최형규/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 1 Jun 2018 08:11:19 +0000 (17:11 +0900)]
Introduce dummy STRIDED_SLICE kernel to pure_arm_compute (#1497)
- Introduce dummy STRIDED_SLICE kernel to pure_arm_compute
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 1 Jun 2018 08:10:48 +0000 (17:10 +0900)]
Implement ANeuralNetworksModel_addOperationEx in sample runtimes (#1515)
This commit implements ANeuralNetworksModel_addOperationEx in sample
runtimes (logging, template).
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 1 Jun 2018 07:52:59 +0000 (16:52 +0900)]
Add Diff constructor (#1512)
This commit adds a constructor to 'nnfw::util::tensor::Diff' which
initializes expected/obtained fields, and simplifies Comparator
implementation using this new constructor.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
이춘석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 1 Jun 2018 07:47:30 +0000 (16:47 +0900)]
make acl on pureacl arm cross build script (#1514)
BUILD_ACL does not support embeded kernel. To avoid cl_kernels installation on
test job, this commit disables BUILD_ACL and do `make acl` on arm cross build script.
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 1 Jun 2018 07:23:57 +0000 (16:23 +0900)]
Add custom tensorflowmax unittest (#1501)
Add unittest for custom tensorflowmax
2D (float, int32) input => 1D output
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
이상규/동작제어Lab(SR)/Principal Engineer/삼성전자 [Fri, 1 Jun 2018 06:32:52 +0000 (15:32 +0900)]
Fix build error std::function on pure acl runtime (#1505)
I include <functional> explicitly for std::function.
Related Issue: #1503
Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
이춘석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 1 Jun 2018 06:31:28 +0000 (15:31 +0900)]
Remove ARMCL_DIR setting from arm-ubunut pureacl build script (#1504)
This commit removes ARMCL_DIR from docker_build_cross_arm_pureacl.sh
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 1 Jun 2018 06:29:18 +0000 (15:29 +0900)]
Define Max namespace in pure acl runtime (#1506)
Define Max namespace, node, param in pure acl runtime
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 1 Jun 2018 06:14:34 +0000 (15:14 +0900)]
[nnapi_test] Support int32 input generation (#1498)
With this commit, nnapi_test is able to generate int32 input tensors.
Note that random generation is not supported, yet.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 1 Jun 2018 06:13:39 +0000 (15:13 +0900)]
Show used NN runtime filepath (#1496)
This commit shows the filepath of NN runtime being used.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 1 Jun 2018 06:05:46 +0000 (15:05 +0900)]
Set addOperationEx return value in pure acl runtime (#1507)
Set addOperationEx return value in pure acl runtime
To avoid conflict, commit small change.
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 1 Jun 2018 05:51:04 +0000 (14:51 +0900)]
[NNAPI unittest] Add 'add_2' case (#1464)
* [NNAPI unittest] Add 'add_2' case
This commit adds 'add_2' unittest which takes RHS as an input, but LHS
is a constant.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
* Fix format errors
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 1 Jun 2018 05:48:28 +0000 (14:48 +0900)]
Type-based tensor output compare (#1499)
This commit introduces type-based tensor output compare infrastructure
in TfLiteInterpMatchApp.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
최성진/동작제어Lab(SR)/Principal Engineer/삼성전자 [Fri, 1 Jun 2018 04:06:13 +0000 (13:06 +0900)]
[Gather OP] add nnapi delegate for Gather operation (#1465)
This commit adds nnapi delegate for Gather operation.
-axis parameter is added
Signed-off-by: SungJin Choi <lotieye.choi@samsung.com>
이춘석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 1 Jun 2018 03:27:11 +0000 (12:27 +0900)]
Add script to build pure_arm_compute runtime (#1484)
* Add script to build pure_arm_compute runtime
1. Add script to cross-build pure_arm_compute runtime for both arm-ubunut and tizen
To run test, please set LD_LIBRARY_PATH=Product/lib/pureacl instead of
Product/lib.
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 1 Jun 2018 03:17:53 +0000 (12:17 +0900)]
Flexible input generation switch for RandomTestRunner (#1495)
This commit revises RandomTestRunner to make it easy to extend supported
input type.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 1 Jun 2018 02:56:01 +0000 (11:56 +0900)]
More tests for STRIDED_SLICE_EX (#1488)
Add more test for STRIDE_SLICE_EX
It's similar with STRIDE_SLICE (Fix argument)
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 1 Jun 2018 02:35:13 +0000 (11:35 +0900)]
[support.tflite] Generic TensorView (#1494)
This commit revises TensorView class as a generic class.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 1 Jun 2018 02:02:57 +0000 (11:02 +0900)]
Detect NNAPI load failure (#1489)
This commit revises NeuralNetworksShim.h and NeuralNetworksExShim.h to
abort execution when NNAPI functions are unavailable, but used.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
최성진/동작제어Lab(SR)/Principal Engineer/삼성전자 [Fri, 1 Jun 2018 01:18:30 +0000 (10:18 +0900)]
[Topkv2 OP] add nnapi delegate for Topk_v2 operation (#1463)
* [Topkv2 OP] add nnapi delegate for Topk_v2 operation
This commit add nnapi delegate for Topk_v2 operation.
-change output order (output_value is first and output_index is second)
Signed-off-by: SungJin Choi <lotieye.choi@samsung.com>
* [Topkv2 OP] add nnapi delegate for Topk_v2 operation
This commit add nnapi delegate for Topk_v2 operation.
-change output order (output_value is first and output_index is second)
Signed-off-by: SungJin Choi <lotieye.choi@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 1 Jun 2018 01:05:30 +0000 (10:05 +0900)]
[tflite_benchmark] Show benchmark configurations (#1491)
This commit revises tflite_benchmark to show benchmark configurations
before running benchmark.
Currently, the following configurations will be shown:
- The number of threads to be used
- Whether NNAPI is used, or not
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
서상민/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 1 Jun 2018 00:38:50 +0000 (09:38 +0900)]
[PureCL] Support DIV operation (#1482)
* [PureCL] Support DIV operation
For #1338 and #1366
This patch adds DIV operation to the pure ACL runtime.
Signed-off-by: Sangmin Seo <sangmin7.seo@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 1 Jun 2018 00:18:18 +0000 (09:18 +0900)]
Update Android build scripts (#1487)
This commit revises 'build_android_rootfs.sh' to create ndk directory
for tensorflow lite buiild, and updates tensorflow lite CMakeLists.txt
to access this NDK during build.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 31 May 2018 11:49:14 +0000 (20:49 +0900)]
[tflite_run] Generate random input (#1477)
When running `tflite_run` without input argument it runs with values
with 0. This commit will fill the input tensors with random values
instead, same way with `nnapi_test` does.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이춘석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 31 May 2018 11:48:27 +0000 (20:48 +0900)]
Submodule update on internal_acl_build (#1483)
`make acl` does not work on fresh repo. This commit makes submodule update on
target internal_acl_build.
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
최성진/동작제어Lab(SR)/Principal Engineer/삼성전자 [Thu, 31 May 2018 11:46:10 +0000 (20:46 +0900)]
[Gather Op] add axis to runtime_run_android_nn_test (#1479)
This commit adds axis to runtime_run_android_nn_test.
-axis parameter is added newly.
Signed-off-by: SungJin Choi <lotieye.choi@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 31 May 2018 08:43:27 +0000 (17:43 +0900)]
[Pure CL] Introduce Const Tensor View (#1474)
This commit introduces ConstView class to make it easy to write generic
const operand updates.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
최형규/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 31 May 2018 07:56:29 +0000 (16:56 +0900)]
Update comment for STRIDED_SLICE in NeuralNetworks.h (#1460)
- Update comment for STRIDED_SLICE in NeuralNetworks.h,
because operands of STRIDED_SLICE has been implemented by
following parameters defined in the latest Android NN API.
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 31 May 2018 07:56:03 +0000 (16:56 +0900)]
[Pure CL] Fix typo (USE_CL_ARITHMETIC_ADDTION) (#1480)
This commit fixes type in environment variable name:
(USE_CL_ARITHMETIC_ADDTION -> USE_CL_ARITHMETIC_ADDITION)
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
이춘석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 31 May 2018 07:49:03 +0000 (16:49 +0900)]
set ROOTFS_DIR on docker_build_cross_arm_ubuntu.sh (#1467)
To run docker_build_cross_arm_ubuntu, ROOTFS_DIR should be passed.
It takes 6min to complete build.
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 31 May 2018 07:06:03 +0000 (16:06 +0900)]
[Android] Update rootfs build script (#1476)
- Install boost (header only)
- Install arm compute
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 31 May 2018 07:03:38 +0000 (16:03 +0900)]
[Android] Enable build for all runtimes (#1475)
Enable Android build for runtimes including pure_arm_compute
For link library, pthread should be parameterized.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 31 May 2018 06:54:11 +0000 (15:54 +0900)]
Generic Tensor Comparator (#1462)
This commit revises TfLiteTensorComparator as generic tensor comparator,
and move it to nnfw_util.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
서상민/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 31 May 2018 06:23:03 +0000 (15:23 +0900)]
[PureCL] Initialize CLTensors having const data (#1472)
For issue #1366
In the current implementation of Pure CL runtime, CLTensors, which have
const data in their corresponding NNAPI operand, are not initialized.
This patch fixes this issue by initializing them in PlanBuilder's
finalize method.
Signed-off-by: Sangmin Seo <sangmin7.seo@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 31 May 2018 06:22:47 +0000 (15:22 +0900)]
[tfliite_benchmark] Allow Thread configuration (#1473)
This commit revises tflite_benchmark to read thread configuration from
THREAD environment variable.
This commit also changes the default configuration as -1 to enable
multi-threading.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 31 May 2018 06:21:59 +0000 (15:21 +0900)]
[Pure CL] Allow iteration over NNAPI operands (#1471)
This commit introduces iterate method which allows users to iterate over
registered NNAPI operands.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
최성진/동작제어Lab(SR)/Principal Engineer/삼성전자 [Thu, 31 May 2018 05:59:32 +0000 (14:59 +0900)]
[Gather Op] add other operation types (#1329)
* [Gather Op] add uint8 operation types
This commit adds other operation types.
-uint8
Signed-off-by: SungJin Choi <lotieye.choi@samsung.com>
* [Gather Op] add int32 operation types
This commit adds other operation types.
-int32
Signed-off-by: SungJin Choi <lotieye.choi@samsung.com>
* [Gather Op] add an comment that explains gather operation
This commit adds comments.
- Gather of NN Runtime supports TENSOR_FLOAT32, TENSOR_QUANT8_ASYMM, TENSOR_INT32 other than TFLite.(TFLite additionaly supports kTfLiteString.)
Signed-off-by: SungJin Choi <lotieye.choi@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 31 May 2018 05:49:24 +0000 (14:49 +0900)]
[Pure CL] Skeleton for addOperationEx (#1468)
Prepare addOperationEx
It is to avoid merge conflict.
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
윤현식/동작제어Lab(SR)/Principal Engineer/삼성전자 [Thu, 31 May 2018 05:26:22 +0000 (14:26 +0900)]
Code for generic tensor shape in acl/src/shape.cpp (#1404)
Parent issue: #1402
This code follows the explanation in https://arm-software.github.io/ComputeLibrary/latest/architecture.xhtml
```
Tensors are defined by a DataType plus a number of channels (Always expected to be 1 for now)
their dimensions are expressed as [width, height, feature_maps, batch].
In other words, the lower three dimensions of a tensor specify a single input in
[width, height, feature_maps], while any other specified dimension represents a batch
in the appropriate dimension space. For example, a tensor with dimensions [128, 128, 64, 16]
represents a 1D batch space with 16 batches of 128 elements in width and height
and 64 feature maps each.
```
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 31 May 2018 04:50:54 +0000 (13:50 +0900)]
[Pure CL] Extract Various Source Implementations (#1452)
This commit extracts various source implementations from execution.cc.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 31 May 2018 04:20:47 +0000 (13:20 +0900)]
Extract RandomGenerator from RandomTestRunner (#1445)
In `nnfw_util` module, extract RandomGenerator from RandomTestRunner
so RandomGenerator can be used in `tflite_run` as well.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
서상민/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 31 May 2018 02:43:37 +0000 (11:43 +0900)]
[Pure CL Runtime] Handle shape with rank one on setInput/Output (#1449)
This patch revises ANeuralNetworksExecution_setInput/Output to handle
the case of rank one.
Signed-off-by: Sangmin Seo <sangmin7.seo@samsung.com>
서상민/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 31 May 2018 02:42:38 +0000 (11:42 +0900)]
Add a DIV broadcast unit test (#1459)
For issue #1366
This commit adds a test case where DIV broadcast operation is examined.
Signed-off-by: Sangmin Seo <sangmin7.seo@samsung.com>
윤현식/동작제어Lab(SR)/Principal Engineer/삼성전자 [Thu, 31 May 2018 02:42:02 +0000 (11:42 +0900)]
Modify #ifndef macro of SimpleArithmeticAdditionLayer.h (#1458)
from
`#ifndef __SIMPLE_ARITHMETIC_LAYER_H__`
to
`#ifndef __SIMPLE_ARITHMETIC_ADDITION_LAYER_H__`
because we may need more arithmetic layers.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 31 May 2018 02:41:36 +0000 (11:41 +0900)]
[NNAPI test] Introduce asserts on input types (#1457)
This commit introduces asserts on input types as float32 is the only
input type that nnapi_test supports currently.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
최형규/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 31 May 2018 02:33:04 +0000 (11:33 +0900)]
Do not delegate temporary tensors to NNAPI. (#1359)
- also added delegation for MUL, and set the default scale to be 0.0f.
PiperOrigin-RevId:
195288948
This is a cherrypick of https://github.sec.samsung.net/RS7-STAR/Tensorflow/pull/41
with modification to (1) resolve conflict and (2) apply new format.
Therefore this is a modification of cherrypick of origitnal Tensorflow from
https://github.com/tensorflow/tensorflow/commit/
a16ba4fc0d3faec077c689f3f361264978a2d3cb#diff-
f4a5031c0f2ec4e55c3a89bf4c21c7f8
Signed-off-by: Hyung-Kyu Choi <hk0110.choi@samsung.com>
이춘석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 31 May 2018 02:28:49 +0000 (11:28 +0900)]
CAST_EX nnapi_delegate (#1429)
1. Add add_cast_ex_params
2. Add nnapi call from nnapi_delagate.cpp
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
박세희/동작제어Lab(SR)/Principal Engineer/삼성전자 [Thu, 31 May 2018 01:24:49 +0000 (10:24 +0900)]
Build Android TFLite JNI shared library (#1442)
* Build Android TFLite JNI shared library
This will change cmake to drop Android JNI including Tensorflow Lite
- current drop will work same as vanilla TFLite library
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* apply comments
이상규/동작제어Lab(SR)/Principal Engineer/삼성전자 [Thu, 31 May 2018 01:24:25 +0000 (10:24 +0900)]
Fix linking error on arm compute library (#1456)
Recently introduced `ARMCompute` removed link to dl and pthread.
So it produces linking error like following:
```
arm-linux-gnueabihf/bin/ld:
warning: libdl.so.2, needed by
/home/brian/z/nnfw/Product/armv7l-linux.debug/out/lib/libarm_compute.so,
not found (try using -rpath or -rpath-link)
/home/brian/z/nnfw/Product/armv7l-linux.debug/out/lib/libarm_compute_core.so:
undefined reference to `dlopen@GLIBC_2.4'
/home/brian/z/nnfw/Product/armv7l-linux.debug/out/lib/libarm_compute_core.so:
undefined reference to `dlsym@GLIBC_2.4'
/home/brian/z/nnfw/Product/armv7l-linux.debug/out/lib/libarm_compute_core.so:
undefined reference to `dlerror@GLIBC_2.4'
collect2: error: ld returned 1 exit status
```
By specifying dl and pthread as dependency of arm_compute_xxx, the
linking error could be fixed.
Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
서상민/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 31 May 2018 01:21:49 +0000 (10:21 +0900)]
Update ACL submodule hash (#1454)
This patch updates the submodule hash for ACL to the latest one.
Signed-off-by: Sangmin Seo <sangmin7.seo@samsung.com>
이춘석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 31 May 2018 01:05:56 +0000 (10:05 +0900)]
Use submodule on tizen cross build (#1437)
* Use submodule on tizen cross build
1. Let OBS_BUILD have ON/OFF
2. Use submodule on tizen cross build
In summary, tizen cross build works in the same way of arm cross build.
* Change submodule during tizen cross build
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
김용섭/동작제어Lab(SR)/Engineer/삼성전자 [Thu, 31 May 2018 00:59:02 +0000 (09:59 +0900)]
Enable test_driver to catch failed test (#1409)
* Enable test_driver to catch failed test
Enable tools/test_driver/test_driver.sh to catch failed test, so that
now do not pass any failed test cases. To do this, run_benchmark*.sh and
run_unittest.sh will stop when the tests catch fail.
run_frameworktest.sh will stop after the tests run all models.
Signed-off-by: Yongseop Kim <yons.kim@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 31 May 2018 00:56:31 +0000 (09:56 +0900)]
[Pure CL] Support Tensor Sinks (#1440)
This commit revises pure CL runtime to support generic tensor outputs.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
서상민/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 31 May 2018 00:47:06 +0000 (09:47 +0900)]
Make iterate() static inline (#1451)
This patch makes iterate() static inline function in order to avoid
multiple definition issue that could be caused when IndexIterator.h is
included in multiple files.
Signed-off-by: Sangmin Seo <sangmin7.seo@samsung.com>
서상민/동작제어Lab(SR)/Staff Engineer/삼성전자 [Wed, 30 May 2018 23:57:33 +0000 (08:57 +0900)]
Add hasData() to class Object (#1450)
Object::hasData() will be used to check if Object has data internally.
Signed-off-by: Sangmin Seo <sangmin7.seo@samsung.com>
서상민/동작제어Lab(SR)/Staff Engineer/삼성전자 [Wed, 30 May 2018 23:57:16 +0000 (08:57 +0900)]
Add a DIV unit test (#1448)
This patch adds a DIV unit test where two tensors with the same same are
passed as inputs.
Signed-off-by: Sangmin Seo <sangmin7.seo@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Wed, 30 May 2018 23:41:31 +0000 (08:41 +0900)]
Introduce 'tf_test' tool (#1421)
This commit introduce experimental 'tf_test' tool. This tool loads
tensorflow lite model and corresponding tensorflow model, run
inferences, and compare the result.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Wed, 30 May 2018 23:41:01 +0000 (08:41 +0900)]
TensorFlowMax nnapi_delegate (#1432)
Add nnapi call from nnapi_delegate.cpp for TensorFlowMax
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Wed, 30 May 2018 22:24:51 +0000 (07:24 +0900)]
TFLite interpreter: Custom operator max (#1419)
General implementation for custom operator max
Similar with mean
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
오형석/동작제어Lab(SR)/Staff Engineer/삼성전자 [Wed, 30 May 2018 22:23:38 +0000 (07:23 +0900)]
Operator selector: cast operator bug fix (#1428)
Fix bug about generating cast operator's option
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 30 May 2018 22:23:15 +0000 (07:23 +0900)]
[newrt] Apply fix to newrt (#1444)
Apply fixed that were applied to pure_arm_compute
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
최성진/동작제어Lab(SR)/Principal Engineer/삼성전자 [Wed, 30 May 2018 11:06:45 +0000 (20:06 +0900)]
[Topk_v2 Op] support other operation types (#1314)
* [Topk_v2 Op] add uint8 types
This commit adds other types.
-uint8
Signed-off-by: SungJin Choi <lotieye.choi@samsung.com>
* [Topk_v2 Op] add int32 operation types
This commit adds other types.
-int32
Signed-off-by: SungJin Choi <lotieye.choi@samsung.com>
* [Topk_v2 Op] add an comment that explains topk_v2 operation
This commit adds comments.
- TopK_v2 of NN Runtime supports TENSOR_FLOAT32, TENSOR_QUANT8_ASYMM, TENSOR_INT32 other than TFLite.
Signed-off-by: SungJin Choi <lotieye.choi@samsung.com>
* [Top[Topk_v2 Op] add 2D test cases for other types
This commit adds 2D test cases for other types (int32, uint8)
Signed-off-by: SungJin Choi <lotieye.choi@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Wed, 30 May 2018 10:53:59 +0000 (19:53 +0900)]
[Pure CL] Inheric nnfw::utils::tensor::Shape (#1438)
This commit revises internal::tflite::operand::Shape to inherit
nnfw::utils::tensor::Shape (to reduce code duplication).
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Wed, 30 May 2018 08:49:00 +0000 (17:49 +0900)]
[Pure CL runtime] Support reshape/softmax over tensors (#1436)
This commit removes asserts in reshape/softmax compilation
implementation to support ADAS model.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Wed, 30 May 2018 08:04:43 +0000 (17:04 +0900)]
Introduce 'tensor::Diff' class (#1422)
This commit extracts general 'tensor::Diff' clss from
'TfLiteTensorDiff'.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
오형석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Wed, 30 May 2018 07:53:27 +0000 (16:53 +0900)]
Fix format for pure acl runtime (#1435)
Fix format for pure acl runtime
Enable format check for pure acl runtime
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Wed, 30 May 2018 07:47:00 +0000 (16:47 +0900)]
Ignore compiler warnings from ARMCompute (#1434)
This commit adds -Wno-ignored-attributes option when building pure CL
runtime to ignore compiler warnings due to the code in ARMCompute
library (CL.h).
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
이춘석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Wed, 30 May 2018 07:31:16 +0000 (16:31 +0900)]
Fix build break on debug build (#1427)
Fix debug build break for pure acl runtime
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
오형석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Wed, 30 May 2018 06:21:44 +0000 (15:21 +0900)]
Update TensorFlow submodule for topk_v2 (#1420)
This patch updates the submodule hash for tensorflow to handle topk_v2
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Wed, 30 May 2018 05:36:53 +0000 (14:36 +0900)]
[Pure CL runtime] Fix CMakeLists.txt indentation (#1424)
This commit fixes CMakeLists.txt indentation.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Wed, 30 May 2018 02:25:51 +0000 (11:25 +0900)]
Add 'TFLITE_ENSURE' macro (#1418)
This commit introduces 'TFLITE_ENSURE' macro, and use it for tflite_run
and tflite_benchmark.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Wed, 30 May 2018 02:23:54 +0000 (11:23 +0900)]
[Pure CL Runtime] Use SimpleArithmeticAdditionLayer (#1414)
This commit introduces SimpleArithmeticAddationLayer, and uses it by
default kernel on pure CL runtime.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Wed, 30 May 2018 02:20:56 +0000 (11:20 +0900)]
Introduce 'ARMCompute' package (#1400)
This commit introduces 'ARMCompute' package which provides uniform
interface over imported/built ARMCompute libraries.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
김수진/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 30 May 2018 02:12:32 +0000 (11:12 +0900)]
Update xu4 documentations to more detail about minicom (#1334)
* Update xu4 documentations for using keyboard control
This commit updates xu4 documentations to more detail for using keyboard control
Signed-off-by: sjsujinkim <sjsujin.kim@samsung.com>
* Change comments
* Change comments to more cleary
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 30 May 2018 01:32:10 +0000 (10:32 +0900)]
Enable Android build of tflite_benchmark, disable tflite_run (#1413)
Enable/Disable build for Android
- Enable build of tflite_benchmark
- Disable tflite_run since it has boost library build issue
(Related to standard library - gnustl/libc++)
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이춘석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Wed, 30 May 2018 01:30:28 +0000 (10:30 +0900)]
Support Strided Slice Ex on nnapi_delegate.cc (#1399)
* Support Strided Slice Ex on nnapi_delegate.cc
Add Strided Slice Ex conversion to nnapi_delegate.cc
Note that stride_slide on nn runtime does not support 2 parameters : ellipsis_mask
and new_axis_mask.
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
* add CHECK_NN, remove break
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 29 May 2018 22:33:08 +0000 (07:33 +0900)]
Remove CXX_STANDARD workaround (#1405)
* Remove CXX_STANDARD workaround
CXX_STANDARD is no longer necessary as the minimum required version of
CMake is 3.1 (which supports CXX_STANDARD feature).
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
* Apply the same change to 'libs/kernel/acl/CMakeLists.txt'
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 29 May 2018 11:09:00 +0000 (20:09 +0900)]
[Pure CL runtime] Add symbolcheck test (#1397)
This commit introduces 'pure_arm_compute_symbolcheck' executable which
allows us to detect undefined references in pure CL runtimes during
build.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
서상민/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 29 May 2018 05:51:07 +0000 (14:51 +0900)]
Revert "Update validation in runtime (#1351)" (#1406)
This reverts commit
86bbecea1fce90cac87a40782900eb24ff09db95.
PR #1351 caused many test failures, which were not noticed by our CI.
Signed-off-by: Sangmin Seo <sangmin7.seo@samsung.com>
장지섭/동작제어Lab(SR)/Engineer/삼성전자 [Tue, 29 May 2018 02:01:53 +0000 (11:01 +0900)]
Update validation in runtime (#1351)
This commit updates validation in runtime.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
이춘석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 29 May 2018 01:25:00 +0000 (10:25 +0900)]
Install arm_compute_XXX and libnueralnetworks.so on pure_arm_compute … (#1393)
* Install arm_compute_XXX and libnueralnetworks.so on pure_arm_compute build
1. On pure_arm_compute runtime build, install arm_compute_XXX.so into lib
2. DO NOT build NN Runtime if BUILD_ACL=ON
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
* Add option BUILD_PURE_ARM_COMPUTE and change install dir
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
* Remove `BUILD_NN_RUNTIME` option off
Since pure_acl_compute runtime is installed to lib/pureacl, we do not need turn
off option BUILD_NN_RUNTIME.
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
* Apply suggestion
In summary,
1. install arm_compute_xxx to lib
2. install libnn.so to lib/pureacl
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
서상민/동작제어Lab(SR)/Senior Engineer/삼성전자 [Tue, 29 May 2018 01:22:40 +0000 (10:22 +0900)]
Convert TF Lite's DIV to NN API's (#1398)
For #1199 and #1366
This commit links TF Lite's DIV operation to NN API's. It also enables
the Div test in the framework test.
Signed-off-by: Sangmin Seo <sangmin7.seo@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Tue, 29 May 2018 00:47:33 +0000 (09:47 +0900)]
[pure_arm_compute] Fix no return value (Context &) (#1396)
`Context::set()` is non-void but did not return a value.
Make it to return `*this`.
Similar to #1382
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
이상규/동작제어Lab(SR)/Principal Engineer/삼성전자 [Tue, 29 May 2018 00:47:11 +0000 (09:47 +0900)]
Enable SUB NNAPI delegation and SUB broadcast test (#1394)
Tensorflow is updated to v.1.8 so we can run SUB broadcast.
I pave the way to NN runtime for Sub, (It's same to ADD.)
and enabled the corresponding test.
Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Mon, 28 May 2018 11:13:50 +0000 (20:13 +0900)]
Introduce 'Tensorflow' package (#1387)
This commit introduces 'Tensorflow' package which allows users to use
externally built libtensorflow-core.a as if CMake targets.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
윤현식/동작제어Lab(SR)/Principal Engineer/삼성전자 [Mon, 28 May 2018 10:43:11 +0000 (19:43 +0900)]
Make Tensorflow Lite calls MUL CPU fallback (#1360)
1. Modified `NNAPIDelegate` to build graph with `MUL`
2. Test with ADAS 'MUL' graph by nnapi_test
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
서상민/동작제어Lab(SR)/Senior Engineer/삼성전자 [Mon, 28 May 2018 08:32:50 +0000 (17:32 +0900)]
Update TensorFlow submodule to v1.8 (#1388)
For #1383
This patch updates the submodule hash for TensorFlow to v1.8 branch in
RS7-STAR/Tensorflow.
Signed-off-by: Sangmin Seo <sangmin7.seo@samsung.com>
서상민/동작제어Lab(SR)/Senior Engineer/삼성전자 [Mon, 28 May 2018 07:27:21 +0000 (16:27 +0900)]
Include model.h in nnapi_delegate.cpp (#1386)
For #1383
Although TF Lite's `nnapi_delegate.cc` includes
`tensorflow/contrib/lite/model.h`, our nnapi_delegate.cpp did not
include `model.h` because we were able to compile `nnapi_delegate.cpp`
without the header file. However, it is not the case with the latest
TensorFlow v1.8, which we will be using soon. This patch is to prevent
the build failure that could be caused by not including `model.h` when
building with TensorFlow v1.8.
Signed-off-by: Sangmin Seo <sangmin7.seo@samsung.com>
서상민/동작제어Lab(SR)/Senior Engineer/삼성전자 [Mon, 28 May 2018 07:19:56 +0000 (16:19 +0900)]
Introduce nnfw's BuiltinOpResolver (#1384)
For issue #1383
This patch introduces our own BuiltinOpResolver, which is basically
copied from externals/tensorflow, in order to manage which builtin
operators in TF Lite are used with nnfw.
To only use nnfw's BuiltinOpResolver, this patch also excludes
register.cc in externals/tensorflow from the TF Lite build.
Signed-off-by: Sangmin Seo <sangmin7.seo@samsung.com>
이한종/동작제어Lab(SR)/Engineer/삼성전자 [Mon, 28 May 2018 07:09:16 +0000 (16:09 +0900)]
[pure_arm_compute] Fix no return value (#1382)
`Sequence::emplace_back()` is non-void but did not return a variable.
Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Mon, 28 May 2018 06:31:22 +0000 (15:31 +0900)]
[Pure CL runtime] Fix undefined references (#1380)
This commit fixes undefined reference issues on pure CL runtime.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
서상민/동작제어Lab(SR)/Senior Engineer/삼성전자 [Mon, 28 May 2018 05:35:42 +0000 (14:35 +0900)]
Include cstring for memcpy (#1378)
This patch adds `#include <cstring>` to tensor_dumper.cc, since `memcpy`
is used in the file.
Signed-off-by: Sangmin Seo <sangmin7.seo@samsung.com>
이춘석/동작제어Lab(SR)/Senior Engineer/삼성전자 [Mon, 28 May 2018 05:35:23 +0000 (14:35 +0900)]
Add Extended NNAPI shim header (#1365)
* Add Extended NNAPI shim header
Add Extended NNAPI shim header. This header includes
ANEURALNETWORKS_STRIDED_SLICE_EX = 50005
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
서상민/동작제어Lab(SR)/Senior Engineer/삼성전자 [Mon, 28 May 2018 04:33:05 +0000 (13:33 +0900)]
Include cassert in Diff.cpp (#1377)
This patch adds `#include <cassert>` to Diff.cpp, since `assert()` is
used in the file.
Signed-off-by: Sangmin Seo <sangmin7.seo@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Mon, 28 May 2018 04:32:49 +0000 (13:32 +0900)]
Introduce 'nnfw_find_package' (#1372)
This commit introduces 'nnfw_find_package' macro which allows users to
write and use nnfw-internal CMake modules (similarly as modules provided
via find_package).
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Mon, 28 May 2018 04:11:38 +0000 (13:11 +0900)]
Use nnfw-internal NNAPIDelegate (#1370)
This commit revises NNAPISession to use nnfw-internal NNAPIDelegate.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Mon, 28 May 2018 04:10:33 +0000 (13:10 +0900)]
Add tensorflow-based 'detection' example (#1368)
This commit adds 'detection' sample app which loads T/F model (which
takes 1x320x320x3 input whose name is 'input_node') and measures elapsed
time for 5 runs.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Mon, 28 May 2018 04:07:48 +0000 (13:07 +0900)]
Use common benchmark framework (#1371)
Thsi commit revises tflite_run and tflite_benchmark to use common
benchmark framework.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Mon, 28 May 2018 04:06:45 +0000 (13:06 +0900)]
Do NOT install libtensorflow-lite.a (#1375)
libtensrflow-lite.a is used only during nnfw build.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Senior Engineer/삼성전자 [Mon, 28 May 2018 02:39:55 +0000 (11:39 +0900)]
[Pure CL runtime] Support general reshape (#1353)
* [Pure CL runtime] Support general reshape
This commit removes height/width restriction on reshape operation in
pure CL runtime.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>