Yongjoo Ahn [Mon, 15 Jan 2024 07:51:39 +0000 (16:51 +0900)]
[test] Fix snpe filter test to support two API versions
- Add dlc model files for SNPE2.
- Rename SNPE1 model files.
- Fix unittest to set proper model files.
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Yongjoo Ahn [Mon, 15 Jan 2024 07:49:21 +0000 (16:49 +0900)]
[filter] Add SNPE filter supports API ver 2
- Add a SNPE filter source `tensor_filter_snpe.cc` to support SNPE API version 2
- It utilizes new C APIs.
- Let meson decide cpp source file to compile by checking snpe_api_version
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
MyungJoo Ham [Mon, 22 Jan 2024 08:17:42 +0000 (17:17 +0900)]
actions: add cppcheck, doxygen-build, executable, ...
The following TAOS-CI pr-prebuild checks are imported
- cppcheck
- doxygen-build
- executable
- hardcoded-path
- misspelling
- nobody
- timestamp
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
MyungJoo Ham [Fri, 19 Jan 2024 07:12:25 +0000 (16:12 +0900)]
actions: static/rpm-spec check added
RPM-SPEC runs rpmlint if .spec file is updated.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
MyungJoo Ham [Fri, 19 Jan 2024 06:39:10 +0000 (15:39 +0900)]
actions: static/doxygen-tag style clean
Make the code style consistant with other recent checks.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
MyungJoo Ham [Fri, 19 Jan 2024 06:32:37 +0000 (15:32 +0900)]
actions: more static checks from TAOC-CI
- indent
- newline
- pylint
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Jaeyun Jung [Thu, 25 Jan 2024 12:12:27 +0000 (21:12 +0900)]
[CodeClean] fix svace issue
1. change data type to handle gst-memory.
2. use util function to get nth tensor info.
3. add null-init in filter-edgetpu.
4. fix use-after-free case when parsing accel in tensor-filter.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
MyungJoo Ham [Wed, 17 Jan 2024 06:40:06 +0000 (15:40 +0900)]
github-action: rebuild only if it's required
This adds a workflow routine that checks if rebuild is needed.
Then, if there is no changes in source code, skip rebuilding
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Wook Song [Tue, 23 Jan 2024 05:10:24 +0000 (14:10 +0900)]
GitHub/Workflows: Add an action to check C++ source file format
This patch adds a GitHub Action workflow to check C++ source file
format.
Signed-off-by: Wook Song <wook16.song@samsung.com>
[Imported from deviceMLOps.MLAgent]
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
MyungJoo Ham [Wed, 24 Jan 2024 03:55:15 +0000 (12:55 +0900)]
github-action: do not run static-checks on deleted files.
The git-changed files may include deleted files.
Do not run static-cchecks on deletec files.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
MyungJoo Ham [Tue, 23 Jan 2024 02:14:31 +0000 (11:14 +0900)]
test: nondeterministic unit test error.
Due to multithread nondeterminism, we get
occasional unit test error from the "new_data_cb"
callback.
Create a lock and make data local to the function.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Jaeyun Jung [Mon, 22 Jan 2024 05:07:44 +0000 (14:07 +0900)]
[Filter] remove space of model file
To prevent not-found error, remove space of model path.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
MyungJoo Ham [Fri, 12 Jan 2024 08:45:27 +0000 (17:45 +0900)]
actions: add static checks
Add static checkers & verifiers to github-actions.
Initially, clang-format for C++ files and file size checker
are migrated from TAOS-CI.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
MyungJoo Ham [Tue, 7 Feb 2023 07:22:09 +0000 (16:22 +0900)]
github-action/risc-v test update
Use dockerfile of nnstreamer's fork designed for nnstreamer packages.
Enable gtest, too.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Suyeon Kim [Thu, 18 Jan 2024 02:01:58 +0000 (11:01 +0900)]
[fix/svace] fix svace issue
Fix covertiy and SVACE issue
- model_path isn't initialized. fixed it.
Signed-off-by: Suyeon Kim <suyeon5.kim@samsung.com>
Suyeon Kim [Wed, 17 Jan 2024 05:33:59 +0000 (14:33 +0900)]
[add/onnxruntime] Update onnxruntime_support on
add true default onnxruntime option for Tizen releases
Signed-off-by: Suyeon Kim <suyeon5.kim@samsung.com>
hyunil park [Tue, 16 Jan 2024 09:56:49 +0000 (18:56 +0900)]
[tensor_trainer] Move the function of checking invalid parameters
- Move gst_tensor_trainer_check_invalid_param to enable setting properties in PAUSED states
Signed-off-by: hyunil park <hyunil46.park@samsung.com>
Jaeyun Jung [Fri, 12 Jan 2024 11:53:24 +0000 (20:53 +0900)]
[CodeClean] util to get list size
Code clean, fix indent and util function to get list size.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun Jung [Fri, 12 Jan 2024 11:51:48 +0000 (20:51 +0900)]
[Filter/ArmNN] fix mem leak
Fix mem leak, free model path in destructor.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Yongjoo Ahn [Fri, 12 Jan 2024 07:35:59 +0000 (16:35 +0900)]
[build] Let meson check API version of SNPE
- Let meson.build check SNPE API version.
- Added compile flag `SNPE_VERSION_MAJOR` would be used in future commits.
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Yongjoo Ahn [Fri, 12 Jan 2024 07:52:47 +0000 (16:52 +0900)]
[filter] Rename `tensor_filter_snpe.cc` to `tensor_filter_snpe_v1.cc`
- Current source code targets SNPE version 1.
- Future commits will add source code targets SNPE version 2.
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Yelin Jeong [Fri, 12 Jan 2024 06:07:35 +0000 (15:07 +0900)]
[tests] Add SSAT test for padding mode in tensor_transform
This patch adds SSAT test for padding mode in tensor_transform.
Signed-off-by: Yelin Jeong <yelini.jeong@samsung.com>
Yelin Jeong [Fri, 12 Jan 2024 05:00:09 +0000 (14:00 +0900)]
[transform] Add tensor padding mode in tensor_transform
This patch adds tensor padding mode in tensor_transform.
tensor padding has options like this.
left, right, top, bottom, front, back, type (NCHW or NHWC).
Signed-off-by: Yelin Jeong <yelini.jeong@samsung.com>
Yelin Jeong [Fri, 12 Jan 2024 04:47:36 +0000 (13:47 +0900)]
[common] Move enum tensor_layout
This patch moves tensor_layout enum to tensor_typedef.
tensor_layout can be used in other elements.
Signed-off-by: Yelin Jeong <yelini.jeong@samsung.com>
Yongjoo Ahn [Fri, 5 Jan 2024 09:20:45 +0000 (18:20 +0900)]
[trivial] Fix multiline comments to meet doxygen requiremnent
- Fix multiline comments in unittest_mlagent.cc to satisfy CI
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Yongjoo Ahn [Fri, 5 Jan 2024 09:17:27 +0000 (18:17 +0900)]
[mlagent] Let `mlagent_get_model_path_from` returns uri in fallback
- Let the function returns properly allocated gchar* in fallback case.
- Fix memleak (g_strfreev).
- Fix memleak in a testcase.
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Yongjoo Ahn [Fri, 5 Jan 2024 08:41:38 +0000 (17:41 +0900)]
[test] Add a missing header in mock_mlagent.h
- Add a missing header <utility> in mock_mlagent.h
- It declares `std::exchange`
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Suyeon Kim [Tue, 9 Jan 2024 01:26:24 +0000 (10:26 +0900)]
[fix/onnxruntime-filter] fixed tensor with negative value in shape
Some onnx models had free dimensions for batch. ex) [None,3,720,720]
It fixed that creating tensor with negative value in shape '''Ort::Value''' is treated as 1.
Signed-off-by: Suyeon Kim <suyeon5.kim@samsung.com>
Jaeyun Jung [Mon, 8 Jan 2024 10:34:30 +0000 (19:34 +0900)]
[Py/CodeClean] pylock after checking ptr
Code clean, apply gil after checking null ptr.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Yongjoo Ahn [Fri, 5 Jan 2024 08:20:36 +0000 (17:20 +0900)]
[build] Remove unnecessary BuildRequires
- The latest tflite2 tizen package has its own flatbuffers header
internally. Let's remove the `BuildRequires: flatbuffers-debvel`
- Let meson check the header when build_platform is not tizen
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Jaeyun Jung [Mon, 8 Jan 2024 10:36:01 +0000 (19:36 +0900)]
[Pkg] fix library name for nns internal
nnstreamer-internal pkg requires nns-single library.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
MyungJoo Ham [Wed, 13 Dec 2023 05:41:54 +0000 (14:41 +0900)]
Centralize python interpreter init/fini for subplugins
Python subplugins of decoder, converter, filter should
share the init/fini status.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
MyungJoo Ham [Wed, 13 Dec 2023 04:44:46 +0000 (13:44 +0900)]
filter/python3: revise GIL access.
Control GIL at C callbacks.
Control mutex at C++ class.
As in decoder/converter changes.
This is the last step before the final step for #3834
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
MyungJoo Ham [Tue, 12 Dec 2023 10:10:06 +0000 (19:10 +0900)]
converter::python make them thread safe with GIL
Apply GIL control to converter::python3 as in we do with filter::python3.
Fixes converter part of #3834
Overrides #3872
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Jaeyun Jung [Tue, 2 Jan 2024 05:29:09 +0000 (14:29 +0900)]
[CodeClean/grpc] clearly define release function
The type of data pushed into queue is gst-buffer.
Clearly set the release function when calling free.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun Jung [Wed, 3 Jan 2024 05:01:50 +0000 (14:01 +0900)]
[CodeClean] fix indent and codestyle
Trivial, fix indent and codestyle for line comment.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Yongjoo Ahn [Mon, 11 Dec 2023 09:03:38 +0000 (18:03 +0900)]
[build] Halt meson setup when NOT-GCC is provided for pytorch subplugin
- Halt meson setup procedure when NOT-GCC is given and pytorch-support
is enabled. Rather than warning and go.
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Jaeyun Jung [Tue, 2 Jan 2024 09:40:23 +0000 (18:40 +0900)]
[Pkg] header file for devel package
1. move util for trainer (single-devel > devel)
2. update library (link to nns-single, internal devel package)
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Wook Song [Thu, 14 Dec 2023 04:08:03 +0000 (13:08 +0900)]
Dist/Tizen: Add an option to control MLAgent support
This patch adds an option for MLAgent support control to the Tizen RPM
spec file.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Wook Song [Thu, 14 Dec 2023 03:57:55 +0000 (12:57 +0900)]
Filter/MLAgent: Add unit test cases for parsing MLAgent URIs
This patch adds test cases for unit testing of parsing the MLAgent URIs.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Wook Song [Tue, 28 Nov 2023 04:38:19 +0000 (13:38 +0900)]
Filter/MLAgent: Implement the JSON parsing logic to get model's path
This patch adds the logic to parse the C-style stringified JSON data
gathered from the ML Agent in order to convert the given URI to the
model's path.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Wook Song [Mon, 27 Nov 2023 10:33:37 +0000 (19:33 +0900)]
Filter/MLAgent: Do not use g_autoptr for a const pointer
This is a trivial bug fix that removes the uneccessary attribute,
g_autofree, for the c string gathered from GObject, which is a const
pointer.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Wook Song [Tue, 21 Nov 2023 04:47:04 +0000 (13:47 +0900)]
Filter/MLAgent: Use a proper data type argument for ml_agent_model_get
This patch is a trivial bug fix that changes the data type of an
argument for ml_agent_model_get.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Yongjoo Ahn [Tue, 26 Dec 2023 07:53:38 +0000 (16:53 +0900)]
[build] Let meson check ncnn pkg-config
- Let meson check ncnn pc file first, and do fallback check then.
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Jaeyun Jung [Thu, 21 Dec 2023 09:41:50 +0000 (18:41 +0900)]
[Build/TF-lite-custom] gen shared lib in build dir
Code clean, generate tflite-custom library in build directory.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
gichan2-jang [Tue, 26 Dec 2023 05:41:07 +0000 (14:41 +0900)]
Remove flatbuf-python dependency
Remove flatbuf-python dependency from all Tizen profile.
Signed-off-by: gichan2-jang <gichan2.jang@samsung.com>
Jaeyun Jung [Thu, 21 Dec 2023 10:22:04 +0000 (19:22 +0900)]
[Protobuf] missing tensor type
Add missing tensor type 'float16' when generating protobuf and flatbuf code.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun Jung [Tue, 19 Dec 2023 12:18:13 +0000 (21:18 +0900)]
[Build/Edge] check nns-edge dependency
Code clean, use pkgconfig to check nnstreamer-edge dependency.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
MyungJoo Ham [Fri, 22 Dec 2023 04:52:33 +0000 (13:52 +0900)]
Ubuntu: clean up control file.
We have multiple control files for different distros.
Clean up ubuntu-ppa control file.
Reported by Q2 of #4357
Fixes the first half of #4358
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Jaeyun Jung [Wed, 20 Dec 2023 10:49:30 +0000 (19:49 +0900)]
[Build] dependency to gstreamer
Remove unnecessary dependency of gstreamer when building subplugins.
(e.g., we should support filter-subplugin for single-shot, without gst.)
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun Jung [Wed, 20 Dec 2023 12:42:17 +0000 (21:42 +0900)]
[Doc] fix typo
Trivial, update docs and lib name.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Sungbin Jo [Mon, 18 Dec 2023 18:48:01 +0000 (03:48 +0900)]
Update code based on feedback
Incorporate the feedback given from the code review process.
Signed-off-by: Sungbin Jo <goranmoomin@daum.net>
chosanglyul [Mon, 18 Dec 2023 04:25:44 +0000 (13:25 +0900)]
Add documention for the ncnn subplugin
Update usage examples according to nnstreamer-example
Signed-off-by: chosanglyul <chosanglyul@gmail.com>
Kijun Shin [Sat, 16 Dec 2023 17:48:01 +0000 (02:48 +0900)]
Fix ncnn unit test error
Signed-off-by: Kijun Shin <sharaelong.shin@gmail.com>
Kijun Shin [Sat, 16 Dec 2023 16:42:55 +0000 (01:42 +0900)]
Add unit test cases
Signed-off-by: Kijun Shin <sharaelong.shin@gmail.com>
Kijun Shin [Sat, 16 Dec 2023 16:37:54 +0000 (01:37 +0900)]
WIP ncnn unit test
Signed-off-by: Kijun Shin <sharaelong.shin@gmail.com>
chosanglyul [Sat, 16 Dec 2023 10:31:27 +0000 (19:31 +0900)]
Fix compilation error when vulkan is disabled
Signed-off-by: chosanglyul <chosanglyul@gmail.com>
chosanglyul [Sat, 16 Dec 2023 10:10:14 +0000 (19:10 +0900)]
Add options for the ncnn subplugin
Signed-off-by: chosanglyul <chosanglyul@gmail.com>
chosanglyul [Sat, 9 Dec 2023 12:36:24 +0000 (21:36 +0900)]
Add main ncnn inference code
Signed-off-by: chosanglyul <chosanglyul@gmail.com>
Sungbin Jo [Mon, 18 Sep 2023 03:00:14 +0000 (12:00 +0900)]
Add ncnn subplugin skeleton code
Signed-off-by: Sungbin Jo <goranmoomin@daum.net>
Yelin Jeong [Tue, 19 Dec 2023 07:58:31 +0000 (16:58 +0900)]
[nnfw] Check NNFW_MAX_RANK in nnfw filter
This patch adds to check NNFW_MAX_RANK in nnfw filter.
If tensor info's rank is larger than NNFW_MAX_RANK and min_rank is smaller, use min_rank instead.
Signed-off-by: Yelin Jeong <yelini.jeong@samsung.com>
MyungJoo Ham [Tue, 12 Dec 2023 10:20:54 +0000 (19:20 +0900)]
decoder::python make them thread safe with GIL.
Apply GIL control to decoder::python3 as in we do with filter::python3.
Fixes decoder part of #3834
Overrides #3872
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
MyungJoo Ham [Fri, 25 Nov 2022 01:56:52 +0000 (10:56 +0900)]
Add SSAT to RISC-V github-action
Get SSAT and build SSAT for RISC-V
Then, Run!
@TODO: SSAT TC fails in RISC-V.
Either fix failed cases or let it ignore in RISC-V build.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
MyungJoo Ham [Tue, 12 Dec 2023 10:28:41 +0000 (19:28 +0900)]
filter::python3: do not break converter/decoder
A python-global reference counter is required for py-finalize.
This is workaround before applying such a global reference counter.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Yongjoo Ahn [Mon, 18 Dec 2023 07:57:29 +0000 (16:57 +0900)]
[tensor_transform] Fix confusing usage of transpose
- Correct usage is `... option=1:2:0:3` when a user want to change tensor
layout NHWC into NCHW.
- Fix document abuout it.
- Fix incorrect usage in a SSAT test.
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Jaeyun Jung [Mon, 18 Dec 2023 05:07:02 +0000 (14:07 +0900)]
[CodeClean] fix svace issue
Fix svace issue, prevent index error case.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Sangjung Woo [Fri, 15 Dec 2023 05:32:58 +0000 (14:32 +0900)]
[Test] Remove the MOBILE and WEARABLE tags in header file
Since both mobile and wearable profiles in Tizen project are removed,
related tags also should be removed. This patch removes them at Tizen
PM's request.
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
Yongjoo Ahn [Mon, 11 Dec 2023 08:55:04 +0000 (17:55 +0900)]
[build] Check flatbuffers header files before build tflite subplugin
- TFLite API requires flatbuffers header. Let meson check required header
- Add `BuildRequires` flatbuffers-devel explicitly in spec
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
MyungJoo Ham [Thu, 17 Aug 2023 06:02:17 +0000 (15:02 +0900)]
Suppress valgrind errors due to system libraries
Added a valgrind-suppression configuration that ignores Valgrind errors from:
- Python runtime
- Between glib/gstreamer
- Not freeing memory allocated by gstreamer/gobject class init.
This works with SSAT 1.3.0+.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
MyungJoo Ham [Fri, 22 Apr 2022 07:57:15 +0000 (16:57 +0900)]
github action: add tests of valgrind, gtest, ssat.
Add test runs for github-action - Ubuntu clean build.
However, do not block PRs for valgrind errors until
we get most valgrind errors fixed.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Jaeyun Jung [Fri, 8 Dec 2023 08:50:02 +0000 (17:50 +0900)]
[Filter/TF-lite] fill 1 when changing input dimension
After implementing 0-init dims, tf-lite subplugin fills zero when changing input dimension.
Fill 1 if given dimension is 0 dim value.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun Jung [Mon, 4 Dec 2023 07:48:42 +0000 (16:48 +0900)]
[Common] check tensor dim to append previous caps
Prevent appending same caps if tensor dimension has min rank 4.
(e.g., given dim is 1001:1, previous caps with rank 4 is unnecessary.)
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun Jung [Mon, 4 Dec 2023 08:22:59 +0000 (17:22 +0900)]
[Filter] condition to update dimension
For caps negotiation, compare tensor dimension if caps is tensor stream and already configured.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun Jung [Mon, 4 Dec 2023 06:58:39 +0000 (15:58 +0900)]
[QueryClient] handle tensor buffer
1. Decrease the ref count of incoming buffer if an error occurs.
2. Check the number of requested buffer before creating edge-data handle, and try to get the output from query-server.
3. Append gst-buffer using util function if current caps is tensor stream.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Yongjoo Ahn [Fri, 8 Dec 2023 09:24:25 +0000 (18:24 +0900)]
[build] Set cpp_std=c++17 (changed from c++14)
- Recent neural network frameworks needs c++17 standards to use its APIs
- Let nnstreamer be built with c++17.
- Add a Wno flag to snpe filter.
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
MyungJoo Ham [Wed, 6 Dec 2023 08:28:15 +0000 (17:28 +0900)]
RISCV/lua: disable an exception handling test case
Somehow, in Ubuntu 20.04 filter::lua@RISCV cannot handle
exceptions as x64/aarch64 handles.
Issue #4342 will track this.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
MyungJoo Ham [Mon, 19 Dec 2022 03:28:04 +0000 (12:28 +0900)]
Configure unit test environments
Unit test requires a few more configurations for plugins.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
MyungJoo Ham [Fri, 16 Dec 2022 08:01:11 +0000 (17:01 +0900)]
github-action/riscv64: update build name
The build name was incorrect. It's Ubuntu now!
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
SuYeon [Mon, 4 Dec 2023 05:25:47 +0000 (14:25 +0900)]
[Test/onnxruntime] Add unit tests and ssat tests for ONNX Runtime
- Add gtest based unit test for ONNX Runtime
- Add SSAT test cases for ONNX Runtime tensor filter subplugin
Signed-off-by: SuYeon <suyeon5.kim@samsung.com>
SuYeon [Mon, 4 Dec 2023 05:10:19 +0000 (14:10 +0900)]
[Filter/onnxruntime] Add filter subplugin for ONNX Runtime
- Add onnxruntime filter with ONNX runtime
- To build ONNX Runtime: https://onnxruntime.ai/
TODO:
- Only supports CPU. GPU and oter hardware support is NYI.
- Only supports float32 with input/output. Other types are NYI.
Signed-off-by: SuYeon <suyeon5.kim@samsung.com>
Jaeyun Jung [Fri, 1 Dec 2023 04:58:26 +0000 (13:58 +0900)]
[Doc] remove unnecessary lines from ml-api
Code clean, now we do not define data structure in ML API repo and use common functions of nnstreamer.
Remove unnecessary lines defined in api repo.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
hyunil park [Thu, 30 Nov 2023 00:40:14 +0000 (09:40 +0900)]
[tensor_trainer] Modify description of model-load-path
Change to 'pre-trained' instead of 'existing' expression
Signed-off-by: hyunil park <hyunil46.park@samsung.com>
Jaeyun Jung [Wed, 22 Nov 2023 08:22:44 +0000 (17:22 +0900)]
[Transform] available caps without peer
tensor-transform cannot determine the caps, this element can set out-caps by using incoming caps.
Remove unnecessary query for peer caps and update types of out-caps if input has tensors.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
hyunil park [Thu, 23 Nov 2023 07:30:18 +0000 (16:30 +0900)]
[datareposrc] Bugfix: Fixed removing JSON file before unittest
- The Filename is changed for each TC because sometimes the JSON file of
unittest(fps30ReadFlexibleTensors) is deleted by g_remove() in the previous unittest
Signed-off-by: hyunil park <hyunil46.park@samsung.com>
Yongjoo Ahn [Wed, 22 Nov 2023 06:37:30 +0000 (15:37 +0900)]
[trivial] Fix some typos
- Fix some typos in some old source code.
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Yongjoo Ahn [Wed, 22 Nov 2023 06:36:40 +0000 (15:36 +0900)]
[build] Enable clang with some deps
- Fix compile error with clang
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Jaeyun Jung [Fri, 10 Nov 2023 03:43:17 +0000 (12:43 +0900)]
[Common] macro for max number of tensors
Change definition for the max number of tensors (256), and update related code.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Yongjoo Ahn [Thu, 23 Nov 2023 03:54:11 +0000 (12:54 +0900)]
[doc] Fix outdated urls
- Fix outdated urls for Tizen releases.
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Jaeyun Jung [Wed, 22 Nov 2023 11:12:04 +0000 (20:12 +0900)]
[Test] tensorflow-lite version in testcase
Fix version of tensorflow-lite.
Both version 2.x and 1.x can be used in testcase if tensor-filter property framework is tensorflow-lite.
Default version is v2.x, and it has higher priority.
Also, added minor patch to fix indent.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun Jung [Wed, 22 Nov 2023 12:12:37 +0000 (21:12 +0900)]
[Reviewers] update reviewers
Update reviewers list in nnstreamer repo.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Yongjoo Ahn [Wed, 22 Nov 2023 06:27:08 +0000 (15:27 +0900)]
[build] Remove unnecessary compile args for SNPE sub-plugin
- Remove those args thus it's not needed with recent snpe sdk.
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Jaeyun Jung [Mon, 20 Nov 2023 12:00:41 +0000 (21:00 +0900)]
[CodeClean/Element] util function to handle tensors
Code clean, use common util function to handle tensor information and memory.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun Jung [Mon, 20 Nov 2023 10:59:12 +0000 (19:59 +0900)]
[Common/CodeClean] update description
Code clean, update doc and API description.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
gichan2-jang [Tue, 21 Nov 2023 01:58:20 +0000 (10:58 +0900)]
[TEST] Add tensor query test for drop prop
- Add tensor query test for drop prop.
- Add framerate for tensor query server.
Signed-off-by: gichan2-jang <gichan2.jang@samsung.com>
gichan2-jang [Tue, 21 Nov 2023 01:07:38 +0000 (10:07 +0900)]
[Query] Add drop prop to query client
Currently, the query client sends all input buffers to the query server.
If the processing speed of the server is slower than the client,
The result time of the server is gradually different from the input time.
To prevent this, the server processes only two buffers, and drop the remaining input buffers.
Signed-off-by: gichan2-jang <gichan2.jang@samsung.com>
Jaeyun Jung [Tue, 7 Nov 2023 11:54:06 +0000 (20:54 +0900)]
[Subplugin/flatbuf/protobuf] util to handle tensor data
Use common util function to handle tensor data.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Wook Song [Fri, 17 Nov 2023 05:28:00 +0000 (14:28 +0900)]
Common/API: Introduce g_auto for automatic memory cleaning of "gchar *"
This patch introduces g_autofree and g_auto for automatic memory
cleaning of "gchar *" and "gchar **", respectively. Note that "GStrv" is
a typedef alias for "gchar **".
Signed-off-by: Wook Song <wook16.song@samsung.com>
Jaeyun Jung [Mon, 13 Nov 2023 05:31:40 +0000 (14:31 +0900)]
[Edge] util to handle tensor info
Code clean, fix indent and use util function if current caps is tensor stream.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun Jung [Thu, 9 Nov 2023 08:55:29 +0000 (17:55 +0900)]
[Test/CodeClean] util to append tensors
Code clean, remove duplicated code and use util function to append tensor memory.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun Jung [Fri, 17 Nov 2023 06:32:16 +0000 (15:32 +0900)]
[DataRepo/CodeClean] parse gst-caps
Code clean,
- reduce function call, parse caps and update stream type.
- update log message.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>