남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 7 Aug 2019 00:03:26 +0000 (09:03 +0900)]
[moco-tf] Introduce function to create BiasAdd operation (#6292)
This commit will introduce a new function to create `BiasAdd` operation
and insert it after preceding operation.
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
Add default fallback for Visitor::visit (#5889)
This will help to escape overriding a lot of visit()
with default body: empty body or `throw std::runtime_error("NYI")`:
just override this new method
additional changes:
* added file with all mir operation headers
* refined IVisitor
Signed-off-by: Dilshodzhon Poshshoev <d.poshshoev@samsung.com>
[mir_onnx] Replace BiasAdd and Scale with Elementwise equivalents (#6298)
`BiasAdd` and `Scale` are restricted versions of equivalent Elementwise ops and are going to be removed.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[mir2loco] Replace BiasAdd and Scale with Elementwise equivalents (#6296)
`BiasAdd` and `Scale` are restricted versions of equivalent Elementwise ops and are going to be removed.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[mir_caffe] Replace BiasAdd and Scale with elementwise equivalents (#6295)
`BiasAdd` and `Scale` are restricted versions of equivalent Elementwise ops and are going to be removed.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[mir_tflite] Replace BiasAdd and Scale with Elementwise equivalents (#6294)
`BiasAdd` and `Scale` are restricted versions of equivalent Elementwise ops and are going to be removed.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[mir_caffe2] Replace BiasAdd and Scale with Elementwise equivalents (#6293)
`BiasAdd` and `Scale` are restricted versions of equivalent Elementwise ops and are going to be removed.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[mir_onnx] Add NodeConverters for all supported operations from OpCreator (#6291)
* Added OpConverters
* Register all converters
* Remove OpCreator
Signed-off-by: Pavel Iliutchenko <p.iliutchenk@samsung.com>
[mir_onnx] Introduce NodeConverterRegistry and remove operation hash (#6241)
* Introduce NodeConverterRegistry
* Remove operation hash
* Replace operation hash uses with NodeConverter
Signed-off-by: Pavel Iliutchenko <p.iliutchenk@samsung.com>
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 6 Aug 2019 10:06:56 +0000 (19:06 +0900)]
[moco-tf] Divide into function for fused convolution node in FuseBinaryIntoPreceding (#6273)
* [moco-tf] Divide into function for fused convolution node in FuseBinaryIntoPreceding
This commit will introduce `fused_conv_node` function for dividing long code.
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
* add missing code
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 6 Aug 2019 09:37:51 +0000 (18:37 +0900)]
[loco] Use link helper (#6271)
Let's use link helper instead of deprecated node method.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 6 Aug 2019 09:36:10 +0000 (18:36 +0900)]
[loco] Update TensorShaped classes with initializer_list (#6284)
Now, "shape" setter in TensorShaped mixin takes initializer_list
as its input.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 6 Aug 2019 09:33:46 +0000 (18:33 +0900)]
[mocotest-tf] Register Squeeze tests (#6265)
This commit register and enable valid Squeeze tests for moco tf
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 6 Aug 2019 09:33:30 +0000 (18:33 +0900)]
[res] Add negative squeeze dimension test case (#6262)
This commit add Squeeze network test case with negative squeeze dimension
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 6 Aug 2019 09:33:07 +0000 (18:33 +0900)]
[moco-tf] Extract common Pad calculation method (#6252)
* [moco-tf] Extract common Pad calculation method
This will extract common Pad calculation methods for Canonical and TF dialect nodes
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* use const
* use int64_t
* add guard
이상규/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 6 Aug 2019 08:38:29 +0000 (17:38 +0900)]
nnpackager will verify when verify option is given (#6276)
nnpackger always verified. Now, it verifies if verify is given.
Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 6 Aug 2019 08:16:14 +0000 (17:16 +0900)]
[loco] Implement AvgPool2D shape inference (#6251)
* [loco] Implement AvgPool2D shape inference
This commit extends CanonicalShapeInferenceRule to accept AvgPool2D nodes.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
* Update testcase
* Introduce FeatureShapeUpdater
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 6 Aug 2019 08:14:15 +0000 (17:14 +0900)]
[moco-tf] Use public member tbl() for Squeeze log (#6282)
This commit use public tbl() member funciton instead of to-be-deprecated
private member _tbl
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 6 Aug 2019 08:13:10 +0000 (17:13 +0900)]
[loco] TypeInferenceRule to use inference rules of multiple dialects (#6198)
* [loco] TypeInferenceRule to use inference rules of multiple dialects
This commit adds MultiDialectTypeInferenceRule that uses inference rules of multiple dialects.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* using loco::link(..)
* using loco::link(..) for output too
* assert
* bind method
* struct to class
* //// -> ///
* modifying assert
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 6 Aug 2019 08:10:22 +0000 (17:10 +0900)]
[exo-tflite] Exporting custom op (#6269)
This commit enables exportation of custom op (Operator Codes, input, output, op with attr) into tflite file.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 6 Aug 2019 08:09:41 +0000 (17:09 +0900)]
[moco-tf] Introduce TFSub IR (#6275)
This will introduce TFSub IR and required implementations
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
이춘석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 6 Aug 2019 06:04:23 +0000 (15:04 +0900)]
Update TFLITE_MINOR_VER to 13 (#6274)
- Tizen has tensorflow lite version 1.13.1
- Update TFLITE_MINOR_VER to 13 in nnfw.spec
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 6 Aug 2019 05:34:15 +0000 (14:34 +0900)]
Make to distinguish frontend layout from backend layout in Graph (#6253)
This commit makes to distinguish frontend layout fron backend layout in Gragh.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 6 Aug 2019 04:52:34 +0000 (13:52 +0900)]
[moco-tf] Introduce TFRealDiv IR (#6270)
* [moco-tf] Introduce TFRealDiv IR
This will introduce TFRealDiv IR and required implementations
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* fix comment
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 6 Aug 2019 03:53:56 +0000 (12:53 +0900)]
[exo-tflite] Export EltwiseSub and EltwiseDiv (#6263)
This will enable export of EltwiseSub and EltwiseDiv node
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 6 Aug 2019 03:53:31 +0000 (12:53 +0900)]
[exo-tflite] Shape inf. for EltwiseSub and EltwiseDiv (#6264)
This will enable shape inference of EltwiseSub and EltwiseDiv node
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 6 Aug 2019 03:39:05 +0000 (12:39 +0900)]
[moco-tf] Enable Squeeze Canonicalizer (#6261)
This commit register and enables SqueezeCanonicalizer
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 6 Aug 2019 03:32:31 +0000 (12:32 +0900)]
[loco] Shape Inference for Eltwise Arithmetic Nodes (#6256)
This commit implements shape inference for eltwise arithmetic nodes.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 6 Aug 2019 01:54:21 +0000 (10:54 +0900)]
[locomotiv] Support EltwiseDiv (#6258)
This will support EltwiseDiv node
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 6 Aug 2019 01:54:01 +0000 (10:54 +0900)]
[locomotiv] Support EltwiseSub (#6257)
This will support EltwiseSub node
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 6 Aug 2019 01:53:02 +0000 (10:53 +0900)]
Update replaced scripts path (#6255)
Update replaced scripts path reference in script or document: scripts/standalone -> infra/scripts
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 6 Aug 2019 01:46:21 +0000 (10:46 +0900)]
[moco-tf] Squeeze Canonicalizer (#6234)
Squeeze canonicalizer is to canonicalize TF-dialect TFSqueeze into
canonical FixedReshape.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 6 Aug 2019 01:45:47 +0000 (10:45 +0900)]
[exo-tflite] Convert FeatureShape to ShapeDescription (#6254)
This commit introduces "to_shape_description" helper which constructs
ShapeDescription (which exo-tflite internally uses) from FeatureShape.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
이춘석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 6 Aug 2019 00:36:11 +0000 (09:36 +0900)]
Upgrade Arm Compute Lib to 19.05 (#5611)
- upgrade Arm Compute Library to v19.05
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 6 Aug 2019 00:16:58 +0000 (09:16 +0900)]
[moco-tf] Use loco::link helper (#6228)
Let's use loco::link helper instead of deprecated node method!
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 5 Aug 2019 23:25:10 +0000 (08:25 +0900)]
[moco-tf] Fix pad for TFAvgPool in fix shape (#6225)
This will merge fix pad for TFAvgPool node in fix shape transform
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 5 Aug 2019 23:23:19 +0000 (08:23 +0900)]
[moco-tf] Introduce FuseType and divide functions using it (#6227)
This commit will introduce `FuseType` and divide functions using `FuseType`
for each case
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 5 Aug 2019 23:22:18 +0000 (08:22 +0900)]
[moco-tf] Fix pad for MaxPool2D in fix shape (#6223)
This will merge fix pad for MaxPool2D node in fix shape transform
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 5 Aug 2019 23:21:59 +0000 (08:21 +0900)]
[loco] EltwiseDiv IR (#6229)
This will introduce canonical EltwiseDiv IR
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 5 Aug 2019 23:21:46 +0000 (08:21 +0900)]
[exo-tflite] Registering custom op before exporting op code table (#6235)
This adds custom op before exporting op code table.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 5 Aug 2019 23:17:51 +0000 (08:17 +0900)]
[res] RealDiv_000 (#6233)
This will introduce TensorFlow RealDiv_000 test model with one RealDiv node
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 5 Aug 2019 23:17:11 +0000 (08:17 +0900)]
[res] TF Sub_000 (#6231)
This will introduce Sub_000 TensorFlow test model with one Sub node
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
[mir] Remove deserializer header (#6249)
This is a leftover after removing stale serialization support.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
Selecting all helper functions in a separate file (#6237)
* Move all helper functions in one file
* Fix other code
Signed-off-by: Pavel Iliutchenko <p.iliutchenk@samsung.com>
[mir2loco] Rename DTYPE to DataType (#6247)
Replace deprecated `DTYPE` with `DataType`.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[nnc] Rename DTYPE to DataType (#6248)
Replace deprecated `DTYPE` with `DataType`.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[mir_tflite] Rename DTYPE to DataType (#6246)
Replace deprecated `DTYPE` with `DataType`.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[mir_caffe2] Rename DTYPE to DataType (#6245)
Replace deprecated `DTYPE` with `DataType`.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[mir_caffe] Rename DTYPE to DataType (#6244)
Replace deprecated `DTYPE` with `DataType`.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[mir_tflite] Rename namespace nnc to mir_tflite in TFLite importer (#6240)
* Rename namespace and include guards in TFLite importer
* Fix usages of the namespace in `nnc`
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[mir_caffe] Rename namespace nnc to mir_caffe in Caffe importer (#6239)
* Rename namespace and include guards in Caffe importer
* Fix usages of the namespace in `nnc`
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[mir_caffe2] Rename namespace nnc to mir_caffe2 in Caffe2 importer (#6238)
* Rename namespace and include guards in Caffe2 importer
* Fix usages of the namespace in `nnc`
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[mir] Rename DTYPE to DataType (#6243)
Rename enum class `DTYPE` to `DataType` to conform to the naming rules.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
Rename ncc namespace into mir_onnx in ONNX importer (#6236)
* Rename namespace in importer
* Fix onnx_frontend in NNC
Signed-off-by: Pavel Iliutchenko <p.iliutchenk@samsung.com>
ACL NEON Strided Slice (#5946)
Add support of Strided Slice in ACL NEON
Signed-off-by: Sizov Nikita <n.sizov@samsung.com>
[mir2loco] Support Conv2D operation (#5899)
* Implemented `mir::Conv2DOp` transformation to `loco::Conv2D`
* Added test for this
Signed-off-by: Pavel Iliutchenko <p.iliutchenk@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 5 Aug 2019 11:12:18 +0000 (20:12 +0900)]
Add layout as a condition for merging subgraphs (#6216)
This commit adds layout as a condition for merging subgraph.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
[mir] Remove Deserializer unittest (#6178)
This is a leftover after removing serialization support.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[mir] Normalize variable names (#6179)
Normalize variable names:
* `_snake_case` for private members;
* `snake_case` for local variables (and parameters).
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 5 Aug 2019 11:02:54 +0000 (20:02 +0900)]
[mir2loco] Use link helper (#6211)
This commit revises mir2loco to use link helper function instead of
deprecated node method in GraphInput and GraphOutput.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 5 Aug 2019 10:20:59 +0000 (19:20 +0900)]
[res] Add selective Squeeze test (#6230)
Add Squeeze test that has selective squeeze dimension
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
이상규/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 5 Aug 2019 09:48:01 +0000 (18:48 +0900)]
[nnpackage] Introduce nnpackager (#6145)
This PR introduces nnpackager, which provides verify function only.
It checks:
- the existence of nnpackager root directory
- the existence of MANIFEST
- the validity of json format
- the existence of models file
Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 5 Aug 2019 09:40:03 +0000 (18:40 +0900)]
[moco-tf] Fix pad for Conv2D in fix shape (#6221)
This will merge fix pad for Conv2D node in fix shape transform
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 5 Aug 2019 09:39:42 +0000 (18:39 +0900)]
[moco-tf] Enable import as TFRelu and TFRelu6 (#6220)
This will enable import as TRRelu and TFRelu6
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 5 Aug 2019 09:38:03 +0000 (18:38 +0900)]
[loco] Implement ReLU6 shape inference (#6222)
This commit extends CanonicalShapeInferenceRule to accept ReLU6 node.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 5 Aug 2019 09:28:21 +0000 (18:28 +0900)]
[res/TensorFlowTests] renaming compilerInfo.pbtxt to customop.conf (#6215)
[res/TensorFlowTests] renaming compilerInfo.pbtxt to customop.confi
Two compilerInfo.pbtxt files were renamed to customop.conf.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 5 Aug 2019 08:50:01 +0000 (17:50 +0900)]
Rename nnapi-header (#6199)
Rename nnapi-header interface to nnfw-header
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 5 Aug 2019 08:47:59 +0000 (17:47 +0900)]
Use option for coverage and test build flag (#6224)
Use cmake option to set ENABLE_TEST and ENABLE_COVERAGE
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 5 Aug 2019 08:30:40 +0000 (17:30 +0900)]
[moco-tf] Give chance for negative squeeze dim shape inference (#6206)
Previously shape inference for TFSqueeze aborts when it encounters
negative squeeze dimension. From this commit, shape inference,
'fix_shape()', for TFSqueeze just return false, not abort program
to give a chance for later time, possibly after negative dimension
resolved.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 5 Aug 2019 08:17:26 +0000 (17:17 +0900)]
[moco-tf] Fix pad for AvgPool2D in fix shape (#6196)
* [moco-tf] Fix pad for AvgPool2D in fix shape
This will merge fix pad for AvgPool2D node in fix shape transform
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* remove local var only for assert
* remove useless assert
* remove unused header
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 5 Aug 2019 08:13:56 +0000 (17:13 +0900)]
[locoex-customop] Use link helper (#6208)
Let's use link helper instead of deprecated node method.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 5 Aug 2019 08:13:42 +0000 (17:13 +0900)]
[moco-onnx] Use link helper (#6209)
Let's use link helper instead of deprecated node method.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 5 Aug 2019 08:13:18 +0000 (17:13 +0900)]
[moco-tf] Import Squeeze (#6205)
* [moco-tf] Import Squeeze
This commit adds import stage for tensorflow Squeeze operation.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
* Comments on what to test
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 5 Aug 2019 07:53:58 +0000 (16:53 +0900)]
Introduce interface for strict build and coverage (#6207)
Introduce interface nnfw_common and nnfw_coverage in nnfw cmake script
Introduce cmake option variable for strict build and variable setting for coverage build
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 5 Aug 2019 07:52:36 +0000 (16:52 +0900)]
[moco-tf] Canonicalize TFRelu and TFRelu6 (#6217)
This will enable TFRelu and TFRelu6 canonicalization
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 5 Aug 2019 07:52:23 +0000 (16:52 +0900)]
[moco-tf] Revise comments of FuseBinaryIntoPreceding (#6204)
* [moco-tf] Revise comments of FuseBinaryIntoPreceding
This commit will revise some comments of `FuseBinaryIntoPreceding.cpp`
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
* Apply comments
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 5 Aug 2019 07:35:00 +0000 (16:35 +0900)]
[loco] EltwiseSub IR (#6203)
* [loco] EltwiseSub IR
This will introduce canonical EltwiseSub IR
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* todo cmt for shape inference
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 5 Aug 2019 07:33:07 +0000 (16:33 +0900)]
[loco-exporter] Use link helper (#6212)
Let's use link helper instead of deprecated node method.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 5 Aug 2019 07:12:01 +0000 (16:12 +0900)]
[moco-tf] Fix comments for Relu, Relu6 (#6202)
This will fix comments for Relu and Relu6 nodes
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 5 Aug 2019 07:07:54 +0000 (16:07 +0900)]
[moco-tf] Canonicalize TFRelu (#6195)
This will add ReluCanonicalizer that transforms TFRelu node to canonical ReLU node
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 5 Aug 2019 07:05:09 +0000 (16:05 +0900)]
[moco-tf] Hide TFNodeSummaryBuilderBase _tbl (#6168)
* [moco-tf] Hide TFNodeSummaryBuilderBase _tbl
This will change TFNodeSummaryBuilderBase member _tbl to be private to prevent unintended problems
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* add warning message
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 5 Aug 2019 06:48:14 +0000 (15:48 +0900)]
Fix pylint warning in res (#6210)
Fix pylint warning by python files in res
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 5 Aug 2019 06:05:35 +0000 (15:05 +0900)]
[locomotiv] Use link helper (#6197)
Let's use link helper instead of deprecated "node" method in GraphInput/Output.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 5 Aug 2019 06:03:25 +0000 (15:03 +0900)]
[exo-tflite] Use link helper (#6200)
This commit revises exo-tflite to use link helper instead of
deprecated "node" method in GraphInput/GraphOutput.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 5 Aug 2019 04:53:29 +0000 (13:53 +0900)]
[moco-tf] Canonicalize TFRelu6 (#6193)
This will add Relu6Canonicalizier that transforms TFRelu6 node to canonical ReLU6 node
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 5 Aug 2019 04:42:25 +0000 (13:42 +0900)]
Change the way getting layout of node in PermutationInsertionPass (#6148)
This commit changes the way getting layout of node in PermuttationInsertionPass from model::Subgraph to operation::LowerInfo.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 5 Aug 2019 04:42:13 +0000 (13:42 +0900)]
[loco] Implement FeatureCodec shape inference (#6191)
This commit extends CanonicalShaapeInferenRule to accept
FeatureEncode and FeatureDecode nodes.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 5 Aug 2019 04:42:00 +0000 (13:42 +0900)]
Remove softlink directory (#6188)
Remove softlink directory scripts/standalone - made for CI infra workaround
Update format check script
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 5 Aug 2019 04:38:11 +0000 (13:38 +0900)]
[moco-tf] Import as TFRelu (#6182)
This will change import Relu node as TFRelu from the knob
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 5 Aug 2019 04:33:26 +0000 (13:33 +0900)]
[loco] Introduce "link" helper (#6190)
This commit introdues "link" helper which associates
GraphInput/GraphOutout with Pull/Push node, respectively.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 5 Aug 2019 04:16:54 +0000 (13:16 +0900)]
Format check hpp extension (#6187)
Add format check extension: hpp file
Fix format: nnfw_api.hpp
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 5 Aug 2019 04:10:04 +0000 (13:10 +0900)]
[moco-tf] Import as TFRelu6 (#6183)
This will change import Relu6 node as TFRelu6 from the knob
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 5 Aug 2019 04:01:21 +0000 (13:01 +0900)]
[moco-tf] LogHelper to dump Pad and PadData (#6185)
This will introduce LogHelper to dump Pad and PadData values
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 5 Aug 2019 03:53:13 +0000 (12:53 +0900)]
[tflite_benchmark] Insert pause between iterations (#6189)
tflite_benchmark now allows users to insert pause between iterations via
"PAUSE" environment variable.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 5 Aug 2019 03:30:20 +0000 (12:30 +0900)]
[moco-tf] Revise COpCall test (#6184)
This will revisit CopCall node test to remove relation with graph connections
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 5 Aug 2019 02:53:19 +0000 (11:53 +0900)]
[enco] Fix copyright year (#6174)
This commit fixes copyright year for 2019 generated file
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 5 Aug 2019 02:50:39 +0000 (11:50 +0900)]
[loco] Implement Relu shape inference (#6161)
* [loco] Implement Relu shape inference
CanonicalShapeInferenceRule now properly infers the shape of relu node.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
* Add final
이성재/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 5 Aug 2019 01:52:59 +0000 (10:52 +0900)]
Update scheduled deprecated API. (#6163)
* Change old API `gfile.FastGFile` to `tf.gfile.GFile`
Signed-off-by: Sung-Jae Lee <sj925.lee@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 5 Aug 2019 01:32:35 +0000 (10:32 +0900)]
Use external volume mount in standalone scripts (#6136)
It enables using external volume mount in CI test infra
- Reduce external code download overhead
- Remove pre-built acl binary dependency (especially on version up)
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 5 Aug 2019 00:59:03 +0000 (09:59 +0900)]
[moco-tf] Summary for TFSqueeze (#6162)
* [moco-tf] Summary for TFSqueeze
This commit introduces log summary for TFSqueeze
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
* Make use of pepper::str