Gian Marco | 05288a2 | 2017-11-21 10:57:50 +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_CLGEMMLOWPMATRIXMULTIPLYCORE_H__ |
| 25 | #define __ARM_COMPUTE_CLGEMMLOWPMATRIXMULTIPLYCORE_H__ |
| 26 | |
| 27 | #include "arm_compute/core/CL/kernels/CLGEMMInterleave4x4Kernel.h" |
| 28 | #include "arm_compute/core/CL/kernels/CLGEMMLowpMatrixMultiplyKernel.h" |
| 29 | #include "arm_compute/core/CL/kernels/CLGEMMLowpOffsetContributionKernel.h" |
| 30 | #include "arm_compute/core/CL/kernels/CLGEMMLowpReductionKernel.h" |
| 31 | #include "arm_compute/core/CL/kernels/CLGEMMTranspose1xWKernel.h" |
| 32 | #include "arm_compute/runtime/CL/CLMemoryGroup.h" |
| 33 | #include "arm_compute/runtime/CL/CLTensor.h" |
| 34 | #include "arm_compute/runtime/IFunction.h" |
| 35 | |
| 36 | namespace arm_compute |
| 37 | { |
| 38 | class IMemoryManager; |
| 39 | class ICLTensor; |
| 40 | |
| 41 | /** Basic function to execute GEMMLowpMatrixMultiplyCore on OpenCL. This function calls the following OpenCL kernels: |
| 42 | * |
| 43 | * -# @ref CLGEMMInterleave4x4Kernel (if the output tensor is a matrix) |
| 44 | * -# @ref CLGEMMTranspose1xWKernel (if the output tensor is a matrix) |
| 45 | * -# @ref CLGEMMLowpMatrixMultiplyKernel |
| 46 | * -# @ref CLGEMMLowpMatrixAReductionKernel (if the offset of matrix B is not 0) |
| 47 | * -# @ref CLGEMMLowpMatrixBReductionKernel (if the offset of matrix A is not 0) |
| 48 | * -# @ref CLGEMMLowpOffsetContributionKernel |
| 49 | * |
| 50 | */ |
| 51 | class CLGEMMLowpMatrixMultiplyCore : public IFunction |
| 52 | { |
| 53 | public: |
| 54 | /** Constructor */ |
| 55 | CLGEMMLowpMatrixMultiplyCore(std::shared_ptr<IMemoryManager> memory_manager = nullptr); |
| 56 | /** Initialise the kernel's inputs, output |
| 57 | * |
| 58 | * @note GEMM_LOWP: low precision GEMM kernel |
| 59 | * This kernel performs the following computations: |
| 60 | * |
| 61 | * -# Convert a values from QASYMM8 to int32 and add a_offset to each of them. |
| 62 | * -# Convert b values from QASYMM8 to int32 add b_offset to each of them. |
| 63 | * -# Compute the matrix product of the resulting a * b in int32. |
| 64 | * |
Chunosov | 5124be5 | 2017-11-22 20:42:13 +0700 | [diff] [blame] | 65 | * @param[in] a First input tensor (Matrix A). Data type supported: QASYMM8. |
| 66 | * @param[in] b Second input tensor (Matrix B). Data type supported: same as @p a |
| 67 | * @param[out] output Output tensor. Data type supported: Data type supported: S32 |
| 68 | * @param[in] gemm_info (Optional) Specifies if the matrix A and/or matrix B have been reshaped and |
| 69 | * if the reshape of matrix B should be executed only for the first run |
Gian Marco | 05288a2 | 2017-11-21 10:57:50 +0000 | [diff] [blame] | 70 | */ |
Chunosov | 5124be5 | 2017-11-22 20:42:13 +0700 | [diff] [blame] | 71 | void configure(const ICLTensor *a, const ICLTensor *b, ICLTensor *output, const GEMMInfo &gemm_info = GEMMInfo()); |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 72 | /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMLowpMatrixMultiplyCore |
| 73 | * |
| 74 | * @param[in] a First input tensor (Matrix A). Data type supported: QASYMM8. |
| 75 | * @param[in] b Second input tensor (Matrix B). Data type supported: same as @p a |
| 76 | * @param[in] output Output tensor. Data type supported: Data type supported: S32 |
| 77 | * @param[in] gemm_info (Optional) Specifies if the matrix A and/or matrix B have been reshaped and |
| 78 | * if the reshape of matrix B should be executed only for the first run |
| 79 | * |
| 80 | * @return a status |
| 81 | */ |
| 82 | static Status validate(const ITensorInfo *a, const ITensorInfo *b, const ITensorInfo *output, const GEMMInfo &gemm_info = GEMMInfo()); |
Gian Marco | 05288a2 | 2017-11-21 10:57:50 +0000 | [diff] [blame] | 83 | |
| 84 | // Inherited methods overridden: |
| 85 | void run() override; |
| 86 | |
| 87 | private: |
| 88 | CLMemoryGroup _memory_group; |
| 89 | CLGEMMLowpMatrixMultiplyKernel _mm_kernel; |
| 90 | CLGEMMInterleave4x4Kernel _mtx_a_reshape_kernel; |
| 91 | CLGEMMTranspose1xWKernel _mtx_b_reshape_kernel; |
| 92 | CLGEMMLowpMatrixAReductionKernel _mtx_a_reduction_kernel; |
| 93 | CLGEMMLowpMatrixBReductionKernel _mtx_b_reduction_kernel; |
| 94 | CLGEMMLowpOffsetContributionKernel _offset_contribution_kernel; |
| 95 | CLTensor _vector_sum_col; |
| 96 | CLTensor _vector_sum_row; |
| 97 | CLTensor _tmp_a; |
| 98 | CLTensor _tmp_b; |
| 99 | int32_t _a_offset; |
| 100 | int32_t _b_offset; |
| 101 | bool _is_interleaved_transposed; |
Chunosov | 5124be5 | 2017-11-22 20:42:13 +0700 | [diff] [blame] | 102 | bool _is_first_run; |
| 103 | bool _reshape_b_only_on_first_run; |
Gian Marco | 05288a2 | 2017-11-21 10:57:50 +0000 | [diff] [blame] | 104 | }; |
| 105 | } |
| 106 | #endif /*__ARM_COMPUTE_CLGEMMLOWPMATRIXMULTIPLYCORE_H__ */ |