Gian Marco | e75a02b | 2017-11-08 12:24:09 +0000 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2017 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_NEGEMMLOWPOFFSETCONTRIBUTIONKERNEL_H__ |
| 25 | #define __ARM_COMPUTE_NEGEMMLOWPOFFSETCONTRIBUTIONKERNEL_H__ |
| 26 | |
| 27 | #include "arm_compute/core/NEON/INEKernel.h" |
| 28 | |
| 29 | namespace arm_compute |
| 30 | { |
| 31 | class ITensor; |
| 32 | |
Gian Marco | 05288a2 | 2017-11-21 10:57:50 +0000 | [diff] [blame] | 33 | /** NEON kernel used to add the offset contribution after @ref NEGEMMLowpMatrixMultiplyKernel. The computation is performed in-place |
Gian Marco | e75a02b | 2017-11-08 12:24:09 +0000 | [diff] [blame] | 34 | * |
Gian Marco | 05288a2 | 2017-11-21 10:57:50 +0000 | [diff] [blame] | 35 | * This kernel takes a final int32 accumulator value (the output of @ref NEGEMMLowpMatrixMultiplyKernel), |
Gian Marco | e75a02b | 2017-11-08 12:24:09 +0000 | [diff] [blame] | 36 | * and adds to it the offset contribution of matrix A and matrix B in-place. |
| 37 | * |
Gian Marco | 6b77e91 | 2017-11-17 09:27:57 +0000 | [diff] [blame] | 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 | * |
Gian Marco | e75a02b | 2017-11-08 12:24:09 +0000 | [diff] [blame] | 45 | */ |
| 46 | class NEGEMMLowpOffsetContributionKernel : public INEKernel |
| 47 | { |
| 48 | public: |
| 49 | /** Constructor */ |
| 50 | NEGEMMLowpOffsetContributionKernel(); |
| 51 | /** Prevent instances of this class from being copied (As this class contains pointers)*/ |
| 52 | NEGEMMLowpOffsetContributionKernel(const NEGEMMLowpOffsetContributionKernel &) = delete; |
| 53 | /** Prevent instances of this class from being copied (As this class contains pointers)*/ |
| 54 | NEGEMMLowpOffsetContributionKernel &operator=(const NEGEMMLowpOffsetContributionKernel &) = delete; |
| 55 | /** Allow instances of this class to be moved */ |
| 56 | NEGEMMLowpOffsetContributionKernel(NEGEMMLowpOffsetContributionKernel &&) = default; |
| 57 | /** Allow instances of this class to be moved */ |
| 58 | NEGEMMLowpOffsetContributionKernel &operator=(NEGEMMLowpOffsetContributionKernel &&) = default; |
| 59 | /** Initialise the kernel's input and output. |
| 60 | * |
| 61 | * @param[in, out] mm_result Input tensor containing the result of @ref NEGEMMLowpMatrixMultiplyKernel. 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] k Number of matrix A columns or Matrix B rows |
| 67 | * @param[in] a_offset Offset to be added to each element of the matrix A. |
| 68 | * @param[in] b_offset Offset to be added to each element of the matrix B. |
| 69 | */ |
| 70 | void configure(ITensor *mm_result, const ITensor *vector_sum_col, const ITensor *vector_sum_row, int32_t k, int32_t a_offset, int32_t b_offset); |
Georgios Pinitas | a3b1b46 | 2017-11-16 19:24:39 +0000 | [diff] [blame] | 71 | /** Static function to check if given info will lead to a valid configuration of @ref NEGEMMLowpOffsetContributionKernel |
| 72 | * |
| 73 | * @param[in] mm_result Input tensor containing the result of @ref NEGEMMLowpMatrixMultiplyKernel. Data type supported: S32 |
| 74 | * @param[in] vector_sum_col Input row-vector of sums of all the entries in each column of matrix B. |
| 75 | * Note: vector_sum_col can be a nullptr in case a_offset = 0. Data type supported: same as @p mm_result |
| 76 | * @param[in] vector_sum_row Input row-vector of sums of all the entries in each row of matrix A. |
| 77 | * Note: vector_sum_row can be a nullptr in case b_offset = 0. Data type supported: same as @p mm_result |
| 78 | * @param[in] a_offset Offset to be added to each element of the matrix A. |
| 79 | * @param[in] b_offset Offset to be added to each element of the matrix B. |
| 80 | */ |
| 81 | static Error validate(const ITensorInfo *mm_result, const ITensorInfo *vector_sum_col, const ITensorInfo *vector_sum_row, int32_t a_offset, int32_t b_offset); |
Gian Marco | e75a02b | 2017-11-08 12:24:09 +0000 | [diff] [blame] | 82 | |
| 83 | // Inherited methods overridden: |
| 84 | void run(const Window &window, const ThreadInfo &info) override; |
| 85 | |
| 86 | private: |
| 87 | const ITensor *_vector_sum_col; |
| 88 | const ITensor *_vector_sum_row; |
| 89 | ITensor *_mm_result; |
| 90 | int32_t _a_offset; |
| 91 | int32_t _b_offset; |
| 92 | int32_t _k_offset; |
| 93 | bool _slide_vector_sum_col; |
| 94 | }; |
| 95 | } // namespace arm_compute |
| 96 | |
| 97 | #endif /* __ARM_COMPUTE_NEGEMMLOWPOFFSETCONTRIBUTIONKERNEL_H__ */ |