박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Thu, 5 Apr 2018 04:08:09 +0000 (13:08 +0900)]
[Pure ACL NN Runtime] Implement ANeuralNetworksMemory (#424)
This commit implements ANeuralNetworksMemory in experimental pure ACL NN
runtime.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Thu, 5 Apr 2018 02:13:17 +0000 (11:13 +0900)]
Introduce IndexIterator (for feature) (#428)
This commit introduces IndexIterator class and relevent helpers in
nnfw::util::feature namespace to make it easy to iterate over feature
elements.
This commit includes updates on NNAPI Conv1 Unittest to demonstrate how
to use these helpers.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
오형석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Thu, 5 Apr 2018 02:12:00 +0000 (11:12 +0900)]
Print shape info in model parser tool (#449)
* Print shape info in model parser tool
Print shape info in model parser tool to get tensor size
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
* Update README.md
Update README.md to show shape info
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Thu, 5 Apr 2018 02:02:43 +0000 (11:02 +0900)]
Support update in TfLiteFeatureView (#425)
This commit introduces a (in-place) update method into TfLiteFeatureView
class.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Thu, 5 Apr 2018 02:01:17 +0000 (11:01 +0900)]
Introduce 'nnfw::util::fp32' namespace (#445)
This commit introduces 'nnfw::util::fp32' namespace which includes
various helps for handling floating-pointer numbers.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Thu, 5 Apr 2018 01:53:39 +0000 (10:53 +0900)]
Introduce 'nnfw::util::tensor' namespace (#440)
This commit introduces 'nnfw::util::tensor' namespace which will include
various helpers for tensor (ND-array) object.
Currently, 'nnfw::util::tensor' namespace includes the following
classes and funtiont:
- class Index
- class Shape
- class IndexIterator
- method iterate
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
이상규/동작제어Lab(SR)/Principal Engineer/삼성전자 [Wed, 4 Apr 2018 13:46:39 +0000 (22:46 +0900)]
Bring android NN runtime tests (#438)
This PR introduces TestMain.cpp, TestTrivialModel and TestValidation
from Android NN tests.
Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
김수진/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 4 Apr 2018 13:44:12 +0000 (22:44 +0900)]
Enable ModelBuilder::sortIntoRunOrder() (#435)
This commit enables ModelBuilder::sortIntoRunOrder() which is missed in implementation.
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
김수진/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 4 Apr 2018 13:42:57 +0000 (22:42 +0900)]
Comment out the unnecessary assertion in CONV2D Op. (#430)
This commit comments out the unnecessary assertion in CONV2D Op.
Related #403
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Wed, 4 Apr 2018 12:50:57 +0000 (21:50 +0900)]
Introduce 'nnfw::util::vector' namespace (#421)
This commit introduces 'nnfw::util::vector' namespace which includes
various helpers for vector objects.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
장지섭/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 4 Apr 2018 08:42:00 +0000 (17:42 +0900)]
Fix wrong return value (#436)
This commit fixes wrong return value.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
장지섭/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 4 Apr 2018 08:29:23 +0000 (17:29 +0900)]
Remove the dependency on hal and hidl from interfaces related to the memory. (#399)
This commit removes the dependency on hal and hidl from interfaces related to the memory.
- IMemory@1.0
- IAllocator@1.0
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
김용섭/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 4 Apr 2018 08:15:23 +0000 (17:15 +0900)]
Change tools/nnapi_test to install (#429)
This commits enables tools/nnapi_test moved to Product/out/bin on
installed.
Signed-off-by: Yongseop Kim <yons.kim@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 4 Apr 2018 08:06:20 +0000 (17:06 +0900)]
Add md5hash check for test files (#418)
To verify cached files are good
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Wed, 4 Apr 2018 07:15:01 +0000 (16:15 +0900)]
Introduce 'nnfw::util::kernel' namespace (#414)
This commit introduces 'nnfw::util::kernel' namespace which holds
various helpers for handling convolution kernels.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Wed, 4 Apr 2018 06:52:51 +0000 (15:52 +0900)]
[nnapi_test] Show details on test failure (#407)
This commit allows users to check details on test failure via
environment variable (VERBOSE).
Note that this feature is turned off by default.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Wed, 4 Apr 2018 06:52:03 +0000 (15:52 +0900)]
[NNAPI unittest] Show configurations before run (#417)
This commit revises 'nnapi_unittest_conv_1' to show configurations
before run (to make it easy to analyze issues).
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Wed, 4 Apr 2018 06:42:02 +0000 (15:42 +0900)]
Introduce TfLiteFeatureView (#412)
This commit implements TfLiteFeatureView in NN API Unittest (conv_1).
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Wed, 4 Apr 2018 06:41:39 +0000 (15:41 +0900)]
[NNAPI Unittest] Fix environment variable setup (#411)
This commit fix IntVar class to set an internal variable correctly from
environments.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
오형석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Wed, 4 Apr 2018 04:19:12 +0000 (13:19 +0900)]
Fix bug in run_test.sh (#415)
Fix bug when finding test directory
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
박세희/동작제어Lab(SR)/Principal Engineer/삼성전자 [Wed, 4 Apr 2018 04:01:02 +0000 (13:01 +0900)]
Add ActivationLayer to kernel test lab (#409)
This will add ActivationLayer with Relu to convolution kernel test lab
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Wed, 4 Apr 2018 02:03:27 +0000 (11:03 +0900)]
[nnapi_test] Fix float equality condition (#408)
0.0f is not equal to 0.0f in the current equality condition.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
장지섭/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 4 Apr 2018 01:35:49 +0000 (10:35 +0900)]
Remove the dependency on hal and hidl from interfaces of neuralnetworks (#397)
This commit removes the dependency on hal and hidl from interfaces of neuralnetworks.
- version 1.0 : IDevice, IPreparedModel, IExecutionCallcak, IPreparedModelCallback and types
- version 1.1 : IDevice and types
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
최형규/동작제어Lab(SR)/Senior Engineer/삼성전자 [Wed, 4 Apr 2018 01:35:24 +0000 (10:35 +0900)]
Enable remaining helpers in OperationsUtils (#396)
- Enable remaining helpers using activation type in OperationsUtils
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
이춘석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Wed, 4 Apr 2018 00:34:13 +0000 (09:34 +0900)]
Enable googletest for unittesting (#398)
0. Support arm-cross, x64-native, arm-native, and tizen-obs build
1. Introduce googletest at top level
2. Add simple test tflite_test.cc : this should be removed
3. Add howto doc
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 3 Apr 2018 23:26:08 +0000 (08:26 +0900)]
Setup NN API without NN Runtime project (#390)
This commit adds a room for NN API implementation without NN Runtime
(for experiments).
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
최형규/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 3 Apr 2018 23:25:01 +0000 (08:25 +0900)]
Add git to docker image for nnfw (#405)
- Add git package to docker image for nnfw
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
김용섭/동작제어Lab(SR)/Engineer/삼성전자 [Tue, 3 Apr 2018 23:18:24 +0000 (08:18 +0900)]
Add USE_LOGNN option for make check (#400)
Enables having choice for LD_LIBRARY. The default LD_LIBRARY is out/lib
while obj/tools/nnapi_bindings/bindings/logging on USE_LOGNN=1. This is
supported for debugging.
Signed-off-by: Yongseop Kim <yons.kim@samsung.com>
최형규/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 3 Apr 2018 13:53:57 +0000 (22:53 +0900)]
Use docker image of nnfw repo in CI (#401)
- Use docker image of nnfw repo in CI, i.e. nnfw_docker
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 3 Apr 2018 13:51:38 +0000 (22:51 +0900)]
Introduce nnfw::util::feature namespace (#393)
* Introduce nnfw::util::feature namespace
This commit introduces nnfw::util::feature namespace which will contain
various helpers for 3D tensor object.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
* Fix a typo(utils -> util)
* Implement feature::Shape's consturctor in its header
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 3 Apr 2018 13:50:43 +0000 (22:50 +0900)]
Introduce nnfw::util::env::IntAccessor (#392)
This commit introduces nnfw::util::env::IntAccessor which makes it easy
to access environmets (for configuration).
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 3 Apr 2018 13:50:29 +0000 (22:50 +0900)]
Introduce util::benchmark (#391)
This commit extracts benchmark-related helpers (implemented inside
tflite_benchmark) as a shared library.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 3 Apr 2018 13:49:51 +0000 (22:49 +0900)]
Use float-compare on NN API testing (#389)
* Use float-compare on NN API testing
This commit revises nnapi_test to use float-compare on NN API
implementation testing.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
* Use std::isnan instead of isnan
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 3 Apr 2018 13:49:33 +0000 (22:49 +0900)]
Introduce nnapi_template (#388)
This commit introduces nnapi_template which provides a dummy NN API
implementation in order to make it easy to introduce an experimental NN
API implementation.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 3 Apr 2018 13:49:23 +0000 (22:49 +0900)]
Support STRIDE configuration (#386)
This commit revises nnapi_unittest_conv1 to support STRIDE
configuration.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 3 Apr 2018 13:49:11 +0000 (22:49 +0900)]
Manage bias_data lifetime correclty (#385)
T/F Lite Interpreter assumes that buffer passed via
SetTensorParametersReadOnly method lives longer than the interpreter
itself.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
김정현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 3 Apr 2018 13:48:46 +0000 (22:48 +0900)]
Introduce simple .tflite model generator for CONV2D (#381)
This commit introduces a simple .tflite model generator.
This first version only supports CONV2D for now.
Usage:
`$ modelgen.py` generates model.json
`$ modelgen.sh model.json` generates model.tflite
You can modify parameters `kernel_shape` and `input_shape` in modelgen.py
to generate different sizes of the kernel and input.
You can execute modelgen.py and modelgen.sh in any directory because
they find where they located at.
Signed-off-by: Junghyun Kim <jh0822.kim@samsung.com>
이춘석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 3 Apr 2018 07:46:58 +0000 (16:46 +0900)]
build test-suite for test on target (#375)
* build test-suite for test on target
1. test-suite.tar.gz contains tests and Product/out
2. cross-build on CI generates test-suite
최형규/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 3 Apr 2018 06:21:57 +0000 (15:21 +0900)]
Fix SIGFPE of CpuExecutor (#395)
Enable following helpers
- getScalarData(const RunTimeOperandInfo& info)
- IsNullInput(const RunTimeOperandInfo *input) {
This resiolves SIGFPE reported at #370
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
서상민/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 3 Apr 2018 05:47:34 +0000 (14:47 +0900)]
Rename nnapi_unitest_common to nnapi_unittest_common (#384)
This patch adds a missing 't' to the lib name.
Signed-off-by: Sangmin Seo <sangmin7.seo@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 3 Apr 2018 04:44:21 +0000 (13:44 +0900)]
Export nnfw_util's include path (#387)
This commit exports nnfw_util's include path as PUBLIC to make it
easy to use nnfw_util.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
최형규/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 3 Apr 2018 03:56:05 +0000 (12:56 +0900)]
Add Tizen rootfs for arm cross compilation (#378)
- Add Tizen rootfs for arm cross compilation
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 3 Apr 2018 02:18:36 +0000 (11:18 +0900)]
Fix build break on Ubuntu 14.04 (#383)
This commit fixes build break on Ubuntu 14.04.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
장지섭/동작제어Lab(SR)/Engineer/삼성전자 [Tue, 3 Apr 2018 02:13:06 +0000 (11:13 +0900)]
add the hal interfaces of neuralnetworks@1.1 to android_nn of labs (#371)
This commit adds the hal interfaces of the neuralnetworks@1.1 to android_nn of labs.
The interfaces of neuralnetworks have been included in o-mr1-iot-preview-7 android-o-mr1-iot-preview-7 android-p-preview-1.
Ref : https://android.googlesource.com/platform/hardware/interfaces/+/
7df64ea2fd8c8e501ac1ba5cb05def96f8a8d49e
Additionally, This is the history log) related to android nn of o-mr1-iot-preview-7 branch.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Mon, 2 Apr 2018 23:37:13 +0000 (08:37 +0900)]
Introduce add_nnapi_unittest function (#380)
This commit introduces add_nnapi_unittest function in 'CMakeLists.txt'
to make it easy to add new unittests.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
김용섭/동작제어Lab(SR)/Engineer/삼성전자 [Mon, 2 Apr 2018 23:14:43 +0000 (08:14 +0900)]
Fix wrong setting libneuralnetworks.so (#372)
Set right libneuralnetworks.so. To do this, revise some options such as
OPTIONS_LD_LIB_NNAPI and USE_NNAPI.
Signed-off-by: Yongseop Kim <yons.kim@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Mon, 2 Apr 2018 23:11:58 +0000 (08:11 +0900)]
Introduce NNAPI Unittest through T/F Lite (#374)
This commit introduces a convolution NNAPI unittest which
invokes NN API via T/F Lite interpreter and compare it with
the output of T/F Lite pure interpreter.
This testcase is derived from a convolution example in tflite_examples.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박세희/동작제어Lab(SR)/Principal Engineer/삼성전자 [Mon, 2 Apr 2018 23:03:38 +0000 (08:03 +0900)]
Compare with CPU and ACL result in kernel conv test (#368)
This will enable convolution test to get expected from CPU and result from ACL and compare the two.
Some changes was done not directly related with tis change but to fix compile and linking.
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
이춘석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Mon, 2 Apr 2018 08:59:22 +0000 (17:59 +0900)]
install libneuralnetworks.so into lib (#369)
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
김용섭/동작제어Lab(SR)/Engineer/삼성전자 [Mon, 2 Apr 2018 08:16:47 +0000 (17:16 +0900)]
Add make check for run_test.sh (#361)
* Add make check for run_test.sh
This make check executes run_test.sh to test scripts in test/framework/
Signed-off-by: Yongseop Kim <yons.kim@samsung.com>
* Apply other comments
- Separate `make install` & `make check`
- Only tests for x86 (NOT cross-build for arm)
- Add LD_LIBRARY_PATH on USE_NNAPI=1
박세희/동작제어Lab(SR)/Principal Engineer/삼성전자 [Mon, 2 Apr 2018 04:17:24 +0000 (13:17 +0900)]
Add compareData for kernel test lab (#362)
This will add compareData that compares result and expected of convolution.
Success will do nothing and show error with exit -1 when fails.
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
이상규/동작제어Lab(SR)/Principal Engineer/삼성전자 [Mon, 2 Apr 2018 02:53:40 +0000 (11:53 +0900)]
Get rid of cmake find_package(Threads) (#357)
We cannot use find_package(Threads) since some of our build
system use old version cmake. pthread is linked directly now.
So I removed find_package(Threads) and related statements.
Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
장지섭/동작제어Lab(SR)/Engineer/삼성전자 [Mon, 2 Apr 2018 02:53:17 +0000 (11:53 +0900)]
Enable setInputFromMemory and setOutputFromMemory from Execution (#360)
This commit enables setInputFromMemory and setOutputFromMemory from Execution.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
장지섭/동작제어Lab(SR)/Engineer/삼성전자 [Mon, 2 Apr 2018 02:53:03 +0000 (11:53 +0900)]
Modify output log in simple_model (#359)
This commit modifies output log in simple_model.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
오형석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Mon, 2 Apr 2018 02:10:03 +0000 (11:10 +0900)]
Change test script construction (#354)
* Change test script construction
- Make directory for each test
- Make config.sh file for each test setting
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
박세희/동작제어Lab(SR)/Principal Engineer/삼성전자 [Mon, 2 Apr 2018 02:09:13 +0000 (11:09 +0900)]
Labs - kernel unit test - add convFloat32 method (#352)
* Labs - kernel unit test - add convFloat32 method
This will add convFloat32 method from NN runtime to prepare binding ACL convolution
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* remove empty line
* add space for readability
이춘석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Mon, 2 Apr 2018 02:01:11 +0000 (11:01 +0900)]
docker arm cross build with no acl/rootfs build (#342)
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
박세희/동작제어Lab(SR)/Principal Engineer/삼성전자 [Mon, 2 Apr 2018 01:54:10 +0000 (10:54 +0900)]
Separate Cross build for arm document (#358)
This will separate cross build for arm section in main readme into docs folder file
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/동작제어Lab(SR)/Principal Engineer/삼성전자 [Mon, 2 Apr 2018 01:50:42 +0000 (10:50 +0900)]
Add manual conv with excel (#356)
This will add and simple 3x3 convolution in excel file.
It uses excel built-in SUMPRODUCT() method.
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
장지섭/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 30 Mar 2018 08:32:40 +0000 (17:32 +0900)]
Fix a failure to set RunTimePoolInfo (#355)
* Fix creating ashmem fail
This commit fixes creating ashmem fail.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
* Fix the fail to set RunTimePoolInfo
This commit fixes the fail to set RunTimePoolInfo.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
장지섭/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 30 Mar 2018 08:31:29 +0000 (17:31 +0900)]
Fix creating ashmem fail (#350)
This commit fixes creating ashmem fail.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
오형석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Fri, 30 Mar 2018 08:11:06 +0000 (17:11 +0900)]
Make readme file for tflite file tool (#353)
Make reaeme file for
- model parser
- model generator
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Fri, 30 Mar 2018 06:52:48 +0000 (15:52 +0900)]
T/F Lite Conv Example: Use non-squared IFM & Kernel (#337)
This commit revises T/F Lite Conv Example to use non-squred IFM and
Kernel to make it easy to distinguish dimensions.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
김정현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Fri, 30 Mar 2018 06:52:41 +0000 (15:52 +0900)]
Introduce weights extractor from .tflite (#288)
This commit introduces weight/bias extractor from .tflite file.
You can execute this script like below:
`$ extract_from_tflite.sh inception.tflite`
Then, you can find *.npy files in the current working directory.
Those *.npy files contain weight or bias values.
The name of each file is set to be the name of each layer.
Signed-off-by: Junghyun Kim <jh0822.kim@samsung.com>
김수진/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 30 Mar 2018 03:41:21 +0000 (12:41 +0900)]
Modify add sample (#349)
This commit modifies add sample to fix the bug and show printing results.
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
최형규/동작제어Lab(SR)/Senior Engineer/삼성전자 [Fri, 30 Mar 2018 02:24:03 +0000 (11:24 +0900)]
Update TODO of NeuralNetworks.cpp (#344)
- Update TODO comment to use REF-ANN and TODO-NNRT
- Implment some TODO-NNRT
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
최형규/동작제어Lab(SR)/Senior Engineer/삼성전자 [Fri, 30 Mar 2018 02:23:56 +0000 (11:23 +0900)]
Enable MUL of CpuExecutor (#345)
- Enable MUL of CpuExecutor
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
장지섭/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 30 Mar 2018 02:07:51 +0000 (11:07 +0900)]
Enable the methods of `RunTimePoolInfo` to prepare to use memory (#343)
This commit enables the methods of `RunTimePoolInfo` to prepare to use memory.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
최형규/동작제어Lab(SR)/Senior Engineer/삼성전자 [Fri, 30 Mar 2018 01:45:54 +0000 (10:45 +0900)]
Enable StepExecutor::startComputeOnCpu (#341)
- Enable StepExecutor::startComputeOnCpu
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
이춘석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Fri, 30 Mar 2018 01:31:51 +0000 (10:31 +0900)]
Skip pthread TRY_RUN when cross build (#340)
최형규/동작제어Lab(SR)/Senior Engineer/삼성전자 [Fri, 30 Mar 2018 01:26:11 +0000 (10:26 +0900)]
Enable Conv2D and FullyConnected operations (#333)
- Introduce Conv2D and FullyConnected operations
- Enable OperationType::CONV_2D and OperationType::FULLY_CONNECTED
of CpuExecutor
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
박세희/동작제어Lab(SR)/Principal Engineer/삼성전자 [Fri, 30 Mar 2018 01:26:02 +0000 (10:26 +0900)]
Fix NN_RUNTIME build option (#336)
This will move BUILD_NN_RUNTIME option into src of cmake as src has separate util project
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
이춘석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Fri, 30 Mar 2018 01:08:00 +0000 (10:08 +0900)]
Fix link fail for pthread on obs (#338)
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
박세희/동작제어Lab(SR)/Principal Engineer/삼성전자 [Fri, 30 Mar 2018 00:40:32 +0000 (09:40 +0900)]
Rename experiments to labs (#296)
As external and experiments both start with ex, it makes more typing.
This will change experiments folder name to labs to work more efficiently.
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
Hyung-Kyu Choi [Fri, 30 Mar 2018 00:01:10 +0000 (09:01 +0900)]
Fix name of target in nn/runtime/CMakeLists.txt
- Fix name of target in nn/runtime/CMakeLists.txt
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
Hyung-Kyu Choi [Thu, 29 Mar 2018 15:00:53 +0000 (00:00 +0900)]
Enable three operation of CpuExecutor
- Introduce Concatenation operation
- Enable OperationType::CONCATENATION, OperationType::SOFTMAX
and OperationType::MAX_POOL_2D of CpuExecutor
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
이춘석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Thu, 29 Mar 2018 23:55:45 +0000 (08:55 +0900)]
Add target external_acl for docker build (#325)
To skip acl build on docker env, `make external_acl` will copy shared libs
to lib folder.
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
Sanggyu Lee [Thu, 29 Mar 2018 10:34:55 +0000 (19:34 +0900)]
Fix compilation errors in StepExecutor::startComputeOnCpu
- Link pthread for NNRuntime test executables
- Add using statement for Status
- IExecutionCallback to ExecutionCallback
Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
Hyung-Kyu Choi [Thu, 29 Mar 2018 08:52:26 +0000 (17:52 +0900)]
Enable Pooling of CpuExecutor
- Enable Poolinig of CpuExecutor
- Enable OperationType::AVERAGE_POOL_2D of CpuExecutor
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
jiseob.jang [Thu, 29 Mar 2018 10:38:05 +0000 (19:38 +0900)]
Copy `Status.cpp` from libhidl of Android
This commit copies `Status.cpp` from libhidl of Android.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
Jonghyun Park [Thu, 29 Mar 2018 09:17:54 +0000 (18:17 +0900)]
T/F Lite Conv Example: Use NN API (if possible)
This commit revises T/F Lite Conv Example to use NN API (if possible).
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Thu, 29 Mar 2018 09:40:31 +0000 (18:40 +0900)]
Do not build ACL static libraries (#306)
* Do not build ACL static libraries
This commit updates acl.cmake not to build static libraries in order to
reduce build time.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
이춘석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Thu, 29 Mar 2018 08:58:01 +0000 (17:58 +0900)]
Add arm ubuntu cross build script (#312)
* Add arm ubuntu cross build script
build step
1. prepare rootfs for arm ubuntu
2. run docker/docker_build_cross_arm_ubuntu.sh
TODO
Cache ACL build artifacts
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
* Add install target
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
* apply comment and extract long command string
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Thu, 29 Mar 2018 08:43:50 +0000 (17:43 +0900)]
Introduce 'tflite_conv_example' (#317)
This commit introduce 'tflite_conv_example' which creates and runs a simple
NN model with one convolution with T/F Lite interpreter.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
Hyung-Kyu Choi [Thu, 29 Mar 2018 08:07:50 +0000 (17:07 +0900)]
Enable SimpleMath of CpuExecutor
- Enable SimpleMath of CpuExecutor
- Enable OperationType::ADD of CpuExecutor
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
sjsujinkim [Thu, 29 Mar 2018 08:14:22 +0000 (17:14 +0900)]
Change comment
sjsujinkim [Thu, 29 Mar 2018 08:06:48 +0000 (17:06 +0900)]
Add TODO comment into ExecutionBuilder::startCompute
This commit adds TODO comment into ExecutionBuilder::startCompute for future plan.
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
sjsujinkim [Thu, 29 Mar 2018 07:39:01 +0000 (16:39 +0900)]
Remove dummy return in Execution::create()
This commit removes dummy return which causes 'NeuralNetworksExecution_setInput passed a nullptr' in Execution::create().
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
jiseob.jang [Thu, 29 Mar 2018 07:23:12 +0000 (16:23 +0900)]
Implement Memory::create()
This commit implements Memory::create().
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
Hyung-Kyu Choi [Thu, 29 Mar 2018 07:33:12 +0000 (16:33 +0900)]
Introduce CPU operation of NN runtime
- Introduce CPU operation of NN runtime
- Also introduce eigen and gemmlowp in depend/external
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
sjsujinkim [Thu, 29 Mar 2018 07:02:04 +0000 (16:02 +0900)]
Remove dummy return in CompilationBuilder::createExecution
This commit removes dummy return which causes 'NeuralNetworksExecution_setInput passed a nullptr' in CompilationBuilder::createExecution.
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
sjsujinkim [Thu, 29 Mar 2018 06:46:19 +0000 (15:46 +0900)]
Add std::endl before logging by LOG
Currently, we can see the log logging by LOG only in one line. It is very hard to read.
This commit adds std::endl before logging by LOG helper for easy to read.
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
sjsujinkim [Thu, 29 Mar 2018 04:49:50 +0000 (13:49 +0900)]
Implement Execution_setOutput
This commit implements Execution_setOutput.
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
박세희/동작제어Lab(SR)/Principal Engineer/삼성전자 [Thu, 29 Mar 2018 07:19:23 +0000 (16:19 +0900)]
Add empty template code for kernel testing (#311)
This will add initial empty code for kernel testing programs
Main purpose of this change is to add add_kerneltesting method in cmake
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
SaeHie Park [Thu, 29 Mar 2018 06:35:11 +0000 (15:35 +0900)]
Fix acl expierments cmake
This will fix cmake of acl experiment to use common variables defined in root cmake
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
오형석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Thu, 29 Mar 2018 06:49:28 +0000 (15:49 +0900)]
Tflite file operator selection tool (#310)
User select operators in input tflite model, and write these operators' number in text file.
This tool read input tflite model and text file,
then generate tflite model including selected operator only
This version can select operations in inception v3 nonslim
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
Hyung-Kyu Choi [Thu, 29 Mar 2018 05:12:15 +0000 (14:12 +0900)]
Introduce OperationsUtils
- Introduce OperationsUtils
- Make use of types introduced from OperationUtils in CpuExecutor
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
sjsujinkim [Wed, 28 Mar 2018 10:52:23 +0000 (19:52 +0900)]
Implement Execution_startCompute for only CPU
This commit implements Execution_startCompute for only CPU.
- Add StepExecutor(const ExecutionBuilder* executionBuilder, const ModelBuilder* model)
- enable StepExecutor::mapInputsAndOutputsTrivially
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
sjsujinkim [Thu, 29 Mar 2018 04:37:39 +0000 (13:37 +0900)]
Implement Execution_setInput
This commit implements Execution_setInput.
- Enables
- ModelArgumentInfo::setFromPointer
- ModelArgumentInfo::updateDimensionInfo
- DataLocation
박세희/동작제어Lab(SR)/Principal Engineer/삼성전자 [Thu, 29 Mar 2018 05:57:17 +0000 (14:57 +0900)]
Introduce cmake vars for acl (#304)
This will introduce NNFW_ACL_INCLUDES and NNFW_ALC_LIBS to cmake
Signed-off-by: SaeHie Park <saehie.park@samsung.com>