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 | #include "arm_compute/core/CL/kernels/CLGEMMLowpOffsetContributionKernel.h" |
| 25 | |
| 26 | #include "arm_compute/core/AccessWindowStatic.h" |
| 27 | #include "arm_compute/core/CL/ICLTensor.h" |
| 28 | #include "arm_compute/core/Error.h" |
| 29 | #include "arm_compute/core/Helpers.h" |
| 30 | #include "arm_compute/core/TensorInfo.h" |
| 31 | #include "arm_compute/core/Types.h" |
| 32 | #include "arm_compute/core/Utils.h" |
| 33 | #include "arm_compute/core/Validate.h" |
| 34 | #include "arm_compute/core/Window.h" |
| 35 | #include "support/ToolchainSupport.h" |
| 36 | |
| 37 | #include <cstddef> |
| 38 | #include <cstdint> |
| 39 | |
| 40 | using namespace arm_compute; |
| 41 | |
| 42 | namespace arm_compute |
| 43 | { |
| 44 | class Coordinates; |
| 45 | } // namespace arm_compute |
| 46 | |
| 47 | CLGEMMLowpOffsetContributionKernel::CLGEMMLowpOffsetContributionKernel() |
| 48 | : _vector_sum_col(nullptr), _vector_sum_row(nullptr), _mm_result(nullptr) |
| 49 | { |
| 50 | } |
| 51 | |
| 52 | void CLGEMMLowpOffsetContributionKernel::configure(ICLTensor *mm_result, const ICLTensor *vector_sum_col, const ICLTensor *vector_sum_row, int32_t k, int32_t a_offset, int32_t b_offset) |
| 53 | { |
| 54 | ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(mm_result, 1, DataType::S32); |
| 55 | |
| 56 | // Set the arguments to pass at compile time |
| 57 | CLBuildOptions build_opts; |
| 58 | |
| 59 | // If a_offset == 0, vector_sum_col can be a nullptr |
| 60 | if(a_offset != 0) |
| 61 | { |
| 62 | ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(vector_sum_col, 1, DataType::S32); |
| 63 | ARM_COMPUTE_ERROR_ON(vector_sum_col->info()->dimension(0) != mm_result->info()->dimension(0)); |
| 64 | |
Gian Marco | 05288a2 | 2017-11-21 10:57:50 +0000 | [diff] [blame] | 65 | build_opts.add_option("-DA_OFFSET=" + support::cpp11::to_string(a_offset)); |
Chunosov | 5124be5 | 2017-11-22 20:42:13 +0700 | [diff] [blame] | 66 | build_opts.add_option_if(vector_sum_col->info()->tensor_shape().num_dimensions() > 1, "-DSUM_COL_HAS_BATCHES"); |
Gian Marco | 05288a2 | 2017-11-21 10:57:50 +0000 | [diff] [blame] | 67 | } |
| 68 | |
| 69 | // If b_offset == 0, vector_sum_row can be a nullptr |
| 70 | if(b_offset != 0) |
| 71 | { |
| 72 | ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(vector_sum_row, 1, DataType::S32); |
| 73 | ARM_COMPUTE_ERROR_ON(vector_sum_row->info()->dimension(0) != mm_result->info()->dimension(1)); |
| 74 | |
Georgios Pinitas | 45bcc3a | 2017-11-29 11:06:49 +0000 | [diff] [blame] | 75 | // Validate batches |
| 76 | TensorShape output_shape = mm_result->info()->tensor_shape(); |
| 77 | if(output_shape.num_dimensions() > 1) |
Gian Marco | 05288a2 | 2017-11-21 10:57:50 +0000 | [diff] [blame] | 78 | { |
Georgios Pinitas | 45bcc3a | 2017-11-29 11:06:49 +0000 | [diff] [blame] | 79 | TensorShape vector_sum_row_shape = vector_sum_row->info()->tensor_shape(); |
| 80 | vector_sum_row_shape.collapse_from(1); |
| 81 | output_shape.collapse_from(2); |
Gian Marco | 05288a2 | 2017-11-21 10:57:50 +0000 | [diff] [blame] | 82 | |
Georgios Pinitas | 45bcc3a | 2017-11-29 11:06:49 +0000 | [diff] [blame] | 83 | ARM_COMPUTE_ERROR_ON_MSG(vector_sum_row_shape[1] != output_shape[2], "mm_result tensor must have the same number of batches of output tensor"); |
| 84 | |
| 85 | if(a_offset != 0) |
| 86 | { |
| 87 | TensorShape vector_sum_col_shape = vector_sum_col->info()->tensor_shape(); |
| 88 | vector_sum_col_shape.collapse_from(1); |
| 89 | |
| 90 | ARM_COMPUTE_ERROR_ON_MSG(vector_sum_col_shape[1] != 1 |
| 91 | && vector_sum_col_shape[1] != vector_sum_row_shape[1], |
| 92 | "vector_sum_col tensor must have the same number of batches of vector_sum_row_shape or the number of batches must be set to 1"); |
| 93 | } |
Gian Marco | 05288a2 | 2017-11-21 10:57:50 +0000 | [diff] [blame] | 94 | } |
| 95 | |
| 96 | build_opts.add_option("-DB_OFFSET=" + support::cpp11::to_string(b_offset)); |
| 97 | } |
| 98 | |
| 99 | build_opts.add_option("-DK_OFFSET=" + support::cpp11::to_string(a_offset * b_offset * k)); |
| 100 | |
| 101 | // Create kernel |
| 102 | _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel("gemmlowp_offset_contribution", build_opts.options())); |
| 103 | |
| 104 | _vector_sum_col = vector_sum_col; |
| 105 | _vector_sum_row = vector_sum_row; |
| 106 | _mm_result = mm_result; |
| 107 | |
| 108 | constexpr unsigned int num_elems_processed_per_iteration = 16; |
| 109 | |
| 110 | // Configure kernel window |
| 111 | Window win = calculate_max_window(*mm_result->info(), Steps(num_elems_processed_per_iteration)); |
| 112 | |
| 113 | AccessWindowHorizontal mm_result_access(mm_result->info(), 0, num_elems_processed_per_iteration); |
| 114 | |
| 115 | update_window_and_padding(win, mm_result_access); |
| 116 | |
| 117 | if(a_offset != 0) |
| 118 | { |
| 119 | AccessWindowHorizontal vector_sum_col_access(vector_sum_col->info(), 0, num_elems_processed_per_iteration); |
| 120 | update_window_and_padding(win, vector_sum_col_access); |
| 121 | } |
| 122 | |
| 123 | if(b_offset != 0) |
| 124 | { |
| 125 | AccessWindowStatic vector_sum_row_access(vector_sum_row->info(), 0, 0, vector_sum_row->info()->dimension(0), 0); |
| 126 | update_window_and_padding(win, vector_sum_row_access); |
| 127 | } |
| 128 | |
| 129 | ICLKernel::configure(win); |
| 130 | } |
| 131 | |
| 132 | void CLGEMMLowpOffsetContributionKernel::run(const Window &window, cl::CommandQueue &queue) |
| 133 | { |
| 134 | ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this); |
| 135 | ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window); |
| 136 | |
| 137 | Window collapsed = window.collapse_if_possible(ICLKernel::window(), Window::DimZ); |
| 138 | Window slice = collapsed.first_slice_window_3D(); |
| 139 | |
| 140 | // Set window for vector_sum_col |
| 141 | Window win_vector_sum_col = slice; |
| 142 | win_vector_sum_col.set(Window::DimY, Window::Dimension(0, 0, 0)); |
| 143 | |
| 144 | // Set window for vector_sum_row |
| 145 | Window win_vector_sum_row = slice; |
| 146 | win_vector_sum_row.set(Window::DimX, Window::Dimension(0, 0, 0)); |
| 147 | win_vector_sum_row.set(Window::DimY, Window::Dimension(0, 0, 0)); |
| 148 | |
| 149 | do |
| 150 | { |
| 151 | unsigned int idx = 0; |
| 152 | add_3D_tensor_argument(idx, _mm_result, slice); |
| 153 | if(_vector_sum_col != nullptr) |
| 154 | { |
| 155 | add_2D_tensor_argument(idx, _vector_sum_col, win_vector_sum_col); |
| 156 | } |
| 157 | if(_vector_sum_row != nullptr) |
| 158 | { |
| 159 | add_2D_tensor_argument(idx, _vector_sum_row, win_vector_sum_row); |
| 160 | } |
| 161 | enqueue(queue, *this, slice); |
| 162 | } |
| 163 | while(collapsed.slide_window_slice_3D(slice)); |
| 164 | } |