Hyoung Joo Ahn [Thu, 25 Jun 2020 10:39:34 +0000 (19:39 +0900)]
[Filter] fix a memory leak
the allocated memory(when `allocate_in_invoke` is FALSE) should be free when `invoke()` returns not 0(OK)
Signed-off-by: Hyoung Joo Ahn <hello.ahn@samsung.com>
Wook Song [Wed, 24 Jun 2020 07:09:21 +0000 (16:09 +0900)]
[Tests/Filter/EdgeTPU] Enable test() in the meson build script
This patch enables test(), which has been commented out, in the meson
build script for the EdgeTPU filter.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Wook Song [Wed, 24 Jun 2020 06:49:51 +0000 (15:49 +0900)]
[Meson] Add missing environment variable, NNSTREAMER_SOURCE_ROOT_PATH
This patch adds the missing environment variable,
NNSTREAMER_SOURCE_ROOT_PATH, to the test environment object in the
meson build script. This variable is necessary to run the test cases
via 'ninja test'.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Jaeyun [Fri, 19 Jun 2020 06:19:55 +0000 (15:19 +0900)]
[Mux/Merge] code clean
code clean, prepare new sync-mode.
- remove unnecessary param
- reduce code complexity
- change return val (true to push buffer to src pad)
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Tue, 23 Jun 2020 11:17:32 +0000 (20:17 +0900)]
[CodeClean] remove unnecessary defines
code clean, remove unnecessary predefinition to set pad template.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
gichan-jang [Wed, 24 Jun 2020 11:04:43 +0000 (20:04 +0900)]
[TEST] Fix nnfw unit test
When testing the operation of the multi-framework, if tflite is disabled, the test fails.
If tflite is disabled, skip the test.
Signed-off-by: gichan-jang <gichan2.jang@samsung.com>
Wook Song [Wed, 24 Jun 2020 05:57:51 +0000 (14:57 +0900)]
[Tests] Rename NNSTREAMER_BUILD_ROOT_PATH with NNSTREAMER_SOURCE_ROOT_PATH
In most cases, the environment variable, NNSTREAMER_BUILD_ROOT_PATH,
is used for indicating the root directory of the path to which the
'test_models' directory belongs, which means the source root path. To
avoid confusion, this patch renames it with NNSTREAMER_SOURCE_ROOT_PATH.
Signed-off-by: Wook Song <wook16.song@samsung.com>
gichan-jang [Wed, 26 Feb 2020 11:30:28 +0000 (20:30 +0900)]
[Decoder] Add tensor decoder subplugin flatbuffers
Add tensor decoder subplugin for flatbuffer
flatbuffers package can be found here :
https://launchpad.net/~nnstreamer/+archive/ubuntu/ppa-build-test/+packages
Signed-off-by: gichan-jang <gichan2.jang@samsung.com>
Yongjoo Ahn [Tue, 23 Jun 2020 05:20:38 +0000 (14:20 +0900)]
[Test] Add test for imageseg-decoder and tensor_filter profiling
- Add test cases for newly added option `snpe-deeplab` and `snpe-depth`
- It also check profiling option of tensor_filter
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Dongju Chae [Thu, 18 Jun 2020 05:43:18 +0000 (14:43 +0900)]
[Conf/MemLeak] Add missing g_free() to nnstreamer_conf.c
This patch adds missing `g_free()` to `nnstreamer_conf.c`
`nnsconf_get_custom_value_string()` allocates `hashkey` in the first
line. However, if a value is found in the first hash lookup or a value
does not exist in anywhere, `hashkey` is not inserted to the hash table,
which results in memory leak.
Signed-off-by: Dongju Chae <dongju.chae@samsung.com>
Jaeyun [Fri, 19 Jun 2020 06:45:48 +0000 (15:45 +0900)]
[RepoSink] ignore sync and preroll
repo-sink may not receive a buffer when pipeline constructed.
set sync and async option to ignore this case.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
gichan-jang [Fri, 19 Jun 2020 08:01:00 +0000 (17:01 +0900)]
[Coverage] Exclude generated files by flatbuf
Exclude generated files by flatbuf from coverage report
Signed-off-by: gichan-jang <gichan2.jang@samsung.com>
Wook Song [Thu, 18 Jun 2020 02:57:06 +0000 (11:57 +0900)]
[Tests/EdgeTPU] Use dummy device mode instead of dummy library
This patch revises the unit test cases for EdgeTPU to use dummy device
mode instead of the dummy EdgeTPU library.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Jaeyun [Wed, 17 Jun 2020 03:45:50 +0000 (12:45 +0900)]
[C-Api] handle invalid info when creating data handle
When creating the data handle, return error if given tensors-info is invalid.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
MyungJoo Ham [Fri, 5 Jun 2020 05:51:55 +0000 (14:51 +0900)]
TAOS-CI / Enable clang-format for C++ codes
1. Enable clang TAOS-CI module
2. Add "C++ ONLY" mode for it.
Note that I need to update TAOS-CI to make it accept such options.
This depends on https://github.com/nnstreamer/TAOS-CI/pull/628
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Jaeyun [Tue, 16 Jun 2020 07:35:35 +0000 (16:35 +0900)]
[Test/C-Api] add testcase of internal functions
add simple testcases to handle the case of invalid param.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Tue, 16 Jun 2020 07:26:21 +0000 (16:26 +0900)]
[C-Api] error code when failed to allocate tensor data
Remove unnecessary error code (stream-pipe).
If failed to allocate mem for tensors-data handle, API returns out-of-mem error.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Mon, 15 Jun 2020 12:47:53 +0000 (21:47 +0900)]
[Meson] converter sub-plugin path in test-env
Update sub-plugin path in test-env, set converter plugin path.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Dongju Chae [Tue, 16 Jun 2020 11:04:20 +0000 (20:04 +0900)]
[Fix] Remove unncessary arg checking before calling g_strdup()
This patch removes uncessary argument checking before calling g_strdup()
If an argument of g_strdup() is NULL, its return value is also NULL.
So, checking its argument before calling is duplicated and useless.
Signed-off-by: Dongju Chae <dongju.chae@samsung.com>
Hyoung Joo Ahn [Tue, 16 Jun 2020 03:13:41 +0000 (12:13 +0900)]
[Repo/Test] make it easier to understand
applying line change at the complicated pipeline to understand easier.
Signed-off-by: Hyoung Joo Ahn <hello.ahn@samsung.com>
Wook Song [Thu, 11 Jun 2020 07:47:17 +0000 (16:47 +0900)]
[Filter/EdgeTPU] Add a 'dummy' device mode for the test purpose
This patch adds a 'dummy' device mode for the test purpose instead of
using the dummy shared library that pretends to be libedgetpu.so.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Jaeyun [Wed, 10 Jun 2020 04:11:21 +0000 (13:11 +0900)]
[Android] do not convert meta in invoke method
To improve performance, it is necessary to remove metadata conversion in invoke method.
Do not handle dynamic mode in invoke method and update description to change the tensor info.
Also, if output info is changed using set-info method, update cached instance in private data.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Wed, 10 Jun 2020 04:03:22 +0000 (13:03 +0900)]
[Android] create info handle in private data
create info handle when allocating private data.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
junmin-kim [Tue, 16 Jun 2020 05:46:30 +0000 (14:46 +0900)]
[Filter/tflite] Fix typo
Assignment should be ended with semicolon if there is no special reason.
Replace `,` with `;`
Signed-off-by: junmin-kim <junmindd.kim@samsung.com>
Jaeyun [Wed, 10 Jun 2020 09:08:33 +0000 (18:08 +0900)]
[Android] constructor with info
In tensors-data class, update private constructor with tensor-info.
This can remove method call to set metadata in native.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Dongju Chae [Fri, 12 Jun 2020 07:36:37 +0000 (16:36 +0900)]
[Bugfix/Auto] Fix the bug to access freed memory region in auto fwname
This patch fixes the memory bug in auto tensor filter detection.
`gst_tensor_filter_get_available_framework` assumes that `fw_name` and
`prop->fwname` are pointing different memory regions.
However, `_gtfc_setprop_Model` has passed the argument `fw_name` as
`prop->fwname`, which may cause the invalid memory access.
In this case, `fw_name` is being accessed after `prop->fwname` is freed.
Signed-off-by: Dongju Chae <dongju.chae@samsung.com>
gichan-jang [Thu, 14 May 2020 02:25:12 +0000 (11:25 +0900)]
[Tensor Converter] Add tensor converter subplugin flatbuffer
Add tensor converter subplugin flatbuffer
Signed-off-by: gichan-jang <gichan2.jang@samsung.com>
Jaeyun [Wed, 10 Jun 2020 10:35:40 +0000 (19:35 +0900)]
[Filter/Custom] copy metadata in registration
In the filter custom-easy, prevent error case when developer tries to register custom-filter with local variables.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Wook Song [Wed, 10 Jun 2020 10:33:42 +0000 (19:33 +0900)]
[Filter/EdgeTPU] Handle the error case when a compiled model is given
This patch is to handle the error case when a model compiled by
edgetpu-compiler is given. In general, a higher version of
tensorflow-lite (> 1.13) could support more various types of the tflite
models.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Jaeyun [Fri, 12 Jun 2020 10:30:16 +0000 (19:30 +0900)]
[Converter] reduce complexity
code clean to reduce complexity
1. internal function to push multiple buffers.
2. remove unnecessary args
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Fri, 12 Jun 2020 08:27:56 +0000 (17:27 +0900)]
[Android/Doc] build options
Update document, build options for android library.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Wook Song [Mon, 15 Jun 2020 05:29:53 +0000 (14:29 +0900)]
[Tests/Meson] Fix the unknown variable error related to python filters
This patch fixes the unknown variable error related to the filter
sub-plugins for python 2 and 3.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Wook Song [Mon, 15 Jun 2020 05:12:12 +0000 (14:12 +0900)]
[Meson] Remove trailing whitespaces
This is a trivial patch to eliminate trailing whitespaces of the meson
build script in the source root.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Jaeyun [Mon, 15 Jun 2020 05:05:10 +0000 (14:05 +0900)]
[Doc] fix wrong file name
fix wrong file name in converter header file.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Fri, 12 Jun 2020 09:18:50 +0000 (18:18 +0900)]
[Doc] fix invalid document
typo correction and fix document for converter sub-plugin api.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Sangjung Woo [Thu, 11 Jun 2020 02:12:11 +0000 (11:12 +0900)]
[Tests/Filter] Add test case for NNS_custom_easy_unregister()
This patch newl adds test cases for NNS_custom_easy_unregister().
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
Sangjung Woo [Thu, 11 Jun 2020 01:58:56 +0000 (10:58 +0900)]
[Filter/Custom/Easy] Add unregister function
This patch newly adds 'NNS_custom_easy_unregister()' function to remove
registered function so that pipeline itself can be reused without
changing custom model name.
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
gichan-jang [Wed, 10 Jun 2020 05:40:26 +0000 (14:40 +0900)]
[Debian/control] Add flatbuf-dev build dependent package
As flatbuf headers are separated from tflite-dev pkg, flatbuf-dev pkg is added.
Signed-off-by: gichan-jang <gichan2.jang@samsung.com>
gichan-jang [Wed, 10 Jun 2020 10:23:41 +0000 (19:23 +0900)]
[Converter] Change find external converter function
Change to search for sub-plugins according to the mime type regardless of the name of the library
Signed-off-by: gichan-jang <gichan2.jang@samsung.com>
gichan-jang [Fri, 29 May 2020 07:09:17 +0000 (16:09 +0900)]
[Converter] Change prototype of subplugin
Change subplugin function prototype of tensor converter
Signed-off-by: gichan-jang <gichan2.jang@samsung.com>
gichan-jang [Thu, 28 May 2020 01:45:33 +0000 (10:45 +0900)]
[Converter] Change tensor config to tensors config
Change tensor config to tensors config. Assume single tensor only except subplugin.
Signed-off-by: gichan-jang <gichan2.jang@samsung.com>
gichan-jang [Mon, 18 May 2020 01:37:16 +0000 (10:37 +0900)]
[Converter] Prepare to register sub-plugin
Preprare to register sub-plugin for flatbuffers and protobuf
Signed-off-by: gichan-jang <gichan2.jang@samsung.com>
Wook Song [Wed, 10 Jun 2020 07:54:43 +0000 (16:54 +0900)]
[Filter/C++] Do not go further when failed to open NN framework
This patch adds more catch blocks that return negative error code in
cpp_open so that the execution does not go further when it failed to
open NN framework.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Wook Song [Wed, 10 Jun 2020 04:51:24 +0000 (13:51 +0900)]
[Dist/Debian] Provide a sub-package for the Edge TPU extension
This patch provides a sub-package for the Edge TPU extension.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Wook Song [Wed, 10 Jun 2020 02:50:10 +0000 (11:50 +0900)]
[Ext/Filter/Meson] Revise resolving build dependency for Edge TPU
This patch revises the mechanism that resolves the build dependency on
libedgetpu and header files.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Wook Song [Wed, 10 Jun 2020 02:27:56 +0000 (11:27 +0900)]
[Ext/Filter/Meson] Elminate white spaces
This is a trivial fix that eliminates white spaces in the meson build
script for the filter extensions.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Sangjung Woo [Wed, 10 Jun 2020 01:16:46 +0000 (10:16 +0900)]
[Tests] Add 'nosync' option into mux
To fix `framerate` issue in mux, it set sync-mode to slowest as default.
However, This LSTM test case is hanging as a side-effect since it uses
both tensor_repo and tensor_mux. This patch adds 'nosync' option into
mux so that the test pipeline starts and finishes without hanging.
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
Sangjung Woo [Tue, 9 Jun 2020 05:10:26 +0000 (14:10 +0900)]
[Mux] Set sync-mode to slowest as default
After passing mux / demux, the framerate is extremely bigger than
before. (Issue #2440) This is mainly because sync_mode of tensor_mux
filter is SYNC_NOSYNC. To fix this bug, this patch set sync_mode to
SYNC_SLOWEST as default.
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
MyungJoo Ham [Tue, 9 Jun 2020 10:29:48 +0000 (19:29 +0900)]
Filter/Custom/Easy: fix doxygen entry
Fix easy-custom-filter API's doxygen entry.
It gets output dimenesion info as an input, not output.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Jaeyun [Tue, 9 Jun 2020 06:22:21 +0000 (15:22 +0900)]
[Android] implement on-load in native
register native functions when loading native library.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Tue, 9 Jun 2020 04:24:12 +0000 (13:24 +0900)]
[Android] return to static build
To execute model faster, return to static build.
Make all sub-plugins static libs and build one native so file.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Yongjoo Ahn [Fri, 5 Jun 2020 07:15:54 +0000 (16:15 +0900)]
[Fix][Filter/Prop] Fix performance property (latency) for tensorfilter
- Initialize with zero value.
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Dongju Chae [Thu, 4 Jun 2020 11:25:54 +0000 (20:25 +0900)]
[Decoder/Neon] Support neon accl for SNPE deeplab/depth seg decoder
This patch supports neon accl for SNPE deeplab/depth seg decoder
Signed-off-by: Dongju Chae <dongju.chae@samsung.com>
Dongju Chae [Thu, 4 Jun 2020 10:53:01 +0000 (19:53 +0900)]
[Decoder/Fix] Revise tensordec-imagesegment impl. and fix minor bugs
This patch revises tensordec-imagesegment impl. and fix minor bugs
Signed-off-by: Dongju Chae <dongju.chae@samsung.com>
Jaeyun [Wed, 3 Jun 2020 05:19:06 +0000 (14:19 +0900)]
[Android] private data to set metadata
Each invoke callback gets/returns tensor data object including metadata.
To improve performance, Cache tensor metadata object and avoid duplicated conversion.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
MyungJoo Ham [Thu, 4 Jun 2020 02:50:24 +0000 (11:50 +0900)]
Start development of 1.5.3 as 1.6.0-RC4
As soon as all major known bugs and SVACE/Coverity issues are fixed,
I'm going to release 1.6.0.
Cheers!
```
1.5.2 -> 1.5.3
- Mediapipe's graphs (NNStreamer pipeline equivalent) may be embedded as an element in NNStreamer pipeline
- Qualcomm SNPE is supported (tensor-filter subplugin)
- Verisilicon Vivante is supported (tensor-filter subplugin)
- NNStreamer --> Protobuf decoder added.
- New tensor-filter subplugin API "v1" released. (with v0 backward compatibility)
- Tensor-filter now accepts C++ classes as a subplugin. Edge-TPU subplugin is re-written as an example.
- Meson script re-worked.
- Semantics of hardware-acceleration options for tensor-filter re-worked.
- API/Android: nnfw-runtime (neurun) and SNPE support
- API/Android: usability update.
- API/Android: less invokation latency. (more optimization coming in next versions)
- API/C: bugfixes, architectural upgrade, latency reduction.
- tensor-filter has latency and throughput performance monitors.
- tensor-sink is by default "sync=false". If appsink or tensor_sink in NNStreamer Pipeline API's pipeline has sync=true, emit warning messages.
- Architectural updates: lower CC, less duplication, removed dependency cycles, less complicated #if statements and blocks.
- Test suite updates: timeout handling, arm-arch error fixes, more test cases (supplying more negative cases), floating-point handling. and a lot more fixes and performance (latency) optimizations.
- Build script updates: cleaned up dependencies and applied "feature" meson feature.
- No more essential class in assert() for optimized binaries.
- Much less assertions. Apply error-handling instead.
- Daily build & test activated and published.
- Rules & policy updated to comply with LF/AI.
- Memory leaks removed: from demux, split, and a few more components.
- Known bug: edge-TPU subplugin is not working in Tizen devices recently.
```
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Jaeyun [Fri, 5 Jun 2020 13:41:37 +0000 (22:41 +0900)]
[Android] typo in mk file
fix typo in mk file (jni nnstreamer defines)
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
MyungJoo Ham [Fri, 5 Jun 2020 05:57:22 +0000 (14:57 +0900)]
Tizen/spec: disable mvncsdk/edgeTPU for DA build
DA requested to remove mvncsdk/edgeTPU from the dependency.
Disable them for their builds.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Jaeyun [Thu, 4 Jun 2020 11:16:20 +0000 (20:16 +0900)]
[NNFW] set backend after loading model
recently updated, nnfw set-backend should be called after loading the model.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Geunsik Lim [Tue, 2 Jun 2020 11:10:18 +0000 (20:10 +0900)]
doc: how to enable vivante tensor filter
This commit is to append how to enable the Vivante tensor filter
based on Vivante NPU Toolkit library on Ubuntu/ARM and Tizen/ARM devices.
Signed-off-by: Geunsik Lim <geunsik.lim@samsung.com>
Geunsik Lim [Thu, 4 Jun 2020 11:28:56 +0000 (20:28 +0900)]
Movidius Filter: Fixed typos
This commit is trivial. Let's fix typos in the Movidius filter file.
Signed-off-by: Geunsik Lim <geunsik.lim@samsung.com>
Hyoung Joo Ahn [Thu, 4 Jun 2020 08:38:30 +0000 (17:38 +0900)]
[Filter/TFLite] remove ineffective codes
setting allocation type is not effective anymore.
Signed-off-by: Hyoung Joo Ahn <hello.ahn@samsung.com>
Dongju Chae [Mon, 1 Jun 2020 12:31:08 +0000 (12:31 +0000)]
[Decoder] Extend tensordec-imagesegment to support snpe-deeplab/depth
This patch extends tensordec-imagesegment to support snpe-deeplab/depth.
Also, it include minor code refactoring.
Signed-off-by: Dongju Chae <dongju.chae@samsung.com>
Dongju Chae [Wed, 3 Jun 2020 04:01:35 +0000 (13:01 +0900)]
[Filter/Prop] Add performance properties (lat/fps) for tensor filter
This patch adds performance properties (latency/throughput) for tensor filter.
Those properties will be used for the overlay information, to show the performance
of tensor filter (e.g., SNPE).
Signed-off-by: Dongju Chae <dongju.chae@samsung.com>
Geunsik Lim [Thu, 4 Jun 2020 02:25:57 +0000 (11:25 +0900)]
SPEC: fixed indentation
This commit is trivial. It is to keep an indentation consistently.
Signed-off-by: Geunsik Lim <geunsik.lim@samsung.com>
Jaeyun [Wed, 3 Jun 2020 02:40:36 +0000 (11:40 +0900)]
[Android/Custom] drop input buffer
In custom-filter interface, returning null in invoke-callback will drop the input buffer.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Wed, 3 Jun 2020 05:08:39 +0000 (14:08 +0900)]
[C-Api] function to get pipeline element
Add new function to get the gst-element from pipeline handle.
For native developers, it is necessary to get pipeline element to use gstreamer functions.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Thu, 4 Jun 2020 02:38:10 +0000 (11:38 +0900)]
[Log] add definition log-fatal
define new nns_logf (common fatal-log definition with nns_log prefix)
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Yongjoo Ahn [Tue, 24 Mar 2020 08:48:03 +0000 (17:48 +0900)]
[Decoder/protobuf] Add protobuf tensor_decoder sub-plugin
- Add tensor_decoder sub-plugin "protobuf" to support Protocol Buffers
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
MyungJoo Ham [Wed, 3 Jun 2020 05:25:59 +0000 (14:25 +0900)]
[Filter] bugfix: dangling pointer access.
gfree should be called when you completed accesing it.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
MyungJoo Ham [Wed, 3 Jun 2020 06:26:13 +0000 (15:26 +0900)]
README update: add committer, update hw-accl, ci status
- Add the new committer
- Add the list of hardware accelerators
- Add daily-release status
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
MyungJoo Ham [Wed, 3 Jun 2020 06:21:43 +0000 (15:21 +0900)]
Doc update: component-description
There are a few components completed. Update component list
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Hyoung Joo Ahn [Wed, 3 Jun 2020 10:36:58 +0000 (19:36 +0900)]
[Filter] fix a tiny mem leak
move a few `g_free()` to fix mem leak
Signed-off-by: Hyoung Joo Ahn <hello.ahn@samsung.com>
Dongju Chae [Wed, 3 Jun 2020 09:30:08 +0000 (18:30 +0900)]
[Build/Fix] Add a missing header to 'nnstreamer_cppplugin_api_filter.hh'
This patch adds a missing header, string.h, to
'nnstreamer_cppplugin_api_filter.hh', which uses memcpy().
Signed-off-by: Dongju Chae <dongju.chae@samsung.com>
Jaeyun [Tue, 2 Jun 2020 07:20:26 +0000 (16:20 +0900)]
[Conf] parse sub-plugins name
Basename is unnecessary, set sub-plugins name while loading configuration.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Dongju Chae [Tue, 2 Jun 2020 01:26:21 +0000 (10:26 +0900)]
[Build/Tests] Fix 'tests/meson.build' to always find 'copy'
This patch fixes minor build errors when `libpng_dep` does not exist.
`tests/nnstreamer_filter_extensions_common/meson.build` is using this
variable regardless of `libpng_dep`'s existence.
Signed-off-by: Dongju Chae <dongju.chae@samsung.com>
Parichay Kapoor [Mon, 18 May 2020 18:13:51 +0000 (23:43 +0530)]
[single] Add missing gstTensorFilterFramework v1 support
Add missing support for gstTensorFilterFramework in checking framework availability
for a given backend accelerator/hardware
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Jaeyun [Wed, 20 May 2020 11:49:51 +0000 (20:49 +0900)]
[Android] cache method id
1. get required method ids to improve performance when initializing the pipe info.
2. add internal method to handle arraylist.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Fri, 29 May 2020 03:50:18 +0000 (12:50 +0900)]
[Filter] release mem block when execution failed
release in/out mem blocks when failed to invoke nn framework.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Yongjoo Ahn [Mon, 1 Jun 2020 07:38:17 +0000 (16:38 +0900)]
[Android/SNPE] Decouple SNPE env setting from native-api
- Move function for android SNPE env setting from nnstreamer-native-api.c to tensor_filter_snpe.cc
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Yongjoo Ahn [Wed, 27 May 2020 06:19:28 +0000 (15:19 +0900)]
[SNPE] Add custom properties
- Add Debug message
- Add custom prop for snpe runtime and cpu_fallback options
ex) "custom=Runtime:GPU,CPUFallback:true"
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Yongjoo Ahn [Fri, 15 May 2020 01:15:48 +0000 (10:15 +0900)]
[Android] prepare dsp/npu runtime for snpe filter
- Add required snpe-sdk libs
- Set `jniLibs.srcDirs` to this project explictly contains snpe-sdk libs for dsp/npu
- Make function `nns_set_env` to set env "ADSP_LIBRARY_PATH" (https://developer.qualcomm.com/docs/snpe/dsp_runtime.html)
- Log SNPE version
- Prevent build SNAP and SNPE together in android build script and
Android.mk
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
gichan-jang [Fri, 29 May 2020 07:50:25 +0000 (16:50 +0900)]
[Aggregator] Remove duplicated function to set caps
Remove duplicated works to set source pad cap when sink event is received
Signed-off-by: gichan-jang <gichan2.jang@samsung.com>
Sangjung Woo [Fri, 22 May 2020 05:26:47 +0000 (14:26 +0900)]
[Unittest] Fix a potential bug of unit testcase
This patch fixes the potential bug (i.e. pipeline parsing error) of unit
testcase when using Protobuf v3.8.
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
Parichay Kapoor [Fri, 15 May 2020 10:28:47 +0000 (15:58 +0530)]
[subplugins] Update subplugins accl support
Update subplugin's accelerators support as per updated rules for auto/default
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Geunsik Lim [Fri, 15 May 2020 05:00:50 +0000 (14:00 +0900)]
[Tensor Filter] Added a sub-plugin filter for VIM3/Vivante NPU
This commit is to append the Vivante sub-plugin based on the tensor filter
of NNStreamer to support the Vivante NPU driver.
**Changelog**
* Version 10:
* Added license clause
* Improved statements of the nnsreamer.spec file
* Version 1-9:
* Fixed memory leak (e.g., g_free)
* FIxed security (e.g., nnstreamer.nanifest) for Tizen packaging
* doc: Added how to build and run the Vivante filter on Ubuntu 18.04 (ARM64)
* Updated the invoke function to support various models
(e.g., Inception, Yolo, Mobile SSD)
* Added Tizen package script for Tizen platform
* Fixed incorrect var type (e.g., unsigned int for num_tensors)
* Added annotations for code maintenance
* Added run.sh for Ubuntu18.04 aarch64 + VIM3 board
* Added convert function for vivante tensor data type
* Added vsi_nn_GetTensor of OpenVX apis to obtain tensor data (input, output)
* Added info[i].name with dummy data (e.g., tensor id)
* Removed deprecated variables and functions
* Updated the vivante filter with "multi-files" structure for re-usability
- "model=/opt/vivante/model/inception_v3.nb,
/opt/vivante/model/libinception_v3.so
* Updated the existing vivante tensor filter based on dlopen() scheme
* Added a functions to check the dlopen operation to avoid undefined symbol issue
* Updated annotaiton (e.g., image size is 299:299)
* Added Vivante functions (common APIs) in the vivante_open function
* Added Vivante functions (common APIs) in the vivante_invoke function
* Added Vivante functions (common APIs) in the vivante_close function
* Update annotation to avoid misunderstanding
* Fixed typos (GstTensorsInfo)
* Add todo list (e.g., glib's time functions)
* Added vsi_nn_pub.h header file to use libraries of acuity-ovxlib-dev tool
* Added a skeleton of the Vivate tensor filter
* Imported the major implementation of the Vivante (inceptionv3 model)
for Nnstreamer pipeline. libvivante_v3.so will be used by "model=***"
interface to support a generic approach for re-usability.
* Added vivante_inception_v3.h header file for .so file of a NN model
* Added a data structure (e.g., vivante_data) to manage Vivante data
Signed-off-by: Geunsik Lim <geunsik.lim@samsung.com>
Jaeyun [Thu, 21 May 2020 05:32:20 +0000 (14:32 +0900)]
[Android] fix warning
fix build warning, remove unnecessary code.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Mon, 25 May 2020 06:02:44 +0000 (15:02 +0900)]
[Merge] reduce complexity
code clean in tensor-merge element, reduce code complexity.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Wed, 27 May 2020 10:08:37 +0000 (19:08 +0900)]
[Android/Test] sample for media format
test/sample to input media data format into appsrc.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Mon, 25 May 2020 10:53:44 +0000 (19:53 +0900)]
[Android/Doc] link proper method
link proper method, for the interfaces defined in Pipeline/CustomFilter.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Sangjung Woo [Fri, 22 May 2020 08:23:10 +0000 (17:23 +0900)]
[Spec] Fix the wrong meson option for armnn
Because of the typo for meson option, below warning occurs when
building.
* WARNING: Unknown options: "armnn_support"
This patch fixes this warning.
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
Jaeyun [Thu, 21 May 2020 05:34:51 +0000 (14:34 +0900)]
[Test] use test-env
update unittest arg, use test-env in cpp-method test.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
gichan-jang [Thu, 21 May 2020 02:23:32 +0000 (11:23 +0900)]
[CONF] Fix nnstreamer conf load function
Since key_file is declared using g_autoptr macro, it does not need to cleaning up variables.
It is related issue #2393.
Signed-off-by: gichan-jang <gichan2.jang@samsung.com>
Jaeyun [Tue, 19 May 2020 09:27:57 +0000 (18:27 +0900)]
[Android] enable custom filter in singleshot
register filter subplugins (custom-filter support) when building single-shot library
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Parichay Kapoor [Tue, 12 May 2020 21:18:20 +0000 (02:48 +0530)]
[filter] Handle auto/default semantics for tensor_filter v1
Patch to handle auto/default semantics for tensor_filter v1
getFrameworkInfo adds two more parameters - accl_default and accl_auto with the same characteristics as v0
Passed user input is parsed and if auto/default is used, corresponding accelerators are set and used.
This resolves #2320 for tensor_filter v1
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Parichay Kapoor [Tue, 12 May 2020 20:09:19 +0000 (01:39 +0530)]
[filter] Handle auto/default semantics for tensor_filter v0
Patch to handle auto/default semantics for tensor_filter
Each tensor_filter should now pass its default/auto accelerator
If not passed, the value defaults to first entry from supported list of accelerators
Passed user input is parsed and if auto/default is used, corresponding accelerators
are set and used.
This resolves #2320 for v0
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Jaeyun [Tue, 19 May 2020 09:00:00 +0000 (18:00 +0900)]
[Android] fix lint warning
fix lint report : set default locale
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Wook Song [Wed, 22 Apr 2020 05:38:40 +0000 (14:38 +0900)]
[Meson] Provide an environment object to test()
In order to make test() in the meson.build script work, this patch
defines an environment object for the unit test cases and provides it to
those test() functions as an argument.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Sangjung Woo [Wed, 13 May 2020 05:26:44 +0000 (14:26 +0900)]
[Android] Update document for Android Library
This patch newly updates the guide for Android Library.
* Add installration guide for required package
* Add proxy setting guide simply without specific IP and port info
* Fix the typoes
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
Geunsik Lim [Thu, 7 May 2020 10:52:59 +0000 (19:52 +0900)]
[Android] Updated document for nnstreamer multi-model app
Fixed issue nnstreamer/nnstreamer#2349.
This commit is to update the existing document in order to re-upload
the nnstreamer multi-model application because the application management
of Google Store is changed.
Signed-off-by: Geunsik Lim <geunsik.lim@samsung.com>