Remove space_to_batch gtest from skiplist armv7l-linux (#3079)
This patch will remove passing space_to_batch gtest from skiplist.
Signed-off-by: shubham <shub98.gupta@samsung.com>
남궁석/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 12 Oct 2018 06:37:59 +0000 (15:37 +0900)]
Enable to check operand status in pureacl (#2438)
This commit will enable to check if operand exists or not
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 12 Oct 2018 03:36:04 +0000 (12:36 +0900)]
[neurun] Rename IBackendConfig to IConfig (#3098)
Rename `IBackendConfig` to `IConfig` since it is in namespace
`neurun::backend` so "Backend" in the class name is redundant.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
장지섭/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 12 Oct 2018 02:06:24 +0000 (11:06 +0900)]
Seperate functions of Cast.h into the declaration and the definition (#3104)
This commit seperates functions of Cast.h into the declaration and the definition.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
김수진/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 11 Oct 2018 22:14:27 +0000 (07:14 +0900)]
[neurun] Show other inputs on dot graph (#3078)
* [neurun] Show other inputs on dot graph
This commit shows other inputs(kernel,bias..) on dot graph.
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
* Add options to switch showing range
* merge condition
Fix max_pool op to support quant type (#3016)
This patch will fix the max_pool_quant8_3 Gtest
And adds support for Quantization.
Signed-off-by: shubham <shub98.gupta@samsung.com>
Move TRANSPOSE_CONV_EX case to AddOperationEx in model.cc (#3103)
This patch moves TRANSPOSE_CONV_EX case to AddOperationEx since it's an Ex operation.
Signed-off-by: prasannar <prasanna.r@samsung.com>
[neurun] Make codegen impl less verbose (#3084)
Since we have the inputs and outputs as operand::Index we do not need the full
scope resolution, hence changing those to auto for codegen.
Signed-off-by: Praveen D N <praveen.dn@samsung.com>
Add NNAPI Delegation Changes for L2_Normalization op (#3080)
This patch adds NNAPI Delegation Changes for L2_Normalization op.
Signed-off-by: prasannar <prasanna.r@samsung.com>
NNAPI delegation changes for Batch_To_Space_ND op (#3057)
This patch gives the required NNAPI_Delegation changes for Batch_To_Space_op.
Signed-off-by: prasannar <prasanna.r@samsung.com>
Remove Unsupported Datatye Cases in SpaceToDepth (#3077)
NN api defines only for U8 and F32 dataype for spacetodepth op.
Hence, leaving code only for supported dataypes.
Signed-off-by: shubham <shub98.gupta@samsung.com>
Remove S8,U32,S32 dataype cases from DepthToSpace (#3076)
This patch will remove S8, U32 and S32 cases since nnapi doesnot defines for them in DepthToSpace
Signed-off-by: shubham <shub98.gupta@samsung.com>
Add LRN codes in NE-ACL to supp both odd/even val (#2908)
This patch contains RS7-RuntimeNTools/ComputeLibrary#47 PR
For NEON-ACL LRN support
Updated license contents for NENormalizationLayerEx
Signed-off-by: shubham <shub98.gupta@samsung.com>
Fix for avg_pool_quant8_4 Gtest (#3017)
This patch will fix avg_pool_quant8_4 gtest
Signed-off-by: shubham <shub98.gupta@samsung.com>
nnfw: RSQRT support for Extended ACL. (#3039)
SQRT is supported as an Activation function in ACL, so RSQRT is also
added as in Activation Function in case of Extended ACL for nnfw.
Full framework for Activation function is added with this patch so
that if similar functionality is required for some other operator
then it can be added easily.
Signed-off-by: Tanuj Tekriwal <tanuj.tekri@samsung.com>
Introduce BATCH_TO_SPACE_ND in runtime. (#3058)
This patch introduces BATCH_TO_SPACE_ND in PACL runtime.
Signed-off-by: prasannar <prasanna.r@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 11 Oct 2018 04:18:23 +0000 (13:18 +0900)]
Remove cmake for acl build (#3092)
Remove cmake for acl build (not used anymore)
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
Add NNAPI_Delegation changes for HashTableLookup op (#3082)
This patch adds NNAPI_Delegation changes for HashTableLookup op
Signed-off-by: prasannar <prasanna.r@samsung.com>
Introduce Custom(Abs) op in runtime (#3035)
This commit introduces Custom(Abs) op in PACL runtime.
Related issue: #2828
Signed-off-by: Siva Sai <siva.sai@samsung.com>
윤현식/동작제어Lab(SR)/Principal Engineer/삼성전자 [Wed, 10 Oct 2018 10:54:35 +0000 (19:54 +0900)]
[PureACL] remove unused code (tanh activation code) (#3072)
Remove tanh fuseCode since we don't have tanh supported.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
Updating skiplist for HashtableLookup (#2907)
With #2655 and #2708 both the removed gtests are passing,
hence removing them from the skiplist.
Signed-off-by: Siva Sai <siva.sai@samsung.com>
김수진/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 10 Oct 2018 10:33:01 +0000 (19:33 +0900)]
[neurun] Fix condition typo in lowering (#3071)
This commit fixes condition typo in lowering.
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
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>
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>
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>
Add CL Kernels for SquaredDifference op. (#3037)
This patch adds CL Kernels for SquaredDifference op.
Signed-off-by: prasannar <prasanna.r@samsung.com>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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
Add Apache License (#3005)
This patch adds Apache License to SimpleEmbedding lookup and SquaredDiff layer
Signed-off-by: shubham <shub98.gupta@samsung.com>
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>
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>
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>
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>
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>
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>
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>
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>
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
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>
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>
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>
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>
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>
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>
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>
Add Apache License to L2Normalization Op (#2931)
This patch adds Apache License to L2Normalization Op.
Signed-off-by: prasannar <prasanna.r@samsung.com>
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>
Add Apache License to SquaredDifference Op (#2930)
This patch adds Apache License to Squared Difference Op
Signed-off-by: prasannar <prasanna.r@samsung.com>
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>
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>
Add Apache License to HashTableLookup op (#2933)
This patch adds Apache License to HashTableLookup op.
Signed-off-by: prasannar <prasanna.r@samsung.com>
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>
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>
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>
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>
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>
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>
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>
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>
Reverting the execution permission (#2905)
This commit reverse the change of mode for the following files.
Signed-off-by: shubham <shub98.gupta@samsung.com>
Removing the execution permission for NeuralNetowrksEx.h (#2909)
Removing the execution permission for NeuralNetowrksEx.h
Signed-off-by: Siva Sai <siva.sai@samsung.com>
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>
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>
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>
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>
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>
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>
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
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>