_filter_radius;
+
+ float in_pixel_influence_lowerbound = (out_pixel_influence_lowerbound + out_shift) / scale_ratio;
+ float in_pixel_influence_upperbound = (out_pixel_influence_upperbound + out_shift) / scale_ratio;
+
+ *in_center_of_out = (out_pixel_center + out_shift) / scale_ratio;
+ *in_first_pixel = (int)(floor(in_pixel_influence_lowerbound + 0.5));
+ *in_last_pixel = (int)(floor(in_pixel_influence_upperbound - 0.5));
+}
+
+// What output pixels does this input pixel contribute to?
+static void stbir__calculate_sample_range_downsample(int n, float in_pixels_radius, float scale_ratio, float out_shift, int* out_first_pixel, int* out_last_pixel, float* out_center_of_in)
+{
+ float in_pixel_center = (float)n + 0.5f;
+ float in_pixel_influence_lowerbound = in_pixel_center - in_pixels_radius;
+ float in_pixel_influence_upperbound = in_pixel_center + in_pixels_radius;
+
+ float out_pixel_influence_lowerbound = in_pixel_influence_lowerbound * scale_ratio - out_shift;
+ float out_pixel_influence_upperbound = in_pixel_influence_upperbound * scale_ratio - out_shift;
+
+ *out_center_of_in = in_pixel_center * scale_ratio - out_shift;
+ *out_first_pixel = (int)(floor(out_pixel_influence_lowerbound + 0.5));
+ *out_last_pixel = (int)(floor(out_pixel_influence_upperbound - 0.5));
+}
+
+static void stbir__calculate_coefficients_upsample(stbir_filter filter, float scale, int in_first_pixel, int in_last_pixel, float in_center_of_out, stbir__contributors* contributor, float* coefficient_group)
+{
+ int i;
+ float total_filter = 0;
+ float filter_scale;
+
+ STBIR_ASSERT(in_last_pixel - in_first_pixel <= (int)ceil(stbir__filter_info_table[filter].support(1/scale) * 2)); // Taken directly from stbir__get_coefficient_width() which we can't call because we don't know if we're horizontal or vertical.
+
+ contributor->n0 = in_first_pixel;
+ contributor->n1 = in_last_pixel;
+
+ STBIR_ASSERT(contributor->n1 >= contributor->n0);
+
+ for (i = 0; i <= in_last_pixel - in_first_pixel; i++)
+ {
+ float in_pixel_center = (float)(i + in_first_pixel) + 0.5f;
+ coefficient_group[i] = stbir__filter_info_table[filter].kernel(in_center_of_out - in_pixel_center, 1 / scale);
+
+ // If the coefficient is zero, skip it. (Don't do the <0 check here, we want the influence of those outside pixels.)
+ if (i == 0 && !coefficient_group[i])
+ {
+ contributor->n0 = ++in_first_pixel;
+ i--;
+ continue;
+ }
+
+ total_filter += coefficient_group[i];
+ }
+
+ STBIR_ASSERT(stbir__filter_info_table[filter].kernel((float)(in_last_pixel + 1) + 0.5f - in_center_of_out, 1/scale) == 0);
+
+ STBIR_ASSERT(total_filter > 0.9);
+ STBIR_ASSERT(total_filter < 1.1f); // Make sure it's not way off.
+
+ // Make sure the sum of all coefficients is 1.
+ filter_scale = 1 / total_filter;
+
+ for (i = 0; i <= in_last_pixel - in_first_pixel; i++)
+ coefficient_group[i] *= filter_scale;
+
+ for (i = in_last_pixel - in_first_pixel; i >= 0; i--)
+ {
+ if (coefficient_group[i])
+ break;
+
+ // This line has no weight. We can skip it.
+ contributor->n1 = contributor->n0 + i - 1;
+ }
+}
+
+static void stbir__calculate_coefficients_downsample(stbir_filter filter, float scale_ratio, int out_first_pixel, int out_last_pixel, float out_center_of_in, stbir__contributors* contributor, float* coefficient_group)
+{
+ int i;
+
+ STBIR_ASSERT(out_last_pixel - out_first_pixel <= (int)ceil(stbir__filter_info_table[filter].support(scale_ratio) * 2)); // Taken directly from stbir__get_coefficient_width() which we can't call because we don't know if we're horizontal or vertical.
+
+ contributor->n0 = out_first_pixel;
+ contributor->n1 = out_last_pixel;
+
+ STBIR_ASSERT(contributor->n1 >= contributor->n0);
+
+ for (i = 0; i <= out_last_pixel - out_first_pixel; i++)
+ {
+ float out_pixel_center = (float)(i + out_first_pixel) + 0.5f;
+ float x = out_pixel_center - out_center_of_in;
+ coefficient_group[i] = stbir__filter_info_table[filter].kernel(x, scale_ratio) * scale_ratio;
+ }
+
+ STBIR_ASSERT(stbir__filter_info_table[filter].kernel((float)(out_last_pixel + 1) + 0.5f - out_center_of_in, scale_ratio) == 0);
+
+ for (i = out_last_pixel - out_first_pixel; i >= 0; i--)
+ {
+ if (coefficient_group[i])
+ break;
+
+ // This line has no weight. We can skip it.
+ contributor->n1 = contributor->n0 + i - 1;
+ }
+}
+
+static void stbir__normalize_downsample_coefficients(stbir__contributors* contributors, float* coefficients, stbir_filter filter, float scale_ratio, int input_size, int output_size)
+{
+ int num_contributors = stbir__get_contributors(scale_ratio, filter, input_size, output_size);
+ int num_coefficients = stbir__get_coefficient_width(filter, scale_ratio);
+ int i, j;
+ int skip;
+
+ for (i = 0; i < output_size; i++)
+ {
+ float scale;
+ float total = 0;
+
+ for (j = 0; j < num_contributors; j++)
+ {
+ if (i >= contributors[j].n0 && i <= contributors[j].n1)
+ {
+ float coefficient = *stbir__get_coefficient(coefficients, filter, scale_ratio, j, i - contributors[j].n0);
+ total += coefficient;
+ }
+ else if (i < contributors[j].n0)
+ break;
+ }
+
+ STBIR_ASSERT(total > 0.9f);
+ STBIR_ASSERT(total < 1.1f);
+
+ scale = 1 / total;
+
+ for (j = 0; j < num_contributors; j++)
+ {
+ if (i >= contributors[j].n0 && i <= contributors[j].n1)
+ *stbir__get_coefficient(coefficients, filter, scale_ratio, j, i - contributors[j].n0) *= scale;
+ else if (i < contributors[j].n0)
+ break;
+ }
+ }
+
+ // Optimize: Skip zero coefficients and contributions outside of image bounds.
+ // Do this after normalizing because normalization depends on the n0/n1 values.
+ for (j = 0; j < num_contributors; j++)
+ {
+ int range, max, width;
+
+ skip = 0;
+ while (*stbir__get_coefficient(coefficients, filter, scale_ratio, j, skip) == 0)
+ skip++;
+
+ contributors[j].n0 += skip;
+
+ while (contributors[j].n0 < 0)
+ {
+ contributors[j].n0++;
+ skip++;
+ }
+
+ range = contributors[j].n1 - contributors[j].n0 + 1;
+ max = stbir__min(num_coefficients, range);
+
+ width = stbir__get_coefficient_width(filter, scale_ratio);
+ for (i = 0; i < max; i++)
+ {
+ if (i + skip >= width)
+ break;
+
+ *stbir__get_coefficient(coefficients, filter, scale_ratio, j, i) = *stbir__get_coefficient(coefficients, filter, scale_ratio, j, i + skip);
+ }
+
+ continue;
+ }
+
+ // Using min to avoid writing into invalid pixels.
+ for (i = 0; i < num_contributors; i++)
+ contributors[i].n1 = stbir__min(contributors[i].n1, output_size - 1);
+}
+
+// Each scan line uses the same kernel values so we should calculate the kernel
+// values once and then we can use them for every scan line.
+static void stbir__calculate_filters(stbir__contributors* contributors, float* coefficients, stbir_filter filter, float scale_ratio, float shift, int input_size, int output_size)
+{
+ int n;
+ int total_contributors = stbir__get_contributors(scale_ratio, filter, input_size, output_size);
+
+ if (stbir__use_upsampling(scale_ratio))
+ {
+ float out_pixels_radius = stbir__filter_info_table[filter].support(1 / scale_ratio) * scale_ratio;
+
+ // Looping through out pixels
+ for (n = 0; n < total_contributors; n++)
+ {
+ float in_center_of_out; // Center of the current out pixel in the in pixel space
+ int in_first_pixel, in_last_pixel;
+
+ stbir__calculate_sample_range_upsample(n, out_pixels_radius, scale_ratio, shift, &in_first_pixel, &in_last_pixel, &in_center_of_out);
+
+ stbir__calculate_coefficients_upsample(filter, scale_ratio, in_first_pixel, in_last_pixel, in_center_of_out, stbir__get_contributor(contributors, n), stbir__get_coefficient(coefficients, filter, scale_ratio, n, 0));
+ }
+ }
+ else
+ {
+ float in_pixels_radius = stbir__filter_info_table[filter].support(scale_ratio) / scale_ratio;
+
+ // Looping through in pixels
+ for (n = 0; n < total_contributors; n++)
+ {
+ float out_center_of_in; // Center of the current out pixel in the in pixel space
+ int out_first_pixel, out_last_pixel;
+ int n_adjusted = n - stbir__get_filter_pixel_margin(filter, scale_ratio);
+
+ stbir__calculate_sample_range_downsample(n_adjusted, in_pixels_radius, scale_ratio, shift, &out_first_pixel, &out_last_pixel, &out_center_of_in);
+
+ stbir__calculate_coefficients_downsample(filter, scale_ratio, out_first_pixel, out_last_pixel, out_center_of_in, stbir__get_contributor(contributors, n), stbir__get_coefficient(coefficients, filter, scale_ratio, n, 0));
+ }
+
+ stbir__normalize_downsample_coefficients(contributors, coefficients, filter, scale_ratio, input_size, output_size);
+ }
+}
+
+static float* stbir__get_decode_buffer(stbir__info* stbir_info)
+{
+ // The 0 index of the decode buffer starts after the margin. This makes
+ // it okay to use negative indexes on the decode buffer.
+ return &stbir_info->decode_buffer[stbir_info->horizontal_filter_pixel_margin * stbir_info->channels];
+}
+
+#define STBIR__DECODE(type, colorspace) ((type) * (STBIR_MAX_COLORSPACES) + (colorspace))
+
+static void stbir__decode_scanline(stbir__info* stbir_info, int n)
+{
+ int c;
+ int channels = stbir_info->channels;
+ int alpha_channel = stbir_info->alpha_channel;
+ int type = stbir_info->type;
+ int colorspace = stbir_info->colorspace;
+ int input_w = stbir_info->input_w;
+ size_t input_stride_bytes = stbir_info->input_stride_bytes;
+ float* decode_buffer = stbir__get_decode_buffer(stbir_info);
+ stbir_edge edge_horizontal = stbir_info->edge_horizontal;
+ stbir_edge edge_vertical = stbir_info->edge_vertical;
+ size_t in_buffer_row_offset = stbir__edge_wrap(edge_vertical, n, stbir_info->input_h) * input_stride_bytes;
+ const void* input_data = (char *) stbir_info->input_data + in_buffer_row_offset;
+ int max_x = input_w + stbir_info->horizontal_filter_pixel_margin;
+ int decode = STBIR__DECODE(type, colorspace);
+
+ int x = -stbir_info->horizontal_filter_pixel_margin;
+
+ // special handling for STBIR_EDGE_ZERO because it needs to return an item that doesn't appear in the input,
+ // and we want to avoid paying overhead on every pixel if not STBIR_EDGE_ZERO
+ if (edge_vertical == STBIR_EDGE_ZERO && (n < 0 || n >= stbir_info->input_h))
+ {
+ for (; x < max_x; x++)
+ for (c = 0; c < channels; c++)
+ decode_buffer[x*channels + c] = 0;
+ return;
+ }
+
+ switch (decode)
+ {
+ case STBIR__DECODE(STBIR_TYPE_UINT8, STBIR_COLORSPACE_LINEAR):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = ((float)((const unsigned char*)input_data)[input_pixel_index + c]) / stbir__max_uint8_as_float;
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT8, STBIR_COLORSPACE_SRGB):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = stbir__srgb_uchar_to_linear_float[((const unsigned char*)input_data)[input_pixel_index + c]];
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ decode_buffer[decode_pixel_index + alpha_channel] = ((float)((const unsigned char*)input_data)[input_pixel_index + alpha_channel]) / stbir__max_uint8_as_float;
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT16, STBIR_COLORSPACE_LINEAR):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = ((float)((const unsigned short*)input_data)[input_pixel_index + c]) / stbir__max_uint16_as_float;
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT16, STBIR_COLORSPACE_SRGB):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = stbir__srgb_to_linear(((float)((const unsigned short*)input_data)[input_pixel_index + c]) / stbir__max_uint16_as_float);
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ decode_buffer[decode_pixel_index + alpha_channel] = ((float)((const unsigned short*)input_data)[input_pixel_index + alpha_channel]) / stbir__max_uint16_as_float;
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT32, STBIR_COLORSPACE_LINEAR):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = (float)(((double)((const unsigned int*)input_data)[input_pixel_index + c]) / stbir__max_uint32_as_float);
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT32, STBIR_COLORSPACE_SRGB):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = stbir__srgb_to_linear((float)(((double)((const unsigned int*)input_data)[input_pixel_index + c]) / stbir__max_uint32_as_float));
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ decode_buffer[decode_pixel_index + alpha_channel] = (float)(((double)((const unsigned int*)input_data)[input_pixel_index + alpha_channel]) / stbir__max_uint32_as_float);
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_FLOAT, STBIR_COLORSPACE_LINEAR):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = ((const float*)input_data)[input_pixel_index + c];
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_FLOAT, STBIR_COLORSPACE_SRGB):
+ for (; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+ int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+ for (c = 0; c < channels; c++)
+ decode_buffer[decode_pixel_index + c] = stbir__srgb_to_linear(((const float*)input_data)[input_pixel_index + c]);
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ decode_buffer[decode_pixel_index + alpha_channel] = ((const float*)input_data)[input_pixel_index + alpha_channel];
+ }
+
+ break;
+
+ default:
+ STBIR_ASSERT(!"Unknown type/colorspace/channels combination.");
+ break;
+ }
+
+ if (!(stbir_info->flags & STBIR_FLAG_ALPHA_PREMULTIPLIED))
+ {
+ for (x = -stbir_info->horizontal_filter_pixel_margin; x < max_x; x++)
+ {
+ int decode_pixel_index = x * channels;
+
+ // If the alpha value is 0 it will clobber the color values. Make sure it's not.
+ float alpha = decode_buffer[decode_pixel_index + alpha_channel];
+#ifndef STBIR_NO_ALPHA_EPSILON
+ if (stbir_info->type != STBIR_TYPE_FLOAT) {
+ alpha += STBIR_ALPHA_EPSILON;
+ decode_buffer[decode_pixel_index + alpha_channel] = alpha;
+ }
+#endif
+ for (c = 0; c < channels; c++)
+ {
+ if (c == alpha_channel)
+ continue;
+
+ decode_buffer[decode_pixel_index + c] *= alpha;
+ }
+ }
+ }
+
+ if (edge_horizontal == STBIR_EDGE_ZERO)
+ {
+ for (x = -stbir_info->horizontal_filter_pixel_margin; x < 0; x++)
+ {
+ for (c = 0; c < channels; c++)
+ decode_buffer[x*channels + c] = 0;
+ }
+ for (x = input_w; x < max_x; x++)
+ {
+ for (c = 0; c < channels; c++)
+ decode_buffer[x*channels + c] = 0;
+ }
+ }
+}
+
+static float* stbir__get_ring_buffer_entry(float* ring_buffer, int index, int ring_buffer_length)
+{
+ return &ring_buffer[index * ring_buffer_length];
+}
+
+static float* stbir__add_empty_ring_buffer_entry(stbir__info* stbir_info, int n)
+{
+ int ring_buffer_index;
+ float* ring_buffer;
+
+ stbir_info->ring_buffer_last_scanline = n;
+
+ if (stbir_info->ring_buffer_begin_index < 0)
+ {
+ ring_buffer_index = stbir_info->ring_buffer_begin_index = 0;
+ stbir_info->ring_buffer_first_scanline = n;
+ }
+ else
+ {
+ ring_buffer_index = (stbir_info->ring_buffer_begin_index + (stbir_info->ring_buffer_last_scanline - stbir_info->ring_buffer_first_scanline)) % stbir_info->ring_buffer_num_entries;
+ STBIR_ASSERT(ring_buffer_index != stbir_info->ring_buffer_begin_index);
+ }
+
+ ring_buffer = stbir__get_ring_buffer_entry(stbir_info->ring_buffer, ring_buffer_index, stbir_info->ring_buffer_length_bytes / sizeof(float));
+ memset(ring_buffer, 0, stbir_info->ring_buffer_length_bytes);
+
+ return ring_buffer;
+}
+
+
+static void stbir__resample_horizontal_upsample(stbir__info* stbir_info, float* output_buffer)
+{
+ int x, k;
+ int output_w = stbir_info->output_w;
+ int channels = stbir_info->channels;
+ float* decode_buffer = stbir__get_decode_buffer(stbir_info);
+ stbir__contributors* horizontal_contributors = stbir_info->horizontal_contributors;
+ float* horizontal_coefficients = stbir_info->horizontal_coefficients;
+ int coefficient_width = stbir_info->horizontal_coefficient_width;
+
+ for (x = 0; x < output_w; x++)
+ {
+ int n0 = horizontal_contributors[x].n0;
+ int n1 = horizontal_contributors[x].n1;
+
+ int out_pixel_index = x * channels;
+ int coefficient_group = coefficient_width * x;
+ int coefficient_counter = 0;
+
+ STBIR_ASSERT(n1 >= n0);
+ STBIR_ASSERT(n0 >= -stbir_info->horizontal_filter_pixel_margin);
+ STBIR_ASSERT(n1 >= -stbir_info->horizontal_filter_pixel_margin);
+ STBIR_ASSERT(n0 < stbir_info->input_w + stbir_info->horizontal_filter_pixel_margin);
+ STBIR_ASSERT(n1 < stbir_info->input_w + stbir_info->horizontal_filter_pixel_margin);
+
+ switch (channels) {
+ case 1:
+ for (k = n0; k <= n1; k++)
+ {
+ int in_pixel_index = k * 1;
+ float coefficient = horizontal_coefficients[coefficient_group + coefficient_counter++];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ }
+ break;
+ case 2:
+ for (k = n0; k <= n1; k++)
+ {
+ int in_pixel_index = k * 2;
+ float coefficient = horizontal_coefficients[coefficient_group + coefficient_counter++];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
+ }
+ break;
+ case 3:
+ for (k = n0; k <= n1; k++)
+ {
+ int in_pixel_index = k * 3;
+ float coefficient = horizontal_coefficients[coefficient_group + coefficient_counter++];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
+ output_buffer[out_pixel_index + 2] += decode_buffer[in_pixel_index + 2] * coefficient;
+ }
+ break;
+ case 4:
+ for (k = n0; k <= n1; k++)
+ {
+ int in_pixel_index = k * 4;
+ float coefficient = horizontal_coefficients[coefficient_group + coefficient_counter++];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
+ output_buffer[out_pixel_index + 2] += decode_buffer[in_pixel_index + 2] * coefficient;
+ output_buffer[out_pixel_index + 3] += decode_buffer[in_pixel_index + 3] * coefficient;
+ }
+ break;
+ default:
+ for (k = n0; k <= n1; k++)
+ {
+ int in_pixel_index = k * channels;
+ float coefficient = horizontal_coefficients[coefficient_group + coefficient_counter++];
+ int c;
+ STBIR_ASSERT(coefficient != 0);
+ for (c = 0; c < channels; c++)
+ output_buffer[out_pixel_index + c] += decode_buffer[in_pixel_index + c] * coefficient;
+ }
+ break;
+ }
+ }
+}
+
+static void stbir__resample_horizontal_downsample(stbir__info* stbir_info, float* output_buffer)
+{
+ int x, k;
+ int input_w = stbir_info->input_w;
+ int channels = stbir_info->channels;
+ float* decode_buffer = stbir__get_decode_buffer(stbir_info);
+ stbir__contributors* horizontal_contributors = stbir_info->horizontal_contributors;
+ float* horizontal_coefficients = stbir_info->horizontal_coefficients;
+ int coefficient_width = stbir_info->horizontal_coefficient_width;
+ int filter_pixel_margin = stbir_info->horizontal_filter_pixel_margin;
+ int max_x = input_w + filter_pixel_margin * 2;
+
+ STBIR_ASSERT(!stbir__use_width_upsampling(stbir_info));
+
+ switch (channels) {
+ case 1:
+ for (x = 0; x < max_x; x++)
+ {
+ int n0 = horizontal_contributors[x].n0;
+ int n1 = horizontal_contributors[x].n1;
+
+ int in_x = x - filter_pixel_margin;
+ int in_pixel_index = in_x * 1;
+ int max_n = n1;
+ int coefficient_group = coefficient_width * x;
+
+ for (k = n0; k <= max_n; k++)
+ {
+ int out_pixel_index = k * 1;
+ float coefficient = horizontal_coefficients[coefficient_group + k - n0];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ }
+ }
+ break;
+
+ case 2:
+ for (x = 0; x < max_x; x++)
+ {
+ int n0 = horizontal_contributors[x].n0;
+ int n1 = horizontal_contributors[x].n1;
+
+ int in_x = x - filter_pixel_margin;
+ int in_pixel_index = in_x * 2;
+ int max_n = n1;
+ int coefficient_group = coefficient_width * x;
+
+ for (k = n0; k <= max_n; k++)
+ {
+ int out_pixel_index = k * 2;
+ float coefficient = horizontal_coefficients[coefficient_group + k - n0];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
+ }
+ }
+ break;
+
+ case 3:
+ for (x = 0; x < max_x; x++)
+ {
+ int n0 = horizontal_contributors[x].n0;
+ int n1 = horizontal_contributors[x].n1;
+
+ int in_x = x - filter_pixel_margin;
+ int in_pixel_index = in_x * 3;
+ int max_n = n1;
+ int coefficient_group = coefficient_width * x;
+
+ for (k = n0; k <= max_n; k++)
+ {
+ int out_pixel_index = k * 3;
+ float coefficient = horizontal_coefficients[coefficient_group + k - n0];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
+ output_buffer[out_pixel_index + 2] += decode_buffer[in_pixel_index + 2] * coefficient;
+ }
+ }
+ break;
+
+ case 4:
+ for (x = 0; x < max_x; x++)
+ {
+ int n0 = horizontal_contributors[x].n0;
+ int n1 = horizontal_contributors[x].n1;
+
+ int in_x = x - filter_pixel_margin;
+ int in_pixel_index = in_x * 4;
+ int max_n = n1;
+ int coefficient_group = coefficient_width * x;
+
+ for (k = n0; k <= max_n; k++)
+ {
+ int out_pixel_index = k * 4;
+ float coefficient = horizontal_coefficients[coefficient_group + k - n0];
+ STBIR_ASSERT(coefficient != 0);
+ output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+ output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
+ output_buffer[out_pixel_index + 2] += decode_buffer[in_pixel_index + 2] * coefficient;
+ output_buffer[out_pixel_index + 3] += decode_buffer[in_pixel_index + 3] * coefficient;
+ }
+ }
+ break;
+
+ default:
+ for (x = 0; x < max_x; x++)
+ {
+ int n0 = horizontal_contributors[x].n0;
+ int n1 = horizontal_contributors[x].n1;
+
+ int in_x = x - filter_pixel_margin;
+ int in_pixel_index = in_x * channels;
+ int max_n = n1;
+ int coefficient_group = coefficient_width * x;
+
+ for (k = n0; k <= max_n; k++)
+ {
+ int c;
+ int out_pixel_index = k * channels;
+ float coefficient = horizontal_coefficients[coefficient_group + k - n0];
+ STBIR_ASSERT(coefficient != 0);
+ for (c = 0; c < channels; c++)
+ output_buffer[out_pixel_index + c] += decode_buffer[in_pixel_index + c] * coefficient;
+ }
+ }
+ break;
+ }
+}
+
+static void stbir__decode_and_resample_upsample(stbir__info* stbir_info, int n)
+{
+ // Decode the nth scanline from the source image into the decode buffer.
+ stbir__decode_scanline(stbir_info, n);
+
+ // Now resample it into the ring buffer.
+ if (stbir__use_width_upsampling(stbir_info))
+ stbir__resample_horizontal_upsample(stbir_info, stbir__add_empty_ring_buffer_entry(stbir_info, n));
+ else
+ stbir__resample_horizontal_downsample(stbir_info, stbir__add_empty_ring_buffer_entry(stbir_info, n));
+
+ // Now it's sitting in the ring buffer ready to be used as source for the vertical sampling.
+}
+
+static void stbir__decode_and_resample_downsample(stbir__info* stbir_info, int n)
+{
+ // Decode the nth scanline from the source image into the decode buffer.
+ stbir__decode_scanline(stbir_info, n);
+
+ memset(stbir_info->horizontal_buffer, 0, stbir_info->output_w * stbir_info->channels * sizeof(float));
+
+ // Now resample it into the horizontal buffer.
+ if (stbir__use_width_upsampling(stbir_info))
+ stbir__resample_horizontal_upsample(stbir_info, stbir_info->horizontal_buffer);
+ else
+ stbir__resample_horizontal_downsample(stbir_info, stbir_info->horizontal_buffer);
+
+ // Now it's sitting in the horizontal buffer ready to be distributed into the ring buffers.
+}
+
+// Get the specified scan line from the ring buffer.
+static float* stbir__get_ring_buffer_scanline(int get_scanline, float* ring_buffer, int begin_index, int first_scanline, int ring_buffer_num_entries, int ring_buffer_length)
+{
+ int ring_buffer_index = (begin_index + (get_scanline - first_scanline)) % ring_buffer_num_entries;
+ return stbir__get_ring_buffer_entry(ring_buffer, ring_buffer_index, ring_buffer_length);
+}
+
+
+static void stbir__encode_scanline(stbir__info* stbir_info, int num_pixels, void *output_buffer, float *encode_buffer, int channels, int alpha_channel, int decode)
+{
+ int x;
+ int n;
+ int num_nonalpha;
+ stbir_uint16 nonalpha[STBIR_MAX_CHANNELS];
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_PREMULTIPLIED))
+ {
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ float alpha = encode_buffer[pixel_index + alpha_channel];
+ float reciprocal_alpha = alpha ? 1.0f / alpha : 0;
+
+ // unrolling this produced a 1% slowdown upscaling a large RGBA linear-space image on my machine - stb
+ for (n = 0; n < channels; n++)
+ if (n != alpha_channel)
+ encode_buffer[pixel_index + n] *= reciprocal_alpha;
+
+ // We added in a small epsilon to prevent the color channel from being deleted with zero alpha.
+ // Because we only add it for integer types, it will automatically be discarded on integer
+ // conversion, so we don't need to subtract it back out (which would be problematic for
+ // numeric precision reasons).
+ }
+ }
+
+ // build a table of all channels that need colorspace correction, so
+ // we don't perform colorspace correction on channels that don't need it.
+ for (x = 0, num_nonalpha = 0; x < channels; ++x)
+ {
+ if (x != alpha_channel || (stbir_info->flags & STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ {
+ nonalpha[num_nonalpha++] = (stbir_uint16)x;
+ }
+ }
+
+ #define STBIR__ROUND_INT(f) ((int) ((f)+0.5))
+ #define STBIR__ROUND_UINT(f) ((stbir_uint32) ((f)+0.5))
+
+ #ifdef STBIR__SATURATE_INT
+ #define STBIR__ENCODE_LINEAR8(f) stbir__saturate8 (STBIR__ROUND_INT((f) * stbir__max_uint8_as_float ))
+ #define STBIR__ENCODE_LINEAR16(f) stbir__saturate16(STBIR__ROUND_INT((f) * stbir__max_uint16_as_float))
+ #else
+ #define STBIR__ENCODE_LINEAR8(f) (unsigned char ) STBIR__ROUND_INT(stbir__saturate(f) * stbir__max_uint8_as_float )
+ #define STBIR__ENCODE_LINEAR16(f) (unsigned short) STBIR__ROUND_INT(stbir__saturate(f) * stbir__max_uint16_as_float)
+ #endif
+
+ switch (decode)
+ {
+ case STBIR__DECODE(STBIR_TYPE_UINT8, STBIR_COLORSPACE_LINEAR):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < channels; n++)
+ {
+ int index = pixel_index + n;
+ ((unsigned char*)output_buffer)[index] = STBIR__ENCODE_LINEAR8(encode_buffer[index]);
+ }
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT8, STBIR_COLORSPACE_SRGB):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < num_nonalpha; n++)
+ {
+ int index = pixel_index + nonalpha[n];
+ ((unsigned char*)output_buffer)[index] = stbir__linear_to_srgb_uchar(encode_buffer[index]);
+ }
+
+ if (!(stbir_info->flags & STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ ((unsigned char *)output_buffer)[pixel_index + alpha_channel] = STBIR__ENCODE_LINEAR8(encode_buffer[pixel_index+alpha_channel]);
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT16, STBIR_COLORSPACE_LINEAR):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < channels; n++)
+ {
+ int index = pixel_index + n;
+ ((unsigned short*)output_buffer)[index] = STBIR__ENCODE_LINEAR16(encode_buffer[index]);
+ }
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT16, STBIR_COLORSPACE_SRGB):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < num_nonalpha; n++)
+ {
+ int index = pixel_index + nonalpha[n];
+ ((unsigned short*)output_buffer)[index] = (unsigned short)STBIR__ROUND_INT(stbir__linear_to_srgb(stbir__saturate(encode_buffer[index])) * stbir__max_uint16_as_float);
+ }
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ ((unsigned short*)output_buffer)[pixel_index + alpha_channel] = STBIR__ENCODE_LINEAR16(encode_buffer[pixel_index + alpha_channel]);
+ }
+
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT32, STBIR_COLORSPACE_LINEAR):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < channels; n++)
+ {
+ int index = pixel_index + n;
+ ((unsigned int*)output_buffer)[index] = (unsigned int)STBIR__ROUND_UINT(((double)stbir__saturate(encode_buffer[index])) * stbir__max_uint32_as_float);
+ }
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_UINT32, STBIR_COLORSPACE_SRGB):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < num_nonalpha; n++)
+ {
+ int index = pixel_index + nonalpha[n];
+ ((unsigned int*)output_buffer)[index] = (unsigned int)STBIR__ROUND_UINT(((double)stbir__linear_to_srgb(stbir__saturate(encode_buffer[index]))) * stbir__max_uint32_as_float);
+ }
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ ((unsigned int*)output_buffer)[pixel_index + alpha_channel] = (unsigned int)STBIR__ROUND_INT(((double)stbir__saturate(encode_buffer[pixel_index + alpha_channel])) * stbir__max_uint32_as_float);
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_FLOAT, STBIR_COLORSPACE_LINEAR):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < channels; n++)
+ {
+ int index = pixel_index + n;
+ ((float*)output_buffer)[index] = encode_buffer[index];
+ }
+ }
+ break;
+
+ case STBIR__DECODE(STBIR_TYPE_FLOAT, STBIR_COLORSPACE_SRGB):
+ for (x=0; x < num_pixels; ++x)
+ {
+ int pixel_index = x*channels;
+
+ for (n = 0; n < num_nonalpha; n++)
+ {
+ int index = pixel_index + nonalpha[n];
+ ((float*)output_buffer)[index] = stbir__linear_to_srgb(encode_buffer[index]);
+ }
+
+ if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+ ((float*)output_buffer)[pixel_index + alpha_channel] = encode_buffer[pixel_index + alpha_channel];
+ }
+ break;
+
+ default:
+ STBIR_ASSERT(!"Unknown type/colorspace/channels combination.");
+ break;
+ }
+}
+
+static void stbir__resample_vertical_upsample(stbir__info* stbir_info, int n)
+{
+ int x, k;
+ int output_w = stbir_info->output_w;
+ stbir__contributors* vertical_contributors = stbir_info->vertical_contributors;
+ float* vertical_coefficients = stbir_info->vertical_coefficients;
+ int channels = stbir_info->channels;
+ int alpha_channel = stbir_info->alpha_channel;
+ int type = stbir_info->type;
+ int colorspace = stbir_info->colorspace;
+ int ring_buffer_entries = stbir_info->ring_buffer_num_entries;
+ void* output_data = stbir_info->output_data;
+ float* encode_buffer = stbir_info->encode_buffer;
+ int decode = STBIR__DECODE(type, colorspace);
+ int coefficient_width = stbir_info->vertical_coefficient_width;
+ int coefficient_counter;
+ int contributor = n;
+
+ float* ring_buffer = stbir_info->ring_buffer;
+ int ring_buffer_begin_index = stbir_info->ring_buffer_begin_index;
+ int ring_buffer_first_scanline = stbir_info->ring_buffer_first_scanline;
+ int ring_buffer_length = stbir_info->ring_buffer_length_bytes/sizeof(float);
+
+ int n0,n1, output_row_start;
+ int coefficient_group = coefficient_width * contributor;
+
+ n0 = vertical_contributors[contributor].n0;
+ n1 = vertical_contributors[contributor].n1;
+
+ output_row_start = n * stbir_info->output_stride_bytes;
+
+ STBIR_ASSERT(stbir__use_height_upsampling(stbir_info));
+
+ memset(encode_buffer, 0, output_w * sizeof(float) * channels);
+
+ // I tried reblocking this for better cache usage of encode_buffer
+ // (using x_outer, k, x_inner), but it lost speed. -- stb
+
+ coefficient_counter = 0;
+ switch (channels) {
+ case 1:
+ for (k = n0; k <= n1; k++)
+ {
+ int coefficient_index = coefficient_counter++;
+ float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length);
+ float coefficient = vertical_coefficients[coefficient_group + coefficient_index];
+ for (x = 0; x < output_w; ++x)
+ {
+ int in_pixel_index = x * 1;
+ encode_buffer[in_pixel_index + 0] += ring_buffer_entry[in_pixel_index + 0] * coefficient;
+ }
+ }
+ break;
+ case 2:
+ for (k = n0; k <= n1; k++)
+ {
+ int coefficient_index = coefficient_counter++;
+ float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length);
+ float coefficient = vertical_coefficients[coefficient_group + coefficient_index];
+ for (x = 0; x < output_w; ++x)
+ {
+ int in_pixel_index = x * 2;
+ encode_buffer[in_pixel_index + 0] += ring_buffer_entry[in_pixel_index + 0] * coefficient;
+ encode_buffer[in_pixel_index + 1] += ring_buffer_entry[in_pixel_index + 1] * coefficient;
+ }
+ }
+ break;
+ case 3:
+ for (k = n0; k <= n1; k++)
+ {
+ int coefficient_index = coefficient_counter++;
+ float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length);
+ float coefficient = vertical_coefficients[coefficient_group + coefficient_index];
+ for (x = 0; x < output_w; ++x)
+ {
+ int in_pixel_index = x * 3;
+ encode_buffer[in_pixel_index + 0] += ring_buffer_entry[in_pixel_index + 0] * coefficient;
+ encode_buffer[in_pixel_index + 1] += ring_buffer_entry[in_pixel_index + 1] * coefficient;
+ encode_buffer[in_pixel_index + 2] += ring_buffer_entry[in_pixel_index + 2] * coefficient;
+ }
+ }
+ break;
+ case 4:
+ for (k = n0; k <= n1; k++)
+ {
+ int coefficient_index = coefficient_counter++;
+ float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length);
+ float coefficient = vertical_coefficients[coefficient_group + coefficient_index];
+ for (x = 0; x < output_w; ++x)
+ {
+ int in_pixel_index = x * 4;
+ encode_buffer[in_pixel_index + 0] += ring_buffer_entry[in_pixel_index + 0] * coefficient;
+ encode_buffer[in_pixel_index + 1] += ring_buffer_entry[in_pixel_index + 1] * coefficient;
+ encode_buffer[in_pixel_index + 2] += ring_buffer_entry[in_pixel_index + 2] * coefficient;
+ encode_buffer[in_pixel_index + 3] += ring_buffer_entry[in_pixel_index + 3] * coefficient;
+ }
+ }
+ break;
+ default:
+ for (k = n0; k <= n1; k++)
+ {
+ int coefficient_index = coefficient_counter++;
+ float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length);
+ float coefficient = vertical_coefficients[coefficient_group + coefficient_index];
+ for (x = 0; x < output_w; ++x)
+ {
+ int in_pixel_index = x * channels;
+ int c;
+ for (c = 0; c < channels; c++)
+ encode_buffer[in_pixel_index + c] += ring_buffer_entry[in_pixel_index + c] * coefficient;
+ }
+ }
+ break;
+ }
+ stbir__encode_scanline(stbir_info, output_w, (char *) output_data + output_row_start, encode_buffer, channels, alpha_channel, decode);
+}
+
+static void stbir__resample_vertical_downsample(stbir__info* stbir_info, int n)
+{
+ int x, k;
+ int output_w = stbir_info->output_w;
+ stbir__contributors* vertical_contributors = stbir_info->vertical_contributors;
+ float* vertical_coefficients = stbir_info->vertical_coefficients;
+ int channels = stbir_info->channels;
+ int ring_buffer_entries = stbir_info->ring_buffer_num_entries;
+ float* horizontal_buffer = stbir_info->horizontal_buffer;
+ int coefficient_width = stbir_info->vertical_coefficient_width;
+ int contributor = n + stbir_info->vertical_filter_pixel_margin;
+
+ float* ring_buffer = stbir_info->ring_buffer;
+ int ring_buffer_begin_index = stbir_info->ring_buffer_begin_index;
+ int ring_buffer_first_scanline = stbir_info->ring_buffer_first_scanline;
+ int ring_buffer_length = stbir_info->ring_buffer_length_bytes/sizeof(float);
+ int n0,n1;
+
+ n0 = vertical_contributors[contributor].n0;
+ n1 = vertical_contributors[contributor].n1;
+
+ STBIR_ASSERT(!stbir__use_height_upsampling(stbir_info));
+
+ for (k = n0; k <= n1; k++)
+ {
+ int coefficient_index = k - n0;
+ int coefficient_group = coefficient_width * contributor;
+ float coefficient = vertical_coefficients[coefficient_group + coefficient_index];
+
+ float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length);
+
+ switch (channels) {
+ case 1:
+ for (x = 0; x < output_w; x++)
+ {
+ int in_pixel_index = x * 1;
+ ring_buffer_entry[in_pixel_index + 0] += horizontal_buffer[in_pixel_index + 0] * coefficient;
+ }
+ break;
+ case 2:
+ for (x = 0; x < output_w; x++)
+ {
+ int in_pixel_index = x * 2;
+ ring_buffer_entry[in_pixel_index + 0] += horizontal_buffer[in_pixel_index + 0] * coefficient;
+ ring_buffer_entry[in_pixel_index + 1] += horizontal_buffer[in_pixel_index + 1] * coefficient;
+ }
+ break;
+ case 3:
+ for (x = 0; x < output_w; x++)
+ {
+ int in_pixel_index = x * 3;
+ ring_buffer_entry[in_pixel_index + 0] += horizontal_buffer[in_pixel_index + 0] * coefficient;
+ ring_buffer_entry[in_pixel_index + 1] += horizontal_buffer[in_pixel_index + 1] * coefficient;
+ ring_buffer_entry[in_pixel_index + 2] += horizontal_buffer[in_pixel_index + 2] * coefficient;
+ }
+ break;
+ case 4:
+ for (x = 0; x < output_w; x++)
+ {
+ int in_pixel_index = x * 4;
+ ring_buffer_entry[in_pixel_index + 0] += horizontal_buffer[in_pixel_index + 0] * coefficient;
+ ring_buffer_entry[in_pixel_index + 1] += horizontal_buffer[in_pixel_index + 1] * coefficient;
+ ring_buffer_entry[in_pixel_index + 2] += horizontal_buffer[in_pixel_index + 2] * coefficient;
+ ring_buffer_entry[in_pixel_index + 3] += horizontal_buffer[in_pixel_index + 3] * coefficient;
+ }
+ break;
+ default:
+ for (x = 0; x < output_w; x++)
+ {
+ int in_pixel_index = x * channels;
+
+ int c;
+ for (c = 0; c < channels; c++)
+ ring_buffer_entry[in_pixel_index + c] += horizontal_buffer[in_pixel_index + c] * coefficient;
+ }
+ break;
+ }
+ }
+}
+
+static void stbir__buffer_loop_upsample(stbir__info* stbir_info)
+{
+ int y;
+ float scale_ratio = stbir_info->vertical_scale;
+ float out_scanlines_radius = stbir__filter_info_table[stbir_info->vertical_filter].support(1/scale_ratio) * scale_ratio;
+
+ STBIR_ASSERT(stbir__use_height_upsampling(stbir_info));
+
+ for (y = 0; y < stbir_info->output_h; y++)
+ {
+ float in_center_of_out = 0; // Center of the current out scanline in the in scanline space
+ int in_first_scanline = 0, in_last_scanline = 0;
+
+ stbir__calculate_sample_range_upsample(y, out_scanlines_radius, scale_ratio, stbir_info->vertical_shift, &in_first_scanline, &in_last_scanline, &in_center_of_out);
+
+ STBIR_ASSERT(in_last_scanline - in_first_scanline + 1 <= stbir_info->ring_buffer_num_entries);
+
+ if (stbir_info->ring_buffer_begin_index >= 0)
+ {
+ // Get rid of whatever we don't need anymore.
+ while (in_first_scanline > stbir_info->ring_buffer_first_scanline)
+ {
+ if (stbir_info->ring_buffer_first_scanline == stbir_info->ring_buffer_last_scanline)
+ {
+ // We just popped the last scanline off the ring buffer.
+ // Reset it to the empty state.
+ stbir_info->ring_buffer_begin_index = -1;
+ stbir_info->ring_buffer_first_scanline = 0;
+ stbir_info->ring_buffer_last_scanline = 0;
+ break;
+ }
+ else
+ {
+ stbir_info->ring_buffer_first_scanline++;
+ stbir_info->ring_buffer_begin_index = (stbir_info->ring_buffer_begin_index + 1) % stbir_info->ring_buffer_num_entries;
+ }
+ }
+ }
+
+ // Load in new ones.
+ if (stbir_info->ring_buffer_begin_index < 0)
+ stbir__decode_and_resample_upsample(stbir_info, in_first_scanline);
+
+ while (in_last_scanline > stbir_info->ring_buffer_last_scanline)
+ stbir__decode_and_resample_upsample(stbir_info, stbir_info->ring_buffer_last_scanline + 1);
+
+ // Now all buffers should be ready to write a row of vertical sampling.
+ stbir__resample_vertical_upsample(stbir_info, y);
+
+ STBIR_PROGRESS_REPORT((float)y / stbir_info->output_h);
+ }
+}
+
+static void stbir__empty_ring_buffer(stbir__info* stbir_info, int first_necessary_scanline)
+{
+ int output_stride_bytes = stbir_info->output_stride_bytes;
+ int channels = stbir_info->channels;
+ int alpha_channel = stbir_info->alpha_channel;
+ int type = stbir_info->type;
+ int colorspace = stbir_info->colorspace;
+ int output_w = stbir_info->output_w;
+ void* output_data = stbir_info->output_data;
+ int decode = STBIR__DECODE(type, colorspace);
+
+ float* ring_buffer = stbir_info->ring_buffer;
+ int ring_buffer_length = stbir_info->ring_buffer_length_bytes/sizeof(float);
+
+ if (stbir_info->ring_buffer_begin_index >= 0)
+ {
+ // Get rid of whatever we don't need anymore.
+ while (first_necessary_scanline > stbir_info->ring_buffer_first_scanline)
+ {
+ if (stbir_info->ring_buffer_first_scanline >= 0 && stbir_info->ring_buffer_first_scanline < stbir_info->output_h)
+ {
+ int output_row_start = stbir_info->ring_buffer_first_scanline * output_stride_bytes;
+ float* ring_buffer_entry = stbir__get_ring_buffer_entry(ring_buffer, stbir_info->ring_buffer_begin_index, ring_buffer_length);
+ stbir__encode_scanline(stbir_info, output_w, (char *) output_data + output_row_start, ring_buffer_entry, channels, alpha_channel, decode);
+ STBIR_PROGRESS_REPORT((float)stbir_info->ring_buffer_first_scanline / stbir_info->output_h);
+ }
+
+ if (stbir_info->ring_buffer_first_scanline == stbir_info->ring_buffer_last_scanline)
+ {
+ // We just popped the last scanline off the ring buffer.
+ // Reset it to the empty state.
+ stbir_info->ring_buffer_begin_index = -1;
+ stbir_info->ring_buffer_first_scanline = 0;
+ stbir_info->ring_buffer_last_scanline = 0;
+ break;
+ }
+ else
+ {
+ stbir_info->ring_buffer_first_scanline++;
+ stbir_info->ring_buffer_begin_index = (stbir_info->ring_buffer_begin_index + 1) % stbir_info->ring_buffer_num_entries;
+ }
+ }
+ }
+}
+
+static void stbir__buffer_loop_downsample(stbir__info* stbir_info)
+{
+ int y;
+ float scale_ratio = stbir_info->vertical_scale;
+ int output_h = stbir_info->output_h;
+ float in_pixels_radius = stbir__filter_info_table[stbir_info->vertical_filter].support(scale_ratio) / scale_ratio;
+ int pixel_margin = stbir_info->vertical_filter_pixel_margin;
+ int max_y = stbir_info->input_h + pixel_margin;
+
+ STBIR_ASSERT(!stbir__use_height_upsampling(stbir_info));
+
+ for (y = -pixel_margin; y < max_y; y++)
+ {
+ float out_center_of_in; // Center of the current out scanline in the in scanline space
+ int out_first_scanline, out_last_scanline;
+
+ stbir__calculate_sample_range_downsample(y, in_pixels_radius, scale_ratio, stbir_info->vertical_shift, &out_first_scanline, &out_last_scanline, &out_center_of_in);
+
+ STBIR_ASSERT(out_last_scanline - out_first_scanline + 1 <= stbir_info->ring_buffer_num_entries);
+
+ if (out_last_scanline < 0 || out_first_scanline >= output_h)
+ continue;
+
+ stbir__empty_ring_buffer(stbir_info, out_first_scanline);
+
+ stbir__decode_and_resample_downsample(stbir_info, y);
+
+ // Load in new ones.
+ if (stbir_info->ring_buffer_begin_index < 0)
+ stbir__add_empty_ring_buffer_entry(stbir_info, out_first_scanline);
+
+ while (out_last_scanline > stbir_info->ring_buffer_last_scanline)
+ stbir__add_empty_ring_buffer_entry(stbir_info, stbir_info->ring_buffer_last_scanline + 1);
+
+ // Now the horizontal buffer is ready to write to all ring buffer rows.
+ stbir__resample_vertical_downsample(stbir_info, y);
+ }
+
+ stbir__empty_ring_buffer(stbir_info, stbir_info->output_h);
+}
+
+static void stbir__setup(stbir__info *info, int input_w, int input_h, int output_w, int output_h, int channels)
+{
+ info->input_w = input_w;
+ info->input_h = input_h;
+ info->output_w = output_w;
+ info->output_h = output_h;
+ info->channels = channels;
+}
+
+static void stbir__calculate_transform(stbir__info *info, float s0, float t0, float s1, float t1, float *transform)
+{
+ info->s0 = s0;
+ info->t0 = t0;
+ info->s1 = s1;
+ info->t1 = t1;
+
+ if (transform)
+ {
+ info->horizontal_scale = transform[0];
+ info->vertical_scale = transform[1];
+ info->horizontal_shift = transform[2];
+ info->vertical_shift = transform[3];
+ }
+ else
+ {
+ info->horizontal_scale = ((float)info->output_w / info->input_w) / (s1 - s0);
+ info->vertical_scale = ((float)info->output_h / info->input_h) / (t1 - t0);
+
+ info->horizontal_shift = s0 * info->output_w / (s1 - s0);
+ info->vertical_shift = t0 * info->output_h / (t1 - t0);
+ }
+}
+
+static void stbir__choose_filter(stbir__info *info, stbir_filter h_filter, stbir_filter v_filter)
+{
+ if (h_filter == 0)
+ h_filter = stbir__use_upsampling(info->horizontal_scale) ? STBIR_DEFAULT_FILTER_UPSAMPLE : STBIR_DEFAULT_FILTER_DOWNSAMPLE;
+ if (v_filter == 0)
+ v_filter = stbir__use_upsampling(info->vertical_scale) ? STBIR_DEFAULT_FILTER_UPSAMPLE : STBIR_DEFAULT_FILTER_DOWNSAMPLE;
+ info->horizontal_filter = h_filter;
+ info->vertical_filter = v_filter;
+}
+
+static stbir_uint32 stbir__calculate_memory(stbir__info *info)
+{
+ int pixel_margin = stbir__get_filter_pixel_margin(info->horizontal_filter, info->horizontal_scale);
+ int filter_height = stbir__get_filter_pixel_width(info->vertical_filter, info->vertical_scale);
+
+ info->horizontal_num_contributors = stbir__get_contributors(info->horizontal_scale, info->horizontal_filter, info->input_w, info->output_w);
+ info->vertical_num_contributors = stbir__get_contributors(info->vertical_scale , info->vertical_filter , info->input_h, info->output_h);
+
+ // One extra entry because floating point precision problems sometimes cause an extra to be necessary.
+ info->ring_buffer_num_entries = filter_height + 1;
+
+ info->horizontal_contributors_size = info->horizontal_num_contributors * sizeof(stbir__contributors);
+ info->horizontal_coefficients_size = stbir__get_total_horizontal_coefficients(info) * sizeof(float);
+ info->vertical_contributors_size = info->vertical_num_contributors * sizeof(stbir__contributors);
+ info->vertical_coefficients_size = stbir__get_total_vertical_coefficients(info) * sizeof(float);
+ info->decode_buffer_size = (info->input_w + pixel_margin * 2) * info->channels * sizeof(float);
+ info->horizontal_buffer_size = info->output_w * info->channels * sizeof(float);
+ info->ring_buffer_size = info->output_w * info->channels * info->ring_buffer_num_entries * sizeof(float);
+ info->encode_buffer_size = info->output_w * info->channels * sizeof(float);
+
+ STBIR_ASSERT(info->horizontal_filter != 0);
+ STBIR_ASSERT(info->horizontal_filter < STBIR__ARRAY_SIZE(stbir__filter_info_table)); // this now happens too late
+ STBIR_ASSERT(info->vertical_filter != 0);
+ STBIR_ASSERT(info->vertical_filter < STBIR__ARRAY_SIZE(stbir__filter_info_table)); // this now happens too late
+
+ if (stbir__use_height_upsampling(info))
+ // The horizontal buffer is for when we're downsampling the height and we
+ // can't output the result of sampling the decode buffer directly into the
+ // ring buffers.
+ info->horizontal_buffer_size = 0;
+ else
+ // The encode buffer is to retain precision in the height upsampling method
+ // and isn't used when height downsampling.
+ info->encode_buffer_size = 0;
+
+ return info->horizontal_contributors_size + info->horizontal_coefficients_size
+ + info->vertical_contributors_size + info->vertical_coefficients_size
+ + info->decode_buffer_size + info->horizontal_buffer_size
+ + info->ring_buffer_size + info->encode_buffer_size;
+}
+
+static int stbir__resize_allocated(stbir__info *info,
+ const void* input_data, int input_stride_in_bytes,
+ void* output_data, int output_stride_in_bytes,
+ int alpha_channel, stbir_uint32 flags, stbir_datatype type,
+ stbir_edge edge_horizontal, stbir_edge edge_vertical, stbir_colorspace colorspace,
+ void* tempmem, size_t tempmem_size_in_bytes)
+{
+ size_t memory_required = stbir__calculate_memory(info);
+
+ int width_stride_input = input_stride_in_bytes ? input_stride_in_bytes : info->channels * info->input_w * stbir__type_size[type];
+ int width_stride_output = output_stride_in_bytes ? output_stride_in_bytes : info->channels * info->output_w * stbir__type_size[type];
+
+#ifdef STBIR_DEBUG_OVERWRITE_TEST
+#define OVERWRITE_ARRAY_SIZE 8
+ unsigned char overwrite_output_before_pre[OVERWRITE_ARRAY_SIZE];
+ unsigned char overwrite_tempmem_before_pre[OVERWRITE_ARRAY_SIZE];
+ unsigned char overwrite_output_after_pre[OVERWRITE_ARRAY_SIZE];
+ unsigned char overwrite_tempmem_after_pre[OVERWRITE_ARRAY_SIZE];
+
+ size_t begin_forbidden = width_stride_output * (info->output_h - 1) + info->output_w * info->channels * stbir__type_size[type];
+ memcpy(overwrite_output_before_pre, &((unsigned char*)output_data)[-OVERWRITE_ARRAY_SIZE], OVERWRITE_ARRAY_SIZE);
+ memcpy(overwrite_output_after_pre, &((unsigned char*)output_data)[begin_forbidden], OVERWRITE_ARRAY_SIZE);
+ memcpy(overwrite_tempmem_before_pre, &((unsigned char*)tempmem)[-OVERWRITE_ARRAY_SIZE], OVERWRITE_ARRAY_SIZE);
+ memcpy(overwrite_tempmem_after_pre, &((unsigned char*)tempmem)[tempmem_size_in_bytes], OVERWRITE_ARRAY_SIZE);
+#endif
+
+ STBIR_ASSERT(info->channels >= 0);
+ STBIR_ASSERT(info->channels <= STBIR_MAX_CHANNELS);
+
+ if (info->channels < 0 || info->channels > STBIR_MAX_CHANNELS)
+ return 0;
+
+ STBIR_ASSERT(info->horizontal_filter < STBIR__ARRAY_SIZE(stbir__filter_info_table));
+ STBIR_ASSERT(info->vertical_filter < STBIR__ARRAY_SIZE(stbir__filter_info_table));
+
+ if (info->horizontal_filter >= STBIR__ARRAY_SIZE(stbir__filter_info_table))
+ return 0;
+ if (info->vertical_filter >= STBIR__ARRAY_SIZE(stbir__filter_info_table))
+ return 0;
+
+ if (alpha_channel < 0)
+ flags |= STBIR_FLAG_ALPHA_USES_COLORSPACE | STBIR_FLAG_ALPHA_PREMULTIPLIED;
+
+ if (!(flags&STBIR_FLAG_ALPHA_USES_COLORSPACE) || !(flags&STBIR_FLAG_ALPHA_PREMULTIPLIED))
+ STBIR_ASSERT(alpha_channel >= 0 && alpha_channel < info->channels);
+
+ if (alpha_channel >= info->channels)
+ return 0;
+
+ STBIR_ASSERT(tempmem);
+
+ if (!tempmem)
+ return 0;
+
+ STBIR_ASSERT(tempmem_size_in_bytes >= memory_required);
+
+ if (tempmem_size_in_bytes < memory_required)
+ return 0;
+
+ memset(tempmem, 0, tempmem_size_in_bytes);
+
+ info->input_data = input_data;
+ info->input_stride_bytes = width_stride_input;
+
+ info->output_data = output_data;
+ info->output_stride_bytes = width_stride_output;
+
+ info->alpha_channel = alpha_channel;
+ info->flags = flags;
+ info->type = type;
+ info->edge_horizontal = edge_horizontal;
+ info->edge_vertical = edge_vertical;
+ info->colorspace = colorspace;
+
+ info->horizontal_coefficient_width = stbir__get_coefficient_width (info->horizontal_filter, info->horizontal_scale);
+ info->vertical_coefficient_width = stbir__get_coefficient_width (info->vertical_filter , info->vertical_scale );
+ info->horizontal_filter_pixel_width = stbir__get_filter_pixel_width (info->horizontal_filter, info->horizontal_scale);
+ info->vertical_filter_pixel_width = stbir__get_filter_pixel_width (info->vertical_filter , info->vertical_scale );
+ info->horizontal_filter_pixel_margin = stbir__get_filter_pixel_margin(info->horizontal_filter, info->horizontal_scale);
+ info->vertical_filter_pixel_margin = stbir__get_filter_pixel_margin(info->vertical_filter , info->vertical_scale );
+
+ info->ring_buffer_length_bytes = info->output_w * info->channels * sizeof(float);
+ info->decode_buffer_pixels = info->input_w + info->horizontal_filter_pixel_margin * 2;
+
+#define STBIR__NEXT_MEMPTR(current, newtype) (newtype*)(((unsigned char*)current) + current##_size)
+
+ info->horizontal_contributors = (stbir__contributors *) tempmem;
+ info->horizontal_coefficients = STBIR__NEXT_MEMPTR(info->horizontal_contributors, float);
+ info->vertical_contributors = STBIR__NEXT_MEMPTR(info->horizontal_coefficients, stbir__contributors);
+ info->vertical_coefficients = STBIR__NEXT_MEMPTR(info->vertical_contributors, float);
+ info->decode_buffer = STBIR__NEXT_MEMPTR(info->vertical_coefficients, float);
+
+ if (stbir__use_height_upsampling(info))
+ {
+ info->horizontal_buffer = NULL;
+ info->ring_buffer = STBIR__NEXT_MEMPTR(info->decode_buffer, float);
+ info->encode_buffer = STBIR__NEXT_MEMPTR(info->ring_buffer, float);
+
+ STBIR_ASSERT((size_t)STBIR__NEXT_MEMPTR(info->encode_buffer, unsigned char) == (size_t)tempmem + tempmem_size_in_bytes);
+ }
+ else
+ {
+ info->horizontal_buffer = STBIR__NEXT_MEMPTR(info->decode_buffer, float);
+ info->ring_buffer = STBIR__NEXT_MEMPTR(info->horizontal_buffer, float);
+ info->encode_buffer = NULL;
+
+ STBIR_ASSERT((size_t)STBIR__NEXT_MEMPTR(info->ring_buffer, unsigned char) == (size_t)tempmem + tempmem_size_in_bytes);
+ }
+
+#undef STBIR__NEXT_MEMPTR
+
+ // This signals that the ring buffer is empty
+ info->ring_buffer_begin_index = -1;
+
+ stbir__calculate_filters(info->horizontal_contributors, info->horizontal_coefficients, info->horizontal_filter, info->horizontal_scale, info->horizontal_shift, info->input_w, info->output_w);
+ stbir__calculate_filters(info->vertical_contributors, info->vertical_coefficients, info->vertical_filter, info->vertical_scale, info->vertical_shift, info->input_h, info->output_h);
+
+ STBIR_PROGRESS_REPORT(0);
+
+ if (stbir__use_height_upsampling(info))
+ stbir__buffer_loop_upsample(info);
+ else
+ stbir__buffer_loop_downsample(info);
+
+ STBIR_PROGRESS_REPORT(1);
+
+#ifdef STBIR_DEBUG_OVERWRITE_TEST
+ STBIR_ASSERT(memcmp(overwrite_output_before_pre, &((unsigned char*)output_data)[-OVERWRITE_ARRAY_SIZE], OVERWRITE_ARRAY_SIZE) == 0);
+ STBIR_ASSERT(memcmp(overwrite_output_after_pre, &((unsigned char*)output_data)[begin_forbidden], OVERWRITE_ARRAY_SIZE) == 0);
+ STBIR_ASSERT(memcmp(overwrite_tempmem_before_pre, &((unsigned char*)tempmem)[-OVERWRITE_ARRAY_SIZE], OVERWRITE_ARRAY_SIZE) == 0);
+ STBIR_ASSERT(memcmp(overwrite_tempmem_after_pre, &((unsigned char*)tempmem)[tempmem_size_in_bytes], OVERWRITE_ARRAY_SIZE) == 0);
+#endif
+
+ return 1;
+}
+
+
+static int stbir__resize_arbitrary(
+ void *alloc_context,
+ const void* input_data, int input_w, int input_h, int input_stride_in_bytes,
+ void* output_data, int output_w, int output_h, int output_stride_in_bytes,
+ float s0, float t0, float s1, float t1, float *transform,
+ int channels, int alpha_channel, stbir_uint32 flags, stbir_datatype type,
+ stbir_filter h_filter, stbir_filter v_filter,
+ stbir_edge edge_horizontal, stbir_edge edge_vertical, stbir_colorspace colorspace)
+{
+ stbir__info info;
+ int result;
+ size_t memory_required;
+ void* extra_memory;
+
+ stbir__setup(&info, input_w, input_h, output_w, output_h, channels);
+ stbir__calculate_transform(&info, s0,t0,s1,t1,transform);
+ stbir__choose_filter(&info, h_filter, v_filter);
+ memory_required = stbir__calculate_memory(&info);
+ extra_memory = STBIR_MALLOC(memory_required, alloc_context);
+
+ if (!extra_memory)
+ return 0;
+
+ result = stbir__resize_allocated(&info, input_data, input_stride_in_bytes,
+ output_data, output_stride_in_bytes,
+ alpha_channel, flags, type,
+ edge_horizontal, edge_vertical,
+ colorspace, extra_memory, memory_required);
+
+ STBIR_FREE(extra_memory, alloc_context);
+
+ return result;
+}
+
+STBIRDEF int stbir_resize_uint8( const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels)
+{
+ return stbir__resize_arbitrary(NULL, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,-1,0, STBIR_TYPE_UINT8, STBIR_FILTER_DEFAULT, STBIR_FILTER_DEFAULT,
+ STBIR_EDGE_CLAMP, STBIR_EDGE_CLAMP, STBIR_COLORSPACE_LINEAR);
+}
+
+STBIRDEF int stbir_resize_float( const float *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ float *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels)
+{
+ return stbir__resize_arbitrary(NULL, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,-1,0, STBIR_TYPE_FLOAT, STBIR_FILTER_DEFAULT, STBIR_FILTER_DEFAULT,
+ STBIR_EDGE_CLAMP, STBIR_EDGE_CLAMP, STBIR_COLORSPACE_LINEAR);
+}
+
+STBIRDEF int stbir_resize_uint8_srgb(const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags)
+{
+ return stbir__resize_arbitrary(NULL, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,alpha_channel,flags, STBIR_TYPE_UINT8, STBIR_FILTER_DEFAULT, STBIR_FILTER_DEFAULT,
+ STBIR_EDGE_CLAMP, STBIR_EDGE_CLAMP, STBIR_COLORSPACE_SRGB);
+}
+
+STBIRDEF int stbir_resize_uint8_srgb_edgemode(const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode)
+{
+ return stbir__resize_arbitrary(NULL, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,alpha_channel,flags, STBIR_TYPE_UINT8, STBIR_FILTER_DEFAULT, STBIR_FILTER_DEFAULT,
+ edge_wrap_mode, edge_wrap_mode, STBIR_COLORSPACE_SRGB);
+}
+
+STBIRDEF int stbir_resize_uint8_generic( const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space,
+ void *alloc_context)
+{
+ return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,alpha_channel,flags, STBIR_TYPE_UINT8, filter, filter,
+ edge_wrap_mode, edge_wrap_mode, space);
+}
+
+STBIRDEF int stbir_resize_uint16_generic(const stbir_uint16 *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ stbir_uint16 *output_pixels , int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space,
+ void *alloc_context)
+{
+ return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,alpha_channel,flags, STBIR_TYPE_UINT16, filter, filter,
+ edge_wrap_mode, edge_wrap_mode, space);
+}
+
+
+STBIRDEF int stbir_resize_float_generic( const float *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ float *output_pixels , int output_w, int output_h, int output_stride_in_bytes,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space,
+ void *alloc_context)
+{
+ return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,alpha_channel,flags, STBIR_TYPE_FLOAT, filter, filter,
+ edge_wrap_mode, edge_wrap_mode, space);
+}
+
+
+STBIRDEF int stbir_resize( const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ stbir_datatype datatype,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical,
+ stbir_filter filter_horizontal, stbir_filter filter_vertical,
+ stbir_colorspace space, void *alloc_context)
+{
+ return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,NULL,num_channels,alpha_channel,flags, datatype, filter_horizontal, filter_vertical,
+ edge_mode_horizontal, edge_mode_vertical, space);
+}
+
+
+STBIRDEF int stbir_resize_subpixel(const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ stbir_datatype datatype,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical,
+ stbir_filter filter_horizontal, stbir_filter filter_vertical,
+ stbir_colorspace space, void *alloc_context,
+ float x_scale, float y_scale,
+ float x_offset, float y_offset)
+{
+ float transform[4];
+ transform[0] = x_scale;
+ transform[1] = y_scale;
+ transform[2] = x_offset;
+ transform[3] = y_offset;
+ return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ 0,0,1,1,transform,num_channels,alpha_channel,flags, datatype, filter_horizontal, filter_vertical,
+ edge_mode_horizontal, edge_mode_vertical, space);
+}
+
+STBIRDEF int stbir_resize_region( const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+ void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+ stbir_datatype datatype,
+ int num_channels, int alpha_channel, int flags,
+ stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical,
+ stbir_filter filter_horizontal, stbir_filter filter_vertical,
+ stbir_colorspace space, void *alloc_context,
+ float s0, float t0, float s1, float t1)
+{
+ return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes,
+ output_pixels, output_w, output_h, output_stride_in_bytes,
+ s0,t0,s1,t1,NULL,num_channels,alpha_channel,flags, datatype, filter_horizontal, filter_vertical,
+ edge_mode_horizontal, edge_mode_vertical, space);
+}
+
+#endif // STB_IMAGE_RESIZE_IMPLEMENTATION
+
+/*
+------------------------------------------------------------------------------
+This software is available under 2 licenses -- choose whichever you prefer.
+------------------------------------------------------------------------------
+ALTERNATIVE A - MIT License
+Copyright (c) 2017 Sean Barrett
+Permission is hereby granted, free of charge, to any person obtaining a copy of
+this software and associated documentation files (the "Software"), to deal in
+the Software without restriction, including without limitation the rights to
+use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
+of the Software, and to permit persons to whom the Software is furnished to do
+so, subject to the following conditions:
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+------------------------------------------------------------------------------
+ALTERNATIVE B - Public Domain (www.unlicense.org)
+This is free and unencumbered software released into the public domain.
+Anyone is free to copy, modify, publish, use, compile, sell, or distribute this
+software, either in source code form or as a compiled binary, for any purpose,
+commercial or non-commercial, and by any means.
+In jurisdictions that recognize copyright laws, the author or authors of this
+software dedicate any and all copyright interest in the software to the public
+domain. We make this dedication for the benefit of the public at large and to
+the detriment of our heirs and successors. We intend this dedication to be an
+overt act of relinquishment in perpetuity of all present and future rights to
+this software under copyright law.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+------------------------------------------------------------------------------
+*/
--- /dev/null
+#! /usr/bin/env python3
+
+# Copyright 2017 Intel Corporation.
+# The source code, information and material ("Material") contained herein is
+# owned by Intel Corporation or its suppliers or licensors, and title to such
+# Material remains with Intel Corporation or its suppliers or licensors.
+# The Material contains proprietary information of Intel or its suppliers and
+# licensors. The Material is protected by worldwide copyright laws and treaty
+# provisions.
+# No part of the Material may be used, copied, reproduced, modified, published,
+# uploaded, posted, transmitted, distributed or disclosed in any way without
+# Intel's prior express written permission. No license under any patent,
+# copyright or other intellectual property rights in the Material is granted to
+# or conferred upon you, either expressly, by implication, inducement, estoppel
+# or otherwise.
+# Any license under such intellectual property rights must be express and
+# approved by Intel in writing.
+
+from mvnc import mvncapi as mvnc
+import sys
+import numpy
+import cv2
+import time
+import csv
+import os
+import sys
+
+dim=(227,227)
+EXAMPLES_BASE_DIR='../../'
+
+# ***************************************************************
+# get labels
+# ***************************************************************
+labels_file=EXAMPLES_BASE_DIR+'data/ilsvrc12/synset_words.txt'
+labels=numpy.loadtxt(labels_file,str,delimiter='\t')
+
+# ***************************************************************
+# configure the NCS
+# ***************************************************************
+mvnc.SetGlobalOption(mvnc.GlobalOption.LOG_LEVEL, 2)
+
+# ***************************************************************
+# Get a list of ALL the sticks that are plugged in
+# ***************************************************************
+devices = mvnc.EnumerateDevices()
+if len(devices) == 0:
+ print('No devices found')
+ quit()
+
+# ***************************************************************
+# Pick the first stick to run the network
+# ***************************************************************
+device = mvnc.Device(devices[0])
+
+# ***************************************************************
+# Open the NCS
+# ***************************************************************
+device.OpenDevice()
+
+network_blob='graph'
+
+#Load blob
+with open(network_blob, mode='rb') as f:
+ blob = f.read()
+
+graph = device.AllocateGraph(blob)
+
+# ***************************************************************
+# Load the image
+# ***************************************************************
+ilsvrc_mean = numpy.load(EXAMPLES_BASE_DIR+'data/ilsvrc12/ilsvrc_2012_mean.npy').mean(1).mean(1) #loading the mean file
+img = cv2.imread(EXAMPLES_BASE_DIR+'data/images/nps_electric_guitar.png')
+img=cv2.resize(img,dim)
+img = img.astype(numpy.float32)
+img[:,:,0] = (img[:,:,0] - ilsvrc_mean[0])
+img[:,:,1] = (img[:,:,1] - ilsvrc_mean[1])
+img[:,:,2] = (img[:,:,2] - ilsvrc_mean[2])
+
+# ***************************************************************
+# Send the image to the NCS
+# ***************************************************************
+graph.LoadTensor(img.astype(numpy.float16), 'user object')
+
+# ***************************************************************
+# Get the result from the NCS
+# ***************************************************************
+output, userobj = graph.GetResult()
+
+# ***************************************************************
+# Print the results of the inference form the NCS
+# ***************************************************************
+order = output.argsort()[::-1][:6]
+print('\n------- predictions --------')
+for i in range(0,5):
+ print ('prediction ' + str(i) + ' (probability ' + str(output[order[i]]*100) + '%) is ' + labels[order[i]] + ' label index is: ' + str(order[i]) )
+
+
+# ***************************************************************
+# Clean up the graph and the device
+# ***************************************************************
+graph.DeallocateGraph()
+device.CloseDevice()
+
+
+
+
--- /dev/null
+../../data/ilsvrc12/synset_words.txt
\ No newline at end of file
--- /dev/null
+
+ifneq ($(findstring movidius, $(PYTHONPATH)), movidius)
+ export PYTHONPATH:=/opt/movidius/caffe/python:/opt/movidius/mvnc/python:$(PYTHONPATH)
+endif
+
+.PHONY: help
+help:
+ @echo "Possible Make targets"
+ @echo " make help - shows this message"
+ @echo " make clean - Removes all temp files from all directories"
+
+TOPTARGETS := clean
+
+SUBDIRS := $(wildcard il*/.)
+
+$(TOPTARGETS): $(SUBDIRS)
+$(SUBDIRS):
+ $(MAKE) -C $@ $(MAKECMDGOALS)
+
+.PHONY: $(TOPTARGETS) $(SUBDIRS)
+
+all compile profile check run:
+
--- /dev/null
+python_version_full := $(wordlist 2,4,$(subst ., ,$(shell python3 --version 2>&1)))
+python_version_minor := $(word 2,${python_version_full})
+
+ILSVRC_SH = get_ilsvrc_aux.sh
+GET_ILSVRC_AUX = wget https://raw.githubusercontent.com/BVLC/caffe/master/data/ilsvrc12/get_ilsvrc_aux.sh
+.PHONY: all
+
+all:
+ @test -f ${ILSVRC_SH} || (${GET_ILSVRC_AUX};chmod +x ${ILSVRC_SH};./${ILSVRC_SH});
+ @test -f ilsvrc_2012_mean.npy || wget https://raw.githubusercontent.com/BVLC/caffe/master/python/caffe/imagenet/ilsvrc_2012_mean.npy
+
+.PHONY: clean
+clean:
+ @rm -f ${ILSVRC_SH}
+ @rm -f test.txt
+ @rm -f synsets.txt
+ @rm -f val.txt
+ @rm -f train.txt
+ @rm -f synset_words.txt
+ @rm -f det_synset_words.txt
+ @rm -f imagenet.bet.pickle
+ @rm -f imagenet_mean.binaryproto
+ @rm -f ._*
+ @rm -f caffe_ilsvrc12.tar.gz
+ @rm -f ilsvrc_2012_mean.npy
--- /dev/null
+
+ifneq ($(findstring movidius, $(PYTHONPATH)), movidius)
+ export PYTHONPATH:=/opt/movidius/caffe/python:/opt/movidius/mvnc/python:$(PYTHONPATH)
+endif
+
+
+SUBDIRS := $(wildcard */.)
+TOPTARGETS := all clean check compile profile run
+
+$(TOPTARGETS): $(SUBDIRS)
+$(SUBDIRS):
+ $(MAKE) -C $@ $(MAKECMDGOALS)
+
+.PHONY: $(TOPTARGETS) $(SUBDIRS)
+
+.PHONY: help
+help:
+ @echo "Possible Make targets"
+ @echo " make help - shows this message"
+ @echo " make all - Makes all targets"
+ @echo " make clean - Removes all temp files from all directories"
+ @echo " make check - Runs check on all caffe/tensorflow models"
+ @echo " make profile - Runs profile on all caffe/tensorflow models"
+ @echo " make compile - Runs compile on all caffe/tensorflow models"
+ @echo " make run - Runs all caffe/tensorflow/apps"
--- /dev/null
+
+ifneq ($(findstring movidius, $(PYTHONPATH)), movidius)
+ export PYTHONPATH:=/opt/movidius/caffe/python:/opt/movidius/mvnc/python:$(PYTHONPATH)
+endif
+
+NCCOMPILE = mvNCCompile
+NCPROFILE = mvNCProfile
+NCCHECK = mvNCCheck
+
+.IGNORE: profile check compile
+
+WEIGHTS_FILENAME = inception_v1_2016_08_28.tar.gz
+GET_WEIGHTS = (wget http://download.tensorflow.org/models/${WEIGHTS_FILENAME} && tar zxf ${WEIGHTS_FILENAME} && rm ${WEIGHTS_FILENAME})
+
+MODEL_FILENAME = output/inception-v1.meta
+CONV_SCRIPT = ./inception-v1.py
+
+INPUT_NODE_FLAG = -in=input
+OUTPUT_NODE_FLAG = -on=InceptionV1/Logits/Predictions/Reshape_1
+
+.PHONY: all
+all: profile check compile
+
+.PHONY: prereqs
+prereqs:
+ (cd ../../data/ilsvrc12; make)
+ @sed -i 's/\r//' run.py
+ @chmod +x run.py
+
+.PHONY: profile
+profile: weights
+ ${NCPROFILE} -s 12 ${MODEL_FILENAME} ${INPUT_NODE_FLAG} ${OUTPUT_NODE_FLAG}
+
+.PHONY: browse_profile
+browse_profile: weights profile
+ @if [ -e output_report.html ] ; \
+ then \
+ firefox output_report.html & \
+ else \
+ @echo "***\nError - output_report.html not found" ; \
+ fi ;
+
+.PHONY: weights
+weights:
+ @sed -i 's/\r//' ${CONV_SCRIPT}
+ @chmod +x ${CONV_SCRIPT}
+ test -f ${MODEL_FILENAME} || (${GET_WEIGHTS} && ${CONV_SCRIPT})
+
+.PHONY: compile
+compile: weights
+ test -f graph || ${NCCOMPILE} -s 12 ${MODEL_FILENAME} ${INPUT_NODE_FLAG} ${OUTPUT_NODE_FLAG}
+
+.PHONY: check
+check: weights
+ -${NCCHECK} -s 12 ${MODEL_FILENAME} ${INPUT_NODE_FLAG} ${OUTPUT_NODE_FLAG} -i ../../data/images/cat.jpg -id 829 -S 2 -M 128 -cs 0,1,2
+
+.PHONY: run
+run: compile
+ ./run.py
+
+.PHONY: run_py
+run_py: compile
+ ./run.py
+
+.PHONY: help
+help:
+ @echo "possible make targets: ";
+ @echo " make help - shows this message";
+ @echo " make all - makes the following: prototxt, profile, compile, check, cpp, run_py, run_cpp";
+ @echo " make weights - downloads the trained model";
+ @echo " make compile - runs SDK compiler tool to compile the NCS graph file for the network";
+ @echo " make check - runs SDK checker tool to verify an NCS graph file";
+ @echo " make profile - runs the SDK profiler tool to profile the network creating output_report.html";
+ @echo " make browse_profile - runs the SDK profiler tool and brings up report in browser.";
+ @echo " make run_py - runs the run.py python example program";
+ @echo " make clean - removes all created content"
+
+
+clean:
+ rm -f output.gv
+ rm -f output.gv.svg
+ rm -f output_report.html
+ rm -f output_expected.npy
+ rm -f *.ckpt
+ rm -f output_result.npy
+ rm -f output_val.csv
+ rm -rf output
+
--- /dev/null
+# Introduction
+The [inception v1](https://github.com/tensorflow/models/tree/master/slim/nets) network can be used for image classification. The provided Makefile does the following
+1. Downloads the TensorFlow checkpoint file
+2. Runs the conversion/save python script to generate network.meta file.
+3. Profiles, Compiles and Checks the network using the Neural Compute SDK.
+4. There is a run.py provided that does a single inference on a provided image as an example on how to use the network using the Neural Compute API
+
+# Makefile
+Provided Makefile describes various targets that help with the above mentioned tasks.
+
+## make all
+Runs ncprofile, ncompile and run.
+
+## make profile
+Runs the provided network on the NCS and generates per layer statistics that are helpful for understanding the performance of the network on the Neural Compute Stick.
+
+## make compile
+Uses the network description and the trained weights files to generate a Movidius internal 'graph' format file. This file is later used for loading the network on to the Neural Compute Stick and executing the network.
+
+## make run
+Runs the provided run.py file which sends a single image to the Neural Compute Stick and receives and displays the inference results.
+
+## make check
+Runs the network on Caffe on CPU and runs the network on the Neural Compute Stick. Check then compares the two results to make sure they are consistent with each other.
+
+## make clean
+Removes all the temporary files that are created by the Makefile
+
--- /dev/null
+classes
+background
+tench, Tinca tinca
+goldfish, Carassius auratus
+great white shark, white shark, man-eater, man-eating shark, Carcharodon carcharias
+tiger shark, Galeocerdo cuvieri
+hammerhead, hammerhead shark
+electric ray, crampfish, numbfish, torpedo
+stingray
+cock
+hen
+ostrich, Struthio camelus
+brambling, Fringilla montifringilla
+goldfinch, Carduelis carduelis
+house finch, linnet, Carpodacus mexicanus
+junco, snowbird
+indigo bunting, indigo finch, indigo bird, Passerina cyanea
+robin, American robin, Turdus migratorius
+bulbul
+jay
+magpie
+chickadee
+water ouzel, dipper
+kite
+bald eagle, American eagle, Haliaeetus leucocephalus
+vulture
+great grey owl, great gray owl, Strix nebulosa
+European fire salamander, Salamandra salamandra
+common newt, Triturus vulgaris
+eft
+spotted salamander, Ambystoma maculatum
+axolotl, mud puppy, Ambystoma mexicanum
+bullfrog, Rana catesbeiana
+tree frog, tree-frog
+tailed frog, bell toad, ribbed toad, tailed toad, Ascaphus trui
+loggerhead, loggerhead turtle, Caretta caretta
+leatherback turtle, leatherback, leathery turtle, Dermochelys coriacea
+mud turtle
+terrapin
+box turtle, box tortoise
+banded gecko
+common iguana, iguana, Iguana iguana
+American chameleon, anole, Anolis carolinensis
+whiptail, whiptail lizard
+agama
+frilled lizard, Chlamydosaurus kingi
+alligator lizard
+Gila monster, Heloderma suspectum
+green lizard, Lacerta viridis
+African chameleon, Chamaeleo chamaeleon
+Komodo dragon, Komodo lizard, dragon lizard, giant lizard, Varanus komodoensis
+African crocodile, Nile crocodile, Crocodylus niloticus
+American alligator, Alligator mississipiensis
+triceratops
+thunder snake, worm snake, Carphophis amoenus
+ringneck snake, ring-necked snake, ring snake
+hognose snake, puff adder, sand viper
+green snake, grass snake
+king snake, kingsnake
+garter snake, grass snake
+water snake
+vine snake
+night snake, Hypsiglena torquata
+boa constrictor, Constrictor constrictor
+rock python, rock snake, Python sebae
+Indian cobra, Naja naja
+green mamba
+sea snake
+horned viper, cerastes, sand viper, horned asp, Cerastes cornutus
+diamondback, diamondback rattlesnake, Crotalus adamanteus
+sidewinder, horned rattlesnake, Crotalus cerastes
+trilobite
+harvestman, daddy longlegs, Phalangium opilio
+scorpion
+black and gold garden spider, Argiope aurantia
+barn spider, Araneus cavaticus
+garden spider, Aranea diademata
+black widow, Latrodectus mactans
+tarantula
+wolf spider, hunting spider
+tick
+centipede
+black grouse
+ptarmigan
+ruffed grouse, partridge, Bonasa umbellus
+prairie chicken, prairie grouse, prairie fowl
+peacock
+quail
+partridge
+African grey, African gray, Psittacus erithacus
+macaw
+sulphur-crested cockatoo, Kakatoe galerita, Cacatua galerita
+lorikeet
+coucal
+bee eater
+hornbill
+hummingbird
+jacamar
+toucan
+drake
+red-breasted merganser, Mergus serrator
+goose
+black swan, Cygnus atratus
+tusker
+echidna, spiny anteater, anteater
+platypus, duckbill, duckbilled platypus, duck-billed platypus, Ornithorhynchus anatinus
+wallaby, brush kangaroo
+koala, koala bear, kangaroo bear, native bear, Phascolarctos cinereus
+wombat
+jellyfish
+sea anemone, anemone
+brain coral
+flatworm, platyhelminth
+nematode, nematode worm, roundworm
+conch
+snail
+slug
+sea slug, nudibranch
+chiton, coat-of-mail shell, sea cradle, polyplacophore
+chambered nautilus, pearly nautilus, nautilus
+Dungeness crab, Cancer magister
+rock crab, Cancer irroratus
+fiddler crab
+king crab, Alaska crab, Alaskan king crab, Alaska king crab, Paralithodes camtschatica
+American lobster, Northern lobster, Maine lobster, Homarus americanus
+spiny lobster, langouste, rock lobster, crawfish, crayfish, sea crawfish
+crayfish, crawfish, crawdad, crawdaddy
+hermit crab
+isopod
+white stork, Ciconia ciconia
+black stork, Ciconia nigra
+spoonbill
+flamingo
+little blue heron, Egretta caerulea
+American egret, great white heron, Egretta albus
+bittern
+crane
+limpkin, Aramus pictus
+European gallinule, Porphyrio porphyrio
+American coot, marsh hen, mud hen, water hen, Fulica americana
+bustard
+ruddy turnstone, Arenaria interpres
+red-backed sandpiper, dunlin, Erolia alpina
+redshank, Tringa totanus
+dowitcher
+oystercatcher, oyster catcher
+pelican
+king penguin, Aptenodytes patagonica
+albatross, mollymawk
+grey whale, gray whale, devilfish, Eschrichtius gibbosus, Eschrichtius robustus
+killer whale, killer, orca, grampus, sea wolf, Orcinus orca
+dugong, Dugong dugon
+sea lion
+Chihuahua
+Japanese spaniel
+Maltese dog, Maltese terrier, Maltese
+Pekinese, Pekingese, Peke
+Shih-Tzu
+Blenheim spaniel
+papillon
+toy terrier
+Rhodesian ridgeback
+Afghan hound, Afghan
+basset, basset hound
+beagle
+bloodhound, sleuthhound
+bluetick
+black-and-tan coonhound
+Walker hound, Walker foxhound
+English foxhound
+redbone
+borzoi, Russian wolfhound
+Irish wolfhound
+Italian greyhound
+whippet
+Ibizan hound, Ibizan Podenco
+Norwegian elkhound, elkhound
+otterhound, otter hound
+Saluki, gazelle hound
+Scottish deerhound, deerhound
+Weimaraner
+Staffordshire bullterrier, Staffordshire bull terrier
+American Staffordshire terrier, Staffordshire terrier, American pit bull terrier, pit bull terrier
+Bedlington terrier
+Border terrier
+Kerry blue terrier
+Irish terrier
+Norfolk terrier
+Norwich terrier
+Yorkshire terrier
+wire-haired fox terrier
+Lakeland terrier
+Sealyham terrier, Sealyham
+Airedale, Airedale terrier
+cairn, cairn terrier
+Australian terrier
+Dandie Dinmont, Dandie Dinmont terrier
+Boston bull, Boston terrier
+miniature schnauzer
+giant schnauzer
+standard schnauzer
+Scotch terrier, Scottish terrier, Scottie
+Tibetan terrier, chrysanthemum dog
+silky terrier, Sydney silky
+soft-coated wheaten terrier
+West Highland white terrier
+Lhasa, Lhasa apso
+flat-coated retriever
+curly-coated retriever
+golden retriever
+Labrador retriever
+Chesapeake Bay retriever
+German short-haired pointer
+vizsla, Hungarian pointer
+English setter
+Irish setter, red setter
+Gordon setter
+Brittany spaniel
+clumber, clumber spaniel
+English springer, English springer spaniel
+Welsh springer spaniel
+cocker spaniel, English cocker spaniel, cocker
+Sussex spaniel
+Irish water spaniel
+kuvasz
+schipperke
+groenendael
+malinois
+briard
+kelpie
+komondor
+Old English sheepdog, bobtail
+Shetland sheepdog, Shetland sheep dog, Shetland
+collie
+Border collie
+Bouvier des Flandres, Bouviers des Flandres
+Rottweiler
+German shepherd, German shepherd dog, German police dog, alsatian
+Doberman, Doberman pinscher
+miniature pinscher
+Greater Swiss Mountain dog
+Bernese mountain dog
+Appenzeller
+EntleBucher
+boxer
+bull mastiff
+Tibetan mastiff
+French bulldog
+Great Dane
+Saint Bernard, St Bernard
+Eskimo dog, husky
+malamute, malemute, Alaskan malamute
+Siberian husky
+dalmatian, coach dog, carriage dog
+affenpinscher, monkey pinscher, monkey dog
+basenji
+pug, pug-dog
+Leonberg
+Newfoundland, Newfoundland dog
+Great Pyrenees
+Samoyed, Samoyede
+Pomeranian
+chow, chow chow
+keeshond
+Brabancon griffon
+Pembroke, Pembroke Welsh corgi
+Cardigan, Cardigan Welsh corgi
+toy poodle
+miniature poodle
+standard poodle
+Mexican hairless
+timber wolf, grey wolf, gray wolf, Canis lupus
+white wolf, Arctic wolf, Canis lupus tundrarum
+red wolf, maned wolf, Canis rufus, Canis niger
+coyote, prairie wolf, brush wolf, Canis latrans
+dingo, warrigal, warragal, Canis dingo
+dhole, Cuon alpinus
+African hunting dog, hyena dog, Cape hunting dog, Lycaon pictus
+hyena, hyaena
+red fox, Vulpes vulpes
+kit fox, Vulpes macrotis
+Arctic fox, white fox, Alopex lagopus
+grey fox, gray fox, Urocyon cinereoargenteus
+tabby, tabby cat
+tiger cat
+Persian cat
+Siamese cat, Siamese
+Egyptian cat
+cougar, puma, catamount, mountain lion, painter, panther, Felis concolor
+lynx, catamount
+leopard, Panthera pardus
+snow leopard, ounce, Panthera uncia
+jaguar, panther, Panthera onca, Felis onca
+lion, king of beasts, Panthera leo
+tiger, Panthera tigris
+cheetah, chetah, Acinonyx jubatus
+brown bear, bruin, Ursus arctos
+American black bear, black bear, Ursus americanus, Euarctos americanus
+ice bear, polar bear, Ursus Maritimus, Thalarctos maritimus
+sloth bear, Melursus ursinus, Ursus ursinus
+mongoose
+meerkat, mierkat
+tiger beetle
+ladybug, ladybeetle, lady beetle, ladybird, ladybird beetle
+ground beetle, carabid beetle
+long-horned beetle, longicorn, longicorn beetle
+leaf beetle, chrysomelid
+dung beetle
+rhinoceros beetle
+weevil
+fly
+bee
+ant, emmet, pismire
+grasshopper, hopper
+cricket
+walking stick, walkingstick, stick insect
+cockroach, roach
+mantis, mantid
+cicada, cicala
+leafhopper
+lacewing, lacewing fly
+dragonfly, darning needle, devil's darning needle, sewing needle, snake feeder, snake doctor, mosquito hawk, skeeter hawk
+damselfly
+admiral
+ringlet, ringlet butterfly
+monarch, monarch butterfly, milkweed butterfly, Danaus plexippus
+cabbage butterfly
+sulphur butterfly, sulfur butterfly
+lycaenid, lycaenid butterfly
+starfish, sea star
+sea urchin
+sea cucumber, holothurian
+wood rabbit, cottontail, cottontail rabbit
+hare
+Angora, Angora rabbit
+hamster
+porcupine, hedgehog
+fox squirrel, eastern fox squirrel, Sciurus niger
+marmot
+beaver
+guinea pig, Cavia cobaya
+sorrel
+zebra
+hog, pig, grunter, squealer, Sus scrofa
+wild boar, boar, Sus scrofa
+warthog
+hippopotamus, hippo, river horse, Hippopotamus amphibius
+ox
+water buffalo, water ox, Asiatic buffalo, Bubalus bubalis
+bison
+ram, tup
+bighorn, bighorn sheep, cimarron, Rocky Mountain bighorn, Rocky Mountain sheep, Ovis canadensis
+ibex, Capra ibex
+hartebeest
+impala, Aepyceros melampus
+gazelle
+Arabian camel, dromedary, Camelus dromedarius
+llama
+weasel
+mink
+polecat, fitch, foulmart, foumart, Mustela putorius
+black-footed ferret, ferret, Mustela nigripes
+otter
+skunk, polecat, wood pussy
+badger
+armadillo
+three-toed sloth, ai, Bradypus tridactylus
+orangutan, orang, orangutang, Pongo pygmaeus
+gorilla, Gorilla gorilla
+chimpanzee, chimp, Pan troglodytes
+gibbon, Hylobates lar
+siamang, Hylobates syndactylus, Symphalangus syndactylus
+guenon, guenon monkey
+patas, hussar monkey, Erythrocebus patas
+baboon
+macaque
+langur
+colobus, colobus monkey
+proboscis monkey, Nasalis larvatus
+marmoset
+capuchin, ringtail, Cebus capucinus
+howler monkey, howler
+titi, titi monkey
+spider monkey, Ateles geoffroyi
+squirrel monkey, Saimiri sciureus
+Madagascar cat, ring-tailed lemur, Lemur catta
+indri, indris, Indri indri, Indri brevicaudatus
+Indian elephant, Elephas maximus
+African elephant, Loxodonta africana
+lesser panda, red panda, panda, bear cat, cat bear, Ailurus fulgens
+giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca
+barracouta, snoek
+eel
+coho, cohoe, coho salmon, blue jack, silver salmon, Oncorhynchus kisutch
+rock beauty, Holocanthus tricolor
+anemone fish
+sturgeon
+gar, garfish, garpike, billfish, Lepisosteus osseus
+lionfish
+puffer, pufferfish, blowfish, globefish
+abacus
+abaya
+academic gown, academic robe, judge's robe
+accordion, piano accordion, squeeze box
+acoustic guitar
+aircraft carrier, carrier, flattop, attack aircraft carrier
+airliner
+airship, dirigible
+altar
+ambulance
+amphibian, amphibious vehicle
+analog clock
+apiary, bee house
+apron
+ashcan, trash can, garbage can, wastebin, ash bin, ash-bin, ashbin, dustbin, trash barrel, trash bin
+assault rifle, assault gun
+backpack, back pack, knapsack, packsack, rucksack, haversack
+bakery, bakeshop, bakehouse
+balance beam, beam
+balloon
+ballpoint, ballpoint pen, ballpen, Biro
+Band Aid
+banjo
+bannister, banister, balustrade, balusters, handrail
+barbell
+barber chair
+barbershop
+barn
+barometer
+barrel, cask
+barrow, garden cart, lawn cart, wheelbarrow
+baseball
+basketball
+bassinet
+bassoon
+bathing cap, swimming cap
+bath towel
+bathtub, bathing tub, bath, tub
+beach wagon, station wagon, wagon, estate car, beach waggon, station waggon, waggon
+beacon, lighthouse, beacon light, pharos
+beaker
+bearskin, busby, shako
+beer bottle
+beer glass
+bell cote, bell cot
+bib
+bicycle-built-for-two, tandem bicycle, tandem
+bikini, two-piece
+binder, ring-binder
+binoculars, field glasses, opera glasses
+birdhouse
+boathouse
+bobsled, bobsleigh, bob
+bolo tie, bolo, bola tie, bola
+bonnet, poke bonnet
+bookcase
+bookshop, bookstore, bookstall
+bottlecap
+bow
+bow tie, bow-tie, bowtie
+brass, memorial tablet, plaque
+brassiere, bra, bandeau
+breakwater, groin, groyne, mole, bulwark, seawall, jetty
+breastplate, aegis, egis
+broom
+bucket, pail
+buckle
+bulletproof vest
+bullet train, bullet
+butcher shop, meat market
+cab, hack, taxi, taxicab
+caldron, cauldron
+candle, taper, wax light
+cannon
+canoe
+can opener, tin opener
+cardigan
+car mirror
+carousel, carrousel, merry-go-round, roundabout, whirligig
+carpenter's kit, tool kit
+carton
+car wheel
+cash machine, cash dispenser, automated teller machine, automatic teller machine, automated teller, automatic teller, ATM
+cassette
+cassette player
+castle
+catamaran
+CD player
+cello, violoncello
+cellular telephone, cellular phone, cellphone, cell, mobile phone
+chain
+chainlink fence
+chain mail, ring mail, mail, chain armor, chain armour, ring armor, ring armour
+chain saw, chainsaw
+chest
+chiffonier, commode
+chime, bell, gong
+china cabinet, china closet
+Christmas stocking
+church, church building
+cinema, movie theater, movie theatre, movie house, picture palace
+cleaver, meat cleaver, chopper
+cliff dwelling
+cloak
+clog, geta, patten, sabot
+cocktail shaker
+coffee mug
+coffeepot
+coil, spiral, volute, whorl, helix
+combination lock
+computer keyboard, keypad
+confectionery, confectionary, candy store
+container ship, containership, container vessel
+convertible
+corkscrew, bottle screw
+cornet, horn, trumpet, trump
+cowboy boot
+cowboy hat, ten-gallon hat
+cradle
+crane
+crash helmet
+crate
+crib, cot
+Crock Pot
+croquet ball
+crutch
+cuirass
+dam, dike, dyke
+desk
+desktop computer
+dial telephone, dial phone
+diaper, nappy, napkin
+digital clock
+digital watch
+dining table, board
+dishrag, dishcloth
+dishwasher, dish washer, dishwashing machine
+disk brake, disc brake
+dock, dockage, docking facility
+dogsled, dog sled, dog sleigh
+dome
+doormat, welcome mat
+drilling platform, offshore rig
+drum, membranophone, tympan
+drumstick
+dumbbell
+Dutch oven
+electric fan, blower
+electric guitar
+electric locomotive
+entertainment center
+envelope
+espresso maker
+face powder
+feather boa, boa
+file, file cabinet, filing cabinet
+fireboat
+fire engine, fire truck
+fire screen, fireguard
+flagpole, flagstaff
+flute, transverse flute
+folding chair
+football helmet
+forklift
+fountain
+fountain pen
+four-poster
+freight car
+French horn, horn
+frying pan, frypan, skillet
+fur coat
+garbage truck, dustcart
+gasmask, respirator, gas helmet
+gas pump, gasoline pump, petrol pump, island dispenser
+goblet
+go-kart
+golf ball
+golfcart, golf cart
+gondola
+gong, tam-tam
+gown
+grand piano, grand
+greenhouse, nursery, glasshouse
+grille, radiator grille
+grocery store, grocery, food market, market
+guillotine
+hair slide
+hair spray
+half track
+hammer
+hamper
+hand blower, blow dryer, blow drier, hair dryer, hair drier
+hand-held computer, hand-held microcomputer
+handkerchief, hankie, hanky, hankey
+hard disc, hard disk, fixed disk
+harmonica, mouth organ, harp, mouth harp
+harp
+harvester, reaper
+hatchet
+holster
+home theater, home theatre
+honeycomb
+hook, claw
+hoopskirt, crinoline
+horizontal bar, high bar
+horse cart, horse-cart
+hourglass
+iPod
+iron, smoothing iron
+jack-o'-lantern
+jean, blue jean, denim
+jeep, landrover
+jersey, T-shirt, tee shirt
+jigsaw puzzle
+jinrikisha, ricksha, rickshaw
+joystick
+kimono
+knee pad
+knot
+lab coat, laboratory coat
+ladle
+lampshade, lamp shade
+laptop, laptop computer
+lawn mower, mower
+lens cap, lens cover
+letter opener, paper knife, paperknife
+library
+lifeboat
+lighter, light, igniter, ignitor
+limousine, limo
+liner, ocean liner
+lipstick, lip rouge
+Loafer
+lotion
+loudspeaker, speaker, speaker unit, loudspeaker system, speaker system
+loupe, jeweler's loupe
+lumbermill, sawmill
+magnetic compass
+mailbag, postbag
+mailbox, letter box
+maillot
+maillot, tank suit
+manhole cover
+maraca
+marimba, xylophone
+mask
+matchstick
+maypole
+maze, labyrinth
+measuring cup
+medicine chest, medicine cabinet
+megalith, megalithic structure
+microphone, mike
+microwave, microwave oven
+military uniform
+milk can
+minibus
+miniskirt, mini
+minivan
+missile
+mitten
+mixing bowl
+mobile home, manufactured home
+Model T
+modem
+monastery
+monitor
+moped
+mortar
+mortarboard
+mosque
+mosquito net
+motor scooter, scooter
+mountain bike, all-terrain bike, off-roader
+mountain tent
+mouse, computer mouse
+mousetrap
+moving van
+muzzle
+nail
+neck brace
+necklace
+nipple
+notebook, notebook computer
+obelisk
+oboe, hautboy, hautbois
+ocarina, sweet potato
+odometer, hodometer, mileometer, milometer
+oil filter
+organ, pipe organ
+oscilloscope, scope, cathode-ray oscilloscope, CRO
+overskirt
+oxcart
+oxygen mask
+packet
+paddle, boat paddle
+paddlewheel, paddle wheel
+padlock
+paintbrush
+pajama, pyjama, pj's, jammies
+palace
+panpipe, pandean pipe, syrinx
+paper towel
+parachute, chute
+parallel bars, bars
+park bench
+parking meter
+passenger car, coach, carriage
+patio, terrace
+pay-phone, pay-station
+pedestal, plinth, footstall
+pencil box, pencil case
+pencil sharpener
+perfume, essence
+Petri dish
+photocopier
+pick, plectrum, plectron
+pickelhaube
+picket fence, paling
+pickup, pickup truck
+pier
+piggy bank, penny bank
+pill bottle
+pillow
+ping-pong ball
+pinwheel
+pirate, pirate ship
+pitcher, ewer
+plane, carpenter's plane, woodworking plane
+planetarium
+plastic bag
+plate rack
+plow, plough
+plunger, plumber's helper
+Polaroid camera, Polaroid Land camera
+pole
+police van, police wagon, paddy wagon, patrol wagon, wagon, black Maria
+poncho
+pool table, billiard table, snooker table
+pop bottle, soda bottle
+pot, flowerpot
+potter's wheel
+power drill
+prayer rug, prayer mat
+printer
+prison, prison house
+projectile, missile
+projector
+puck, hockey puck
+punching bag, punch bag, punching ball, punchball
+purse
+quill, quill pen
+quilt, comforter, comfort, puff
+racer, race car, racing car
+racket, racquet
+radiator
+radio, wireless
+radio telescope, radio reflector
+rain barrel
+recreational vehicle, RV, R.V.
+reel
+reflex camera
+refrigerator, icebox
+remote control, remote
+restaurant, eating house, eating place, eatery
+revolver, six-gun, six-shooter
+rifle
+rocking chair, rocker
+rotisserie
+rubber eraser, rubber, pencil eraser
+rugby ball
+rule, ruler
+running shoe
+safe
+safety pin
+saltshaker, salt shaker
+sandal
+sarong
+sax, saxophone
+scabbard
+scale, weighing machine
+school bus
+schooner
+scoreboard
+screen, CRT screen
+screw
+screwdriver
+seat belt, seatbelt
+sewing machine
+shield, buckler
+shoe shop, shoe-shop, shoe store
+shoji
+shopping basket
+shopping cart
+shovel
+shower cap
+shower curtain
+ski
+ski mask
+sleeping bag
+slide rule, slipstick
+sliding door
+slot, one-armed bandit
+snorkel
+snowmobile
+snowplow, snowplough
+soap dispenser
+soccer ball
+sock
+solar dish, solar collector, solar furnace
+sombrero
+soup bowl
+space bar
+space heater
+space shuttle
+spatula
+speedboat
+spider web, spider's web
+spindle
+sports car, sport car
+spotlight, spot
+stage
+steam locomotive
+steel arch bridge
+steel drum
+stethoscope
+stole
+stone wall
+stopwatch, stop watch
+stove
+strainer
+streetcar, tram, tramcar, trolley, trolley car
+stretcher
+studio couch, day bed
+stupa, tope
+submarine, pigboat, sub, U-boat
+suit, suit of clothes
+sundial
+sunglass
+sunglasses, dark glasses, shades
+sunscreen, sunblock, sun blocker
+suspension bridge
+swab, swob, mop
+sweatshirt
+swimming trunks, bathing trunks
+swing
+switch, electric switch, electrical switch
+syringe
+table lamp
+tank, army tank, armored combat vehicle, armoured combat vehicle
+tape player
+teapot
+teddy, teddy bear
+television, television system
+tennis ball
+thatch, thatched roof
+theater curtain, theatre curtain
+thimble
+thresher, thrasher, threshing machine
+throne
+tile roof
+toaster
+tobacco shop, tobacconist shop, tobacconist
+toilet seat
+torch
+totem pole
+tow truck, tow car, wrecker
+toyshop
+tractor
+trailer truck, tractor trailer, trucking rig, rig, articulated lorry, semi
+tray
+trench coat
+tricycle, trike, velocipede
+trimaran
+tripod
+triumphal arch
+trolleybus, trolley coach, trackless trolley
+trombone
+tub, vat
+turnstile
+typewriter keyboard
+umbrella
+unicycle, monocycle
+upright, upright piano
+vacuum, vacuum cleaner
+vase
+vault
+velvet
+vending machine
+vestment
+viaduct
+violin, fiddle
+volleyball
+waffle iron
+wall clock
+wallet, billfold, notecase, pocketbook
+wardrobe, closet, press
+warplane, military plane
+washbasin, handbasin, washbowl, lavabo, wash-hand basin
+washer, automatic washer, washing machine
+water bottle
+water jug
+water tower
+whiskey jug
+whistle
+wig
+window screen
+window shade
+Windsor tie
+wine bottle
+wing
+wok
+wooden spoon
+wool, woolen, woollen
+worm fence, snake fence, snake-rail fence, Virginia fence
+wreck
+yawl
+yurt
+web site, website, internet site, site
+comic book
+crossword puzzle, crossword
+street sign
+traffic light, traffic signal, stoplight
+book jacket, dust cover, dust jacket, dust wrapper
+menu
+plate
+guacamole
+consomme
+hot pot, hotpot
+trifle
+ice cream, icecream
+ice lolly, lolly, lollipop, popsicle
+French loaf
+bagel, beigel
+pretzel
+cheeseburger
+hotdog, hot dog, red hot
+mashed potato
+head cabbage
+broccoli
+cauliflower
+zucchini, courgette
+spaghetti squash
+acorn squash
+butternut squash
+cucumber, cuke
+artichoke, globe artichoke
+bell pepper
+cardoon
+mushroom
+Granny Smith
+strawberry
+orange
+lemon
+fig
+pineapple, ananas
+banana
+jackfruit, jak, jack
+custard apple
+pomegranate
+hay
+carbonara
+chocolate sauce, chocolate syrup
+dough
+meat loaf, meatloaf
+pizza, pizza pie
+potpie
+burrito
+red wine
+espresso
+cup
+eggnog
+alp
+bubble
+cliff, drop, drop-off
+coral reef
+geyser
+lakeside, lakeshore
+promontory, headland, head, foreland
+sandbar, sand bar
+seashore, coast, seacoast, sea-coast
+valley, vale
+volcano
+ballplayer, baseball player
+groom, bridegroom
+scuba diver
+rapeseed
+daisy
+yellow lady's slipper, yellow lady-slipper, Cypripedium calceolus, Cypripedium parviflorum
+corn
+acorn
+hip, rose hip, rosehip
+buckeye, horse chestnut, conker
+coral fungus
+agaric
+gyromitra
+stinkhorn, carrion fungus
+earthstar
+hen-of-the-woods, hen of the woods, Polyporus frondosus, Grifola frondosa
+bolete
+ear, spike, capitulum
+toilet tissue, toilet paper, bathroom tissue
--- /dev/null
+#! /usr/bin/env python3
+
+import numpy as np
+import tensorflow as tf
+
+from tensorflow.contrib.slim.nets import inception
+
+slim = tf.contrib.slim
+
+def run(name, image_size, num_classes):
+ with tf.Graph().as_default():
+ image = tf.placeholder("float", [1, image_size, image_size, 3], name="input")
+ with slim.arg_scope(inception.inception_v1_arg_scope()):
+ logits, _ = inception.inception_v1(image, num_classes, is_training=False, spatial_squeeze=False)
+ probabilities = tf.nn.softmax(logits)
+ init_fn = slim.assign_from_checkpoint_fn('inception_v1.ckpt', slim.get_model_variables('InceptionV1'))
+
+ with tf.Session() as sess:
+ init_fn(sess)
+ saver = tf.train.Saver(tf.global_variables())
+ saver.save(sess, "output/"+name)
+
+run('inception-v1', 224, 1001)
--- /dev/null
+224
\ No newline at end of file
--- /dev/null
+#! /usr/bin/env python3
+
+# Copyright 2017 Intel Corporation.
+# The source code, information and material ("Material") contained herein is
+# owned by Intel Corporation or its suppliers or licensors, and title to such
+# Material remains with Intel Corporation or its suppliers or licensors.
+# The Material contains proprietary information of Intel or its suppliers and
+# licensors. The Material is protected by worldwide copyright laws and treaty
+# provisions.
+# No part of the Material may be used, copied, reproduced, modified, published,
+# uploaded, posted, transmitted, distributed or disclosed in any way without
+# Intel's prior express written permission. No license under any patent,
+# copyright or other intellectual property rights in the Material is granted to
+# or conferred upon you, either expressly, by implication, inducement, estoppel
+# or otherwise.
+# Any license under such intellectual property rights must be express and
+# approved by Intel in writing.
+
+from mvnc import mvncapi as mvnc
+import sys
+import numpy
+import cv2
+
+path_to_networks = './'
+path_to_images = '../../data/images/'
+graph_filename = 'graph'
+image_filename = path_to_images + 'nps_electric_guitar.png'
+
+#mvnc.SetGlobalOption(mvnc.GlobalOption.LOGLEVEL, 2)
+devices = mvnc.EnumerateDevices()
+if len(devices) == 0:
+ print('No devices found')
+ quit()
+
+device = mvnc.Device(devices[0])
+device.OpenDevice()
+
+#Load graph
+with open(path_to_networks + graph_filename, mode='rb') as f:
+ graphfile = f.read()
+
+#Load preprocessing data
+mean = 128
+std = 1/128
+
+#Load categories
+categories = []
+with open(path_to_networks + 'categories.txt', 'r') as f:
+ for line in f:
+ cat = line.split('\n')[0]
+ if cat != 'classes':
+ categories.append(cat)
+ f.close()
+ print('Number of categories:', len(categories))
+
+#Load image size
+with open(path_to_networks + 'inputsize.txt', 'r') as f:
+ reqsize = int(f.readline().split('\n')[0])
+
+graph = device.AllocateGraph(graphfile)
+
+img = cv2.imread(image_filename).astype(numpy.float32)
+
+dx,dy,dz= img.shape
+delta=float(abs(dy-dx))
+if dx > dy: #crop the x dimension
+ img=img[int(0.5*delta):dx-int(0.5*delta),0:dy]
+else:
+ img=img[0:dx,int(0.5*delta):dy-int(0.5*delta)]
+
+img = cv2.resize(img, (reqsize, reqsize))
+
+img=cv2.cvtColor(img,cv2.COLOR_BGR2RGB)
+
+for i in range(3):
+ img[:,:,i] = (img[:,:,i] - mean) * std
+
+print('Start download to NCS...')
+graph.LoadTensor(img.astype(numpy.float16), 'user object')
+output, userobj = graph.GetResult()
+
+top_inds = output.argsort()[::-1][:5]
+
+print(''.join(['*' for i in range(79)]))
+print('inception-v1 on NCS')
+print(''.join(['*' for i in range(79)]))
+for i in range(5):
+ print(top_inds[i], categories[top_inds[i]], output[top_inds[i]])
+
+print(''.join(['*' for i in range(79)]))
+graph.DeallocateGraph()
+device.CloseDevice()
+print('Finished')
--- /dev/null
+
+ifneq ($(findstring movidius, $(PYTHONPATH)), movidius)
+ export PYTHONPATH:=/opt/movidius/caffe/python:/opt/movidius/mvnc/python:$(PYTHONPATH)
+endif
+
+NCCOMPILE = mvNCCompile
+NCPROFILE = mvNCProfile
+NCCHECK = mvNCCheck
+
+.IGNORE: profile check compile
+
+WEIGHTS_FILENAME = inception_v3_2016_08_28.tar.gz
+GET_WEIGHTS = (wget http://download.tensorflow.org/models/${WEIGHTS_FILENAME} && tar zxf ${WEIGHTS_FILENAME} && rm ${WEIGHTS_FILENAME})
+
+MODEL_FILENAME = output/inception-v3.meta
+CONV_SCRIPT = ./inception-v3.py
+
+INPUT_NODE_FLAG = -in=input
+OUTPUT_NODE_FLAG = -on=InceptionV3/Predictions/Reshape_1
+
+.PHONY: all
+all: profile check compile
+
+.PHONY: prereqs
+prereqs:
+ (cd ../../data/ilsvrc12; make)
+ @sed -i 's/\r//' run.py
+ @chmod +x run.py
+
+.PHONY: profile
+profile: weights
+ ${NCPROFILE} -s 12 ${MODEL_FILENAME} ${INPUT_NODE_FLAG} ${OUTPUT_NODE_FLAG}
+
+.PHONY: browse_profile
+browse_profile: weights
+ ${NCPROFILE} -s 12 ${MODEL_FILENAME} ${INPUT_NODE_FLAG} ${OUTPUT_NODE_FLAG}
+ firefox output_report.html &
+
+.PHONY: weights
+weights:
+ @sed -i 's/\r//' ${CONV_SCRIPT}
+ @chmod +x ${CONV_SCRIPT}
+ test -f ${MODEL_FILENAME} || (${GET_WEIGHTS} && ${CONV_SCRIPT})
+
+.PHONY: compile
+compile: weights
+ test -f graph || ${NCCOMPILE} -s 12 ${MODEL_FILENAME} ${INPUT_NODE_FLAG} ${OUTPUT_NODE_FLAG}
+
+.PHONY: check
+check: weights
+ -${NCCHECK} -s 12 ${MODEL_FILENAME} ${INPUT_NODE_FLAG} ${OUTPUT_NODE_FLAG} -i ../../data/images/cat.jpg -id 917 -M 128 -S 2 -cs 0,1,2
+
+.PHONY: run
+run: compile
+ ./run.py
+
+.PHONY: run_py
+run_py: compile
+ ./run.py
+
+.PHONY: help
+help:
+ @echo "possible make targets: ";
+ @echo " make help - shows this message";
+ @echo " make all - makes the following: prototxt, profile, compile, check, cpp, run_py, run_cpp";
+ @echo " make weights - downloads the trained model";
+ @echo " make compile - runs SDK compiler tool to compile the NCS graph file for the network";
+ @echo " make check - runs SDK checker tool to verify an NCS graph file";
+ @echo " make profile - runs the SDK profiler tool to profile the network creating output_report.html";
+ @echo " make browse_profile - runs the SDK profiler tool and brings up report in browser.";
+ @echo " make run_py - runs the run.py python example program";
+ @echo " make clean - removes all created content"
+
+clean:
+ rm -f output.gv
+ rm -f output.gv.svg
+ rm -f output_report.html
+ rm -f output_expected.npy
+ rm -f *.ckpt
+ rm -f output_result.npy
+ rm -f output_val.csv
+ rm -rf output
+
--- /dev/null
+# Introduction
+The [inception v3](https://github.com/tensorflow/models/tree/master/research/slim) network can be used for image classification. The provided Makefile does the following
+1. Downloads the TensorFlow checkpoint file
+2. Runs the conversion/save python script to generate network.meta file.
+3. Profiles, Compiles and Checks the network using the Neural Compute SDK.
+4. There is a run.py provided that does a single inference on a provided image as an example on how to use the network using the Neural Compute API
+
+# Makefile
+Provided Makefile describes various targets that help with the above mentioned tasks.
+
+## make all
+Runs ncprofile, nccompile and run.
+
+## make profile
+Runs the provided network on the NCS and generates per layer statistics that are helpful for understanding the performance of the network on the Neural Compute Stick.
+
+## make compile
+Uses the network description and the trained weights files to generate a Movidius internal 'graph' format file. This file is later used for loading the network on to the Neural Compute Stick and executing the network.
+
+## make run
+Runs the provided run.py file which sends a single image to the Neural Compute Stick and receives and displays the inference results.
+
+## make check
+Runs the network on Caffe on CPU and runs the network on the Neural Compute Stick. Check then compares the two results to make sure they are consistent with each other.
+
+## make clean
+Removes all the temporary files that are created by the Makefile
--- /dev/null
+classes
+background
+tench, Tinca tinca
+goldfish, Carassius auratus
+great white shark, white shark, man-eater, man-eating shark, Carcharodon carcharias
+tiger shark, Galeocerdo cuvieri
+hammerhead, hammerhead shark
+electric ray, crampfish, numbfish, torpedo
+stingray
+cock
+hen
+ostrich, Struthio camelus
+brambling, Fringilla montifringilla
+goldfinch, Carduelis carduelis
+house finch, linnet, Carpodacus mexicanus
+junco, snowbird
+indigo bunting, indigo finch, indigo bird, Passerina cyanea
+robin, American robin, Turdus migratorius
+bulbul
+jay
+magpie
+chickadee
+water ouzel, dipper
+kite
+bald eagle, American eagle, Haliaeetus leucocephalus
+vulture
+great grey owl, great gray owl, Strix nebulosa
+European fire salamander, Salamandra salamandra
+common newt, Triturus vulgaris
+eft
+spotted salamander, Ambystoma maculatum
+axolotl, mud puppy, Ambystoma mexicanum
+bullfrog, Rana catesbeiana
+tree frog, tree-frog
+tailed frog, bell toad, ribbed toad, tailed toad, Ascaphus trui
+loggerhead, loggerhead turtle, Caretta caretta
+leatherback turtle, leatherback, leathery turtle, Dermochelys coriacea
+mud turtle
+terrapin
+box turtle, box tortoise
+banded gecko
+common iguana, iguana, Iguana iguana
+American chameleon, anole, Anolis carolinensis
+whiptail, whiptail lizard
+agama
+frilled lizard, Chlamydosaurus kingi
+alligator lizard
+Gila monster, Heloderma suspectum
+green lizard, Lacerta viridis
+African chameleon, Chamaeleo chamaeleon
+Komodo dragon, Komodo lizard, dragon lizard, giant lizard, Varanus komodoensis
+African crocodile, Nile crocodile, Crocodylus niloticus
+American alligator, Alligator mississipiensis
+triceratops
+thunder snake, worm snake, Carphophis amoenus
+ringneck snake, ring-necked snake, ring snake
+hognose snake, puff adder, sand viper
+green snake, grass snake
+king snake, kingsnake
+garter snake, grass snake
+water snake
+vine snake
+night snake, Hypsiglena torquata
+boa constrictor, Constrictor constrictor
+rock python, rock snake, Python sebae
+Indian cobra, Naja naja
+green mamba
+sea snake
+horned viper, cerastes, sand viper, horned asp, Cerastes cornutus
+diamondback, diamondback rattlesnake, Crotalus adamanteus
+sidewinder, horned rattlesnake, Crotalus cerastes
+trilobite
+harvestman, daddy longlegs, Phalangium opilio
+scorpion
+black and gold garden spider, Argiope aurantia
+barn spider, Araneus cavaticus
+garden spider, Aranea diademata
+black widow, Latrodectus mactans
+tarantula
+wolf spider, hunting spider
+tick
+centipede
+black grouse
+ptarmigan
+ruffed grouse, partridge, Bonasa umbellus
+prairie chicken, prairie grouse, prairie fowl
+peacock
+quail
+partridge
+African grey, African gray, Psittacus erithacus
+macaw
+sulphur-crested cockatoo, Kakatoe galerita, Cacatua galerita
+lorikeet
+coucal
+bee eater
+hornbill
+hummingbird
+jacamar
+toucan
+drake
+red-breasted merganser, Mergus serrator
+goose
+black swan, Cygnus atratus
+tusker
+echidna, spiny anteater, anteater
+platypus, duckbill, duckbilled platypus, duck-billed platypus, Ornithorhynchus anatinus
+wallaby, brush kangaroo
+koala, koala bear, kangaroo bear, native bear, Phascolarctos cinereus
+wombat
+jellyfish
+sea anemone, anemone
+brain coral
+flatworm, platyhelminth
+nematode, nematode worm, roundworm
+conch
+snail
+slug
+sea slug, nudibranch
+chiton, coat-of-mail shell, sea cradle, polyplacophore
+chambered nautilus, pearly nautilus, nautilus
+Dungeness crab, Cancer magister
+rock crab, Cancer irroratus
+fiddler crab
+king crab, Alaska crab, Alaskan king crab, Alaska king crab, Paralithodes camtschatica
+American lobster, Northern lobster, Maine lobster, Homarus americanus
+spiny lobster, langouste, rock lobster, crawfish, crayfish, sea crawfish
+crayfish, crawfish, crawdad, crawdaddy
+hermit crab
+isopod
+white stork, Ciconia ciconia
+black stork, Ciconia nigra
+spoonbill
+flamingo
+little blue heron, Egretta caerulea
+American egret, great white heron, Egretta albus
+bittern
+crane
+limpkin, Aramus pictus
+European gallinule, Porphyrio porphyrio
+American coot, marsh hen, mud hen, water hen, Fulica americana
+bustard
+ruddy turnstone, Arenaria interpres
+red-backed sandpiper, dunlin, Erolia alpina
+redshank, Tringa totanus
+dowitcher
+oystercatcher, oyster catcher
+pelican
+king penguin, Aptenodytes patagonica
+albatross, mollymawk
+grey whale, gray whale, devilfish, Eschrichtius gibbosus, Eschrichtius robustus
+killer whale, killer, orca, grampus, sea wolf, Orcinus orca
+dugong, Dugong dugon
+sea lion
+Chihuahua
+Japanese spaniel
+Maltese dog, Maltese terrier, Maltese
+Pekinese, Pekingese, Peke
+Shih-Tzu
+Blenheim spaniel
+papillon
+toy terrier
+Rhodesian ridgeback
+Afghan hound, Afghan
+basset, basset hound
+beagle
+bloodhound, sleuthhound
+bluetick
+black-and-tan coonhound
+Walker hound, Walker foxhound
+English foxhound
+redbone
+borzoi, Russian wolfhound
+Irish wolfhound
+Italian greyhound
+whippet
+Ibizan hound, Ibizan Podenco
+Norwegian elkhound, elkhound
+otterhound, otter hound
+Saluki, gazelle hound
+Scottish deerhound, deerhound
+Weimaraner
+Staffordshire bullterrier, Staffordshire bull terrier
+American Staffordshire terrier, Staffordshire terrier, American pit bull terrier, pit bull terrier
+Bedlington terrier
+Border terrier
+Kerry blue terrier
+Irish terrier
+Norfolk terrier
+Norwich terrier
+Yorkshire terrier
+wire-haired fox terrier
+Lakeland terrier
+Sealyham terrier, Sealyham
+Airedale, Airedale terrier
+cairn, cairn terrier
+Australian terrier
+Dandie Dinmont, Dandie Dinmont terrier
+Boston bull, Boston terrier
+miniature schnauzer
+giant schnauzer
+standard schnauzer
+Scotch terrier, Scottish terrier, Scottie
+Tibetan terrier, chrysanthemum dog
+silky terrier, Sydney silky
+soft-coated wheaten terrier
+West Highland white terrier
+Lhasa, Lhasa apso
+flat-coated retriever
+curly-coated retriever
+golden retriever
+Labrador retriever
+Chesapeake Bay retriever
+German short-haired pointer
+vizsla, Hungarian pointer
+English setter
+Irish setter, red setter
+Gordon setter
+Brittany spaniel
+clumber, clumber spaniel
+English springer, English springer spaniel
+Welsh springer spaniel
+cocker spaniel, English cocker spaniel, cocker
+Sussex spaniel
+Irish water spaniel
+kuvasz
+schipperke
+groenendael
+malinois
+briard
+kelpie
+komondor
+Old English sheepdog, bobtail
+Shetland sheepdog, Shetland sheep dog, Shetland
+collie
+Border collie
+Bouvier des Flandres, Bouviers des Flandres
+Rottweiler
+German shepherd, German shepherd dog, German police dog, alsatian
+Doberman, Doberman pinscher
+miniature pinscher
+Greater Swiss Mountain dog
+Bernese mountain dog
+Appenzeller
+EntleBucher
+boxer
+bull mastiff
+Tibetan mastiff
+French bulldog
+Great Dane
+Saint Bernard, St Bernard
+Eskimo dog, husky
+malamute, malemute, Alaskan malamute
+Siberian husky
+dalmatian, coach dog, carriage dog
+affenpinscher, monkey pinscher, monkey dog
+basenji
+pug, pug-dog
+Leonberg
+Newfoundland, Newfoundland dog
+Great Pyrenees
+Samoyed, Samoyede
+Pomeranian
+chow, chow chow
+keeshond
+Brabancon griffon
+Pembroke, Pembroke Welsh corgi
+Cardigan, Cardigan Welsh corgi
+toy poodle
+miniature poodle
+standard poodle
+Mexican hairless
+timber wolf, grey wolf, gray wolf, Canis lupus
+white wolf, Arctic wolf, Canis lupus tundrarum
+red wolf, maned wolf, Canis rufus, Canis niger
+coyote, prairie wolf, brush wolf, Canis latrans
+dingo, warrigal, warragal, Canis dingo
+dhole, Cuon alpinus
+African hunting dog, hyena dog, Cape hunting dog, Lycaon pictus
+hyena, hyaena
+red fox, Vulpes vulpes
+kit fox, Vulpes macrotis
+Arctic fox, white fox, Alopex lagopus
+grey fox, gray fox, Urocyon cinereoargenteus
+tabby, tabby cat
+tiger cat
+Persian cat
+Siamese cat, Siamese
+Egyptian cat
+cougar, puma, catamount, mountain lion, painter, panther, Felis concolor
+lynx, catamount
+leopard, Panthera pardus
+snow leopard, ounce, Panthera uncia
+jaguar, panther, Panthera onca, Felis onca
+lion, king of beasts, Panthera leo
+tiger, Panthera tigris
+cheetah, chetah, Acinonyx jubatus
+brown bear, bruin, Ursus arctos
+American black bear, black bear, Ursus americanus, Euarctos americanus
+ice bear, polar bear, Ursus Maritimus, Thalarctos maritimus
+sloth bear, Melursus ursinus, Ursus ursinus
+mongoose
+meerkat, mierkat
+tiger beetle
+ladybug, ladybeetle, lady beetle, ladybird, ladybird beetle
+ground beetle, carabid beetle
+long-horned beetle, longicorn, longicorn beetle
+leaf beetle, chrysomelid
+dung beetle
+rhinoceros beetle
+weevil
+fly
+bee
+ant, emmet, pismire
+grasshopper, hopper
+cricket
+walking stick, walkingstick, stick insect
+cockroach, roach
+mantis, mantid
+cicada, cicala
+leafhopper
+lacewing, lacewing fly
+dragonfly, darning needle, devil's darning needle, sewing needle, snake feeder, snake doctor, mosquito hawk, skeeter hawk
+damselfly
+admiral
+ringlet, ringlet butterfly
+monarch, monarch butterfly, milkweed butterfly, Danaus plexippus
+cabbage butterfly
+sulphur butterfly, sulfur butterfly
+lycaenid, lycaenid butterfly
+starfish, sea star
+sea urchin
+sea cucumber, holothurian
+wood rabbit, cottontail, cottontail rabbit
+hare
+Angora, Angora rabbit
+hamster
+porcupine, hedgehog
+fox squirrel, eastern fox squirrel, Sciurus niger
+marmot
+beaver
+guinea pig, Cavia cobaya
+sorrel
+zebra
+hog, pig, grunter, squealer, Sus scrofa
+wild boar, boar, Sus scrofa
+warthog
+hippopotamus, hippo, river horse, Hippopotamus amphibius
+ox
+water buffalo, water ox, Asiatic buffalo, Bubalus bubalis
+bison
+ram, tup
+bighorn, bighorn sheep, cimarron, Rocky Mountain bighorn, Rocky Mountain sheep, Ovis canadensis
+ibex, Capra ibex
+hartebeest
+impala, Aepyceros melampus
+gazelle
+Arabian camel, dromedary, Camelus dromedarius
+llama
+weasel
+mink
+polecat, fitch, foulmart, foumart, Mustela putorius
+black-footed ferret, ferret, Mustela nigripes
+otter
+skunk, polecat, wood pussy
+badger
+armadillo
+three-toed sloth, ai, Bradypus tridactylus
+orangutan, orang, orangutang, Pongo pygmaeus
+gorilla, Gorilla gorilla
+chimpanzee, chimp, Pan troglodytes
+gibbon, Hylobates lar
+siamang, Hylobates syndactylus, Symphalangus syndactylus
+guenon, guenon monkey
+patas, hussar monkey, Erythrocebus patas
+baboon
+macaque
+langur
+colobus, colobus monkey
+proboscis monkey, Nasalis larvatus
+marmoset
+capuchin, ringtail, Cebus capucinus
+howler monkey, howler
+titi, titi monkey
+spider monkey, Ateles geoffroyi
+squirrel monkey, Saimiri sciureus
+Madagascar cat, ring-tailed lemur, Lemur catta
+indri, indris, Indri indri, Indri brevicaudatus
+Indian elephant, Elephas maximus
+African elephant, Loxodonta africana
+lesser panda, red panda, panda, bear cat, cat bear, Ailurus fulgens
+giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca
+barracouta, snoek
+eel
+coho, cohoe, coho salmon, blue jack, silver salmon, Oncorhynchus kisutch
+rock beauty, Holocanthus tricolor
+anemone fish
+sturgeon
+gar, garfish, garpike, billfish, Lepisosteus osseus
+lionfish
+puffer, pufferfish, blowfish, globefish
+abacus
+abaya
+academic gown, academic robe, judge's robe
+accordion, piano accordion, squeeze box
+acoustic guitar
+aircraft carrier, carrier, flattop, attack aircraft carrier
+airliner
+airship, dirigible
+altar
+ambulance
+amphibian, amphibious vehicle
+analog clock
+apiary, bee house
+apron
+ashcan, trash can, garbage can, wastebin, ash bin, ash-bin, ashbin, dustbin, trash barrel, trash bin
+assault rifle, assault gun
+backpack, back pack, knapsack, packsack, rucksack, haversack
+bakery, bakeshop, bakehouse
+balance beam, beam
+balloon
+ballpoint, ballpoint pen, ballpen, Biro
+Band Aid
+banjo
+bannister, banister, balustrade, balusters, handrail
+barbell
+barber chair
+barbershop
+barn
+barometer
+barrel, cask
+barrow, garden cart, lawn cart, wheelbarrow
+baseball
+basketball
+bassinet
+bassoon
+bathing cap, swimming cap
+bath towel
+bathtub, bathing tub, bath, tub
+beach wagon, station wagon, wagon, estate car, beach waggon, station waggon, waggon
+beacon, lighthouse, beacon light, pharos
+beaker
+bearskin, busby, shako
+beer bottle
+beer glass
+bell cote, bell cot
+bib
+bicycle-built-for-two, tandem bicycle, tandem
+bikini, two-piece
+binder, ring-binder
+binoculars, field glasses, opera glasses
+birdhouse
+boathouse
+bobsled, bobsleigh, bob
+bolo tie, bolo, bola tie, bola
+bonnet, poke bonnet
+bookcase
+bookshop, bookstore, bookstall
+bottlecap
+bow
+bow tie, bow-tie, bowtie
+brass, memorial tablet, plaque
+brassiere, bra, bandeau
+breakwater, groin, groyne, mole, bulwark, seawall, jetty
+breastplate, aegis, egis
+broom
+bucket, pail
+buckle
+bulletproof vest
+bullet train, bullet
+butcher shop, meat market
+cab, hack, taxi, taxicab
+caldron, cauldron
+candle, taper, wax light
+cannon
+canoe
+can opener, tin opener
+cardigan
+car mirror
+carousel, carrousel, merry-go-round, roundabout, whirligig
+carpenter's kit, tool kit
+carton
+car wheel
+cash machine, cash dispenser, automated teller machine, automatic teller machine, automated teller, automatic teller, ATM
+cassette
+cassette player
+castle
+catamaran
+CD player
+cello, violoncello
+cellular telephone, cellular phone, cellphone, cell, mobile phone
+chain
+chainlink fence
+chain mail, ring mail, mail, chain armor, chain armour, ring armor, ring armour
+chain saw, chainsaw
+chest
+chiffonier, commode
+chime, bell, gong
+china cabinet, china closet
+Christmas stocking
+church, church building
+cinema, movie theater, movie theatre, movie house, picture palace
+cleaver, meat cleaver, chopper
+cliff dwelling
+cloak
+clog, geta, patten, sabot
+cocktail shaker
+coffee mug
+coffeepot
+coil, spiral, volute, whorl, helix
+combination lock
+computer keyboard, keypad
+confectionery, confectionary, candy store
+container ship, containership, container vessel
+convertible
+corkscrew, bottle screw
+cornet, horn, trumpet, trump
+cowboy boot
+cowboy hat, ten-gallon hat
+cradle
+crane
+crash helmet
+crate
+crib, cot
+Crock Pot
+croquet ball
+crutch
+cuirass
+dam, dike, dyke
+desk
+desktop computer
+dial telephone, dial phone
+diaper, nappy, napkin
+digital clock
+digital watch
+dining table, board
+dishrag, dishcloth
+dishwasher, dish washer, dishwashing machine
+disk brake, disc brake
+dock, dockage, docking facility
+dogsled, dog sled, dog sleigh
+dome
+doormat, welcome mat
+drilling platform, offshore rig
+drum, membranophone, tympan
+drumstick
+dumbbell
+Dutch oven
+electric fan, blower
+electric guitar
+electric locomotive
+entertainment center
+envelope
+espresso maker
+face powder
+feather boa, boa
+file, file cabinet, filing cabinet
+fireboat
+fire engine, fire truck
+fire screen, fireguard
+flagpole, flagstaff
+flute, transverse flute
+folding chair
+football helmet
+forklift
+fountain
+fountain pen
+four-poster
+freight car
+French horn, horn
+frying pan, frypan, skillet
+fur coat
+garbage truck, dustcart
+gasmask, respirator, gas helmet
+gas pump, gasoline pump, petrol pump, island dispenser
+goblet
+go-kart
+golf ball
+golfcart, golf cart
+gondola
+gong, tam-tam
+gown
+grand piano, grand
+greenhouse, nursery, glasshouse
+grille, radiator grille
+grocery store, grocery, food market, market
+guillotine
+hair slide
+hair spray
+half track
+hammer
+hamper
+hand blower, blow dryer, blow drier, hair dryer, hair drier
+hand-held computer, hand-held microcomputer
+handkerchief, hankie, hanky, hankey
+hard disc, hard disk, fixed disk
+harmonica, mouth organ, harp, mouth harp
+harp
+harvester, reaper
+hatchet
+holster
+home theater, home theatre
+honeycomb
+hook, claw
+hoopskirt, crinoline
+horizontal bar, high bar
+horse cart, horse-cart
+hourglass
+iPod
+iron, smoothing iron
+jack-o'-lantern
+jean, blue jean, denim
+jeep, landrover
+jersey, T-shirt, tee shirt
+jigsaw puzzle
+jinrikisha, ricksha, rickshaw
+joystick
+kimono
+knee pad
+knot
+lab coat, laboratory coat
+ladle
+lampshade, lamp shade
+laptop, laptop computer
+lawn mower, mower
+lens cap, lens cover
+letter opener, paper knife, paperknife
+library
+lifeboat
+lighter, light, igniter, ignitor
+limousine, limo
+liner, ocean liner
+lipstick, lip rouge
+Loafer
+lotion
+loudspeaker, speaker, speaker unit, loudspeaker system, speaker system
+loupe, jeweler's loupe
+lumbermill, sawmill
+magnetic compass
+mailbag, postbag
+mailbox, letter box
+maillot
+maillot, tank suit
+manhole cover
+maraca
+marimba, xylophone
+mask
+matchstick
+maypole
+maze, labyrinth
+measuring cup
+medicine chest, medicine cabinet
+megalith, megalithic structure
+microphone, mike
+microwave, microwave oven
+military uniform
+milk can
+minibus
+miniskirt, mini
+minivan
+missile
+mitten
+mixing bowl
+mobile home, manufactured home
+Model T
+modem
+monastery
+monitor
+moped
+mortar
+mortarboard
+mosque
+mosquito net
+motor scooter, scooter
+mountain bike, all-terrain bike, off-roader
+mountain tent
+mouse, computer mouse
+mousetrap
+moving van
+muzzle
+nail
+neck brace
+necklace
+nipple
+notebook, notebook computer
+obelisk
+oboe, hautboy, hautbois
+ocarina, sweet potato
+odometer, hodometer, mileometer, milometer
+oil filter
+organ, pipe organ
+oscilloscope, scope, cathode-ray oscilloscope, CRO
+overskirt
+oxcart
+oxygen mask
+packet
+paddle, boat paddle
+paddlewheel, paddle wheel
+padlock
+paintbrush
+pajama, pyjama, pj's, jammies
+palace
+panpipe, pandean pipe, syrinx
+paper towel
+parachute, chute
+parallel bars, bars
+park bench
+parking meter
+passenger car, coach, carriage
+patio, terrace
+pay-phone, pay-station
+pedestal, plinth, footstall
+pencil box, pencil case
+pencil sharpener
+perfume, essence
+Petri dish
+photocopier
+pick, plectrum, plectron
+pickelhaube
+picket fence, paling
+pickup, pickup truck
+pier
+piggy bank, penny bank
+pill bottle
+pillow
+ping-pong ball
+pinwheel
+pirate, pirate ship
+pitcher, ewer
+plane, carpenter's plane, woodworking plane
+planetarium
+plastic bag
+plate rack
+plow, plough
+plunger, plumber's helper
+Polaroid camera, Polaroid Land camera
+pole
+police van, police wagon, paddy wagon, patrol wagon, wagon, black Maria
+poncho
+pool table, billiard table, snooker table
+pop bottle, soda bottle
+pot, flowerpot
+potter's wheel
+power drill
+prayer rug, prayer mat
+printer
+prison, prison house
+projectile, missile
+projector
+puck, hockey puck
+punching bag, punch bag, punching ball, punchball
+purse
+quill, quill pen
+quilt, comforter, comfort, puff
+racer, race car, racing car
+racket, racquet
+radiator
+radio, wireless
+radio telescope, radio reflector
+rain barrel
+recreational vehicle, RV, R.V.
+reel
+reflex camera
+refrigerator, icebox
+remote control, remote
+restaurant, eating house, eating place, eatery
+revolver, six-gun, six-shooter
+rifle
+rocking chair, rocker
+rotisserie
+rubber eraser, rubber, pencil eraser
+rugby ball
+rule, ruler
+running shoe
+safe
+safety pin
+saltshaker, salt shaker
+sandal
+sarong
+sax, saxophone
+scabbard
+scale, weighing machine
+school bus
+schooner
+scoreboard
+screen, CRT screen
+screw
+screwdriver
+seat belt, seatbelt
+sewing machine
+shield, buckler
+shoe shop, shoe-shop, shoe store
+shoji
+shopping basket
+shopping cart
+shovel
+shower cap
+shower curtain
+ski
+ski mask
+sleeping bag
+slide rule, slipstick
+sliding door
+slot, one-armed bandit
+snorkel
+snowmobile
+snowplow, snowplough
+soap dispenser
+soccer ball
+sock
+solar dish, solar collector, solar furnace
+sombrero
+soup bowl
+space bar
+space heater
+space shuttle
+spatula
+speedboat
+spider web, spider's web
+spindle
+sports car, sport car
+spotlight, spot
+stage
+steam locomotive
+steel arch bridge
+steel drum
+stethoscope
+stole
+stone wall
+stopwatch, stop watch
+stove
+strainer
+streetcar, tram, tramcar, trolley, trolley car
+stretcher
+studio couch, day bed
+stupa, tope
+submarine, pigboat, sub, U-boat
+suit, suit of clothes
+sundial
+sunglass
+sunglasses, dark glasses, shades
+sunscreen, sunblock, sun blocker
+suspension bridge
+swab, swob, mop
+sweatshirt
+swimming trunks, bathing trunks
+swing
+switch, electric switch, electrical switch
+syringe
+table lamp
+tank, army tank, armored combat vehicle, armoured combat vehicle
+tape player
+teapot
+teddy, teddy bear
+television, television system
+tennis ball
+thatch, thatched roof
+theater curtain, theatre curtain
+thimble
+thresher, thrasher, threshing machine
+throne
+tile roof
+toaster
+tobacco shop, tobacconist shop, tobacconist
+toilet seat
+torch
+totem pole
+tow truck, tow car, wrecker
+toyshop
+tractor
+trailer truck, tractor trailer, trucking rig, rig, articulated lorry, semi
+tray
+trench coat
+tricycle, trike, velocipede
+trimaran
+tripod
+triumphal arch
+trolleybus, trolley coach, trackless trolley
+trombone
+tub, vat
+turnstile
+typewriter keyboard
+umbrella
+unicycle, monocycle
+upright, upright piano
+vacuum, vacuum cleaner
+vase
+vault
+velvet
+vending machine
+vestment
+viaduct
+violin, fiddle
+volleyball
+waffle iron
+wall clock
+wallet, billfold, notecase, pocketbook
+wardrobe, closet, press
+warplane, military plane
+washbasin, handbasin, washbowl, lavabo, wash-hand basin
+washer, automatic washer, washing machine
+water bottle
+water jug
+water tower
+whiskey jug
+whistle
+wig
+window screen
+window shade
+Windsor tie
+wine bottle
+wing
+wok
+wooden spoon
+wool, woolen, woollen
+worm fence, snake fence, snake-rail fence, Virginia fence
+wreck
+yawl
+yurt
+web site, website, internet site, site
+comic book
+crossword puzzle, crossword
+street sign
+traffic light, traffic signal, stoplight
+book jacket, dust cover, dust jacket, dust wrapper
+menu
+plate
+guacamole
+consomme
+hot pot, hotpot
+trifle
+ice cream, icecream
+ice lolly, lolly, lollipop, popsicle
+French loaf
+bagel, beigel
+pretzel
+cheeseburger
+hotdog, hot dog, red hot
+mashed potato
+head cabbage
+broccoli
+cauliflower
+zucchini, courgette
+spaghetti squash
+acorn squash
+butternut squash
+cucumber, cuke
+artichoke, globe artichoke
+bell pepper
+cardoon
+mushroom
+Granny Smith
+strawberry
+orange
+lemon
+fig
+pineapple, ananas
+banana
+jackfruit, jak, jack
+custard apple
+pomegranate
+hay
+carbonara
+chocolate sauce, chocolate syrup
+dough
+meat loaf, meatloaf
+pizza, pizza pie
+potpie
+burrito
+red wine
+espresso
+cup
+eggnog
+alp
+bubble
+cliff, drop, drop-off
+coral reef
+geyser
+lakeside, lakeshore
+promontory, headland, head, foreland
+sandbar, sand bar
+seashore, coast, seacoast, sea-coast
+valley, vale
+volcano
+ballplayer, baseball player
+groom, bridegroom
+scuba diver
+rapeseed
+daisy
+yellow lady's slipper, yellow lady-slipper, Cypripedium calceolus, Cypripedium parviflorum
+corn
+acorn
+hip, rose hip, rosehip
+buckeye, horse chestnut, conker
+coral fungus
+agaric
+gyromitra
+stinkhorn, carrion fungus
+earthstar
+hen-of-the-woods, hen of the woods, Polyporus frondosus, Grifola frondosa
+bolete
+ear, spike, capitulum
+toilet tissue, toilet paper, bathroom tissue
--- /dev/null
+#! /usr/bin/env python3
+
+import numpy as np
+import tensorflow as tf
+
+from tensorflow.contrib.slim.nets import inception
+
+slim = tf.contrib.slim
+
+def run(name, image_size, num_classes):
+ with tf.Graph().as_default():
+ image = tf.placeholder("float", [1, image_size, image_size, 3], name="input")
+ with slim.arg_scope(inception.inception_v3_arg_scope()):
+ logits, _ = inception.inception_v3(image, num_classes, is_training=False, spatial_squeeze=False)
+ probabilities = tf.nn.softmax(logits)
+ init_fn = slim.assign_from_checkpoint_fn('inception_v3.ckpt', slim.get_model_variables('InceptionV3'))
+
+ with tf.Session() as sess:
+ init_fn(sess)
+ saver = tf.train.Saver(tf.global_variables())
+ saver.save(sess, "output/"+name)
+
+run('inception-v3', 299, 1001)
--- /dev/null
+299
\ No newline at end of file
--- /dev/null
+#! /usr/bin/env python3
+
+# Copyright 2017 Intel Corporation.
+# The source code, information and material ("Material") contained herein is
+# owned by Intel Corporation or its suppliers or licensors, and title to such
+# Material remains with Intel Corporation or its suppliers or licensors.
+# The Material contains proprietary information of Intel or its suppliers and
+# licensors. The Material is protected by worldwide copyright laws and treaty
+# provisions.
+# No part of the Material may be used, copied, reproduced, modified, published,
+# uploaded, posted, transmitted, distributed or disclosed in any way without
+# Intel's prior express written permission. No license under any patent,
+# copyright or other intellectual property rights in the Material is granted to
+# or conferred upon you, either expressly, by implication, inducement, estoppel
+# or otherwise.
+# Any license under such intellectual property rights must be express and
+# approved by Intel in writing.
+
+from mvnc import mvncapi as mvnc
+import sys
+import numpy
+import cv2
+
+path_to_networks = './'
+path_to_images = '../../data/images/'
+graph_filename = 'graph'
+image_filename = path_to_images + 'nps_electric_guitar.png'
+
+#mvnc.SetGlobalOption(mvnc.GlobalOption.LOGLEVEL, 2)
+devices = mvnc.EnumerateDevices()
+if len(devices) == 0:
+ print('No devices found')
+ quit()
+
+device = mvnc.Device(devices[0])
+device.OpenDevice()
+
+#Load graph
+with open(path_to_networks + graph_filename, mode='rb') as f:
+ graphfile = f.read()
+
+#Load preprocessing data
+mean = 128
+std = 1/128
+
+#Load categories
+categories = []
+with open(path_to_networks + 'categories.txt', 'r') as f:
+ for line in f:
+ cat = line.split('\n')[0]
+ if cat != 'classes':
+ categories.append(cat)
+ f.close()
+ print('Number of categories:', len(categories))
+
+#Load image size
+with open(path_to_networks + 'inputsize.txt', 'r') as f:
+ reqsize = int(f.readline().split('\n')[0])
+
+graph = device.AllocateGraph(graphfile)
+
+img = cv2.imread(image_filename).astype(numpy.float32)
+dx,dy,dz= img.shape
+delta=float(abs(dy-dx))
+if dx > dy: #crop the x dimension
+ img=img[int(0.5*delta):dx-int(0.5*delta),0:dy]
+else:
+ img=img[0:dx,int(0.5*delta):dy-int(0.5*delta)]
+img = cv2.resize(img, (reqsize, reqsize))
+
+img=cv2.cvtColor(img,cv2.COLOR_BGR2RGB)
+
+for i in range(3):
+ img[:,:,i] = (img[:,:,i] - mean) * std
+
+print('Start download to NCS...')
+graph.LoadTensor(img.astype(numpy.float16), 'user object')
+output, userobj = graph.GetResult()
+
+top_inds = output.argsort()[::-1][:5]
+
+print(''.join(['*' for i in range(79)]))
+print('inception-v3 on NCS')
+print(''.join(['*' for i in range(79)]))
+for i in range(5):
+ print(top_inds[i], categories[top_inds[i]], output[top_inds[i]])
+
+print(''.join(['*' for i in range(79)]))
+graph.DeallocateGraph()
+device.CloseDevice()
+print('Finished')
--- /dev/null
+# Examples showing TensorFlow usage
--- /dev/null
+#! /bin/bash
+
+#test for python already installed for opencv
+python3 -c "import cv2" > /dev/null 2>&1
+if [ $? -eq 0 ] ;
+then
+ echo "";
+ echo "OpenCV already setup for python";
+ echo "";
+ exit 0
+fi;
+
+## Install OpenCV
+if [[ `lsb_release -d` =~ .*Raspbian.* ]]
+then
+ echo ""
+ echo "************************ Please confirm *******************************"
+ echo " Installing OpenCV on Raspberry Pi may take a long time. "
+ echo " You may skip this part of the installation in which case some examples "
+ echo " may not work without modifications but the rest of the SDK will still "
+ echo " be functional. Select n to skip OpenCV installation or y to install it."
+ read -p " Continue installing OpenCV (y/n) ? " CONTINUE
+ if [[ "$CONTINUE" == "y" || "$CONTINUE" == "Y" ]]; then
+ echo "";
+ echo "Installing OpenCV";
+ echo "";
+ sudo apt-get update -y && sudo apt-get upgrade -y
+ sudo apt-get install -y build-essential cmake pkg-config
+ sudo apt-get install -y libjpeg-dev libtiff5-dev libjasper-dev libpng12-dev
+ sudo apt-get install -y libavcodec-dev libavformat-dev libswscale-dev libv4l-dev
+ sudo apt-get install -y libxvidcore-dev libx264-dev
+ sudo apt-get install -y libgtk2.0-dev libgtk-3-dev
+ sudo apt-get install -y libatlas-base-dev gfortran
+ sudo apt-get install -y python2.7-dev python3-dev
+
+ cd ~
+ wget -O opencv.zip https://github.com/Itseez/opencv/archive/3.3.0.zip
+ unzip opencv.zip
+ wget -O opencv_contrib.zip https://github.com/Itseez/opencv_contrib/archive/3.3.0.zip
+ unzip opencv_contrib.zip
+ cd ~/opencv-3.3.0/
+ mkdir build
+ cd build
+ cmake -D CMAKE_BUILD_TYPE=RELEASE \
+ -D CMAKE_INSTALL_PREFIX=/usr/local \
+ -D INSTALL_PYTHON_EXAMPLES=OFF \
+ -D OPENCV_EXTRA_MODULES_PATH=~/opencv_contrib-3.3.0/modules \
+ -D BUILD_EXAMPLES=OFF ..
+ make
+ sudo make install
+ sudo ldconfig
+ else
+ echo "";
+ echo "Skipping OpenCV installation";
+ echo "";
+ fi
+else
+ echo "Installing opencv python for non-Raspbian";
+ sudo -H pip3 install opencv-python
+ sudo -H pip3 install opencv-contrib-python
+fi
--- /dev/null
+#! /bin/bash
+
+# download the file
+cp ncsdk.conf /tmp
+
+rm -f /tmp/ncsdk_redirector.txt
+
+if [ -f ncsdk_redirector.txt ]
+then
+ cp ncsdk_redirector.txt /tmp
+ cd /tmp
+else
+ cd /tmp
+ wget --no-cache http://ncs-forum-uploads.s3.amazonaws.com/ncsdk/ncsdk_01_09/ncsdk_redirector.txt
+fi
+
+download_filename=NCSDK-1.09.tar.gz
+
+# redirector is the url from redirector text file
+redirector=$(<ncsdk_redirector.txt)
+
+# ncsdk_link is the url with '\r' deleted if it exists
+# would probably be from editing in windows/dos
+ncsdk_link=$(echo "$redirector" | tr -d '\r')
+
+# download the payload from the redirector link
+# and save it the download_filename no matter what the original filename was
+wget --no-cache -O ${download_filename} $ncsdk_link
+
+# Get the filenames, ncsdk_archive is just the download filename
+# and ncsdk_pkg is the filename without the .tar.gz
+#ncsdk_archive=$(echo "$ncsdk_link" | grep -o '[^/]*$')
+ncsdk_archive=${download_filename}
+ncsdk_pkg=${ncsdk_archive%%.tar.gz}
+
+# now install-ncsdk.sh is doing the uninstallation if necessary.
+# Remove older installation dirs
+#if [ -f /opt/movidius/uninstall-ncsdk.sh ]
+#then
+# echo "Removing previous install..."
+# /opt/movidius/uninstall-ncsdk.sh
+#fi
+
+# Create Required installation dirs
+sudo mkdir -p /opt/movidius
+sudo cp $ncsdk_archive /opt/movidius/ncsdk.tar.gz
+cd /opt/movidius/
+
+# untar the new install and run the install script
+sudo tar zxvf ./ncsdk.tar.gz
+sudo rm -rf NCSDK
+sudo mv $ncsdk_pkg* NCSDK
+cd /opt/movidius/NCSDK
+cp /tmp/ncsdk.conf .
+./install-ncsdk.sh
+
+# leave the uninstall script on the target
+#sudo cp ./uninstall-ncsdk.sh ../ncsdk/
+
+# cleanup
+sudo rm ./install-ncsdk.sh
+cd ..
+sudo rm ncsdk.tar.gz
+#sudo rm -r NCSDK
+sudo rm -f /tmp/${ncsdk_archive}
+
+
+
+
--- /dev/null
+MAKE_PROCS=1
+SETUPDIR=/opt/movidius
+VERBOSE=yes
+SYSTEM_INSTALL=yes
+INTEL_CAFFE=no
+CAFFE_USE_CUDA=no
+INSTALL_TENSORFLOW=yes
+INSTALL_TOOLKIT=yes
--- /dev/null
+#! /bin/bash
+
+if [ -f /opt/movidius/uninstall-ncsdk.sh ]; then
+ echo "running uninstall-ncsdk.sh"
+ /opt/movidius/uninstall-ncsdk.sh
+else
+ echo "could not find uninstall-ncsdk.sh - nothing to uninstall"
+fi
+
+sudo rm -rf /opt/movidius/ncsdk
+
projects / platform / adaptation / npu / intel-libmvnc.git / commitdiff
