1 /* 2 * Copyright (c) 2017-2020 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_CLGEMMLOWPOFFSETCONTRIBUTIONKERNEL_H 25 #define ARM_COMPUTE_CLGEMMLOWPOFFSETCONTRIBUTIONKERNEL_H 26 27 #include "src/core/CL/ICLKernel.h" 28 29 namespace arm_compute 30 { 31 class ICLTensor; 32 33 /** OpenCL kernel used to add the offset contribution after the matrix multiplication. The computation is performed in-place 34 * 35 * This kernel takes a final int32 accumulator value (the output of the matrix multiplication), 36 * and adds to it the offset contribution of matrix A and matrix B in-place. 37 * 38 * The final result is: 39 * 40 * mm_result[i][k] = mm_result[i][k] + 41 * (vector_sum_col[k] * a_offset) + 42 * (vector_sum_row[i] * b_offset) + 43 * (a_offset * b_offset * k) 44 * 45 */ 46 class CLGEMMLowpOffsetContributionKernel : public ICLKernel 47 { 48 public: 49 /** Constructor */ 50 CLGEMMLowpOffsetContributionKernel(); 51 /** Prevent instances of this class from being copied (As this class contains pointers)*/ 52 CLGEMMLowpOffsetContributionKernel(const CLGEMMLowpOffsetContributionKernel &) = delete; 53 /** Prevent instances of this class from being copied (As this class contains pointers)*/ 54 CLGEMMLowpOffsetContributionKernel &operator=(const CLGEMMLowpOffsetContributionKernel &) = delete; 55 /** Allow instances of this class to be moved */ 56 CLGEMMLowpOffsetContributionKernel(CLGEMMLowpOffsetContributionKernel &&) = default; 57 /** Allow instances of this class to be moved */ 58 CLGEMMLowpOffsetContributionKernel &operator=(CLGEMMLowpOffsetContributionKernel &&) = default; 59 /** Initialise the kernel's input and output. 60 * 61 * @param[in, out] mm_result Input tensor containing the result of the matrix multiplication. Data type supported: S32 62 * @param[in] vector_sum_col Input row-vector of sums of all the entries in each column of matrix B. 63 * Note: vector_sum_col can be a nullptr in case a_offset = 0. Data type supported: same as @p mm_result 64 * @param[in] vector_sum_row Input row-vector of sums of all the entries in each row of matrix A. 65 * Note: vector_sum_row can be a nullptr in case b_offset = 0. Data type supported: same as @p mm_result 66 * @param[in] bias Biases tensor. Only shared biases supported and it can be a nullptr if the addition of biases is not required. 67 * Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input. 68 * @param[in] k Number of matrix A columns or Matrix B rows 69 * @param[in] a_offset Offset to be added to each element of the matrix A. 70 * @param[in] b_offset Offset to be added to each element of the matrix B. 71 */ 72 void configure(ICLTensor *mm_result, const ICLTensor *vector_sum_col, const ICLTensor *vector_sum_row, const ICLTensor *bias, int32_t k, int32_t a_offset, int32_t b_offset); 73 /** Initialise the kernel's input and output. 74 * 75 * @param[in] compile_context The compile context to be used. 76 * @param[in, out] mm_result Input tensor containing the result of the matrix multiplication. Data type supported: S32 77 * @param[in] vector_sum_col Input row-vector of sums of all the entries in each column of matrix B. 78 * Note: vector_sum_col can be a nullptr in case a_offset = 0. Data type supported: same as @p mm_result 79 * @param[in] vector_sum_row Input row-vector of sums of all the entries in each row of matrix A. 80 * Note: vector_sum_row can be a nullptr in case b_offset = 0. Data type supported: same as @p mm_result 81 * @param[in] bias Biases tensor. Only shared biases supported and it can be a nullptr if the addition of biases is not required. 82 * Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input. 83 * @param[in] k Number of matrix A columns or Matrix B rows 84 * @param[in] a_offset Offset to be added to each element of the matrix A. 85 * @param[in] b_offset Offset to be added to each element of the matrix B. 86 */ 87 void configure(const CLCompileContext &compile_context, ICLTensor *mm_result, const ICLTensor *vector_sum_col, const ICLTensor *vector_sum_row, const ICLTensor *bias, int32_t k, int32_t a_offset, 88 int32_t b_offset); 89 /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMLowpOffsetContributionKernel 90 * 91 * @param[in] mm_result Input tensor containing the result of @ref CLGEMMLowpOffsetContributionKernel. Data type supported: S32 92 * @param[in] vector_sum_col Input row-vector of sums of all the entries in each column of matrix B. 93 * Note: vector_sum_col can be a nullptr in case a_offset = 0. Data type supported: same as @p mm_result 94 * @param[in] vector_sum_row Input row-vector of sums of all the entries in each row of matrix A. 95 * Note: vector_sum_row can be a nullptr in case b_offset = 0. Data type supported: same as @p mm_result 96 * @param[in] bias Biases tensor. Only shared biases supported and it can be a nullptr if the addition of biases is not required. 97 * Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input. 98 * @param[in] a_offset Offset to be added to each element of the matrix A. 99 * @param[in] b_offset Offset to be added to each element of the matrix B. 100 * 101 * @return a status 102 */ 103 static Status validate(const ITensorInfo *mm_result, const ITensorInfo *vector_sum_col, const ITensorInfo *vector_sum_row, const ITensorInfo *bias, int32_t a_offset, int32_t b_offset); 104 105 // Inherited methods overridden: 106 void run(const Window &window, cl::CommandQueue &queue) override; 107 108 private: 109 const ICLTensor *_vector_sum_col; 110 const ICLTensor *_vector_sum_row; 111 ICLTensor *_mm_result; 112 const ICLTensor *_bias; 113 }; 114 } // namespace arm_compute 115 116 #endif /* ARM_COMPUTE_CLGEMMLOWPOFFSETCONTRIBUTIONKERNEL_H */ 117