platform/core/ml/nnfw.git
6 years ago[neurun] Remove dot dump env verbose (#3073)
김수진/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 10 Oct 2018 10:23:30 +0000 (19:23 +0900)]
[neurun] Remove dot dump env verbose (#3073)

This commit removes dot dump env verbose which is not needed anymore.

Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
6 years agoFix wrong return of EnvVar::asBool (#3075)
윤현식/동작제어Lab(SR)/Principal Engineer/삼성전자 [Wed, 10 Oct 2018 10:15:10 +0000 (19:15 +0900)]
Fix wrong return of EnvVar::asBool (#3075)

Previously `asBool` always return `1`. Fixed the bug.

Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
6 years ago[neurun] Combine generateWeight and generateBias (#3066)
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 10 Oct 2018 08:25:51 +0000 (17:25 +0900)]
[neurun] Combine generateWeight and generateBias (#3066)

Combine `generateWeight` and `generateBias` into one `generate` method
in `IInitializerGenerator`.

The type `Initializer` is now a list of a pair of Index and Initializer
function. To keep the legacy code as much as possible, the two functions
are still there as private and non-virtual.

Resolve #3042

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
6 years agoAdd CL Kernels for SquaredDifference op. (#3037)
Prasanna R/System SW /SRI-Bangalore/Engineer/삼성전자 [Wed, 10 Oct 2018 08:04:59 +0000 (13:34 +0530)]
Add CL Kernels for SquaredDifference op. (#3037)

This patch adds CL Kernels for SquaredDifference op.

Signed-off-by: prasannar <prasanna.r@samsung.com>
6 years ago[neurun] Move backend interfaces to interface dir (#3041)
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 10 Oct 2018 05:28:55 +0000 (14:28 +0900)]
[neurun] Move backend interfaces to interface dir (#3041)

Move backend interfaces to interface dir.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
6 years agoSetting quantization info to fix bug (#3020)
Shubham Gupta/System SW /SRI-Bangalore/Engineer/삼성전자 [Wed, 10 Oct 2018 05:04:36 +0000 (10:34 +0530)]
Setting quantization info to fix bug (#3020)

This patch will set quantization info for the quant8 supported ops.

Signed-off-by: shubham <shub98.gupta@samsung.com>
6 years agoRemove passed Generated Tests from skiplist armv7l-linux (#3049)
Prasanna R/System SW /SRI-Bangalore/Engineer/삼성전자 [Wed, 10 Oct 2018 04:49:12 +0000 (10:19 +0530)]
Remove passed Generated Tests from skiplist armv7l-linux (#3049)

This patch removes the passed Generated Tests
```
tensorflowmax_ex*
space_to_depth*
```
from the skiplist ```runtime_run_android_nn_test.skip.armv7l-linux```.

Signed-off-by: prasannar <prasanna.r@samsung.com>
6 years agoAdd Generated Test for Squared Difference op (#3019)
Prasanna R/System SW /SRI-Bangalore/Engineer/삼성전자 [Wed, 10 Oct 2018 04:37:28 +0000 (10:07 +0530)]
Add Generated Test for Squared Difference op (#3019)

Add squared_difference_ex_1D_float for testing 1d tensor.
Add squared_difference_ex_2D_float for testing 2d tensor.

Signed-off-by: prasannar <prasanna.r@samsung.com>
6 years agoIncluded file is not required (#2910)
Shubham Gupta/System SW /SRI-Bangalore/Engineer/삼성전자 [Wed, 10 Oct 2018 03:05:48 +0000 (08:35 +0530)]
Included file is not required (#2910)

This patch will remove #include CLArithmeticSubtraction which is not getting used at all.

Signed-off-by: shubham <shub98.gupta@samsung.com>
6 years agoChanges in Generated files of gather gTests (#3045)
Prasanna R/System SW /SRI-Bangalore/Engineer/삼성전자 [Wed, 10 Oct 2018 03:05:12 +0000 (08:35 +0530)]
Changes in Generated files of gather gTests (#3045)

The files
```
gather_1D_quant8.example.cpp
gather_2D_quant8.example.cpp
gather_1D_quant8.model.cpp
gather_2D_quant8.model.cpp
```
have changes that is not reflected in the nnfw repo.
This patch points out those necessary changes.

Signed-off-by: prasannar <prasanna.r@samsung.com>
6 years ago[neurun] Only one item for codegen operand Context (#3043)
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 10 Oct 2018 02:53:43 +0000 (11:53 +0900)]
[neurun] Only one item for codegen operand Context (#3043)

* [neurun] Only one item for codegen operand Context

Now that Permutation insertion has changed that there must be only one
`backend::operand::Object` for each operand. So this commit removes
vector container which made it one-to-many relationship between graph
operand and backend operand.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
* Add assertion

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
6 years agoAdd NNAPI Delegation and Custom Op for Equal (#2984)
Prasanna R/System SW /SRI-Bangalore/Engineer/삼성전자 [Wed, 10 Oct 2018 02:33:26 +0000 (08:03 +0530)]
Add NNAPI Delegation and Custom Op for Equal (#2984)

This patch adds NNAPI Delegation and Custom op definition for Equal.

Signed-off-by: prasannar <prasanna.r@samsung.com>
6 years agoRemove depth_to_space gtest from skiplist armv7l-linux (#3052)
Shubham Gupta/System SW /SRI-Bangalore/Engineer/삼성전자 [Wed, 10 Oct 2018 02:10:49 +0000 (07:40 +0530)]
Remove depth_to_space gtest from skiplist armv7l-linux (#3052)

This patch will remove passing depth_to_space gtest from skiplist.

Signed-off-by: shubham <shub98.gupta@samsung.com>
6 years agoImplementation of Space_to_BatchND op in PACL (#2854)
Shubham Gupta/System SW /SRI-Bangalore/Engineer/삼성전자 [Wed, 10 Oct 2018 02:10:00 +0000 (07:40 +0530)]
Implementation of Space_to_BatchND op in PACL (#2854)

This patch will acc CPU implemenation of SPACETOBATCHND for rank=4 tensors.

Signed-off-by: shubham <shub98.gupta@samsung.com>
6 years ago[neurun] Simplify graph dot dumper implementation (#3034)
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 10 Oct 2018 02:07:34 +0000 (11:07 +0900)]
[neurun] Simplify graph dot dumper implementation (#3034)

* [neurun] Simplify graph dot dumper implementation

This commit unifies DotBuilder update methods into one, and remove
redundant shape info in the parameter list.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
* Simplify addNode

6 years agoAdd Apache License (#3005)
Shubham Gupta/System SW /SRI-Bangalore/Engineer/삼성전자 [Wed, 10 Oct 2018 01:25:20 +0000 (06:55 +0530)]
Add Apache License (#3005)

This patch adds Apache License to SimpleEmbedding lookup and SquaredDiff layer

Signed-off-by: shubham <shub98.gupta@samsung.com>
6 years ago[neurun] Remove unnecessary header includes (#3053)
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 10 Oct 2018 01:17:19 +0000 (10:17 +0900)]
[neurun] Remove unnecessary header includes (#3053)

Remove unnecessary header includes in
`frontend/wrapper/compilation.cc`.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
6 years ago[neurun] Update Plan and TensorBuilder role (#3011)
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Tue, 9 Oct 2018 23:56:55 +0000 (08:56 +0900)]
[neurun] Update Plan and TensorBuilder role (#3011)

Update `ITensorBuilder::prepare` not to depend on `Plan` object.

In order to do that, Introduce `wrapTensor` and `iterate` interfaces for
`ITensorBuilder`.

Resolve #2746

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
6 years agoPatch bug in ArmComputeEx NEG_Ex (#3018)
Prasanna R/System SW /SRI-Bangalore/Engineer/삼성전자 [Mon, 8 Oct 2018 05:25:17 +0000 (10:55 +0530)]
Patch bug in ArmComputeEx NEG_Ex (#3018)

This patch solves the bug in NEG_Ex in ArmComputeEx.
Related issue: #2920

Signed-off-by: prasannar <prasanna.r@samsung.com>
6 years agoFix run script (#3024)
박세희/동작제어Lab(SR)/Principal Engineer/삼성전자 [Mon, 8 Oct 2018 02:52:41 +0000 (11:52 +0900)]
Fix run script (#3024)

This will fix run script that was incorrecly fixed

Signed-off-by: SaeHie Park <saehie.park@samsung.com>
6 years ago[neurun] Add default ctor for operand::Shape (#3025)
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Mon, 8 Oct 2018 02:51:17 +0000 (11:51 +0900)]
[neurun] Add default ctor for operand::Shape (#3025)

Add default ctor for `graph::operand::Shape` to remove meaningless
initialization with rank param like `Shape{1}`.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
6 years agoFix a bug of StridedSliceCPU (#3015)
장지섭/동작제어Lab(SR)/Engineer/삼성전자 [Mon, 8 Oct 2018 01:34:57 +0000 (10:34 +0900)]
Fix a bug of StridedSliceCPU (#3015)

This commit fixes a bug of StridedSliceCPU.
- Fix StridedSliceCPU to work with `ICLTensor` even if the `ICLTensor`s have padding.

Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
6 years agoFix a bug of ReduceSum (#3013)
장지섭/동작제어Lab(SR)/Engineer/삼성전자 [Mon, 8 Oct 2018 01:34:42 +0000 (10:34 +0900)]
Fix a bug of ReduceSum (#3013)

This commit fixes a bug of ReduceSum.
- Fix ReduceSum to work with `ICLTensor` even if the `ICLTensor`s have padding.

Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
6 years agoFix a bug of ReduceMax for cpu (#3012)
장지섭/동작제어Lab(SR)/Engineer/삼성전자 [Mon, 8 Oct 2018 01:34:30 +0000 (10:34 +0900)]
Fix a bug of ReduceMax for cpu (#3012)

This commit fixes a bug of ReduceMax for cpu.
- Fix ReduceMax to work with `ICLTensor` even if the `ICLTensor`s have padding.

Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
6 years ago[neurun] Implement to run the dot dumper (#2912)
김수진/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 5 Oct 2018 07:14:50 +0000 (16:14 +0900)]
[neurun] Implement to run the dot dumper (#2912)

* [neurun] Implement to run the dot dumper

This commit implements to run the dot dumper with env GRAPH_DOT_DUMP.

Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
* add

* Extract dump method to DotDumper class

* add license to DotDumper.cc

6 years agoAdd Apache License to Local Response Normalization op (#2932)
Prasanna R/System SW /SRI-Bangalore/Engineer/삼성전자 [Fri, 5 Oct 2018 06:25:36 +0000 (11:55 +0530)]
Add Apache License to Local Response Normalization op (#2932)

This patch adds Apache License to Local Response Normalization op.

Signed-off-by: prasannar <prasanna.r@samsung.com>
6 years agoIntroduce TRANSPOSE_CONV operation on runtime (#2941)
Devansh Bansal/System SW /SRI-Bangalore/Engineer/삼성전자 [Fri, 5 Oct 2018 06:05:26 +0000 (11:35 +0530)]
Introduce TRANSPOSE_CONV operation on runtime (#2941)

This patch introduces TRANSPOSE_CONV operation on runtime.
This operator is required by grocery recognition system model.

Signed-off-by: b.devansh <b.devansh@samsung.com>
6 years ago[neurun] Rewrite the way for addShapeConstr (#2937)
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 5 Oct 2018 04:55:41 +0000 (13:55 +0900)]
[neurun] Rewrite the way for addShapeConstr (#2937)

Now that we have input/outputs as `operation::Node` interface, we do not
need to implement calls to `addShapeConstr` for each node. Instead, it
is done in general way for all nodes from `Linear::markTensors`. And for
this, there was some changes to `ITensorBuilder` interface.

Reslove #2741

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
6 years ago[tfltool] OP selector: Support more operator (#2975)
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 5 Oct 2018 04:14:51 +0000 (13:14 +0900)]
[tfltool] OP selector: Support more operator (#2975)

Support selecting operators: NEG, EXP

Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
6 years agoChange input arguments of SquaredDifference operator (#2903)
Prasanna R/System SW /SRI-Bangalore/Engineer/삼성전자 [Fri, 5 Oct 2018 04:10:32 +0000 (09:40 +0530)]
Change input arguments of SquaredDifference operator (#2903)

While testing grocery model, below errors occured,
```
tflite_run: /home/jyoung/git/star/nnfw/runtimes/pure_arm_compute/src/internal/op/SquaredDifference.cc:33: internal::tflite::op::SquaredDifference::Param::Param(uint32_t, const uint32_t*, uint32_t, const uint32_t*): Assertion `inputCount == 3 && outputCount == 1' failed.
```
There is no need of an activation index as SquaredDifference op takes only two inputs X & Y.
This patch makes those necessary changes.

Signed-off-by: prasannar <prasanna.r@samsung.com>
6 years agoUpdate nnapi_delegation for Space_To_BatchND (#2904)
Shubham Gupta/System SW /SRI-Bangalore/Engineer/삼성전자 [Fri, 5 Oct 2018 04:06:59 +0000 (09:36 +0530)]
Update nnapi_delegation for Space_To_BatchND (#2904)

This patch will add nnapi delegaion for Space_to_BatchND

Signed-off-by: shubham <shub98.gupta@samsung.com>
6 years agoAdd generated tests for NEG_EX (#2869)
Prasanna R/System SW /SRI-Bangalore/Engineer/삼성전자 [Fri, 5 Oct 2018 01:21:36 +0000 (06:51 +0530)]
Add generated tests for NEG_EX (#2869)

Add neg_ex_1D_float for testing 1d tensor.
Add neg_ex_2D_float for testing 2d tensor.

Signed-off-by: prasannar <prasanna.r@samsung.com>
6 years agoAdd Apache License to L2Normalization Op (#2931)
Prasanna R/System SW /SRI-Bangalore/Engineer/삼성전자 [Fri, 5 Oct 2018 01:20:33 +0000 (06:50 +0530)]
Add Apache License to L2Normalization Op (#2931)

This patch adds Apache License to L2Normalization Op.

Signed-off-by: prasannar <prasanna.r@samsung.com>
6 years ago[neurun] Introduce Enum `Input` for Conv2D (#2943)
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 4 Oct 2018 10:59:30 +0000 (19:59 +0900)]
[neurun] Introduce Enum `Input` for Conv2D (#2943)

Introduce Enum `Input` for FullyConnected to avoid using magic numbers
from input list.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
6 years agoAdd Apache License to SquaredDifference Op (#2930)
Prasanna R/System SW /SRI-Bangalore/Engineer/삼성전자 [Thu, 4 Oct 2018 10:58:27 +0000 (16:28 +0530)]
Add Apache License to SquaredDifference Op (#2930)

This patch adds Apache License to Squared Difference Op

Signed-off-by: prasannar <prasanna.r@samsung.com>
6 years ago[neurun] Introduce Enum `Input` for MaxPool2D (#2945)
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 4 Oct 2018 10:30:59 +0000 (19:30 +0900)]
[neurun] Introduce Enum `Input` for MaxPool2D (#2945)

Introduce Enum `Input` for MaxPool2D to avoid using magic numbers
from input list.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
6 years ago[neurun] Introduce Enum `Input` for the rest ops (#2947)
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 4 Oct 2018 10:30:48 +0000 (19:30 +0900)]
[neurun] Introduce Enum `Input` for the rest ops (#2947)

Introduce Enum `Input` for Reshape and Softmax to avoid using magic
numbers from input list.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
6 years agoAdd Apache License to HashTableLookup op (#2933)
Prasanna R/System SW /SRI-Bangalore/Engineer/삼성전자 [Thu, 4 Oct 2018 10:29:52 +0000 (15:59 +0530)]
Add Apache License to HashTableLookup op (#2933)

This patch adds Apache License to HashTableLookup op.

Signed-off-by: prasannar <prasanna.r@samsung.com>
6 years agoIntroduce ABS_EX op in NeuralNetworksEx.h (#2949)
Siva Sai Vaddipati/System SW /SRI-Bangalore/Engineer/삼성전자 [Thu, 4 Oct 2018 10:16:21 +0000 (15:46 +0530)]
Introduce ABS_EX op in NeuralNetworksEx.h (#2949)

This commit introduces ABS_EX op in NeuralNetworksEx.h
Related issue: #2828

Signed-off-by: Siva Sai <siva.sai@samsung.com>
6 years agoIntroduce EQUAL_EX operation on runtime (#2872)
Prasanna R/System SW /SRI-Bangalore/Engineer/삼성전자 [Thu, 4 Oct 2018 06:58:20 +0000 (12:28 +0530)]
Introduce EQUAL_EX operation on runtime (#2872)

This patch introduces EQUAL_EX operation on runtime.
The related ACL implementation is done in subsequent patches.

Signed-off-by: prasannar <prasanna.r@samsung.com>
6 years ago[neurun] Remove unnecessary code in Linear.h (#2901)
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 4 Oct 2018 05:54:09 +0000 (14:54 +0900)]
[neurun] Remove unnecessary code in Linear.h (#2901)

Remove unnecessary code in `linear/Linear.h`.

- Incorrect `virtual` keyword
- Meaningless `private` keyword

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
6 years ago[neurun] Remove outdated comment (#2900)
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 4 Oct 2018 05:53:49 +0000 (14:53 +0900)]
[neurun] Remove outdated comment (#2900)

Remove outdated TODO comment that has been fixed already.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
6 years ago[neurun] Introduce Enum `Input` for Conv2D (#2899)
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 4 Oct 2018 05:53:20 +0000 (14:53 +0900)]
[neurun] Introduce Enum `Input` for Conv2D (#2899)

Introduce Enum `Input` for Conv2D to avoid using magic numbers from
input list.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
6 years ago[neurun] Remove redundant code (#2898)
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 4 Oct 2018 05:53:02 +0000 (14:53 +0900)]
[neurun] Remove redundant code (#2898)

Remove redundant code. This was supposed to be removed from #2870,
however it was not removed by mistake.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
6 years ago[neurun] Introduce Enum `Input` for AvgPool2D (#2897)
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 4 Oct 2018 05:52:39 +0000 (14:52 +0900)]
[neurun] Introduce Enum `Input` for AvgPool2D (#2897)

Introduce Enum `Input` for AvgPool to avoid using magic numbers for
input list.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
6 years ago[neurun] Introduce EdgeConsistencyChecker (#2893)
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 4 Oct 2018 05:52:20 +0000 (14:52 +0900)]
[neurun] Introduce EdgeConsistencyChecker (#2893)

Introduce EdgeConsistencyChecker which verifies the consistency between
operation-side edges and operand-side edges.

This commit also makes DAGChecker to be an assert condition.

Resolve #2706

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
6 years agoReverting the execution permission (#2905)
Shubham Gupta/System SW /SRI-Bangalore/Engineer/삼성전자 [Thu, 4 Oct 2018 01:38:36 +0000 (07:08 +0530)]
Reverting the execution permission (#2905)

This commit reverse the change of mode for the following files.

Signed-off-by: shubham <shub98.gupta@samsung.com>
6 years agoRemoving the execution permission for NeuralNetowrksEx.h (#2909)
Siva Sai Vaddipati/System SW /SRI-Bangalore/Engineer/삼성전자 [Thu, 4 Oct 2018 00:51:11 +0000 (06:21 +0530)]
Removing the execution permission for NeuralNetowrksEx.h (#2909)

Removing the execution permission for NeuralNetowrksEx.h

Signed-off-by: Siva Sai <siva.sai@samsung.com>
6 years agoAdd generated test for TENSORFLOW_SUM_EX (#2883)
윤지영/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 3 Oct 2018 23:52:28 +0000 (08:52 +0900)]
Add generated test for TENSORFLOW_SUM_EX (#2883)

* Add generated test for TENSORFLOW_SUM_EX

This commit adds generated test for TENSORFLOW_SUM_EX

Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
* Add tensorflowsum_ex_* generated tests to skip list (neurun, tizen and x86_64)

Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
6 years agoAssert moved:exec to compile time for SpaceToDepth (#2878)
Shubham Gupta/System SW /SRI-Bangalore/Engineer/삼성전자 [Tue, 2 Oct 2018 09:27:35 +0000 (14:57 +0530)]
Assert moved:exec to compile time for SpaceToDepth (#2878)

If shape constraint has done well at compilation time, then there is no
need of assert at execution time.

Signed-off-by: shubham <shub98.gupta@samsung.com>
6 years ago[neurun] Remove unused TensorConvert kernels (#2876)
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Tue, 2 Oct 2018 08:00:39 +0000 (17:00 +0900)]
[neurun] Remove unused TensorConvert kernels (#2876)

Remove unused TensorCovert kernels since now we have PermuteLayer
instead.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
6 years agoAssert moved:exec to compile time for DepthToSpace (#2877)
Shubham Gupta/System SW /SRI-Bangalore/Engineer/삼성전자 [Tue, 2 Oct 2018 07:59:36 +0000 (13:29 +0530)]
Assert moved:exec to compile time for DepthToSpace (#2877)

If shape constraint has done well at compilation, then there is no
need of assert at execution time.

Signed-off-by: shubham <shub98.gupta@samsung.com>
6 years agoFix custom op's output resize (#2873)
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Tue, 2 Oct 2018 06:53:20 +0000 (15:53 +0900)]
Fix custom op's output resize (#2873)

Fix custom op TensorFlowMax & TensorFlowSum implementation
Output resize policy: check dimension info

Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
6 years ago[neurun] Move namespace of arm_compute::feature (#2886)
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Tue, 2 Oct 2018 06:47:08 +0000 (15:47 +0900)]
[neurun] Move namespace of arm_compute::feature (#2886)

Move namespace `internal::arm_compute::feature` to
`neurun::util::feature::nhwc`.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
6 years ago[neurun] Introduce .dot dumper (#2787)
김수진/동작제어Lab(SR)/Engineer/삼성전자 [Tue, 2 Oct 2018 06:23:14 +0000 (15:23 +0900)]
[neurun] Introduce .dot dumper (#2787)

* [neurun] Introduce .dot dumper

Part of : #2773

This commit introduces `.dot` dumper, which includes below.

- DotNodeInfo : operation info for dot graph
- DotOperandInfo : operand info for dot graph
- DotDumper : make dot graph and stream dump

Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
* introduce dot_shape()

* change asInt to value()

* Change the directory of dot dumper

6 years ago[neurun] Bring the part that uses source/sink in execution. (#2788)
장지섭/동작제어Lab(SR)/Engineer/삼성전자 [Tue, 2 Oct 2018 05:24:53 +0000 (14:24 +0900)]
[neurun] Bring the part that uses source/sink in execution. (#2788)

This commit bring the part that uses source/sink in execution.
- Bring the part from pureacl.
- Change sourse/sink for 2D/3D to not use `execute_window_loop` in order to avoid unused-param error.

Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
6 years ago[neurun] Fix build bug (#2885)
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Tue, 2 Oct 2018 05:12:14 +0000 (14:12 +0900)]
[neurun] Fix build bug (#2885)

Fix build bug by conflict: shape construction

Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
6 years agoAdd generated tests for EXP_EX (#2858)
윤지영/동작제어Lab(SR)/Engineer/삼성전자 [Tue, 2 Oct 2018 04:46:47 +0000 (13:46 +0900)]
Add generated tests for EXP_EX (#2858)

* Add generated tests for EXP_EX

- Add exp_ex_1D_float for testing 1d tensor
- Add exp_ex_2D_float for testing 2d tensor

Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
* Update generated test skip list for armv7l-tizen, x86_64-linux and armv7l-linux.neurun

Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
6 years ago[neurun] Introduce PermutationInsertionPass (#2861)
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Tue, 2 Oct 2018 04:28:17 +0000 (13:28 +0900)]
[neurun] Introduce PermutationInsertionPass (#2861)

Introduce PermutationInsertionPass that enables to run the model with
heterogeneous backends. This commit also applies this pass in
`Graph::lower`.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
6 years ago[neurun] Rename IVerifier.h to Verifier.h (#2871)
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Tue, 2 Oct 2018 01:51:41 +0000 (10:51 +0900)]
[neurun] Rename IVerifier.h to Verifier.h (#2871)

Rename the filename `IVerifier.h` to `Verifier.h` since it contains not
only `IVerifier` interface but also its derivatives.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
6 years ago[neurun] Move namespace of nnapi::feature classes (#2875)
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Tue, 2 Oct 2018 01:26:43 +0000 (10:26 +0900)]
[neurun] Move namespace of nnapi::feature classes (#2875)

Move namespace `internal::nnapi::feature` to
`neurun::util::feature::nhwc`. This will introduce the namespace
`neurun::util`.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
6 years ago[neurun] Unify converting functions. (#2806)
장지섭/동작제어Lab(SR)/Engineer/삼성전자 [Tue, 2 Oct 2018 01:22:28 +0000 (10:22 +0900)]
[neurun] Unify converting functions. (#2806)

This commit unifies converting(to ACL) functions.
- Bring Swizzle.h from pure_arm_compute.
- Bring converting functions from pure_arm_compute.
- Modify applying the function.

Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
6 years agoSupport ReduceSum operation on CPU (#2862)
윤지영/동작제어Lab(SR)/Engineer/삼성전자 [Tue, 2 Oct 2018 01:18:59 +0000 (10:18 +0900)]
Support ReduceSum operation on CPU (#2862)

* Support ReduceSum operation on CPU

This commit supports ReduceSum operation on ACL.
But this patch uses a CPU rather than GPU.

Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
* Add assertion in TensorFlowSum compilation stage

Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
6 years agoAdd ArithmeticSubtraction codes for support broadcast of SUB (#2864)
윤지영/동작제어Lab(SR)/Engineer/삼성전자 [Tue, 2 Oct 2018 01:18:41 +0000 (10:18 +0900)]
Add ArithmeticSubtraction codes for support broadcast of SUB (#2864)

* Add ArithmeticSubtraction codes for support broadcast of SUB

This patch contains RS7-RuntimeNTools/ComputeLibrary:#11 PR.

Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
* Update license contents for CLArithmeticSubtractionEx

Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
* Remove unnecessary artimetic_add function in acl kernel

Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
6 years agoSupport ReduceMax operation up to 4 dimetions (#2850)
장지섭/동작제어Lab(SR)/Engineer/삼성전자 [Tue, 2 Oct 2018 01:06:10 +0000 (10:06 +0900)]
Support ReduceMax operation up to 4 dimetions (#2850)

This commit supports ReduceMax operation up to 4 dimentions.

Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
6 years agoImplementation of Neg operator in PACL (#2834)
Prasanna R/System SW /SRI-Bangalore/Engineer/삼성전자 [Tue, 2 Oct 2018 01:02:48 +0000 (06:32 +0530)]
Implementation of Neg operator in PACL (#2834)

This patch implements Neg operator in PACL.
NEG operator is used in mtcnn model.
Related issue : #2167

Signed-off-by: prasannar <prasanna.r@samsung.com>
6 years agoIntroduce EQUAL_EX in NeuralNetworksEx.h (#2846)
Prasanna R/System SW /SRI-Bangalore/Engineer/삼성전자 [Mon, 1 Oct 2018 23:54:40 +0000 (05:24 +0530)]
Introduce EQUAL_EX in NeuralNetworksEx.h (#2846)

This patch introduces EQUAL_EX in NeuralNetworksEx.h.
Related issue : #2610

Signed-off-by: prasannar <prasanna.r@samsung.com>
6 years ago[neurun] Refine StageGenerator implementation (#2870)
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Mon, 1 Oct 2018 23:54:08 +0000 (08:54 +0900)]
[neurun] Refine StageGenerator implementation (#2870)

Change index type of each Param for each Node to use
`graph::operand::Index` rather than raw integer. This is for both
acl_cl and cpu backend.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
6 years ago[neurun] Implement Planner with Permute (#2860)
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Mon, 1 Oct 2018 10:06:46 +0000 (19:06 +0900)]
[neurun] Implement Planner with Permute (#2860)

Implement `Planner::visit` method for Permute operation.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
6 years agoSupport HashtableLookup op in PACL runtime (#2708)
Siva Sai Vaddipati/System SW /SRI-Bangalore/Engineer/삼성전자 [Mon, 1 Oct 2018 10:04:51 +0000 (15:34 +0530)]
Support HashtableLookup op in PACL runtime (#2708)

This commit adds support for HashtableLookup op on cpu.
Related issue: #2654
For: #2589

Signed-off-by: Siva Sai <siva.sai@samsung.com>
6 years agoUpdate ACL cross build guide (#2804)
이상규/동작제어Lab(SR)/Principal Engineer/삼성전자 [Mon, 1 Oct 2018 06:20:44 +0000 (15:20 +0900)]
Update ACL cross build guide (#2804)

Related Issue: #2795, #2783

ACL build requires `CROSS_BUILD=1`.

Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
6 years agoRemoving unneccessary TODO for Squeeze (#2855)
Shubham Gupta/System SW /SRI-Bangalore/Engineer/삼성전자 [Mon, 1 Oct 2018 06:13:22 +0000 (11:43 +0530)]
Removing unneccessary TODO for Squeeze (#2855)

Since all Squeeze gtest are passing, this TODO is not required.

Signed-off-by: shubham <shub98.gupta@samsung.com>
6 years agoImplement SquaredDifference in PACL (#2815)
Prasanna R/System SW /SRI-Bangalore/Engineer/삼성전자 [Mon, 1 Oct 2018 06:08:57 +0000 (11:38 +0530)]
Implement SquaredDifference in PACL (#2815)

This patch implements Squared Difference using Sub and Square kernel.
This patch is added as a replacement to previous patch #2570, which realizes Squared Difference using Sub and Mul.

Signed-off-by: prasannar <prasanna.r@samsung.com>
6 years agoAdd generated tests for TENSORFLOW_MAX_EX (#2851)
장지섭/동작제어Lab(SR)/Engineer/삼성전자 [Mon, 1 Oct 2018 04:53:37 +0000 (13:53 +0900)]
Add generated tests for TENSORFLOW_MAX_EX (#2851)

This commit adds generated tests for TENSORFLOW_MAX_EX.

Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
6 years ago[neurun] Revise insertOperation to insertPermute (#2826)
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Mon, 1 Oct 2018 04:25:46 +0000 (13:25 +0900)]
[neurun] Revise insertOperation to insertPermute (#2826)

This commit removes the method `Graph::insertOperation` and adds
`Graph::insertPermute`. Since with `insertOperation` method is a general
method for inserting any node but it can be done with limited situation.
Instead, with `insertPermute` method, it can only Permute node but for
any operand. However since it requires Graph to have dependency on
Backend and Permute node, so it should be extracted to another module
later.

Here are the differences

- insertPermute inserts `Permute` node only while insertOperation can
  insert any type of the node given by parameter,
- insertPermute only adds a node that uses given operand as input
  while insertOperation connects the newly created node's output and the
  original input operation
- insertPermute can be called at LOWERED phase only

This commit also removes the tests for `insertOperation` since
`insertPermute` can be only called at LOWERED phase.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
6 years agoAdd copyright and license for benchmark directory (#2847)
이상규/동작제어Lab(SR)/Principal Engineer/삼성전자 [Mon, 1 Oct 2018 02:49:44 +0000 (11:49 +0900)]
Add copyright and license for benchmark directory (#2847)

- Add Apache 2.0 in our source (Benchmark.*)
- Add our copyright in our modification to ACL (benchmark*.cpp)

Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
6 years ago[neurun] Remove misleading comment (#2829)
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Mon, 1 Oct 2018 02:38:14 +0000 (11:38 +0900)]
[neurun] Remove misleading comment (#2829)

Remove misleading comment in BackendManager - which has been already
fixed, but the comment is not updated.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
6 years agoSupport Exp kernel on ACL (#2835)
윤지영/동작제어Lab(SR)/Engineer/삼성전자 [Mon, 1 Oct 2018 01:49:10 +0000 (10:49 +0900)]
Support Exp kernel on ACL (#2835)

* Support Exp kernel on ACL

This commit implements an exp kernel moduel and related codes on ACL.

Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
* Update License contents

Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
6 years ago[neurun] Make cpu backend impl less verbose (#2848)
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Mon, 1 Oct 2018 01:28:23 +0000 (10:28 +0900)]
[neurun] Make cpu backend impl less verbose (#2848)

Now that we have inputs/outputs/params as `operand::Index` type so we do
not have to write the full type info for backend implementation.
So this commit revises those to use `auto` for cpu backend.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
6 years ago[neurun] Make acl_cl backend impl less verbose (#2849)
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Mon, 1 Oct 2018 01:28:12 +0000 (10:28 +0900)]
[neurun] Make acl_cl backend impl less verbose (#2849)

Now that we have inputs/outputs/params as operand::Index type so we do
not have to write the full type info for backend implementation.
So this commit revises those to use auto for acl_cl backend.

Likewise for cpu backend : #2848

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
6 years agoReverting the execution permission (#2853)
Siva Sai Vaddipati/System SW /SRI-Bangalore/Engineer/삼성전자 [Sun, 30 Sep 2018 23:26:40 +0000 (04:56 +0530)]
Reverting the execution permission (#2853)

After the recent changes, the modes have been changed for the following files.
This commit revokes them.

Signed-off-by: Siva Sai <siva.sai@samsung.com>
6 years agoRevering the mode change (#2852)
Siva Sai Vaddipati/System SW /SRI-Bangalore/Engineer/삼성전자 [Sun, 30 Sep 2018 23:26:27 +0000 (04:56 +0530)]
Revering the mode change (#2852)

This commit reverse the change of mode in NeuralNetworksEx.h

Signed-off-by: Siva Sai <siva.sai@samsung.com>
6 years ago[neurun] Update type of Node Params (#2843)
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 28 Sep 2018 06:01:48 +0000 (15:01 +0900)]
[neurun] Update type of Node Params (#2843)

Params for each node has used raw integer, this commit updates it to
have use `operand::Index`.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
6 years agoUpdate LICENSE and tizen spec (#2842)
이상규/동작제어Lab(SR)/Principal Engineer/삼성전자 [Fri, 28 Sep 2018 06:00:15 +0000 (15:00 +0900)]
Update LICENSE and tizen spec (#2842)

- Eigen is not used any more so BSD and MPL license are removed.
- Some sources under MIT are introduced.

Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
6 years agoChange license MIT to Apache 2.0 for our ARMComputeEx sources (*.cl) (#2840)
이상규/동작제어Lab(SR)/Principal Engineer/삼성전자 [Fri, 28 Sep 2018 05:59:56 +0000 (14:59 +0900)]
Change license MIT to Apache 2.0 for our ARMComputeEx sources (*.cl) (#2840)

*.cl's licenses are changed that was missing in #2704.

Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
6 years ago[neurun] Introduce default constructor for Index (#2831)
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 28 Sep 2018 04:07:46 +0000 (13:07 +0900)]
[neurun] Introduce default constructor for Index (#2831)

Introduce default constructor for `Graph::Index` so wrapping objects can
have a default constructor.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
6 years agoIntroduce TENSORFLOW_SUM_EX operation in PACL (#2836)
윤지영/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 28 Sep 2018 02:45:10 +0000 (11:45 +0900)]
Introduce TENSORFLOW_SUM_EX operation in PACL (#2836)

Add nnapi delegation of TENSORFLOW_SUM_EX operation.
Add param and node for TENSORFLOW_SUM_EX operation.

Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
6 years ago[Tools] Revise operator option generation in select operator (#2769)
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 28 Sep 2018 02:17:04 +0000 (11:17 +0900)]
[Tools] Revise operator option generation in select operator (#2769)

Revise operator option generation in select operator
- Reorder by option type's enum value
- Move import

Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
6 years ago[neurun] Remove duplicated include (#2833)
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 28 Sep 2018 01:43:14 +0000 (10:43 +0900)]
[neurun] Remove duplicated include (#2833)

This commit removes the line for duplicated include.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
6 years agoImplementation of DepthToSpace op in PACL for CPU (#2808)
Shubham Gupta/System SW /SRI-Bangalore/Engineer/삼성전자 [Fri, 28 Sep 2018 01:21:31 +0000 (06:51 +0530)]
Implementation of DepthToSpace op in PACL for CPU (#2808)

This patch will add DepthToSpace op

Signed-off-by: shubham <shub98.gupta@samsung.com>
6 years agoIntroduce TRANSPOSE_CONV_EX in NeuralNetworksEx.h (#2825)
Devansh Bansal/System SW /SRI-Bangalore/Engineer/삼성전자 [Fri, 28 Sep 2018 01:21:07 +0000 (06:51 +0530)]
Introduce TRANSPOSE_CONV_EX in NeuralNetworksEx.h (#2825)

This patch introduces TRANSPOSE_CONV_EX in NeuralNetworksEx.h
Related Issue: #2809

Signed-off-by: b.devansh <b.devansh@samsung.com>
6 years agoIntroduce SpaceToBatchND in PACL (#2810)
Shubham Gupta/System SW /SRI-Bangalore/Engineer/삼성전자 [Fri, 28 Sep 2018 01:20:39 +0000 (06:50 +0530)]
Introduce SpaceToBatchND in PACL (#2810)

This patch introduces SpaceToBatchND op in PACL

Signed-off-by: shubham <shub98.gupta@samsung.com>
6 years agoIntroduce EXP_EX operation in PACL (#2817)
윤지영/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 27 Sep 2018 23:37:32 +0000 (08:37 +0900)]
Introduce EXP_EX operation in PACL (#2817)

Add nnapi delegation of EXP_EX operation.
Add param and node for EXP_EX operation.

Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
6 years agoIntroduce TENSORFLOW_SUM_EX operation on NeuralNetworks.h (#2818)
윤지영/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 27 Sep 2018 09:40:23 +0000 (18:40 +0900)]
Introduce TENSORFLOW_SUM_EX operation on NeuralNetworks.h (#2818)

Add TENSORFLOW_SUM_EX operation in NeuralNetworks.h.

Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
6 years agoIntroduce NEG_EX operation in PACL (#2816)
윤지영/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 27 Sep 2018 09:39:45 +0000 (18:39 +0900)]
Introduce NEG_EX operation in PACL (#2816)

Add nnapi delegation of NEG_EX operation.
Add param and node for NEG_EX operation.

Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
6 years ago[Tools] Fix parser error (#2790)
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 27 Sep 2018 06:12:10 +0000 (15:12 +0900)]
[Tools] Fix parser error (#2790)

Fix parser error when parsing operator option table

Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
6 years agoIntroduce EXP operation on NeuralNetworks.h (#2814)
Prasanna R/System SW /SRI-Bangalore/Engineer/삼성전자 [Thu, 27 Sep 2018 05:59:42 +0000 (11:29 +0530)]
Introduce EXP operation on NeuralNetworks.h (#2814)

This patch introduces ANEURALNETWORKS_EXP_EX in NeuralNetworks.h
EXP operator is used in mtcnn model.
Related issue : #2167

Signed-off-by: prasannar <prasanna.r@samsung.com>
6 years agoIntroduce NEG operation on NeuralNetworks.h (#2813)
Prasanna R/System SW /SRI-Bangalore/Engineer/삼성전자 [Thu, 27 Sep 2018 02:22:31 +0000 (07:52 +0530)]
Introduce NEG operation on NeuralNetworks.h (#2813)

This patch introduces ```ANEURALNETWORKS_NEG_EX``` in NeuralNetworks.h
NEG operator is used in mtcnn model.
Related issue : #2167

Signed-off-by: prasannar <prasanna.r@samsung.com>
6 years agoAdd TensorFlowSum custom operator in tflite interpreter (#2797)
윤지영/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 27 Sep 2018 02:03:18 +0000 (11:03 +0900)]
Add TensorFlowSum custom operator in tflite interpreter (#2797)

This commit registers TensorFlowSum operator in tflite interpreter.
It contains the actual behavior of TensorFlowSum.

Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
6 years agoIntroduce DEPTH_TO_SPACE operation into PACL (#2781)
Shubham Gupta/System SW /SRI-Bangalore/Engineer/삼성전자 [Mon, 24 Sep 2018 10:00:20 +0000 (15:30 +0530)]
Introduce DEPTH_TO_SPACE operation into PACL (#2781)

This patch will introduce Depth_To_Space into pure_arm_compute (#2765)

Signed-off-by: shubham <shub98.gupta@samsung.com>