이한종/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 25 Jan 2019 06:46:58 +0000 (15:46 +0900)]
Avoid redefinition of macro functions (#4320)
Avoid redefinition of CAST_CL and CAST_NE that generates warning.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 25 Jan 2019 06:24:33 +0000 (15:24 +0900)]
Fix concat for not rank 4 (#4317)
Now concat operation rank is not 4 can run using cpu implementation
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 25 Jan 2019 00:40:39 +0000 (09:40 +0900)]
[neurun] Move Model class to `model` (#4303)
Move Model class to `model` from `graph` since `Model` represents pure
neural network primitive elements.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 24 Jan 2019 03:33:56 +0000 (12:33 +0900)]
Update the description of Gather (#4277)
This commit updates the description of the types that CLGather and CLGatherKerenl support.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 24 Jan 2019 00:00:56 +0000 (09:00 +0900)]
Introduce BUILD_UBEN option (#4309)
* Introduce BUILD_UBEN option
This commit introduces BUILD_UBEN option to fix Tizen build break.
* Set default value
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 23 Jan 2019 23:44:24 +0000 (08:44 +0900)]
[PACL] Replace CLReduceOperation with CLReduceMean. (#4304)
This commit replaces CLReduceOperation with CLReduceMean in case of mean operation.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 23 Jan 2019 23:44:12 +0000 (08:44 +0900)]
[PACL] Replace CLLogicalNot with CLBitwiseNot (#4299)
This commit replaces CLLogicalNot of ARMComputeEx with CLBitwiseNot of ARMCompute
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Wed, 23 Jan 2019 23:39:33 +0000 (08:39 +0900)]
Move input and output setting into Executor class (#4293)
* Move input and output setting into Executor class
Move source (input setting) and sink (output setting) into Executor class
* Fix source list name to same as directory name
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 23 Jan 2019 23:39:13 +0000 (08:39 +0900)]
[neurun] Extract sizeOfDataType as a function (#4307)
Extract sizeOfDataType as a function which were implemented in class
`Object`.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Wed, 23 Jan 2019 10:24:14 +0000 (19:24 +0900)]
Introduce uben (#4301)
* Introduce uben
This commit introduces uben (Micro BENchmark) suite based on nonius
benchmark framework.
Initial version includes convolution 3x3 (with unit stride) benchmark
over ARM Compute CL Direct/GEMM/Winograd kernels.
* Fix a bug in ker_info method
* Use CLWinogradConvolutionLayer correctly
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 23 Jan 2019 07:35:21 +0000 (16:35 +0900)]
[PACL] Replace CLPixelWiseDivision with CLArithmeticDivision (#4302)
This commit replaces CLPixelWiseDivision of ARMComputeEx with CLArithmeticDivision of ARMCompute.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
김수진/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 23 Jan 2019 05:04:05 +0000 (14:04 +0900)]
[neurun] Support activation for AvgPool2D (#4297)
Related : #2147
This commit supports unimplemented activation for AvgPool2D.
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 23 Jan 2019 04:59:15 +0000 (13:59 +0900)]
[PACL] Add the declaration of asTensorInfo (#4298)
This commit adds the declaration of asTensorInfo.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
김수진/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 22 Jan 2019 09:59:13 +0000 (18:59 +0900)]
[neurun] Support explicit padding for AvgPool2D (#4278)
* [neurun] Support explicit padding for AvgPool2D
Related : #2147
This commit supports explicit padding for `AvgPool2D`.
- Merge `Implicit` and `Explicit` `Node` compared to `PACL`
- Remove related test skip list
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
* move explicit_padding to param, add TODO comment
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 22 Jan 2019 06:28:22 +0000 (15:28 +0900)]
Move source and sink in wrapper (#4290)
* Move source and sink in wrapper
Move source and sink setting and call in wrapper into Executor class
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
* Remove header include
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 22 Jan 2019 04:14:12 +0000 (13:14 +0900)]
[PACL] Replace CLBatchToSpaceND with CLBatchToSpaceLayer (#4283)
This commit replaces CLBatchToSpaceND of ARMComputeEx with CLBatchToSpaceLayer of ARMCompute
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Tue, 22 Jan 2019 03:50:40 +0000 (09:20 +0530)]
Provide Q8 sup and comments correction for Arg operation (#4268)
1. Provide Q8 sup
2. comments correction
Signed-off-by: shubham <shub98.gupta@samsung.com>
김수진/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 22 Jan 2019 02:37:02 +0000 (11:37 +0900)]
[neurun] Support Relu1, Relu6 activation (#4287)
I've tried to pass `GeneratedTests.avg_pool_float_4`
However I met the missmatching in result.
I figured out it is the problem of `activation` Relu6.
This commit supports `Relu1` and `Relu6` activation in `ActivationBuilder`.
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
[neurun] Move kernels into backends (#4126)
* [neurun] Move kernels into backends
Kernels are part of backends, so they should be in subdir of their own
backends
* [neurun] Change namespaces of kernels to include backends
Exchange namespaces of backends and kernels in kernels
* [neurun] Add "backend" namespace to kernels
Add "backend" namespace to kernels
Signed-off-by: Poshshoev Dilshodzhon <d.poshshoev@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 22 Jan 2019 01:57:17 +0000 (10:57 +0900)]
[PACL] Add an overloading function of asTensorInfo (#4284)
This commit addes an overloading function of asTensorInfo that can be set arm_compute::DataType optionally.
Some kernels of ARMCompute require data type that does not exist in nnapi spec. However, the PACL currently
sets the data type of ARMCompute to a mapped data type of nnapi. There is a need for a function to optionally set
the data type of ARMCompute.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 22 Jan 2019 01:46:07 +0000 (10:46 +0900)]
[neurun] Move State member to Compiler from Plan (#4286)
Move State member variable to Compiler from Plan.
Plan is just part of compilation so it would be more proper.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 22 Jan 2019 01:38:08 +0000 (10:38 +0900)]
Introduce executor class (#4281)
Introduce executor class to control execution phase on runtime internal
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 22 Jan 2019 01:37:21 +0000 (10:37 +0900)]
Update frontend execution API (#4282)
- Free execution object
- Return bad state when setInput & setOutput using memory object (NYI)
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Mon, 21 Jan 2019 10:06:30 +0000 (15:36 +0530)]
Remove Passing gtests from skiplist (#4269)
After integration of ACL v18.11, some gtests are
passing which were earlier failing.
This patch will remove all passing gtests from
skiplist.
Signed-off-by: shubham <shub98.gupta@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Mon, 21 Jan 2019 10:06:17 +0000 (15:36 +0530)]
Changes in CLPermuteExKernel.cpp (#4273)
1. move calc shape func of validation in inside of
if condition since it is getting used only when that if
condition is true.
2. remove tensorshape local func since misc/shapecalc/utils has
same function which can be directly used.
Signed-off-by: shubham <shub98.gupta@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 21 Jan 2019 08:16:28 +0000 (17:16 +0900)]
Update validation check on frontend execution (#4267)
* Update validation check on frontend execution
Update validation check on NNAPI frontend execution's input & output setting
- Check length
- Check shape
- Print error message
Abstract to access model
- Operand index of input & output
- Type & shape compare
* Fix bug
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
이성재/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 21 Jan 2019 07:03:39 +0000 (16:03 +0900)]
Update broken link in `howoto` document. (#4280)
- Updated link to Dockerfile which broken by directory restructuring.
Signed-off-by: Sung-Jae Lee <sj925.lee@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 21 Jan 2019 04:40:31 +0000 (13:40 +0900)]
Remove CLArithmeticSubtractionEx from ArmComputeEx (#4275)
This commit removes CLArithmeticSubtractionEx from ArmComputeEx.
- remove CLArithmeticSubtractionEx
- replace CLArithmeticSubtractionEx with CLArithmeticSubtraction since CLArithmeticSubtraction supports the broadcast.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
이성재/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 21 Jan 2019 03:49:32 +0000 (12:49 +0900)]
Update broken links to model repository in `howto.md`. (#4274)
- Update broken model zoo links as `tflite` leaves contrib.
Signed-off-by: Sung-Jae Lee <sj925.lee@samsung.com>
김수진/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 18 Jan 2019 06:20:02 +0000 (15:20 +0900)]
[neurun] Remove skip list for TrivialTest.AddThree (#4266)
Related : #4211
This commit removes skip list for `TrivialTest.AddThree`, since #4263 and #4257 are merged.
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
김수진/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 18 Jan 2019 03:22:06 +0000 (12:22 +0900)]
[neurun] Support to initialize 2D constant (#4257)
* [neurun] Support to initialize 2D constant
Until now, `ConstantInitializer` in `neurun` has supported only for `FullyConnected`.
This commit supports to initialize 2D constant for all other operations.
* Change some comment
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 18 Jan 2019 03:20:24 +0000 (12:20 +0900)]
Move type converter (#4253)
* Move type converter
Move type converter from graph to util
It makes more independent internal code with NNAPI
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
* Fix test and remove deprecated include
* Fix comment
* Divide function declaration and definition
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 18 Jan 2019 00:53:06 +0000 (09:53 +0900)]
Use config manager for logging (#4261)
Use config manager for neurun logging enable/disable
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Fri, 18 Jan 2019 00:51:29 +0000 (06:21 +0530)]
Correction in comment in Pad kernel (#4262)
this patch will correct the comment which tells
what the pad kernel does.
Signed-off-by: shubham <shub98.gupta@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Fri, 18 Jan 2019 00:51:16 +0000 (06:21 +0530)]
Renaming ArgMinMax to ArgOperation (#4238)
In order to have genericity so that any arg op
can be added, this patch will rename occurances
of ArgMinMax to ArgOperation.
Also, this patch will correct few of the comments.
Signed-off-by: shubham <shub98.gupta@samsung.com>
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 18 Jan 2019 00:46:44 +0000 (09:46 +0900)]
[neurun] Fix when `Graph::lower` is called twice (#4263)
This commit fixes to handle `Graph::lower` is called multiple times.
Fix #4258
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
김수진/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 17 Jan 2019 22:24:44 +0000 (07:24 +0900)]
[neurun] Remove skip list for TrivialTest.AddTwo (#4260)
Related : #4211
This commit removes skip lit for `TrivialTest.AddTwo`, since #4249 is merged.
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 17 Jan 2019 09:40:46 +0000 (18:40 +0900)]
Remove dependence of data format(NHWC) of PACL from kernels of PAD (#4014)
This commit removes dependece of data format(NHWC) of PACL runtime from kernels of PAD.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
Prasanna R/SNAP /SRI-Bangalore/Engineer/삼성전자 [Thu, 17 Jan 2019 09:37:52 +0000 (15:07 +0530)]
Add Broadcast support for PReLU in PACL (#4072)
This patch adds broadcast support for PReLU in PACL.
Signed-off-by: prasannar <prasanna.r@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Thu, 17 Jan 2019 09:32:48 +0000 (15:02 +0530)]
Optimisation in Permute kernel (#4202)
Since many of the assignment operations are present
in kernel which can be easily avoided and hence can
lead to some optmisation.
This patch will remove the unnecessary assignemnt operations.
Signed-off-by: shubham <shub98.gupta@samsung.com>
김수진/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 17 Jan 2019 08:12:50 +0000 (17:12 +0900)]
[neurun] Support other types in PermuteSource/PermuteSink (#4249)
* [neurun] Support other types in PermuteSource/PermuteSink
This commit supports other types in `PermuteSource`/`PermuteSink`.
These would make some mismatching TCs to match such as `TrivialTest.AddTwo`.
* Remove TODO
* Remove casting value already casted
* Remove unnecessary casting
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 17 Jan 2019 05:45:25 +0000 (14:45 +0900)]
Update enabled validation test list (#4256)
Update enabled validation tests by supporting ADD operation on armv7l acl-cl backend
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
Support LogcalNot in runtime and nnapi_delegate (#4048)
* Support LogcalNot in runtime and nnapi_delegate
Add support of LogcalNot in PACL runtime and nnapi_delegate
Signed-off-by: Poshshoev Dilshodzhon <d.poshshoev@samsung.com>
* Add logical_not to _kernel_program_map and remove from skip list
Add logical_not to _kernel_program_map and remove from skip list
Signed-off-by: Poshshoev Dilshodzhon <d.poshshoev@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 17 Jan 2019 03:53:19 +0000 (12:53 +0900)]
[PACL] Fix reshape bug. (#4254)
This commit fixes reshape bug that reshape does not support non 4-D input and 4-D output in PACL.
- Fix GenericReshapeLayer bug
- Set the apply_dim_correction input shape of GenericFullyConnectedLayer to `false`
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 17 Jan 2019 03:36:08 +0000 (12:36 +0900)]
Fix NNAPI validation check for compilation preference (#4255)
Fix NNAPI frontend validation check for compilation preference
- Bad state: try to set after finish is called
- Bad preference value
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 17 Jan 2019 01:06:59 +0000 (10:06 +0900)]
[PACL] Support TransposeConv for GPU (#4223)
This commit supports TransposeConv operation for GPU.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 16 Jan 2019 09:45:35 +0000 (18:45 +0900)]
Bring CLDeconvolutionLayer from ACL into libARMComputeEx (#4250)
This commit brings CLDeconvolutionLayer in order to fix a bug computing wrong the output's dimension value.
- Bring CLDeconvolutionLayer
- Bring a function that computes the output's dimension value
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Wed, 16 Jan 2019 06:57:39 +0000 (12:27 +0530)]
Remove unused num_elem variable from Arg CL Host file (#4234)
This patch will remove unused num_elem variable from
CLArgMinMaxKernel.cpp
Signed-off-by: shubham <shub98.gupta@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 15 Jan 2019 09:35:54 +0000 (18:35 +0900)]
Test packaging for x64 build (#4247)
Test packaging for x64 build. It's to prepare neurun x64 test
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 15 Jan 2019 09:10:43 +0000 (18:10 +0900)]
More strict acl extention lib debug build option (#4241)
More strict acl extention lib debug build option: unused-parameter and unused-function check
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
김수진/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 15 Jan 2019 06:58:47 +0000 (15:58 +0900)]
[neurun] Remove skip list and add framework test for ADD op (#4216)
* [neurun] Remove skip list and add framework test for ADD op
This commit removes skip lit and add framework test for ADD op.
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
* add GeneratedTests.add_broadcast_4D_2D_after_nops_float_nnfw, remove framework tests
김수진/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 15 Jan 2019 04:38:36 +0000 (13:38 +0900)]
[neurun] Fix broadcast bug for ADD op and remove skip list (#4243)
This commit fixes broadcast bug for ADD op and remove skip list `TrivialTest.BroadcastAddTwo`.
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 15 Jan 2019 03:41:35 +0000 (12:41 +0900)]
Framework test list for neurun cpu and acl (#4244)
Add framework test list for neurun armv7l cpu and acl cl
Add framework test list for neurun x86-64 cpu
Workaround test packaging on makefile for CI test
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
김수진/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 15 Jan 2019 03:07:47 +0000 (12:07 +0900)]
[neurun] Correct wrong assumptions for Reshape and Concat (#4210)
* [neurun] Correct wrong assumptions for Reshape and Concat
This commit corrects wrong assumptions for `Reshape` and `Concat` (refer to belows)
- Reshape : https://developer.android.com/ndk/reference/group/neural-networks#group___neural_networks_1ggaabbe492c60331b13038e39d4207940e0a535e7e99383ee49456c8671843b93a59
- Concat : https://developer.android.com/ndk/reference/group/neural-networks#group___neural_networks_1ggaabbe492c60331b13038e39d4207940e0a44cbea825c4b224dd3ea757e9b1f65ed
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
* Correct concat count check
* Change concat input range
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 15 Jan 2019 01:05:49 +0000 (10:05 +0900)]
Fix nnapi delegate (#4240)
Fix nnapi delegate bug for gbs build
Fix switch-case indent
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 15 Jan 2019 00:18:18 +0000 (09:18 +0900)]
Use more strict compile option for tflite library (#4196)
Use more strict compile option for tflite library
Remove use of size_t as possible
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 15 Jan 2019 00:17:45 +0000 (09:17 +0900)]
Fix infering wrong padding in TransposeConv (#4237)
This commit fixes infering wrong padding in TransposeConv.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 14 Jan 2019 11:07:17 +0000 (20:07 +0900)]
Remove deprecated test for transpose convolution (#4235)
Remove deprecated test: int32 transpose convolution
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
Vishal Keshav/SNAP /SRI-Bangalore/Engineer/삼성전자 [Mon, 14 Jan 2019 11:04:56 +0000 (16:34 +0530)]
Transpose Convolution fix (#4137)
Fixed the incorrect filter index
Pad change reverted
Signed-off-by: Vishal Keshav <vishal.k1@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 14 Jan 2019 10:56:11 +0000 (19:56 +0900)]
Update NNAPI extend spec not supporting int32 tensor operation (#4229)
Change int32 tensor to quantized int tensor for ANEURALNETWORKS_TRANSPOSE_CONV_EX
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
이춘석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 14 Jan 2019 10:53:33 +0000 (19:53 +0900)]
Fix build error by unused variable on CLLogicalNotKernel.cpp (#4233)
- Fix build error by unused variable on CLLogicalNotKernel.cpp
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 14 Jan 2019 10:39:59 +0000 (19:39 +0900)]
Fix permute vaildation check bug (#4226)
Fix permute validation check to check validate rank only
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
Add CL Kernel of LogicalNot (#3636)
This commit adds implementation of LogicalNot in CL kernel
Related issues: #3596
Signed-off-by: Poshshoev Dilshodzhon <d.poshshoev@samsung.com>
Refactoring of Shape::broadcast (#4132)
Remove 4 loops and creating two temp vectors and handle all checks in
one loop
Signed-off-by: Poshshoev Dilshodzhon <d.poshshoev@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Mon, 14 Jan 2019 08:16:44 +0000 (13:46 +0530)]
Bug fix for output shape in Permute (#4201)
This patch will fix the bug in calc output shape in Permute
Signed-off-by: shubham <shub98.gupta@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Mon, 14 Jan 2019 01:23:26 +0000 (06:53 +0530)]
Add more validations in Permute CL host file (#4188)
This patch will add validations to check input
and permutation vector.
Signed-off-by: shubham <shub98.gupta@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Sun, 13 Jan 2019 21:50:32 +0000 (03:20 +0530)]
Correction in variable names in CL Arg op files (#4125)
Since somewhere in code variable name is argminmax_axis and
somewhere it is just axis along with comments having name as
axis.So, this patch will corrects the comments and replaces
all argminmax_axis to axis to have a uniformity in variable
name for axis.
Signed-off-by: shubham <shub98.gupta@samsung.com>
김수진/On-Device Lab(SR)/Engineer/삼성전자 [Sun, 13 Jan 2019 21:48:36 +0000 (06:48 +0900)]
[neurun] Support broadcast for ADD op (#4203)
Related : #4173
This commit supports broadcast for ADD op and enable the gtest.
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
Synch tools/nnapi_quickcheck with current changes (#4197)
Since we don't build tools/nnapi_quickcheck, it's far behind from
current state. So, this change updates all files of this dir to be able
to build and run them
Signed-off-by: Poshshoev Dilshodzhon <d.poshshoev@samsung.com>
[neurun] Fix CLTensor::setBuffer to work with new acl (#4199)
Related issue: 4185
In new versions of acl that we use now, its functions used in this
method were either updated or removed. This change updates
CLTensor::setBuffer to be able to address these new changes
Signed-off-by: Poshshoev Dilshodzhon <d.poshshoev@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Fri, 11 Jan 2019 01:00:29 +0000 (06:30 +0530)]
Remove unused func from CLArgMinMaxKernel file (#4200)
This patch will remove unused func from CLArgMinMaxKernel.h
Signed-off-by: shubham <shub98.gupta@samsung.com>
김수진/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 11 Jan 2019 00:50:18 +0000 (09:50 +0900)]
[neurun] Enable basic ADD op (#4186)
* [neurun] Enable basic ADD op
Related : #4173
This commit migrate basic `ADD` op that is already exist in `PACL` to `neurun`.
- Neon cannot be supported now.
- TensorBuilder should has not only `arm_compute::CLTensor` but also `arm_compute::Tensor`.
- Remove skip test `GeneratedTests.add`
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
* Remove neon codes
* Remove unused codes
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 11 Jan 2019 00:45:01 +0000 (09:45 +0900)]
[neurun] Extract Operand LowerInfo from Node (#4184)
* [neurun] Extract Operand LowerInfo from Node
As `model::operand::Node` is pure node info so LowerInfo should not be
Node class. `graph::Graph` holds them instead.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
* Do not pass Graph object for DotOperandInfo
Pass only `LowerInfo` instead
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
김수진/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 11 Jan 2019 00:33:27 +0000 (09:33 +0900)]
[PACL] Extract internal kernels to libs (#4144)
* [PACL] draft: Extract internal kernels to libs
Related : #4060
- Introduce `libs/kernels` and `libs/kernels/simple`
- Move `ARMComputeEx` to `libs/kernels`
- Move internal kernels of `PACL` to `libs/kernels/simple`
- Remove `FeatureLoggingLayer`
- Some utils are copied from `PACL` to `libs/kernels/simple`
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
* remove *.clembed
* Revert kernels location
* Fix build failure
* Change directory to misc
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 10 Jan 2019 08:14:36 +0000 (17:14 +0900)]
Define more skiplist and install (#4195)
Define more skiplist: neurun x86, neurun armv7l cpu
Install all skiplists for target platform
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 10 Jan 2019 04:38:21 +0000 (13:38 +0900)]
Remove warning from acl on gbs build (#4194)
Remove ignored-attributes warning from acl ex header on gbs build
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 10 Jan 2019 03:59:04 +0000 (12:59 +0900)]
Use more strict compile option for misc and profiling (#4193)
Use more strict compile option for misc and profiling
Remove use of size_t as possible
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Thu, 10 Jan 2019 01:10:38 +0000 (06:40 +0530)]
Correcting name to asPadStrideInfo from asPadStringInfo (#4190)
Since by asPadStringInfo, one cant think if it is having
stride info or not. Hence, this patch wil rename this func
name to asPadStrideInfo.
Signed-off-by: shubham <shub98.gupta@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 10 Jan 2019 01:07:23 +0000 (10:07 +0900)]
Strict build option for acl (#4163)
* Strict build option for acl
- Add build option for more strict
- Except warning unused-parameter and unused-function for validation check
- Add assert, log, debug build option for debug configure
- Fix bug
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
* Fix validatoin check & broadcast bug
* Fix update_window_and_padding call using or operation
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 10 Jan 2019 00:28:07 +0000 (09:28 +0900)]
[neurun] Remove `Index::asInt` (#4189)
Remove `asInt` method from `Index` since it should return its underlying
type. Calls to `asInt()` has been replaced with `value()`.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Wed, 9 Jan 2019 09:32:29 +0000 (18:32 +0900)]
Enable neurun x86 build (#4187)
Enable neurun x86 build - use cpu kernel only
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 9 Jan 2019 07:54:45 +0000 (16:54 +0900)]
[neurun] Do not use raw int in operand::Context (#4182)
This commit revises the type of key for `compiler::operand::Context` to
be `Index` class rather than raw int. Since Index does not support
`operator<` the conainer became `unordered_map`.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 9 Jan 2019 07:28:50 +0000 (16:28 +0900)]
Change macro `DEBUG` to `ARM_COMPUTE_DEBUG_ENABLED` (#4183)
This commit changes `DEBUG` option to `ARM_COMPUTE_DEBUG_ENABLED` in ARMComputeEx.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 9 Jan 2019 04:32:05 +0000 (13:32 +0900)]
[neurun] Move Backend info to PermuteNode (#4180)
In order for StageGenerator to not have dependency on backends.
This commit moves the backend info for input/output from StageGenerator
to creation of PermuteNode.
(which is created only in PermutationInsertionPass)
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Wed, 9 Jan 2019 04:15:33 +0000 (13:15 +0900)]
Use -fPIC build flag for tflite only (#4175)
* Use -fPIC build flag for tflite only
Instead of global setting, use -fPIC build flag for tenroflow lite only
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
* Specify to build tensorflow lite library static
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Wed, 9 Jan 2019 00:51:25 +0000 (06:21 +0530)]
Correction in validation func in CL Host files (#4139)
Since ARM_COMPUTE_RETURN_ERROR_ON_MSG() is an "ERROR_ON" macro,
which means an error will be thrown if the expression evaluates to "true"
Hence, checks need to be inverted.
This patch will correct the checks in th following files:
1. CLBatchToSpaceNDKernel.cpp
2. CLDepthToSpaceKernel.cpp
3. CLPadLayerKernel.cpp
4. CLReduceOperationKernel.cpp
5. CLSpaceToDepthKernel.cpp
Signed-off-by: shubham <shub98.gupta@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Wed, 9 Jan 2019 00:48:51 +0000 (09:48 +0900)]
Remove warning from external acl and tensorflow (#4181)
Remove warning from external acl and tensorflow by using SYSTEM keyword
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 8 Jan 2019 10:12:50 +0000 (19:12 +0900)]
Use acl pre-built binary (#4179)
Remove workaround to avoid build fail by acl version update
Use pre-built acl v18.11 binary
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
장지섭/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 8 Jan 2019 07:10:18 +0000 (16:10 +0900)]
Update version of ARMComputeEx to 18.11 (#3877)
* Update version of ARMComputeEx to 18.11
This commit updates the version of ARMComputeEx to 18.11.
- Change the name of the kernel of StridedSlice and PadLayer to StridedSliceEx and PadLayerEx to prevent duplicated name.
- Remove the code related to fixed point(QS8, QS16).
- Update CLKernelLibraryEx to 18.11.
- Apply using the same context for CLKernelLibraryEx as CLScheduler.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
* Change config files to build with acl v18.11
This commit changes config files to build with acl v18.11.
- Change download url of acl from v18.05 to v18.11
- Change acl to always build
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
* Enable the build of neurun with v18.11
This commit enables the build of neurun with v18.11.
For detail : https://github.sec.samsung.net/STAR/nnfw/pull/3877#issuecomment-227170
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
* acl benchmark patch
* Fix missing cl kernel function
This commit adds missing cl kernel function to map list.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 8 Jan 2019 03:42:05 +0000 (12:42 +0900)]
Remove acl link for neurun ineternal (#4174)
Remove acl link for neurun ineternal
Becasue neurun internal is independent with acl, we can remove acl link
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 8 Jan 2019 03:41:40 +0000 (12:41 +0900)]
Environment variable to disable compile (#4172)
Introduce environment variable to disable compile
Return with bad state in exection phase when compile state is NOT_COMPILED
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 8 Jan 2019 01:08:19 +0000 (10:08 +0900)]
[neurun] Move Permute Type struct to PermuteNode (#4166)
In order for StageGenerator to not have dependency on backends.
This commit moves Permute Type decision from StageGenerator to creation
of PermuteNode(which is created only in PermutationInsertionPass).
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이한종/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 8 Jan 2019 01:03:15 +0000 (10:03 +0900)]
[neurun] Remove StageGenerator log prints (#4167)
Remove StageGenerator log prints since it is not useful anymore.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 7 Jan 2019 08:42:11 +0000 (17:42 +0900)]
Revert acl cmake build (#4164)
Revert acl cmake build to update acl v18.11
This reverts commit
5e4e7e27d7b4a20ef300bb4577c4e14369db7e06.
This reverts commit
7d484e5c3acd4b3ad8039226a25f78d26748be0a.
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
이상규/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 7 Jan 2019 02:50:12 +0000 (11:50 +0900)]
Add pause time between each tests in benchmark-op (#4158)
To reduce the impact of high temperature due to prior running,
this patch will add pause time between each run.
Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Mon, 7 Jan 2019 02:16:30 +0000 (07:46 +0530)]
Correcting assertion in transposeConv CPU implementation (#4153)
This patch will correct the position of bracket which has
been wrongly placed.
Signed-off-by: shubham <shub98.gupta@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Fri, 4 Jan 2019 03:09:36 +0000 (08:39 +0530)]
Remove unnecessary headers from CLReduce and CLArg ops (#4124)
This patch will remove unnecessary headers from CLReduce
and CLArg op files. And Replace CLArray to CLScheduler
since only CLScheduler header file is enough.
Signed-off-by: shubham <shub98.gupta@samsung.com>
Shubham Gupta/SNAP /SRI-Bangalore/Engineer/삼성전자 [Fri, 4 Jan 2019 01:23:29 +0000 (06:53 +0530)]
Remove unused variables in CLArgMinMaxKernel file (#4127)
This patch will remove unused variables from
CLArgMinMaxKernel.cpp
Signed-off-by: shubham <shub98.gupta@samsung.com>
오형석/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 4 Jan 2019 01:21:42 +0000 (10:21 +0900)]
Block gtest install (#4146)
- Update add_extdirectory macro for EXCLUDE_FROM_ALL setting
- Disable gtest install using add_subdirectory with EXCLUDE_FROM_ALL
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
Removing ambiguity from pack gtests (#4074)
This patch rectifies the name of one of the inputs.
Signed-off-by: b.devansh <b.devansh@samsung.com>