Hanjoung Lee [Tue, 25 Aug 2020 03:45:33 +0000 (12:45 +0900)]
[onert] Set dyn_tensor_manager for static tensors (#3951)
Set `dynamic_tensor_manager` for static tensors in controlflow backend.
This is needed as any static tensors could turn into dynamic.
% Same approach has been applied for cpu backend, but not for controflow
yet.
ONE-DCO-1.0-Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
SaeHie Park [Tue, 25 Aug 2020 02:16:11 +0000 (11:16 +0900)]
[luci] Extract export_node to reduce class LOC 2 (#3960)
This will revise OperationExporter with extracting class methods to functions to reduce class LOC
ONE-DCO-1.0-Signed-off-by: SaeHie Park <saehie.park@gmail.com>
Hanjoung Lee [Tue, 25 Aug 2020 02:01:08 +0000 (11:01 +0900)]
[nnfwapi] Simple WHILE test (#3953)
Add simple WHILE tests. As WHILE alone cannot make a meaningful test,
Add and Less operator are also used.
Introduce multi-subgraph support for `CircleGen`.
ONE-DCO-1.0-Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
seongwoo chae [Tue, 25 Aug 2020 01:56:19 +0000 (10:56 +0900)]
[TensorFlowLiteRecipes] Add test.rule to Net_TConv_BN_000 (#3959)
This commit adds test.rule to Net_TConv_BN_000.
ONE-DCO-1.0-Signed-off-by: seongwoo <mhs4670go@naver.com>
seongwoo chae [Tue, 25 Aug 2020 01:56:07 +0000 (10:56 +0900)]
[luci] Add bias attribute to CircleTrasnposeConv IR (#3958)
This commit adds bias attribute to CircleTransposeConv IR.
ONE-DCO-1.0-Signed-off-by: seongwoo <mhs4670go@naver.com>
SaeHie Park [Tue, 25 Aug 2020 01:14:56 +0000 (10:14 +0900)]
[luci] Extract export_node to reduce class LOC 1 (#3957)
This will revise OperationExporter with extracting class methods to functions to reduce class LOC
ONE-DCO-1.0-Signed-off-by: SaeHie Park <saehie.park@gmail.com>
seongwoo chae [Tue, 25 Aug 2020 00:28:47 +0000 (09:28 +0900)]
[tflite2circle] Support SparsityParameters (#3947)
This commit enables tflite2circle to support SparsityParameters.
ONE-DCO-1.0-Signed-off-by: seongwoo mhs4670go@naver.com
Hyukjin Jeong [Tue, 25 Aug 2020 00:28:21 +0000 (09:28 +0900)]
[circle-quantizer] Support requantization from int8 to uint8 (#3950)
This supports requantization from int8 to uint8
ONE-DCO-1.0-Signed-off-by: Hyukjin Jeong <hj1.jeong@samsung.com>
SaeHie Park [Tue, 25 Aug 2020 00:23:34 +0000 (09:23 +0900)]
[luci] Shape inf common for Pad (#3955)
This will provide common methods for MirrorPad, Pad and PadV2 nodes to reduce duplicate codes
ONE-DCO-1.0-Signed-off-by: SaeHie Park <saehie.park@gmail.com>
SaeHie Park [Tue, 25 Aug 2020 00:23:13 +0000 (09:23 +0900)]
[luci] Extract export class methods to functions (#3954)
This will extract export class methods to functions to reduce class LOC
ONE-DCO-1.0-Signed-off-by: SaeHie Park <saehie.park@gmail.com>
seongwoo chae [Mon, 24 Aug 2020 21:16:32 +0000 (06:16 +0900)]
[luci-interpreter] Add assert condition of CircleTransposeConv (#3940)
This commit adds assert condition of CircleTransposeConv.
ONE-DCO-1.0-Signed-off-by: seongwoo <mhs4670go@naver.com>
Alexander Moiseev [Mon, 24 Aug 2020 12:42:59 +0000 (15:42 +0300)]
[luci-interpreter] Add support for sqrt op (#3904)
Add support for sqrt element-wise op in luci-interpreter.
ONE-DCO-1.0-Signed-off-by: Alexander Moiseev <a.moiseev@samsung.com>
Hyeongseok Oh [Mon, 24 Aug 2020 11:30:05 +0000 (20:30 +0900)]
[onert/test] Operation backend test on one_op_tests (#3939)
- Introduce backend setting method in GenModelTest
- Backend test on one_op_tests
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
Jang Jiseob [Mon, 24 Aug 2020 07:49:21 +0000 (16:49 +0900)]
[onert] Unify element-wise unary operations to ElementwiseUnary. (#3933)
This commit unifies element-wise activation operations to ElementwiseActivation for reducing code duplication.
Signed-off-by: ragmani <ragmani0216@gmail.com>
Hyeongseok Oh [Mon, 24 Aug 2020 07:38:44 +0000 (16:38 +0900)]
[onert] Copyright fix (#3945)
- Add copyright to auto generated files
- Fix copyright checker to ignore NNAPI file
- Fix copyright checker to allow (year-year)
- Add copyright to ported code with small fix
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
seongwoo chae [Mon, 24 Aug 2020 07:37:45 +0000 (16:37 +0900)]
[tools] Introduce github release tool (#3900)
This commit introduces github release tool that helps you to automate the release.
ONE-DCO-1.0-Signed-off-by: seongwoo <mhs4670go@naver.com>
SaeHie Park [Mon, 24 Aug 2020 07:32:43 +0000 (16:32 +0900)]
[luci] SummaryBuilder reduce LOC 2 (#3946)
This will relocate implementation of summary builder to reduce class LOC
ONE-DCO-1.0-Signed-off-by: SaeHie Park <saehie.park@gmail.com>
SaeHie Park [Mon, 24 Aug 2020 06:36:24 +0000 (15:36 +0900)]
[luci] Shape inf reduce LOC S- (#3944)
This will relocate implementation of shape inference to reduce class LOC
ONE-DCO-1.0-Signed-off-by: SaeHie Park <saehie.park@gmail.com>
SaeHie Park [Mon, 24 Aug 2020 06:11:36 +0000 (15:11 +0900)]
[luci] SummaryBuilder reduce LOC 1 (#3942)
This will relocate implementation of summary builder to reduce class LOC
ONE-DCO-1.0-Signed-off-by: SaeHie Park <saehie.park@gmail.com>
SaeHie Park [Mon, 24 Aug 2020 06:06:27 +0000 (15:06 +0900)]
[luci] Shape inf reduce LOC M-R (#3941)
This will relocate implementation of shape inference to reduce class LOC
ONE-DCO-1.0-Signed-off-by: SaeHie Park <saehie.park@gmail.com>
YiHyunJin [Mon, 24 Aug 2020 06:04:33 +0000 (15:04 +0900)]
[onert] Fix conv2d op for dilation (#3932)
When using MultiThread, both dilation data must be 1.
When obtaining padding data, dilation data is used.
Move additional conditions of dilation into usableMultiThreaded
Signed-off-by: YiHyunjin <hj0412.yi@samsung.com>
Hanjoung Lee [Mon, 24 Aug 2020 05:58:32 +0000 (14:58 +0900)]
[nnapi] Unify comparison ops in OperationFactory (#3875)
Unify comparison ops in OperationFactory
ONE-DCO-1.0-Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
SaeHie Park [Mon, 24 Aug 2020 04:48:30 +0000 (13:48 +0900)]
[luci] Shape inf reduce for virtual outputs (#3938)
This will relocate implementation of shape inference to reduce class LOC
ONE-DCO-1.0-Signed-off-by: SaeHie Park <saehie.park@gmail.com>
Jang Jiseob [Mon, 24 Aug 2020 04:37:46 +0000 (13:37 +0900)]
[onert] Unify element-wise activation operations to ElementwiseActivation for reducing code duplication (#3901)
This commit unifies element-wise activation operations to ElementwiseActivation for reducing code duplication.
Signed-off-by: ragmani <ragmani0216@gmail.com>
Sanggyu Lee [Mon, 24 Aug 2020 02:44:15 +0000 (11:44 +0900)]
[tflitefile_tool] select_operator supports Select op (#3915)
select_operator will support Select op.
ONE-DCO-1.0-Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
Hanjoung Lee [Mon, 24 Aug 2020 02:40:52 +0000 (11:40 +0900)]
[onert] Remove ITensorBuilder::supportDynamicTensor (#3929)
Remove `ITensorBuilder::supportDynamicTensor`.
This is as adding dynamic alloc info for all has no harm, and eventually
adding dynamic alloc info will be removed.
ONE-DCO-1.0-Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
Jang Jiseob [Mon, 24 Aug 2020 02:27:55 +0000 (11:27 +0900)]
[onert] Move the time of choosing kernel to configure (#3924)
This commit moves the time of choosing kernel to configure in some cpu layers
- cpu layers : BinaryArithmeticLayer, ElementwiseBinaryLayer, PoolLayer, ReduceLayer
Signed-off-by: ragmani <ragmani0216@gmail.com>
Hyeongseok Oh [Mon, 24 Aug 2020 02:26:09 +0000 (11:26 +0900)]
[onert] Copy and package armcompute only in overlay (#3885)
Copy to installed path and pacakge arm compute library only from overlay (ignore boost build)
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
SaeHie Park [Mon, 24 Aug 2020 01:56:19 +0000 (10:56 +0900)]
[luci] Shape inf reduce LOC D-M (#3937)
This will relocate implementation of shape inference to reduce class LOC
ONE-DCO-1.0-Signed-off-by: SaeHie Park <saehie.park@gmail.com>
Hyeongseok Oh [Mon, 24 Aug 2020 01:55:42 +0000 (10:55 +0900)]
[onert/test] Update nnpackage test list (#3928)
Update nnpackage test list file
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
Hanjoung Lee [Mon, 24 Aug 2020 01:52:02 +0000 (10:52 +0900)]
[nnfwapi] Improve GenTest (#3931)
- Aggregate variables into `_test_data` that test makers must generate
- Support multiple test cases
ONE-DCO-1.0-Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
SaeHie Park [Mon, 24 Aug 2020 01:15:50 +0000 (10:15 +0900)]
[luci] Relocate shape inf common functions (#3936)
This will relocate shape inference common functions to upper side of the file
ONE-DCO-1.0-Signed-off-by: SaeHie Park <saehie.park@gmail.com>
SaeHie Park [Mon, 24 Aug 2020 01:06:37 +0000 (10:06 +0900)]
[luci] Import validate helpers to reduce dup codes (#3920)
This will introduce ValidateHelpers to reduce duplicate codes
ONE-DCO-1.0-Signed-off-by: SaeHie Park <saehie.park@gmail.com>
SaeHie Park [Mon, 24 Aug 2020 00:35:30 +0000 (09:35 +0900)]
[luci] Shape inf reduce LOC A-B (#3935)
This will relocate implementation of shape inference to reduce class LOC
ONE-DCO-1.0-Signed-off-by: SaeHie Park <saehie.park@gmail.com>
Alexander Moiseev [Sun, 23 Aug 2020 21:21:22 +0000 (00:21 +0300)]
[circledump] No opt comment for NonMaxSupressionV5 (#3796)
Add comment about no option in NonMaxSuppressionV5.
ONE-DCO-1.0-Signed-off-by: Alexander Moiseev <a.moiseev@samsung.com>
dr-venkman [Fri, 21 Aug 2020 13:44:41 +0000 (22:44 +0900)]
[luci/tests] Add Unique operation (#3919)
Parent Issue : #2568
Draft PR : #2829
This commit will add unique operation to luci tests
ONE-DCO-1.0-Signed-off-by: venkat.iyer <venkat.iyer@samsung.com>
Sanggyu Lee [Fri, 21 Aug 2020 11:03:42 +0000 (20:03 +0900)]
[tflitefile_tool] support dilation in select_operator.py (#3923)
Currently select_operator ignores dilation factors.
Now, it will handle them.
ONE-DCO-1.0-Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
Hyeongseok Oh [Fri, 21 Aug 2020 07:29:40 +0000 (16:29 +0900)]
[onert/test] Split GenModelTests.cc (#3911)
* [onert/test] Split GenModelTests.cc
- Header file for GenModelTest class
- Divide test for each operation
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
* Rename directory and header
OneOp test directory: one_op_tests
Header file name: GenModelTest.h
Hanjoung Lee [Fri, 21 Aug 2020 07:07:34 +0000 (16:07 +0900)]
[onert] Remove DynamicTensorMan from DynAllocInfo (#3903)
Now that there is this method `ITensor::dynamic_tensor_manager`, no need
to have `DynamicTensorManager` in `DynAllocInfo`.
% Eventually `DynAllocInfo` will be removed.
ONE-DCO-1.0-Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
seongwoo chae [Fri, 21 Aug 2020 05:54:04 +0000 (14:54 +0900)]
[common-artifacts] Remove redundant space (#3927)
This commit removes redundant space in cmake file.
ONE-DCO-1.0-Signed-off-by: seongwoo <mhs4670go@naver.com>
struss [Fri, 21 Aug 2020 05:52:22 +0000 (14:52 +0900)]
[luci-interpreter]Update to use Tensorflow 2.3.0 on luci-interpreter. (#3857)
* [luci-interpreter]Update to use Tensorflow 2.3.0 on luci-interpreter.
This commit update to use Tensorflow 2.3.0 on luci-interpreter.
1. Update cmake file in luci-interpreter kernel to include some file in Tensorflow(Mean kernel use this.)
2. Update kernels for changed spec from 2.1.0 to 2.3.0.(`LeakyRelu`, `Mul`)
`LeakyRelu` : Arguments are changed.
`Mul` : kernel changed to reference to optimized.
3. Update Kernel that accept more tensor.(handle 3~4 Tensor both).`TransposeConv`
currently, accept nullptr on 4th Tensor. it will update later.
ONE-DCO-1.0-Signed-off-by: KiDeuk Bang <rrstrous@nate.com>
* Update Testcases on TransposeConv.
ONE-DCO-1.0-Signed-off-by: KiDeuk Bang <rrstrous@nate.com>
SaeHie Park [Fri, 21 Aug 2020 04:54:43 +0000 (13:54 +0900)]
[locomotiv] Reduce class codes D (#3917)
This will rename methods to reduce class lines of codes of NodeExecution
ONE-DCO-1.0-Signed-off-by: SaeHie Park <saehie.park@gmail.com>
Sanggyu Lee [Fri, 21 Aug 2020 03:42:17 +0000 (12:42 +0900)]
[tflitefile_tool] select_operator will show unsupported op info (#3912)
Currently select_operator does not show any information about
what operator is not supported.
It will print BuiltinOptions index to find out easily what operator is
not supported.
ONE-DCO-1.0-Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
Sanggyu Lee [Fri, 21 Aug 2020 03:42:02 +0000 (12:42 +0900)]
[nnpkg_run] use same format for dumping boolean (#3910)
nnpackage_run has stored bool as I8.
It should be same format (=U8) of python and nnpackage_run load.
ONE-DCO-1.0-Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
Hyeongseok Oh [Fri, 21 Aug 2020 03:36:36 +0000 (12:36 +0900)]
[onert] Enable L2Norm and PadV2 operation loading (#3887)
- Enable L2Normalization and PadV2 operation in base_loader.h
- Add model generation test
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
SaeHie Park [Fri, 21 Aug 2020 03:31:09 +0000 (12:31 +0900)]
[luci] Reduce duplicate codes in Conv2D shape-inf (#3914)
This will reduce duplicate codes in Conv2D type shape inference
ONE-DCO-1.0-Signed-off-by: SaeHie Park <saehie.park@gmail.com>
SaeHie Park [Fri, 21 Aug 2020 02:33:51 +0000 (11:33 +0900)]
[locomotiv] Reduce class codes T (#3909)
This will rename methods to reduce class lines of codes of NodeExecution
ONE-DCO-1.0-Signed-off-by: SaeHie Park <saehie.park@gmail.com>
SaeHie Park [Fri, 21 Aug 2020 02:15:28 +0000 (11:15 +0900)]
[locomotiv] Reduce class codes M-S (#3908)
This will rename methods to reduce class lines of codes of NodeExecution
ONE-DCO-1.0-Signed-off-by: SaeHie Park <saehie.park@gmail.com>
SaeHie Park [Fri, 21 Aug 2020 02:09:03 +0000 (11:09 +0900)]
[locomotiv] Reduce class codes F-M (#3907)
This will rename methods to reduce class lines of codes of NodeExecution
ONE-DCO-1.0-Signed-off-by: SaeHie Park <saehie.park@gmail.com>
SaeHie Park [Fri, 21 Aug 2020 02:08:38 +0000 (11:08 +0900)]
[locomotiv] Reduce class codes B-D (#3906)
This will rename methods to reduce class lines of codes of NodeExecution
ONE-DCO-1.0-Signed-off-by: SaeHie Park <saehie.park@gmail.com>
seongwoo chae [Fri, 21 Aug 2020 01:49:36 +0000 (10:49 +0900)]
[circle-quantizer] verify circle file (#3894)
This commit enables circle-quantizer to verify circle file.
ONE-DCO-1.0-Signed-off-by: seongwoo <mhs4670go@naver.com>
seongwoo chae [Thu, 20 Aug 2020 23:34:34 +0000 (08:34 +0900)]
[record-minmax] verify circle file (#3893)
This commit enables record-minmax to verify circle file.
ONE-DCO-1.0-Signed-off-by: seongwoo <mhs4670go@naver.com>
Alexander Moiseev [Thu, 20 Aug 2020 21:20:58 +0000 (00:20 +0300)]
[luci/tests] Add NonMaxSuppressionV5 luci tests (#3804)
Add testcases for NonMaxSuppressionV5 op in luci.
ONE-DCO-1.0-Signed-off-by: Alexander Moiseev <a.moiseev@samsung.com>
Co-authored-by: Alexander Efimov <efimov.alexander@gmail.com>
Hanjoung Lee [Thu, 20 Aug 2020 11:13:32 +0000 (20:13 +0900)]
[onert] Revise OpSequence dump method (#3890)
- Make the method to be outside of the class
- Remove message parameter
ONE-DCO-1.0-Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
Hanjoung Lee [Thu, 20 Aug 2020 09:25:43 +0000 (18:25 +0900)]
[onert] Use TensorRegistry directly in controlflow (#3899)
Use TensorRegistry directly in controlflow KernelGenerator.
ONE-DCO-1.0-Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
Alexander Moiseev [Thu, 20 Aug 2020 09:01:22 +0000 (12:01 +0300)]
[res] Add NonMaxSuppressionV5 recipes (#3803)
Add recipes for NonMaxSuppressionV5 op.
ONE-DCO-1.0-Signed-off-by: Alexander Moiseev <a.moiseev@samsung.com>
a-zh [Thu, 20 Aug 2020 08:47:57 +0000 (11:47 +0300)]
[luci] Enable PadV2 operator in tests (#3675)
This enables tests for PadV2 operator
Parent Issue: #3362
Draft PR: #3566
ONE-DCO-1.0-Signed-off-by: Anton Zhukov <a.zhukov@samsung.com>
seongwoo chae [Thu, 20 Aug 2020 08:40:42 +0000 (17:40 +0900)]
[tflite2circle] verify tflite file (#3892)
This commit enables tflite2circle to verify tflite file.
ONE-DCO-1.0-Signed-off-by: seongwoo <mhs4670go@naver.com>
Hanjoung Lee [Thu, 20 Aug 2020 08:10:41 +0000 (17:10 +0900)]
[onert] cpu KernelGen use _tensor_reg directly (#3895)
cpu KernelGen use _tensor_reg directly
ONE-DCO-1.0-Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
Hyeongseok Oh [Thu, 20 Aug 2020 08:05:16 +0000 (17:05 +0900)]
[onert] Remove unused Mean IR (#3898)
Mean IR is merged into Reduce IR
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
Hanjoung Lee [Thu, 20 Aug 2020 07:05:57 +0000 (16:05 +0900)]
[doc] Write compute.md (#3879)
This commit fills `compute.md` which is about compute directory.
ONE-DCO-1.0-Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
Hyeongseok Oh [Thu, 20 Aug 2020 07:02:34 +0000 (16:02 +0900)]
[onert/test] Fix test command warning (#3888)
Fix test command warning: unary operator expected
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
dr-venkman [Thu, 20 Aug 2020 06:04:33 +0000 (15:04 +0900)]
[luci/service] Correct Unique operation (#3876)
Parent Issue : #2568
Draft PR : #2829
This commit corrects shapeinferencing for unique operation
ONE-DCO-1.0-Signed-off-by: venkat.iyer <venkat.iyer@samsung.com>
Hanjoung Lee [Thu, 20 Aug 2020 05:48:15 +0000 (14:48 +0900)]
[onert] Remove `ITensorBuilder::tensorAt` (#3880)
Remove `ITensorBuilder::tensorAt` to distinguish the roles of
`ITensorBuilder` and `ITensorRegistry`. A builder is a builder and a
registry contains built tensors.
Note that KernelGenerator and ConstantInitializer now use TensorRegistry
to find tensor objects which takes most part of the changes.
ONE-DCO-1.0-Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
Hanjoung Lee [Thu, 20 Aug 2020 03:41:12 +0000 (12:41 +0900)]
[onert] Simplify CF KernelGenerator::getTensor (#3869)
Simplify `backend::controlflow::KernelGenerator::getTensor`.
`TensorBuilders` has the exact function already.
ONE-DCO-1.0-Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
Hyeongseok Oh [Thu, 20 Aug 2020 02:59:01 +0000 (11:59 +0900)]
[infra] Update eigen download url (#3883)
bitbucket no more support source code download service
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
YiHyunJin [Wed, 19 Aug 2020 09:26:14 +0000 (18:26 +0900)]
[onert] Support dilation for Conv2d (#3845)
It applied dilation data in conv2d And Added test case
for conv2d with dilation data
Signed-off-by: YiHyunjin <hj0412.yi@samsung.com>
SaeHie Park [Wed, 19 Aug 2020 09:09:16 +0000 (18:09 +0900)]
[circle2circle] Fix segfault for invalid input (#3874)
This will fix to handle segment fault error for invalid input file
ONE-DCO-1.0-Signed-off-by: SaeHie Park <saehie.park@gmail.com>
Hyeongseok Oh [Wed, 19 Aug 2020 09:07:47 +0000 (18:07 +0900)]
[tool/model2nnpkg] Fix string comparison bug (#3871)
Fix string comparison test bug
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
a-zh [Wed, 19 Aug 2020 08:55:53 +0000 (11:55 +0300)]
[res] TensorFlowLiteRecipes for PadV2 (#3676)
This adds PadV2 model recipe
Parent Issue: #3362
Draft PR: #3566
ONE-DCO-1.0-Signed-off-by: Anton Zhukov <a.zhukov@samsung.com>
Alexander Moiseev [Wed, 19 Aug 2020 08:49:35 +0000 (11:49 +0300)]
[luci/export] Support NonMaxSuppressionV5 op (#3802)
Add support for NonMaxSuppressionV5 op in luci exporter.
ONE-DCO-1.0-Signed-off-by: Alexander Moiseev <a.moiseev@samsung.com>
Alexander Moiseev [Wed, 19 Aug 2020 08:46:41 +0000 (11:46 +0300)]
[luci/import] Add NonMaxSuppressionV5 op (#3801)
Support NonMaxSuppressionV5 in luci importer.
ONE-DCO-1.0-Signed-off-by: Alexander Moiseev <a.moiseev@samsung.com>
Alexander Moiseev [Wed, 19 Aug 2020 08:46:29 +0000 (11:46 +0300)]
[luci/service] Support NonMaxSuppressionV5 op (#3800)
Add shape and type inference for NonMaxSuppressionV5 in luci service.
ONE-DCO-1.0-Signed-off-by: Alexander Moiseev <a.moiseev@samsung.com>
Alexander Moiseev [Wed, 19 Aug 2020 08:46:17 +0000 (11:46 +0300)]
[luci/logex] Support NonMaxSuppressionV5 op (#3799)
Add support for NonMaxSuppressionV5 op in luci/logex.
ONE-DCO-1.0-Signed-off-by: Alexander Moiseev <a.moiseev@samsung.com>
Hanjoung Lee [Wed, 19 Aug 2020 08:34:06 +0000 (17:34 +0900)]
[onert] Add ifdef guard to controlflow TensorReg (#3868)
Add ifdef guard to controlflow TensorRegistry.
ONE-DCO-1.0-Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
Hyukjin Jeong [Wed, 19 Aug 2020 08:28:10 +0000 (17:28 +0900)]
[luci] Support for requantization from int8 to uint8 (#3830)
This supports requantization from int8 to uin8
ONE-DCO-1.0-Signed-off-by: Hyukjin Jeong <hj1.jeong@samsung.com>
dr-venkman [Wed, 19 Aug 2020 08:21:43 +0000 (17:21 +0900)]
[res/TensorFlowLiteRecipes] Correct Unique recipe (#3862)
Parent Issue : #1606
Draft PR : #2829
This commit adds a correction for unique recipes that were introduced in #3410
ONE-DCO-1.0-Signed-off-by: venkat.iyer <venkat.iyer@samsung.com>
Sanggyu Lee [Wed, 19 Aug 2020 08:13:54 +0000 (17:13 +0900)]
[onert] build nnpkg_run by default for android (#3860)
nnpackage_run will be built by default for android.
Note: nnpackage_run will be built without feature of loading and dumping
hdf5 if you don't specify pre-built hdf5 libraries.
ONE-DCO-1.0-Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
Hyeongseok Oh [Wed, 19 Aug 2020 07:50:32 +0000 (16:50 +0900)]
[infra/cmake] Boost config and Install path (#3867)
- Introduce boost source download mirror environment: BOOST_URL
- Changebuild path to intermediate workspace (obj/externals/boost)
- Change user-config.jam generation path to intermediate workspace (obj/externals/boost)
- Change built boost install path to overlay
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
Hanjoung Lee [Wed, 19 Aug 2020 06:59:55 +0000 (15:59 +0900)]
[onert] Add TensorRegistry for acl backends (#3865)
- Make `AclTensorBuilder:tensorRegistry` return a tensor registry
- Simplify getting tensor of `controlflow::getTensorBuilder`
This is to make all the backends have `TensorRegistry`.
Part of #3809
ONE-DCO-1.0-Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
Hyeongseok Oh [Wed, 19 Aug 2020 06:49:36 +0000 (15:49 +0900)]
[tool/nnpkg] Set nnpacakge name (#3861)
Allow to set nnpackage name by option -p
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
Jang Jiseob [Wed, 19 Aug 2020 06:30:21 +0000 (15:30 +0900)]
[onert] Unify pooling operations to Pool2D (#3853)
This commit unifies pooling operations to Pool2D for reducing code duplication.
Signed-off-by: ragmani <ragmani0216@gmail.com>
Yongseop Kim [Wed, 19 Aug 2020 06:10:46 +0000 (15:10 +0900)]
[onert/util] Introduce MDTable write type of ChromeTracingObserver (#3858)
* [onert/util] Introduce MDTable write type of ChromeTracingObserver
Introduce new write type of ChromeTracingObserver - MD Table
ONE-DCO-1.0-Signed-off-by: Yongseop Kim <yons.kim@samsung.com>
* Add virtual destructor
ONE-DCO-1.0-Signed-off-by: Yongseop Kim <yons.kim@samsung.com>
* Revert change of WriteFormat
ONE-DCO-1.0-Signed-off-by: Yongseop Kim <yons.kim@samsung.com>
Sanggyu Lee [Wed, 19 Aug 2020 03:16:22 +0000 (12:16 +0900)]
[nnpackage_tool] update tflite2circle (#3852)
- Update tensorflow schema version v1.12.0 -> v1.13.1
- Remove unnecessary default value writings in json
ONE-DCO-1.0-Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
Hanjoung Lee [Wed, 19 Aug 2020 02:18:31 +0000 (11:18 +0900)]
[onert] Fix HEScheduler and running script (#3855)
- Fix HEScheduler to make it work
- Update testing script to make it work
ONE-DCO-1.0-Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
Alexander Moiseev [Tue, 18 Aug 2020 21:13:41 +0000 (00:13 +0300)]
[common-artifacts] Add NonMaxSuppressionV5 tests (#3798)
Exclude tests for NonMaxSuppressionV5 op in common artifacts.
ONE-DCO-1.0-Signed-off-by: Alexander Moiseev <a.moiseev@samsung.com>
a-zh [Tue, 18 Aug 2020 17:18:51 +0000 (20:18 +0300)]
[luci] Enable PadV2 operator in services (#3677)
This enables PadV2 support in services
Parent Issue: #3362
Draft PR: #3566
ONE-DCO-1.0-Signed-off-by: Anton Zhukov <a.zhukov@samsung.com>
Alexander Moiseev [Tue, 18 Aug 2020 16:12:01 +0000 (19:12 +0300)]
[tflite2circle] Enable NonMaxSuppressionV5 op (#3794)
Support NonMaxSuppressionV5 op in tflite2circle.
ONE-DCO-1.0-Signed-off-by: Alexander Moiseev <a.moiseev@samsung.com>
Alexander Moiseev [Tue, 18 Aug 2020 16:02:59 +0000 (19:02 +0300)]
[luci/lang] Support NonMaxSuppressionV5 op (#3797)
Add support for NonMaxSuppressionV5 op in luci IR.
ONE-DCO-1.0-Signed-off-by: Alexander Moiseev <a.moiseev@samsung.com>
Alexander Moiseev [Tue, 18 Aug 2020 15:55:57 +0000 (18:55 +0300)]
[tfldump] No opt comment for NonMaxSuppressionV5 (#3795)
Comment about no option in NonMaxSuppressionV5.
ONE-DCO-1.0-Signed-off-by: Alexander Moiseev <a.moiseev@samsung.com>
Alexander Moiseev [Tue, 18 Aug 2020 15:55:24 +0000 (18:55 +0300)]
[tflchef] Support NonMaxSuppressionV5 op (#3793)
Support NonMaxSuppressionV5 op in tflchef.
ONE-DCO-1.0-Signed-off-by: Alexander Moiseev <a.moiseev@samsung.com>
JiHwan Yeo [Tue, 18 Aug 2020 09:42:31 +0000 (18:42 +0900)]
[benchmark] Measure memory usage at the end of each phase (#3843)
- Typically, the memory usage is largest at the end of the run
- However, memory usage is not measured at the end of each phase
- This PR measures memory usage at the end of each phase
ONE-DCO-1.0-Signed-off-by: JiHwan Yeo <jihwan.yeo@samsung.com>
Hanjoung Lee [Tue, 18 Aug 2020 07:45:13 +0000 (16:45 +0900)]
[onert] Remove method `TensorBuilder::iterate` (#3842)
Remove method `TensorBuilder::iterate` which is never used.
ONE-DCO-1.0-Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
struss [Tue, 18 Aug 2020 06:50:37 +0000 (15:50 +0900)]
[nncc/infra]Add GEMMLowp and Ruy source for Tensorflow 2.3.0 (#3849)
This commit add GEMMLowp and Ruy source for Tensorflow 2.3.0
ONE-DCO-1.0-Signed-off-by: KiDeuk Bang <rrstrous@nate.com>
Hanjoung Lee [Tue, 18 Aug 2020 03:24:01 +0000 (12:24 +0900)]
[onert] Throw when no backends loaded (#3829)
Throw when no backends were loaded rather than assertion.
Add a regression test.
Resolve #3826
ONE-DCO-1.0-Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
SaeHie Park [Tue, 18 Aug 2020 01:42:36 +0000 (10:42 +0900)]
[luci] Apply disable_validation to optimized model (#3847)
This will apply disable_validation option to optimized model validation
- for real model with dynamic shape in Concat
ONE-DCO-1.0-Signed-off-by: SaeHie Park <saehie.park@gmail.com>
SaeHie Park [Tue, 18 Aug 2020 01:41:34 +0000 (10:41 +0900)]
[luci] Fix typo in command help (#3848)
This will fix typo in command help of disable_validation
ONE-DCO-1.0-Signed-off-by: SaeHie Park <saehie.park@gmail.com>
Hyukjin Jeong [Mon, 17 Aug 2020 21:33:22 +0000 (06:33 +0900)]
[pota-value-test] Add test data for channel-wise quant (#3841)
This adds test data for CWQ of following Ops
- Conv2D, DepthwiseConv2D, FullyConnected, TransposeConv
ONE-DCO-1.0-Signed-off-by: Hyukjin Jeong <hj1.jeong@samsung.com>
Seok Namkoong [Fri, 14 Aug 2020 08:10:01 +0000 (17:10 +0900)]
[tflite2circle] Add shape signature during conversion (#3834)
Until now, `tflite2circle` didn't add shape_signature of TFLite model.
This commit will add shape_signature to circle model.
ONE-DCO-1.0-Signed-off-by: Seok NamKoong <seok9311@naver.com>
Jang Jiseob [Fri, 14 Aug 2020 07:41:26 +0000 (16:41 +0900)]
[onert] Unify element-wise binary operations to ElementwiseBinary (#3791)
This commit unifies element-wise binary operations to ElementwiseBinary for reducing code duplication.
- ElementwiseBinary ops: LogicalAnd, LogicalOr, Maximum, Minimum
Signed-off-by: ragmani <ragmani0216@gmail.com>