1 /* 2 * Copyright (c) 2016-2020 Arm Limited. 3 * 4 * SPDX-License-Identifier: MIT 5 * 6 * Permission is hereby granted, free of charge, to any person obtaining a copy 7 * of this software and associated documentation files (the "Software"), to 8 * deal in the Software without restriction, including without limitation the 9 * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or 10 * sell copies of the Software, and to permit persons to whom the Software is 11 * furnished to do so, subject to the following conditions: 12 * 13 * The above copyright notice and this permission notice shall be included in all 14 * copies or substantial portions of the Software. 15 * 16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 19 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 22 * SOFTWARE. 23 */ 24 #ifndef ARM_COMPUTE_NEGEMMMATRIXADDITIONKERNEL_H 25 #define ARM_COMPUTE_NEGEMMMATRIXADDITIONKERNEL_H 26 27 #include "src/core/NEON/INESimpleKernel.h" 28 29 namespace arm_compute 30 { 31 class ITensor; 32 33 /** NEON kernel to perform the in-place matrix addition between 2 matrices taking into account that the second matrix might be weighted by a scalar value beta: 34 * 35 * @note [ MTX_OUT = MTX_0 + beta * MTX_1 ] with MTX_0 and MTX_1 of the same size 36 * 37 * @note This stage is used to finalize the GEMM result and it is computed if and only if beta != 0.0. In case this kernel is used for finalizing GEMM result, we have: 38 * - MTX_0 = A * B * alpha, where MTX_0 is the output of @ref NEGEMMMatrixMultiplyKernel 39 * - MTX_1 = C 40 */ 41 class NEGEMMMatrixAdditionKernel : public INESimpleKernel 42 { 43 public: name()44 const char *name() const override 45 { 46 return "NEGEMMMatrixAdditionKernel"; 47 } 48 /** Constructor */ 49 NEGEMMMatrixAdditionKernel(); 50 /** Prevent instances of this class from being copied */ 51 NEGEMMMatrixAdditionKernel(const NEGEMMMatrixAdditionKernel &) = delete; 52 /** Prevent instances of this class from being copied */ 53 NEGEMMMatrixAdditionKernel &operator=(const NEGEMMMatrixAdditionKernel &) = delete; 54 /** Allow instances of this class to be moved */ 55 NEGEMMMatrixAdditionKernel(NEGEMMMatrixAdditionKernel &&) = default; 56 /** Allow instances of this class to be moved */ 57 NEGEMMMatrixAdditionKernel &operator=(NEGEMMMatrixAdditionKernel &&) = default; 58 /** Default destructor */ 59 ~NEGEMMMatrixAdditionKernel() = default; 60 /** Initialise the kernel's input and output. 61 * 62 * @note The input and output tensor must have the same dimensions 63 * 64 * @param[in] input Input tensor (Matrix C). Data types supported: F16/F32 65 * @param[in, out] output Output tensor. If this kernel is used to finalize the GEMM result, output contains the result obtained by the kernel @ref NEGEMMMatrixMultiplyKernel. Data type supported: the same as @p input. 66 * @param[in] beta Weight of matrix C 67 */ 68 void configure(const ITensor *input, ITensor *output, float beta); 69 /** Static function to check if given info will lead to a valid configuration of @ref NEGEMMMatrixAdditionKernel. 70 * 71 * @note The input and output tensor must have the same dimensions 72 * 73 * @param[in] input Input tensor info (Matrix C). Data types supported: F16/F32 74 * @param[in] output Output tensor info. If this kernel is used to finalize the GEMM result, output contains the result obtained by the kernel @ref NEGEMMMatrixMultiplyKernel. Data type supported: the same as @p input. 75 * @param[in] beta Weight of matrix C 76 * 77 * @return a status 78 */ 79 static Status validate(const ITensorInfo *input, const ITensorInfo *output, float beta); 80 81 // Inherited methods overridden: 82 void run(const Window &window, const ThreadInfo &info) override; 83 84 private: 85 /** Common signature for all the matrix addition functions 86 * 87 * @param[in] input An input tensor. Data types supported: F16/F32 88 * @param[out] output The output tensor. Data type supported: same as @p input 89 * @param[in] window Region on which to execute the kernel. 90 * @param[in] beta Weight of matrix C 91 */ 92 using MatrixAdditionFunction = void(const ITensor *input, ITensor *output, const Window &window, float beta); 93 /** Matrix addition function to use for the particular tensor types passed to configure() */ 94 MatrixAdditionFunction *_func; 95 float _beta; 96 }; 97 } // namespace arm_compute 98 #endif /* ARM_COMPUTE_NEGEMMMATRIXADDITIONKERNEL_H */ 99