남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 29 Apr 2019 22:31:23 +0000 (07:31 +0900)]
[moco/ONNX] Introduce Identity testcase for moco ONNX frontend (#3380)
This commit will introduce a new testcase for moco ONNX frontend,
which is consisted of Identity operation.
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 29 Apr 2019 22:28:36 +0000 (07:28 +0900)]
[locomotiv] Implement 'annot_data', wrapper for 'annot' (#3385)
* [locomotiv] Implement 'annot_data', wrapper for 'annot'
This commit makes implementation of 'annot_data', wrapper function for
annotating NodeDataImpl to loco Node object. It also adds relative
comments and test. 'annot_data' is for project-wide internal use only.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
* Use rvalue as argument for annot_data
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 29 Apr 2019 03:34:07 +0000 (12:34 +0900)]
[locomotiv] Reset unique pointer at NodeDataImpl ctor (#3382)
This commit use reset() of unique pointer for simple and efficient code
at NodeDataImpl class constructor.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 29 Apr 2019 03:31:05 +0000 (12:31 +0900)]
[locomotiv] Use explicit float as constant (#3381)
This commit use explicit float constant for float NodeData test,
instead of double.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 29 Apr 2019 02:22:13 +0000 (11:22 +0900)]
[locomotiv] Internal implementation of 'NodeData' (#3374)
Interal 'NodeDataImpl' implements public API of 'NodeData'. This commit
also introduces test framework and test for NodeData.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
윤지영/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 29 Apr 2019 02:07:21 +0000 (11:07 +0900)]
[nnkit] Add Status class for onnx backend (#3369)
* [nnkit] Add Status class for onnx backend
This class is for auto-release OrtStatus.
Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
* Return the lhs by reference in assignment operator overloading function
* Release status before assignment
* Change function name and add description
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 29 Apr 2019 02:06:41 +0000 (11:06 +0900)]
[moco/ONNX] Convert onnx graph to loco graph (#3368)
* [moco/ONNX] Convert onnx graph to loco graph
This commit will enable converting ONNX graph to loco graph
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
* fix type name
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 29 Apr 2019 01:49:03 +0000 (10:49 +0900)]
[loco] Visit each node only once (#3378)
The current implementation of postorder_traversal may visits the same
node multiple times.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 25 Apr 2019 23:42:02 +0000 (08:42 +0900)]
[nnkit/moco/tf] skeleton of backend for nnkit moco tf (#3371)
This commits adds early version of backend for nnkit moco tf and build script.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
Ivan Vagin/AI Tools Lab /SRR/Engineer/삼성전자 [Thu, 25 Apr 2019 18:05:57 +0000 (21:05 +0300)]
[nnc] Introduced 'FuseArithmeticOps' graph optimization (#3340)
Main purpose of this optimization - is to fuse 'Conv->BatchNorm' into 'Conv'
Currently 'BatchNorm' splitting by NNC frontends into 'Scale->Scale->BiasAdd'
This optimization performing in two steps (repeated while graph changing):
1. Merging two successive operations with constant weights into one (ex: 'Scale->Scale' becomes 'Scale')
2. Sinking 'BiasAdd' through 'Scale' (so 'Conv->BiasAdd->Scale' becomes 'Conv->Scale->BiasAdd')
This optimization increases performance by ~7.5% on caffe inception_v3 (which has ~90 Conv->BatchNorm edges)
Signed-off-by: Ivan Vagin <ivan.vagin@samsung.com>
채성우/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 25 Apr 2019 06:58:34 +0000 (15:58 +0900)]
[CMake] Fix typo in TensorFlowConfig.cmake (#3367)
This commit fix typo in TensorFlowConfig.cmake
Signed-off-by: seongwoo <sw4670.chae@samsung.com>
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 25 Apr 2019 06:49:51 +0000 (15:49 +0900)]
[moco/ONNX] Load onnx model in moco (#3366)
This commit will enable loading ONNX model file in moco onnx frontend
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 25 Apr 2019 05:00:14 +0000 (14:00 +0900)]
[nnkit] Moving TensorInfoParser & ParsedTensor into util dir (#3356)
* [nnkit] Moving TensorInfoParser & ParsedTensor into util dir
This commit moves TensorInfoParser & ParsedTensor into util dir. These two files are used by `nnkit tf` and also `nnkit moco tf`. So it would be better to move these into common directory for better reuse.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* modify dir name from `util` to `tftestinfo`
* util -> tftestinfo
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 25 Apr 2019 04:28:14 +0000 (13:28 +0900)]
[moco/ONNX] Introduce Identity operation (#3365)
This commit will introduce Identity operation and add it to GraphBuilderRegisry
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 25 Apr 2019 04:17:33 +0000 (13:17 +0900)]
[locomotiv] Set graph input data by NodeData (#3363)
This commit changes how Session gets its graph input. Using NodeData
instead of Buffer is for matching input and output type, and to avoid
use of template on header.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 25 Apr 2019 04:01:21 +0000 (13:01 +0900)]
[moco] remove TF frontend op headers (#3364)
This will remove op headers as not needed anymore
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 25 Apr 2019 03:32:06 +0000 (12:32 +0900)]
[moco] use const string for map key (#3362)
This will add const keyword to string for op map key
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 25 Apr 2019 02:36:52 +0000 (11:36 +0900)]
[locomotiv] Introduce NodeData class (#3349)
NodeData class is responsible to data type and data value of node. It
serves as input and output interface for 'Session' class. Its
implementation would be hidden from outside.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 25 Apr 2019 02:05:21 +0000 (11:05 +0900)]
[moco] Support operator self register for TF frontent (#3359)
* [moco] Support operator self register for TF frontent
This will provide self registeration for operator builder of TensorFlow frontend
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* use const
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 25 Apr 2019 01:25:33 +0000 (10:25 +0900)]
[loco] Add "postorder_traversal" helper (#3354)
This commit adds "postorder_traversal" helper which generates a
postorder traversal sequence from a given "roots".
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 24 Apr 2019 22:22:23 +0000 (07:22 +0900)]
[moco/ONNX] Introduce symbolTable in GraphBuilderContext (#3353)
This commit will introduce symbolTable in GraphBuilderContext
symbolTable is needed for linking each nodes
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
Андрей Шедько/AI Tools Lab /SRR/Engineer/삼성전자 [Wed, 24 Apr 2019 17:55:25 +0000 (20:55 +0300)]
[mir] Introduce copy with changed input method to Operation (#3346)
Introduce "copy with changed input" method to Operation
to facilitate easier creation of optimizations
Signed-off-by: Andrei Shedko <a.shedko@samsung.com>
[nnc] Simplify implementation of Softmax in interpreter (#3347)
Do not use `Fill` and `Reduce` in Softmax implementation in interpreter.
Fix coding style.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 24 Apr 2019 07:02:25 +0000 (16:02 +0900)]
[enco] Skip tflite model generation if possible (#3351)
* [enco] Skip tflite model generation if possible
This will update TF lite test framework to skip code generation if there is no effective change
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* remove target depend
Андрей Шедько/AI Tools Lab /SRR/Engineer/삼성전자 [Tue, 23 Apr 2019 12:39:00 +0000 (15:39 +0300)]
[nnc] Dump Graph Impl (#3001)
Add Option to Dump graph after pass
Dumping now creates `dot` files with meaningful names:
`0_importer.dot`, `1_opt_combine_transposes.dot`
Since backend do not return a dumpable graph, we do not dump them
Signed-off-by: Andrei Shedko <a.shedko@samsung.com>
[nnc] Extract Elementwise interpreter logic into a separate file (#3339)
Extract handling of `ElementwiseOp` in interpreter into a separate file.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 23 Apr 2019 04:53:12 +0000 (13:53 +0900)]
[moco] update test comment for shape (#3345)
This will update the comment of shape related test with output node
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 23 Apr 2019 03:39:39 +0000 (12:39 +0900)]
[moco] add test of graph from load (#3343)
This will add test with graph from result of load
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 23 Apr 2019 01:46:53 +0000 (10:46 +0900)]
[moco] add include file (#3342)
This will add include file that is used in the source file
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 23 Apr 2019 01:44:08 +0000 (10:44 +0900)]
[enco] skip enco-basic test code gen if possible (#3344)
This will make skip enco-basic test code generation if there is no effective change
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 22 Apr 2019 23:51:28 +0000 (08:51 +0900)]
[pp] README.md (#3315)
* [pp] Readme.md
This adds Readme.md file in contrib/pp.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* added code examples
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 22 Apr 2019 23:51:01 +0000 (08:51 +0900)]
[ann-ref] Add README.md (#3279)
Let's introduce README.md with a brief introduction.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 22 Apr 2019 07:36:24 +0000 (16:36 +0900)]
Remove add_nncc_library function (#3328)
This commit removes deprecated add_nncc_library function from the
top-level CMakeLists.txt.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 22 Apr 2019 07:23:34 +0000 (16:23 +0900)]
[moco] change to use string and add header (#3335)
This will change message of runtime_error() to use string concatenation
Additionally add header for using assert()
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 22 Apr 2019 05:37:00 +0000 (14:37 +0900)]
[locomotiv] Introduce Session class API (#3326)
This commit defines basic interpreter API using Session class. Session
class is responsible to loco graph inference. NodeData class is to store
node information that would be passed as output.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 22 Apr 2019 04:30:27 +0000 (13:30 +0900)]
[moco] provide istream as input of conversion (#3334)
This will add another load() to parse from istream and a test code
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 22 Apr 2019 03:26:48 +0000 (12:26 +0900)]
[nnkit] Revision of README.md (#3287)
* Draft: README.md for nnkit
This is a draft for README.md for nnkit.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* more content into design
* remove '\'
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 22 Apr 2019 02:15:09 +0000 (11:15 +0900)]
[moco/ONNX] Introduce GraphBuilder and GraphBuilderRegistry (#3322)
This commit will introduce GraphBuilder for building graph
and GraphBuilderRegistry for registering each operation's graph building
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 22 Apr 2019 01:49:24 +0000 (10:49 +0900)]
[moco] register op and do convert (#3329)
This will register two operators conversion and enable conversion
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 22 Apr 2019 00:50:12 +0000 (09:50 +0900)]
[moco] convert two nodes (#3283)
This will fill two Placeholder and Identity node converters
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Sun, 21 Apr 2019 23:26:25 +0000 (08:26 +0900)]
[encodump] Introduce README.md (#3325)
This commit will introduction, howto and examples of encodump
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Sun, 21 Apr 2019 23:26:13 +0000 (08:26 +0900)]
Update TensorFlow Lite 1.7 Package (#3327)
Now, it is possible to build TensorFlow Lite 1.7 and 1.12 at the same
time (if necessary).
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Sun, 21 Apr 2019 23:26:02 +0000 (08:26 +0900)]
[cwrap] Update README.md (#3324)
This commit will update README.md
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 19 Apr 2019 06:49:22 +0000 (15:49 +0900)]
[mir] Remove add_nncc_library call (#3294)
This commit removes add_nncc_library call from mir build script.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 19 Apr 2019 06:49:12 +0000 (15:49 +0900)]
[nnc] Remove add_nncc_library call (#3295)
This commit removes add_nncc_library call from nnc build script.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 19 Apr 2019 05:49:24 +0000 (14:49 +0900)]
[locomotiv] Rename loco interpreter project (#3323)
locomotiv, loco interperter outside of loco itself
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 19 Apr 2019 03:16:20 +0000 (12:16 +0900)]
[onnxkit] Update README.md (#3321)
This commit will update README.md for onnxkit
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 19 Apr 2019 02:15:39 +0000 (11:15 +0900)]
[moco/ONNX] Introduce ONNX GraphBuilderContext (#3300)
This commit will introduce GraphBuilderContext for ONNX
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 19 Apr 2019 01:35:14 +0000 (10:35 +0900)]
[enco] Skip Caffe codegen if possible (#3320)
This commit updates enco Caffe test framework to skip codegen if there
is no effective change.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 19 Apr 2019 01:00:05 +0000 (10:00 +0900)]
[enco] Skip caffemodel generation if possible (#3312)
With this commit, enco caffe test framework skips caffemodel generation
if there is no change on the input prototxt itself.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 19 Apr 2019 00:39:32 +0000 (09:39 +0900)]
[coco] Update README.md (#3317)
README.md states that coco is a compiler, but it does not hold. coco is
just an intermediate representation for compilers.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 19 Apr 2019 00:39:21 +0000 (09:39 +0900)]
Copy comment related with FlatBuffer version (#3319)
This commit moves the comment related with FlatBuffers vesion from
FlatBuffersSource package into TensorFlow Lite package as the first
step to remove non-versioned FlatBuffersSource package.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 19 Apr 2019 00:32:09 +0000 (09:32 +0900)]
Introduce FlatBuffersSource v1.8 package (#3318)
This commit introduces FlatBuffersSource v1.8 package.
NOTE TensorFlow Lite 1.7 depends on this version.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 19 Apr 2019 00:00:02 +0000 (09:00 +0900)]
[nnkit/TF] tensor name and shape validation (#3288)
* [nnkit/TF] tensor name and shape validation
This code adds validation code for tensor name and shape.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* more comment about the structure of tensor name
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 18 Apr 2019 09:10:44 +0000 (18:10 +0900)]
[angkor] Remove Tensor IndexRange (#3314)
This commit removes deprecated code related with Tensor IndexRange
iteration.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 18 Apr 2019 08:00:22 +0000 (17:00 +0900)]
[cli] README.md (#3313)
This is a readme file for `contrib/cli`.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 18 Apr 2019 07:46:42 +0000 (16:46 +0900)]
[nnkit] Use IndexEnumerator instead of IndexRange (#3310)
This commit replaces all the occurrences of IndexRange (range function)
in nnkit code with IndexEnumerator.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 18 Apr 2019 07:46:00 +0000 (16:46 +0900)]
[nnkit/TF] TF backend working with zero input (#3303)
This commit adds TF backend working with zero input.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 18 Apr 2019 07:32:39 +0000 (16:32 +0900)]
[stdex] README.md file (#3298)
* [stdex] README.md file
This commids adds README.md for stdex.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* revision
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 18 Apr 2019 07:21:05 +0000 (16:21 +0900)]
[loco] Introduce NodeAnnotation (#3296)
This commit introduces NodeAnnotation class and attach these annotations
into Node (via inheriting AnnotatedItem).
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 18 Apr 2019 07:13:37 +0000 (16:13 +0900)]
[loco-interp] Initial commit (#3306)
This would introduce loco interpreter empty project
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 18 Apr 2019 07:13:17 +0000 (16:13 +0900)]
[moco/test/tf] Testcase where there is no input (#3305)
This is to check if moco works with a model that has no input.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 18 Apr 2019 07:12:45 +0000 (16:12 +0900)]
Use FlatBufferSource with Explicit Version (#3302)
This commit updates FlatBuffers package to mention 1.10 explicitly.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 18 Apr 2019 07:10:42 +0000 (16:10 +0900)]
[enco] Do NOT copy test.prototxt if there is no change (#3301)
With this commit, enco's caffe test frmework will not copy test.prototxt
if there is no change.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 18 Apr 2019 05:39:19 +0000 (14:39 +0900)]
Build TensorFlow Lite with FlatBuffer source (#3299)
It turns out that TensorFlow Lite does not require FlatBuffer library.
Rather TensorFlow Lite needs only FlatBuffer header.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 18 Apr 2019 03:59:38 +0000 (12:59 +0900)]
Introduce FlatBuffersSource (v1.10) package (#3286)
This commit allows users to explicitly access FlatBuffers (v1.10) source
code via nncc_find_package.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 18 Apr 2019 02:12:27 +0000 (11:12 +0900)]
[loco] Introduce AnnotatedItem class (#3289)
* [loco] Introduce AnnotatedItem class
This commit introduces AnnotatedItem<T> class which makes it easy to
define classs with annotation feature.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
* Fix a typo (stdext -> stdex)
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 18 Apr 2019 01:57:50 +0000 (10:57 +0900)]
[angkor] Do not use add_nncc_library (#3293)
This commit removes add_nncc_library call in angkor build script.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Wed, 17 Apr 2019 23:57:42 +0000 (08:57 +0900)]
Remove add_nncc_example_executable (#3291)
This commit removes add_nncc_example_executable as it is no longer used.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Wed, 17 Apr 2019 10:56:18 +0000 (19:56 +0900)]
[nnc] Remove add_nncc_example_executable calls (#3290)
This commit replaces all the occurrences of add_nncc_example_executable
with add_executable.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 17 Apr 2019 06:48:02 +0000 (15:48 +0900)]
[moco] Introduce convert_graph method (#3284)
This will introduce convert_graph method that will convert TensorFlow graph to loco graph
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 17 Apr 2019 03:10:28 +0000 (12:10 +0900)]
[moco] Add SymbolTable to GraphBuilderContext (#3281)
* [moco] Add SymbolTable to GraphBuilderContext
This will add two SymbolTable to GraphBuilderContext
- nodes to find nodes from string name
- input_names to find string names of a node
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* apply comment
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Wed, 17 Apr 2019 01:47:02 +0000 (10:47 +0900)]
Remove "ann" project (#3278)
Now there is no project that depends on "ann" itself.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 17 Apr 2019 01:34:09 +0000 (10:34 +0900)]
[moco] Introduce SymbolTable (#3265)
* [moco] Introduce SymbolTable
This will introduce SymbolTable class to record nodes and string names while building loco graph
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* add comments
* use node_name
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Wed, 17 Apr 2019 01:18:04 +0000 (10:18 +0900)]
[enco] Use ann-api and ann-ref directly (#3274)
"ann" is now just a meta package of "ann-api" and "ann-ref". Let's
remove this indirection and use ann-api and ann-ref directly.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Wed, 17 Apr 2019 01:10:55 +0000 (10:10 +0900)]
Remove add_nncc_test function (#3276)
This commit removes add_nncc_test function from the top-level CMakeLists.txt.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 17 Apr 2019 00:46:26 +0000 (09:46 +0900)]
[moco] remove cmake message print (#3273)
Remove message dump line that seems to be not intended
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 16 Apr 2019 10:48:30 +0000 (19:48 +0900)]
[mir] Remove add_nncc_test from build script (#3246)
This commit removes add_nncc_test from mib build script as a step to
deprecate add_nncc_test.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 16 Apr 2019 08:24:15 +0000 (17:24 +0900)]
Extract ann-ref from ann (#3258)
This commit extracts ann-ref code and build scripts from ann.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 16 Apr 2019 05:54:50 +0000 (14:54 +0900)]
[nnkit/TF] Enabling TF backend to handle multiple inputs and outputs (#3263)
* [nnkit/TF] Enabling TF backend to handle multiple inputs and outputs
This commits enables TF backend to receive multiple inputs and produce multiple outputs.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* remove test files in moco
* add error check for string::find(":")
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 16 Apr 2019 05:51:18 +0000 (14:51 +0900)]
[nnkit/TF] C++ version of parser for test.info (#3261)
* [nnkit/TF] C++ version of parser for test.info
Previously it was written in C, having resource destruction issue. Now it was converted into C++ version.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* add const, add parenthesis when macro param is used
* unnamed namespace
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 16 Apr 2019 04:50:53 +0000 (13:50 +0900)]
[moco/test/tf] tc with multiple inputs and outputs (#3270)
This commit adds a test case with multiple inputs and outputs.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 16 Apr 2019 04:42:51 +0000 (13:42 +0900)]
[moco/ONNX] Introduce empty moco ONNX frontend (#3267)
* [moco/ONNX] Introduce empty moco ONNX frontend
This commit will introduce empty `moco` ONNX frontend
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
* fix some words
* fix variable name
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 16 Apr 2019 02:53:41 +0000 (11:53 +0900)]
[moco] Support istream for loading (#3268)
This will add support methods for loading from istream
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 16 Apr 2019 01:45:25 +0000 (10:45 +0900)]
[moco] Introduce ModelSignature (#3264)
This will separate input and outputs of Frontend to ModelSignature
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 15 Apr 2019 12:10:09 +0000 (21:10 +0900)]
[nnc] Remove add_nncc_test from build script (#3247)
This commit removes add_nncc_test from build script as a step to remove
add_nncc_test from the top-level CMakeLists.txt.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 15 Apr 2019 09:48:58 +0000 (18:48 +0900)]
[moco] Provide TF/Frontend to set graph inputs,outputs (#3262)
This will add method to set graph input and output nodes by name
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 15 Apr 2019 02:04:04 +0000 (11:04 +0900)]
[ann] Temporarily disable runtime build (#3257)
This commit temporarily disables runtime build as a step to extract
ann-ref as an independent proejct.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 15 Apr 2019 02:03:54 +0000 (11:03 +0900)]
Introduce onnxkit (#3256)
* Introduce onnxkit
This commit introduces onnxkit which allows users to encode and decode
ONNX model files.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
* Fix a typo
* Fix comments & license
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 15 Apr 2019 02:03:42 +0000 (11:03 +0900)]
[nnkit/TF] Backend implementation for nnkit tensorflow backend (#3251)
* [nnkit/TF] Backend.h for nnkit tensorflow backend
This commit adds Backend.h which reads input, parse test.info, run tensorflow, and produce output.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* impl to cpp
* remove useless header. add #include <memory>
* headers
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Sun, 14 Apr 2019 23:12:09 +0000 (08:12 +0900)]
[ann] Add a dummy require for backward compatibility (#3255)
This commit inserts a dummy require from ann to ann-ref to maintain
backward compatiblity.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 12 Apr 2019 08:50:28 +0000 (17:50 +0900)]
Introduce ONNXSource 1.4.1 package (#3252)
Now, users are able to access the source of ONNX 1.4.1 release.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 12 Apr 2019 07:43:28 +0000 (16:43 +0900)]
[loco] Introduce "Forward" node (#3250)
This commit introduces "Forward" node which corresponds to "Identity"
op in TensorFlow.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 12 Apr 2019 07:07:06 +0000 (16:07 +0900)]
Create ann-ref project (#3249)
This commit introduces (empty) ann-ref project as the first step to
extract ann-ref from ann.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 12 Apr 2019 06:22:31 +0000 (15:22 +0900)]
[moco] Return graph of load (#3253)
This will change load function to return loco::Graph
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 11 Apr 2019 23:41:08 +0000 (08:41 +0900)]
[tflchef] Remove OwnedFileDescriptor (#3238)
This commit replaces all the occurrences of OwnedFileDescriptor with
cwrap::Fildes.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 11 Apr 2019 08:02:07 +0000 (17:02 +0900)]
[nnkit/TF] adding TensorContext (#3235)
* [nnkit/TF] adding TensorContext for Tensorflow background
This commit adds child class of TensorContext for Tensorflow backend.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* deleted move & copy construcors
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 11 Apr 2019 07:52:56 +0000 (16:52 +0900)]
[nnkit] Remove THREADS_PREFER_PTHREAD_FLAG setting (#3221)
The top-level CMakeLists.txt now sets THREADS_PREFER_PTHREAD_FLAG
value.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 11 Apr 2019 07:10:37 +0000 (16:10 +0900)]
Remove LLVM build from Dockerfile (#3236)
This commit removes scripts realted with LLVM build as it increases
Docker image build time too much (and there is no code that uses LLVM,
yet).
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
윤지영/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 11 Apr 2019 07:07:36 +0000 (16:07 +0900)]
[nnkit] Implement empty onnx backend (#3237)
This patch has an empty onnx backend.
It will be implemented with the functions to load and run the onnx file.
Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>