[mir] New interface for PadOp (#6708)
* Remove redundant `num_dims` parameter of constructor.
* Make padding arguments consistent with other operations (two vectors instead of vector of pairs).
* Change the type of the value to be used for padding from `Scalar` to `float` (it is `float` in all supported input frameworks).
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
채성우/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 20 Aug 2019 10:29:21 +0000 (19:29 +0900)]
[exo-tflite] Export DepthwiseConv2D operation (#6720)
This commit introduce DepthwiseConv2D operation to exporter.
Signed-off-by: seongwoo <sw4670.chae@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 20 Aug 2019 10:17:11 +0000 (19:17 +0900)]
Check format changed files (#6683)
Check and fix format with warning message when there is changed files
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 20 Aug 2019 09:58:41 +0000 (18:58 +0900)]
Fix coverage test suite (#6725)
Add lost coverage test suite packaging files: nnas, nnfw script
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 20 Aug 2019 09:56:54 +0000 (18:56 +0900)]
Introduce ThirdPartyTools module (#6730)
* Introduce ThirdPartyTools module
This commit introduces ThiryPartyTools module. The current version
includes ThirdParty_URL function which manages URLs via file.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 20 Aug 2019 09:45:10 +0000 (18:45 +0900)]
[moco-tf] Shape inference on canonical optimize stage (#6722)
This commit adds shape inference at canonical optimize stage of moco.
This is required for ResolveRedundantReshape transform.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 20 Aug 2019 09:35:24 +0000 (18:35 +0900)]
[moco-tf] TFNodeSummary for TFStopGradient (#6738)
This will add TFStopGradient to TFNodeSummaryBuilder
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 20 Aug 2019 09:12:28 +0000 (18:12 +0900)]
[moco-tf] Update shape inference in Canonicalizers (#6732)
This will update shape inference to use loco service in Canonicalizers of Add,Mul,Realdiv and Sub nodes
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 20 Aug 2019 09:12:05 +0000 (18:12 +0900)]
[moco-tf] Canonicalizer for TFStopGradient (#6731)
This will introduce Canonicalizer for TFStopGradient that converts to Canonical Forward node
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 20 Aug 2019 08:16:44 +0000 (17:16 +0900)]
[moco-tf] GraphBuilder for StopGradient (#6729)
This will introduce GraphBuilder for StopGradient node to import as TFStopGradient IR
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 20 Aug 2019 08:05:43 +0000 (17:05 +0900)]
[moco-tf] TFNodeSummaryBuilder for TFConcatV2 (#6727)
This will add TFNodeSummaryBuilder summary method for TFConcatV2
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 20 Aug 2019 08:03:10 +0000 (17:03 +0900)]
[moco-tf] Turn on Knob for ImportAsTFConcatV2 (#6728)
This will turn on ImportAsTFConcatV2 knob so as to make ConcatV2 is imported as TF dialect TFConcatV2
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 20 Aug 2019 07:46:48 +0000 (16:46 +0900)]
[moco-tf] Canonicalizer for TFRsqrt (#6723)
This will introduce Canonicalizer for TFRsqrt node that converts to "1/Sqrt(x)"
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 20 Aug 2019 07:46:23 +0000 (16:46 +0900)]
[moco_tf] Do shape,type inference before Canonicalizer (#6718)
This will run shape and type inference before Canonicalizer and use Restart type phase when any node type has changed
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 20 Aug 2019 07:44:05 +0000 (16:44 +0900)]
Make OperationValidator supports multi-layout for frontend (#6714)
This commit makes OperationValidator supports multi-layout for frontend.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 20 Aug 2019 07:38:12 +0000 (16:38 +0900)]
[moco-tf] Register shape inference pass (#6726)
This commit registers shape inference pass to optimize stage transforms
list.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 20 Aug 2019 07:34:00 +0000 (16:34 +0900)]
[locomotiv] Remove GraphInput/Output node method calls (#6719)
This commit removes all the GraphInput::node and GraphOutput::node
method calls in locomotiv. This is a step toward loco GraphInput/Output
class redesign.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 20 Aug 2019 07:08:23 +0000 (16:08 +0900)]
[moco-tf] Canonicalize TFConcatV2 (#6703)
* [moco-tf] Canonicalize TFConcatV2
This will implement Canonicalize of TFConcatV2 to TensorConcat
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* fix log msg
Add ONNX sources v1.5 (#6642)
Add a cmake configs for fetching ONNX v1.5 sources.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 20 Aug 2019 07:05:03 +0000 (16:05 +0900)]
[exo-tflite] Do NOT invoke GraphInput/Output node (#6717)
This commit removes all the GraphInput::node and GraphOutput::node
method calls in exo-tflite. This is a step toward loco GraphInput/Output
class redesign.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 20 Aug 2019 07:04:36 +0000 (16:04 +0900)]
[moco-tf] Introduce TFStopGradient (#6715)
* [moco-tf] Introduce TFStopGradient
This will introduce TFStopGradient IR for TensorFlow StopGradient node and required codes
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* fix comment
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 20 Aug 2019 07:02:38 +0000 (16:02 +0900)]
[moco-tf] Do NOT invoke GraphInput/Output node (#6716)
As a step to loco redesign, this commit removes all
GraphInput::node and GraphOutput::node method calls in moco-tf.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
Ivan Vagin/AI Tools Lab /SRR/Engineer/삼성전자 [Tue, 20 Aug 2019 06:43:55 +0000 (15:43 +0900)]
[neurun] Introduced ITimer, CPUTimer and CLTimer (#6690)
* [neurun] Introduced ITimer, CPUTimer and CLTimer
* Introduced ITimer, CPUTimer and CLTimer
* Added timer getters to backends config
* Made ProfileObserver to measure time with ITimer
Signed-off-by: Ivan Vagin <ivan.vagin@samsung.com>
* Review fixes
Signed-off-by: Ivan Vagin <ivan.vagin@samsung.com>
채성우/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 20 Aug 2019 06:42:07 +0000 (15:42 +0900)]
[loco] Implement DepthwiseConv2D shape inference (#6569)
* [loco] Implement DepthwiseConv2D shape inference
This commit extends CanonicalShapeInferenceRule to accept
DepthwiseConv2D nodes.
Signed-off-by: seongwoo <sw4670.chae@samsung.com>
* fix a typo.
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 20 Aug 2019 05:01:52 +0000 (14:01 +0900)]
[res] TensorFlow StopGradient_000 (#6710)
This will introduce TensorFlow StopGradient_000 test material for basic StopGradient operator test
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 20 Aug 2019 05:01:30 +0000 (14:01 +0900)]
[mocotest-tf] Enable Sqrt and SquaredDifference (#6711)
This will enable Sqrt and SquaredDifference basic testing
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 20 Aug 2019 05:01:09 +0000 (14:01 +0900)]
[tf2tflite] Enable Sqrt and SquaredDifference (#6712)
This will enable Sqrt and SquaredDifference basic testing
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 20 Aug 2019 04:44:02 +0000 (13:44 +0900)]
[moco-tf] Introduce TypeInferencePass (#6702)
This will introduce TypeInferencePass class to provide type inference as a Transform
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 20 Aug 2019 04:38:39 +0000 (13:38 +0900)]
[moco-tf] Introduce ShapeInferencePass (#6701)
This will introduce ShapeInferencePass class to provide shape inference as a Transform
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 20 Aug 2019 04:10:01 +0000 (13:10 +0900)]
Make FeatureShape supports NCHW layout (#6705)
This commit makes FeatureShape supports NCHW layout.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
Ivan Vagin/AI Tools Lab /SRR/Engineer/삼성전자 [Tue, 20 Aug 2019 04:09:27 +0000 (13:09 +0900)]
[neurun] Removed some dead code (#6709)
Removed some dead code from ExecTime test
Signed-off-by: Ivan Vagin <ivan.vagin@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 20 Aug 2019 04:04:37 +0000 (13:04 +0900)]
[exo-tflite] Basic Classes for TFLDialect (#6682)
These commit adds basic classes for TFLDialect.
Namespace is 'locoex' since this is an extension of `loco`.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 20 Aug 2019 04:01:29 +0000 (13:01 +0900)]
[moco-tf] Type inference for TFSoftmax (#6700)
This will add type inference for TFSoftmax node
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 20 Aug 2019 03:46:25 +0000 (12:46 +0900)]
Use unified coverage report command (#6537)
- Use unified coverage report command in docker_coverage_report.sh
- Update Makefile.template to package infra
- Update test_coverage.sh to archive test data and codes
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 20 Aug 2019 02:29:19 +0000 (11:29 +0900)]
[loco] Introduce push_node/pull_node helper (#6684)
This commit introduces push_node/pull_node helper which serves as an
alternative to "node" method in GraphInput/GraphOutput class.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 20 Aug 2019 00:42:07 +0000 (09:42 +0900)]
Replace nnapi shim headers into libs (#6588)
Introduce libs/nnapi
Replace NeuralNetworksShims.h, NeuralNetworksExShims.h NeuralNetworksLoadHelpers.h to libs/nnapi
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 20 Aug 2019 00:41:20 +0000 (09:41 +0900)]
[moco-tf] GraphBuilder for TFConcatV2 (#6673)
This will fill GraphBuilder for TFConcatV2 and tests. Also ShapeInference implementation that is required by the test code.
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 19 Aug 2019 23:34:59 +0000 (08:34 +0900)]
[tf2tflite] Test REGRESSION_0000 (#6681)
From this commit, REGRESSION_0000 is tested by tf2tflite
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
[mir] Remove Dropout operation (#6675)
This is a no-op in inference mode.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[mir_onnx] Remove GivenTensorFill converter (#6666)
This operation is not a part of the ONNX specification.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
Add missing include guard (#6655)
Added missing include guard in onnx to mir converter
Signed-off-by: Efimov Alexander <a.efimov@samsung.com>
Павел Ильютченко/AI Tools Lab /SRR/Engineer/삼성전자 [Mon, 19 Aug 2019 17:35:00 +0000 (20:35 +0300)]
[mir2loco] Support DepthwiseConv operation in transformer (#6574)
* Added DepthwiseConv2DOp convertion and test for this
Signed-off-by: Pavel Iliutchenko <p.iliutchenk@samsung.com>
이상규/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 19 Aug 2019 10:13:23 +0000 (19:13 +0900)]
[nnpkg_run] add load option (#6688)
This patch adds `load` option.
It does not load input files yet.
Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 19 Aug 2019 10:08:42 +0000 (19:08 +0900)]
[mocotest-tf] Test REGRESSION_0000 (#6678)
REGRESSION_0000 test is now run for mocotest-tf
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 19 Aug 2019 10:07:46 +0000 (19:07 +0900)]
[moco-tf] Introduce ShapeInference (#6679)
This will introduce ShapeInference for loco graph and query for loco node that will do shape inference for all dialects used in moco
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 19 Aug 2019 09:07:29 +0000 (18:07 +0900)]
[mocotest-tf] Add divide comment for refactoring (#6680)
This comment will be removed after test case convention refactoring
project.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 19 Aug 2019 08:46:02 +0000 (17:46 +0900)]
[tf2tflite] Enable Softmax_001 test (#6687)
This commit will enable `Softmax_001` test in `tf2tflite`
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 19 Aug 2019 08:45:52 +0000 (17:45 +0900)]
[mocotest-tf] Enable Softmax_001 test (#6685)
This commit will enable `Softmax_001` test
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 19 Aug 2019 08:02:18 +0000 (17:02 +0900)]
Find ignore directory using ls-files (#6676)
Instead of using find command, use git ls-files to find .FORMATDENY file to make format check ignore directory list
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
Vladimir Plazun/AI Tools Lab /SRR/Engineer/삼성전자 [Mon, 19 Aug 2019 07:57:29 +0000 (10:57 +0300)]
[custom op] Update node for new custom op design (#6286)
Now `_id` and `userdata` are stored directly in `CustomNode` as now custom ops are defined using pure C api
- Use string id
- return id by const reference
Signed-off-by: Vladimir Plazun <v.plazun@samsung.com>
채성우/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 19 Aug 2019 07:55:40 +0000 (16:55 +0900)]
[loco] Introduce DepthwiseConv2D Layer (#6662)
This commit introduce DepthwiseConv2D layer and related GraphTestcase.
Related: #6569
Signed-off-by: seongwoo <sw4670.chae@samsung.com>
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 19 Aug 2019 07:38:16 +0000 (16:38 +0900)]
[exo-tflite] Add TensorSoftmax (#6623)
* [exo-tflite] Add TensorSoftmax
This commit will add `TensorSoftmax` in `exo-tflite`
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
* add assert
* add axis assert
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 19 Aug 2019 07:37:29 +0000 (16:37 +0900)]
[moco-tf] Support TFSoftmax Canonicalization (#6672)
* [moco-tf] Support TFSoftmax Canonicalization
This commit will enable TFSoftmax Canonicalization
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
* remove comment
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 19 Aug 2019 07:31:10 +0000 (16:31 +0900)]
[moco-tf] Introduce TFShapeInferenceRule (#6665)
* [moco-tf] Introduce TFShapeInferenceRule
This will introduce empty TFShapeInferenceRule that utilizies existing ShapeInferenceData
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* fix comment, remove override
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 19 Aug 2019 07:06:57 +0000 (16:06 +0900)]
[res] Regression test for Add (#6513)
This commit adds Regression test for Add network with Placeholder and
Const operations as inputs.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
채성우/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 19 Aug 2019 06:40:56 +0000 (15:40 +0900)]
[loco] Introduce DepthwiseFilerEncode layer (#6668)
This commit introduce DepthwiseFilterEncode layer to GraphBuilder.
Signed-off-by: seongwoo <sw4670.chae@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 19 Aug 2019 05:45:19 +0000 (14:45 +0900)]
[moco_tf] Remover override for final method (#6671)
This will remove override keyword for final method that is redundant
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 19 Aug 2019 05:28:44 +0000 (14:28 +0900)]
[locomotiv] Support TensorSoftmax (#6620)
* [locomotiv] Support TensorSoftmax
This commit will enable supporting `TensorSoftmax` in `locomotiv`
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
* Modify domain assert
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 19 Aug 2019 05:05:22 +0000 (14:05 +0900)]
[moco-tf] Introduce Restart PhaseRunner (#6667)
This will introduce Restart type PhaseRunner that restarts from the beginning when there is any change in the transformation list
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 19 Aug 2019 04:51:14 +0000 (13:51 +0900)]
[moco-tf] Update TFConcatV2 IR to have variadic arity (#6663)
This will update TFConcatV2 IR to have variadic arity
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 19 Aug 2019 04:17:57 +0000 (13:17 +0900)]
[loco] Eq operato for NodeShape (#6661)
This will introduce Eq(==) operator for NodeShape
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 19 Aug 2019 02:21:03 +0000 (11:21 +0900)]
[neurun] Write comments for class Index (#6608)
This commit adds comments for class `util::Index`.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
[neurun] Clearify Scheduler::ESTAndExecTime with comments (#6649)
Add comments to clearify return values in case of non-parallel-executor
Signed-off-by: Poshshoev Dilshodzhon <d.poshshoev@samsung.com>
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 19 Aug 2019 01:42:04 +0000 (10:42 +0900)]
[moco-tf] Add TFFormattedGraph for TFSoftmax (#6647)
* [moco-tf] Add TFFormattedGraph for TFSoftmax
This commit will add `TFSoftmax` summary function in `TFFormattedGraph.cpp`
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
* fix name
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 19 Aug 2019 01:12:20 +0000 (10:12 +0900)]
Introduce a map storing layouts into TensorBuilders (#6641)
This commit introduces a map storing layouts for each operand into TensorBuilders.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 19 Aug 2019 01:10:57 +0000 (10:10 +0900)]
Fix to proper check on non-empty string (#6659)
Previously checking whether shell variable is non-empty worked not as
expected. This commit fixes to do this function as expected.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 19 Aug 2019 00:47:21 +0000 (09:47 +0900)]
[moco-tf] GraphBuilder for SquaredDifference (#6631)
* [moco-tf] GraphBuilder for SquaredDifference
This will introduce GraphBuilder for SquaredDifference to import SquaredDifference node
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* fix comment
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Sun, 18 Aug 2019 23:30:57 +0000 (08:30 +0900)]
[locop] Add TensorSoftmax visit function (#6648)
This commit will add `TensorSoftmax` visit function
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Sun, 18 Aug 2019 23:30:08 +0000 (08:30 +0900)]
[res] Remove obsolete TF_SMALL_NET_XXXX tests (#6650)
This commit removes obsolete TF_SMALL_NET_XXXX test cases. Note that
they are copied to NET_XXXX
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
채성우/On-Device Lab(SR)/Engineer/삼성전자 [Sun, 18 Aug 2019 23:29:05 +0000 (08:29 +0900)]
[loco] Introduce constgen layer (#6645)
This commit introduce constgen layer to GraphBuilder.
Signed-off-by: seongwoo <sw4670.chae@samsung.com>
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Sun, 18 Aug 2019 23:28:16 +0000 (08:28 +0900)]
[moco-tf] Enable importing Softmax (#6646)
This commit will enable importing `Softmax` in `moco-tf`
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
[mir] Remove Gemm operation (#6654)
`GemmOp` is almost the same operation as `FullyConnectedOp`.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[mir_onnx] Remove redundant include (#6652)
Remove inclusion of `DropoutOp.h` popped up as a result of merge conflict.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[nnc] Remove support for GemmOp (#6625)
`GemmOp` is almost the same as `FullyConnectedOp`. It will be removed.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
Павел Ильютченко/AI Tools Lab /SRR/Engineer/삼성전자 [Fri, 16 Aug 2019 09:07:57 +0000 (12:07 +0300)]
[mir_onnx] Introduce ConverterContext for using in NodeConverters (#6457)
* Implemented ConverterContext
* Fix NodeConverter interface to accept ConverterContext
* Fix all NodeConverters according to new interface
* Fix ONNX IR version checking.
Signed-off-by: Pavel Iliutchenko <p.iliutchenk@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 16 Aug 2019 05:55:43 +0000 (14:55 +0900)]
[tf2tflite] Use NET_XXXX for test list (#6639)
This commit use NET_XXXX for tf2tflite test listing instead of
TF_SMALL_NET_XXXX
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 16 Aug 2019 05:55:33 +0000 (14:55 +0900)]
[mocotest-tf] Use NET_XXXX for test list (#6638)
This commit uses NET_XXXX for mocotest-tf test list instead of
TF_SMALL_NET_XXXX
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
이상규/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 16 Aug 2019 04:39:51 +0000 (13:39 +0900)]
[nnpackage_run] Introduce tensor_dumper (#6602)
- Introduce `dump` option in args.*
- Introduce tensor_dumper
- Use tensor_dumper in nnpackage_run
Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
이상규/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 16 Aug 2019 04:28:12 +0000 (13:28 +0900)]
[nnpackage_run] Fix wrong allocation size for output (#6640)
outputs[i] is vector<float>. sizeof(float) should not be multiplied.
Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
이상규/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 16 Aug 2019 04:27:08 +0000 (13:27 +0900)]
Add missing copyright for tflite_run's tensor dumper (#6600)
Add copyright in tensor_dumper.cc and tensor_dumper.h .
It has 2018 year since this files were created in 2018.
Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
Ivan Vagin/AI Tools Lab /SRR/Engineer/삼성전자 [Fri, 16 Aug 2019 04:12:49 +0000 (13:12 +0900)]
[neurun] Refactor scheduler a little bit (#6636)
* Simplified makeRank method
* Renamed some variables
* Fixed some typos
Signed-off-by: Ivan Vagin <ivan.vagin@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 16 Aug 2019 03:43:31 +0000 (12:43 +0900)]
[moco-tf] Enable SquaredDifference canonicalizer (#6629)
This will enable SquaredDifference canonicalizer
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 16 Aug 2019 03:43:06 +0000 (12:43 +0900)]
[moco-tf] Enable type inf for TFSquaredDifference (#6628)
This will enable type inference for TFSquaredDifference node
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 16 Aug 2019 03:41:45 +0000 (12:41 +0900)]
[exo-tflite] Enable EltwiseSqrt (#6610)
* [exo-tflite] Enable EltwiseSqrt
This will enable export for EltwiseSqrt node
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* remove for shapeinf
* remove not related new line
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 16 Aug 2019 01:46:33 +0000 (10:46 +0900)]
[res] NET_XXXX to replace SMALL_NET_XXXX (#6621)
This commit adds NET_XXXX tests which copies SMALL_NET_XXXX tests
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 16 Aug 2019 01:43:42 +0000 (10:43 +0900)]
[moco-tf] Register Reshape elimination transforms (#6624)
This commit registers following Reshape elimination transforms to
canonical optimization stage:
- ResolveDuplicateReshape
- ResolveRedundantReshape
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 16 Aug 2019 01:23:20 +0000 (10:23 +0900)]
[moco-tf] Introduce TFDialect for Softmax (#6568)
* [moco-tf] Introduce TFDialect for Softmax
This commit will introduce `TFSoftmax`
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
* add fix transforms
* remove fix_pad
* use logits instead of input
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 16 Aug 2019 00:37:03 +0000 (09:37 +0900)]
Make helper converting functions support NCHW (#6606)
This commit makes converting functions support NCHW.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
Vladimir Plazun/AI Tools Lab /SRR/Engineer/삼성전자 [Fri, 16 Aug 2019 00:35:10 +0000 (03:35 +0300)]
[bug] [custom op] Add back _t suffix for size_t type (#6619)
Was previously removed by error
Signed-off-by: Vladimir Plazun <v.plazun@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 16 Aug 2019 00:34:35 +0000 (09:34 +0900)]
Make cpu KernelGenerator supports NCHW frontend layout (#6605)
This commit makes cpu KernelGenerator supports NCHW frontend layout.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 16 Aug 2019 00:33:54 +0000 (09:33 +0900)]
Make ContantInitializers support NCHW frontend layout (#6555)
This commit makes ContantInitializers support NCHW frontend layout.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
Павел Ильютченко/AI Tools Lab /SRR/Engineer/삼성전자 [Thu, 15 Aug 2019 19:08:05 +0000 (22:08 +0300)]
[mir_onnx] Fix border type in Average pooling (#6637)
* Fix come from debuging InceptionV3
Signed-off-by: Pavel Iliutchenko <p.iliutchenk@samsung.com>
[mir_onnx] Remove Scale converter (#6633)
This operation is not a part of the ONNX specification. It is probably a copy&paste from the Caffe2 importer.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[nnc] Remove batchnorm code snippet (#6626)
It was left behind when the rest of the support of `BatchNormOp` was removed.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 14 Aug 2019 12:01:47 +0000 (21:01 +0900)]
[neurun] Refine neurun-core CMakeLists (#6635)
Refine `target_link_libraries`.
- One target per line
- Remove duplicated lib - `dl`
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 14 Aug 2019 11:46:21 +0000 (20:46 +0900)]
[neurun] Let xdata and xprobe be installed (#6634)
Update `CMakeLists.txt` to install xdata and xprobe.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 14 Aug 2019 11:22:58 +0000 (20:22 +0900)]
[neurun] Introduce CMakeLists.txt for test lib (#6615)
This commit moves test library build from `neurun/CMakeLists.txt` to
`neurun/test/CMakeLists.txt` which is newly created.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 14 Aug 2019 11:22:11 +0000 (20:22 +0900)]
[neurun] Show dlerror on dlopen/dlsym failure (#6632)
Show dlerror on dlopen/dlsym failure
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
채성우/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 14 Aug 2019 10:15:46 +0000 (19:15 +0900)]
[loco] add ctor of DepthwiseFilter PermutingEncoder (#6627)
This commit adds ctor of DepthwiseFilter PermutingEncoder.
Signed-off-by: seongwoo <sw4670.chae@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 14 Aug 2019 10:05:34 +0000 (19:05 +0900)]
[moco-tf] Canonicalizer for TFSquaredDifference (#6612)
* [moco-tf] Canonicalizer for TFSquaredDifference
This will introduce Canonicalizer for TFSquaredDifference node
For now this will decompose to EltwiseSub and EltwiseMul
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* apply comments