MyungJoo Ham [Wed, 17 Nov 2021 08:58:02 +0000 (17:58 +0900)]
C/common: update error reporting macros
1. Error continue-report only if the op has returned a non-zero value.
2. Commonize formatting macro.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Sangjung Woo [Thu, 25 Nov 2021 11:43:09 +0000 (20:43 +0900)]
[C-Api] Fix the omitted Requires section in capi-ml-inference.pc
Because of the bug in the meson file, Requires section in
capi-ml-inference.pc is empty. It causes the compile-time error when
using ML APIs. This patch fixes this bug.
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
Yongjoo Ahn [Mon, 22 Nov 2021 06:31:04 +0000 (15:31 +0900)]
[fix/trivial] Fix unnecessary null check
- Remove unnecessary null check
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Yongjoo Ahn [Mon, 22 Nov 2021 06:17:07 +0000 (15:17 +0900)]
[fix/trivial] Fix unreacable code
- Remove unreachable code
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
MyungJoo Ham [Wed, 3 Nov 2021 03:36:36 +0000 (12:36 +0900)]
C/Service: introduction of the new API set.
Declaration of what the "ML Service APIs" will provide.
This is the start of designing https://github.com/nnstreamer/nnstreamer/issues/3473
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
MyungJoo Ham [Tue, 26 Oct 2021 04:47:19 +0000 (13:47 +0900)]
C-API/Common: more elaborated error reporting.
As an example for other API implementations,
error messages of ml-api-common.c are elaborated
with the new error reporting infra.
Unit test cases will follow later.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
MyungJoo Ham [Tue, 26 Oct 2021 04:46:41 +0000 (13:46 +0900)]
C-API/Common: macro for formatting error messages.
New macros for error reporting infrastructure with some more info.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Yongjoo Ahn [Tue, 16 Nov 2021 00:54:01 +0000 (09:54 +0900)]
[android/test] Add pipeline api test for mqtt elements
- Add a test for mqtt elements. Because it needs running mqtt broker,
it should be ignored at this moment.
- At least it is passed in configured environment (mqtt broker is
installed in the android device and running locally)
- Add permisiion for internet in AndroidManifest.xml
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Yongjoo Ahn [Mon, 15 Nov 2021 09:55:23 +0000 (18:55 +0900)]
[android/mqtt] Add MQTT support
- Add MQTT support
- It requires paho.mqtt.c prebuilt lib
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Jaeyun [Tue, 16 Nov 2021 08:06:28 +0000 (17:06 +0900)]
[Build/CodeClean] std headers
Code clean, it is recommended to not use stddef for NDK build, use stdlib instead.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Sangjung Woo [Tue, 16 Nov 2021 05:56:02 +0000 (14:56 +0900)]
[Spec] Fix the typo in Requires section
This patch fixes the typo in Requires section.
* capi-machine-larning-common -> capi-machine-learning-common
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
Jihoon Lee [Mon, 15 Nov 2021 08:03:04 +0000 (17:03 +0900)]
[Singleshot] Enhance dim inference for nntrainer
There was a voc that `ml_single_open(&single, 'model.ini', NULL, NULL,
ML_NNFW_TYPE_NNTR_INF, ML_NNFW_HW_ANY);` should work, which is giving
NULL for the tensor_input info and tensor output info.
This was not feasible because the nntrainer inference filter is not
defining `getInputDim()`, `getOutputDim()` on purpose to make it able to
infer output dimension from input dimension in pipeline scenario.
This patch introduce the dimension inferencing mechanism for nntrainer
to make it coherent with the pipeline api.
Below rules apply when input_info or output_info is nullptr for
nntrainer.
1. After opening the framework, allow calling set_input_dim with input_info with
count==0 in case of nntrainer.
2. nntrainer checks if input_info->count==0, if true, nntrainer assums
it should be default dimension and set input/output dimension accordingly.
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: Jihoon Lee <jhoon.it.lee@samsung.com>
MyungJoo Ham [Tue, 26 Oct 2021 04:16:55 +0000 (13:16 +0900)]
C-API/Common: add _ prefix for internal functions.
Add _ prefix for all internal functions
so that developers can distinguish external (API) functions
and internal functions easily.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
MyungJoo Ham [Tue, 26 Oct 2021 01:49:56 +0000 (10:49 +0900)]
C-API/Common: add _ prefix to internal interface - log
Rename mlapi_log* --> _ml_log*
This is the first of adding _ prefix to internal interfaces series.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
MyungJoo Ham [Mon, 25 Oct 2021 10:52:46 +0000 (19:52 +0900)]
C-API/Common: error reporting API implementation.
API implementation:
ml_error(), ml_strerror()
Internal interface for API implementation:
_ml_error_report()
TODO: apply _ prefix for all internal interfaces.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
MyungJoo Ham [Thu, 21 Oct 2021 08:21:03 +0000 (17:21 +0900)]
C-API/Common: error reporting function skeleton.
dlerror()-like function that can provide error descriptions and
strerror()-like function that provide general descriptions for error codes.
This is to review its format and behavior.
The contents and internal APIs for API implementation will
follow after this API is approved.
This addresses (starts addressing) https://github.com/nnstreamer/nnstreamer/issues/3528
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
MyungJoo Ham [Tue, 19 Oct 2021 09:54:21 +0000 (18:54 +0900)]
[C/Inference] Refactoring C-API module relations.
ML C-API interfaces and implementations are being refactored for better relations.
ml-common will be no longer depending on nnstreamer.
Then, in the future, nntrainer will be no longer depending on nnstreamer except for its nnstreamer-nntrainer subplugin.
1. Renamed headers and sources (except for Tizen Studio's files, which we need to tend later)
2. Refactored inter-file dependency and relations.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Jihoon Lee [Wed, 20 Oct 2021 04:59:31 +0000 (13:59 +0900)]
[Headers] Fix missing include in ml-api-common
As tensor data structure now being moved to ml-api-common.
`size_t` is now used in `int ml_tensors_info_get_tensor_size`.
Does including <stddef.h> here.
Please refer to below error in http://nnstreamer.mooo.com/nntrainer/ci/repo-workers/pr-checker/1648-
202110201315200.
81315588951111-
c03945380718edd44c6a19b56056dcf80a64636d/report/build_log_1648_android_error.txt
```
error: unknown type name 'size_t'
int ml_tensors_info_get_tensor_size (ml_tensors_info_h info, int index, size_t *data_size);
```
Signed-off-by: Jihoon Lee <jhoon.it.lee@samsung.com>
MyungJoo Ham [Mon, 18 Oct 2021 10:03:40 +0000 (19:03 +0900)]
[Common] refactor headers to clean up dependency
Data handling APIs are used by ML.training as well as ML.inference.
Move such APIs to common header.
TODO: Move implementation to another .c file so that
we can refactor subpackage dependencies, too.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Gichan Jang [Fri, 24 Sep 2021 06:55:51 +0000 (15:55 +0900)]
Release of ML-API 1.8.0
- Release of ML-API 1.8.0.
Signed-off-by: Gichan Jang <gichan2.jang@samsung.com>
Gichan Jang [Fri, 24 Sep 2021 06:39:08 +0000 (15:39 +0900)]
[Single] Don't close single handle during invoke process
Don't close single handle during invoke process.
If single handle is released during invoke prcoess, an interpreter may
access invalid memory.
Signed-off-by: Gichan Jang <gichan2.jang@samsung.com>
Jihoon Lee [Thu, 16 Sep 2021 10:22:15 +0000 (19:22 +0900)]
[Coverage] Set coverage badge
This patch enables coverage badge. Behind the scene, daily coverage
check is enabled :)
**Self evaluation:**
1. Build test: [ ]Passed [ ]Failed [X]Skipped
2. Run test: [ ]Passed [ ]Failed [X]Skipped
Signed-off-by: Jihoon Lee <jhoon.it.lee@samsung.com>
gichan [Wed, 1 Sep 2021 10:05:50 +0000 (19:05 +0900)]
[TEST] Add checking nnfw availability util
Add checking nnfw availability util.
Signed-off-by: gichan <gichan2.jang@samsung.com>
MyungJoo Ham [Fri, 6 Aug 2021 07:19:08 +0000 (16:19 +0900)]
Unittest: data type consistency with nnstreamer/gstreamer
There are type re-definitions in C ML-API, which
should inherit values from nnstreamer and gstreamer.
Add unittests that ensures the consistency.
Fixes https://github.com/nnstreamer/nnstreamer/issues/2587
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Chunseok Lee [Thu, 26 Aug 2021 08:06:25 +0000 (17:06 +0900)]
nnfw_ver to 1.17.0
nnfw(ONE) v1.17.0 is released. This will bump up nnfw version to 1.17.0
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
Jaeyun [Fri, 20 Aug 2021 06:35:39 +0000 (15:35 +0900)]
[C-Api] function to check hw availability
Add new function to check hw availability with custom option string.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Wed, 18 Aug 2021 10:04:21 +0000 (19:04 +0900)]
[Android/Build] remove release flag
Remove release flag and related source.
Uploading library to jcenter is not available anymore.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
gichan [Tue, 17 Aug 2021 01:45:07 +0000 (10:45 +0900)]
[C-API] Change src-callback name
Change src-callback function name.
- Names of callback function types should have the '_cb' suffix.
- A modile providing callbacks called on events should support exactly one of two registration schemes.
> single callback registration: {module}_set_{details}_cb
Signed-off-by: gichan <gichan2.jang@samsung.com>
Jaeyun [Fri, 13 Aug 2021 10:12:05 +0000 (19:12 +0900)]
[Android/Build] dependency to glib
clearly denote dependency to glib in android build.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Sangjung Woo [Fri, 6 Aug 2021 06:51:40 +0000 (15:51 +0900)]
[C-API] Use ML_NNFW_TYPE_TRIX_ENGINE instead of ML_NNFW_TYPE_TRIx_ENGINE
According to the Tizen naming convention, the names of the entities must
be composed of upper case letters. This patch fixes this issue.
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
Jaeyun [Wed, 4 Aug 2021 10:05:07 +0000 (19:05 +0900)]
[Test/C-Api] src-callback test
Add new testcases for src callback.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Sangjung Woo [Wed, 4 Aug 2021 05:52:41 +0000 (14:52 +0900)]
[C-API] Rename srnpu to trix-engine
The patch updates the name 'srnpu' to 'trix-engine'.
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
Jaeyun [Mon, 5 Jul 2021 10:11:24 +0000 (19:11 +0900)]
[Test] update flex-tensor caps
Update caps for flexible tensor, old mime type for flex-tensor is removed.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Tue, 3 Aug 2021 03:26:46 +0000 (12:26 +0900)]
[C-Api] implement src callback
Implement new API - appsrc callbacks.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Yongjoo Ahn [Mon, 2 Aug 2021 11:11:49 +0000 (20:11 +0900)]
[Android] Test SNAP with tensorflow lite model
- SNAP v3.0 supports tensorflow lite model.
- Add pipeline and singleshot test using it.
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Jaeyun [Mon, 19 Jul 2021 06:51:21 +0000 (15:51 +0900)]
[C-Api] framework priority from conf
Load framework priority from nnstreamer conf file.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Mon, 19 Jul 2021 05:05:49 +0000 (14:05 +0900)]
[C-Api] remove warning while checking framework
When checking fw with model files is failed, it returns null and this will print warning message even if this works properly.
Update macro to not print this message. (requested from VD)
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Gichan Jang [Tue, 13 Jul 2021 05:49:10 +0000 (14:49 +0900)]
[TEST] Add tensor clone test
- Add tensor clone test
Signed-off-by: Gichan Jang <gichan2.jang@samsung.com>
Gichan Jang [Wed, 30 Jun 2021 05:46:29 +0000 (14:46 +0900)]
[debian] Add jni debian package
Add jni wrapper debian package.
Signed-off-by: Gichan Jang <gichan2.jang@samsung.com>
Jaeyun [Wed, 30 Jun 2021 07:58:19 +0000 (16:58 +0900)]
[JNI] remove unnecessary feature in native
We should call subplugin registration only on Android.
Remove unnecessary feature in native code.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Sangjung Woo [Tue, 29 Jun 2021 07:34:40 +0000 (16:34 +0900)]
[C-Api/Util] Add ML_NNFW_HW_NPU_SLSI case to converting function
To support VD product, this patch adds the ML_NNFW_HW_NPU_SLSI case
to ml_nnfw_to_accl_hw() function.
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
Jaeyun [Thu, 24 Jun 2021 05:44:49 +0000 (14:44 +0900)]
[Build/Java] add script to build jar
Add new script to build java library.
To prevent build error, mark prefix in java files and change it while building the library.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Tue, 22 Jun 2021 10:42:07 +0000 (19:42 +0900)]
[Java] prepare ubuntu build
Prepare Java API on ubuntu, add feature for android and remove android annotation.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Fri, 23 Apr 2021 11:17:35 +0000 (20:17 +0900)]
[C-Api] handle flexible tensor
In flexible tensor stream, sink and src handle cannot set exact tensor info from pad caps.
To handle flex-tensor, ignore tensor info while configuring the element handle.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Wed, 23 Jun 2021 07:00:17 +0000 (16:00 +0900)]
[Android/Build] check target abi
Check target abi when build android library, and set dir name to get gstreamer prebuilt libs.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Wed, 23 Jun 2021 06:11:24 +0000 (15:11 +0900)]
[Java] build JNI wrapper
Build JNI wrapper using meson.
The source for JNI wrapper is same with Android native code.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Wed, 9 Jun 2021 05:11:53 +0000 (14:11 +0900)]
[Api/TensorsData] hold tensors info in data handle
Clone tensors-info handle and hold it in tensors-data handle.
Also, pass info handle to parse data object in Android native build.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Fri, 28 May 2021 07:29:40 +0000 (16:29 +0900)]
[C-Api/Util] remove duplicated source
Add private util function and remove duplicated source.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Wed, 23 Jun 2021 05:20:03 +0000 (14:20 +0900)]
[Android] min sdk version to 24
For android release, set minimum sdk version to 24.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Gichan Jang [Wed, 23 Jun 2021 06:02:00 +0000 (15:02 +0900)]
[Data] Unlock data and remove unreachable code
- Unlock data when policy neither ML_PIPELINE_BUF_POLICY_AUTO_FREE nor
ML_PIPELINE_BUF_POLICY_DO_NOT_FREE.
- Remove unreachable code
Signed-off-by: Gichan Jang <gichan2.jang@samsung.com>
Jaeyun [Mon, 21 Jun 2021 10:47:29 +0000 (19:47 +0900)]
[Android] nnfw feature
Change nnfw feature definition in android build, same to c-api define.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Yongjoo Ahn [Fri, 18 Jun 2021 07:59:10 +0000 (16:59 +0900)]
[android/build] Fix android build script for wget
- Fix wget urls to support git lfs files
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Parichay Kapoor [Mon, 14 Jun 2021 07:01:21 +0000 (16:01 +0900)]
[capi] Update on thread safety for ml_tensors_info
This patch fixes the case of Time of Check Time of Use bug in
ml_pipeline_custom_easy_filter_register().
The provided arguments `in` and `out` are first checked for validity
under their locks and then used for clone again under their locks.
However, between the two operations, both the variable are left lock
free, and the two variables can become invalid.
This patch adds the corresponding fix by adding support for
checking validity of ml_tensors_info_s lockfree which is done
under the lock acquired inside ml_tensors_info_clone().
Further, as a minor fix, redundant locking and unlocking
has been removed from ml_pipeline_custom_easy_filter_register().
See also #42 https://github.com/nnstreamer/api/pull/42#issuecomment-
840265534
Resolves https://github.com/nnstreamer/nnstreamer/issues/2600
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Sangjung Woo [Thu, 10 Jun 2021 08:16:07 +0000 (17:16 +0900)]
[C-API] Fix the coverity issue
This patch fixes the coverity issue: "waiting while holding a lock".
Before entering the loop, mutex lock is release so that other thread can
use the CPU.
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
MyungJoo Ham [Tue, 11 May 2021 10:59:50 +0000 (19:59 +0900)]
Thread safety for ml_tensors_data APIs
Fixes #nnstreamer/2600
Note that locally created-and-freed ml-tensors-data
are not required to be protected.
Another fix: do not destroy user data even if it is AUTO_FREE
if push_buffer is NOT called.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Gichan Jang [Wed, 9 Jun 2021 07:32:40 +0000 (16:32 +0900)]
[Flatbuf] Change version check of the flatc
Change version check of the flatc from error to warning.
Even if the flatbuffers compiler has different versions, compatibility can be maintained.
Tested with
- System flatc version: 2.0.0
- Android flatc version: 1.12.0
Signed-off-by: Gichan Jang <gichan2.jang@samsung.com>
Jaeyun [Mon, 7 Jun 2021 06:05:41 +0000 (15:05 +0900)]
[Android] tf-lite custom options
update docs to set custom option for tensorflow-lite model.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Gichan Jang [Mon, 7 Jun 2021 02:07:49 +0000 (11:07 +0900)]
[git] Add gitignore list
Add build directory to git ignore list.
Signed-off-by: Gichan Jang <gichan2.jang@samsung.com>
Jihoon Lee [Fri, 4 Jun 2021 05:39:17 +0000 (14:39 +0900)]
[CAPI/ml-common] Add require to base-common
When a developer tries to add dependency to `capi-ml-common` they have
to care about adding dependency to `capi-base-common` as well because
`capi-ml-common` is actually depending upon `capi-base-common`
This patch resolves the issue by adding require to `capi-base-common`
in the pc file.
resolves #57
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: Jihoon Lee <jhoon.it.lee@samsung.com>
MyungJoo Ham [Fri, 28 May 2021 06:50:25 +0000 (15:50 +0900)]
NNStreamer-devel-internal headers are external.
Headers supplied by nnstreamer-devel-internal are external headers.
Use < > instead of " ".
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
MyungJoo Ham [Fri, 23 Apr 2021 05:49:33 +0000 (14:49 +0900)]
[API/Skeleton] App Src Callback support
Refer: https://github.com/nnstreamer/nnstreamer/issues/2581
This provides API signature and code skeleton for
callback support of app-src elements in ML-API pipelines.
CC: @junmin-kim
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Yongjoo Ahn [Mon, 17 May 2021 08:41:31 +0000 (17:41 +0900)]
[Android] Fix android ml api build script with patches for gstreamer build script
- Let android build script use patch files to modify gstreamer build script
from https://gstreamer.freedesktop.org/data/pkg/android/1.16.2/
- Fix tracing feature to add GstShark
https://github.com/RidgeRun/gst-shark/
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Jaeyun [Mon, 24 May 2021 08:33:16 +0000 (17:33 +0900)]
[clang] add clang format
add clang format in api repository.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Dongju Chae [Mon, 5 Apr 2021 11:24:25 +0000 (20:24 +0900)]
[SRNPU/NNFW] Add NNFW Type for SRNPU
This patch adds NNFW Type for SRNPU.
Signed-off-by: Dongju Chae <dongju.chae@samsung.com>
Sangjung Woo [Tue, 11 May 2021 02:40:06 +0000 (11:40 +0900)]
[Spec] Add the Provides option in case of 64bit
If architecture is aarch64 or x86_64, the 64bit option should be added.
If not, the wrong so file is provided as below.
$ rpm -qp --provides capi-machine-learning-inference-1.7.2-0.aarch64.rpm
capi-machine-learning-inference = 1.7.2-0
capi-machine-learning-inference(aarch-64) = 1.7.2-0
libcapi-nnstreamer.so
libcapi-nnstreamer.so.1()(64bit)
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
Jaeyun [Mon, 10 May 2021 11:17:32 +0000 (20:17 +0900)]
[C-Api] enum for AIFW
add new enum value (for AIFW and SLSI NPU)
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
MyungJoo Ham [Thu, 29 Apr 2021 06:45:21 +0000 (15:45 +0900)]
Thread safety for ml_tensors_info APIs
Fixes #nnstreamer/2600
This supports thread safety for ml_tensors_info related APIs.
TODO: ml_tensors_data related APIs
https://github.com/nnstreamer/nnstreamer/issues/2600
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Yongjoo Ahn [Fri, 9 Apr 2021 02:26:01 +0000 (11:26 +0900)]
[Android/TFLite] Add XNNPACK delegate support
- Add XNNPACK delegate in tensorflow-lite
- tensorflow-lite utilize XNNPACK as their default engine since v2.3.0
REF:
- https://github.com/google/XNNPACK
- https://github.com/tensorflow/tensorflow/tree/master/tensorflow/lite/delegates/xnnpack
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Jaeyun [Thu, 6 May 2021 11:19:40 +0000 (20:19 +0900)]
[Android/Build] use default std
Remove flag to set cpp std in android native build.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Thu, 6 May 2021 10:54:00 +0000 (19:54 +0900)]
[Android] update tf-lite and nnfw version
update default version, tensorflow-lite v2.3.0 and nnfw v1.15.0.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Thu, 6 May 2021 11:22:02 +0000 (20:22 +0900)]
[Android/Test] remove unnecessary constructor
Code clean, remove unnecessary constructor in testcases.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Sangjung Woo [Thu, 6 May 2021 08:27:00 +0000 (17:27 +0900)]
[Spec] Add the provides section
Since rpmbuild explicitly exclude the name from the library rpm,
libcapi-nnstreamer.so is not provided as below.
$ rpm -qp --provides capi-machine-learning-inference-1.7.2-0.armv7l.rpm
capi-machine-learning-inference = 1.7.2-0
capi-machine-learning-inference(armv7l-32) = 1.7.2-0
libcapi-nnstreamer.so.1
This patch adds the provides section to capi-machine-learning-inference
package so that capi-machine-learning-inference provides
libcapi-nnstreamer.so.
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
Yongjoo Ahn [Mon, 3 May 2021 08:42:09 +0000 (17:42 +0900)]
[Android] Build GStreamer with tracing
- Add Android GStreamer build option for tracing in build script
- Add patch for file `gstreamer_android-1.0.c.in` to enable this feature
REF: https://gstreamer.freedesktop.org/documentation/additional/design/tracing.html?gi-language=c
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Yongjoo Ahn [Thu, 6 May 2021 04:20:29 +0000 (13:20 +0900)]
[Test/Flatbuf] Add Pipeline API test for Flatbuf
- Add a test for Flatbuf which check eqaulity of input buffer with output of flatbuf decoder and converter
- input -> decoder -> other/flatbuf-tensor -> converter -> output)
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Yongjoo Ahn [Mon, 3 May 2021 08:32:16 +0000 (17:32 +0900)]
[C-API] Fix Pipekline Flush API
- Some gstreamer elements fail to send `gst_event_new_flush_start`
- Spilt `_flush_start` and `_flush_stop` to call `_flush_stop`
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Jaeyun [Tue, 6 Apr 2021 05:00:52 +0000 (14:00 +0900)]
[Test] testcase for pipeline flush
Add testcase for new API - pipeline flush.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Fri, 2 Apr 2021 11:39:25 +0000 (20:39 +0900)]
[Android] method to flush the pipeline
NNStreamer Android lib, add new method to flush the pipeline.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Fri, 2 Apr 2021 10:31:09 +0000 (19:31 +0900)]
[C-Api] new API to flush the pipeline
Add new API to flush the pipeline.
If new input should be run without old data in pipeline, this is necessary to clear all.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Sangjung Woo [Thu, 29 Apr 2021 05:28:32 +0000 (14:28 +0900)]
[TCM] Rename TCM config file
To run the TCM on AHub infra, the default name of config file should be
'config.yaml'.
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
Jaeyun [Mon, 26 Apr 2021 10:29:09 +0000 (19:29 +0900)]
[Android/Pipeline] clone data from src element
We cannot guarantee the data object from java is available while running the pipeline.
Set clone when pushing data to src element.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Sangjung Woo [Fri, 23 Apr 2021 05:40:17 +0000 (14:40 +0900)]
[Spec] Move libcapi-nnstreamer.so to capi-machine-learning-inference package
To support the backward compatibility of Tizen 6.0,
libcapi-nnstreamer.so is moved to capi-machine-learning-inference
package.
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
Gichan Jang [Tue, 20 Apr 2021 01:26:54 +0000 (10:26 +0900)]
[Indent] Fix indent style
Fix indent style.
The "pointer" qualifier, '*', should be with the variable.
This is commit body message
Signed-off-by: Gichan Jang <gichan2.jang@samsung.com>
Jaeyun [Thu, 8 Apr 2021 05:20:06 +0000 (14:20 +0900)]
[Tizen/Build] unnecessary build require
remove unnecessary feature definition and build requires.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Thu, 8 Apr 2021 09:33:54 +0000 (18:33 +0900)]
[Android/Test] junit ver
Fix avas scan report, set junit ver 4.13.1.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Wed, 7 Apr 2021 10:02:28 +0000 (19:02 +0900)]
[Build/Test] coverage report
update script to generate coverage report.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Gichan Jang [Tue, 6 Apr 2021 08:00:05 +0000 (17:00 +0900)]
[TCM] Add TCM config files
Add TCM config filres for tizen and android.
Signed-off-by: Gichan Jang <gichan2.jang@samsung.com>
Jaeyun [Tue, 6 Apr 2021 06:02:23 +0000 (15:02 +0900)]
[Test] change property name
change property name sync mode, nnstreamer will change sync mode and option.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
MyungJoo Ham [Thu, 25 Feb 2021 01:11:22 +0000 (10:11 +0900)]
[Tizen/Privilege] Use newly exported MMFW headers
This requires
https://review.tizen.org/gerrit/#/c/platform/core/multimedia/libmm-camcorder/+/254182/ (OLD)
https://review.tizen.org/gerrit/#/c/platform/core/multimedia/libmm-camcorder/+/255956/ (NEW)
to be merged and deployed. Or maintainers may do group-SR.
Fixes https://github.com/nnstreamer/nnstreamer/issues/1734
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Gichan Jang [Tue, 30 Mar 2021 04:32:55 +0000 (13:32 +0900)]
[Flatbuffers] Change subplugins name
Change flatbuffers and flexbuffers subplugins name.
Signed-off-by: Gichan Jang <gichan2.jang@samsung.com>
Sangjung Woo [Fri, 26 Mar 2021 04:19:41 +0000 (13:19 +0900)]
[Spec] Use tar archiving instead of tar.gz
This patch uses tar archiving instead of tar.gz instead of VD QBuild
system.
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
Jaeyun [Thu, 25 Mar 2021 10:36:31 +0000 (19:36 +0900)]
[Android] sub-plugins for flatbuf/flexbuf
Update script to build Android lib with flatbuf and flexbuf subplugins.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
MyungJoo Ham [Mon, 29 Mar 2021 10:53:38 +0000 (19:53 +0900)]
Fix unit test script
In main routine, return cannot be used.
Return values by calling exit instead.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Jaeyun [Mon, 29 Mar 2021 04:35:32 +0000 (13:35 +0900)]
[Etc] add reviewers
add reviewers for API repository.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Hyoung Joo Ahn [Wed, 24 Mar 2021 11:29:02 +0000 (20:29 +0900)]
[Log] divide a log message
Since the length of the pipeline string is too long, the error message cannot be checked.
For easier debugging, this log message should be divided.
Signed-off-by: Hyoung Joo Ahn <hello.ahn@samsung.com>
Jaeyun [Mon, 22 Mar 2021 03:47:09 +0000 (12:47 +0900)]
[Build] version info
Add version information for Android ML API.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Yongjoo Ahn [Mon, 8 Mar 2021 05:20:15 +0000 (14:20 +0900)]
[Android] Add android build script for nnstreamer
- Add `java/android/build-nnstreamer-android-lib.sh`
- Trim Android.mk and Android-nnstreamer.mk
- Set ML_API_VERSION
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Jihoon Lee [Wed, 10 Mar 2021 12:20:22 +0000 (21:20 +0900)]
[Singleshot] Support nntrainer inference
This patch enables singleshot inference from `ML_NNFW_TYPE_NNTR_INF`
I read code throughly to look where to change but it was only one line,
Please review if there are more things to be done.
Signed-off-by: Jihoon Lee <jhoon.it.lee@samsung.com>
Jaeyun [Thu, 4 Mar 2021 04:13:52 +0000 (13:13 +0900)]
[Android] upload nnstreamer code
Upload nnstreamer android files.
TODO: setup android build in ci and add build script.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Fri, 5 Mar 2021 06:33:18 +0000 (15:33 +0900)]
[C-Api] add ref-count to custom filter/if handle
When developer constructs a pipeline with custom filter or if-condition, add ref count to custom handle.
Prevent custom handle unregistration if pipeline is running.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Mon, 8 Mar 2021 06:55:21 +0000 (15:55 +0900)]
[Test] tensor name in tf-lite tc
Ignore tensor name comparison in tf-lite testcase.
NNStreamer tf-lite plugin will set tensor name.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>