남궁석/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 21 May 2019 02:11:20 +0000 (11:11 +0900)]
[moco/ONNX] Change parameter of convert_graph (#3547)
For checking opset version of ONNX, onnx ModelProto is needed.
But until now, convert_graph has GraphProto, not ModelProto.
This commit will pass ModelProto parameter instead of GraphProto
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 21 May 2019 02:11:06 +0000 (11:11 +0900)]
[loco] Introduce shape helper method (#3544)
This commit extends Push and Pull node with shape helper method.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 21 May 2019 02:10:01 +0000 (11:10 +0900)]
[moco] finalize node conversion inherit classes (#3546)
This will finalize GraphBuilder and GraphUpdate inherited classes
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 21 May 2019 02:08:26 +0000 (11:08 +0900)]
[moco] use GraphUpdate to set input (#3540)
This will change to use GraphUpdate to set inputs of the node
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 20 May 2019 22:17:14 +0000 (07:17 +0900)]
[moco] Implement UpdateQueue::enroll() (#3538)
This will add UpdateQueue::enroll() implementation
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 20 May 2019 09:29:12 +0000 (18:29 +0900)]
[moco] Add UpdateQueue member (#3537)
This will add UpdateQueue attribute in GraphBuilderContext
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 20 May 2019 08:05:42 +0000 (17:05 +0900)]
[moco] Add two GraphUpdate inherits (#3536)
* [moco] Add two GraphUpdate inherits
This will add two GraphUpdate class inherits that needs graph update(input settings), Identity and Relu.
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* add const
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 20 May 2019 07:38:23 +0000 (16:38 +0900)]
[locomotiv] Session's graph input/output size (#3531)
* [locomotiv] Session's graph input/output size
This commit introduces Session::input_size() and Session::output_size()
which are getter for graph input and output size it possesses.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
* Make them const method
* Use const variable for magic number
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 20 May 2019 07:32:27 +0000 (16:32 +0900)]
[moco] add const method type (#3535)
This will add const for method using const SymbolTable
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 20 May 2019 06:55:00 +0000 (15:55 +0900)]
[moco] Introduce UpdateQueue (#3533)
This will introduce UpdageQueue class to store graph update objects
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 20 May 2019 04:59:30 +0000 (13:59 +0900)]
[moco] Introduce GraphUpdate interface (#3525)
* [moco] Introduce GraphUpdate interface
This will introduce GraphUpdate interface
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* add default destructor
* add const
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 20 May 2019 04:23:14 +0000 (13:23 +0900)]
Verify ONNXSource 1.4.1 archive checksum (#3528)
This commit updates ONNXSource 1.4.1 package to verify its archive after
download.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 20 May 2019 04:02:19 +0000 (13:02 +0900)]
Verify ONNXSource 1.3.0 archive checksum (#3527)
This commit updates ONNXSource 1.3.0 package to verify its archive after
downloading.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 20 May 2019 04:02:09 +0000 (13:02 +0900)]
[locomotiv] How to use locomotiv (#3524)
* [locomotiv] How to use locomotiv
This commit introduces documentation for how to use locomotiv.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
* Review feedback updated
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 20 May 2019 03:43:35 +0000 (12:43 +0900)]
[loco] Introduce Feature Permuting Codec (#3517)
* [loco] Introduce Feature Permuting Codec
This commit implements a basic feature codec based on axis permutation.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
* Update comments & add a missing assert message
* Check whether permutation axes are all distinct
* Fix a typo (Encoder -> Decoder)
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 20 May 2019 03:42:27 +0000 (12:42 +0900)]
[loco] Introduce IR verification infrastructure (#3526)
This commit introduces basic verifier infrastructure for loco.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 20 May 2019 03:20:56 +0000 (12:20 +0900)]
Verify downloaded archive (#3512)
With this commit, users are able to verify downloaded archive via
ExternalSource_Download with checksum.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 20 May 2019 00:45:47 +0000 (09:45 +0900)]
[loco] Introduce all_nodes helper (#3522)
This commit introduces all_nodes helper which allows users to enumerate
all the nodes in a given loco graph.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 17 May 2019 05:08:23 +0000 (14:08 +0900)]
[locomotiv] Support ConstGen execution (#3498)
This commit starts to support ConstGen node execution.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 17 May 2019 05:05:14 +0000 (14:05 +0900)]
[locomotiv] How to support new node execution (#3484)
* [locomotiv] How to support new node execution
This commit introduces a guide how to support new loco node execution.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
* Review feedback updated
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 17 May 2019 03:57:36 +0000 (12:57 +0900)]
Remove version-less ONNXSource package (#3518)
This commit removes version-less ONNXSource package to prevent possible
versioning issue.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 17 May 2019 02:22:43 +0000 (11:22 +0900)]
Use a specific ONNX version in ONNXProto (#3513)
This commit updates ONNXProto CMake package to use a specific version of
ONNX source code (1.3.0).
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 17 May 2019 02:22:31 +0000 (11:22 +0900)]
[nnc] Use versioned ONNXSource package (#3515)
This commit updates nnc build script to find a specific version of
ONNXSource package.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 17 May 2019 02:19:02 +0000 (11:19 +0900)]
[loco] Introduce asserts inside postorder_traversal (#3506)
This commit introduces several asserts inside postorder_traversal in
order to faciliate debugging.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 17 May 2019 02:14:49 +0000 (11:14 +0900)]
[moco/tf] Show executed commands (#3510)
With this commit, runall test script records executed commands into the log file.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 17 May 2019 02:14:38 +0000 (11:14 +0900)]
[locomotiv] Relocate codes that may be duplicates (#3503)
This will relocate relu codes that may be duplicates in the future.
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 16 May 2019 23:51:20 +0000 (08:51 +0900)]
[loco] Introduce Domain enum class (#3500)
* [loco] Introduce Domain enum class
This commit introduces Domain enum class which denotes the kind of
a loco value.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
* Update comment
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 16 May 2019 07:50:27 +0000 (16:50 +0900)]
[locomotiv] Remove data update duplication in Forward execution (#3507)
This commit removes NodeData update duplication in switch statement in
Forward execution.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 16 May 2019 06:57:41 +0000 (15:57 +0900)]
[locomotiv] Remove data update duplication in Push execution (#3496)
* [locomotiv] Remove data update duplication in Push execution
This commit removes NodeData update duplication in switch statement in
Push execution.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
* Add assertion
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 16 May 2019 06:26:04 +0000 (15:26 +0900)]
[loco] Introduce Feature Encode/Decode node (#3495)
This commit extends loco IR with FeatureEncode/FeatureDecode nodes.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 16 May 2019 04:54:25 +0000 (13:54 +0900)]
[locomotiv] Introduce ReLU (#3485)
* [locomotive] Introduce ReLU
This will enable ReLU FLOAT32 execution
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* remove unused using
* use index enumerator
* remove unused code
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 16 May 2019 04:29:23 +0000 (13:29 +0900)]
[moco] import TensorFlow Const for some float values (#3494)
This will enable TensorFlow Const conversion where value is given with float value but with partial elements.
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 16 May 2019 03:55:44 +0000 (12:55 +0900)]
[moco] import TensorFlow Const for binary encoded (#3492)
This will enable TensorFlow Const conversion where value is given as tensor_content.
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 16 May 2019 03:49:28 +0000 (12:49 +0900)]
[moco] enable TensoFlow Relu node import (#3488)
This will enable TensorFlow Relu node import in frontend.
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 16 May 2019 03:47:07 +0000 (12:47 +0900)]
[loco] Introduce Feature Encoder & Decoder (#3479)
* [loco] Introduce Feature Encoder & Decoder
This commit introduces Feature Encoder & Decoder interface.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
* Update comments per feedback
* Another comment update
* Remove unnecessary space
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 16 May 2019 03:43:29 +0000 (12:43 +0900)]
[moco] test material for TensorFlow Relu (#3490)
This will add fist unit test material for importing TensorFlow Relu node.
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 16 May 2019 03:37:59 +0000 (12:37 +0900)]
[locomotiv] Docstring for NodeData and make_data() (#3483)
* [locomotiv] Docstring for NodeData and make_data()
This commit adds docstring for NodeData class and make_data() function.
NodeData is read-only, and not modifiable after creation.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
* Review feedback updated
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 16 May 2019 03:36:13 +0000 (12:36 +0900)]
[moco] import TensorFlow Const for one value data (#3491)
This will enable TensorFlow Const conversion where there is only one float value.
And add a test code for this case.
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 16 May 2019 01:54:15 +0000 (10:54 +0900)]
[moco] import Const attributes and value data (#3487)
This will enable import for Const attributes and data where there are same number of const value data.
Add test code for this case.
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 16 May 2019 00:51:47 +0000 (09:51 +0900)]
[moco] introduce test helper header (#3476)
* [moco] introduce test helper header
This will add some helpers for node unit testing
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* move outside namespace
* use istream
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Wed, 15 May 2019 10:37:59 +0000 (19:37 +0900)]
Add ONNXSource 1.3.0 package (#3481)
This commit introduces ONNXSource v1.3.0 package.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
[loco] Eliminate CRTP from Named class and rename it to NameInfo (#3464)
* rename Named class to NameInfo
* remove virtual destructor because this class is not designed for public inheritance
Signed-off-by: Roman Rusyaev <r.rusyaev@samsung.com>
Ivan Vagin/AI Tools Lab /SRR/Engineer/삼성전자 [Wed, 15 May 2019 09:14:32 +0000 (12:14 +0300)]
[nnkit/moco/tf] Build moco_test_tf only when TensorFlow is found (#3465)
Build moco_test_tf only when TensorFlow is found
Signed-off-by: Ivan Vagin <ivan.vagin@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 15 May 2019 08:28:29 +0000 (17:28 +0900)]
[moco] introduce empty TF Const converter (#3477)
This will introduce empty Const node converter
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 15 May 2019 06:42:42 +0000 (15:42 +0900)]
[moco/tf] Introduce get_tensor_attr (#3468)
This will introduce get_tensor_attr method to get tensor attribure from node
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 15 May 2019 06:42:26 +0000 (15:42 +0900)]
[loco] Introduce ReLU node (#3472)
This will extend loco IR with ReLU node.
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Wed, 15 May 2019 06:42:15 +0000 (15:42 +0900)]
[loco] Introduce MaxPool2D node (#3473)
This commit introduces MaxPool2D node.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Wed, 15 May 2019 05:23:13 +0000 (14:23 +0900)]
Distinguish unversioned ONNXSource package (#3469)
This commit introduces ONNXSourceConfigVersion.cmake for unversioned
package in order to eliminate ambiguity in package searching.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 14 May 2019 10:59:59 +0000 (19:59 +0900)]
[loco] Introduce Window class (#3459)
This commit declares (generic) Window class, and provides its
specialization over 2D Window.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 14 May 2019 10:59:49 +0000 (19:59 +0900)]
[loco] Introduce Stride class (#3461)
* [loco] Introduce Stride class
This commit declares (generic) Stride class, and provides its
specialization over 2D.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
* Mark as a final class
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 14 May 2019 09:50:38 +0000 (18:50 +0900)]
[loco] Introduce FeatureIndex class (#3458)
This commit introduces FeatureIndex class which is the offset of an
element in a feature map.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 14 May 2019 03:55:39 +0000 (12:55 +0900)]
[nnkit/moco/tf] `runall` and build script (#3449)
This commit adds `runall` and build script, working for `Placeholder_000` test case.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Tue, 14 May 2019 03:50:38 +0000 (12:50 +0900)]
[loco] Introduce FeatureShape class (#3448)
This commit introduces FeautreShape class which describes the shape of
Feature Map.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 14 May 2019 00:50:06 +0000 (09:50 +0900)]
[nnkit/moco/tf] Adds getConstFloatTensor for InputTensorContext (#3447)
This adds getConstFloatTensor() method body for InputTensorContext, which is called by HDF5::Export.cpp
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 13 May 2019 23:35:46 +0000 (08:35 +0900)]
[loco] Introduce TensorIndex class (#3445)
* [loco] Introduce TensorIndex class
This commit introduces TensorIndex class to loco IR.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
* Fix a typo
Андрей Шедько/AI Tools Lab /SRR/Engineer/삼성전자 [Mon, 13 May 2019 16:23:55 +0000 (19:23 +0300)]
[MIR] Added Readme (#3271)
Added a readme for Model IR
Signed-off-by: Andrei Shedko <a.shedko@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 13 May 2019 10:12:11 +0000 (19:12 +0900)]
[nnkit/moco/tf] Fix bug while calling ModelSignature (#3446)
With this commit, node name (without, e.g., ":0") is passed to ModelSignature, not a tensor name with, e.g., ":0".
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 13 May 2019 09:46:33 +0000 (18:46 +0900)]
[nnkit/moco/tf] Restructuring 'using' (#3438)
`using` is restructured.
ParsedTensors are common so move `using` with ParsedTensors to top level. Others were moved to method level.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 13 May 2019 09:13:37 +0000 (18:13 +0900)]
[nnkit/moco/tf] backend of nnkit (#3444)
* [nnkit/moco/tf] backend of nnkit
This commit adds backend of nnkit for nnkit/moco/tf.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* indentation
채성우/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 13 May 2019 09:13:14 +0000 (18:13 +0900)]
[CMake] Check TensorFlow version (#3373)
* [CMake] Check TensorFlow version
This commit introduce check of tensorflow version.
Signed-off-by: seongwoo <sw4670.chae@samsung.com>
* This commit introduce .FORMATCHECKED
Signed-off-by: seongwoo <sw4670.chae@samsung.com>
* This commit apply format.patch
Signed-off-by: seongwoo <sw4670.chae@samsung.com>
* This commit make code safe.
Signed-off-by: seongwoo <sw4670.chae@samsung.com>
* This commit remove modification of 'format'
Signed-off-by: seongwoo <sw4670.chae@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 13 May 2019 07:46:07 +0000 (16:46 +0900)]
[locomotiv] Session frees node annotations on destruction (#3441)
* [locomotiv] Session frees node annotations on destruction
This commit introduces destructor of Session. Previously node
annotations set by Session, like NodeDataImpl, remain even after
Session closed. They were not memory leak by definition, as they can
be accessed through another Session. However node annotations maybe
preserved until program ends, so it is like memory leak. This commit
would prevent that situation.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
* Delete unused using Index
* Session is final class
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 13 May 2019 07:42:03 +0000 (16:42 +0900)]
[loco] Introduce TensorShape class (#3443)
This commit introduces TensorShape class to loco IR.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 13 May 2019 07:29:28 +0000 (16:29 +0900)]
Extract nnkit-tflite (#3442)
This commit extracts nnkit modules related with TensorFlow Lite as
nnkit-tflite.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 13 May 2019 05:29:26 +0000 (14:29 +0900)]
[nnkit/tf] fix wrong name of method (#3440)
This commit renames isInt32Tensor to isS32Tensor.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
박세희/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 13 May 2019 04:48:32 +0000 (13:48 +0900)]
[tfldump] Add readme (#3297)
* [tfldump] Add readme
This will add readme for tfldump project
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* fix typo and some comments
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 13 May 2019 03:46:54 +0000 (12:46 +0900)]
[locomotiv] Support Forward node (#3436)
This commit starts to support loco::Forward node. In terms of node
execution, Forward calculation is same with Push's.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 13 May 2019 03:44:58 +0000 (12:44 +0900)]
[locomotiv] Use break rather than return (#3437)
For Push node execution, let's use break rather than return inside
switch statement, as it feels more natural.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Mon, 13 May 2019 03:44:38 +0000 (12:44 +0900)]
[nnkit] int32 support for nnkit::TensorContext (#3413)
* [nnkit] int32 support for nnkit::TensorContext
This commit adds methods to support int32 for nnkit::TensorContext
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* Int32->S32, added header
* cluster S32 function tegether
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Mon, 13 May 2019 02:12:39 +0000 (11:12 +0900)]
Extract nnkit-caffe (#3435)
This commit extracts nnkit-caffe from nnkit, and simplifes the depedency
check in CMakeLists.txt.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Mon, 13 May 2019 02:12:01 +0000 (11:12 +0900)]
[locomotiv] Use macro to iterate supported Nodes (#3429)
* [locomotiv] Use macro to iterate supported Nodes
This commit introduces list of supported Nodes for execution, and use
them with macro to iterate and do same thing through Nodes.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
* clang-format off for comment indentation
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Sun, 12 May 2019 23:47:43 +0000 (08:47 +0900)]
[loco] Introduce ConstGen node (#3425)
* [loco] Introduce ConstGen node
This commit extends loco IR with ConstGen node.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
* Add assert
* Fix a bug and add a corresponding test
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Sun, 12 May 2019 23:45:58 +0000 (08:45 +0900)]
[nnkit/moco/tf] backend implementation (#3424)
* [nnkit/moco/tf] backend implementation
This commit adds backend implementation for nnkit/moco/tf.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* move using into class
* fixed for locomotiv set_input
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 10 May 2019 09:49:30 +0000 (18:49 +0900)]
Extract "nnkit-intf" from "nnkit" (#3433)
This commit extracts "nnkit-intf" from "nnkit" as the first step toward
nnkit directory restructuring.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
윤지영/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 10 May 2019 08:11:10 +0000 (17:11 +0900)]
[nnkit] Run onnx model using onnx runtime (#3432)
This patch has TensorContext class to run pre and post actions.
Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 10 May 2019 06:41:54 +0000 (15:41 +0900)]
[loco] Introduce input_nodes/output_nodes helper (#3428)
* [loco] Introduce input_nodes/output_nodes helper
This commit introduces input_nodes/output_nodes helper functions which
allow users to easily enumerate graph-level input/output nodes.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
* Update test name
윤지영/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 10 May 2019 05:52:43 +0000 (14:52 +0900)]
[nnkit] Add TensorSet class for managing TensorData of onnx model (#3427)
This patch adds TensorSet class.
And it also support to allocate and release the input and output tensor datas.
Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
윤지영/On-Device Lab(SR)/Staff Engineer/삼성전자 [Fri, 10 May 2019 05:43:32 +0000 (14:43 +0900)]
[nnkit] Remove finalize function of nnkit::Backend (#3430)
This patch does not remove the plugin unique_ptr directly.
Instead, it passes it in so that it is removed according to the unique_ptr policy.
Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 9 May 2019 07:40:33 +0000 (16:40 +0900)]
[locomotiv] Implement Session::get_output() (#3423)
This commit implements how session get its output. Related test for this
feature added as well.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
윤지영/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 9 May 2019 07:33:21 +0000 (16:33 +0900)]
[nnkit] Add OnnxRunner for managing onnx runtime (#3417)
* [nnkit] Add OnnxRunner for managing onnx runtime
This patch adds the OnnxRunner class.
This class is for managing the onnx runtime to prepare input and output and to run the model.
Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
* Do not throw exception in constructor.
In this case, it will make an assertion.
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 9 May 2019 06:57:46 +0000 (15:57 +0900)]
[locomotiv] Implement inference for session (#3419)
This commit implements Session::infer() which does inference. It also
shows test how to use this.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Thu, 9 May 2019 05:33:00 +0000 (14:33 +0900)]
[nnkit] Removed unused methods in TensorContext (#3418)
Two methods in TensorContext seems not to be used. So removed.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Thu, 9 May 2019 01:46:31 +0000 (10:46 +0900)]
[locomotiv] Support Pull node (#3415)
This commit introduces support for Pull node execution. Pull node
basically do nothing, but aborts if input data annotation not ready.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
박종현/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 9 May 2019 00:24:25 +0000 (09:24 +0900)]
[loco] Introduce DataTypeImpl trait (#3399)
This commit introduces DataTypeImpl traits which returns C++ type
corresponding to each DataType enum value.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
채성우/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 8 May 2019 23:45:59 +0000 (08:45 +0900)]
Add .c file to format-check-candidate (#3414)
This commit adds .c file to format-check-candidate.
Signed-off-by: seongwoo <sw4670.chae@samsung.com>
Андрей Шедько/AI Tools Lab /SRR/Engineer/삼성전자 [Wed, 8 May 2019 13:37:32 +0000 (16:37 +0300)]
[MIR] Added a Remove Dead Ends Pass (#3272)
Added a `RemoveDeadEnds` pass
that decreases compiler runtime memory usage.
Signed-off-by: Andrei Shedko <a.shedko@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 8 May 2019 04:46:23 +0000 (13:46 +0900)]
[locomotiv] Implement Push node execution (#3408)
This commit implements Push node execution and introduces related tests.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 8 May 2019 03:54:50 +0000 (12:54 +0900)]
[moco/test/tf] test cases with Const (#3409)
This commit adds test cases with Const operation.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Wed, 8 May 2019 03:54:24 +0000 (12:54 +0900)]
[moco/test/tf] test files for reshape (#3410)
This adds test files for reshape.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
윤지영/On-Device Lab(SR)/Staff Engineer/삼성전자 [Wed, 8 May 2019 03:50:30 +0000 (12:50 +0900)]
[nnkit] Add Allocator for onnx backend (#3372)
* [nnkit] Add Allocator for onnx backend
This class is for management onnx tensor memory.
Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
* Add comments
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Wed, 8 May 2019 01:29:58 +0000 (10:29 +0900)]
[locomotiv] Introduce NodeExecution helper class (#3402)
This commit introduces NodeExecution class. This class serves as helper
class for Session::infer(). This class is responsible to calculate one
specific node.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
Андрей Шедько/AI Tools Lab /SRR/Engineer/삼성전자 [Tue, 7 May 2019 15:59:12 +0000 (18:59 +0300)]
[nncc-optimization] Sink transpose optimization added (#3292)
- Added the transpose sinking optimization
- Moved swapAdjacent to OptimizationUtils
Signed-off-by: Andrei Shedko <a.shedko@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 7 May 2019 08:12:57 +0000 (17:12 +0900)]
[nnkit/moco/tf] TensorContext for output tensor (#3406)
This commit adds OutputTensorContext, which is a child of TensorContext for output tensor.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 7 May 2019 03:18:21 +0000 (12:18 +0900)]
[nnkit/moco/tf] Make cmake work only when Tensorflow is found (#3405)
This commits makes cmake builds nnkit/moco/tf only when TF is found
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
[nnc] Refine Reduce operation implementation in interpreter (#3358)
Rename `ReduceFOp` to `ReduceOp`, `ReduceN` to `Reduce`.
Speed up `ReduceOp` operation implementation in interpreter.
Reformat according to coding style.
Remove unused files.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 3 May 2019 09:23:03 +0000 (18:23 +0900)]
[locomotiv] Introduce NodeDataImpl annotation erase (#3400)
This commit introduces erase_annot_data() function to erase already
annotated NodeDataImpl. It also includes test for added function.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Fri, 3 May 2019 08:38:05 +0000 (17:38 +0900)]
[nnkit/moco/tf] TensorContext for input tensor (#3392)
* [nnkit/moco/tf] TensorContext for input tensor
This commit adds InputTensorContext, which is a child of TensorContext for input tensor.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* absolute path for include dir to relative path
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Fri, 3 May 2019 02:18:31 +0000 (11:18 +0900)]
[locomotiv] Implement 'set_input' of Session (#3395)
* [locomotiv] Implement 'set_input' of Session
This commit introduces implementation of Session::set_input() and its
relevant test. This method now returns bool according to accomplishment.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
* Add docstring for 'set_input'
윤지영/On-Device Lab(SR)/Staff Engineer/삼성전자 [Thu, 2 May 2019 06:58:19 +0000 (15:58 +0900)]
[nnkit] Error handling for dlopen/dlsym and dlclose (#3355)
* [nnkit] Error handling for dlopen/dlsym and dlclose
This patch allows to handle some error case of dlopen/dlsym and dlclose.
Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
* Remove runtime_error exception in comments
* Second version for error handling of dl related functions
박천교/On-Device Lab(SR)/Engineer/삼성전자 [Tue, 30 Apr 2019 06:47:57 +0000 (15:47 +0900)]
[locomotiv] const type qualified fuction arguments (#3391)
This commit adds const type qualifier for relevant function arguments
for NodeData.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
윤현식/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 30 Apr 2019 00:35:38 +0000 (09:35 +0900)]
[nnkit/moco/tf] bug fix in CMakeLists.txt (#3388)
This commit fixes wrong directory name in CMakeLists.txt.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>