Prasanna R/SNAP /SRI-Bangalore/Engineer/삼성전자 [Fri, 14 Dec 2018 01:15:15 +0000 (06:45 +0530)]
Remove AND GTests from skiplist (#3982)
This patch removes AND GTests from skiplists.
Signed-off-by: prasannar <prasanna.r@samsung.com>
Prasanna R/SNAP /SRI-Bangalore/Engineer/삼성전자 [Fri, 14 Dec 2018 01:13:54 +0000 (06:43 +0530)]
Add DataTypes supported by BinaryLogicalOp CL Kernel (#4012)
This patch adds DataType supported by BinaryLogicalOp CL Kernel
Signed-off-by: prasannar <prasanna.r@samsung.com>
Prasanna R/SNAP /SRI-Bangalore/Engineer/삼성전자 [Fri, 14 Dec 2018 01:13:29 +0000 (06:43 +0530)]
Add CL Kernel for PRELU (#3961)
This patch adds single CL Kernel for PReLU implemented as,
f(x) = alpha * x for x < 0, f(x) = x for x >= 0.
As of now (#3901 ) PReLU implemented as,
f(x) = Relu(x) + (negative_alpha * Neg(x, activation=Relu))
Scope:
- Optimization
- No switch between CPU & GPU
Signed-off-by: prasannar <prasanna.r@samsung.com>
Add tests for LogicalNot (#3600)
This commit adds tests for LogicalNot op
Signed-off-by: Poshshoev Dilshodzhon <d.poshshoev@samsung.com>
Prasanna R/SNAP /SRI-Bangalore/Engineer/삼성전자 [Fri, 14 Dec 2018 00:58:59 +0000 (06:28 +0530)]
Propose unified kernel for Comparison operators and bug fix (#3914)
This patch proposed unified CL Kernel for Comparison operator
This patch includes:
- CL kernel for EQUAL/NOT_EQUAL
- TRUTH value bug fix
- CL Kernel that can be easily extended for GT, LT, GE, LE
Signed-off-by: prasannar <prasanna.r@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 13 Dec 2018 10:14:56 +0000 (19:14 +0900)]
Update namespace and ifdef for libs/misc (#4022)
Update ifdef for libs/misc
Update namespace libs/misc as nnfw::misc
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 13 Dec 2018 09:42:59 +0000 (18:42 +0900)]
Move tools/test_driver to tests/scripts (#4019)
* Move tools/test_driver to tests/scripts
Move tools/test_driver to tests/scripts
Update test path
Workaround for CI
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
* Comment in Makefile for workaround
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 13 Dec 2018 09:39:44 +0000 (18:39 +0900)]
[neurun] Introduce ConfigManager (#4021)
Introduce ConfigManager that manages global configuration data.
The values are from Environment Variables.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
Prasanna R/SNAP /SRI-Bangalore/Engineer/삼성전자 [Thu, 13 Dec 2018 09:36:13 +0000 (15:06 +0530)]
Correction in GTests of Equal_Ex with truth value as 0xff (255) (#3927)
This patch corrects GTests of Equal_Ex with truth value as 0xff (255)
Related issue: #3823 #3851
As quoted by @hseok82-oh
>IMO, use 0xff (255) for true, and 0x00 (0) for false in test.
Signed-off-by: prasannar <prasanna.r@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Thu, 13 Dec 2018 09:27:16 +0000 (14:57 +0530)]
Add gtest to test broadcasting in Notequal op (#3717)
This patch will test broadcasting in notequal op
Signed-off-by: shubham <shub98.gupta@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 13 Dec 2018 07:37:15 +0000 (16:37 +0900)]
Update namespace in libs/tflite/ext (#4015)
Update namespace in libs/tflite/ext to nnfw::tflite
Reduce namespace depth for custom op
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
김수진/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 13 Dec 2018 04:02:25 +0000 (13:02 +0900)]
[neurun] Use IObject in PermuteLayer (#4005)
This commit uses `IObject` insted of `ITensor` in `PermuteLayer`, since we don't want to use `map/unmap` functions explicitly and to get any `ACL` dependencies.
Related : #3279
Part of : #3873
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 13 Dec 2018 03:21:05 +0000 (12:21 +0900)]
Fix external tensorflow lite build cmake (#4007)
- Remove BUILD_TFLITE_BENCHMARK_MODEL check for profiling source code
- Fix typo
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 13 Dec 2018 03:20:41 +0000 (12:20 +0900)]
Rename tensorflow lite interpreter test (#4009)
* Rename tensorflow lite interpreter test
Rename tensorflow lite interpreter test by comparing with result of tensorflow
: tf_test -> tflite_test
* Fix build option description
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
윤지영/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 13 Dec 2018 02:14:31 +0000 (11:14 +0900)]
Move how to run guideline into readme file (#3987)
This patch moves the how to run guideline into readme file
Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 13 Dec 2018 01:43:53 +0000 (10:43 +0900)]
Add argmax quant8 to skiplist (#4004)
Add argmax quant8 to skiplist: test fail
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 13 Dec 2018 01:35:01 +0000 (10:35 +0900)]
Move libs/misc/profiling to libs/profiling (#3997)
Move libs/misc/profiling to libs/profiling for profiling library
Update cmake to link profiling library
Update header path for profiling
Update cmake for tensorflow build to support build with BUILD_TFLITE_BENCHMARK_MODEL option
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
ArgMinMax CL kernel for ArgMax and ArgMin (#3867)
The changes introduce the argmax and argmin CL kernel in one cl file.
Signed-off-by: Praveen D N <praveen.dn@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Thu, 13 Dec 2018 00:53:26 +0000 (06:23 +0530)]
Add gtest to test LogicalAnd op for 2D and 3D tensors (#3938)
This patch will add gtests to test logicalAnd op for 2D and 3D tensors
Signed-off-by: shubham <shub98.gupta@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Thu, 13 Dec 2018 00:53:06 +0000 (06:23 +0530)]
Add gtest to test LogicalOr for 2D and 3D tensors (#3939)
This patch will add gtests to test logicalOr for 2d and 3d tensors
Signed-off-by: shubham <shub98.gupta@samsung.com>
윤지영/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 13 Dec 2018 00:51:13 +0000 (09:51 +0900)]
Pass explicitly local workgroup size to enqueue in acl (#3996)
This patch allows the local workgroup size(_lws_hint) value pass to enqueue function
Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 13 Dec 2018 00:17:01 +0000 (09:17 +0900)]
[neurun] Introduce Config.lst (#3999)
Config.lst is a macro list with fields "Name", "Type", "Default value".
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
김수진/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 12 Dec 2018 11:16:02 +0000 (20:16 +0900)]
[neurun] Fix wrong size calculation when permuting (#3998)
This commit fixes wrong size calculation when permuting.
It should be data size not pointer size.
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
윤지영/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 12 Dec 2018 10:42:03 +0000 (19:42 +0900)]
Apply dim correction to neg operator (#3986)
This allows to apply dim correction to Neg operator
Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
김수진/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 12 Dec 2018 10:13:08 +0000 (19:13 +0900)]
[neurun] Change ifdef guard in Convert.h (#3990)
This commit changes ifdef guard in `Convert.h` to correct.
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
[neurun] Use host memory in acl_cl (#3974)
Related issue:3735
Add CLTensor::setBuffer to be able to use host memory from the acl_cl
Signed-off-by: Poshshoev Dilshodzhon <d.poshshoev@samsung.com>
Prasanna R/SNAP /SRI-Bangalore/Engineer/삼성전자 [Wed, 12 Dec 2018 09:02:19 +0000 (14:32 +0530)]
Add GTests to test broadcast of LogicalAND op (#3950)
This patch adds GTests to test broadcast of LogicalAND op.
Signed-off-by: prasannar <prasanna.r@samsung.com>
Prasanna R/SNAP /SRI-Bangalore/Engineer/삼성전자 [Wed, 12 Dec 2018 09:02:02 +0000 (14:32 +0530)]
ADD 1D, 2D, 3D, 4D GTests for RSQRT (#3932)
This patch adds 1D, 2D, 3D, 4D GTests for RSQRT
Signed-off-by: prasannar <prasanna.r@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Wed, 12 Dec 2018 08:56:23 +0000 (17:56 +0900)]
Fix build bug by tflite library namespace (#3993)
Fix build bug by tflite library namespace change
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Wed, 12 Dec 2018 08:39:07 +0000 (17:39 +0900)]
Update namespace and ifdef for tflite library (#3968)
* Update namespace and ifdef for tflite library
Make tflite/extend directory for ported codes from public tensorflow
Update namespace except extended directory
Update ifdef in header
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
* Rename tflite/extend to tflite/ext
* Fix format
* Update ifdef in ext
* Flatten namespace nnfw::tflite::interp to nnfw::tflite
* Add comment for endif
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Wed, 12 Dec 2018 08:38:08 +0000 (14:08 +0530)]
Add missing axises comment (#3979)
This patch will add missing axises comment.
Signed-off-by: shubham <shub98.gupta@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Wed, 12 Dec 2018 08:27:54 +0000 (13:57 +0530)]
Fix wrong radius of LRN in NENormalization (#3977)
Fix wrong radius setting for LRN
Signed-off-by: shubham <shub98.gupta@samsung.com>
이춘석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Wed, 12 Dec 2018 07:38:50 +0000 (16:38 +0900)]
Make TestGenerated file for each version (#3975)
* Make TestGenerated file for each version
On coverage build, build have failed since all_generated_tests.cpp is too
large. This patch will generate a test file for each NNAPI version.
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
* Extract common code from TestGenerated_XX.cpp
- Extract common code from TestGenerated_XX.cpp
윤지영/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 12 Dec 2018 05:13:21 +0000 (14:13 +0900)]
Support boolean input param in tflite_run (#3958)
This patch assigns random boolean value to boolean input param.
Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Wed, 12 Dec 2018 04:03:42 +0000 (09:33 +0530)]
Add gtest to test Exp op for 3D/4D tesnors (#3797)
This patch will add gtest to test Exp op for tesnors having
dimensions other than 4
Signed-off-by: shubham <shub98.gupta@samsung.com>
Prasanna R/SNAP /SRI-Bangalore/Engineer/삼성전자 [Wed, 12 Dec 2018 02:00:35 +0000 (07:30 +0530)]
Add CL Kernel calls for LogicalAnd op (#3970)
This patch adds CL Kernel calls for LogicalAnd op
Signed-off-by: prasannar <prasanna.r@samsung.com>
Prasanna R/SNAP /SRI-Bangalore/Engineer/삼성전자 [Wed, 12 Dec 2018 01:59:53 +0000 (07:29 +0530)]
Remove setting of false to apply_dim_correction in compilation,cc for Broadcast support ops (#3976)
This patch removes setting of false to apply_dim_correction in compilation,cc for Broadcast support ops.
Signed-off-by: prasannar <prasanna.r@samsung.com>
Prasanna R/SNAP /SRI-Bangalore/Engineer/삼성전자 [Wed, 12 Dec 2018 01:52:17 +0000 (07:22 +0530)]
Add GTests to test broadcast of LogicalOR (#3949)
This patch adds GTests to test broadcast of LogicalOR.
Signed-off-by: prasannar <prasanna.r@samsung.com>
Prasanna R/SNAP /SRI-Bangalore/Engineer/삼성전자 [Wed, 12 Dec 2018 01:14:51 +0000 (06:44 +0530)]
Redefine REDUCE_MIN in NeuralNetworkEx.h (#3925)
This patch redefines REDUCE_MIN in NeuralNetworkEx.h
explaining it supports INT32 also
Reference tf.reduce_min
input_tensor: The tensor to reduce. Should have numeric type.
Signed-off-by: prasannar <prasanna.r@samsung.com>
Prasanna R/SNAP /SRI-Bangalore/Engineer/삼성전자 [Wed, 12 Dec 2018 01:14:03 +0000 (06:44 +0530)]
Redefine REDUCE_MIN in NeuralNetworkEx.h (#3926)
This patch redefines REDUCE_MIN in NeuralNetworkEx.h
explaining it supports INT32 also
Reference tf.reduce_max
```
input_tensor: The tensor to reduce. Should have numeric type.
```
Signed-off-by: prasannar <prasanna.r@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Wed, 12 Dec 2018 01:13:43 +0000 (06:43 +0530)]
Add gtest to test neg op for int datatype (#3930)
This patch will add gtest to test neg for int datatype
Signed-off-by: shubham <shub98.gupta@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Wed, 12 Dec 2018 01:11:20 +0000 (06:41 +0530)]
Correction in GTests of NotEqual_Ex with truth value as 0xff (255) (#3929)
This patch corrects GTests of NotEqual_Ex with truth value as 0xff (255)
Related issue: #3823 #3851
Signed-off-by: shubham <shub98.gupta@samsung.com>
Prasanna R/SNAP /SRI-Bangalore/Engineer/삼성전자 [Tue, 11 Dec 2018 07:04:53 +0000 (12:34 +0530)]
Fix Typo in internal/op/* files (#3973)
This patch fixes typo issues in internal/op/*.h files
changes:
```defualt -> default```
Signed-off-by: prasannar <prasanna.r@samsung.com>
이춘석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Tue, 11 Dec 2018 06:15:18 +0000 (15:15 +0900)]
Fix typo on SimpleArgMinMax.cc (#3972)
fix typo : defualy -> default
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
윤지영/동작제어Lab(SR)/Engineer/삼성전자 [Tue, 11 Dec 2018 05:15:19 +0000 (14:15 +0900)]
Use TFLiteConverter instead of deprecated API in tensorflow_model_freezer (#3917)
Add prerequires section in readme
Use tflite v1.12 tool to support the extended operators
Fix some tc errors
Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
윤지영/동작제어Lab(SR)/Engineer/삼성전자 [Tue, 11 Dec 2018 04:35:50 +0000 (13:35 +0900)]
Add sample py example for some operator (#3919)
Added operators:
ARGMAX, ARGMIN, LOGICAL_AND, LOGICAL_NOT, LOGICAL_OR, STACK, UNSTACK
Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Tue, 11 Dec 2018 02:41:31 +0000 (11:41 +0900)]
Update tflite library name in cmake (#3967)
Update tflite library name in cmake to nnfw_lib_tflite
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
김수진/동작제어Lab(SR)/Engineer/삼성전자 [Tue, 11 Dec 2018 02:08:18 +0000 (11:08 +0900)]
[neurun] Remove default values in View.h (#3966)
This commit removes default values of `feature_index_to_byte_offset`'s parameter in `View.h`.
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Tue, 11 Dec 2018 00:39:06 +0000 (09:39 +0900)]
Update namespace of nnapi library (#3965)
Update namespace of nnapi library: nnfw::support::nnapi to nnfw::nnapi
Update header ifdef
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
김수진/동작제어Lab(SR)/Engineer/삼성전자 [Mon, 10 Dec 2018 11:26:29 +0000 (20:26 +0900)]
[neurun] Fix wrong parameter in nchw View.h (#3963)
This commit fixes wrong parameter for 3 rank in nchw `View.h`.
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Mon, 10 Dec 2018 11:23:19 +0000 (20:23 +0900)]
Rename libs/util to libs/misc and update cmake (#3962)
Rename libs/util to libs/misc
Use static build only
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
윤지영/동작제어Lab(SR)/Engineer/삼성전자 [Mon, 10 Dec 2018 08:26:57 +0000 (17:26 +0900)]
Resolve inconsistencies in the boolean data values (#3957)
* Resolve inconsistencies in the boolean data values
Since the data is stored in the index value of the input/output tensors,
it is modified to use the index value when it accesses the data.
Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
* Use meaningful variable names
Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Mon, 10 Dec 2018 06:40:20 +0000 (15:40 +0900)]
Move libs/support/tflite to libs/tflite (#3920)
Move libs/support/tflite to libs/tflite
Update header include path
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Mon, 10 Dec 2018 06:32:43 +0000 (15:32 +0900)]
Change nnapi library name and update cmake (#3956)
Chane nnapi library name: nnfw_support_nnapi -> nnfw_lib_nnapi
Specify library type as static
Remove unused nnapi library link
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
Prasanna R/SNAP /SRI-Bangalore/Engineer/삼성전자 [Mon, 10 Dec 2018 06:26:02 +0000 (11:56 +0530)]
Remove redundant .cl files regarding reduce operations (#3952)
This patch removes redundant .cl files regarding reduce operations
In issue: #3448
Since Unification of reduce kernel is merged,
removal of existing implementation can be removed
Hence removing the same.
Signed-off-by: prasannar <prasanna.r@samsung.com>
이춘석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Mon, 10 Dec 2018 06:02:11 +0000 (15:02 +0900)]
Update skiplist for tizen (#3953)
Now, the skiplist of tizen is the same as linux's
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
Prasanna R/SNAP /SRI-Bangalore/Engineer/삼성전자 [Mon, 10 Dec 2018 05:13:42 +0000 (10:43 +0530)]
Add PReLU CL Kernel calls from PACL (#3916)
This patch adds PReLU CL Kernel calls from PACL.
Signed-off-by: prasannar <prasanna.r@samsung.com>
Prasanna R/SNAP /SRI-Bangalore/Engineer/삼성전자 [Mon, 10 Dec 2018 05:07:17 +0000 (10:37 +0530)]
Add 3D 4D Gtests for SquaredDifferenceEx op (#3933)
This patch adds 3D 4D Gtests for SquaredDifferenceEx op.
Signed-off-by: prasannar <prasanna.r@samsung.com>
Prasanna R/SNAP /SRI-Bangalore/Engineer/삼성전자 [Mon, 10 Dec 2018 01:58:05 +0000 (07:28 +0530)]
Add GTests for SQRT_EX 3D 4D case (#3934)
This patch GTests for SQRT_EX 3D 4D case
Signed-off-by: prasannar <prasanna.r@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Mon, 10 Dec 2018 01:56:28 +0000 (07:26 +0530)]
Add validations for Neg op in CLKernel.cpp (#3931)
This patch will add validations req to be tested
before executing neg op
Signed-off-by: shubham <shub98.gupta@samsung.com>
Prasanna R/SNAP /SRI-Bangalore/Engineer/삼성전자 [Mon, 10 Dec 2018 01:55:18 +0000 (07:25 +0530)]
Remove unwanted apache license from RSQRT Gtest (#3937)
This patch removes unwanted apache license from RSQRT Gtest ```rsqrt_ex_float1.mod.py```
Signed-off-by: prasannar <prasanna.r@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Mon, 10 Dec 2018 01:53:41 +0000 (07:23 +0530)]
Implement LogicalOr op in PACL (#3944)
This patch will add implementation of LogicalOr op
in PACL
Signed-off-by: shubham <shub98.gupta@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Mon, 10 Dec 2018 01:44:09 +0000 (07:14 +0530)]
Add gtest to test quant8 datatype for NotEqual Op (#3757)
This patch will add gtest to test quant8 datatype for NotEqual Op
Signed-off-by: shubham <shub98.gupta@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Mon, 10 Dec 2018 01:42:50 +0000 (07:12 +0530)]
Add NNAPI delegation for LOGICAL_OR (#3904)
This patch adds NNAPI delegation for LOGICAL_OR.
Related issue: #3459
Signed-off-by: shubham <shub98.gupta@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Mon, 10 Dec 2018 01:28:33 +0000 (06:58 +0530)]
Generic EmbeddingLookup kernel to handle tensors upto rank=4 (#3794)
This patch unifies the 3 different embedding lookup kernel into
one kernel so as to make a generic cl to support upto 4d tensors.
Signed-off-by: shubham <shub98.gupta@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Mon, 10 Dec 2018 01:28:15 +0000 (06:58 +0530)]
Defining default constructor in Pack and Unpack layer (#3907)
This patch will define default constructor in Pack and
Unpack Layers
Signed-off-by: shubham <shub98.gupta@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Mon, 10 Dec 2018 01:27:55 +0000 (06:57 +0530)]
Generic HashTableLookup kernel to handle tensors upto rank=4 (#3798)
This patch unifies the 3 different hashtable lookup kernel into
one kernel so as to make a generic cl to support upto 4d tensors.
Signed-off-by: shubham <shub98.gupta@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Mon, 10 Dec 2018 01:26:14 +0000 (06:56 +0530)]
Add gtest to test Argmax op for tensors of QASYMM8 dataype (#3802)
This patch will add gtest for Argmax to test for tensors of
QASYMM8 dataype
Signed-off-by: shubham <shub98.gupta@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Mon, 10 Dec 2018 01:23:18 +0000 (06:53 +0530)]
Remove passing gtests from skiplist (#3908)
This patch will remove all passing gtests from skiplist
- local_response_norm_float_4
- GeneratedTests.unpack*
Signed-off-by: shubham <shub98.gupta@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Mon, 10 Dec 2018 01:22:13 +0000 (06:52 +0530)]
Add license comments in SimpleUnpackLayer* (#3905)
This patch will add license commnets in SimpleUnpackLayer.cc
andSimpleUnpackLayer.h
Signed-off-by: shubham <shub98.gupta@samsung.com>
Devansh Bansal/SNAP /SRI-Bangalore/Engineer/삼성전자 [Mon, 10 Dec 2018 01:21:57 +0000 (06:51 +0530)]
Add gtest for split with negative axis. (#3915)
This patch adds a test case for split with negative axis.
Signed-off-by: b.devansh <b.devansh@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Mon, 10 Dec 2018 00:59:59 +0000 (09:59 +0900)]
Remove generated test cpp (#3921)
Remove generated test cpp as these files are unused since #3874.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 7 Dec 2018 07:48:57 +0000 (16:48 +0900)]
Move Padding module from `internal` to `util`. (#3882)
Move Padding module from `internal` to `util`.
Part of #3764
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
Prasanna R/SNAP /SRI-Bangalore/Engineer/삼성전자 [Fri, 7 Dec 2018 07:48:06 +0000 (13:18 +0530)]
Add NNAPI delegation for LOGICAL_AND (#3903)
This patch adds NNAPI delegation for LOGICAL_AND.
Related issue: #3459
Signed-off-by: prasannar <prasanna.r@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 7 Dec 2018 07:46:20 +0000 (16:46 +0900)]
Validation check for addOperation (#3911)
- Check operation type is validate
- Add try-catch exception and return ANEURALNETWORKS_BAD_STATE when exception is raised
- Set return value when addOperationEx is not failed
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 7 Dec 2018 07:15:23 +0000 (16:15 +0900)]
Add ExtendCMakeFunction module and add_subdirectories function (#3912)
- Add ExtendCMakeFunction module to simplify cmake
- Introduce add_subdirectories function to support adding all subdirectories at once. (It's comes from nncc repo)
- Use add_subdirectories function in libs
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
Prasanna R/SNAP /SRI-Bangalore/Engineer/삼성전자 [Fri, 7 Dec 2018 07:06:03 +0000 (12:36 +0530)]
Unify REDUCE_SUM and REDUCE_MAX operation in CL kernel (#3885)
This patch unifies REDUCE_SUM and REDUCE_MAX operation in CL kernel
This is done to reduce code redundancy.
Related issue: #3771
Signed-off-by: prasannar <prasanna.r@samsung.com>
Prasanna R/SNAP /SRI-Bangalore/Engineer/삼성전자 [Fri, 7 Dec 2018 06:44:13 +0000 (12:14 +0530)]
Add comments to define alpha input in internal/op/PReLU.cc (#3900)
This patch adds comments to define alpha input in internal/op/PReLU.cc.
Signed-off-by: prasannar <prasanna.r@samsung.com>
Prasanna R/SNAP /SRI-Bangalore/Engineer/삼성전자 [Fri, 7 Dec 2018 06:44:04 +0000 (12:14 +0530)]
Change in kernel name for LogicalBinaryOp (#3913)
This patch corrects the change in kernel name for LogicalBinaryOps (AND & OR)
Signed-off-by: prasannar <prasanna.r@samsung.com>
Prasanna R/SNAP /SRI-Bangalore/Engineer/삼성전자 [Fri, 7 Dec 2018 06:43:21 +0000 (12:13 +0530)]
Add PReLU layer in PACL (#3901)
This patch adds PReLU layer in PACL.
Issue: #3459 for Epic4
PReLU implemented as,
f(x) = Relu(x) + (negative_alpha * Neg(x, activation=Relu))
Signed-off-by: prasannar <prasanna.r@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 7 Dec 2018 03:24:37 +0000 (12:24 +0900)]
Move libs/support/nnapi to libs/nnapi (#3895)
Move libs/support/nnapi to libs/nnapi
Update header include path
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 7 Dec 2018 02:13:20 +0000 (11:13 +0900)]
Update namespace and header define for cpp14 library (#3896)
Update namespace to nnfw::cpp14
Update define to __NNFW_CPP14_xxx__
Update usage of cpp14 library to updated namespace
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Fri, 7 Dec 2018 01:20:10 +0000 (06:50 +0530)]
Remove unused headers from Unpack op in internal/layers (#3906)
This patch will remove unused headers included in Unpack
related files in internal/layers
Signed-off-by: shubham <shub98.gupta@samsung.com>
Prasanna R/SNAP /SRI-Bangalore/Engineer/삼성전자 [Fri, 7 Dec 2018 01:14:24 +0000 (06:44 +0530)]
Add GTests for PReLU op (#3899)
This patch adds GTest for PReLU op
Issue: #3459.
Signed-off-by: prasannar <prasanna.r@samsung.com>
Adding assertions in TransposeConv op. (#3887)
This patch adds necessary assertions in TransposeConv.
Signed-off-by: b.devansh <b.devansh@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 7 Dec 2018 00:50:43 +0000 (09:50 +0900)]
Return error for not finished model compile (#3897)
When NNAPI call try to compile model not finished, return BAD_STATE error
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 6 Dec 2018 09:08:34 +0000 (18:08 +0900)]
Remove template runtime (#3845)
Remove experimental template runtime in runtimes/template
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
Vishal Keshav/SNAP /SRI-Bangalore/Engineer/삼성전자 [Thu, 6 Dec 2018 08:27:47 +0000 (13:57 +0530)]
Implementation of CUSTOM Unpack op (#3669)
ACL implementation
Used CLSubtensor and CLReshape
CLPermute to overcome backward swizzle
Signed-off-by: Vishal Keshav <vishal.k1@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 6 Dec 2018 08:11:38 +0000 (17:11 +0900)]
Define GENERATE_FILES for test generation (#3890)
Introduce GENERATE_FILES environment variable to skip running generate_test.sh
If you set GENERATE_FILES 0, makefile skip running generate_test.sh. Default is 1 (run generate_test.sh default)
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 6 Dec 2018 08:09:41 +0000 (17:09 +0900)]
Move include/nnfw/std (#3892)
Move include/nnfw/std to libs/cpp14/
Update include path
Update cmake to include cpp14 library using interface
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 6 Dec 2018 08:08:54 +0000 (17:08 +0900)]
[neurun] Move Swizzle.h to acl backend dir (#3893)
Move Swizzle.h to acl backend dir since it is only used there.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
[neurun] Remove backend/cpu dependancy from PermutateSource/Sink (#3889)
Related issue: #3888
Remove backend/cpu dependancy from PermutateSource/Sink
Signed-off-by: Poshshoev Dilshodzhon <d.poshshoev@samsung.com>
이춘석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 6 Dec 2018 06:39:36 +0000 (15:39 +0900)]
Fix wrong radius of LRN cl kernel (#3861)
* Fix wrong radius setting of LRN cl kernel
Fix wrong radius setting for LRN
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
* Remove unused variables
Prasanna R/SNAP /SRI-Bangalore/Engineer/삼성전자 [Thu, 6 Dec 2018 06:07:22 +0000 (11:37 +0530)]
Change the order of Unpack Input parameter in NeuralNetworksEx.h (#3865)
This patch changes the order of Unpack Input parameter in NeuralNetworksEx.h
The num is seconds and axis is third param.
Signed-off-by: prasannar <prasanna.r@samsung.com>
Prasanna R/SNAP /SRI-Bangalore/Engineer/삼성전자 [Thu, 6 Dec 2018 06:06:56 +0000 (11:36 +0530)]
Redefine PACK_EX in NeuralNetworksEx.h (#3864)
As per @hseok82-oh's words
>I found out runtime is updated to use additional parameter num and this PR will use this change. Please update comment PACK_EX in NeuralNetworksEx.h using another PR.
This patch redefines PACK_EX in NeuralNetworksEx.h.
Signed-off-by: prasannar <prasanna.r@samsung.com>
[neurun] Do backends as shared and update BackendManager to support plugin (#3793)
Related issue: #3703
This is the initial step of converting backends to plugins
Update BackendManager to creage a map using plugins
Signed-off-by: Poshshoev Dilshodzhon <d.poshshoev@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Thu, 6 Dec 2018 06:04:19 +0000 (11:34 +0530)]
Remove #failed comment from skiplist (#3878)
This pacth will remove #failed comment from skiplist
Signed-off-by: shubham <shub98.gupta@samsung.com>
Prasanna R/SNAP /SRI-Bangalore/Engineer/삼성전자 [Thu, 6 Dec 2018 05:13:31 +0000 (10:43 +0530)]
Introduce enum for EQUALITY op. (#3853)
This patch introduces an enum for EQUALITY operations
like EQUAL, NOT_EQUAL.
This is related to #3851.
This is reduce code redundancy to a good extent in future.
Signed-off-by: prasannar <prasanna.r@samsung.com>
Prasanna R/SNAP /SRI-Bangalore/Engineer/삼성전자 [Thu, 6 Dec 2018 04:37:23 +0000 (10:07 +0530)]
Disable nnapi_test of Custom/TensorflowMax (#3850)
This patch disables nnapi_test of Custom/TensorflowMax.
Since TensorFlowMax is builtin op as REDUCE_MAX from V1.12,
we are removing the trace of TensorFlowMax as a custom op.
This is required for PRs #3849 #3846
Signed-off-by: prasannar <prasanna.r@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 6 Dec 2018 01:49:02 +0000 (10:49 +0900)]
Remove UPDATE_MODULE in makefile (#3891)
Remove UPDATE_MODULE environment variable to ignore submodule update: deprecated feature because we don't use submodule
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>