From c6f4ec377027b21a67061efd21b65609079f98f9 Mon Sep 17 00:00:00 2001 From: Manuel Bottini Date: Tue, 18 May 2021 18:41:56 +0100 Subject: Port CLWinogradConvolutionLayer with ClWinogradConv2d Port CLWinogradInputTransformKernel Port CLWinogradFilterTransformKernel Port CLWinogradOutputTransformKernel Resolves: COMPMID-4504 Change-Id: I3177dda0b9c2f56b36cb317027e94abe8d47229e Signed-off-by: Manuel Bottini Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5680 Reviewed-by: Georgios Pinitas Tested-by: Arm Jenkins Comments-Addressed: Arm Jenkins --- arm_compute/runtime/CL/CLFunctions.h | 1 - .../CL/functions/CLWinogradConvolutionLayer.h | 30 ++---- .../CL/functions/CLWinogradInputTransform.h | 111 --------------------- 3 files changed, 9 insertions(+), 133 deletions(-) delete mode 100644 arm_compute/runtime/CL/functions/CLWinogradInputTransform.h (limited to 'arm_compute') diff --git a/arm_compute/runtime/CL/CLFunctions.h b/arm_compute/runtime/CL/CLFunctions.h index 9d4b2fa050..62c94152e8 100644 --- a/arm_compute/runtime/CL/CLFunctions.h +++ b/arm_compute/runtime/CL/CLFunctions.h @@ -111,6 +111,5 @@ #include "arm_compute/runtime/CL/functions/CLTranspose.h" #include "arm_compute/runtime/CL/functions/CLUnstack.h" #include "arm_compute/runtime/CL/functions/CLWinogradConvolutionLayer.h" -#include "arm_compute/runtime/CL/functions/CLWinogradInputTransform.h" #endif /* ARM_COMPUTE_CLFUNCTIONS_H */ diff --git a/arm_compute/runtime/CL/functions/CLWinogradConvolutionLayer.h b/arm_compute/runtime/CL/functions/CLWinogradConvolutionLayer.h index 7b42932f82..4b351267e3 100644 --- a/arm_compute/runtime/CL/functions/CLWinogradConvolutionLayer.h +++ b/arm_compute/runtime/CL/functions/CLWinogradConvolutionLayer.h @@ -25,31 +25,29 @@ #define ARM_COMPUTE_CLWINOGRADCONVOLUTIONLAYER_H #include "arm_compute/core/Types.h" -#include "arm_compute/runtime/CL/functions/CLGEMM.h" -#include "arm_compute/runtime/CL/functions/CLWinogradInputTransform.h" #include "arm_compute/runtime/IFunction.h" +#include "arm_compute/runtime/IMemoryManager.h" + +#include namespace arm_compute { class CLCompileContext; -class CLWinogradFilterTransformKernel; -class CLWinogradOutputTransformKernel; class ICLTensor; class ITensorInfo; /** Basic function to execute Winograd-based convolution on OpenCL. This function calls the following OpenCL functions/kernels: * - * -# @ref CLWinogradInputTransform - * -# @ref CLWinogradFilterTransformKernel (only once) - * -# @ref CLGEMM - * -# @ref CLWinogradOutputTransformKernel + * -# @ref opencl::ClWinogradConv2d * */ class CLWinogradConvolutionLayer : public IFunction { public: - /** Default constructor */ + /** Default Constructor */ CLWinogradConvolutionLayer(std::shared_ptr memory_manager = nullptr); + /** Default Destructor */ + ~CLWinogradConvolutionLayer(); /** Prevent instances of this class from being copied (As this class contains pointers) */ CLWinogradConvolutionLayer(const CLWinogradConvolutionLayer &) = delete; /** Default move constructor */ @@ -58,8 +56,6 @@ public: CLWinogradConvolutionLayer &operator=(const CLWinogradConvolutionLayer &) = delete; /** Default move assignment operator */ CLWinogradConvolutionLayer &operator=(CLWinogradConvolutionLayer &&) = default; - /** Default destructor */ - ~CLWinogradConvolutionLayer(); /** Set the input and output tensors. * * Valid data layouts: @@ -136,16 +132,8 @@ public: void prepare() override; private: - MemoryGroup _memory_group; - CLGEMM _batched_mm; - CLWinogradInputTransform _input_transform; - std::unique_ptr _filter_transform; - std::unique_ptr _output_transform; - CLTensor _input0; - CLTensor _input1; - CLTensor _batched_mm_output; - const ICLTensor *_original_weights; - bool _is_prepared; + struct Impl; + std::unique_ptr _impl; }; } // namespace arm_compute #endif /* ARM_COMPUTE_CLWINOGRADCONVOLUTIONLAYER_H */ diff --git a/arm_compute/runtime/CL/functions/CLWinogradInputTransform.h b/arm_compute/runtime/CL/functions/CLWinogradInputTransform.h deleted file mode 100644 index d644591b57..0000000000 --- a/arm_compute/runtime/CL/functions/CLWinogradInputTransform.h +++ /dev/null @@ -1,111 +0,0 @@ -/* - * Copyright (c) 2018-2021 Arm Limited. - * - * SPDX-License-Identifier: MIT - * - * Permission is hereby granted, free of charge, to any person obtaining a copy - * of this software and associated documentation files (the "Software"), to - * deal in the Software without restriction, including without limitation the - * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or - * sell copies of the Software, and to permit persons to whom the Software is - * furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice shall be included in all - * copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE - * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER - * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, - * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE - * SOFTWARE. - */ -#ifndef ARM_COMPUTE_CLWINOGRADINPUTTRANSFORM_H -#define ARM_COMPUTE_CLWINOGRADINPUTTRANSFORM_H - -#include "arm_compute/core/Types.h" -#include "arm_compute/runtime/CL/ICLSimpleFunction.h" - -#include - -namespace arm_compute -{ -class CLCompileContext; -class ICLTensor; -class ITensorInfo; - -/** Basic function to execute a @ref CLWinogradInputTransformKernel. */ -class CLWinogradInputTransform : public ICLSimpleFunction -{ -public: - /** Set the input and output tensors. - * - * Valid data layouts: - * - NHWC - * - NCHW - * - * Valid data type configurations: - * |src |dst | - * |:--------------|:--------------| - * |F16 |F16 | - * |F32 |F32 | - * - * @note Winograd input transform supports the following configurations for NCWH data layout - * F(output tile, kernel size):F(2x2, 3x3), F(2x1, 3x1), F(1x2, 1x3), - * F(4x4, 3x3), F(4x1, 3x1), F(1x4, 1x3), - * F(4x4, 5x5), F(4x1, 5x1), F(1x4, 1x5) - * - * @note Winograd input transform supports the following configurations for NHWC data layout - * F(output tile, kernel size):F(4x4, 3x3), F(4x1, 3x1), F(1x4, 1x3), - * F(4x4, 5x5), F(4x1, 5x1), F(1x4, 1x5) - * - * Strides: only unit strides - * - * @param[in] input The input tensor to transform. Data types supported: F16,F32 - * @param[in] output The output tensor. The shape for this tensor can be calculated using the utility function @p compute_winograd_input_transform_shape. Data types supported: Same as @p input - * @param[in] winograd_info Contains Winograd's information described in @ref WinogradInfo. - */ - void configure(ICLTensor *input, ICLTensor *output, const WinogradInfo &winograd_info); - /** Set the input and output tensors. - * - * @note Winograd input transform supports the following configurations for NCWH data layout - * F(output tile, kernel size):F(2x2, 3x3), F(2x1, 3x1), F(1x2, 1x3), - * F(4x4, 3x3), F(4x1, 3x1), F(1x4, 1x3), - * F(4x4, 5x5), F(4x1, 5x1), F(1x4, 1x5) - * - * @note Winograd input transform supports the following configurations for NHWC data layout - * F(output tile, kernel size):F(4x4, 3x3), F(4x1, 3x1), F(1x4, 1x3), - * F(4x4, 5x5), F(4x1, 5x1), F(1x4, 1x5) - * - * Strides: only unit strides - * - * @param[in] compile_context The compile context to be used. - * @param[in] input The input tensor to transform. Data types supported: F16,F32 - * @param[in] output The output tensor. The shape for this tensor can be calculated using the utility function @p compute_winograd_input_transform_shape. Data types supported: Same as @p input - * @param[in] winograd_info Contains Winograd's information described in @ref WinogradInfo. - */ - void configure(const CLCompileContext &compile_context, ICLTensor *input, ICLTensor *output, const WinogradInfo &winograd_info); - /** Static function to check if given info will lead to a valid configuration of @ref CLWinogradInputTransform. - * - * @note Winograd input transform supports the following configurations for NCWH data layout - * F(output tile, kernel size):F(2x2, 3x3), F(2x1, 3x1), F(1x2, 1x3), - * F(4x4, 3x3), F(4x1, 3x1), F(1x4, 1x3), - * F(4x4, 5x5), F(4x1, 5x1), F(1x4, 1x5) - * - * @note Winograd input transform supports the following configurations for NHWC data layout - * F(output tile, kernel size):F(4x4, 3x3), F(4x1, 3x1), F(1x4, 1x3), - * F(4x4, 5x5), F(4x1, 5x1), F(1x4, 1x5) - * - * Strides: only unit strides - * - * @param[in] input The input tensor to transform. Data types supported: F16,F32 - * @param[in] output The output tensor. The shape for this tensor can be calculated using the utility function @p compute_winograd_input_transform_shape. Data types supported: Same as @p input - * @param[in] winograd_info Contains Winograd's information described in @ref WinogradInfo. - * - * @return a status - */ - static Status validate(const ITensorInfo *input, const ITensorInfo *output, const WinogradInfo &winograd_info); -}; -} -#endif /*ARM_COMPUTE_CLWINOGRADINPUTTRANSFORM_H */ -- cgit v1.2.1