박천교/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 5 Nov 2019 06:35:46 +0000 (15:35 +0900)]
[exo] Type & shape inference for TFLMean (#8739)
This commit introduces type and shape inference for TFLMean
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 5 Nov 2019 06:17:38 +0000 (15:17 +0900)]
[tf2tflite-dredd-pbtxt-test] Adding CMakeLists.txt and runner.sh (#8744)
* [tf2tflite-dredd-pbtxt-test] Adding CMakeLists.txt and runner.sh
CMakeLists.txt and runner.sh files were added.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* modify comment
* rules -> rule
* TARGET_FILENAME -> TARGET_TESTNAME
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 5 Nov 2019 06:09:18 +0000 (15:09 +0900)]
[exo] Refactoring EltwiseBinaryConverter using DomainConverter (#8752)
This refactors EltwiseBinaryConverter using DomainConverter (and input handler).
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
[neurun/cker] Add Elementwise sub support in cpu (#8554)
* [neurun/cker] Add Elementwise sub support in cpu
Added Sub operation to neurun cpu backend and kernel to cker.
It is copied from Add and renamed -i add to sub and changed
+ to -
Related issue: 7814
Signed-off-by: Poshshoev Dilshodzhon <d.poshshoev@samsung.com>
* Remove sub from framework test since just broadcast is tests
Signed-off-by: Poshshoev Dilshodzhon <d.poshshoev@samsung.com>
* Templatize Sub and Add for float
Signed-off-by: Poshshoev Dilshodzhon <d.poshshoev@samsung.com>
* Fix rebase issue because of renaming runtimes to runtime
Signed-off-by: Poshshoev Dilshodzhon <d.poshshoev@samsung.com>
[neurun/cker] Add Logistic operation support in cpu (#8572)
* [neurun/cker]Add Logistic operation support in cpu
Added support of logistic operation (which is sigmoid)
Signed-off-by: Poshshoev Dilshodzhon <d.poshshoev@samsung.com>
* Address feedback: use std::exp instead of exp
Signed-off-by: Poshshoev Dilshodzhon <d.poshshoev@samsung.com>
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 5 Nov 2019 05:38:25 +0000 (14:38 +0900)]
[neurun] Print error when a test was not found (#8728)
Print error when a framework test was not found. It is printed in file
`framework_test.log`.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 5 Nov 2019 05:35:35 +0000 (14:35 +0900)]
[moco] Set graph input shape in Importer (#8746)
This will enable graph input shape setting after loading in Importer
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 5 Nov 2019 03:22:09 +0000 (12:22 +0900)]
Remove oneco-integration-test (#8747)
This commit removes oneco-integration-test, which is replaced by
oneco-value-pbtxt-test
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 5 Nov 2019 03:21:56 +0000 (12:21 +0900)]
Remove tf2tflite-integration-test (#8748)
This commit removes tf2tflite-integration-test, which is replaced by
tf2tflite-value-pbtxt-test.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 5 Nov 2019 02:30:38 +0000 (11:30 +0900)]
Remove tf2tflite-model-test (#8749)
This commit removes tf2tflite-model-test, which is replaced by
tf2tflite-value-pb-test
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 5 Nov 2019 02:30:25 +0000 (11:30 +0900)]
Remove moco-integration-test (#8750)
This commit removes moco-integration-test, which replaced by
moco-value-pbtxt-test
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 5 Nov 2019 01:53:22 +0000 (10:53 +0900)]
[exo] Register Fuse Rsqrt and SquaredDifference (#8705)
This commit registers FuseRsqrtPass and FuseSquaredDifferencePass to exo
TFL optimize stage. Note that FuseSquaredDifferencePass is turn off by
default.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 5 Nov 2019 01:49:10 +0000 (10:49 +0900)]
[nncc] Install tf2circle (#8725)
Let's include tf2circle in nncc package. Please note that tf2nnpkg
still uses tf2ftlite and tflite2circle instead of this tf2circle.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 5 Nov 2019 01:40:43 +0000 (10:40 +0900)]
[exo] Export TFLMean (#8740)
This commit implements export stage of TFLMean for both tflite and
circle output
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 5 Nov 2019 00:57:52 +0000 (09:57 +0900)]
[nncc] Show configure progress (#8679)
* [nncc] Show configure progress
From now on, running nncc configure shows configuration progress for
each individual components.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
* Fix a typo
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 5 Nov 2019 00:25:01 +0000 (09:25 +0900)]
Introduce oneco-value-pbtxt-test (#8737)
This commit introduces oneco value test for pbtxt input model. It copies
existing oneco-integration-test only with renaming.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 5 Nov 2019 00:20:25 +0000 (09:20 +0900)]
Introduce moco-value-pbtxt-test (#8738)
This commit introduces moco value test for pbtxt input model. It copies
existing moco-integration-test only with renaming.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 5 Nov 2019 00:18:41 +0000 (09:18 +0900)]
Introduce tf2tflite-value-pb-test (#8735)
This commit introduces tf2tflite value test for pb input model. It copies
existing tf2tflite-model-test only with renaming.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 5 Nov 2019 00:18:28 +0000 (09:18 +0900)]
Introduce tf2tflite-value-pbtxt-test (#8736)
This commit introduces tf2tflite value test for pbtxt input model. It copies
existing tf2tflite-integration-test only with renaming.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 5 Nov 2019 00:14:19 +0000 (09:14 +0900)]
[exo] introducing DomainConverter (#8745)
This adds DomainConverter and InputHandler, which are helpers used while converting a canonical node in Feature domain into a TFL node (Tensor domain).
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 4 Nov 2019 23:22:54 +0000 (08:22 +0900)]
[moco] GraphOutputIndexQueryService in TF dialect (#8726)
This will add GraphOutputIndexQueryService in TF dialect
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 4 Nov 2019 09:50:36 +0000 (18:50 +0900)]
Enable ncnn(srcn) tensor to support NCHW (#8723)
This commit enables ncnn(srcn) tensor to support NCHW(row major).
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 4 Nov 2019 08:54:56 +0000 (17:54 +0900)]
[tf2tflite-dredd-pbtxt-test] renaming tf2tflite-quality-test to tf2tflite-dredd-pbtxt-test (#8717)
* [tf2tflite-dredd-pbtxt-test] renaming tf2tflite-quality-test
As discussed offlie, tf2tflite-qualit-test is renamed to tf2tflite-dredd-pbtxt-test.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* removed compiler/tf2tflite-quality-test/.gitignore
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 4 Nov 2019 08:06:59 +0000 (17:06 +0900)]
[loader] Fix infinite recursive call (#8724)
Explicit call parent class's method
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 4 Nov 2019 07:45:10 +0000 (16:45 +0900)]
[moco] TFPlaceholder helpers for index query (#8720)
This will add two helper methods for GraphInputIndex query service
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
[neurun] Replace OperandIndex in DepthToSpace and SpaceToDepth Params (#8707)
Replace `OperandIndex` in `DepthToSpace::Param` and `SpaceToDepth::Param` with `int32_t`.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 4 Nov 2019 07:12:46 +0000 (16:12 +0900)]
Introduce a configuration for layout of ncnn (#8621)
This commit introduces a configuration for layout of ncnn.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 4 Nov 2019 06:48:17 +0000 (15:48 +0900)]
[loco] GraphInputIndexQueryService and input_nodes() (#8713)
This will introduce GraphInputIndexQueryService interface and input_nodes() method to gather Graph Input nodes
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 4 Nov 2019 04:23:58 +0000 (13:23 +0900)]
Make ExecutorBase supports NCHW frontend layout (#8692)
This commit makes ExecutorBase supports NCHW frontend layout.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
[neurun] Replace OperandIndex in LocalResponseNormalization::Param (#8708)
Replace `OperandIndex` in `LocalResponseNormalization::Param` with `int32_t` / `float`.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 4 Nov 2019 02:27:36 +0000 (11:27 +0900)]
[lib/ncnn] Fix depthwise conv NCHW kernel bug (#8711)
Fix depthwise convolution NCHW 3x3 weight same padding kernel bug
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 4 Nov 2019 01:09:26 +0000 (10:09 +0900)]
[base_loader] Throw when file open failed (#8706)
Throw with a brief message when file open failed
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Sun, 3 Nov 2019 22:09:46 +0000 (07:09 +0900)]
[exo] TFLMean IR (#8698)
This commit introduces TFLMean IR
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 1 Nov 2019 08:24:27 +0000 (17:24 +0900)]
[moco] Fix typo on TFMean shape inference (#8699)
This commit fixes typo on TFMean shape inference
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
[nnpackage_run] Use HDF5 C++ interface (#8671)
Use HDF5 C++ interface instead of C.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 1 Nov 2019 07:46:24 +0000 (16:46 +0900)]
Do NOT install locomotiv (#8690)
locomotiv is now built as a static library, and thus it is no longer
necessary to install this library.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 1 Nov 2019 07:35:39 +0000 (16:35 +0900)]
[moco] Use moco Importer.h (#8702)
This will update to use moco/Importer.h in moco and remove old file in sub folder
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 1 Nov 2019 07:29:52 +0000 (16:29 +0900)]
[test] Fix misleading path (#8693)
Nit: fix path in `test_arm_neurun_acl_cl.sh`.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
여지환/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 1 Nov 2019 07:23:55 +0000 (16:23 +0900)]
[neurun] Introduce InstanceNorm op (#8526)
This commit introduces InstanceNorm op into runtime frontend.
- Introduce InstanceNormNode into neurun
- Introduce InstanceNorm op into frontend
- Use default value if value of epsilon is zero
Signed-off-by: JiHwan Yeo <jihwan.yeo@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 1 Nov 2019 07:15:05 +0000 (16:15 +0900)]
[exo] Knob for fuse SquaredDifference (#8696)
This commit introduces UseFuseSquaredDifferencePass knob
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 1 Nov 2019 07:14:51 +0000 (16:14 +0900)]
[exo] Register EltwiseSqrt Converter (#8697)
This commit registers EltwiseSqrt Converter to exo conversion stage
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 1 Nov 2019 07:14:01 +0000 (16:14 +0900)]
[exo] Remove redundant steps of type and shape inference (#8695)
This removes redundant steps of type and shape inference.
Note that convert() and optimize() has already performed type and shape inference.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 1 Nov 2019 05:44:48 +0000 (14:44 +0900)]
[angkor] Change num_elements behavior (#8684)
The current implementation treats rank-0 tensors as a special tensor
which has no elements.
This semantics has a conflict with the well-known standard which treats
a rank-0 tensor as a scalar.
This commit resolves this semantic conflict.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 1 Nov 2019 05:36:01 +0000 (14:36 +0900)]
[exo] Fuse Squared Difference (#8654)
This commit introduces a pass to fuse Squared Difference pattern
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 1 Nov 2019 05:20:59 +0000 (14:20 +0900)]
[moco-tf] Remove alias and use moco (#8688)
This will remove moco::tf alias for moco and update to use moco types
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 1 Nov 2019 04:17:23 +0000 (13:17 +0900)]
Introduce Readers for nchw (#8663)
This commit introduces Readers for nchw.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 1 Nov 2019 04:13:12 +0000 (13:13 +0900)]
[exo] Export Sqrt, Rsqrt, SquaredDifference (#8652)
This commit implements export stage for TFLSqrt, TFLRsqrt and
TFLSquaredDifference. Note that both TFLite exporter and Circle exporter
implemented.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 1 Nov 2019 04:10:45 +0000 (13:10 +0900)]
[moco-tf] Use moco/Importer.h (#8686)
This will update to use moco/Importer.h in moco-tf
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 1 Nov 2019 03:29:38 +0000 (12:29 +0900)]
[nnkit-mocotf] Use moco namespace types not alias (#8682)
This will update to not to use alias moco::tf namespace
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 1 Nov 2019 03:29:03 +0000 (12:29 +0900)]
[tf2circle] Use moco namespace types not alias (#8681)
This will update to not to use alias moco::tf namespace
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 1 Nov 2019 03:03:38 +0000 (12:03 +0900)]
[nnkit-tf] Throw T/F error message (#8673)
The exception thrown inside run method now contains detailed message
that comes from TensorFlow itself.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 1 Nov 2019 03:01:37 +0000 (12:01 +0900)]
[neurun] Update ifdef guards of operations (#8683)
Update `#ifdef` guards of operations which were not updated when it was
renamed.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
[nnapi] Remove skipping Gather operation on cpu tests (#8667)
* Enabled tests for Gather operation on cpu backends
Signed-off-by: Pavel Iliutchenko <p.iliutchenk@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 1 Nov 2019 01:54:20 +0000 (10:54 +0900)]
[tf2tflite] Use moco namespace types not alias (#8680)
This will update to not to use alias moco::tf namespace
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 1 Nov 2019 01:49:07 +0000 (10:49 +0900)]
[moco] Tidy import unused dependency (#8676)
This will remove unused dependency headers and libraries
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 1 Nov 2019 01:40:07 +0000 (10:40 +0900)]
[exo] Fuse Rsqrt (#8655)
This commit introduces a pass to fuse Rsqrt pattern
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 1 Nov 2019 01:19:19 +0000 (10:19 +0900)]
[exo] Convert EltwiseSqrt (#8653)
This commit implements conversion from canonical EltwiseSqrt to
TFL-dialect TFLSqrt
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 1 Nov 2019 01:08:30 +0000 (10:08 +0900)]
[moco] Provide Importer header in moco folder (#8672)
This will provide Importer header in moco folder
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 1 Nov 2019 00:58:26 +0000 (09:58 +0900)]
[neurun] Rename operation classes (#8666)
This commit removes suffix 'Node' from operation class names since
redundant.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 31 Oct 2019 12:09:13 +0000 (21:09 +0900)]
[tflite_loader] Introduce tflite_loader test list (#8660)
Itnroduce tflite_loader test list for tflite_loader validation test
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 31 Oct 2019 10:30:06 +0000 (19:30 +0900)]
[runtime] Move runtimes and rename runtime directory (#8472)
Move logging and pacl runtime into contrib
Rename runtime directory: runtimes -> runtime
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 31 Oct 2019 10:04:52 +0000 (19:04 +0900)]
[moco] Update Importer comment (#8657)
This will update Importer comment with input and output nodes
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 31 Oct 2019 10:04:30 +0000 (19:04 +0900)]
[moco-tf] Remove output node handle for Push (#8656)
This will remove not needed anymore handle for Canonical Push as output node
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 31 Oct 2019 09:57:59 +0000 (18:57 +0900)]
[base_loader] Remove unused variable (#8659)
Remove unused variable in `BaseLoader::loadOperand`.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
[neurun] Inline Graph::linearize (#8627)
This is to remove dependency of `Graph` on `Linear`.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 31 Oct 2019 09:30:56 +0000 (18:30 +0900)]
Make Readers and Views support both ITensor and buffer (#8644)
This commit makes Readers and Views support both ITensor and buffer.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 31 Oct 2019 09:15:08 +0000 (18:15 +0900)]
[neurun] Allow same inputs multiple times (#8567)
Allow same inputs multiple times for an operation.
Remove redundant assertion
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이상규/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 31 Oct 2019 09:11:53 +0000 (18:11 +0900)]
[neurun] Support 1x1x1x1 input for Squeeze (#8646)
Squeeze will handle 1x1x1x1 input using ShapeFixer.
Scalar output will be fixed to rank-1 one element tensor.
The output value will be transfer to app without trouble since nnfw
runtime API get output values using buffer pointer and buffer length.
Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 31 Oct 2019 07:47:27 +0000 (16:47 +0900)]
[moco] Produce TFPush for graph output (#8647)
* [moco] Produce TFPush for graph output
This will update to produce TFPush for graph output node
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* update comment
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 31 Oct 2019 07:19:45 +0000 (16:19 +0900)]
[exo] Type & shape infer. for Sqrt, Rsqrt, SquaredDifference (#8649)
This commit introduces type and shape inferences or TFLSqrt, TFLRsqrt,
TFLSquaredDifference.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 31 Oct 2019 06:40:49 +0000 (15:40 +0900)]
[circle-verify] Add VerifyFlatBuffers (#8648)
This will add and enable VerifyFlatBuffers verify Circle file in flatbuffer file format level
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 31 Oct 2019 06:37:52 +0000 (15:37 +0900)]
[exo] Note on Pass name convention (#8645)
This commit adds note on Pass name convention
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 31 Oct 2019 05:19:19 +0000 (14:19 +0900)]
[exo] Rename to have Pass suffix (#8643)
This commit renames following classes (and related files) to have 'Pass'
suffix like other Passes:
- FoldReshapeOfConst -> FoldReshapeOfConstPass
- FoldTransposeOfConst -> FoldTransposeOfConstPass
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 31 Oct 2019 05:14:49 +0000 (14:14 +0900)]
[exo] TFL nodes for Sqrt, Rsqrt, SquaredDifference (#8623)
This commit introduces simple TFL-dialect nodes:
- locoex::TFLSqrt
- locoex::TFLRsqrt
- locoex::TFLSquaredDifference
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 31 Oct 2019 05:07:27 +0000 (14:07 +0900)]
[circle-verify] Introduce Model and empty Driver (#8641)
* [circle-verify] Introduce Model and empty Driver
This will introdue Model that will provide mmapped buffer and empty Driver
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* apply comments
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 31 Oct 2019 04:43:20 +0000 (13:43 +0900)]
[moco-tf] Prepare to handle TFPush instead of Push (#8636)
This will update Frontend to support output as TFPush and Push while updating to produce TFPush as Output
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
[neurun] Remove workaround after switching to ACL v19.05 (#8629)
Prior versions had a bug which did not take into accout the activation fused into CLDepthwiseConvolutionLayer. The workaround for this is now removed.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[nnpkg_run] Fix hdf5 file dump (#8575)
Value group have to be created only once for file
Remove blank line
Signed-off-by: Vladimir Plazun <v.plazun@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 31 Oct 2019 04:15:17 +0000 (13:15 +0900)]
Change the dimension order of Reader and Views in nhwc (#8509)
This commit changes dimension order of Reader and Views in nhwc to nhwc.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 31 Oct 2019 01:56:01 +0000 (10:56 +0900)]
[circle-verify] Readme (#8635)
* [circle-verify] Readme
This will introduce circle-verify readme file explaining it's usage
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* edit about
* change output format
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 31 Oct 2019 01:55:25 +0000 (10:55 +0900)]
[moco-tf] Add TFPush NodeSummaryBuilder (#8634)
This will add TFPush node to NodeSummmaryBuilder
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 31 Oct 2019 01:47:44 +0000 (10:47 +0900)]
[moco-integration-test] adding tests related to depthwiseConv2D bias fusion (#8543)
This adds tests related to depthwiseConv2D bias fusion.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 31 Oct 2019 01:34:15 +0000 (10:34 +0900)]
[tfl-inspect] Enable nullptr for data in buffer_info (#8632)
This will enable buffer_info() to accept nullptr in buffer_data for use cases where only return size is needed
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 31 Oct 2019 01:31:52 +0000 (10:31 +0900)]
[moco] TF dialect OutputIndexQueryService (#8630)
This will enable TF dialect GraphOutputIndexQueryService
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 31 Oct 2019 01:29:03 +0000 (10:29 +0900)]
[tfl-inspect] Enable conv2d_weight option (#8633)
This will enable conv2d_weight option that shows Conv2D series node with non Const weight input
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 31 Oct 2019 01:27:08 +0000 (10:27 +0900)]
[moco-tf] Enable TFPush canonicalizer (#8631)
* [moco-tf] Enable TFPush canonicalizer
This will enable TFPushCanonicalizer in Canonicalizer phase
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* fix typo
이상규/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 31 Oct 2019 01:03:48 +0000 (10:03 +0900)]
[neurun] fix loadSqueeze and SqueezeNode. (#8602)
- `loadSqueeze` should read `squeeze_dims` from options.
- `squeeze_dims` should be array in SqueezeNode.
Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 30 Oct 2019 23:30:41 +0000 (08:30 +0900)]
[exo] Clean empty line for conversion registry (#8622)
This commit clears empty line in Convert.h, which registers all
conversion passes. This is mainly to enhance readability.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 30 Oct 2019 22:01:08 +0000 (07:01 +0900)]
[compiler] introducing tf2tflite-quality-test (#8614)
This introduces a new project called `tf2tflite-quality-test`, whose previous name was `dredd`.
The basic structure of this is very similar with `tf2tflite-model-test`.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 30 Oct 2019 22:00:39 +0000 (07:00 +0900)]
[exo] revising TestGraph.h (#8618)
This revises TestGraph.h to 1) remove build() and 2) remove graph().
1) build() is moved into constructor
2) graph() is already defined in parent
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 30 Oct 2019 21:56:01 +0000 (06:56 +0900)]
[moco-tf] Push canonicalizer (#8617)
This will introduce Canonicalizer for TFPush node
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
[neurun] Support Gather operation on cpu backend (#8399)
* Added Gather kernel on cker
* Changed axis Gather param
* Fixed other parts with Gather
Signed-off-by: Pavel Iliutchenko <p.iliutchenk@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Wed, 30 Oct 2019 10:11:17 +0000 (19:11 +0900)]
[nnfw-gbs] nnfw-devel package config file (#8518)
Introduce nnfw-dev package config file (nnfw.pc)
Change include path to {prefix}/include/nnfw
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 30 Oct 2019 10:06:26 +0000 (19:06 +0900)]
[exo] Removing EnableTFLDialect knob (#8615)
This removes EnableTFLDialect knob - related files are modified.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 30 Oct 2019 10:05:33 +0000 (19:05 +0900)]
[moco] push_node implementation (#8616)
This will introduce push_node implementation that retuns TFPush node with given GraphOutputIndex
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 30 Oct 2019 10:05:10 +0000 (19:05 +0900)]
[tfl-inspect] Introduce DumpConv2DWeight (#8611)
* [tfl-inspect] Introduce DumpConv2DWeight
This will introduce DumpConv2DWeight to dump Conv2D series node weight input type
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* fix namespace
* update comment
* use buildin_code
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 30 Oct 2019 09:37:58 +0000 (18:37 +0900)]
[moco] Shape and type inference for TFPush (#8613)
This will add shape and type inference for TFPush node
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 30 Oct 2019 07:53:44 +0000 (16:53 +0900)]
[moco] Introduce TFPush IR (#8598)
This will introduce TFPush IR that will be used as graph output
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Wed, 30 Oct 2019 07:14:09 +0000 (16:14 +0900)]
[lib/misc] Handle scalar input/output (#8607)
Fix error when object is scalar
This will fix error nnapi_test when input/output is scalar
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 30 Oct 2019 06:51:00 +0000 (15:51 +0900)]
[tfl-inspect] Readme section for operators option (#8605)
This will introduce section in Readme file for operators option
Signed-off-by: SaeHie Park <saehie.park@samsung.com>