platform/core/ml/nnfw.git
4 years agoApply optimized cpu kernel of MaxPoolFloat32 (#7886)
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 2 Oct 2019 05:35:42 +0000 (14:35 +0900)]
Apply optimized cpu kernel of MaxPoolFloat32 (#7886)

This commit applies optimized cpu kernel of MaxPoolFloat32.
  - Introduce optimized cpu kernel of MaxPool for float
  - Apply that kernel for cpu backend of neurun

Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
4 years ago[nnfw-api] rename api related files (#7895)
이상규/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 2 Oct 2019 05:01:52 +0000 (14:01 +0900)]
[nnfw-api] rename api related files (#7895)

- nnfw_dev.cc is renamed to nnfw_api.cc.
  - nnfw_dev.cc looks like the implementation for `nnfw_dev.h`. (i.e. APIs for developers like custom op)
  - However, it is a binding from C APIs to C++ implementation for WHOLE APIs.
- nnfw_api.cc and nnfw_api.hpp are C++ implementation.
  - nnfw_api.cc is already used for a binding implementation the above.
  - Thus, they are renamed to nnfw_api_internal.*.
- nnfw_api.hpp is renamed to nnfw_api_internal.h.
  - It was the only file that uses the suffix `*.hpp`.
- wrapper directory is removed to make structure simple.

Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
4 years ago[android] Disable STRICT_BUILD (#7900)
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 2 Oct 2019 04:29:35 +0000 (13:29 +0900)]
[android] Disable STRICT_BUILD (#7900)

Disable STRICT_BUILD to suppress warnings from ARMComputeEx.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
4 years ago[moco] Introduce Conv2D series IR (#7891)
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 2 Oct 2019 03:38:18 +0000 (12:38 +0900)]
[moco] Introduce Conv2D series IR (#7891)

This will introduce IR for Conv2D, Conv2DBackpropInput and DepthwiseConv2dNative node

Signed-off-by: SaeHie Park <saehie.park@samsung.com>
4 years ago[moco] Introduce IR for Add AvgPool BiasAdd (#7890)
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 2 Oct 2019 03:38:00 +0000 (12:38 +0900)]
[moco] Introduce IR for Add AvgPool BiasAdd (#7890)

This will introduce IR for Add, AvgPool and BiasAdd node

Signed-off-by: SaeHie Park <saehie.park@samsung.com>
4 years ago[android] Disable building unsupported libraries (#7898)
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 2 Oct 2019 03:28:05 +0000 (12:28 +0900)]
[android] Disable building unsupported libraries (#7898)

Disable building libraries that are not supported due to dependency on
Boost library.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
4 years agoSegregate interface of tflite_loader (#7668)
Сергей Баранников/AI Tools Lab /SRR/Engineer/삼성전자 [Wed, 2 Oct 2019 02:29:02 +0000 (05:29 +0300)]
Segregate interface of tflite_loader (#7668)

* Segregate interface of tflite_loader

Split `tflite_loader` component into two parts -- interface and implementation.
Put the loader into `neurun` namespace.
Remove a couple of unused functions that triggered compiler warnings.

Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
4 years ago[moco] Introduce TFNodeVisitor (#7887)
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 2 Oct 2019 01:01:53 +0000 (10:01 +0900)]
[moco] Introduce TFNodeVisitor (#7887)

This will introduce TFNodeVisitor and revise TFNode to support visitor pattern

Signed-off-by: SaeHie Park <saehie.park@samsung.com>
4 years ago[exo-tflite] add TensorTransposeConverter into the list of pass (#7877)
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 1 Oct 2019 22:06:58 +0000 (07:06 +0900)]
[exo-tflite] add TensorTransposeConverter into the list of pass (#7877)

This adds TensorTransposeConverter into the list of pass.
Please turns on the knob to convert TensorTranspose to TFLTransport.

Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
4 years ago[mir] Fix TensorVariant constructor used in broadcasting (#7889)
Павел Ильютченко/AI Tools Lab /SRR/Engineer/삼성전자 [Tue, 1 Oct 2019 17:04:12 +0000 (20:04 +0300)]
[mir] Fix TensorVariant constructor used in broadcasting (#7889)

* Fix TensorType creation with old shape

Signed-off-by: Pavel Iliutchenko <p.iliutchenk@samsung.com>
4 years agoChange all calls to MIR importers in [nnc] (#7879)
Гусев Дмитрий Викторович/AI Tools Lab /SRR/Engineer/삼성전자 [Tue, 1 Oct 2019 12:15:42 +0000 (15:15 +0300)]
Change all calls to MIR importers in [nnc] (#7879)

According with interface changes, we now use loadModel(...) functions.

Signed-off-by: Dmitry Gusev <d.gusev@partner.samsung.com>
4 years ago[moco-tf] ShapeInf for Add AvgPool BiasAdd (#7885)
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 1 Oct 2019 10:36:13 +0000 (19:36 +0900)]
[moco-tf] ShapeInf for Add AvgPool BiasAdd (#7885)

This will enable shape inference for Add AvgPool and BiasAdd node

Signed-off-by: SaeHie Park <saehie.park@samsung.com>
4 years ago[moco-tf] Use TFShapeInferenceHelper fox FixShape (#7884)
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 1 Oct 2019 10:35:54 +0000 (19:35 +0900)]
[moco-tf] Use TFShapeInferenceHelper fox FixShape (#7884)

This will make FixShapeTransform to use TFShapeInferenceHelper methods

Signed-off-by: SaeHie Park <saehie.park@samsung.com>
4 years ago[nnfw-cmake] Remove runtime variable (#7881)
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 1 Oct 2019 10:26:37 +0000 (19:26 +0900)]
[nnfw-cmake] Remove runtime variable (#7881)

Remove unused runtime name variable
Move set LIB_NEURUN variable

Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
4 years ago[moco-tf] Refactor Conv2DCanonicalizer (#7874)
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 1 Oct 2019 09:47:32 +0000 (18:47 +0900)]
[moco-tf] Refactor Conv2DCanonicalizer (#7874)

This will update Conv2DCanonicalizer to do shape, pad inference and NOT to use annotations

Signed-off-by: SaeHie Park <saehie.park@samsung.com>
4 years ago[moco-tf] Refactor AvgPoolCanonicalizer (#7872)
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 1 Oct 2019 09:47:11 +0000 (18:47 +0900)]
[moco-tf] Refactor AvgPoolCanonicalizer (#7872)

This will update AvgPoolCanonicalizer to do shape, pad inference and NOT to use annotations

Signed-off-by: SaeHie Park <saehie.park@samsung.com>
4 years ago[moco] Introduce TFNode and related (#7870)
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 1 Oct 2019 09:42:21 +0000 (18:42 +0900)]
[moco] Introduce TFNode and related (#7870)

This will introduce TFNode base class and related codes

Signed-off-by: SaeHie Park <saehie.park@samsung.com>
4 years ago[moco] Introduce requires.cmake (#7880)
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 1 Oct 2019 09:41:49 +0000 (18:41 +0900)]
[moco] Introduce requires.cmake (#7880)

This will introduce moco requires.cmake file

Signed-off-by: SaeHie Park <saehie.park@samsung.com>
4 years ago[nnfw-cmake] Replace fp16 source package config (#7840)
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 1 Oct 2019 09:24:52 +0000 (18:24 +0900)]
[nnfw-cmake] Replace fp16 source package config (#7840)

Move fp16 source package config to common directory

Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
4 years ago[neurun] move api out of frontend (#7873)
이상규/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 1 Oct 2019 09:20:37 +0000 (18:20 +0900)]
[neurun] move api out of frontend (#7873)

- api is now under `neurun`, instead of `neurun/frontend`.
- add dependency for nnfw-header in custom op kernels and apps.

Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
4 years ago[moco-tf] broadcast_shape to helper (#7844)
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 1 Oct 2019 09:12:25 +0000 (18:12 +0900)]
[moco-tf] broadcast_shape to helper (#7844)

* [moco-tf] broadcast_shape to helper

This will copy broadcast_shape method and related codes to helper file

Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* add todo comment

4 years ago[moco-tf] Introduce ShapeInferenceAlgorithm (#7871)
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 1 Oct 2019 08:18:49 +0000 (17:18 +0900)]
[moco-tf] Introduce ShapeInferenceAlgorithm (#7871)

This will introduce ShapeInferenceAlgorithm to provide shape inferece by type of TFNode

Signed-off-by: SaeHie Park <saehie.park@samsung.com>
4 years agoApply optimized cpu kernel for AvgPoolFloat32 (#7834)
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 1 Oct 2019 07:35:56 +0000 (16:35 +0900)]
Apply optimized cpu kernel for AvgPoolFloat32 (#7834)

This commit applies optimized cpu kernel for AvgPoolFloat32.
  - Introduce introduce optimized cpu kernel of AvgPool2D op
  - Apply that kernel for AvgPoolFloat32

Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
4 years ago[res] Summary for NET_0025 (#7867)
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 1 Oct 2019 07:14:06 +0000 (16:14 +0900)]
[res] Summary for NET_0025 (#7867)

This commit adds summary for NET_0025, CBR network test case.

Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
4 years ago[tf2tflite] Test NET_0025 (#7868)
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 1 Oct 2019 07:13:34 +0000 (16:13 +0900)]
[tf2tflite] Test NET_0025 (#7868)

This commit registers NET_0025 unit test, CBR network, to tf2tflite

Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
4 years ago[moco-tf] PlaneShape for shape inference (#7866)
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 1 Oct 2019 07:10:42 +0000 (16:10 +0900)]
[moco-tf] PlaneShape for shape inference (#7866)

This will introduce PlaneShape, PlaneInference and related codes for plane shape inference for Pool and Conv2D series nodes

Signed-off-by: SaeHie Park <saehie.park@samsung.com>
4 years ago[moco-tf] window_of for Conv2D (#7864)
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 1 Oct 2019 07:10:18 +0000 (16:10 +0900)]
[moco-tf] window_of for Conv2D (#7864)

This will introduce window_of method used for Conv2D type nodes

Signed-off-by: SaeHie Park <saehie.park@samsung.com>
4 years ago[logo] traversing using active_nodes instead of all_nodes (#7865)
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 1 Oct 2019 07:09:59 +0000 (16:09 +0900)]
[logo] traversing using active_nodes instead of all_nodes (#7865)

Using all_nodes makes simplifying orphant subgraph and cause error.

Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
4 years ago[moco] Introduce lang sub project and TFDialect (#7863)
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 1 Oct 2019 07:08:44 +0000 (16:08 +0900)]
[moco] Introduce lang sub project and TFDialect (#7863)

This will introduce moco lang sub project with TFDialect class

Signed-off-by: SaeHie Park <saehie.park@samsung.com>
4 years ago[pacl] Additional NNAPI implementation for update NNAPI (#7860)
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 1 Oct 2019 03:06:11 +0000 (12:06 +0900)]
[pacl] Additional NNAPI implementation for update NNAPI (#7860)

- PACL's NNAPI implementation for updated NNAPI generated test
- Update skiplist for updated validation check

Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
4 years ago[nnfw-cmake] Rename NNFW_OVERLAY_DIR (#7848)
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 1 Oct 2019 02:01:31 +0000 (11:01 +0900)]
[nnfw-cmake] Rename NNFW_OVERLAY_DIR (#7848)

Rename NNFW_OVERLAY_DIR to NNAS_OVERLAY_DIR  to share cmake package config with nnfw

Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
4 years ago[nnfw-cmake] Replace nonius package config (#7839)
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 1 Oct 2019 01:56:38 +0000 (10:56 +0900)]
[nnfw-cmake] Replace nonius package config (#7839)

Move nonius and source package config to common directory

Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
4 years ago[exo-tflite] Make S32 tensor to be exported (#7827)
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 1 Oct 2019 01:45:05 +0000 (10:45 +0900)]
[exo-tflite] Make S32 tensor to be exported (#7827)

* [exo-exporter] Make S32 tensor to be exported

Previously only FLOAT32 tensor could be exported. This commit enables S32 tensor to be exported.

Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* throw exception when not-supported dtype is found

4 years ago[moco] Initial readme (#7813)
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 1 Oct 2019 01:17:07 +0000 (10:17 +0900)]
[moco] Initial readme (#7813)

* [moco] Initial readme

This will introduce readme for moco project and moco-lang sub project

Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* change folder to lang

* update moco

4 years ago[nncc-cmake] Introduce NNAS_OVERLAY_DIR (#7846)
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 1 Oct 2019 00:53:36 +0000 (09:53 +0900)]
[nncc-cmake] Introduce NNAS_OVERLAY_DIR (#7846)

Introduce NNAS_OVERLAY_DIR to share cmake package config with nnfw

Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
4 years ago[exo-tflite] code to export TFLTranspose to tflite file (#7806)
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 30 Sep 2019 23:21:59 +0000 (08:21 +0900)]
[exo-tflite] code to export TFLTranspose to tflite file (#7806)

Code to export TFLTranspose to tflite file was added into OperationExporter.

Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
4 years ago[exo-tflite] shape & type inference for TFLConv2D (#7835)
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 30 Sep 2019 23:21:46 +0000 (08:21 +0900)]
[exo-tflite] shape & type inference for TFLConv2D (#7835)

This adds shape inference and type inference for TFLConv2D.

Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
4 years ago[exo-tfltie] converting loco::Conv2D to locoex::TFLConv2D (#7845)
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 30 Sep 2019 23:21:33 +0000 (08:21 +0900)]
[exo-tfltie] converting loco::Conv2D to locoex::TFLConv2D (#7845)

This adds a Conv2DConverter that converts loco::Conv2D to locoex::TFLConv2D.

Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
4 years ago[moco-tf] Revose ResolveReshapeWildcardDim (#7843)
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 30 Sep 2019 22:21:32 +0000 (07:21 +0900)]
[moco-tf] Revose ResolveReshapeWildcardDim (#7843)

This will revise ResolveReshapeWildcardDim to use NodeShape instead of ShapeInferenceData

Signed-off-by: SaeHie Park <saehie.park@samsung.com>
4 years ago[moco-tf] Revise ResolveConstantShape (#7842)
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 30 Sep 2019 22:21:06 +0000 (07:21 +0900)]
[moco-tf] Revise ResolveConstantShape (#7842)

This will revise ResolveConstantShape to use NodeShape instead of ShapeInferenceData

Signed-off-by: SaeHie Park <saehie.park@samsung.com>
4 years ago[moco-tf] Revise FuseBinaryIntoPreceding (#7841)
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 30 Sep 2019 22:14:55 +0000 (07:14 +0900)]
[moco-tf] Revise FuseBinaryIntoPreceding (#7841)

This will revise FuseBinaryIntoPreceding to use NodeShape instead of ShapeInferenceData

Signed-off-by: SaeHie Park <saehie.park@samsung.com>
4 years ago[mir2loco] Support Broadcast on Elementwise operations (#7665)
Павел Ильютченко/AI Tools Lab /SRR/Engineer/삼성전자 [Mon, 30 Sep 2019 15:34:17 +0000 (18:34 +0300)]
[mir2loco] Support Broadcast on Elementwise operations (#7665)

* Supported new elemetwise operations
* Implemented TensorBroadcast creation
* Fixed all functions for broadcasting

Signed-off-by: Pavel Iliutchenko <p.iliutchenk@samsung.com>
4 years agoExpand [mir-caffe-importer] interface (#7850)
Гусев Дмитрий Викторович/AI Tools Lab /SRR/Engineer/삼성전자 [Mon, 30 Sep 2019 15:10:40 +0000 (18:10 +0300)]
Expand [mir-caffe-importer] interface (#7850)

Added new interface function loadModel in caffe importer

Signed-off-by: Dmitry Gusev <d.gusev@partner.samsung.com>
4 years agoExpand [mir-caffe2-importer] interface (#7855)
Гусев Дмитрий Викторович/AI Tools Lab /SRR/Engineer/삼성전자 [Mon, 30 Sep 2019 15:08:48 +0000 (18:08 +0300)]
Expand [mir-caffe2-importer] interface (#7855)

A new function loadModel was added.

Signed-off-by: Dmitry Gusev <d.gusev@partner.samsung.com>
4 years agoExpand [mir-onnx-importer] interface (#7856)
Гусев Дмитрий Викторович/AI Tools Lab /SRR/Engineer/삼성전자 [Mon, 30 Sep 2019 15:06:40 +0000 (18:06 +0300)]
Expand [mir-onnx-importer] interface (#7856)

A new function loadModel was added.

Signed-off-by: Dmitry Gusev <d.gusev@partner.samsung.com>
4 years agoImporter interface was expanded. (#7857)
Гусев Дмитрий Викторович/AI Tools Lab /SRR/Engineer/삼성전자 [Mon, 30 Sep 2019 15:05:19 +0000 (18:05 +0300)]
Importer interface was expanded. (#7857)

A new function loadModel was added.

Signed-off-by: Dmitry Gusev <d.gusev@partner.samsung.com>
4 years ago[mir] Introduce TensorType (#7854)
Сергей Баранников/AI Tools Lab /SRR/Engineer/삼성전자 [Mon, 30 Sep 2019 14:59:14 +0000 (17:59 +0300)]
[mir] Introduce TensorType (#7854)

* Add `TensorType` class which combines tensor element type and shape.
* Use it in `TensorVariant` and `Operation::Output`.

Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
4 years ago[nnapi] Update NeuralNetworks.Wrapper.h (#7829)
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 30 Sep 2019 11:04:00 +0000 (20:04 +0900)]
[nnapi] Update NeuralNetworks.Wrapper.h (#7829)

Use NeuralNetworks.Wrapper.h in android-10 nn runtime
Fix to use boost::optional instead of std::optional (c++17 feature)

Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
4 years ago[moco-tf] TFShapeInferenceRule V2 (#7825)
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 30 Sep 2019 10:00:24 +0000 (19:00 +0900)]
[moco-tf] TFShapeInferenceRule V2 (#7825)

This will revise TFShapeInferenceRule to support V2 interface

Signed-off-by: SaeHie Park <saehie.park@samsung.com>
4 years ago[moco-tf] Fix build error for TFShapeInferenceHelper path (#7837)
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 30 Sep 2019 09:36:20 +0000 (18:36 +0900)]
[moco-tf] Fix build error for TFShapeInferenceHelper path (#7837)

This will fix build error for recent TFShapeInferenceHelper path change

Signed-off-by: SaeHie Park <saehie.park@samsung.com>
4 years ago[exo-tflite] converting loco::TensorTranspose to locoex::TFLTranspose (#7824)
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 30 Sep 2019 09:07:15 +0000 (18:07 +0900)]
[exo-tflite] converting loco::TensorTranspose to locoex::TFLTranspose (#7824)

* [exo-tflite] converter to convert loco::TensorTranspose to locoex::TFLTranspose

A converter to convert loco::TensorTranspose to locoex::TFLTranspose is introduced.

Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* adding more validation check + using std::is_permutation(..)

4 years ago[moco-tf] Revise TFNodeSummaryBuilder for TFConst (#7826)
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 30 Sep 2019 09:02:35 +0000 (18:02 +0900)]
[moco-tf] Revise TFNodeSummaryBuilder for TFConst (#7826)

This will remove shape information for TFConst as shape is shown in loco

Signed-off-by: SaeHie Park <saehie.park@samsung.com>
4 years ago[moco-tf] Convert TF strides and ksize (#7812)
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 30 Sep 2019 08:59:48 +0000 (17:59 +0900)]
[moco-tf] Convert TF strides and ksize (#7812)

* [moco-tf] Convert TF strides and ksize

This will introduce two methods to convert TensorFlow Strides and KSize vector to loco Stride and Window in TFShapeInferenceHelper

Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* change folder to src

* recover changes

4 years ago[nnfw-cmake] Replace NNPACK and related package config (#7798)
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 30 Sep 2019 08:33:24 +0000 (17:33 +0900)]
[nnfw-cmake] Replace NNPACK and related package config (#7798)

Move NNPACK and related source package config to common directory

Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
4 years ago[lib-tflite] Remove tflite version definition depdency code (#7807)
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 30 Sep 2019 08:23:17 +0000 (17:23 +0900)]
[lib-tflite] Remove tflite version definition depdency code (#7807)

- Remove TFLITE_MAJOR_VER and TFLITE_MINOR_VER setting in nnfw.spec and cmake script
- Remove TFLITE_MAJOR_VER and TFLITE_MINOR_VER dependency code in tflite extension library

Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
4 years ago[exo-tflite] helper funtion to create filter (#7828)
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 30 Sep 2019 08:07:00 +0000 (17:07 +0900)]
[exo-tflite] helper funtion to create filter (#7828)

* [exo-tflite] helper fultion to create filter

This adds helper functions to easily create FilterEncode and FilterDecode of two filter layout.

Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* remove blank

4 years ago[exo-tflite] Unit test getOpPadding (#7802)
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 30 Sep 2019 08:06:11 +0000 (17:06 +0900)]
[exo-tflite] Unit test getOpPadding (#7802)

This commit adds unit test of getOpPadding() for valid, same
and custom padding cases.

Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
4 years agoMake to support Gather op for acl neon (#7746)
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 30 Sep 2019 06:58:48 +0000 (15:58 +0900)]
Make to support Gather op for acl neon (#7746)

This commit makes to support Gather op for acl neon.
  - Introduce NEGatherEx and NEGatherKernelEx
  - Apply NEGatherEx layer for neurun

Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
4 years ago[cmd] Move doxygen command (#7821)
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 30 Sep 2019 05:20:35 +0000 (14:20 +0900)]
[cmd] Move doxygen command (#7821)

Move doxygen command from nnfw to nnas
Use user-setting workspace for docygen output
Add document for help

Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
4 years ago[exo-tflite] Fix getOpPadding for SAME padding case (#7787)
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 30 Sep 2019 04:18:15 +0000 (13:18 +0900)]
[exo-tflite] Fix getOpPadding for SAME padding case (#7787)

* [exo-tflite] Fix getOpPadding for SAME padding case

Previously getOpPadding() checks only necessary condition for SAME
padding case. This commit checks fully equivalent logic for this.

Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
* Use explicit cast

4 years ago[moco-tf] Remove ShapeInferenceData in Canonicalizers (#7810)
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 30 Sep 2019 03:09:33 +0000 (12:09 +0900)]
[moco-tf] Remove ShapeInferenceData in Canonicalizers (#7810)

This will revise Canonicalizers not to use ShapeInferenceData and use node_shape instead

Signed-off-by: SaeHie Park <saehie.park@samsung.com>
4 years ago[moco-tf] Remove dump shape after import (#7811)
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 30 Sep 2019 03:05:40 +0000 (12:05 +0900)]
[moco-tf] Remove dump shape after import (#7811)

This is to remove usage of ShapeInferenceData.
As loco can display shape information of the nodes, we do not need to dump the shape information after import.

Signed-off-by: SaeHie Park <saehie.park@samsung.com>
4 years agoRename cpu backend Shape (#5534)
Alexander Efimov/AI Tools Lab/./Samsung Electronics [Mon, 30 Sep 2019 02:24:01 +0000 (05:24 +0300)]
Rename cpu backend Shape (#5534)

Rename cpu backend Shape class into TensorDescriptor

Signed-off-by: Efimov Alexander <a.efimov@samsung.com>
4 years agoRoll-back and fix typo in CONTRIBUTORS (#7809)
이성재/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 27 Sep 2019 08:43:49 +0000 (17:43 +0900)]
Roll-back and fix typo in CONTRIBUTORS (#7809)

- Roll-back CONTRIBUTUTORS file from mis committed.
- Fix typo in filename CONTRIBUTUTORS to CONTRIBUTORS.

Signed-off-by: Sung-Jae Lee <sj925.lee@samsung.com>
4 years ago[exo-tflite] shape / type inference for TFLTranspose (#7803)
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 27 Sep 2019 08:28:22 +0000 (17:28 +0900)]
[exo-tflite] shape / type inference for TFLTranspose (#7803)

* [exo-tflite] shape / type inference for TFLTranspose

This adds shape and type inference for TFLTranspose and a test case.

Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* Fix wrong test case. Previously it is testing loco::Transpose, now it tests locoex::TFLTranspose

* remove method needed for loco::TensorTranspose

* typo

4 years ago[loco] Introducing PermutingDecoder for filter (#7801)
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 27 Sep 2019 08:22:28 +0000 (17:22 +0900)]
[loco] Introducing PermutingDecoder for filter (#7801)

* [loco] Introducing PermutingDecoder for filter

This adds PermutingDecoder for filter and related test case.

Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* move PermutingDecoder from DepthwiseFilter area to filter area

4 years agoMake to support Cast op for acl neon (#7800)
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 27 Sep 2019 08:22:14 +0000 (17:22 +0900)]
Make to support Cast op for acl neon (#7800)

This commit make to support Cast op for acl neon.
  - Add NECast layer and NECastKernel
  - Apply NECast layer for neurun

Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
4 years agoEnable some nnapi tests on ACL neon (#7756)
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 27 Sep 2019 08:20:46 +0000 (17:20 +0900)]
Enable some nnapi tests on ACL neon (#7756)

This commit enables some nnapi tests on ACL neon.
  - dequantize*
  - TrivialTest.BroadcastMulTwo

Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
4 years ago[loco] Type and Shape inference for FilterDecode (#7804)
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 27 Sep 2019 07:50:52 +0000 (16:50 +0900)]
[loco] Type and Shape inference for FilterDecode (#7804)

This adds Type and Shape inference for FilterDecode.

Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
4 years ago[nnfw-gbs] Use pre-generated test code (#7805)
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 27 Sep 2019 07:31:24 +0000 (16:31 +0900)]
[nnfw-gbs] Use pre-generated test code (#7805)

Update nnfw.spec to use pre-generated nnapi test code

Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
4 years ago[cmd] Fix build docker image command help (#7742)
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 27 Sep 2019 06:52:58 +0000 (15:52 +0900)]
[cmd] Fix build docker image command help (#7742)

Remove deprecated option: tizen build option was removed

Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
4 years ago[nnfw-cmake] Revise tflite version setting (#7793)
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 27 Sep 2019 06:39:32 +0000 (15:39 +0900)]
[nnfw-cmake] Revise tflite version setting (#7793)

Define option for tflite external code version
Replace ported tflite files to support two or more version
Prepare to remove TFLITE_MAJOR_VER and TFLITE_MINOR_VER setting

Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
4 years ago[ARMCompute] Fix for Android build (#7791)
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 27 Sep 2019 06:02:17 +0000 (15:02 +0900)]
[ARMCompute] Fix for Android build (#7791)

- Do not build ARMCompute by default
- Fix `CMAKE_FIND_ROOT_PATH_MODE_LIBRARY` to search in the host dirs

For now, using pre-built library is recommended and building ARMCompute
is not enabled. It can be used by specifying `EXT_ACL_FOLDER` in
Makefile or pass `ARMCompute_EXTDIR` to CMake.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
4 years ago[android_tflite] Remove finding NDK_DIR (#7799)
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 27 Sep 2019 06:01:59 +0000 (15:01 +0900)]
[android_tflite] Remove finding NDK_DIR (#7799)

Remove automatic finding from CMake. Users must specify NDK_DIR from now
on.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
4 years ago[exo-tflite] fix wrong #endif guard (#7784)
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 27 Sep 2019 04:41:46 +0000 (13:41 +0900)]
[exo-tflite] fix wrong #endif guard (#7784)

Wrong guard was fixed. __CONVERSION_GRAPH_BLOCK_H__ to __GRAPH_BLOCK_H__

Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
4 years ago[loco] TensorTranspose: Shape and Type inference (#7796)
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 27 Sep 2019 04:41:33 +0000 (13:41 +0900)]
[loco] TensorTranspose: Shape and Type inference (#7796)

* [loco] TensorTranspose: Shape and Type inference

This enables shape and type inference for loco::TensorTranspose. A TC for shape inference was also added.

Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* Transpose -> TensorTranspose

4 years ago[exo-tflite] knob for TFLTranspose (#7797)
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 27 Sep 2019 04:40:29 +0000 (13:40 +0900)]
[exo-tflite] knob for TFLTranspose (#7797)

A knob for TFLTranspose was added.

Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
4 years ago[exo-tflite] Introducing TFLConv2D (#7795)
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 27 Sep 2019 04:38:50 +0000 (13:38 +0900)]
[exo-tflite] Introducing TFLConv2D (#7795)

* [exo-tflite] Introducing TFLConv2D

This adds TFLConv2D, which is a representation of Tensorflow Lite CONV_2D.

Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* ( ? : ) operator to switch

4 years ago[loco] Introducing TensorTranspose node (#7772)
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 27 Sep 2019 03:29:32 +0000 (12:29 +0900)]
[loco] Introducing TensorTranspose node (#7772)

* [loco] Introducing TensorTranspose node

This adds TensorTranspose node, which permutes a tensor.

Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* remove initial values of perm

4 years ago[moco-tf] Introduce TFShapeInferenceHelper (#7790)
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 27 Sep 2019 03:28:51 +0000 (12:28 +0900)]
[moco-tf] Introduce TFShapeInferenceHelper (#7790)

* [moco-tf] Introduce TFShapeInferenceHelper

This will introduce TFShapeInferenceHelper to provide helper methods for smooth migrarion to using loco shape inference service and other functionalities for shape related helper functions.

Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* remove dup comments

4 years ago[nnfw-cmake] Revise build option setting (#7780)
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 27 Sep 2019 02:13:08 +0000 (11:13 +0900)]
[nnfw-cmake] Revise build option setting (#7780)

- Reorder correct place for default in CfgOptionFlags.cmake
- Remove redundant platform specific setting - same with default
- Remove redundant option setting in tizen cross build script
- Remove BUILD_TOOLS option
- Introduce BUILD_OPENCL_TOOL to set build option for each tool

Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
4 years ago[nnfw-cmake] Replace boost package config (#7786)
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 27 Sep 2019 00:18:14 +0000 (09:18 +0900)]
[nnfw-cmake] Replace boost package config (#7786)

Move boost & source package config to common directory

Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
4 years ago[nnfw-cmake] Replace acl package config (#7785)
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 27 Sep 2019 00:17:30 +0000 (09:17 +0900)]
[nnfw-cmake] Replace acl package config (#7785)

Move acl & source package config to common directory

Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
4 years ago[exo-tflite] Introducing TFLTranspose (#7783)
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 26 Sep 2019 23:39:41 +0000 (08:39 +0900)]
[exo-tflite] Introducing TFLTranspose (#7783)

TFLTranspose is added to represent Tensorflow Lite Transpose operation.

Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
4 years ago[loco] Introducing FilterDecode (#7778)
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 26 Sep 2019 11:48:06 +0000 (20:48 +0900)]
[loco] Introducing FilterDecode (#7778)

This adds FilterDecode, which converts a filter to a tensor.

Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
4 years ago[exo-tflite] Refactoring ReluConverterTest (#7781)
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 26 Sep 2019 11:46:49 +0000 (20:46 +0900)]
[exo-tflite] Refactoring ReluConverterTest (#7781)

1) using TestGraph to shorten code; 2) first test does not check loco::Pull

Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
4 years ago[tf2tflite] Enable NET_0028 test (#7767)
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 26 Sep 2019 11:33:48 +0000 (20:33 +0900)]
[tf2tflite] Enable NET_0028 test (#7767)

Let's add NET_0028 to tf2tflite integration testsuite.

Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
4 years ago[moco] Enable NET_0028 test (#7766)
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 26 Sep 2019 11:33:29 +0000 (20:33 +0900)]
[moco] Enable NET_0028 test (#7766)

Let's add NET_0028 to moco integration testsuite.

Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
4 years ago[Dist/Tizen] Do not install libraries with devel packages. (#7769)
함명주/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 26 Sep 2019 09:50:35 +0000 (18:50 +0900)]
[Dist/Tizen] Do not install libraries with devel packages. (#7769)

Do not install duplicated libraries with devel packages; it incur build errors in applications with indirect dependencies on nnfw.
For example, if A Requires B and B BuildRequires nnfw, building A indirect-BuildRequires nnfw's *.so files, which cannot be resolved by GBS or OBS because both nnfw and nnfw-devel have nnfw's *.so files.
Thus, if nnfw has all *.so files in _libdir, do not install them with devel except for representative symlinks.

Fixes #7714

Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
4 years agoRevise tizen build (#7755)
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 26 Sep 2019 09:46:37 +0000 (18:46 +0900)]
Revise tizen build (#7755)

Remove special setting for gbs/obs build
Install armcl package in tizen rootfs
Move default setting for tizen into cmake option

Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
4 years ago[tfldump] Use external TensorFlow 1.14 schema file (#7762)
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 26 Sep 2019 08:29:19 +0000 (17:29 +0900)]
[tfldump] Use external TensorFlow 1.14 schema file (#7762)

This commit uses tflite schema file from TensorFlow 1.14 external source
code.

Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
4 years ago[moco-tf] Remove Phase (#7760)
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 26 Sep 2019 07:28:12 +0000 (16:28 +0900)]
[moco-tf] Remove Phase (#7760)

logo replaces this internal implementation of Phase.

Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
4 years ago[moco-tf] Introduce CodecHelper (#7757)
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 26 Sep 2019 07:27:44 +0000 (16:27 +0900)]
[moco-tf] Introduce CodecHelper (#7757)

This commit introduces CodecHelper module which provides
set_feature_enc/set_feature_dec helpers.

This change reduces duplication in code base.

Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
4 years ago[loco] Remove unnecessary friend declarations (#7753)
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 26 Sep 2019 07:27:27 +0000 (16:27 +0900)]
[loco] Remove unnecessary friend declarations (#7753)

This commit removes unnecessary friend declarations from loco Graph
class.

These friend declarations are introduced for migration, but no longer
required.

Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
4 years ago[CI] Exit with fail code when build fail (#7758)
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 26 Sep 2019 07:23:57 +0000 (16:23 +0900)]
[CI] Exit with fail code when build fail (#7758)

Fix build script to check build fail

Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
4 years ago[nncc-cmake] Replace package config files (#7761)
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 26 Sep 2019 06:54:58 +0000 (15:54 +0900)]
[nncc-cmake] Replace package config files (#7761)

Replace package config files and related files to common directory with nnfw

Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
4 years ago[moco-tf] Reduce Code Duplication (#7743)
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 26 Sep 2019 05:50:34 +0000 (14:50 +0900)]
[moco-tf] Reduce Code Duplication (#7743)

This commit rewrites canonizliazation passes to use SimpleNodeTransform
in order to reduce code duplication.

Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
4 years ago[nncc-cmake] Rename nncc_find_package to nnas_find_package (#7740)
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 26 Sep 2019 04:20:15 +0000 (13:20 +0900)]
[nncc-cmake] Rename nncc_find_package to nnas_find_package (#7740)

Rename nncc_find_package to nnas_find_package to share package config with nnfw

Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
4 years ago[tflite_bencmark] Let users can turn On/Off build (#7750)
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 26 Sep 2019 02:44:47 +0000 (11:44 +0900)]
[tflite_bencmark] Let users can turn On/Off build (#7750)

* [tflite_bencmark] Let users can turn On/Off build

Introduce CMake variable `BUILD_TFLITE_BENCHMARK` which is a flag to
indicate do the build or not.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
* Update CfgOptionFlags.cmake

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
4 years ago[ARMComputeEx] Fix target_link_libraries dependency (#7745)
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 26 Sep 2019 01:36:23 +0000 (10:36 +0900)]
[ARMComputeEx] Fix target_link_libraries dependency (#7745)

This library actually needs `arm_compute` rather than
`arm_compute_core`. Without this, Android NDK build fails.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>