[tf2tflite] enable Const_000 (#6598)

This will enable Const_000 that was fixed in exo-tflite

Signed-off-by: SaeHie Park <saehie.park@samsung.com>

[moco-tf] GraphBuilder for TFRsqrt (#6579)

* [moco-tf] GraphBuilder for TFRsqrt

This will introduce GraphBuilder for TFRsqrt to import Rsqrt node

Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* rename var

[loco/Service] FixedReshape type inference (#6595)

This commit introduces type inference of FixedReshape

Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>

[locoex/customop] ShapeInference rule for custom op (#6585)

* [locoex/customop] ShapeInference rule for custom op

This commit adds COpShapeInference rule, which is shape inferring rule for custom op.

Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* add TODO

* wrong comment

* remove 'canonical' from comment

* wrong comment again: Type -> Shape

* wrong comment again and again... COP -> COp

[exo-exporter] Remove unused header (#6596)

Let's remove unused header in OperationExporter.cpp

Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>

[moco-tf] Introduce ResolveRedundantReshape transform (#6571)

* [moco-tf] Introduce ResolveRedundantReshape transform

This commit introduces ResolveRedundantReshape transform of canonical
optimize stage.

Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
* Use const arg

* Review fix: fine grained logic

[moco-tf] Introduce TFSquaredDifference IR (#6580)

This will introduce TFSquaredDifference IR and required minimum changes

Signed-off-by: SaeHie Park <saehie.park@samsung.com>

[exo-tflite] Fix buffer export with adding empty (#6576)

* [exo-tflite] Fix buffer export with adding empty

This will add one empty buffer as first buffer to follow TF lite behavior
- This makes unit Const_000 test pass

Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* remove test list change

[moco-tf] Fix missing changed tag (#6583)

This commit fixes missing 'changed' setting on ResolveDuplicateReshape
transform

Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>

[moco-tf] Comment for TFRsqrt IR (#6578)

This will add comment for TFRsqrt IR

Signed-off-by: SaeHie Park <saehie.park@samsung.com>

[locomotiv] EltwiseSqrt (#6532)

* [locomotiv] EltwiseSqrt

This will introduce execution code for EltwiseSqrt node

Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* add test

* update comment

[custom op] api addition (#6219)

* [custom op] Api addition to support custom op registration

Add nnfw_dev.h header with types required to define custom operations

Signed-off-by: Vladimir Plazun <v.plazun@samsung.com>
* Roll back changes to nnfw.h, add separate header for custom kernel development

* format fix

* use limited dim count for typeinfo struct

* remove _t suffix for types

* fix missed _t suffixes for function types

[loco] FixedReshape graph testcase (#6545)

This commit adds graph builder layer and testcase for FixedReshape

Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>

[exo-tflite] Export FixedReshape operation (#6565)

This commit introduces export of FixedReshape operation

Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>

[res] Introduce Softmax testcases (#6562)

* [res] Introduce Softmax testcases

This commit will introduce `Softmax` testcases

Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
* remove versions

[xtrace] Hoist loop invariant code (#6566)

Hoist loop invariant code in `benchmark_runner.cc` for readability.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>

Make acl_cl KernelGenerator supports NCHW frontend layout (#6552)

This commit makes acl_cl KernelGenerator supports NCHW frontend layout.

Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>

Make acl_neon KernelGenerator supports NCHW frontend layout (#6551)

This commit makes acl_neon KernelGenerator supports NCHW frontend layout.

Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>

[loco] Introduce services for TensorSoftmax (#6564)

This commit will introduce loco services for `TensorSoftmax`

Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>

[moco-tf] Clear ConcatV2 graph builder (#6547)

This will erase ConcatV2 graph builder implementation to reduce complexity of diff for migration to using VariadicArityNode

Signed-off-by: SaeHie Park <saehie.park@samsung.com>

[moco-tf] Introduce TFSqrt IR (#6553)

This will introduce TFSqrt IR to represent TensorFlow Sqrt node and required minimum implementation

Signed-off-by: SaeHie Park <saehie.park@samsung.com>

[mir_onnx] Remove uses of DropoutOp (#6560)

This operation is a no-op.

Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>

[mir2loco] Support Softmax operation in transformer (#6575)

* Added SoftmaxOp convertion and test for this

Signed-off-by: Pavel Iliutchenko <p.iliutchenk@samsung.com>

[mir2loco] Support Mul operation in transformer (#6573)

* Added MulOp convertion and test for this

Signed-off-by: Pavel Iliutchenko <p.iliutchenk@samsung.com>

[mir_onnx] Fix output name for softbackend (#6572)

* Setting empty output node name

Signed-off-by: Pavel Iliutchenko <p.iliutchenk@samsung.com>

[mir] Introduce data format (#6570)

* Add `DataFormat` enum class.
* Use it in convolutional and pooling operations.

Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>

[mir_caffe2] Remove uses of DropoutOp (#6561)

This operation is a no-op.

Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>

[mir_caffe] Remove uses of DropoutOp (#6558)

This operation is a no-op.

Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>

[nnc] Remove uses of DropoutOp (#6556)

This operation is a no-op.

Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>

[nnc] Eliminate IVisitor in nnc (#6519)

Eliminate abstract IVisitor uses in nnc componet and use Visitor instead

Signed-off-by: Efimov Alexander <a.efimov@samsung.com>

Don't build nnpackage_run if neurun is not built (#6557)

nnpackage_run is for neurun.
pacl should not try to build nnpackage_run.

Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>

[loco] Introduce DepthwiseFilterEncode to CanonicalShapeInferenceRule (#6548)

This commit introduces DepthwiseFilterEncode to
CanonicalShapeInferenceRule.

Signed-off-by: seongwoo <sw4670.chae@samsung.com>

[loco/Service] Forward node shape inference (#6542)

This commit implements canonical shape inference for Forward node

Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>

[neurun, nnpackage] Introduce nnpackage_run (#5720)

Initial version of nnpackage.
It supports only `run` with random values.
( store, compare, input are not supported yet. )

Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>

Make acl_cl KernelGenerator supports NCHW frontend layout (#6544)

This commit makes acl_cl KernelGenerator supports NCHW frontend layout.

Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>

[loco] Remove "shape" setter from Push class (#6527)

This commit removes deprecated "shape" setter from Push class.

Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>

[moco-tf] Clear ConcatV2 Canonicalizer (#6521)

This will erase ConcatV2 Canonicalier implementation to reduce complexcity of diff for migration to using VariadicArityNode

Signed-off-by: SaeHie Park <saehie.park@samsung.com>

[moco-tf] Turning on the knob for ReorderDecodeTensorBiasAdd (#6535)

This commit turns on the knob for ReorderDecodeTensorBiasAdd

Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>

[exo-tflite] Introduce DepthwiseFilterEncode with exporting as reshape (#6524)

* [exo-tflite] Introduce DepthwiseFilterEncode with exporting as reshape

This commit introcude DepthwiseFilterEncode with exporting as reshape to
exo-tflite.

Signed-off-by: seongwoo <sw4670.chae@samsung.com>
* remove unused function.

* add some comment.

[moco-tf] Use official FixedReshape alias (#6541)

Now loco::FixedReshape alias is official, so this commit starts to use
the alias in moco-tf module.

Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>

[res] TF_SMALL_NET_0015 with 1.0 / Sqrt (#6539)

This will introduce TF_SMALL_NET_0015 model with Sqrt+RealDiv that works like Rsqrt

Signed-off-by: SaeHie Park <saehie.park@samsung.com>

[neurun] Implement nnfw_input/output_tensorinfo (#6530)

nnfw_input_tensorinfo and nnfw_output_tensorinfo are implemented.

Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>

Apply layout for helper functions converting axis values (#6528)

This commit applies layout for helper functions converting axis values.

Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>

[moco-tf] Clear ConcatV2 fix shape (#6523)

This will erase ConcatV2 fix shape to reduce complexity of diff for migration to using VariadicArityNode

Signed-off-by: SaeHie Park <saehie.park@samsung.com>

[mocotest-tf] Test script runnable from any directory (#6529)

Previously test script partially assumes it runs on WORKDIR, which makes
no harm at all. This commit just explicitly call file path including
WORKDIR to be matched with other lines convention and to be runnable from
any directory.

Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>

[loco] Inference service for EltwiseSqrt (#6526)

This will add Shape and Type inference for EltwiseSqrt

Signed-off-by: SaeHie Park <saehie.park@samsung.com>

[neurun] Implement nnfw_input_size and nnfw_output_size (#6495)

* [neurun] Implement nnfw_input_size and nnfw_output_size

This patch implements nnfw_input_size and nnfw_output_size.

Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
* [neurun] Update int to uint32_t for input/output_size

This patch updates int to uint32_t for input/output_size.

Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>

Set lcov's gcov-tool param (#6525)

For cross build, set lcov's gcov-tool by GCOV_PATH environment variable. Default is gcov (native build).
nnfw's gen-coverage-report has same feature

Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>

[moco-tf] Clear ConcatV2 import test (#6522)

This will erase ConcatV2 TF dialect import test to reduce complexity of diff for migration to using VariadicArityNode

Signed-off-by: SaeHie Park <saehie.park@samsung.com>

[tf2tflite] Test script runnable from any directory (#6514)

Previously test script partially assumes it runs on WORKDIR, which makes
no harm at all. This commit just explicitly call file path including WORKDIR
to be matched with other lines convention and to be runnable from any directory.

Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>

[loco] Using FixedReshape (#6496)

This commit introduces alias for Reshape<ReshapeType::Fixed> loco node.

Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>

[loco] Introduce EltwiseSqrt (#6520)

This will introduce IR for elementwise Sqrt

Signed-off-by: SaeHie Park <saehie.park@samsung.com>

[mlapse-tfl] Implement a command-line argument parser (#6512)

This commit implements a command-line argument parse for mlapse-tfl.

Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>

Fix deadlock when access two acl_cl at once (#6500)

To avoid deadlock when operation need to access two acl_cl at once (source, target), introduce access function to access two tensor

Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>

[moco-tf] Introduce VariadicArityNode (#6472)

This will introduce TF dialect VariadicArityNode to manage Concat node

Signed-off-by: SaeHie Park <saehie.park@samsung.com>

[locomotiv] supporting FeatureBiasAdd (#6504)

This code enables locomotiv to support FeatureBiasAdd.

Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>

[moco-tf] Fix FuseBinaryIntoProceeding logic (#6511)

This commit fixes FuseBinaryIntoProceeding logic. Assertion was
misplace, so that normal addition blocked to this assertion.

Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>

[mir2loco] Remove Push node's shape setter (#6487)

This commit revises all the code that invokes deprecated shape setter.

Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>

[mir2loco] Remove unused visitors and use default visit_fallback (#6515)

* Remove unused visitors
* Override visit_fallback

Signed-off-by: Pavel Iliutchenko <p.iliutchenk@samsung.com>

[mir_caffe] Refactor importing of BatchNorm (#6467)

Structurize the code, fix comments.

Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>

[mir2loco] Supported other types in ConstantOp transformer (#6516)

* Supported F32, F64, S32, S64 data types
* Temporary implementations for F64 and S64

Signed-off-by: Pavel Iliutchenko <p.iliutchenk@samsung.com>

[locop] Ignore the shape attribute for Push nodes (#6485)

CanonicalNodeSummaryBuilder now ignores (deprecated) tensor shape
attributes when it generates a node summary for Push nodes.

Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>

[loco] Introduce Softmax node in loco (#6492)

This commit will introduce `Softmax` node in `loco`

Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>

[tf2tflite] Loading customop.conf file into ModelSignature (#6406)

* [tf2tflite] Loading customop.conf file into ModelSignature

Adds code for loading customop.conf into ModelSignature

Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* 1) class to function 2) remove loco from cmake file 3) remove some headers

* fix wrong lambda comment // use argv[5] directly

[neurun] Do not collect TensorBuilders one by one (#6458)

* [neurun] Do not collect TensorBuilders one by one

To get the full set of tensor builders currently using it used to
collect from all the `LowerInfo`s. Now that `BackendResolver` manages
them all we can just use it.
Allow allocation of 0-byte object

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>

[exo-tflite] support DetphwiseFilterShape in shape_desc (#6491)

* [exo-tflite] support DetphwiseFilterShape in shape_desc

This commit supports DethwiseFilterShape in shape_desc.

Signed-off-by: seongwoo <sw4670.chae@samsung.com>
* apply comment.

[neurun] change int to uint32_t for size of input/output (#6502)

neurun uses uint32_t internally. Runtime API uses int.
This patch will change Runtime API to use uint32_t.
(same type to `neurun` internal).

Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>

[moco-tf] Clean up trivials in FuseBinaryIntoPreceding (#6497)

This commit will clean up some codes
- Remove deprecated function
- Replace and correct comments
- Modify bracket formats

Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>

[moco-tf] Introduce ResolveDuplicateReshape (#6499)

ResolvedDuplicateReshape transform introduced. This transform finds
duplicate FixedReshape pair, and let the later one to bypass the
former.

Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>

[mocotest-tf] Enable TF_SMALL_NET_0004 (#6498)

Since `FuseBinaryIntoPreceding` now supports for `DepthwiseConv2dNative`,
this commit will enable `TF_SMALL_NET_0004`

Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>

[loco] fix a window_of function in CanonicalShapeInferenceRule (#6489)

This commit fix a window_of function.

Related issue:#6486

Signed-off-by: seongwoo <sw4670.chae@samsung.com>

Use nnfw_coverage for gcov build (#6445)

For gcov build release target, use nnfw_coverage link

Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>

Replace coverage report command (#6476)

- Replace nncc's coverage report command into infra/command
- Introduce build.configuration setting
  - Default workspace: build/

Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>

[neurun] Use nnfw api wrapper's return value (#6493)

Currently nnfw implementation returns no error.
Now, it will use the return value of wrapper.

Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>

Remove index parameter from nnfw_input_size (#6488)

index parameter is removed that is not necessary.

Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>

[moco-tf] Support DepthwiseConv2dNative in FuseBinaryIntoPreceding (#6490)

This commit will enable supporting `DepthwiseConv2dNative` in `FuseBinaryIntoPreceding`

Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>

[neurun] Remove class `Plan` (#6450)

`Plan` is nothing more than a wrapping `FunctionSequence` so it is
redundant. It has been replaced with `FunctionSequence`.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>

Print build error on TensorFlowVersionChecker build error (#6477)

TensorFlowConfig.cmake checks running result only.

When prebuilt libtensorflow library is built from Ubuntu 18.04,
 and nncc is going to be built on Ubuntu 16.04, it says:

"you need tensorflow version 1.12.0".

This patch will check build error also.
Then, it prints error message to guide the developers
to use libtensorflow built on same version of Ubuntu.

Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>

Disable downlaoding NNPACK on aarch64-tizen (#6438)

- DOWNLOAD_NNPACK=OFF on aarch64 tizen build

Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>

[moco-tf] Do not reset input for BiasAdd Canonicalizer (#6470)

This will remove reset input for old replaced BiasAdd node in Canonicalization and update the diagram

Signed-off-by: SaeHie Park <saehie.park@samsung.com>

[moco-tf] Do not reset input for Conv2D Canonicalizier (#6471)

This will remove reset input for old replaced Conv2D node in Canonicalization and update the diagram

Signed-off-by: SaeHie Park <saehie.park@samsung.com>

Introduce mlapse toolkit (#6441)

This commit introduces the inital version of mlapse toolkit.

Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>

Fix build error (#6484)

- add missing semicolon on nnfw.h

Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>

API for Input/Output Tensor information (#6153)

* API for Input/Output Tensor information

- Add nnfw_tensor_info
- input/output size getter

Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
* Fix typo, tensor_info -> tensorinfo

[loco] Simplify FeatureCodec Graph Testcase (#6447)

This commit simplifies FeatureCodec Graph Testcase using
GraphBuilder helper.

Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>

[locop] Remove "opcode" helper (#6449)

Only "BuiltinNodeSummaryBuilder" uses this helper currently.

Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>

[mocotest-tf] Tests enabled by ResolveConstantShape transform (#6435)

This commit enables two tests enabled by ResolveConstantShape transform
- TF_SMALL_NET_0013 : Shape test
- TF_SMALL_NET_0014 : Mock of the Epilogue

Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>

[neurun] Break NN API and core API tests (#6448)

"test_neurun" currently includes both NN API and neurun core API tests.

This commit introduces "test_neurun_frontend_nnapi" dedicated to NN API tests.

Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>

[neurun] Remove empty override in ParamChecker (#6454)

Remove empty override(AddNode) in ParamChecker.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>

Fix nnas command build-docker-image (#6451)

`./nnas build-docker-image` did not work due to wrong path. This commit
fixes it. `${NNAS_SCRIPT_PATH}` is undefined.

Signed-off-by: Hanjoung Lee <hanjoung.lee@samsung.com>

[tf2tflite] enable test of fusion of binary op (#6444)

This will enable test of fusion with binary op with Conv2D, BiasAdd

Signed-off-by: SaeHie Park <saehie.park@samsung.com>

[mir_onnx] Refactor importing of BatchNormalization (#6466)

Structurize the code, add comments.

Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>

[mir] Remove BatchNorm op (#6465)

Remove BatchNorm operation which has never been used.

Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>

[mir_onnx] Remove pair return from get attributes functions (#6359)

* Create default values for attributes
* Fix all converters which using this functions

Signed-off-by: Pavel Iliutchenko <p.iliutchenk@samsung.com>

[mir2loco] Remove unused BatchNorm op (#6463)

It will be reimplemented taking into account different variations between frameworks.

Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>

[mir_caffe2] Remove unused BatchNorm op (#6462)

It will be reimplemented taking into accout different variations between frameworks.

Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>

[mir_caffe] Remove unused BatchNorm op (#6461)

It will be reimplemented taking into account different variations between frameworks.

Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>

[nnc] Remove unused BatchNorm op (#6460)

It will be reimplemented taking into account different variations between frameworks.

Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>

[nnc] Reformat using .clang-format (#6455)

Reformat sources to common style.

Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>

[mir] Delete ElementwiseOp (#6418)

It was replaced by several binary elementwise operations.

Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>