최성진/동작제어Lab(SR)/Principal Engineer/삼성전자 [Fri, 18 May 2018 08:05:20 +0000 (17:05 +0900)]
[Gather OP] add gather operation fall-back (#1267)
* [Gather OP] add gather operation fall-back
This commit introduces GATHER operation which is the first version.
- currently, only for float32 and 1D
Signed-off-by: SungJin Choi <lotieye.choi@samsung.com>
* [Gather Op] Gather operation to CPU executor
This commit adds Gather operation in CpuExecutor.
-currently, only for float32 and 1D
Signed-off-by: SungJin Choi <lotieye.choi@samsung.com>
장지섭/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 18 May 2018 06:32:16 +0000 (15:32 +0900)]
Introduce extended CAPI akin to `ANeuralNetworksModel_addOperation()` to support extended operators. (#1256)
This commit introduces extended CAPI that plays a role similar to ANeuralNetworksModel_addOperation() to support extended operators.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
김정현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Fri, 18 May 2018 05:36:29 +0000 (14:36 +0900)]
Ignore submodules for format-checker (#1270)
format-checker is failed when submodule is modified.
This is not what we want, so ignore submodule's state in format-checker.sh
Signed-off-by: Junghyun Kim <jh0822.kim@samsung.com>
오형석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Fri, 18 May 2018 01:06:32 +0000 (10:06 +0900)]
TFLiteFile tool: Fix shape bug, option (#1265)
Fix tensor shape bug when tensor is scalar
Fix generating option for new supported operators
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
이춘석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Fri, 18 May 2018 01:05:24 +0000 (10:05 +0900)]
Arm Cross Toolchain update to 6.3 on nnfw_docker (#1263)
issue: #1251
To support GeneretedTest build, we need at least gcc 6.3.
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
오형석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Fri, 18 May 2018 01:03:18 +0000 (10:03 +0900)]
Fix test tflite file bug (#1259)
Change to fixed tflite file and update md5sum
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
Enabling LOGISTIC operation in CpuExecutor (#1258)
This commit enables the Logistic activation in CpuExecutor and adds the necessary framework tests file to test logistic operation
Signed-off-by: Siva Sai <siva.sai@samsung.com>
서상민/동작제어Lab(SR)/Senior Engineer/삼성전자 [Thu, 17 May 2018 23:33:31 +0000 (08:33 +0900)]
Move libs/util/include/util to include/util (#1255)
To be consistent with other header files, this patch moves
`libs/util/include/util` to `include/util` and accordingly modifies
`libs/util/CMakeLists.txt`.
Signed-off-by: Sangmin Seo <sangmin7.seo@samsung.com>
윤현식/동작제어Lab(SR)/Principal Engineer/삼성전자 [Thu, 17 May 2018 23:31:23 +0000 (08:31 +0900)]
renamed previous mul to matmul (#1254)
In previous PR (#1238), `mul` was added for two matrix multiplication.
However, it turned out that `mul` must be element-wise mul, not matmul (matrix mul).
For this reason previous `mul` was renamed to `matmul` and test was changed to `disabled`.
Also, 'matmul2x2.tflite` was uploaded into `npuarchive.mooo.com`.
(previous `mul2x2.tflite` in `npuarchive.mooo.com` will be deleted after this PR is merged.)
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
이상규/동작제어Lab(SR)/Principal Engineer/삼성전자 [Thu, 17 May 2018 11:14:15 +0000 (20:14 +0900)]
Enable DIV, SUB and STRIDED_SLICE tests (#1257)
SIGSEGV was resolved. Thus related tests are enabled.
Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
김정현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Thu, 17 May 2018 09:25:05 +0000 (18:25 +0900)]
Print exact operation name (#1252)
This commit fixes to print exact operation name using operation index.
If the index is not found, use option_type as before.
Signed-off-by: Junghyun Kim <jh0822.kim@samsung.com>
장지섭/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 17 May 2018 07:53:54 +0000 (16:53 +0900)]
Introduce enum `OperationCodeEx` to support extended operators (#1223)
This commit introduces enum `OperationCodeEx` to support extended operators.
The list of added operations:
- ANEURALNETWORKS_EX_CAST
- ANEURALNETWORKS_EX_GATHER
- ANEURALNETWORKS_EX_TOPK_V2
- ANEURALNETWORKS_EX_TENSORFLOW_MAX
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
김용섭/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 17 May 2018 07:46:29 +0000 (16:46 +0900)]
Introduce profiling feature on test_driver.sh (#1248)
Now `test_driver.sh --profile` runs profiling by `oprofile`. To profile,
`oprofile` should be installed on target machine and needs root
permission.
Signed-off-by: Yongseop Kim <yons.kim@samsung.com>
최형규/동작제어Lab(SR)/Senior Engineer/삼성전자 [Thu, 17 May 2018 04:31:51 +0000 (13:31 +0900)]
Prepare scripts for Coverity (#1236)
* Prepare scripts for Coverity
- Add run_coverity.sh
- Add Coverity_Tizen_SelfVerification_2017.07.0.sh
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
* Download Coverity script from server
- Download Coverity script from server, instead of maintaining in this repo
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
* Receive name of Coverity script as a parameter
- Receive name of Coverity script as a parameter
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
* Use consistent convention for variables
- Add a description about variable name convention
- Rename variables to follow convention
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
윤현식/동작제어Lab(SR)/Principal Engineer/삼성전자 [Thu, 17 May 2018 03:45:53 +0000 (12:45 +0900)]
Add a test for TFLITE to call 'MUL' CPU fallback inside NN API (#1238)
For #1232, a simple tflite file (mul2x2.tflite) was generated and
added into NNAPI_TEST to make sure if tensorflow lite
calls MUL operation (CPU fallback) under NN API layer.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
이상규/동작제어Lab(SR)/Principal Engineer/삼성전자 [Thu, 17 May 2018 02:16:02 +0000 (11:16 +0900)]
Add NN Tests for div, sub, strided_slice (#1246)
Related issue: #1229
Added tests are generated with neural_networks_test/specs/generate_test.sh.
While running these tests, I got SIGSEGV. So I disalbed them for now.
These tests will be enabled after SIGSEGV is fixed.
Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
김수진/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 17 May 2018 01:43:33 +0000 (10:43 +0900)]
Apply format checker to disabled codes in CpuExecutor (#1245)
We are supposed to enable NYI codes later, but the current disabled codes follow android coding rules still.
So if we just remove `#if 0` to enable the codes, it would be hard to compare the difference with before codes for review.
This commit applies format checker to disabled codes in CpuExecutor.
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
최형규/동작제어Lab(SR)/Senior Engineer/삼성전자 [Thu, 17 May 2018 01:43:09 +0000 (10:43 +0900)]
Apply library function to scripts (#1244)
- Apply library function to scripts
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 17 May 2018 01:26:05 +0000 (10:26 +0900)]
[tflite_run] Update README.md (#1243)
Update README to reflect latest features
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
Enable L2_POOL_2D operation (#1222)
Enabling l2pool2d operation along with the tflite test file addition and enabling the test in config file
Signed-off-by: praveen D N <praveen.dn@samsung.com>
이상규/동작제어Lab(SR)/Principal Engineer/삼성전자 [Wed, 16 May 2018 11:52:08 +0000 (20:52 +0900)]
Add DIV, SUB and STRIDED_SLICE enumerations in NeuralNetworks.h (#1239)
Enumeration DIV, SUB and STRIDED_SLICE and the corresponding comments are
copied from https://android.googlesource.com/platform/frameworks/ml/+/efd22b6.
It is the same source that we brought tests under neural_network_tests.
Related issue: #1230
Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
윤현식/동작제어Lab(SR)/Principal Engineer/삼성전자 [Wed, 16 May 2018 11:40:34 +0000 (20:40 +0900)]
Added 'how to run tests' in odroid (#1231)
Added 'how to run tests' in odroid (test_driver.sh)
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
윤현식/동작제어Lab(SR)/Principal Engineer/삼성전자 [Wed, 16 May 2018 08:40:57 +0000 (17:40 +0900)]
ADD vector operation for cl and neon (#1166)
Added ADD operation. This commit handles vector addition.
Scalar and matrix operation will be added in next commits.
(It seems that format check works under more strick rule from this week.
for this reason, util.h was changed)
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
서상민/동작제어Lab(SR)/Senior Engineer/삼성전자 [Wed, 16 May 2018 08:13:04 +0000 (17:13 +0900)]
Use RS7-STAR's TensorFlow again (#1235)
* Revert "Remove scripts/command/change_gitmodule.sh (#1085)"
For issue #1233
This reverts commit
edc95922795cb73dd4fed6ab3783fb959e58a809.
Signed-off-by: Sangmin Seo <sangmin7.seo@samsung.com>
* Use again RS7-STAR's TensorFlow
For issue #1233
This patch changes tensorflow's git url to RS7-STAR's, since we will
work on modification of tensorflow lite there.
Signed-off-by: Sangmin Seo <sangmin7.seo@samsung.com>
최성진/동작제어Lab(SR)/Principal Engineer/삼성전자 [Wed, 16 May 2018 02:49:55 +0000 (11:49 +0900)]
Add CustomOptions for supporting ADAS models (#1227)
This commit introduces CustomOptions to support ADAS model in print_op tool file.
- add custom_options as well as builtin_options_type
Signed-off-by: SungJin Choi <lotieye.choi@samsung.com>
김수진/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 16 May 2018 02:44:49 +0000 (11:44 +0900)]
Apply .FORMATCHECKED to runtimes/nn (#1228)
* Apply .FORMATCHECKED to runtimes/nn
This commit applies .FORMATCHECKED to `runtimes/nn/common`, `runtimes/nn/runtime`.
- Copy .FORMATCHECKED files
- run format-checker.sh
- add licenses to unlicensed files
close : #1186
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
Removing the executign permission for framework/tests files (#1226)
This commit revokes the accidentally added execution permission for the config files present in tests/framework/tests
Signed-off-by: Siva Sai <siva.sai@samsung.com>
최형규/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 15 May 2018 07:49:35 +0000 (16:49 +0900)]
Remove execute permission from source code (#1218)
- Remove execute permission from source code
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
김정현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 15 May 2018 07:49:16 +0000 (16:49 +0900)]
Apply .FORMATCHECKED to libs (#1215)
This commit applies format checker to directory `libs`.
Signed-off-by: Junghyun Kim <jh0822.kim@samsung.com>
서상민/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 15 May 2018 07:47:57 +0000 (16:47 +0900)]
Remove download_dependencies.tar.gz (#1148)
The old version of this patch has already been applied to the
`release/tizen-5.0-m1` branch by PR #1116.
This patch removes packaging/download_dependencies.tar.gz and uses
packages in externals/tensorflow.
Signed-off-by: Sangmin Seo <sangmin7.seo@samsung.com>
최형규/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 15 May 2018 07:10:14 +0000 (16:10 +0900)]
Enable RESIZE_BILENEAR for CPU fallback (#1216)
- Enable RESIZE_BILENEAR for CPU fallback
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Tue, 15 May 2018 07:08:39 +0000 (16:08 +0900)]
[tflite_run] Dump input and output tensor separately (#1214)
After `interpreter.Invoke()` input is no longer valid in tflite interpreter
so this commit revises to dump inputs before it and dump outputs after it.
Also the dumper has changed to have a copy of each tensor.
Resolve #1138
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이춘석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 15 May 2018 07:08:30 +0000 (16:08 +0900)]
Quote pattern on lcov (#1213)
To fix daily-arm-coverage, quote patterns during lcov
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
오형석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 15 May 2018 06:51:59 +0000 (15:51 +0900)]
Add more framework tests (#1211)
Add framework tests
- Casting: 4D tensor integer type to float type
- Resize bilinear: 4D input tensor, 4D output tensor (x2 width, x2 height)
- Custom: TensorFlowMax, 2D input tensor with scalar reduction indice, 1D output
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
최형규/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 15 May 2018 06:49:29 +0000 (15:49 +0900)]
Locate script library correctly (#1210)
- Locate script library correctly
This bug was intrdocued from PR 1183
- Locate script library correctly
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
이상규/동작제어Lab(SR)/Principal Engineer/삼성전자 [Tue, 15 May 2018 04:54:28 +0000 (13:54 +0900)]
Enable floor and tanh from NN generated tests (#1203)
Enable floor and tanh by avoiding name conflict with underscore.
In addition, generator script path in log and generated file is
corrected since the path is changed.
Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
오형석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 15 May 2018 04:34:08 +0000 (13:34 +0900)]
Framework test: arithmetic operations (#1207)
Add framework tests for arithmetic operations: ADD, SUB, MUL, DIV
Set disabled as default
ADD 1D: operand 1 and 2 are 1D tensor
ADD 4D: operand 1 and 2 are 4D tensor
SUB Broadcast: operand 1 is 4D tensor and operand 2 is scalar
MUL Broadcast: operand 1 is 4D tensor and operand 2 is 1D tensor
DIV Broadcast: operand 1 is 4D tensor and operand 2 is scalar
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
Saurav Babu/Advanced SW Team/./삼성전자 [Tue, 15 May 2018 03:49:12 +0000 (09:19 +0530)]
TFLiteSharp: Added spec file to build using gbs (#1162)
This patch adds skeleton code for native c++ binding between c# api and
tflite. This patch also adds packaging to build native cpp binding and
c# api using gbs. It will first build the TFLiteNative package
containing native c++ bindings and then it will build TFLiteSharp(c# api package)
Signed-off-by: Saurav Babu <saurav.babu@samsung.com>
최형규/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 15 May 2018 03:48:06 +0000 (12:48 +0900)]
Introduce a helper for scripts (#1183)
- Introduce docker_helper for scripts
- Apply docker_helper to docker_run_test.sh as a first use case
- Will apply docker_helper to other scripts later
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
이성재/동작제어Lab(SR)/Principal Engineer/삼성전자 [Tue, 15 May 2018 01:07:41 +0000 (10:07 +0900)]
Introduce doxygen for source code documentation. (#1204)
- add `/docs/doxygen` folder and `Doxyfile` for this project
- update `.gitignore` and `.ctags` file
- update Docker file to install `doxygen` and `graphviz` component
```
$ cd $NNFW_HOME/docs/doxygen
$ doxygen
$ <OPEN `html/index.html` with WEB BROWSER>
```
Signed-off-by: Sung-Jae Lee <sj925.lee@samsung.com>
김정현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 15 May 2018 00:20:03 +0000 (09:20 +0900)]
Replace CLUniqueTensor and NEUniqueTensor with UniqueTensor (#1205)
This commit introduces below declarations:
```
using NEUniqueTensor = UniqueTensor<::arm_compute::Tensor>;
using CLUniqueTensor = UniqueTensor<::arm_compute::CLTensor>;
```
instead of implementing NEUniqueTensor and CLUniqueTensor separately.
Signed-off-by: Junghyun Kim <jh0822.kim@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Tue, 15 May 2018 00:18:02 +0000 (09:18 +0900)]
[tflite_run] Load IFM from dumped file (#1202)
* [tflite_run] Load IFM from dumped file
So far we only used default input(all zeros) but now we can use inputs from t
he dumped file. This is reasonable since input and output must be a pair.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
* [tflite_run] Add conflicting options check
Add conflicting options check - `input` and `compare`
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이춘석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Mon, 14 May 2018 23:46:48 +0000 (08:46 +0900)]
Fix typo on DOCKER_ENV_VARS (#1200)
Fix typo on DOCKER_ENV_VARS
1. DOCKER_EVN_VARS -> DOCKER_ENV_VARS
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
서상민/동작제어Lab(SR)/Senior Engineer/삼성전자 [Mon, 14 May 2018 08:07:11 +0000 (17:07 +0900)]
Update the link to NeuralNetworks.h in the HLD doc (#1196)
After updating the source tree structure, the link to NeuralNetworks.h
in the HLD doc was broken. This patch fixes the broken link.
Signed-off-by: Sangmin Seo <sangmin7.seo@samsung.com>
서상민/동작제어Lab(SR)/Senior Engineer/삼성전자 [Mon, 14 May 2018 07:25:14 +0000 (16:25 +0900)]
Enable MobileNet in the integration test (#1195)
For #871
This patch enables MobileNet in the integration test.
Signed-off-by: Sangmin Seo <sangmin7.seo@samsung.com>
Enabling TANH activation in CpuExecutor (#1191)
This commit enables tanh activation operation.
Signed-off-by: Siva Sai Vaddipati <siva.sai@samsung.com>
서상민/동작제어Lab(SR)/Senior Engineer/삼성전자 [Mon, 14 May 2018 04:43:21 +0000 (13:43 +0900)]
Enable MobileNet in the benchmark test (#1194)
For EPIC 2 (#238)
This patch adds MobileNet to the list of benchmark tests.
Signed-off-by: Sangmin Seo <sangmin7.seo@samsung.com>
김정현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Mon, 14 May 2018 03:52:13 +0000 (12:52 +0900)]
Introduce git-push hook (#1184)
This commit introduces git-push hook and its installer.
Originally created by @wateret.
Signed-off-by: Junghyun Kim <jh0822.kim@samsung.com>
오형석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Mon, 14 May 2018 03:16:22 +0000 (12:16 +0900)]
Add gitignore: test report (#1192)
Update gitignore to ignore test report generated by tests/framework/run_test.sh
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
최형규/동작제어Lab(SR)/Senior Engineer/삼성전자 [Mon, 14 May 2018 00:03:53 +0000 (09:03 +0900)]
Update variable name in scripts to keep consistency (#1181)
- Update variable name in scripts to keep consistency
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
김수진/동작제어Lab(SR)/Engineer/삼성전자 [Mon, 14 May 2018 00:03:01 +0000 (09:03 +0900)]
Enable tests using ASharedMemory_create in TestValidation.cpp (#1164)
* Enable tests using ASharedMemory_create in TestValidation.cpp for NN Runtime tests
We've commented out the tests using `ASharedMemory_create` in `TestValidation.cpp` for NN Runtime tests. But I figured out it's ok to use `ashmem_create_region` instead of `ASharedMemory_create` likes [Memory.cpp](https://github.sec.samsung.net/STAR/nnfw/blob/master/runtimes/nn/runtime/Memory.cpp#L34)
So it seems it will be able to enable the tests using `ASharedMemory_create` in `TestValidation.cpp`.
- Add target includes in nn runtime
- Check bad states
- Enable tests
Close : #1163
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
김정현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Fri, 11 May 2018 07:00:17 +0000 (16:00 +0900)]
Introduce UniqueTensor (#1158)
Since NEUniqueTensor and CLUniqueTensor is very similar,
this commit introduce UniqueTensor to merge them into one.
After this commit, the current implementation of NEUniqueTensor
and CLUniqueTensor will be removed and replaced like below:
```
using NEUniqueTensor = UniqueTensor<::arm_compute::Tensor>;
using CLUniqueTensor = UniqueTensor<::arm_compute::CLTensor>;
```
Signed-off-by: Junghyun Kim <jh0822.kim@samsung.com>
이춘석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Fri, 11 May 2018 06:20:19 +0000 (15:20 +0900)]
Provide DOCKER_HOME to gen_coverage_report.sh (#1174)
1. remove `source docker_nnfw` from docker_coverage_report.sh
2. pass DOCKER_HOME to gen_coverage_report.sh
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
이상규/동작제어Lab(SR)/Principal Engineer/삼성전자 [Fri, 11 May 2018 06:14:10 +0000 (15:14 +0900)]
Bring android nn test generator scripts and specs (#1135)
Related issue: #937
run for generating tests
$ runtimes/tests/neural_networks_test/specs/generate_test.sh
Modification to original android NN test:
- Android NN test generator script
: adjusted to our directory structure
- Specs, the inputs to test generator
: V1_1 directory that contains NNAPI 1.1 is removed since it's not used now.
- Directory name
: tools/test_generator -> externals/nnapi_test_generator
since there are many tests in our project
Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 11 May 2018 06:12:40 +0000 (15:12 +0900)]
[tflite_run] Relocate image loading step (#1180)
Relocate image loading step to be more natural code flow.
- AssignTensor should be after generating BinImagge
- BinImage is not used after reading
- BinImage is unused in the lambda function (from `measure`)
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 11 May 2018 06:12:28 +0000 (15:12 +0900)]
[tflite_run] Dump input tensors (#1175)
Dump input tensors with output tensors since the output tensors are depend on
input tensors. The loader's behavior is still valid without changes.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 11 May 2018 05:45:04 +0000 (14:45 +0900)]
[tflite_run] Change variable names (#1179)
Now that the tensor file contains both inputs and outputs output_loader/dumper
is not a proper name for that.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
서상민/동작제어Lab(SR)/Senior Engineer/삼성전자 [Fri, 11 May 2018 05:29:07 +0000 (14:29 +0900)]
Use NeuralNetworks.h instead of nnapi.h (#1176)
For issue #980
This patch replaces use of nnapi.h with NeuralNetworks.h.
Signed-off-by: Sangmin Seo <sangmin7.seo@samsung.com>
최형규/동작제어Lab(SR)/Senior Engineer/삼성전자 [Fri, 11 May 2018 00:42:13 +0000 (09:42 +0900)]
Introduce nnfw_docker_tizen and use it (#1161)
- Introduce nnfw_docker_tizen by extracting code related to nnfw_docker_tizen
- Use it in docker_gbs_build.sh
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
이춘석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Fri, 11 May 2018 00:40:52 +0000 (09:40 +0900)]
Add env variable option when generating coverage report (#1160)
1. add env variable to control coverage gen
- SRC_PREFIX : prefix used on generating lcov info file
- GENHTML_FLAG : flag to pass to genhtml command(e.g. --no-source)
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
이춘석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Fri, 11 May 2018 00:37:41 +0000 (09:37 +0900)]
Add gcov data pulling (#1155)
1. tizen_xu4_test.sh supports gcov data pulling from xu4 tizen.
2. clean up test dir before testing
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
최형규/동작제어Lab(SR)/Senior Engineer/삼성전자 [Thu, 10 May 2018 08:27:49 +0000 (17:27 +0900)]
Use common file for scrtips related to nnfw_docker (#1153)
- Use common file nnfw_docker for scrtips related to nnfw_docker
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
최형규/동작제어Lab(SR)/Senior Engineer/삼성전자 [Thu, 10 May 2018 08:04:44 +0000 (17:04 +0900)]
Use common file for scrtips in CI related to nnfw_docker (#1154)
- Use common file for scrtips in CI related to nnfw_docker
- These scripts are not used for PR-based CI job.
Therefore we have to check it running well in CI manually.
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 10 May 2018 07:53:17 +0000 (16:53 +0900)]
[tflite_run] Dumped file contains tensor indices (#1144)
Store tensor indice numbers in the binary file
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
김용섭/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 10 May 2018 07:42:23 +0000 (16:42 +0900)]
Separate test_driver.sh to small scripts (#1128)
* Separate test_driver.sh to small scripts
test_driver.sh is script file but too big. Separate big script to small
scripts and prepare for converting these to ones in python.
Signed-off-by: Yongseop Kim <yons.kim@samsung.com>
* Add license text and remove `uniq` in print_to_json
Signed-off-by: Yongseop Kim <yons.kim@samsung.com>
최형규/동작제어Lab(SR)/Senior Engineer/삼성전자 [Thu, 10 May 2018 07:10:09 +0000 (16:10 +0900)]
Extract common code of docker image (#1150)
- Introduce nnfw_docker by extracting common code related to docker image nnfw_docker
- Use nnfw_docker in docker_run_test.sh to see it's working
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 10 May 2018 05:47:16 +0000 (14:47 +0900)]
[tflite_run] Remove OutputTensorDumper/Loader (#1145)
Remove old version and use TensorDumper/Loader instead.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이성재/동작제어Lab(SR)/Principal Engineer/삼성전자 [Thu, 10 May 2018 01:24:52 +0000 (10:24 +0900)]
Change unsuitable varable name (#1143)
* Change variable names start from `NPU_COMPILER_` to `NNFW_`
Signed-off-by: Sung-Jae Lee <sj925.lee@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 9 May 2018 23:58:02 +0000 (08:58 +0900)]
[tflite_run] Introduce TensorDumper/Loader (#1139)
Just copy of OutputTensorDumper/Loader and changed only class name.
These classes will dump/load input tensors as well as output tensors which
will eventually replace OutputTensorDumper/Loader.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 9 May 2018 05:58:16 +0000 (14:58 +0900)]
[tflite_run] Reorder print message (#1136)
Elapsed time message comes earlier than dump message.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이성재/동작제어Lab(SR)/Principal Engineer/삼성전자 [Wed, 9 May 2018 02:25:24 +0000 (11:25 +0900)]
Add object allocation info to log message (#1134)
* Add object allocation info to log message
- display `--> (type: address)` information at the end of the log line for each allocation call.
Signed-off-by: Sung-Jae Lee <sj925.lee@samsung.com>
최형규/동작제어Lab(SR)/Senior Engineer/삼성전자 [Wed, 9 May 2018 01:25:12 +0000 (10:25 +0900)]
Use docker image name from common.sh (#1126)
- Use docker image name from common.sh
and use same convention with other scripts.
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
최형규/동작제어Lab(SR)/Senior Engineer/삼성전자 [Wed, 9 May 2018 01:24:29 +0000 (10:24 +0900)]
Update scripts for cross build (#1122)
* Update scripts for cross build
- Update docker_build_cross_arm_ubuntu.sh with changes in
docker_build_cross_arm_ubuntu_without_aclbuild.sh
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
최형규/동작제어Lab(SR)/Senior Engineer/삼성전자 [Wed, 9 May 2018 01:06:55 +0000 (10:06 +0900)]
Update howto document of docker (#1121)
- Update howto document of docker to reflect recent changes
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
이성재/동작제어Lab(SR)/Principal Engineer/삼성전자 [Tue, 8 May 2018 23:32:01 +0000 (08:32 +0900)]
Update LD_LIBRARY_PATH in `howto.md` document (#1132)
* Update LD_LIBRARY_PATH in `howto.md` document
- reflect the path according to directory structure change.
Signed-off-by: Sung-Jae Lee <sjlee@mail.com>
이춘석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 8 May 2018 14:36:55 +0000 (23:36 +0900)]
Install sdb on nnfw_docker_tizen (#1131)
1. install sdb on nnfw_docker_tizen
2. install wget and unzip
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
이성재/동작제어Lab(SR)/Principal Engineer/삼성전자 [Tue, 8 May 2018 11:35:07 +0000 (20:35 +0900)]
Pass env variable LD_LIBRARY_PATH while invoking test driver (#1130)
- introduce `-ldlibrary=` option
- set `$NNFW_HOME/Produce/out/lib` as default value fo LD_LIBRARY_PATH
- pass env variable `LD_LIBRARY_PATH` while invoking test driver
Signed-off-by: Sung-Jae Lee <sjlee@mail.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Tue, 8 May 2018 10:33:23 +0000 (19:33 +0900)]
Support Android build for tflite_run (#1127)
Some include path were added manually since not all tools are supported on
Android yet.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
박세희/동작제어Lab(SR)/Principal Engineer/삼성전자 [Tue, 8 May 2018 06:48:58 +0000 (15:48 +0900)]
Add dependent libs for ACL libs (#1118)
This will add ACL dependent system libs to ACL for compiler versions don't add these for implicitly
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
이성재/동작제어Lab(SR)/Principal Engineer/삼성전자 [Tue, 8 May 2018 06:30:23 +0000 (15:30 +0900)]
Add `ctags` configuration file (#1120)
- exclude binary paths including `Product`, `rootfs`, `cache`
- include external source path
- set --extra=+f as default for easy file searching in project
Signed-off-by: Sung-Jae Lee <sjlee@mail.com>
윤현식/동작제어Lab(SR)/Principal Engineer/삼성전자 [Tue, 8 May 2018 02:43:08 +0000 (11:43 +0900)]
Added content for minicom novice. (#1119)
added 1. where to find minicom setup info and 2. how to find serial device name.
Signed-off-by: Hyun Sik Yoon (hyunsik.yoon@samsung.com)
Kshitiz Bansal/Advanced SW Team/./삼성전자 [Mon, 7 May 2018 07:03:36 +0000 (12:33 +0530)]
TF Lite C# API's Skeleto (#986)
Following are the part of this Pull Request
- TFLiteSharp package
- README.md
- C# api skeleton code
We have built the skeleton of the c# APIs taking Java APIs as the basis, as we discussed.
Signed-off-by: Kshitiz Bansal <kshitiz.b@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 4 May 2018 08:42:10 +0000 (17:42 +0900)]
Add README.md to tflite_run (#1110)
A brief description for tflite_run
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 4 May 2018 08:41:54 +0000 (17:41 +0900)]
Remove unused code from Diff.cpp (#1111)
This code was not removed to minimize diff from another commit.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
박세희/동작제어Lab(SR)/Principal Engineer/삼성전자 [Fri, 4 May 2018 06:28:52 +0000 (15:28 +0900)]
Fix max_pool_float_2 test (#1109)
This will fix max_pool_float_2 test that has N>1
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
서상민/동작제어Lab(SR)/Senior Engineer/삼성전자 [Fri, 4 May 2018 06:25:26 +0000 (15:25 +0900)]
Update the nnfw behavior figure (#1106)
The arrow from the runtime seemed to be directly connected to the
application, which is not correct. This patch modifies the arrow to be
connected only to TF Lite and makes a single output arrow from TF Lite.
Thanks to SaeHie Park for this comment.
Signed-off-by: Sangmin Seo <sangmin7.seo@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 4 May 2018 06:19:54 +0000 (15:19 +0900)]
[tflite_run] Support compare results (#1103)
With `--compare` or `-c` option, user can compare the results with output
tensor file which was dumped by `--dump` option.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
오형석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Fri, 4 May 2018 06:19:12 +0000 (15:19 +0900)]
Fix svace build script bug (#1107)
Fix svace build script bug: check file exist
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
서상민/동작제어Lab(SR)/Senior Engineer/삼성전자 [Fri, 4 May 2018 05:43:13 +0000 (14:43 +0900)]
Update the SW high level design document (#1105)
For issue #942
This commit addresses two comments posted in #921.
- Fix the return path from the runtime in the nnfw_behavior figure
- Clarify the description about what compute interface is used in the
runtime implementation
Signed-off-by: Sangmin Seo <sangmin7.seo@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 4 May 2018 02:14:23 +0000 (11:14 +0900)]
Extract compareSingleTensorView from TfLiteTensorComparator::run (#1072)
So we can compare single TensorView. This method will be used from `tflite_run`.
OutputTensorDumper, which will be introduced later, restores TensorView from a
binary tensor file. So we need an interface for comparing two `TensorView`s.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이춘석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Fri, 4 May 2018 01:57:35 +0000 (10:57 +0900)]
remove internal url from gbs.conf (#1090)
1. remove repo.localacl from gbs.conf
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
서상민/동작제어Lab(SR)/Senior Engineer/삼성전자 [Fri, 4 May 2018 01:56:49 +0000 (10:56 +0900)]
Remove scripts/command/change_gitmodule.sh (#1085)
For issue #1071
This patch removes `scripts/command/change_gitmodule.sh`, since it is no
longer used, and updates its caller sites.
Signed-off-by: Sangmin Seo <sangmin7.seo@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 4 May 2018 01:21:32 +0000 (10:21 +0900)]
[tflite_run] Introduce OutputTensorLoader (#1073)
As a pair with OutputTensorDumper, this class reads and builds
`TensorView<float>` from binary tensor file.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
김정현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Fri, 4 May 2018 01:21:21 +0000 (10:21 +0900)]
Apply insertFusedActivationLayer to NEON layers (#1083)
This commit applies insertFusedActivationLayer to NEON layers
such as Conv2D and Pooling.
Signed-off-by: Junghyun Kim <jh0822.kim@samsung.com>
김정현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Fri, 4 May 2018 01:21:05 +0000 (10:21 +0900)]
Apply insertFusedActivationLayer() to other CL layers. (#1082)
This commit applies insertFusedActivationLayer() to CL layers
(Conv2D and Pooling).
Signed-off-by: Junghyun Kim <jh0822.kim@samsung.com>
서상민/동작제어Lab(SR)/Senior Engineer/삼성전자 [Fri, 4 May 2018 01:02:17 +0000 (10:02 +0900)]
Change the git url for tensorflow submodule (#1070)
For issue #1069
This patch changes the git url and branch for the tensorflow submodule
as Tizen's.
Signed-off-by: Sangmin Seo <sangmin7.seo@samsung.com>
김정현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Fri, 4 May 2018 00:52:07 +0000 (09:52 +0900)]
[BUGFIX] Do not insert activation layer for ANEURALNETWORKS_FUSED_NONE (#1080)
If the activation layer is ANEURALNETWORKS_FUSED_NONE,
just return the function instead of break to avoid
inserting an unconfigured activation layer.
Signed-off-by: Junghyun Kim <jh0822.kim@samsung.com>
오형석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Fri, 4 May 2018 00:30:05 +0000 (09:30 +0900)]
Remove server information (#1075)
Remove server information from script
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
서상민/동작제어Lab(SR)/Senior Engineer/삼성전자 [Fri, 4 May 2018 00:17:44 +0000 (09:17 +0900)]
Use on-device AI instead of embedded AI (#1076)
For issue #939
This patch changes the term, embedded AI, used in the background section
of the requirement specification document as on-device AI for
consistency.
Signed-off-by: Sangmin Seo <sangmin7.seo@samsung.com>
김정현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Thu, 3 May 2018 23:56:11 +0000 (08:56 +0900)]
Apply CLUniqueTensor to Pooling and Softmax (#1051)
This commit applies CLUniqueTensor instead of directly using ACL's Tensor.
Signed-off-by: Junghyun Kim <jh0822.kim@samsung.com>