The key changes are:
* Using VERBATIM to ensure CMake property passes command-line arguments
to child processes.
* Using the INCLUDE_DIRECTORIES property instead of COMPILE_FLAGS to add
include directories, because COMPILE_FLAGS are treated as space-separated
values. (A small side benefit is that this doesn't rely on -I being
the include directory option.)
In addition, some changes had to be made in order to preserve behavior:
* The _GEN_ARGS_LIST variable has to be inlined, because ARCH_SET is a list,
and therefore the "-DXARCH_SET=..." argument gets split into multiple arguments
(this happens to work by coincidence without VERBATIM). IMO, the code looks
better this way anyway.
* It's no longer necessary to replace spaces in XARCH_SET in
cross_compiled_disp_gen.cmake, because those spaces were an artifact of how
the CMake arguments were passed before.
* Fixed framework name attribute for onnx, mxnet.
* Fixed framework name attribute for caffe.
* Removed unnecessary attribute setting from add_opoutput()
* Added identity nodes adding to outputs in mxnet loader.
* Removed unnecessary reformat.
* Removed unnecessary reformat.
* Added check for empty name.
* Used nodes indices instead of node names in loader.
* Code refactoring, small bug fixed.
* refactor: update ie python samples
* python samples: change comment about infer request creation (step 5)
* python sample: add the ability to run object_detection_sample_ssd.py with a model with 2 outputs
* Add batch size usage to python style transfer sample
* Change comment about model reading
* Add output queue to classification async sample
* add reshape for output to catch results with more than 2 dimensions (classification samples)
* Set a log output stream to stdout to pass the hello query device test
* Add comments to the hello query device sample
* Set sys.stdout as a logging stream for all python IE samples
* Add batch size usage to ngraph_function_creation_sample
* Return the ability to read an image from a ubyte file
* Add few comments and function docstrings
* Restore IE python classification samples output
* Add --original_size arg for python style transfer sample
* Change log message to pass tests (object detection ie python sample)
* Return python shebang
* Add comment about a probs array sorting using np.argsort
* [GNA] Add low precision support for some layers (affine, eltwise, ReLU, sigmoid)
* [GNA] Fix convolution and scaleshift regression fail
* [GNA] Fix const scale factor calc
* [GNA] Fix extra whitespace
* [GNA] Eltwise padding and low precision primitive creation support
* [GNA] Changes after review
* [GNA] Remove INPUT_PRECISION GNA flag
Remove INPUT_PRECISION flag, all references to it
and tests using it. Other minor stylistic changes and cleanup.
* Convolution: Enhance dynamic shape inference of validate and infer types method
* Convolution: Change onnx test with dynamic shapes to float element type
* Convolution: Remove test instances with integer precision
* Convolution: Add backticks to types in spec
* Convolution: Change element type variable for output element type
* GroupConvolution: Add backticks to types in spec
* GroupConvolution: Enhance dynamic shape inference of validate and infer types method
* GroupConvolution: Remove serialization test instances with integer precision
* GroupConvolutionBackpropData: Remove serialization test instances with integer precision
* GroupConvolutionBackpropData: Enhance dynamic shape inference of validate and infer types method
* Convolution: Add helper function to validate convolution parameters in ref impl
* Convolution: Rewrite lambda to capture spatial dims of filters in validate and infer types
* GroupConvolution: Refactor reference implementation
* Remove call to old implementation of convolution using dilations
* Added validation method to validate shapes
* GroupConvolutionBackpropData: Add more type_prop unit test and refactor test names
* Convolution: Extended validation of convolution parameters in reference implementation
* GroupConvolution: Extended validation of group convolution parameters in reference implementation
* GroupConvolutionBackpropData: Add helper function to validate convolution backprop parameters in ref impl
* Clean up unnecessary lines
* BinaryConvolution: Use validate helper function from convolution ref impl
* Convolution: Refactor validate and infer types to improve readability
* BinaryConvolution: Refactor validate and infer types to improve readability
* Convolution: Add explicit tensor shape dims for inputs and outputs in spec
* BinaryConvolution: Add explicit tensor shape dims for inputs and outputs in spec
* GroupConvolution: Add explicit tensor shape dims for inputs and outputs in spec
* Add helper function to infer convolution forward output shape
* Convolution: Refactor validate and infer types to use helpers to infer output shape
* BinaryConvolution: Refactor validate and infer types to use helpers to infer output shape
* GroupConvolutionBackpropData: Fix formula to calculate output shape in validation functions
* Remove symbol to export convolution output shape inference function
* GroupConvolution: Add validation checks for input channels dim of data batch and filter shape
* GroupConvolutionBackpropData: clean up type prop tests
* Convolution: Change element type in onnx unit tests with dyn shapes and convolution nodes
* GroupConvolutionBackpropData: Correct layout of filters input
* GroupConvolution: Deduce groups from inputs shape during output shape inference
* Change spec supported types of convolution operations to any numeric type
* Revert "GroupConvolution: Remove serialization test instances with integer precision"
This reverts commit 781c2570d6.
* Revert "GroupConvolutionBackpropData: Remove serialization test instances with integer precision"
This reverts commit 9a6ac23968.
* Revert "Convolution: Remove test instances with integer precision"
This reverts commit 0b07052a62.
* Revert "Convolution: Change element type in onnx unit tests with dyn shapes and convolution nodes"
This reverts commit c9f5944b6b.
* Revert "Convolution: Change onnx test with dynamic shapes to float element type"
This reverts commit 1f4202b010.
* Allow integral types in validate and infer types method for convolution group of operations
* Add i32 precision in single layer tests for convolution group of operations
* BinaryConvolution: Fix shape of input and output tensors in spec
* Address nitpick comments
* Utils: make_try_fold, clone_try_fold. Template node creation and attempt to fold it
* RTTI for ArithmeticReduction(KeepDims)
* Enriched ngraph::get_default_order overloads with ones for dynamic shape and rank
* [ Transpose sinking ] Transpose->FQ->Reduce to FQ->Reduce->Transpose
* Style: deleted empty line
* RTTI in Reduction operations
* RTTI for LogicalReductionKeepDims
* Transpose: optimizations moved from algebraic simplification to TransposeSinking
* renamed file
* Fix test
* keep_dims is always initialized
* Apply suggestions from code review
Co-authored-by: Gleb Kazantaev <gleb.nnstu@gmail.com>
Co-authored-by: Gleb Kazantaev <gleb.nnstu@gmail.com>
* clean the FusedOp from mod operation
* add backend and type_prop tests for mod operator
* convert taking autobrodcast to match binary elementwise arithmetic ops
* add type_prop/mod.cpp to CMakeLists.txt
* fix style
* fix style v2
* remove evaluate method and add backend test for negative numbers
* add copyright for type_prop/mod.cpp
