Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2016, 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_NEGEMMMATRIXADDITIONKERNEL_H__ |
| 25 | #define __ARM_COMPUTE_NEGEMMMATRIXADDITIONKERNEL_H__ |
| 26 | |
| 27 | #include "arm_compute/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: |
| 44 | /** Constructor */ |
| 45 | NEGEMMMatrixAdditionKernel(); |
| 46 | /** Prevent instances of this class from being copied */ |
| 47 | NEGEMMMatrixAdditionKernel(const NEGEMMMatrixAdditionKernel &) = delete; |
| 48 | /** Prevent instances of this class from being copied */ |
| 49 | NEGEMMMatrixAdditionKernel &operator=(const NEGEMMMatrixAdditionKernel &) = delete; |
| 50 | /** Allow instances of this class to be moved */ |
| 51 | NEGEMMMatrixAdditionKernel(NEGEMMMatrixAdditionKernel &&) = default; |
| 52 | /** Allow instances of this class to be moved */ |
| 53 | NEGEMMMatrixAdditionKernel &operator=(NEGEMMMatrixAdditionKernel &&) = default; |
| 54 | /** Initialise the kernel's input and output. |
| 55 | * |
| 56 | * @note The input and output tensor must have the same dimensions |
| 57 | * |
Gian Marco Iodice | bdb6b0b | 2017-06-30 12:21:00 +0100 | [diff] [blame] | 58 | * @param[in] input Input tensor (Matrix C). Data types supported: QS8/QS16/F16/F32 |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 59 | * @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. |
| 60 | * @param[in] beta Weight of matrix C |
| 61 | */ |
| 62 | void configure(const ITensor *input, ITensor *output, float beta); |
| 63 | |
| 64 | // Inherited methods overridden: |
| 65 | void run(const Window &window) override; |
| 66 | |
| 67 | private: |
| 68 | /** Common signature for all the matrix addition functions |
| 69 | * |
Gian Marco Iodice | bdb6b0b | 2017-06-30 12:21:00 +0100 | [diff] [blame] | 70 | * @param[in] input An input tensor. Data types supported: QS8/QS16/F16/F32 |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 71 | * @param[out] output The output tensor. Data type supported: same as @p input |
| 72 | * @param[in] window Region on which to execute the kernel. |
| 73 | * @param[in] beta Weight of matrix C |
| 74 | */ |
| 75 | using MatrixAdditionFunction = void(const ITensor *input, ITensor *output, const Window &window, float beta); |
| 76 | /** Matrix addition function to use for the particular tensor types passed to configure() */ |
| 77 | MatrixAdditionFunction *_func; |
| 78 | float _beta; |
| 79 | }; |
| 80 | } |
| 81 | #endif /* __ARM_COMPUTE_NEGEMMMATRIXADDITIONKERNEL_H__ */ |