Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [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_CLGEMMLOWPMATRIXMULTIPLYKERNEL_H__ |
| 25 | #define __ARM_COMPUTE_CLGEMMLOWPMATRIXMULTIPLYKERNEL_H__ |
| 26 | |
| 27 | #include "arm_compute/core/CL/ICLKernel.h" |
| 28 | |
| 29 | namespace arm_compute |
| 30 | { |
| 31 | class ICLTensor; |
| 32 | |
| 33 | /** OpenCL kernel to compute low precision matrix multiplication kernel |
| 34 | * |
| 35 | * This kernel performs the following computation: |
| 36 | * -# Convert a values from uint8 to int32 and add a_offset to each of them. |
| 37 | * -# Convert b values from uint8 to int32 and add b_offset to each of them. |
| 38 | * -# Compute the int32 matrix product of the resulting a * b. |
| 39 | * -# Add output_offset to each entry of the result. |
| 40 | * -# Multiply each entry of the result and round to the nearest integer |
| 41 | * -# Clamp the resulting int32 values to the [0..255] range and cast to uint8. |
| 42 | */ |
| 43 | class CLGEMMLowpMatrixMultiplyKernel : public ICLKernel |
| 44 | { |
| 45 | public: |
| 46 | /** Default Constructor */ |
| 47 | CLGEMMLowpMatrixMultiplyKernel(); |
| 48 | /** Prevent instances of this class from being copied (As this class contains pointers) */ |
| 49 | CLGEMMLowpMatrixMultiplyKernel(const CLGEMMLowpMatrixMultiplyKernel &) = delete; |
| 50 | /** Prevent instances of this class from being copied (As this class contains pointers) */ |
| 51 | CLGEMMLowpMatrixMultiplyKernel &operator=(const CLGEMMLowpMatrixMultiplyKernel &) = delete; |
| 52 | /** Allow instances of this class to be moved */ |
| 53 | CLGEMMLowpMatrixMultiplyKernel(CLGEMMLowpMatrixMultiplyKernel &&) = default; |
| 54 | /** Allow instances of this class to be moved */ |
| 55 | CLGEMMLowpMatrixMultiplyKernel &operator=(CLGEMMLowpMatrixMultiplyKernel &&) = default; |
| 56 | /** Initialise the kernel's input and output. |
| 57 | * |
| 58 | * The input matrices @p input0 and @p input1 must be the output of the kernels: @ref CLGEMMInterleave4x4Kernel and @ref CLGEMMTranspose1xWKernel. |
| 59 | * These two kernels change the layout of the original matrices to be more cache-friendly. |
| 60 | * |
| 61 | * @param[in] input0 Input tensor containing the interleaved Matrix A. Data types supported: U8 |
| 62 | * @param[in] input1 Input tensor containing the transposed Matrix B. Data types supported: same as @p input0 |
| 63 | * @param[out] output Output tensor to store the result of matrix multiplication, Data types supported: same as @p input0 |
| 64 | * @param[in] a_offset Offset to be added to each element of the matrix A. |
| 65 | * @param[in] b_offset Offset to be added to each element of the matrix B. |
| 66 | * @param[in] output_offset Offset to be added to each element of the output matrix |
| 67 | * @param[in] output_mult_int Offset to be added to each element of the output matrix |
| 68 | * @param[in] shift Number of bits to shift right the result. |
| 69 | */ |
| 70 | void configure(const ICLTensor *input0, const ICLTensor *input1, ICLTensor *output, int32_t a_offset, int32_t b_offset, int32_t output_offset, int32_t output_mult_int, int32_t shift); |
| 71 | |
| 72 | // Inherited methods overridden: |
| 73 | void run(const Window &window, cl::CommandQueue &queue) override; |
| 74 | |
| 75 | private: |
| 76 | const ICLTensor *_input0; |
| 77 | const ICLTensor *_input1; |
| 78 | ICLTensor *_output; |
| 79 | }; |
| 80 | } |
| 81 | #endif /*__ARM_COMPUTE_CLGEMMLOWPMATRIXMULTIPLYKERNEL_H__*/ |