* Update deconv primitive APIs for dilations and output_padding
Signed-off-by: Andrew Park <andrew.park@intel.com>
* Add calc_output_layouts for deconv
Signed-off-by: Andrew Park <andrew.park@intel.com>
* Add deconv shape inference TCs for ov_gpu_unit_tests
Signed-off-by: Andrew Park <andrew.park@intel.com>
* Add deconv dynamic TCs for ov_gpu_func_tests
- Disable pre_replace_deconv for dynamic shape
- Update get_expected_layout for deconv_node during reorder_input
- Add I/O swap of weights for shape infernece
Signed-off-by: Andrew Park <andrew.park@intel.com>
* Add group deconv dynamic TCs for ov_gpu_func_tests
- Update op creation for group deconv
Signed-off-by: Andrew Park <andrew.park@intel.com>
* Fix clang-format
Signed-off-by: Andrew Park <andrew.park@intel.com>
* Update test params with swapped I/O dimensions for ov_gpu_unit_tests
Signed-off-by: Andrew Park <andrew.park@intel.com>
* Update for optional output_shape const/param input support
Signed-off-by: Andrew Park <andrew.park@intel.com>
* Update deconv/group deconv dynamic TCs w/ output_shape for ov_gpu_func_tests
Signed-off-by: Andrew Park <andrew.park@intel.com>
* Update deconv/group deconv shape inference TCs w/ output_shape for ov_gpu_unit_tests
Signed-off-by: Andrew Park <andrew.park@intel.com>
* Apply code review
Signed-off-by: Andrew Park <andrew.park@intel.com>
Signed-off-by: Andrew Park <andrew.park@intel.com>
* Bugfix build issue of onednn test
+ Remove invalid referencing if ENABLE_INTEL_CPU is off
+ Removed unnecessary include
Signed-off-by: Min, Byungil <byungil.min@intel.com>
* Add strided slice dynamic TCs for gpuFunctests
Signed-off-by: Andrew Park <andrew.park@intel.com>
* Update StridedSliceLayerGPUTest for rest inputs's type (parameter)
Signed-off-by: Andrew Park <andrew.park@intel.com>
* remove W/A for strided slice
Signed-off-by: Andrew Park <andrew.park@intel.com>
* primitive API updates for dynamic begin/end/stride inputs support
Signed-off-by: Andrew Park <andrew.park@intel.com>
* Update ocl impl and kernel param, and strided_slice_ref kernel w/o memory_deps
Signed-off-by: Andrew Park <andrew.park@intel.com>
* Update TCs for new API w/ vectors on strided_slice_gpu unittests
Signed-off-by: Andrew Park <andrew.park@intel.com>
* Apply code review
Signed-off-by: Andrew Park <andrew.park@intel.com>
* Apply padding when the size of rest input data is smaller than actual input rank
Signed-off-by: Andrew Park <andrew.park@intel.com>
* Update strided_slice_optimize pass to run onyl when all deps are constant
Signed-off-by: Andrew Park <andrew.park@intel.com>
Signed-off-by: Andrew Park <andrew.park@intel.com>
* fix strided_slice_optimize for 5d new axis output, typo in strided_slice cl kernal, avoid handle_reshape when user is only one
* fix unit test of resample_eltwise_fusing_through/12
* keep Shape subgraph in FP32 for GPU
* replaced ngraph:: -> ov:: namespace
* renamed ConvertModelToFP16ElementType -> ConvertCompressedToMixedPrecision and other passes; removed unnecessary argument for the pass
Signed-off-by: Pavel Esir <pavel.esir@intel.com>
* Update src/common/transformations/src/transformations/common_optimizations/convert_compressed_to_mixed_precision.cpp
* Update src/common/transformations/src/transformations/common_optimizations/convert_compressed_to_mixed_precision.cpp
* rephrase calling of ConvertPrecision to avoid error of an unused variable
* Update convert_compressed_to_mixed_precision.cpp
* placed calling of ConvertCompressedToMixedPrecision into the beginning of CommonOptimizations
* Update transformations_pipeline.cpp
* refactored unit-tests
* cleared obsolete commented line
* code style fix
* Update src/common/transformations/src/transformations/common_optimizations/common_optimizations.cpp
Co-authored-by: Maxim Vafin <maxim.vafin@intel.com>
Signed-off-by: Pavel Esir <pavel.esir@intel.com>
Co-authored-by: Roman Kazantsev <roman.kazantsev@intel.com>
Co-authored-by: Maxim Vafin <maxim.vafin@intel.com>
* Quantize clamp optimization(prepare_quantization.cpp:227) for cldnn is not proper to onednn.
* cast intermediate variable to output type after ~
* cldnn: every fused quantization.
* onednn: all post-ops are processed.
* A little perf regression expected in onednn because clamp is relatively heavier operation than type casting.
* Remove deconv workaround from PR #14139
* Fixed shape infer dep for blocked format
* Set preferred format for shape_of to be made from input rank
- ShapeOf should get the original output layout of the previous node, which is not reorderd.
* updated to get ockable memory for device inputs
* gpuFuncTest to set f64 scalar tensor
* aligned type names
* updated to get lockable memory for some specific precisions
* add a function to get a supported precision
Vladimir Dudnik
Taylor Yeonbok Lee
Vladimir Paramuzov
Andrew Kwangwoong Park
Steve Yoo
Ilya Lavrenov
Min, Byungil
Jade Cho
Wilson Seok
Felix Dohyun Kim
Sergey Shlyapnikov
Wang Wangwangandyanlan song
Eddy Kim
Andrei Gorbachev
Sungeun Kim