이상규/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 10 Oct 2019 04:32:39 +0000 (13:32 +0900)]
[nnpkg-run] set default backend using OP_BACKEND_ALLOPS environment var. (#8003)
`nnpackage-run` (in short, `nnpkg-run`) uses `nnfw_session` from nnfw APIs.
It uses GeneralConfigSource. It means we need to set configuration using nnfw APIs.
Thus, it reads from environment var and set default backends.
Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
여지환/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 10 Oct 2019 03:58:32 +0000 (12:58 +0900)]
Update document for docker image creation (#8016)
Build command changed from 'nnfw' to 'nnas'.
Update https://github.sec.samsung.net/STAR/nnfw/blob/master/docs/nnfw/howto/HowToUseDockerImage.md to use 'nnas' instead of 'nnfw'.
Signed-off-by: JiHwan Yeo <jihwan.yeo@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 10 Oct 2019 03:56:40 +0000 (12:56 +0900)]
[nnapi-test] Port TestMain (#7907)
Port TestMain.cpp from upstream android nn runtime's NNAPI test
Fix
- Always define NNTEST_ONLY_PUBLIC_API to avoid android dependency
- Comment out android logging related code
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 10 Oct 2019 01:52:31 +0000 (10:52 +0900)]
[moco-tf] Revise shape inf for FusedBatchNorm and Identity (#8014)
This will revise shape inference for TFFusedBatchNorm and TFIdentity node to be done in TFShapeInferenceRule
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
채성우/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 10 Oct 2019 01:48:28 +0000 (10:48 +0900)]
[loco] Introduce Pad IR (#7992)
This commit introduces Pad IR to loco.
Signed-off-by: seongwoo <sw4670.chae@samsung.com>
[neurun] Templatize BaseLoader (#8012)
* Templatize `BaseLoader` class.
* Add tflite-specific `LoaderDomain` structure.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 10 Oct 2019 00:05:29 +0000 (09:05 +0900)]
[moco-tf] Revise shape inf for Const, Conv2D and Conv2DBackpropInput (#7999)
This will revise shape inference for TFConst, TFConv2D and TFConv2DBackpropInput node to be done in TFShapeInferenceRule
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 9 Oct 2019 22:07:12 +0000 (07:07 +0900)]
[moco-tf] Introduce node_shape_with_check helper (#7993)
This will introduce node_shape_with_check() helper in TFShapeInferenceRule for getting node_shape of a node and revise for TFBiasAdd node
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
[adtidas] Reformat using clang-format (#8013)
* Reformat the component using clang-format.
* Enable format checking.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[mir_tflite] Change the stream type to char (#8006)
`uint8_t` is not supported directly by the standard library, leading to errors in runtime.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
Enable regression test for tf2tflite utility (#8005)
- Enabled Regression test 2 that checks tf2tfite correctly handles conv2d withour dilation
- Enabled Regression test 1
Signed-off-by: Efimov Alexander <a.efimov@samsung.com>
Enable regression test for moco importer (#8004)
- Enabled regression test 2 that checks that tf-moco correctly handles conv2d withour dilation
- Enabled Regression test 1
Signed-off-by: Efimov Alexander <a.efimov@samsung.com>
김용섭/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 8 Oct 2019 11:04:17 +0000 (20:04 +0900)]
[neurun] Eliminate nnapi tests have different spec (#7290)
Eliminate nnapi tests which have different spec from acl.
nnapi can take weight/bias as input but acl can't.
Signed-off-by: Yongseop Kim <yons.kim@samsung.com>
Regression test for non dilated conv2d (#7970)
Test to check that convolution without dilation attribute compiles successfully
Signed-off-by: Efimov Alexander <a.efimov@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 8 Oct 2019 09:31:21 +0000 (18:31 +0900)]
[moco-tf] Revise shape inference for MaxPool (#7996)
This will revise shape inference for MaxPool to be done in TFShapeInferenceRule
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 8 Oct 2019 09:12:47 +0000 (18:12 +0900)]
[exo-tflite] Fix to use ActivationFunction (#7983)
Until now, `exo-tflite` didn't use `ActivationFunction` when making an options
even if some operations have it.
This commit will fix it to use ActivationFunction
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 8 Oct 2019 08:59:14 +0000 (17:59 +0900)]
[moco] Revise AvgPool Conv2D input names (#7989)
This will fix AvgPool and Conv2D input name to follow that from TensorFlow REGISTER_OP input names
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 8 Oct 2019 07:51:56 +0000 (16:51 +0900)]
[moco] Revise TFFusedBatchNorm input (#7975)
* [moco] Revise TFFusedBatchNorm input
This will revise TFFusedBatchNorm IR input name to follow TensorFlow document
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* use x for input
채성우/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 8 Oct 2019 07:51:37 +0000 (16:51 +0900)]
[moco-tf] Introduce TFPad IR (#7959)
* [moco-tf] Introduce TFPad IR
This commit introduces TFPad IR to moco-tf.
Signed-off-by: seongwoo <sw4670.chae@samsung.com>
* apply comments.
* move location of function declaration alphabetical order.
* rename input method as convention and remove unnecessary assert.
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 8 Oct 2019 07:46:13 +0000 (16:46 +0900)]
[exo-tflite] Introduce 'optimize(..)' that optimize TFL Nodes (#7958)
* [exo-tflite] Introduce 'optimize(..)' that optimize TFL Nodes
This adds 'optimize(..)' that optimize TFL Nodes.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* optimize.cpp to TFLOptimize.cpp
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 8 Oct 2019 07:44:19 +0000 (16:44 +0900)]
[cmd] Docker image build command (#7981)
Add argument in build docker image command for additional dockerfile - ubuntu 18.04
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
채성우/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 8 Oct 2019 07:21:38 +0000 (16:21 +0900)]
[loco] Introduce PaddingND (#7949)
* [loco] Introduce PaddingND
This commit introduces PaddingND class to loco for supporting Pad.
Signed-off-by: seongwoo <sw4670.chae@samsung.com>
* rename size method to rank.
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 8 Oct 2019 07:19:28 +0000 (16:19 +0900)]
[moco-tf] ShapeInf rev for DepthwiseConv2dNative (#7974)
This will revise shape inference for DepthwiseConv2dNative to be done in TFShapeInferenceRule
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 8 Oct 2019 07:07:31 +0000 (16:07 +0900)]
Make to support BatchToSpaceND op for acl cl (#7940)
This commit makes to support BatchToSpaceND op for acl cl.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 8 Oct 2019 07:04:37 +0000 (16:04 +0900)]
Remove layout from OperandInfo (#7962)
This commit removes layout from OperandInfo.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 8 Oct 2019 05:06:39 +0000 (14:06 +0900)]
Update Makefile.template for Android (#7919)
Pass and set additional CMake variables for Android build.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 8 Oct 2019 04:56:38 +0000 (13:56 +0900)]
Introduce Dockerfile based on Ubuntu 18.04 (#7973)
This is an identical file with `infra/docker/Dockerfile` except for
base Ubuntu version. This will be used for Android build which requires
higher version of CMake.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
채성우/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 8 Oct 2019 03:17:34 +0000 (12:17 +0900)]
[loco] Rename pad.test.cpp file (#7952)
This commit renames pad.test.cpp file to Padding2D.test.cpp.
Signed-off-by: seongwoo <sw4670.chae@samsung.com>
[neurun] Make some functions members of `BaseLoader` class (#7966)
* Move standalone functions `convertActivation`, `tensorTypeToDataType` and `lopadOperationIO` inside `BaseLoader` class.
* Rewrite `convertActivation` to use `switch` instead of `static_cast`.
* Improve some error messages a little.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 8 Oct 2019 01:56:11 +0000 (10:56 +0900)]
[res] Add testcases for merging concat (#7961)
This commit will add testcases for `MergeConcat` optimization
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 8 Oct 2019 01:55:12 +0000 (10:55 +0900)]
[exo-tflite] Introduce locoex::FusedActFunc converter (#7964)
This commit will introduce `locoex::FusedActFunc` converter,
to return `tflite::ActivationFunctionType`
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 8 Oct 2019 01:24:29 +0000 (10:24 +0900)]
[moco-tf] ShapeInf for TFConcatV2 (#7947)
* [moco-tf] ShapeInf for TFConcatV2
This will update shape inference to be done in TFShapeInferenceRule for TFConcatV2 node and not use ConcatData annotation
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* apply comments
* simplify assert check
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 8 Oct 2019 01:10:46 +0000 (10:10 +0900)]
[exo-tflite] exporting tflite CONV_2D op by using TFLConv2d (#7941)
This code enables tf2tflite to export tflite CONV_2D op by using TFLConv2d. (when knob is on)
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
[mir2loco] Fix Pad/Window/Stride initialization (#7968)
The order of height / width was messed up.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[mir_tflite] Change the type of `_tensorMap` to `vector`. (#7967)
`vector` is more efficient and less space consuming.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 7 Oct 2019 10:30:28 +0000 (19:30 +0900)]
[moco_tf] Remove FixShapeTransform.test.cpp (#7955)
This will remove FixShapeTransform.test.cpp file
- FixShapeTransform will be replaced by TFShapeInferenceRule
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 7 Oct 2019 10:17:21 +0000 (19:17 +0900)]
Make to support SpaceToBatchND op for acl neon (#7950)
This commit make to support SpaceToBatchND op for acl neon.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
[neurun] Introduce BaseLoader (#7914)
Rename `TFLiteLoader` class to `BaseLoader` and move it from `tflite/tflite_loader.cc` to `base_loader/base_loader.h`.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 7 Oct 2019 10:14:32 +0000 (19:14 +0900)]
[android] Update build_android_ndk.sh (#7946)
Remove unused arguments for this script.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
[mir_tflite] Remove _modelRaw variable (#7956)
Replace private `_modelRaw` variable with local-to-function `model_buffer`.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 7 Oct 2019 07:17:50 +0000 (16:17 +0900)]
[moco-tf] Empty lines for TFShapeInferenceHelper.h (#7948)
This will add some empty lines for readability in TFShapeInferenceHelper.h
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
이성재/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 7 Oct 2019 06:10:58 +0000 (15:10 +0900)]
Fix `./nnfw configure` command (#7939)
- Added INSTALL_PATH creation to satisfy prerequisites for the
`./nnfw install` command.
Signed-off-by: Sung-Jae Lee <sj925.lee@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 7 Oct 2019 05:45:23 +0000 (14:45 +0900)]
[exo-tflite] Rewriting a test using exo::test::TestGraph (#7924)
Test was re-written using exo::test::TestGraph and was simplified.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 7 Oct 2019 05:34:03 +0000 (14:34 +0900)]
Make to support SpaceToBatchND op for acl cl (#7933)
This commit make to support SpaceToBatchND op for acl cl.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
이춘석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 7 Oct 2019 03:53:41 +0000 (12:53 +0900)]
Pad Operation in tflite_loader (#7932)
Pad operaions is now supported by tflite_loader
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
Fix validate check in Conv2d (#7937)
Do not accept conv2d operation with non one dilations
Signed-off-by: Efimov Alexander <a.efimov@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 7 Oct 2019 01:53:56 +0000 (10:53 +0900)]
[exo-tflite] remove unused header (#7923)
This removes unused header in `TFLShapeInferenceRule.cpp`.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 7 Oct 2019 01:51:43 +0000 (10:51 +0900)]
[res] make explain.sh excutable (#7917)
exeplain.sh now can be run (by `chmod +x explain.sh`)
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 7 Oct 2019 00:12:09 +0000 (09:12 +0900)]
[moco] All IRs (#7908)
* [moco] All IRs
This will add rest of IRs of moco for TFDialect
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* fix def
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Sun, 6 Oct 2019 23:54:16 +0000 (08:54 +0900)]
[exo-tflite] Introducing Pass dir and Passes.h (#7909)
This introduces Pass dir, which is for all optimization pass for TFL nodes, and Passes.h, which is a aggregation of passes in Pass dir.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Sun, 6 Oct 2019 23:53:04 +0000 (08:53 +0900)]
[exo-tflite] Remove body of CanonicalNodeConverter.cpp (#7921)
`Run` is now included inside `*.h` so removed from `*.cpp` file.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Sun, 6 Oct 2019 23:49:57 +0000 (08:49 +0900)]
[exo-tflite] Code to export TFLConv2D (#7926)
Now OperationExporter has code to export TFLConv2D.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Sun, 6 Oct 2019 23:46:17 +0000 (08:46 +0900)]
[exo-tflite] Complete node summary for TFLAvergePool2D & TFLMaxPool2D (#7928)
Complete node summary for TFLAvergePool2D & TFLMaxPool2D was added.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Sun, 6 Oct 2019 23:41:49 +0000 (08:41 +0900)]
[exo-tflite] adding an assert to check rank of ifm in TFLShapeInferenceRule (#7922)
This adds an `assert` to check rank of ifm in TFLShapeInferenceRule.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
[mir_tflite] Use object-based API for subgraphs and tensors (#7936)
Object-based API is more convenient.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[mir_tflite] Use object-based API for operators (#7934)
Object-based API is more convenient.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 4 Oct 2019 08:30:43 +0000 (17:30 +0900)]
[res] Adding manifest for NET_0008 and NET_0009 (#7916)
Summary for NET_0008 and NET_0009 was added.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 4 Oct 2019 08:30:02 +0000 (17:30 +0900)]
[exo-tflite] Remove "// TODO TFLNop" (#7925)
Since we don't have `TFLNop`, related comments were removed.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
채성우/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 4 Oct 2019 08:25:50 +0000 (17:25 +0900)]
[res] Add simple Pad test case (#7893)
* [res] Add simple Pad test case
This commit adds simple pad test case.
Signed-off-by: seongwoo <sw4670.chae@samsung.com>
* make manifest.
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 4 Oct 2019 08:20:13 +0000 (17:20 +0900)]
[logo] simplifying FilterEncode-FilterDecode (#7875)
* [logo] simplifying FilterEncode-FilterDecode to Transpose (or to Forward)
This commit simplifying FilterEncode-FilterDecode to Transpose (or to Forward)
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* fix wrong const size
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 4 Oct 2019 08:03:17 +0000 (17:03 +0900)]
[nnfw-build] Revise generate test setting (#7905)
- Remove GENERATE_NNAPI_TESTS in makefile
- Use cmake option to disable generate test
- Default: enable
- For tizen: disable as default for gbs/obs, enable on tizen cross build script
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
[neurun] Schedule branch by branch instead of node (#6965)
* [neurun] Schedule branch by branch instead of node
Related issue: 6963
Schedule whole branch (that can be seen as a subgraph) recursively
before moving to next node
Signed-off-by: Poshshoev Dilshodzhon <d.poshshoev@samsung.com>
* Fix CI failure
Signed-off-by: Poshshoev Dilshodzhon <d.poshshoev@samsung.com>
* Address offline review
Signed-off-by: Poshshoev Dilshodzhon <d.poshshoev@samsung.com>
* Fix review feedbacks
Signed-off-by: Poshshoev Dilshodzhon <d.poshshoev@samsung.com>
* Address feedbacks: rename scheduleRecursively
Signed-off-by: Poshshoev Dilshodzhon <d.poshshoev@samsung.com>
* Fix documentation typo
Signed-off-by: Poshshoev Dilshodzhon <d.poshshoev@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 4 Oct 2019 01:36:08 +0000 (10:36 +0900)]
[nnfw-gbs] Remove outdated code in nnfw.spec (#7906)
Remove pacl related code
Packaging *.so file only and remove useless exclude
Remove install libbackend_acl_common.a
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
[mir-tflite-importer] Hide class TFLiteImporter definition (#7913)
Class moved to cpp in order to simplify interface header.
Signed-off-by: Dmitry Gusev <d.gusev@partner.samsung.com>
[mir-onnx-importer] Hide class OnnxImporterImpl definition (#7912)
Class moved to cpp in order to simplify interface header.
Signed-off-by: Dmitry Gusev <d.gusev@partner.samsung.com>
[mir-caffe2-importer] Hide class Caffe2Importer definition (#7911)
Class moved to cpp in order to simplify interface header.
Signed-off-by: Dmitry Gusev <d.gusev@partner.samsung.com>
[mir-caffe-importer] Hide class CaffeImporter definition (#7910)
In order to simplify interface header, class moved to cpp.
Signed-off-by: Dmitry Gusev <d.gusev@partner.samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Wed, 2 Oct 2019 06:49:35 +0000 (15:49 +0900)]
[nnapi-test] Introduce TestNeuralNetworksWrapper (#7903)
Port TestNeuralNetworksWrapper.h/cpp from upstream android nn runtime's NNAPI test
Fix
- Comment out 'include NeuralNetworksWrapperExtensions.h' and related namespace
- Use boost::optional instead of std::optional
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 2 Oct 2019 05:51:43 +0000 (14:51 +0900)]
[moco] Introduce TensorName and VariadicArityNode (#7899)
This will introduce two helper TensorName and VariadicArityNode used for TFDialect IRs
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
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>
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>
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>
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>
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>
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>
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>
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>
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>
[mir] Fix TensorVariant constructor used in broadcasting (#7889)
* Fix TensorType creation with old shape
Signed-off-by: Pavel Iliutchenko <p.iliutchenk@samsung.com>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>