Gian Marco Iodice | 926afe1 | 2019-03-19 11:44:13 +0000 | [diff] [blame] | 1 | /* |
Gian Marco Iodice | f16eed9 | 2023-04-19 16:29:26 +0100 | [diff] [blame] | 2 | * Copyright (c) 2019-2023 Arm Limited. |
Gian Marco Iodice | 926afe1 | 2019-03-19 11:44:13 +0000 | [diff] [blame] | 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 | */ |
Georgios Pinitas | 856f66e | 2021-04-22 21:13:21 +0100 | [diff] [blame] | 24 | #ifndef ARM_COMPUTE_CL_GEMM_HELPERS_H |
| 25 | #define ARM_COMPUTE_CL_GEMM_HELPERS_H |
Gian Marco Iodice | 926afe1 | 2019-03-19 11:44:13 +0000 | [diff] [blame] | 26 | |
Manuel Bottini | c8e6e2c | 2020-07-02 11:27:27 +0100 | [diff] [blame] | 27 | #include "arm_compute/core/TensorInfo.h" |
Gian Marco Iodice | 926afe1 | 2019-03-19 11:44:13 +0000 | [diff] [blame] | 28 | #include "arm_compute/core/Types.h" |
| 29 | |
| 30 | namespace arm_compute |
| 31 | { |
Georgios Pinitas | 856f66e | 2021-04-22 21:13:21 +0100 | [diff] [blame] | 32 | namespace opencl |
| 33 | { |
| 34 | namespace kernels |
| 35 | { |
| 36 | namespace gemm |
Gian Marco Iodice | 926afe1 | 2019-03-19 11:44:13 +0000 | [diff] [blame] | 37 | { |
Gian Marco Iodice | f16eed9 | 2023-04-19 16:29:26 +0100 | [diff] [blame] | 38 | using GeMMConfigsMatrix = std::vector<std::vector<int32_t>>; |
| 39 | |
Gian Marco Iodice | 926afe1 | 2019-03-19 11:44:13 +0000 | [diff] [blame] | 40 | /** Configure @ref GEMMLHSMatrixInfo and @ref GEMMRHSMatrixInfo |
| 41 | * |
Gian Marco Iodice | ed5fe69 | 2020-07-09 08:41:10 +0100 | [diff] [blame] | 42 | * @param[in] m Number of rows (M) in the LHS matrix not reshaped |
| 43 | * @param[in] n Number of columns (N) in the RHS matrix not reshaped |
| 44 | * @param[in] m0 Number of rows processed by each thread/work-item |
| 45 | * @param[in] n0 Number of columns processed by each thread/work-item |
| 46 | * @param[in] k0 Number of inner accumulation performed by each thread/work-item |
| 47 | * @param[in] v0 Number of vertical blocks of size (m0xk0) stored on the same output row |
| 48 | * @param[in] h0 Number of horizontal blocks of size (k0xn0) stored on the same output row |
| 49 | * @param[in] lhs_interleave True if the v0 (m0xk0) blocks have to be interleaved in the output row |
| 50 | * @param[in] rhs_interleave True if the h0 (k0xn0) blocks have to be interleaved in the output row |
| 51 | * @param[in] lhs_transpose True if the (m0xk0) block has to be transposed before been stored |
| 52 | * @param[in] rhs_transpose True if the (k0xn0) block has to be transposed before been stored |
| 53 | * @param[in] export_to_cl_image (Optional) True if the RHS reshaped matrix has to be exported to cl_image |
Gian Marco Iodice | 926afe1 | 2019-03-19 11:44:13 +0000 | [diff] [blame] | 54 | * |
| 55 | * @return @ref GEMMLHSMatrixInfo and @ref GEMMRHSMatrixInfo |
| 56 | */ |
| 57 | std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_lhs_rhs_info(unsigned int m, unsigned int n, unsigned int m0, unsigned int n0, unsigned int k0, unsigned int v0, unsigned int h0, |
Gian Marco Iodice | ed5fe69 | 2020-07-09 08:41:10 +0100 | [diff] [blame] | 58 | bool lhs_interleave, bool rhs_interleave, bool lhs_transpose, bool rhs_transpose, bool export_to_cl_image = false); |
Manuel Bottini | c8e6e2c | 2020-07-02 11:27:27 +0100 | [diff] [blame] | 59 | |
Gian Marco Iodice | 839e198 | 2020-10-29 13:36:50 +0000 | [diff] [blame] | 60 | /** Select @ref GEMMLHSMatrixInfo and @ref GEMMRHSMatrixInfo |
| 61 | * |
| 62 | * This function accepts two pairs of GEMMLHSMatrixInfo/GEMMRHSMatrixInfo where only the first is with cl_image2d support, |
| 63 | * and selects the valid one validating the GEMMRHSMatrixInfo. If the validation passes, the functions will return |
| 64 | * the first GEMMLHSMatrixInfo/GEMMRHSMatrixInfo pair with cl_image2d support. |
| 65 | * |
| 66 | * @param[in] info_img GEMMLHSMatrixInfo/GEMMRHSMatrixInfo with cl_image2d support |
| 67 | * @param[in] info_buf GEMMLHSMatrixInfo/GEMMRHSMatrixInfo to fall-back if cl_image2d cannot be used |
| 68 | * @param[in] n Number of columns (N) in the RHS matrix not reshaped |
| 69 | * @param[in] k Number of rows (K) in the RHS matrix not reshaped |
| 70 | * @param[in] b Batch size |
| 71 | * @param[in] data_type Data type |
| 72 | * |
| 73 | * @return @ref GEMMLHSMatrixInfo and @ref GEMMRHSMatrixInfo |
| 74 | */ |
| 75 | std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> select_lhs_rhs_info(std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> info_img, |
| 76 | std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> info_buf, |
| 77 | unsigned int n, unsigned int k, unsigned int b, DataType data_type); |
| 78 | |
Manuel Bottini | c8e6e2c | 2020-07-02 11:27:27 +0100 | [diff] [blame] | 79 | /** Update padding required to export the OpenCL buffer to OpenCL image2d |
| 80 | * |
| 81 | * @param[in,out] tensor ITensorInfo of the tensor required to be exported to OpenCL image2d |
| 82 | */ |
| 83 | void update_padding_for_cl_image(ITensorInfo *tensor); |
Gian Marco Iodice | ed5fe69 | 2020-07-09 08:41:10 +0100 | [diff] [blame] | 84 | |
| 85 | /** Utility function to validate the image2d OpenCL object support on the RHS reshaped matrix |
| 86 | * |
| 87 | * @param[in] tensor_reshaped_info TensorInfo for the RHS reshaped matrix |
| 88 | * @param[in] rhs_info @ref GEMMRHSMatrixInfo |
| 89 | * |
| 90 | * @return Status reporting if we can use the image2d OpenCL object on the RHS reshaped matrix |
| 91 | */ |
| 92 | Status validate_image2d_support_on_rhs(const ITensorInfo &tensor_reshaped_info, const GEMMRHSMatrixInfo &rhs_info); |
Gunes Bayir | 4bfc70e | 2021-12-10 16:17:56 +0000 | [diff] [blame] | 93 | |
| 94 | /** Determine if the MMUL kernels should be preferred |
| 95 | * |
| 96 | * @param[in] m Number of rows of the LHS matrix |
| 97 | * @param[in] n Number of columns of the RHS matrix |
| 98 | * @param[in] k Number of columns of the LHS matrix, rows of the RHS matrix |
| 99 | * @param[in] b Batch size |
| 100 | * @param[in] data_type Data type FP32/FP16 |
| 101 | * @param[in, out] best_m0 Suggested M0 (number of rows of the output block) for the kernel |
| 102 | * @param[in, out] best_n0 Suggested N0 (number of columns of the output block) for the kernel |
| 103 | * |
| 104 | * @return true if MMUL kernel is preferred over kernels w/o MMUL, false otherwise |
| 105 | */ |
| 106 | bool is_mmul_kernel_preferred(const unsigned int m, const unsigned int n, const unsigned int k, const unsigned int b, |
| 107 | const DataType data_type, unsigned int &best_m0, unsigned int &best_n0); |
Gian Marco Iodice | f16eed9 | 2023-04-19 16:29:26 +0100 | [diff] [blame] | 108 | |
| 109 | /** Find the preferred configurations for the LHS and RHS tensor using the GeMMConfigsMatrix provided by the user |
| 110 | * |
| 111 | * @param[in] configs List of best configurations for a limited number of GeMM shapes |
| 112 | * @param[in] m Number of rows of the LHS matrix |
| 113 | * @param[in] n Number of columns of the RHS matrix |
| 114 | * @param[in] k Number of columns of the LHS matrix, rows of the RHS matrix |
| 115 | * @param[in] b Batch size |
| 116 | * |
| 117 | * @return @ref GEMMLHSMatrixInfo and @ref GEMMRHSMatrixInfo |
| 118 | */ |
| 119 | std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> find_lhs_rhs_info(const GeMMConfigsMatrix &configs, unsigned int m, unsigned int n, unsigned int k, unsigned int b); |
Georgios Pinitas | 856f66e | 2021-04-22 21:13:21 +0100 | [diff] [blame] | 120 | } // namespace gemm |
| 121 | } // namespace kernels |
| 122 | } // namespace opencl |
Gian Marco Iodice | 926afe1 | 2019-03-19 11:44:13 +0000 | [diff] [blame] | 123 | } // namespace arm_compute |
Georgios Pinitas | 856f66e | 2021-04-22 21:13:21 +0100 | [diff] [blame] | 124 | #endif /* ARM_COMPUTE_CL_GEMM_HELPERS_H */ |