박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Mon, 17 Dec 2018 04:16:26 +0000 (13:16 +0900)]
[encodump] Use stdex::make_unique (#2691)
This commit replaces all the occurrences of make_unique in
nncc::foundation with stdex::make_unique.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
[nnc] Replace ElementwiseOp args number deserialization with generated number (#2678)
[nnc] Issue: SVACE STATIC_OVERFLOW in soft_backend/code_snippets/cpp_operations.def
Signed-off-by: Timur Ablyazimov <t.ablyazimov@samsung.com>
Ivan Vagin/AI Tools Lab /SRR/Engineer/삼성전자 [Fri, 14 Dec 2018 19:25:40 +0000 (22:25 +0300)]
[nnc] Fixed caffe2 importer MIR operator names (#2683)
* Don't name any MIR operators, except for i/o
* Fixed default pooling stride size
Signed-off-by: Ivan Vagin <ivan.vagin@samsung.com>
[nnc] Problem in the ACL backend system tests fixed (#2682)
Fixed a bug in the tensor_gen. nni was renamed to nnkit-run so we had to change this in the CMakeLists.txt of the ACL backend system tests. Also the dependencies of tensor_gen, caffegen and nnkit-run were done obligatory.
Signed-off-by: Timur Ablyazimov <t.ablyazimov@samsung.com>
[nnc] Rectify shared_ptr wrapper around TensorVariant (#2684)
- Get rid of shared_ptr wrapper around TensorVariant class.
- Add const qualifiers to auto reference declarations
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
Андрей Шедько/AI Tools Lab /SRR/Engineer/삼성전자 [Fri, 14 Dec 2018 17:20:47 +0000 (20:20 +0300)]
[nnc] SQRT Operation (#2514)
- Added sqrt op to backends (interpreter and c++ SB)
- Added sqrt to tflite importer
Signed-off-by: Andrei Shedko <a.shedko@partner.samsung.com>
[nnc] Implement Gather operation in interpreter backend (#2681)
Add implementation of Gather operation to interpreter backend.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
updated SDD for completion DR (#2670)
* remove svace defects
Signed-off-by: Roman Rusyaev <r.rusyaev@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 14 Dec 2018 08:10:18 +0000 (17:10 +0900)]
[enco] Build ANN IR from ANNConv2D instruction (#2676)
This commit extends Split pass to build ANN IR from ANNConv2D
instruction.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 14 Dec 2018 05:41:47 +0000 (14:41 +0900)]
[coco] Introduce object casting helpers (#2673)
This commit introduces the following object casting helpers:
- isa<T>
- cast<T>
- safe_cast<T>
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
윤현식/동작제어Lab(SR)/Principal Engineer/삼성전자 [Fri, 14 Dec 2018 05:29:41 +0000 (14:29 +0900)]
[coco] Modified FeatureObject test, working for coco::FeatureShape (#2674)
With #2667, a warning was thrown. (ambiguous operator) This PR resolve the issue.
Signed-off-by: Hyun Sik Yoon hyunsik.yoon@samsung.com
윤현식/동작제어Lab(SR)/Principal Engineer/삼성전자 [Fri, 14 Dec 2018 02:12:50 +0000 (11:12 +0900)]
[coco] "==", "!=" op for FeatureShape (#2667)
This commit adds "==", "!=" operators for FeatureShape.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
윤현식/동작제어Lab(SR)/Principal Engineer/삼성전자 [Fri, 14 Dec 2018 01:11:56 +0000 (10:11 +0900)]
[enco] Validating output shape of Conv2D (#2623)
* [enco] IR validator : validating output shape of conv2D
This commit adds code to check if output shape of conv2D is same with Eval.out
Signed-off-by: Hyun Sik Yoon hyunsik.yoon@samsung.com
* pr fix: comment on Dec 11
* pr fix: assert() for remoted round
* pr fix: string_width, striding_height
* pr fix: various.
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 14 Dec 2018 01:11:04 +0000 (10:11 +0900)]
[nncc/core] Do NOT link unused foundation library (#2672)
nncc_core has no implemenation/declaration that depends on
nncc_foundation.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
윤지영/동작제어Lab(SR)/Engineer/삼성전자 [Fri, 14 Dec 2018 00:51:05 +0000 (09:51 +0900)]
[tflchef] Remove default activation of concat test recipe (#2658)
* [tflchef] Remove default activation of concat test recipe
This patch removes the default activation value of concat test recipe.
It will be assigned NONE type by default.
Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
* Use NONE value instaed of removing that line
Signed-off-by: Jiyoung Yun <jy910.yun@samsung.com>
Андрей Шедько/AI Tools Lab /SRR/Engineer/삼성전자 [Thu, 13 Dec 2018 17:36:52 +0000 (20:36 +0300)]
[nnc] Added Broadcasting (#2417)
- Added Broadcasting to interpreter and shapeInference.
- Fixed interpreter impl and placed const in correct location
- Added broadcasting to softBackend
- Added tests and cpp backend implementations for broadcasted `+, -, *, /, max`
- Added generic implementation of Elementwise Broadcast for ease of
- fixed codestyle; removed shape inference file
Signed-off-by: Andrei Shedko <a.shedko@samsung.com>
Ivan Vagin/AI Tools Lab /SRR/Engineer/삼성전자 [Thu, 13 Dec 2018 16:53:10 +0000 (19:53 +0300)]
[nnc] Caffe2 importer improvements (#2664)
* Tested and fixed operators needed for 'inception': Add, Mul, Concat, SpatialBN
* Supported custom paddings
* Supported custom pooling window shapes
* Several bugs fixed
This PR enables 'TransposeOp' in caffe2 frontend, which is not currently supported by ACL backend.
Signed-off-by: Ivan Vagin <ivan.vagin@samsung.com>
[nnc] Support for Gather operation in soft backend (#2663)
Add support for Gather operation to soft backend.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[nnc] Fix broken ConstantOp (#2671)
Revert recent changes made to ConstantOp that broke it.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
Fix build issue #2668 (#2669)
- Build failure issued by merging of #2646: convertBlob return value changed
- Small code style changes
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[nnc] Support for Caffe Embed layer (#2646)
* Add support for Embed layer.
* Refactor creating of BiasAdd op in Caffe importer.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
Ivan Vagin/AI Tools Lab /SRR/Engineer/삼성전자 [Thu, 13 Dec 2018 10:28:26 +0000 (13:28 +0300)]
[nnc] Initialize uninitialized variable in caffe2 importer (#2665)
Initialize uninitialized variable in caffe2 importer
Signed-off-by: Ivan Vagin <ivan.vagin@samsung.com>
[nnc] New Gemm implementation, GlobalAveragePool and Conv2D updates. (#2622)
The first steps in new Gemm implementation were done: Now it's a standard NN operator with proper shape inference. Interpreter (an d maybe soft backend) support will be added in the next patch.
Pool and Conv operators are using similar input data that's why they were partially merged. As result they now produce proper output shape.
Signed-off-by: Andrew V. Tischenko a.tischenko@partner.samsung.com
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 13 Dec 2018 08:26:18 +0000 (17:26 +0900)]
[enco] Introduce ANNConv2D instruction (#2655)
This commit introduces ANNConv2D instruction class as a part of extended
coco IR.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 13 Dec 2018 08:26:09 +0000 (17:26 +0900)]
[nnsuite] Remove unused includes (#2656)
This commit removes unused includes from nnsuite code.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 13 Dec 2018 08:25:56 +0000 (17:25 +0900)]
[enco] Do NOT use nncc::foundation::make_unique (#2657)
This commit replaces all the occurrences of
nncc::foundation::make_unique with stdex::make_unique.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 13 Dec 2018 03:50:15 +0000 (12:50 +0900)]
[nnkit] Use make_unique in stdex instead of nncc::foundation (#2654)
This commit replaces all the occurrences of nncc::foundation::make_unique
with stdex::make_unique.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 13 Dec 2018 03:16:06 +0000 (12:16 +0900)]
[nnsuite] Use stdex::make_unique (#2653)
This commit replaces all the use of nncc::foundation::make_unique in
nnsuite with stdex::make_unique.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 13 Dec 2018 02:19:20 +0000 (11:19 +0900)]
[cli] Use stdex instead of nncc_foundation (#2650)
This commit replaces all the use of nncc::foundation::make_unique in cli
project with stdex::make_unique.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 13 Dec 2018 02:19:10 +0000 (11:19 +0900)]
[enco] Support bias setting for ANNConv2DAppender (#2652)
With this commit, users are able to a pass bias term to ANNConv2DAppender.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Thu, 13 Dec 2018 00:20:34 +0000 (09:20 +0900)]
[enco] Initialize pointers in ANN Appenders (#2648)
This commit updates ANN Appenders implemented for Split pass to
initialize its pointer fields on construction.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
[nnc] Add new dom to text tests for acl backend (#2641)
Tests added:
- ArtifactId
- ArtifactRef
- ArtifactDeref
- ArtifactFunctionCall
- ArtifactUnaryExpr
- ArtifactBinaryExpr
Signed-off-by: Efimov Alexander <a.efimov@samsung.com>
[nnc] Refine soft backend documentation (#2543)
- Updated interface documentations
- Updated main execution sequence
- Fixed typos
Signed-off-by: Efimov Alexander <a.efimov@samsung.com>
[nnc] Acl backend system tests for Mobilenet and Inception-v3 operations (#2229)
The system tests framework for the ACL soft backend has been created. The test cases for the operations used in the Mobilenet and the Inception-v3 operations have been added.
Signed-off-by: Timur Ablyazimov <t.ablyazimov@samsung.com>
[nnc] Initialize member of AclCppOpGenerator properly (#2642)
AclCppOpGenerator _constrBlock and _infBlock member variables were not initialized
Signed-off-by: Timur Ablyazimov <t.ablyazimov@samsung.com>
Vladimir Plazun/AI Tools Lab /SRR/Engineer/삼성전자 [Wed, 12 Dec 2018 12:50:02 +0000 (15:50 +0300)]
[nnc] Change IrDotDumper handling of misc values (#2542)
Makes all misc values into strings and adds automatic conversion routine
Signed-off-by: Vladimir Plazun <v.plazun@partner.samsung.com>
Vladimir Plazun/AI Tools Lab /SRR/Engineer/삼성전자 [Wed, 12 Dec 2018 10:13:56 +0000 (13:13 +0300)]
[nnc] Update constant support in Graph (#2620)
Do not list constants as inputs
Mark inputs resolved before running visitor
Register constant automatically, no need to keep track in importers
Signed-off-by: Vladimir Plazun <v.plazun@partner.samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Wed, 12 Dec 2018 10:12:46 +0000 (19:12 +0900)]
[nncc/foundation] Remove math functions (#2638)
These math functions are deprecated by "nike" library.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
[nnc] Introduce Gather operation (#2626)
Add GatherOp class to modelIR.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Wed, 12 Dec 2018 08:14:57 +0000 (17:14 +0900)]
[nike] RelativeEpsilonEqual (#2636)
This commit introduce RelativeEpsilonEqual comparison algorithm.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
남궁석/동작제어Lab(SR)/Engineer/삼성전자 [Wed, 12 Dec 2018 07:49:30 +0000 (16:49 +0900)]
[coco] Remove deprecated method (#2624)
This commit will remove deprecated method `Shuffle::at`
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Wed, 12 Dec 2018 07:32:45 +0000 (16:32 +0900)]
[coco] Remove deprecated allocate method (#2635)
This commit removes (deprecated) allocate method which takes "object"
as an argument.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Wed, 12 Dec 2018 05:05:46 +0000 (14:05 +0900)]
[enco] Allocate weight through bag (#2634)
This commit removes all the allocate method call with "object" (which is
deprecated).
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Wed, 12 Dec 2018 01:54:37 +0000 (10:54 +0900)]
[coco/generic] Return allocated span (#2630)
Now, allocate method in PlainWeightContext<T> returns the allocated span
as its return.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
[nnc] Add convenient constructor to mir::Shape class (#2627)
Add a constructor to mir::Shape that accepts rank of the shape.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Tue, 11 Dec 2018 10:45:24 +0000 (19:45 +0900)]
Remove clang-format configurations for JavaScripts (#2625)
There is no JavaScript code to be format checked in nncc.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
남궁석/동작제어Lab(SR)/Engineer/삼성전자 [Tue, 11 Dec 2018 10:16:46 +0000 (19:16 +0900)]
[enco] Substitute deprecated method (#2621)
This commit will substitute deprecated method to the right one
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
남궁석/동작제어Lab(SR)/Engineer/삼성전자 [Tue, 11 Dec 2018 09:37:33 +0000 (18:37 +0900)]
[enco] Remove deprecated intermediates function (#2618)
This commit will remove deprecated function, `intermediates`
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
Fix build bug #2604 (#2616)
* not enable unit test if numpy is not installed
Signed-off-by: Roman Rusyaev <r.rusyaev@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Tue, 11 Dec 2018 09:15:27 +0000 (18:15 +0900)]
[nncc.foundation] Remove classes related Region (#2615)
There is no code that uses these classes.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
[nnc] Support batch in InnerProduct layer in Caffe importer (#2599)
Add support for batch size to InnerProduct layer.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Tue, 11 Dec 2018 07:21:29 +0000 (16:21 +0900)]
Introduce "nike" (#2609)
* Add a "nike" library
This commit introduces a new library called "nike", which will serve a
collection of numeric value comparison routines.
The current implementation includes one comparison routine.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
* Fix a typo (ABSOLUATE -> ABSOLUTE)
* Add a comment over NaN comparision rule
박세희/동작제어Lab(SR)/Principal Engineer/삼성전자 [Tue, 11 Dec 2018 05:44:23 +0000 (14:44 +0900)]
[tflchef] separate tflchef and reverse test (#2613)
* [tflchef] separate tflchef and reverse test
This will separate build and testing of tflchef and tflchef-reverse so that one can develop step-by-step
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* litte edit with note
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Tue, 11 Dec 2018 02:44:05 +0000 (11:44 +0900)]
[enco/tfl] Support feature map concatenation along batch (#2608)
* [enco/tfl] Support feature map concatenation along batch
This commit updates enco tflite frontend to support feature map
concatenation along bath dimension.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
* Fix a typo (unit32_ -> uint32_t)
남궁석/동작제어Lab(SR)/Engineer/삼성전자 [Tue, 11 Dec 2018 02:05:47 +0000 (11:05 +0900)]
[coco] Remove deprecated DefaultVisitor (#2607)
This commit will remove DefaultVisitor and substitute as Visitor
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
윤현식/동작제어Lab(SR)/Principal Engineer/삼성전자 [Tue, 11 Dec 2018 01:52:25 +0000 (10:52 +0900)]
[enco] skeleton of IR validator (#2558)
This is a skeleton code for validating IR in backend.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Tue, 11 Dec 2018 01:41:10 +0000 (10:41 +0900)]
[tflchef] Log test output (#2606)
* [tflchef] Log test output
This commit revises runall to log the output of each test into a
dedicated file.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
* Log running command
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Tue, 11 Dec 2018 00:42:05 +0000 (09:42 +0900)]
[tflchef] Show failed tests on test failure (#2602)
In this commit, runall in tflchef is revised to show the list of failed
tests on test failure.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Tue, 11 Dec 2018 00:40:29 +0000 (09:40 +0900)]
[coco] Use consistent class declarations for Managers (#2603)
This commit revises Manager declarations to use consistent style.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
남궁석/동작제어Lab(SR)/Engineer/삼성전자 [Tue, 11 Dec 2018 00:39:47 +0000 (09:39 +0900)]
[tflchef] Update README and related test (#2430)
This commit will introduce gaussian filler in tflchef
And also recipe of readme will be modified
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Tue, 11 Dec 2018 00:17:13 +0000 (09:17 +0900)]
[coco] Add FeatureShape constructor with explicit arguments (#2592)
This commit adds a FeatureShape constructor which allows users to
specify batch/depth/height/width explicitly.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Tue, 11 Dec 2018 00:16:56 +0000 (09:16 +0900)]
[coco] Create BHWC FeatureLayout with FeatureShape (#2593)
This change allows users to create a BHWC FeatureLayout with non-singular
batch dimension.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박세희/동작제어Lab(SR)/Principal Engineer/삼성전자 [Tue, 11 Dec 2018 00:16:45 +0000 (09:16 +0900)]
[enco] frontend/caffe: remove unused header includes (#2600)
* [enco] frontend/caffe: remove unused header includes
This will remove unused header files in Frontend.cpp
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* remove one more
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Tue, 11 Dec 2018 00:16:06 +0000 (09:16 +0900)]
[nnkit] Remove nni (#2601)
This commit removes deprecated nni project.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
Ivan Vagin/AI Tools Lab /SRR/Engineer/삼성전자 [Mon, 10 Dec 2018 23:08:48 +0000 (02:08 +0300)]
[enco] Fixed build error (#2597)
Fixed error: "no template named 'function' in namespace 'std'"
Signed-off-by: Ivan Vagin <ivan.vagin@samsung.com>
[nnc] Support 'transpose' parameter of InnerProduct layer (#2598)
Add support for 'transpose' parameter of InnerProduct layer.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[nnc] Add arm-c++ target (#2515)
Adds new target option to make possible split of arm and x86 artifact generation in cpu c++ backend
Signed-off-by: Efimov Alexander <a.efimov@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Mon, 10 Dec 2018 18:34:03 +0000 (03:34 +0900)]
[nnc] Use nnkit-run for test (#2585)
Now, infer_testcase.py script will invoke nnkit-run instead of
deprecated nni.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
update Unit Test Report for completion DR (#2596)
* add test round 3 including results of unit tests for ACL C++ code generator
Signed-off-by: Roman Rusyaev <r.rusyaev@samsung.com>
update STD and SDD for completion DR (#2595)
This is a patch from Alexey Kondrashov.
Signed-off-by: Roman Rusyaev <r.rusyaev@samsung.com>
update SW Development Document for completion DR (#2571)
* update SW Development Document for completion DR
* add report of ProtexIP
* add report of SVACE
* add code metrics result
* correct System Test document for completion DR
* add mention about coding style document
Signed-off-by: Roman Rusyaev <r.rusyaev@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Mon, 10 Dec 2018 08:57:41 +0000 (17:57 +0900)]
[coco] Declare coco-internal FeatureShape class (#2591)
This commit declares coco-internal FeatureShape class with "batch"
method.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Mon, 10 Dec 2018 08:15:42 +0000 (17:15 +0900)]
[coco] Remove deprecated FeatureObject::at method (#2590)
"at" method in FeatureObject was introduced when FeatureObject is
responsible to layout management, but this method is no longer used.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
윤현식/동작제어Lab(SR)/Principal Engineer/삼성전자 [Mon, 10 Dec 2018 08:04:22 +0000 (17:04 +0900)]
[enco/tfl/frontend] Remove tensor copying code (#2559)
Frontend now copies tensor copy after building IR. Duplicate code is removed from GraphBuilder.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Mon, 10 Dec 2018 08:02:51 +0000 (17:02 +0900)]
[coco] Remove deprecated FeatureObject constructor (#2588)
This commit removes deprecated FeatureObject constructor with
FeatureShape.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박세희/동작제어Lab(SR)/Principal Engineer/삼성전자 [Mon, 10 Dec 2018 08:02:45 +0000 (17:02 +0900)]
[enco] frontend/caffe: move loop invariant context (#2589)
This will move loop invariant opbuilder_context variable outside of the for loop
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/동작제어Lab(SR)/Principal Engineer/삼성전자 [Mon, 10 Dec 2018 08:02:11 +0000 (17:02 +0900)]
[enco] frontend/caffe: fix header guard directive (#2587)
This will fix input layer header guard with correct directive
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/동작제어Lab(SR)/Principal Engineer/삼성전자 [Mon, 10 Dec 2018 06:44:00 +0000 (15:44 +0900)]
[enco] caffe/frontend: const ref for layer name (#2584)
This will make to use const reference for std::string type layer name for lookup method
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Mon, 10 Dec 2018 06:43:29 +0000 (15:43 +0900)]
[nnkit] Update README.md with nnkit-run (#2581)
This commit replaces all the reference of deprecated "nni" in README.md
with "nnkit-run".
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박세희/동작제어Lab(SR)/Principal Engineer/삼성전자 [Mon, 10 Dec 2018 06:34:27 +0000 (15:34 +0900)]
[enco] caffe/frontend: rename Op to Layer (#2582)
This will rename "Op" to "Layer" for folder name and lookup parameter
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Mon, 10 Dec 2018 06:33:33 +0000 (15:33 +0900)]
[enco/tfl] Use NONE when setting activation (#2583)
This commit rewrites Concat_000 test recipe to use "NONE" instead of
explicit integer (0) to make it easy to understand.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Mon, 10 Dec 2018 06:30:17 +0000 (15:30 +0900)]
[enco/tfl] Support concat along feature height/width (#2579)
* [enco/tfl] Support concat along feature height/width
This commit rewrites enco tflite frontend to support feature map
concatenation along height/width dimensions.
This commit also includes related tests.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
* Use NONE instead of 0
* Fix a typo (compuate -> compute)
박세희/동작제어Lab(SR)/Principal Engineer/삼성전자 [Mon, 10 Dec 2018 06:15:28 +0000 (15:15 +0900)]
[enco] frontend/caffe: separate scale (#2580)
This will move Scale layer handler to separate file
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Mon, 10 Dec 2018 06:00:48 +0000 (15:00 +0900)]
[enco] Use nnkit-run instead of nni for test (#2576)
This commit replaces all the use of nni with nnkit-run in enco test
framework.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Mon, 10 Dec 2018 03:26:11 +0000 (12:26 +0900)]
[enco] Lower ConcatF to Shuffle (#2575)
This commit revises enco to lower ConcatF as Shuffle if is not yet delegated
to Android NN.
Note that the current implementation of "Intrinsic Selection" delegates
ConcatF to Android NN only when ConcatF is depth-wise.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박세희/동작제어Lab(SR)/Principal Engineer/삼성전자 [Mon, 10 Dec 2018 02:25:34 +0000 (11:25 +0900)]
[enco] frontend/caffe: separate eltwise (#2574)
This will move Eltwise(ElementWise) layer handler to separate file
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/동작제어Lab(SR)/Principal Engineer/삼성전자 [Sun, 9 Dec 2018 23:59:00 +0000 (08:59 +0900)]
[enco] frontend/caffe: separate concat (#2567)
This will move Concatenation layer handler to separate file
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/동작제어Lab(SR)/Principal Engineer/삼성전자 [Sun, 9 Dec 2018 23:58:51 +0000 (08:58 +0900)]
[tfldump] Rename file DumpOp to OpPrinter (#2566)
This will rename DumpOp header and source file to OpPrinter
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
[nnc] Add acl backend tests for ops in mobilenet and inception (#2570)
Added tests for operations:
- scale
- bias
- concat
- reshape
- dropout
- conv2d
- depthwise2d
- softmax
- pool
- fc
- relu
Signed-off-by: Efimov Alexander <a.efimov@samsung.com>
[nnc] Refactor weights conversions in Caffe importer (#2209)
* Move conversion of `BlobShape`s to `TensorVariant`s from `createMIRNodesFromLayer to individual layer conversion methods;
* Remove unnecessary type alias `IrTensor`.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
[nnc] Skeleton for acl backend unit tests (#2549)
- added two groups of tests: MIR to DOM transformations and DOM to text transformation
- Declarations of checker functions
- two example tests, one for each group
Signed-off-by: Efimov Alexander <a.efimov@samsung.com>
Ivan Vagin/AI Tools Lab /SRR/Engineer/삼성전자 [Fri, 7 Dec 2018 11:23:02 +0000 (14:23 +0300)]
[nnc] Initial implementation of caffe2_op_creator (#2333)
Initial implementation of caffe2_op_creator.
`mobilenet` supported, to support `inception` model - need to support custom paddings in pooling ops and test not tested operation conversations.
Implemented ops:
- Add
- AveragePool
- Conv
- Concat
- Dropout
- FC
- GivenTensorFill
- MaxPool
- Mul
- Relu
- Softmax
- SpatialBN
- Sum
Not tested ops:
- Add
- Concat
- Mul
- SpatialBN
Signed-off-by: Ivan Vagin <ivan.vagin@samsung.com>
[nnc] Update description of ModelIR in documentation (#2548)
Reflect recent changes in ModelIR in the documentation.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
ACL backend documentation (#2547)
The ACL backend internal structure documentation.
Signed-off-by: Timur Ablyazimov <t.ablyazimov@samsung.com>
박세희/동작제어Lab(SR)/Principal Engineer/삼성전자 [Fri, 7 Dec 2018 08:29:50 +0000 (17:29 +0900)]
[tfldump] rename DumpOp (#2565)
* [tfldump] rename DumpOp
This will rename DumpOp to OrPrinter for class and variables.
- Each inherited classes are also renamed as XxxxPrinter
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* rename variable
박세희/동작제어Lab(SR)/Principal Engineer/삼성전자 [Fri, 7 Dec 2018 08:29:26 +0000 (17:29 +0900)]
[enco] frontend/caffe: separate relu (#2562)
This will move ReLU layer handler to separate file
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박세희/동작제어Lab(SR)/Principal Engineer/삼성전자 [Fri, 7 Dec 2018 07:37:09 +0000 (16:37 +0900)]
[enco] frontend/caffe: separate pooling (#2560)
This will move Pooling layer handler to separate file.
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
박종현/동작제어Lab(SR)/Staff Engineer/삼성전자 [Fri, 7 Dec 2018 07:32:13 +0000 (16:32 +0900)]
[enco] Introduce Intrinsic Selection (#2553)
This commit introduces the basic implementation of Intrinsic Selection pass.
The current implementation is able to rewrite an "Eval" instruction for
"Depthwise concat over features" as "ANNDepthConcatF" instruction.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
박세희/동작제어Lab(SR)/Principal Engineer/삼성전자 [Fri, 7 Dec 2018 02:27:47 +0000 (11:27 +0900)]
[enco] frontend/caffe: separate convolution (#2556)
This will move Convolution layer handler to separate file.
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
윤현식/동작제어Lab(SR)/Principal Engineer/삼성전자 [Fri, 7 Dec 2018 01:48:21 +0000 (10:48 +0900)]
[enco/tfl/frontend] Function to copy tensor value into coco::Data (#2511)
* [enco/tfl/frontend] Function to copy unfilled tensor value into coco::Data
This commit adds a function to copy unfilled tensor value into coco::Data.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* pr fix: added if with continue, assert, int32 message