Павел Ильютченко/AI Tools Lab /SRR/Engineer/삼성전자 [Mon, 23 Sep 2019 15:30:12 +0000 (18:30 +0300)]
[mir2loco] DeConv2D transformation implementation (#7642)
* Supported mir::DeConv2D transformation to loco::TransposedConv2D
* Added test for it
Signed-off-by: Pavel Iliutchenko <p.iliutchenk@samsung.com>
[mir2loco] Remove no longer unused functions (#7666)
Remove shape based `setupWindow` and `setupStride` which are no longer used.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[neurun]HEScheduler ignore backend if not profiled (#7641)
Now HEScheduler throws an exception if there is no profiling info for an
operation on a backend. This change makes HEScheduler not use this
backend for that operation
Signed-off-by: Poshshoev Dilshodzhon <d.poshshoev@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 23 Sep 2019 09:15:57 +0000 (18:15 +0900)]
[nike] Use GTest_AddTest helper (#7689)
This commit revises nike's CMakeLists.txt to use GTest_AddTest helper
instead of explicitly referring "gtest_main" target.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 23 Sep 2019 09:13:41 +0000 (18:13 +0900)]
[moco-tf] Same padding case for Conv2DBackpropInput Canonicalization (#7686)
* [moco-tf] Same padding case for Conv2DBackpropInput Canonicalization
This commit implements SAME padding case for Conv2DBackpropInput
Canonicalization
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
* Clarify comment
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 23 Sep 2019 08:21:40 +0000 (17:21 +0900)]
[moco-int] Enable test SquaredDifference_001 (#7690)
This will enable SquaredDifference_001 unit test
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
Introduce circle_loader (#7667)
Add `circle_loader` library without implementation.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 23 Sep 2019 07:59:37 +0000 (16:59 +0900)]
[moco-tf] Canonicalize rank-0 tensor constants (#7687)
This commit extends ConstCanonicalizer to translate tf.Const nodes that
emits a rank-0 tensor as a canonical.ConstGen node that emits a rank-1
tensor with shape [1].
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 23 Sep 2019 07:54:30 +0000 (16:54 +0900)]
[tf2tflite] Enable test SquaredDifference_001 (#7691)
This will enable SquaredDifference_001 unit test
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 23 Sep 2019 07:35:54 +0000 (16:35 +0900)]
[exo-tflite] Update Reshape exporter (#7680)
* [exo-tflite] Update Reshape exporter
Until now, new shape is exported only by `new_shape` attribute.
However, runtime was receiving new shape by input.
This commit will update `Reshape` exporter to export new shape
not only as attribute but also as input.
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
* add comments
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 23 Sep 2019 06:07:16 +0000 (15:07 +0900)]
[moco-integration-test] Get toolchain from a file (#7678)
This commit revises runall.sh to reads toolchain configuration from
a file instead of command-line arguments.
This change makes it easy to re-run failed tests for diagnostics.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 23 Sep 2019 05:48:48 +0000 (14:48 +0900)]
Change ReduceMean kernel of acl_cl (#7654)
This commit changes ReduceMean kernel of acl_cl.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 23 Sep 2019 05:32:15 +0000 (14:32 +0900)]
Make to support EmbeddingLookup op for acl neon (#7663)
This commit makes to support EmbeddingLookup op for acl neon.
- Introduce NEEmbeddingLookup NEEmbeddingLookupKernel
- Apply NEEmbeddingLookup layer for neurun
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 23 Sep 2019 05:27:20 +0000 (14:27 +0900)]
[moco-tf] Enable ResolveSquaredDifference knob (#7670)
This will enable ResolveSquaredDifference knob that decomposes SquaredDifference to Sub and Mul nodes
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 23 Sep 2019 05:11:05 +0000 (14:11 +0900)]
[tf2tflite] Enable test Div/Sub 001 (#7673)
This will enable unit test of Div/Sub 001 that tests broadcasting
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 23 Sep 2019 04:59:18 +0000 (13:59 +0900)]
[tf2tflite] Enable UNIT_Squeeze_003 test (#7679)
This commit will enable `UNIT_Squeeze_003` test in tf2tflite
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 23 Sep 2019 04:59:03 +0000 (13:59 +0900)]
Add manifest to TensorFlow NET_0028 test (#7672)
This commit adds a manifest for NET_0028 test which includes a
brief summary.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 23 Sep 2019 04:54:25 +0000 (13:54 +0900)]
Introduce `SupportPermutation` to check if backend supports permutation (#7616)
This commit introduces `SupportPermutation` to check if backend supports permutation.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 23 Sep 2019 04:32:51 +0000 (13:32 +0900)]
[nnfw-cmake] Change nnfw variable name used in modules (#7671)
Introduce NNFW_PROJECT_SOURCE_DIR and NNFW_EXTERNALS_DIR to alternate NNFW_XXX variable
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 23 Sep 2019 04:32:46 +0000 (13:32 +0900)]
[moco-integration-test] Enable UNIT_Mean_00x tests (#7656)
This commit will enable `UNIT_Mean_000` ~ `UNIT_Mean_003`
tests in moco-integration-test
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 23 Sep 2019 04:32:26 +0000 (13:32 +0900)]
[tf2tflite] Enable UNIT_Mean_00x tests (#7657)
This commit will enable `UNIT_Mean_000` ~ `UNIT_Mean_003`
tests in tf2tflite
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 23 Sep 2019 04:30:51 +0000 (13:30 +0900)]
[moco_int] Enable Add 005 unit test (#7674)
This will enable Add 005 unit test
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 23 Sep 2019 04:27:28 +0000 (13:27 +0900)]
[exo-tflite] Enable Div,Sub conversion knob (#7677)
This will enable knob for conversion of Div and Sub
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 23 Sep 2019 04:26:29 +0000 (13:26 +0900)]
[neurun] Implement NNAPI get output dimensions (#7653)
Inplement NNAPI ANeuralNetworksExecution_getOutputOperandDimensions and wrapper function
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 23 Sep 2019 02:05:48 +0000 (11:05 +0900)]
Make to support DepthToSpace op for acl neon (#7629)
This commit makes to support DepthToSpace op for acl neon.
- Introduce NEDepthToSpaceLayerEx and NEDepthToSpaceLayerKernelEx
- Apply NEDepthToSpaceLayerEx layer for neurun
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 23 Sep 2019 01:32:23 +0000 (10:32 +0900)]
[exo-tflite] Enable Div,Sub conversion (#7669)
This will enable Eltwise Div/Sub node conversion to TFL node by knob
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 23 Sep 2019 00:49:35 +0000 (09:49 +0900)]
[nncc-cmake] Change nncc variable name used in modules (#7662)
Introduce NNAS_PROJECT_SOURCE_DIR and NNAS_EXTERNALS_DIR to alternate NNCC_XXX variable
Use workaround for NNAS_EXTERNALS_DIR - CI issue
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Sun, 22 Sep 2019 23:55:41 +0000 (08:55 +0900)]
[moco-tf] Padding inference for 'valid' padding case (#7650)
* [moco-tf] Padding inference for 'valid' padding case
This commit implements padding inference for Conv2DBackpropInput
canonicalization, in case of 'valid' padding.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
* bool ready()
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Sun, 22 Sep 2019 22:19:17 +0000 (07:19 +0900)]
[moco-tf] Remove unused function arg (#7607)
This commit removes unused function argument in Conv2D canonicalizer.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 20 Sep 2019 08:34:25 +0000 (17:34 +0900)]
[cmake] Rename nncc_include to nnas_include (#7659)
Rename nncc_include to nnas_include to share modules with nnfw
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 20 Sep 2019 08:31:43 +0000 (17:31 +0900)]
[exo-tflite] pass PhaseStrategy from Saturate to Restart (#7652)
pass strategy was Restart but changed to Saturate by accident. This commit changes it to Restart again.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 20 Sep 2019 07:43:15 +0000 (16:43 +0900)]
[locomotiv] Introduce TensorReduce Operation (#7643)
* [locomotiv] Introduce TensorReduce Operation
This commit will introduce `TensorReduce` operation in `locomotiv`
For now, only supports `ReduceFunc::Mean`
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
* restructure
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 20 Sep 2019 07:27:16 +0000 (16:27 +0900)]
[moco-tf] Register Conv2DBackpropInput canonicalization (#7649)
This commit registers Conv2DBackpropInput canonicalization.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 20 Sep 2019 07:13:34 +0000 (16:13 +0900)]
[cmake] Move cmake modules (#7651)
Move nncc's cmake modules to infra/cmake/modules to share with nnfw
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 20 Sep 2019 07:12:22 +0000 (16:12 +0900)]
[exo-tflite] converting and exporting TFLConst (#7638)
With this, TFLConverter start work and TensorExporter exports tensor using TFLConst when the knob is on.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 20 Sep 2019 07:10:11 +0000 (16:10 +0900)]
Make to support ReduceMax/ReduceMin op for acl neon (#7599)
This commit makes to support ReduceMax/ReduceMin op for acl neon
- Introduce NEReductionOperationKernelEx
- Introduce NEReductionOperationEx and NEReduceOperation
- Apply NEReduceOperation layer for neurun to enable ReduceMax/ReduceMin op
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 20 Sep 2019 07:09:17 +0000 (16:09 +0900)]
[exo-tflite] change TFL node opcode (NOP_CONSTGEN -> CONST) (#7621)
Since use TFLConst (not anything like Nop), I changed the opcode also.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 20 Sep 2019 06:57:25 +0000 (15:57 +0900)]
[neurun] Implement a hash functor for enum class (#7647)
Implement a hash functor for enum class `NodeName` since Android NDK
clang requires it.
* This is a fix for Android Build
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 20 Sep 2019 06:56:57 +0000 (15:56 +0900)]
[nnas-cmd] Use pylint command anywhere (#7648)
Allow use pylint command any path under nnfw
- Use "git ls-files" to find '.FORMATDENY' file
- Use pushd/popd to search python files on nnfw's root path
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
이상규/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 20 Sep 2019 05:27:19 +0000 (14:27 +0900)]
Update tflite_benchmark_model for TF 1.14 (#7645)
It fixes `tflite_benchmark_model` build error that happends after
tensorflow is updated to 1.14.
Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 20 Sep 2019 04:37:26 +0000 (13:37 +0900)]
[nncc-cmake] Check external module path existance (#7646)
Check external module path existance and download if that is not exists
Same with nnfw
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
이춘석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 20 Sep 2019 04:04:58 +0000 (13:04 +0900)]
Make Compatible Shapes for CPU Add (#6540)
- This commit addes routine for making compatible shapes for CPU Add
kernel
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 20 Sep 2019 03:42:00 +0000 (12:42 +0900)]
Implement NNAPI get output rank function (#7581)
Inplement NNAPI ANeuralNetworksExecution_getOutputOperandRank and wrapper function
Introduce private field and method in Execution class to check execution is finished
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 20 Sep 2019 03:32:07 +0000 (12:32 +0900)]
[locomotiv] Layout rule for Matrix domain (#7634)
This commit updates layout rule for Matrix domain in README.md
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 20 Sep 2019 03:27:32 +0000 (12:27 +0900)]
[moco-tf] Implement Conv2DBackpropInput canonicalization partially (#7637)
This commit implements Conv2DBackpropInput canonicalization, except
padding inference.
Павел Ильютченко/AI Tools Lab /SRR/Engineer/삼성전자 [Fri, 20 Sep 2019 01:51:15 +0000 (04:51 +0300)]
[locomotiv] Implement MatrixEncode, MatrixDecode and MatMul (#7604)
* Implemented operations on loco interpreter (locomotiv)
* Added tests for this operations
Signed-off-by: Pavel Iliutchenko <p.iliutchenk@samsung.com>
[mir_onnx] Support ConvTranspose operator (#7610)
Add support for `ConvTranspose` operator.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[nnc] Fix DataFormatSwitcher transformation for DeConv2DOp (#7612)
Rearrange axes of `output_shape` parameter when switching the data format.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[mir_onnx] Fix shape of the bias term in Conv operator converter (#7611)
Reshape the bias term into [1, C, 1, 1] shape to make it broadcastable.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 19 Sep 2019 08:01:12 +0000 (17:01 +0900)]
[exo-tflite] Header-only CanonicalNodeConverter implementation (#7639)
Let's put CanonicalNodeConverter::run method implementation in the
header.
With this change, clients can inherit CanonicalNodeConverter
without "CanonicalNodeConverter.cpp" update.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 19 Sep 2019 07:15:41 +0000 (16:15 +0900)]
[loco] Use GTest_AddTest macro (#7635)
This commit simplifies loco's CMakeLists.txt by using GTest_AddTest macro.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 19 Sep 2019 06:52:44 +0000 (15:52 +0900)]
[exo-tflite] function to convert locoex padding to tflite padding (#7636)
This function converts locoex padding to tflite padding.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 19 Sep 2019 06:24:45 +0000 (15:24 +0900)]
[plier-tf] Enable strict build (#7631)
* [plier-tf] Enable strict build
Let's apply nncc common bulid flags (such as Wall).
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
* Update comment
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 19 Sep 2019 06:06:17 +0000 (15:06 +0900)]
[exo-tflite] adding converter for loco::ConstGen (#7632)
converter and its test for loco::ConstGen (to locoex::TFLConst) was added.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 19 Sep 2019 06:05:55 +0000 (15:05 +0900)]
[locop] Add TensorReduce visit function (#7597)
* [locop] Add TensorReduce visit function
This commit will add `TensorReduce` visit function
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
* apply comments
* fix mistake
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 19 Sep 2019 05:22:20 +0000 (14:22 +0900)]
[exo-tflite] test case for FeatureBiasAddConverter (#7583)
* [exo-tflite] test case for FeatureBiasAddConverter
An unit test to check `FeatureBiasAddConverter` is added.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* fix typo and wrong index of constgen->at(1)
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 19 Sep 2019 05:21:42 +0000 (14:21 +0900)]
[moco-tf] Basic structure for Conv2DBackpropInput canonicalize (#7627)
This commit adds basic structure for TFConv2DBackpropInput canonicalize
stage.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 19 Sep 2019 05:14:29 +0000 (14:14 +0900)]
[plier-tf] Fix copy_shape (#7615)
copy_shape is expected to throw an exception if there is any unknown
dimension (-1), but the current implementation never throws an
exception.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 19 Sep 2019 04:39:47 +0000 (13:39 +0900)]
[res] UNIT_Conv2DBackpropInput_002 (#7606)
* [res] UNIT_Conv2DBackpropInput_002
This commit adds UNIT_Conv2DBackpropInput_002, unit test for
Conv2DBackpropInput with valid padding & non-tight case.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
* Review feedback
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 19 Sep 2019 04:25:30 +0000 (13:25 +0900)]
[res] UNIT_Conv2DBackpropInput_000 (#7605)
This commit adds UNIT_Conv2DBackpropInput_000, unit test for
Conv2DBackpropInput with valid padding & tight fit case.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 19 Sep 2019 04:06:26 +0000 (13:06 +0900)]
Introduce T/F Testcase explain script (#7603)
* Introduce T/F Testcase explain script
This "explain" script generates a summary over commited TensorFlow
testcases.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
* Update HEADER
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 19 Sep 2019 04:05:44 +0000 (13:05 +0900)]
[moco-tf] Support TFMean Canonicalization (#7594)
* [moco-tf] Support TFMean Canonicalization
This commit will enable for supporting `TFMean` canonicalization
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
* apply comments
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 19 Sep 2019 03:55:56 +0000 (12:55 +0900)]
[moco-tf] Import Conv2dBackpropInput (#7579)
This commit introduces import stage of Conv2dBackpropInput.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 19 Sep 2019 03:55:12 +0000 (12:55 +0900)]
[loco] Shape inference for TransposedConv2D (#7564)
This commit introduces shape inference for TransposedConv2D and its test
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 19 Sep 2019 03:54:25 +0000 (12:54 +0900)]
[moco-tf] ResolveSquaredDifference to TFOptimizier (#7593)
This will add ResolveSquaredDifference transform to TFOptimizier phase
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 19 Sep 2019 03:53:38 +0000 (12:53 +0900)]
[nnfw-cmake] Rename cmake module file for subdirecories (#7617)
Use same name with nncc
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 19 Sep 2019 02:13:26 +0000 (11:13 +0900)]
[nncc-cmake] Support excluding install in add_extdirectory cmake macro (#7590)
- Support EXCLUDE_FROM_ALL in add_extdirectory (same with nnfw)
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 19 Sep 2019 02:09:05 +0000 (11:09 +0900)]
[nncc-cmake] Rename cmake module file for subdirecories (#7588)
Change snake-case type file naming to camel-case (same with other modules)
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 19 Sep 2019 02:07:04 +0000 (11:07 +0900)]
Remove boolean flag setting in cmake module (#7589)
Module don't use flag (same with nncc)
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 19 Sep 2019 01:46:28 +0000 (10:46 +0900)]
[nncc-cmake] Revise ExternalSourceTools status message print (#7601)
- Remove remain "--" print and use STATUS mode
- More information when download failed
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 19 Sep 2019 01:32:05 +0000 (10:32 +0900)]
[exo-tflite] Enable convert Div Sub (#7585)
This will enable Canonical Node Convert of Div and Sub node
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 19 Sep 2019 01:31:42 +0000 (10:31 +0900)]
[exo-tflite] Broadcast conversion for Div, Sub (#7591)
This will enable TensorBroadcastConverter to handle Div and Sub node
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 19 Sep 2019 01:31:10 +0000 (10:31 +0900)]
[moco-tf] Reduce dup code for copy_shapedata (#7592)
This will update copy_shapedata() to use get_node_shape() to reduce duplicate codes in FixShapeTransform
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Wed, 18 Sep 2019 23:33:22 +0000 (08:33 +0900)]
Introduce ONNXTests resource (#7565)
* Introduce ONNXTests resource
This commit introduces ONNXTests resource which provide simple ONNX
models for testing.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
* Update comment
[mir_onnx] Support Sqrt operator (#7548)
Add support for `Sqrt` operator.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[mir_onnx] Support Tanh operator (#7608)
Add support for `Tanh` operator.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[mir_tflite] Support Rsqrt operator (#7546)
Add support for `Rsqrt` operator.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[mir_onnx] Support Reciprocal operator (#7598)
Add support for `Reciprocal` operator.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[mir_onnx] Support Sub operator (#7547)
Add support for `Sub` operator.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 18 Sep 2019 11:37:34 +0000 (20:37 +0900)]
Introduce PermutationOperationPass (#7254)
This commit introduces PermutationOperationPass that can change layout of each op
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 18 Sep 2019 11:11:42 +0000 (20:11 +0900)]
[exo-tflite] Add TensorReduce visit function (#7596)
This commit will add `TensorReduce` visit function
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Wed, 18 Sep 2019 10:29:59 +0000 (19:29 +0900)]
[nnfw-cmake] StampTools module (#7595)
Extract StampTools module from ExternalSourceTools module (same with nncc)
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 18 Sep 2019 10:16:05 +0000 (19:16 +0900)]
[moco-tf] Introduce SqueezeReduceNode transformation (#7580)
* [moco-tf] Introduce SqueezeReduceNode transformation
If Reduce node such as `TFMean` do not keep dimensions, `TFSqueeze` should be inserted.
This commit will introduce `SqueezeReduceNode` transformation for that.
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
* revise comments
* add assert
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Wed, 18 Sep 2019 10:09:46 +0000 (19:09 +0900)]
Implement NNAPI synchronous execution function (#7574)
Inplement NNAPI ANeuralNetworksExecution_compute function and wrapper
Wrapper uses neurun's synchronous execution method
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 18 Sep 2019 10:08:01 +0000 (19:08 +0900)]
[exo-tflite] Enable Add, Mul unit tests (#7586)
This will enable Add, Mul unit tests having different input shape
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 18 Sep 2019 09:42:02 +0000 (18:42 +0900)]
[exo-tflite] adding FeatureBiasAddConverter into convert.cpp (#7584)
This makes FeatureBiasAddConverter work when a knob is on.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 18 Sep 2019 08:07:54 +0000 (17:07 +0900)]
[exo-tflite] enclosing namespace exo (#7577)
This commit enclosed codes with `namespace exo`.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
Vladimir Plazun/AI Tools Lab /SRR/Engineer/삼성전자 [Wed, 18 Sep 2019 08:03:48 +0000 (11:03 +0300)]
[custom op] Remove questionable nnfw_core->nnfw_api dependency (#7410)
* [custom op] Remove questionable nnfw_core->nnfw_api dependency
Split KernelRegistry into KernelRegistry and KernelBuilder
Signed-off-by: Vladimir Plazun <v.plazun@samsung.com>
* format fix
* make common frontend into static lib
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Wed, 18 Sep 2019 07:59:05 +0000 (16:59 +0900)]
[encodump] Use safemain (#7578)
Let's reuse safemain's main implemenation.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 18 Sep 2019 07:43:04 +0000 (16:43 +0900)]
[exo-tfltie] type and shape inference for TFLConst (#7567)
This adds type and shape inference for `TFLConst`.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 18 Sep 2019 07:41:39 +0000 (16:41 +0900)]
[moco-tf] Type inference for Conv2DBackpropInput (#7573)
This commit adds type inference for Conv2DBackpropInput.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 18 Sep 2019 07:40:05 +0000 (16:40 +0900)]
Rename to moco-integration-test (#7569)
This commit rename 'mocotest-tf' directory to 'moco-integration-test'.
It also updates README.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 18 Sep 2019 07:36:34 +0000 (16:36 +0900)]
[exo-tflite] Fix TFLMul export option (#7566)
This will fix TFLMul export option that was not corrected from copy-paste
As the option is a dummy, there was no problem with actual execution
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 18 Sep 2019 07:34:55 +0000 (16:34 +0900)]
[moco-tf] Summary for TFConv2DBackpropInput (#7575)
This commit introduces summary for TFConv2DBackpropInput.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
Павел Ильютченко/AI Tools Lab /SRR/Engineer/삼성전자 [Wed, 18 Sep 2019 07:34:44 +0000 (10:34 +0300)]
[loco] Introduce MatrixEncode, MatrixDecode and MatrixMul operations (#7406)
* Added MatrixEncode, MatrixDecode and MatrixMul nodes
* Added tests for them
* Defined CanonicalShapeInferenceRule and TypeInference
Signed-off-by: Pavel Iliutchenko <p.iliutchenk@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Wed, 18 Sep 2019 07:31:33 +0000 (16:31 +0900)]
[plier-tf] Hide Implementation Details (#7576)
No code outside plier-tf uses membuf and imemstream classes.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 18 Sep 2019 06:17:24 +0000 (15:17 +0900)]
[loco] Introduce TensorReduce Operation (#7497)
* [loco] Introduce TensorReduce Operation
This commit will introduce `TensorReduce` operation in loco
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
* fix wrong shapeInference
* revise comments
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Wed, 18 Sep 2019 06:10:34 +0000 (15:10 +0900)]
[tf2tflite] Enable exo loggers with TF2TFLITE knob (#7561)
With this commit, users are able to enable exo loggers with
TF2TFLITE_Log_Backend knob.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 18 Sep 2019 06:07:58 +0000 (15:07 +0900)]
[moco-tf] Shape inference for Conv2DBackpropInput (#7539)
This commit implements shape inference for Conv2DBackpropInput.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 18 Sep 2019 06:05:32 +0000 (15:05 +0900)]
[exo-tflite] revising README.md (#7529)
* [exo-tflite] revising README.md
Steps for adding a new TFL node was added.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* revised