diff options
author | Michele Di Giorgio <michele.digiorgio@arm.com> | 2018-11-01 13:44:05 +0000 |
---|---|---|
committer | Georgios Pinitas <georgios.pinitas@arm.com> | 2018-11-13 14:49:34 +0000 |
commit | 27400b90a9cb3fe028c5b724b58ce0e82d89b5e8 (patch) | |
tree | 4b7dd9d4b29653ada018172dae826fe3e6ef5e08 /arm_compute | |
parent | bb081cac4f386eb6db6e9927fce27c7027dd7be5 (diff) | |
download | ComputeLibrary-27400b90a9cb3fe028c5b724b58ce0e82d89b5e8.tar.gz |
COMPMID-1707: Create 3 special CLWidthConcatenate kernel to concatenate 2/4 and 8 tensors (Part 1)
Creating special cases for concatening 2 and 4 tensors.
Change-Id: I6a739a494ae45011acb65369e353f9ef96970b90
Diffstat (limited to 'arm_compute')
4 files changed, 173 insertions, 1 deletions
diff --git a/arm_compute/core/CL/CLKernels.h b/arm_compute/core/CL/CLKernels.h index 36abb7bd78..df76366a4b 100644 --- a/arm_compute/core/CL/CLKernels.h +++ b/arm_compute/core/CL/CLKernels.h @@ -127,6 +127,8 @@ #include "arm_compute/core/CL/kernels/CLWarpAffineKernel.h" #include "arm_compute/core/CL/kernels/CLWarpPerspectiveKernel.h" #include "arm_compute/core/CL/kernels/CLWeightsReshapeKernel.h" +#include "arm_compute/core/CL/kernels/CLWidthConcatenate2TensorsKernel.h" +#include "arm_compute/core/CL/kernels/CLWidthConcatenate4TensorsKernel.h" #include "arm_compute/core/CL/kernels/CLWidthConcatenateLayerKernel.h" #include "arm_compute/core/CL/kernels/CLWinogradFilterTransformKernel.h" #include "arm_compute/core/CL/kernels/CLWinogradInputTransformKernel.h" diff --git a/arm_compute/core/CL/kernels/CLWidthConcatenate2TensorsKernel.h b/arm_compute/core/CL/kernels/CLWidthConcatenate2TensorsKernel.h new file mode 100644 index 0000000000..cc2eaa25f2 --- /dev/null +++ b/arm_compute/core/CL/kernels/CLWidthConcatenate2TensorsKernel.h @@ -0,0 +1,79 @@ +/* + * Copyright (c) 2018 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_CLWIDTHCONCATENATE_2TENSORS_KERNEL_H__ +#define __ARM_COMPUTE_CLWIDTHCONCATENATE_2TENSORS_KERNEL_H__ + +#include "arm_compute/core/CL/ICLKernel.h" +#include "arm_compute/core/Types.h" + +namespace arm_compute +{ +class ICLTensor; + +/** Interface for the width concatenate kernel of 2 tensors. + * The input1 and input2 tensors will be concatenated into the output tensor. + */ +class CLWidthConcatenate2TensorsKernel : public ICLKernel +{ +public: + /** Default constructor */ + CLWidthConcatenate2TensorsKernel(); + /** Prevent instances of this class from being copied (As this class contains pointers) */ + CLWidthConcatenate2TensorsKernel(const CLWidthConcatenate2TensorsKernel &) = delete; + /** Prevent instances of this class from being copied (As this class contains pointers) */ + CLWidthConcatenate2TensorsKernel &operator=(const CLWidthConcatenate2TensorsKernel &) = delete; + /** Allow instances of this class to be moved */ + CLWidthConcatenate2TensorsKernel(CLWidthConcatenate2TensorsKernel &&) = default; + /** Allow instances of this class to be moved */ + CLWidthConcatenate2TensorsKernel &operator=(CLWidthConcatenate2TensorsKernel &&) = default; + /** Default destructor */ + ~CLWidthConcatenate2TensorsKernel() = default; + /** Initialise the kernel's input1s and output + * + * @param[in] input1 First input tensor. Data types supported: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32 + * @param[in] input2 Second input tensor. Data types supported: same as @p input1 + * @param[out] output Output tensor. Data types supported: Same as @p input1. + */ + void configure(const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output); + /** Static function to check if given info will lead to a valid configuration of @ref CLWidthConcatenate2TensorsKernel + * + * @param[in] input1 First tensor info. Data types supported: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32 + * @param[in] input2 Second tensor info. Data types supported: same as @p input1 + * @param[in] output Output tensor info. Data types supported: Same as @p input1. + * + * @return a status + */ + static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output); + + // Inherited methods overridden: + void run(const Window &window, cl::CommandQueue &queue) override; + +private: + const ICLTensor *_input1; + const ICLTensor *_input2; + ICLTensor *_output; +}; +} // namespace arm_compute +#endif /* __ARM_COMPUTE_CLWIDTHCONCATENATE_2TENSORS_KERNEL_H__ */ diff --git a/arm_compute/core/CL/kernels/CLWidthConcatenate4TensorsKernel.h b/arm_compute/core/CL/kernels/CLWidthConcatenate4TensorsKernel.h new file mode 100644 index 0000000000..952fd99beb --- /dev/null +++ b/arm_compute/core/CL/kernels/CLWidthConcatenate4TensorsKernel.h @@ -0,0 +1,85 @@ +/* + * Copyright (c) 2018 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_CLWIDTHCONCATENATE_4TENSORS_KERNEL_H__ +#define __ARM_COMPUTE_CLWIDTHCONCATENATE_4TENSORS_KERNEL_H__ + +#include "arm_compute/core/CL/ICLKernel.h" +#include "arm_compute/core/Types.h" + +namespace arm_compute +{ +class ICLTensor; + +/** Interface for the width concatenate kernel of 4 tensors. + * All input tensors will be concatenated into the output tensor. + */ +class CLWidthConcatenate4TensorsKernel : public ICLKernel +{ +public: + /** Default constructor */ + CLWidthConcatenate4TensorsKernel(); + /** Prevent instances of this class from being copied (As this class contains pointers) */ + CLWidthConcatenate4TensorsKernel(const CLWidthConcatenate4TensorsKernel &) = delete; + /** Prevent instances of this class from being copied (As this class contains pointers) */ + CLWidthConcatenate4TensorsKernel &operator=(const CLWidthConcatenate4TensorsKernel &) = delete; + /** Allow instances of this class to be moved */ + CLWidthConcatenate4TensorsKernel(CLWidthConcatenate4TensorsKernel &&) = default; + /** Allow instances of this class to be moved */ + CLWidthConcatenate4TensorsKernel &operator=(CLWidthConcatenate4TensorsKernel &&) = default; + /** Default destructor */ + ~CLWidthConcatenate4TensorsKernel() = default; + /** Initialise the kernel's input1s and output + * + * @param[in] input1 First input tensor. Data types supported: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32 + * @param[in] input2 Second input tensor. Data types supported: same as @p input1 + * @param[in] input3 Third input tensor. Data types supported: same as @p input1 + * @param[in] input4 Fourth input tensor. Data types supported: same as @p input1 + * @param[out] output Output tensor. Data types supported: Same as @p input1. + */ + void configure(const ICLTensor *input1, const ICLTensor *input2, const ICLTensor *input3, const ICLTensor *input4, ICLTensor *output); + /** Static function to check if given info will lead to a valid configuration of @ref CLWidthConcatenate4TensorsKernel + * + * @param[in] input1 First tensor info. Data types supported: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32 + * @param[in] input2 Second tensor info. Data types supported: same as @p input1 + * @param[in] input3 Third tensor info. Data types supported: same as @p input1 + * @param[in] input4 Fourth tensor info. Data types supported: same as @p input1 + * @param[in] output Output tensor info. Data types supported: Same as @p input1. + * + * @return a status + */ + static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *input3, const ITensorInfo *input4, const ITensorInfo *output); + + // Inherited methods overridden: + void run(const Window &window, cl::CommandQueue &queue) override; + +private: + const ICLTensor *_input1; + const ICLTensor *_input2; + const ICLTensor *_input3; + const ICLTensor *_input4; + ICLTensor *_output; +}; +} // namespace arm_compute +#endif /* __ARM_COMPUTE_CLWIDTHCONCATENATE_4TENSORS_KERNEL_H__ */ diff --git a/arm_compute/runtime/CL/functions/CLWidthConcatenateLayer.h b/arm_compute/runtime/CL/functions/CLWidthConcatenateLayer.h index 44462b02b2..55b65dadc4 100644 --- a/arm_compute/runtime/CL/functions/CLWidthConcatenateLayer.h +++ b/arm_compute/runtime/CL/functions/CLWidthConcatenateLayer.h @@ -29,6 +29,8 @@ #include "arm_compute/core/Window.h" #include "arm_compute/runtime/IFunction.h" +#include "arm_compute/core/CL/kernels/CLWidthConcatenate2TensorsKernel.h" +#include "arm_compute/core/CL/kernels/CLWidthConcatenate4TensorsKernel.h" #include "arm_compute/core/CL/kernels/CLWidthConcatenateLayerKernel.h" #include <memory> @@ -40,7 +42,9 @@ class ICLTensor; /** Basic function to execute concatenate tensors along x axis. This function calls the following kernel: * - * -# @ref CLDepthConcatenateLayerKernel + * -# @ref CLWidthConcatenateLayerKernel + * -# @ref CLWidthConcatenate2TensorsKernel (if there are exactly 2 input tensors) + * -# @ref CLWidthConcatenate4TensorsKernel (if there are exactly 4 input tensors) * */ class CLWidthConcatenateLayer : public IFunction @@ -74,6 +78,8 @@ public: private: std::unique_ptr<CLWidthConcatenateLayerKernel[]> _concat_kernels_vector; + CLWidthConcatenate2TensorsKernel _concat_x2_kernel; + CLWidthConcatenate4TensorsKernel _concat_x4_kernel; unsigned int _num_inputs; }; } // namespace arm_compute |