Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 1 | /* |
Gian Marco | 36a0a46 | 2018-01-12 10:21:40 +0000 | [diff] [blame] | 2 | * Copyright (c) 2017-2018 ARM Limited. |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [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 | */ |
| 24 | #include "arm_compute/core/CL/kernels/CLGEMMMatrixMultiplyKernel.h" |
| 25 | |
| 26 | #include "arm_compute/core/AccessWindowStatic.h" |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 27 | #include "arm_compute/core/CL/CLHelpers.h" |
| 28 | #include "arm_compute/core/CL/CLKernelLibrary.h" |
Vidhya Sudhan Loganathan | f1f4906 | 2018-05-25 13:21:26 +0100 | [diff] [blame] | 29 | #include "arm_compute/core/CL/CLValidate.h" |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 30 | #include "arm_compute/core/CL/ICLTensor.h" |
| 31 | #include "arm_compute/core/CL/OpenCL.h" |
| 32 | #include "arm_compute/core/Error.h" |
| 33 | #include "arm_compute/core/Helpers.h" |
Isabella Gottardi | d56e770 | 2018-02-28 14:29:36 +0000 | [diff] [blame] | 34 | #include "arm_compute/core/TensorInfo.h" |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 35 | #include "arm_compute/core/Types.h" |
| 36 | #include "arm_compute/core/Utils.h" |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 37 | #include "arm_compute/core/Window.h" |
Gian Marco | 36a0a46 | 2018-01-12 10:21:40 +0000 | [diff] [blame] | 38 | #include "arm_compute/core/utils/misc/ShapeCalculator.h" |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 39 | |
| 40 | #include <set> |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 41 | #include <string> |
| 42 | |
| 43 | using namespace arm_compute; |
Gian Marco | 36a0a46 | 2018-01-12 10:21:40 +0000 | [diff] [blame] | 44 | using namespace arm_compute::misc::shape_calculator; |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 45 | |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 46 | namespace |
| 47 | { |
| 48 | using ElementsProcessed = Steps; |
| 49 | |
Vidhya Sudhan Loganathan | a25d16c | 2018-11-16 11:33:12 +0000 | [diff] [blame] | 50 | inline Status validate_arguments(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *output, bool is_interleaved_transposed, const GEMMReshapeInfo &reshape_info, |
| 51 | bool fp_mixed_precision) |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 52 | { |
Georgios Pinitas | 78c0090 | 2018-01-09 17:33:11 +0000 | [diff] [blame] | 53 | ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input0, input1, output); |
Vidhya Sudhan Loganathan | f1f4906 | 2018-05-25 13:21:26 +0100 | [diff] [blame] | 54 | ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(input0); |
Vidhya Sudhan Loganathan | 7485d5a | 2018-07-04 09:34:00 +0100 | [diff] [blame] | 55 | ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input0, 1, DataType::F16, DataType::F32); |
Gian Marco | 36a0a46 | 2018-01-12 10:21:40 +0000 | [diff] [blame] | 56 | ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input0, input1); |
Vidhya Sudhan Loganathan | a25d16c | 2018-11-16 11:33:12 +0000 | [diff] [blame] | 57 | ARM_COMPUTE_RETURN_ERROR_ON_MSG((fp_mixed_precision && (input0->data_type() != DataType::F16)), "Mixed precision floating point is supported only for F16 data"); |
Isabella Gottardi | 8e74f44 | 2018-03-01 16:42:00 +0000 | [diff] [blame] | 58 | ARM_COMPUTE_RETURN_ERROR_ON_MSG(input0->num_dimensions() > 4, "The number of dimensions for the matrix A must be <= 4"); |
Gian Marco Iodice | d2fab73 | 2018-03-02 11:18:12 +0000 | [diff] [blame] | 59 | ARM_COMPUTE_RETURN_ERROR_ON_MSG(input1->num_dimensions() > 3, "The number of dimensions for the matrix B must be <= 3"); |
Gian Marco Iodice | 68a3f56 | 2018-07-26 11:44:03 +0100 | [diff] [blame] | 60 | ARM_COMPUTE_RETURN_ERROR_ON_MSG(is_interleaved_transposed && reshape_info.reinterpret_input_as_3d(), "The input tensor cannot be reinterpreted as 3D if is_interleaved_transposed is true"); |
Gian Marco Iodice | d39e2b1 | 2018-08-06 14:31:15 +0100 | [diff] [blame] | 61 | ARM_COMPUTE_RETURN_ERROR_ON_MSG(input1->num_dimensions() > 2 && reshape_info.reinterpret_input_as_3d(), "The input1 tensor cannot have more than 2 dimensions if input0 has to be reinterpreted as 3D"); |
Gian Marco | 36a0a46 | 2018-01-12 10:21:40 +0000 | [diff] [blame] | 62 | |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 63 | if(!is_interleaved_transposed) |
| 64 | { |
| 65 | ARM_COMPUTE_RETURN_ERROR_ON(input0->dimension(0) != input1->dimension(1)); |
Gian Marco | 36a0a46 | 2018-01-12 10:21:40 +0000 | [diff] [blame] | 66 | } |
| 67 | else |
| 68 | { |
giuros01 | 8b6b4a9 | 2018-12-18 19:01:33 +0000 | [diff] [blame] | 69 | GEMMRHSMatrixInfo rhs_info; |
| 70 | const int m = reshape_info.m(); |
| 71 | const int n = reshape_info.n(); |
| 72 | const int k = reshape_info.k(); |
| 73 | const int mult_transpose1xW_width = reshape_info.mult_transpose1xW_width(); |
| 74 | const int mult_interleave4x4_height = reshape_info.mult_interleave4x4_height(); |
| 75 | rhs_info.n0 = 16 / input1->element_size(); |
| 76 | rhs_info.k0 = 1; |
| 77 | rhs_info.h0 = mult_transpose1xW_width; |
| 78 | rhs_info.interleave = false; |
| 79 | rhs_info.transpose = false; |
Gian Marco | 36a0a46 | 2018-01-12 10:21:40 +0000 | [diff] [blame] | 80 | |
| 81 | TensorShape tensor_shape0{ input0->tensor_shape() }; |
| 82 | tensor_shape0.set(0, k); |
| 83 | tensor_shape0.set(1, m); |
| 84 | |
| 85 | TensorShape tensor_shape1{ input1->tensor_shape() }; |
| 86 | tensor_shape1.set(0, n); |
| 87 | tensor_shape1.set(1, k); |
| 88 | |
| 89 | const TensorInfo tensor_info0 = input0->clone()->set_tensor_shape(tensor_shape0); |
| 90 | const TensorInfo tensor_info1 = input1->clone()->set_tensor_shape(tensor_shape1); |
| 91 | |
| 92 | const TensorInfo tensor_info_reshaped0 = input0->clone()->set_tensor_shape(compute_interleaved_shape(tensor_info0, mult_interleave4x4_height)); |
giuros01 | 8b6b4a9 | 2018-12-18 19:01:33 +0000 | [diff] [blame] | 93 | const TensorInfo tensor_info_reshaped1 = input1->clone()->set_tensor_shape(compute_rhs_reshaped_shape(tensor_info1, rhs_info)); |
Gian Marco | 36a0a46 | 2018-01-12 10:21:40 +0000 | [diff] [blame] | 94 | |
| 95 | ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input0, &tensor_info_reshaped0); |
| 96 | ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input1, &tensor_info_reshaped1); |
Isabella Gottardi | 8e74f44 | 2018-03-01 16:42:00 +0000 | [diff] [blame] | 97 | } |
Gian Marco | 36a0a46 | 2018-01-12 10:21:40 +0000 | [diff] [blame] | 98 | |
Isabella Gottardi | 8e74f44 | 2018-03-01 16:42:00 +0000 | [diff] [blame] | 99 | if(output->total_size() != 0) |
| 100 | { |
| 101 | const TensorInfo tensor_info_output = output->clone()->set_tensor_shape(compute_mm_shape(*input0, *input1, is_interleaved_transposed, reshape_info)); |
| 102 | ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(output, &tensor_info_output); |
| 103 | ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input0, output); |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 104 | } |
| 105 | |
| 106 | return Status{}; |
| 107 | } |
| 108 | |
| 109 | inline std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input0, ITensorInfo *input1, ITensorInfo *output, |
Gian Marco Iodice | 750641d | 2018-05-08 12:01:57 +0100 | [diff] [blame] | 110 | bool is_interleaved_transposed, const GEMMReshapeInfo &reshape_info, GPUTarget gpu_target, |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 111 | ElementsProcessed &num_elements_processed) |
| 112 | { |
| 113 | bool window_changed = false; |
| 114 | Window win{}; |
Isabella Gottardi | 8e74f44 | 2018-03-01 16:42:00 +0000 | [diff] [blame] | 115 | Window win_out{}; |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 116 | |
| 117 | const DataType data_type = input0->data_type(); |
| 118 | unsigned int &num_elems_processed_per_iteration_x = num_elements_processed[0]; |
| 119 | unsigned int &num_elems_processed_per_iteration_y = num_elements_processed[1]; |
Gian Marco Iodice | d39e2b1 | 2018-08-06 14:31:15 +0100 | [diff] [blame] | 120 | bool reinterpret_input_as_3d = reshape_info.reinterpret_input_as_3d(); |
Gian Marco Iodice | 3139f03 | 2018-11-05 14:26:32 +0000 | [diff] [blame] | 121 | bool reinterpret_output_as_3d = (reshape_info.depth_output_gemm3d() != 0); |
Gian Marco Iodice | d39e2b1 | 2018-08-06 14:31:15 +0100 | [diff] [blame] | 122 | |
| 123 | // In case both input and output have to be reinterpreted as 3D tensors, |
| 124 | // force reinterpret_input_as_3d and reinterpret_output_as_3d to be false. |
| 125 | if(reinterpret_input_as_3d == reinterpret_output_as_3d) |
| 126 | { |
| 127 | reinterpret_input_as_3d = false; |
| 128 | reinterpret_output_as_3d = false; |
| 129 | } |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 130 | |
Gian Marco Iodice | 750641d | 2018-05-08 12:01:57 +0100 | [diff] [blame] | 131 | // Output tensor auto inizialitation if not yet initialized |
Isabella Gottardi | c4f582e | 2018-10-11 19:14:55 +0100 | [diff] [blame] | 132 | auto_init_if_empty(*output, input0->clone()->set_tensor_shape(compute_mm_shape(*input0, *input1, is_interleaved_transposed, reshape_info))); |
Gian Marco Iodice | 750641d | 2018-05-08 12:01:57 +0100 | [diff] [blame] | 133 | |
Isabella Gottardi | 8e74f44 | 2018-03-01 16:42:00 +0000 | [diff] [blame] | 134 | TensorInfo tmp_info(*output); |
| 135 | |
Gian Marco Iodice | d39e2b1 | 2018-08-06 14:31:15 +0100 | [diff] [blame] | 136 | if(reinterpret_output_as_3d) |
Isabella Gottardi | 8e74f44 | 2018-03-01 16:42:00 +0000 | [diff] [blame] | 137 | { |
| 138 | // Since the output tensor has to be reinterpreted as 3D and the execute window is based on a 2D GEMM, |
| 139 | // the window needs to be constructed on the 2D collapsed version of the tensor |
| 140 | TensorShape tmp_shape(output->tensor_shape()); |
| 141 | tmp_shape.collapse(2U, 1U); |
| 142 | tmp_info.set_tensor_shape(tmp_shape); |
| 143 | } |
| 144 | |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 145 | if(is_interleaved_transposed) |
| 146 | { |
Gian Marco Iodice | 68a3f56 | 2018-07-26 11:44:03 +0100 | [diff] [blame] | 147 | // reinterpret_input_as_3d is not supported if is_interleaved_transposed is set |
Isabella Gottardi | c4f582e | 2018-10-11 19:14:55 +0100 | [diff] [blame] | 148 | ARM_COMPUTE_ERROR_ON(reshape_info.reinterpret_input_as_3d()); |
Gian Marco Iodice | 68a3f56 | 2018-07-26 11:44:03 +0100 | [diff] [blame] | 149 | |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 150 | // Configure kernel window |
| 151 | num_elems_processed_per_iteration_x = max_cl_vector_width / data_size_from_type(data_type); |
| 152 | num_elems_processed_per_iteration_y = 4; |
| 153 | |
Isabella Gottardi | 8e74f44 | 2018-03-01 16:42:00 +0000 | [diff] [blame] | 154 | // Note: bottom paddings are calculated manually as the output can be reinterpreted as 3D tensor |
| 155 | // The only way to set properly the paddings, it is to set those explicitly through the AccessWindowStatic |
| 156 | const int m = reshape_info.m(); |
| 157 | const int bottom_pad = (num_elems_processed_per_iteration_y - (m % num_elems_processed_per_iteration_y)) % num_elems_processed_per_iteration_y; |
| 158 | |
| 159 | win = calculate_max_window(tmp_info, Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y)); |
| 160 | win_out = calculate_max_window(*output, Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y)); |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 161 | |
| 162 | AccessWindowRectangle input0_access(input0, 0, 0, num_elems_processed_per_iteration_y, 1, 1.f, 0.25f); |
Georgios Pinitas | 535fedd | 2018-05-04 18:52:25 +0100 | [diff] [blame] | 163 | AccessWindowStatic input1_access(input1, 0, 0, |
| 164 | ceil_to_multiple(input1->dimension(0), num_elems_processed_per_iteration_x), |
| 165 | ceil_to_multiple(input1->dimension(1), num_elems_processed_per_iteration_y)); |
Isabella Gottardi | 8e74f44 | 2018-03-01 16:42:00 +0000 | [diff] [blame] | 166 | AccessWindowStatic output_access(output, 0, 0, |
| 167 | ceil_to_multiple(output->dimension(0), num_elems_processed_per_iteration_x), |
| 168 | output->dimension(1) + bottom_pad); |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 169 | |
Isabella Gottardi | 8e74f44 | 2018-03-01 16:42:00 +0000 | [diff] [blame] | 170 | window_changed = update_window_and_padding(win, input0_access, input1_access) || // window used by the execute_window_loop |
| 171 | update_window_and_padding(win_out, output_access); // window used to update the padding requirements of output tensor |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 172 | |
Isabella Gottardi | 8e74f44 | 2018-03-01 16:42:00 +0000 | [diff] [blame] | 173 | output_access.set_valid_region(win_out, ValidRegion(Coordinates(0, 0), output->tensor_shape())); |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 174 | } |
| 175 | else // The input tensors have not been reshaped |
| 176 | { |
| 177 | // Special case for 1xN, 2xN, 3xN and 4xN input0 tensor. num_elems_processed_per_iteration_x is set up for the default case. |
| 178 | num_elems_processed_per_iteration_x = max_cl_vector_width / data_size_from_type(data_type); |
| 179 | num_elems_processed_per_iteration_y = std::min(static_cast<int>(output->dimension(1)), 4); |
| 180 | |
Isabella Gottardi | 8e74f44 | 2018-03-01 16:42:00 +0000 | [diff] [blame] | 181 | // Note: bottom paddings are calculated manually as the output can be reinterpreted as 3D tensor |
| 182 | // The only way to set properly the paddings, it is to set those explicitly through the AccessWindowStatic |
Gian Marco Iodice | d39e2b1 | 2018-08-06 14:31:15 +0100 | [diff] [blame] | 183 | const int m = reinterpret_input_as_3d ? input0->tensor_shape()[1] * input0->tensor_shape()[2] : input0->tensor_shape()[1]; |
Isabella Gottardi | 8e74f44 | 2018-03-01 16:42:00 +0000 | [diff] [blame] | 184 | const int bottom_pad = (num_elems_processed_per_iteration_y - (m % num_elems_processed_per_iteration_y)) % num_elems_processed_per_iteration_y; |
| 185 | |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 186 | // Create kernels according to the architecture, data type and input size. |
Michalis Spyrou | a967611 | 2018-02-22 18:07:43 +0000 | [diff] [blame] | 187 | GPUTarget arch_target = get_arch_from_target(gpu_target); |
| 188 | if(arch_target == GPUTarget::BIFROST && data_type == DataType::F32) |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 189 | { |
Gian Marco | 1d25ed5 | 2017-12-16 19:33:50 +0000 | [diff] [blame] | 190 | num_elems_processed_per_iteration_x = (input1->dimension(0) <= 1000 && input0->num_dimensions() == 1) ? 2 : 4; |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 191 | } |
| 192 | |
| 193 | // Configure window |
Isabella Gottardi | 8e74f44 | 2018-03-01 16:42:00 +0000 | [diff] [blame] | 194 | win = calculate_max_window(tmp_info, Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y)); |
| 195 | win_out = calculate_max_window(*output, Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y)); |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 196 | |
Gian Marco Iodice | 68a3f56 | 2018-07-26 11:44:03 +0100 | [diff] [blame] | 197 | AccessWindowStatic input0_access(input0, 0, 0, input0->dimension(0), input0->dimension(1) + bottom_pad); |
Isabella Gottardi | 8e74f44 | 2018-03-01 16:42:00 +0000 | [diff] [blame] | 198 | AccessWindowStatic input1_access(input1, 0, 0, ceil_to_multiple(input1->dimension(0), num_elems_processed_per_iteration_x), input1->dimension(1)); |
| 199 | AccessWindowStatic output_access(output, 0, 0, |
| 200 | ceil_to_multiple(output->dimension(0), num_elems_processed_per_iteration_x), |
| 201 | output->dimension(1) + bottom_pad); |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 202 | |
Isabella Gottardi | 8e74f44 | 2018-03-01 16:42:00 +0000 | [diff] [blame] | 203 | window_changed = update_window_and_padding(win, input0_access, input1_access) || // window used by the execute_window_loop |
| 204 | update_window_and_padding(win_out, output_access); // window used to update the padding requirements of output tensor |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 205 | |
| 206 | Coordinates coord; |
| 207 | coord.set_num_dimensions(output->num_dimensions()); |
Isabella Gottardi | 8e74f44 | 2018-03-01 16:42:00 +0000 | [diff] [blame] | 208 | output_access.set_valid_region(win_out, ValidRegion(coord, output->tensor_shape())); |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 209 | } |
| 210 | |
Gian Marco | ae2af74 | 2018-02-15 12:35:44 +0000 | [diff] [blame] | 211 | // Collapse along the Z direction |
| 212 | // This collapse needs to be here in order to tune the Z dimension of LWS |
Gian Marco Iodice | 81b28c4 | 2018-03-29 10:29:36 +0100 | [diff] [blame] | 213 | Window collapsed = win; |
| 214 | const unsigned int dimension_to_collapse = std::min(static_cast<unsigned int>(output->num_dimensions()), 2u); |
| 215 | collapsed = win.collapse(win, dimension_to_collapse); |
Gian Marco | ae2af74 | 2018-02-15 12:35:44 +0000 | [diff] [blame] | 216 | |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 217 | Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{}; |
Gian Marco | ae2af74 | 2018-02-15 12:35:44 +0000 | [diff] [blame] | 218 | return std::make_pair(err, collapsed); |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 219 | } |
| 220 | } // namespace |
| 221 | |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 222 | CLGEMMMatrixMultiplyKernel::CLGEMMMatrixMultiplyKernel() |
Gian Marco Iodice | 68a3f56 | 2018-07-26 11:44:03 +0100 | [diff] [blame] | 223 | : _input0(nullptr), _input1(nullptr), _output(nullptr), _slide_matrix_b(true), _reinterpret_input_as_3d(false), _reinterpret_output_as_3d(false) |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 224 | { |
| 225 | } |
| 226 | |
Vidhya Sudhan Loganathan | a25d16c | 2018-11-16 11:33:12 +0000 | [diff] [blame] | 227 | void CLGEMMMatrixMultiplyKernel::configure(const ICLTensor *input0, const ICLTensor *input1, ICLTensor *output, float alpha, bool is_interleaved_transposed, const GEMMReshapeInfo &reshape_info, |
| 228 | bool fp_mixed_precision) |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 229 | { |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 230 | ARM_COMPUTE_ERROR_ON_NULLPTR(input0, input1, output); |
| 231 | |
| 232 | // Perform validate step |
Vidhya Sudhan Loganathan | a25d16c | 2018-11-16 11:33:12 +0000 | [diff] [blame] | 233 | ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input0->info(), input1->info(), output->info(), is_interleaved_transposed, reshape_info, fp_mixed_precision)); |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 234 | |
Gian Marco Iodice | 68a3f56 | 2018-07-26 11:44:03 +0100 | [diff] [blame] | 235 | _input0 = input0; |
| 236 | _input1 = input1; |
| 237 | _output = output; |
| 238 | _reinterpret_input_as_3d = reshape_info.reinterpret_input_as_3d(); |
Gian Marco Iodice | 3139f03 | 2018-11-05 14:26:32 +0000 | [diff] [blame] | 239 | _reinterpret_output_as_3d = (reshape_info.depth_output_gemm3d() != 0); |
Gian Marco Iodice | 68a3f56 | 2018-07-26 11:44:03 +0100 | [diff] [blame] | 240 | |
Gian Marco Iodice | d39e2b1 | 2018-08-06 14:31:15 +0100 | [diff] [blame] | 241 | // In case both input and output have to be reinterpreted as 3D tensors, |
| 242 | // force reinterpret_input_as_3d and reinterpret_output_as_3d to be false. |
| 243 | if(_reinterpret_input_as_3d == _reinterpret_output_as_3d) |
| 244 | { |
| 245 | _reinterpret_input_as_3d = false; |
| 246 | _reinterpret_output_as_3d = false; |
| 247 | } |
| 248 | |
Gian Marco Iodice | 68a3f56 | 2018-07-26 11:44:03 +0100 | [diff] [blame] | 249 | // Check if we need to slide the matrix B |
| 250 | const unsigned int num_dimensions_input0 = _reinterpret_input_as_3d ? _input0->info()->num_dimensions() - 1 : _input0->info()->num_dimensions(); |
| 251 | |
| 252 | _slide_matrix_b = (_input1->info()->num_dimensions() >= num_dimensions_input0); |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 253 | |
Isabella Gottardi | 8e74f44 | 2018-03-01 16:42:00 +0000 | [diff] [blame] | 254 | const DataType data_type = input0->info()->data_type(); |
Isabella Gottardi | 8e74f44 | 2018-03-01 16:42:00 +0000 | [diff] [blame] | 255 | |
| 256 | // Get target architecture |
| 257 | GPUTarget gpu_target = get_target(); |
| 258 | |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 259 | ElementsProcessed num_elements_processed{}; |
| 260 | |
| 261 | // Configure kernel window |
Gian Marco Iodice | 750641d | 2018-05-08 12:01:57 +0100 | [diff] [blame] | 262 | auto win_config = validate_and_configure_window(input0->info(), input1->info(), output->info(), is_interleaved_transposed, reshape_info, gpu_target, num_elements_processed); |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 263 | ARM_COMPUTE_ERROR_THROW_ON(win_config.first); |
Anthony Barbier | b6eb353 | 2018-08-08 13:20:04 +0100 | [diff] [blame] | 264 | ICLKernel::configure_internal(win_config.second); |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 265 | |
Anton Lokhmotov | 3e80c7f | 2017-11-20 11:02:10 +0000 | [diff] [blame] | 266 | // Create build options |
| 267 | CLBuildOptions build_opts; |
Anton Lokhmotov | 3e80c7f | 2017-11-20 11:02:10 +0000 | [diff] [blame] | 268 | |
Anton Lokhmotov | 3e80c7f | 2017-11-20 11:02:10 +0000 | [diff] [blame] | 269 | // Only define ALPHA when alpha is not 1.0f. This avoids performing unnecessary multiplications. |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 270 | if(std::abs(1.0f - alpha) > 0.00001f) |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 271 | { |
Vidhya Sudhan Loganathan | 7485d5a | 2018-07-04 09:34:00 +0100 | [diff] [blame] | 272 | build_opts.add_option("-DALPHA=" + float_to_string_with_full_precision(alpha)); |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 273 | } |
Gian Marco Iodice | 68a3f56 | 2018-07-26 11:44:03 +0100 | [diff] [blame] | 274 | build_opts.add_option_if(_reinterpret_input_as_3d, "-DREINTERPRET_INPUT_AS_3D"); |
| 275 | build_opts.add_option_if(_reinterpret_output_as_3d, "-DREINTERPRET_OUTPUT_AS_3D"); |
| 276 | build_opts.add_option_if(_reinterpret_input_as_3d || _reinterpret_output_as_3d, "-DHEIGHT_GEMM3D=" + support::cpp11::to_string(output->info()->dimension(1))); |
| 277 | build_opts.add_option_if(_reinterpret_input_as_3d || _reinterpret_output_as_3d, "-DDEPTH_GEMM3D=" + support::cpp11::to_string(output->info()->dimension(2))); |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 278 | |
Gian Marco Iodice | d2fab73 | 2018-03-02 11:18:12 +0000 | [diff] [blame] | 279 | // Do not slide matrix B if _slide_matrix_b = false |
| 280 | build_opts.add_option_if(!_slide_matrix_b, "-DMATRIX_B_DEPTH=" + support::cpp11::to_string(input1->info()->dimension(2))); |
| 281 | |
Gian Marco Iodice | bb36a8e | 2018-04-19 12:05:08 +0100 | [diff] [blame] | 282 | const bool is_bifrost = get_arch_from_target(gpu_target) == GPUTarget::BIFROST; |
| 283 | |
Anton Lokhmotov | 3e80c7f | 2017-11-20 11:02:10 +0000 | [diff] [blame] | 284 | std::string kernel_name; |
Gian Marco Iodice | edfa9f4 | 2017-08-15 11:45:22 +0100 | [diff] [blame] | 285 | if(is_interleaved_transposed) |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 286 | { |
Gian Marco | 36a0a46 | 2018-01-12 10:21:40 +0000 | [diff] [blame] | 287 | const int mult_transpose1xW_width = reshape_info.mult_transpose1xW_width(); |
| 288 | const int mult_interleave4x4_height = reshape_info.mult_interleave4x4_height(); |
| 289 | |
Anton Lokhmotov | 3e80c7f | 2017-11-20 11:02:10 +0000 | [diff] [blame] | 290 | build_opts.add_option("-DCOLS_B=" + support::cpp11::to_string(input1->info()->dimension(0))); |
Gian Marco | 36a0a46 | 2018-01-12 10:21:40 +0000 | [diff] [blame] | 291 | build_opts.add_option("-DMULT_TRANSPOSE1XW_WIDTH=" + support::cpp11::to_string(mult_transpose1xW_width)); |
| 292 | build_opts.add_option("-DMULT_INTERLEAVE4X4_HEIGHT=" + support::cpp11::to_string(mult_interleave4x4_height)); |
| 293 | |
Gian Marco Iodice | bb36a8e | 2018-04-19 12:05:08 +0100 | [diff] [blame] | 294 | if(is_data_type_float(data_type) && is_bifrost) |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 295 | { |
Gian Marco Iodice | bb36a8e | 2018-04-19 12:05:08 +0100 | [diff] [blame] | 296 | kernel_name = "gemm_mm_interleaved_transposed_" + lower_string(string_from_data_type(data_type)) + "_bifrost"; |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 297 | } |
| 298 | else |
| 299 | { |
Anton Lokhmotov | 3e80c7f | 2017-11-20 11:02:10 +0000 | [diff] [blame] | 300 | kernel_name = "gemm_mm_interleaved_transposed_" + lower_string(string_from_data_type(data_type)); |
Vidhya Sudhan Loganathan | 38d93bd | 2018-11-20 15:38:13 +0000 | [diff] [blame] | 301 | if(fp_mixed_precision && data_type == DataType::F16) |
| 302 | { |
| 303 | // currently wider accumulator is only supported for fp16 kernels. |
| 304 | kernel_name += "_acc32"; |
| 305 | } |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 306 | } |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 307 | } |
Gian Marco Iodice | edfa9f4 | 2017-08-15 11:45:22 +0100 | [diff] [blame] | 308 | else // The input tensors have not been reshaped |
| 309 | { |
Anton Lokhmotov | 3e80c7f | 2017-11-20 11:02:10 +0000 | [diff] [blame] | 310 | build_opts.add_option("-DCOLS_A=" + support::cpp11::to_string(input0->info()->dimension(0))); |
Gian Marco Iodice | e52a300 | 2018-04-11 15:59:10 +0100 | [diff] [blame] | 311 | build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(data_type)); |
Gian Marco Iodice | edfa9f4 | 2017-08-15 11:45:22 +0100 | [diff] [blame] | 312 | |
Anton Lokhmotov | 3e80c7f | 2017-11-20 11:02:10 +0000 | [diff] [blame] | 313 | // Create kernels according to the architecture, data type and input size. |
Gian Marco Iodice | bb36a8e | 2018-04-19 12:05:08 +0100 | [diff] [blame] | 314 | if(is_data_type_float(data_type) && is_bifrost) |
Gian Marco Iodice | edfa9f4 | 2017-08-15 11:45:22 +0100 | [diff] [blame] | 315 | { |
Gian Marco Iodice | e52a300 | 2018-04-11 15:59:10 +0100 | [diff] [blame] | 316 | kernel_name = "gemm_mm_floating_point"; |
| 317 | |
| 318 | if(input0->info()->num_dimensions() != 1) |
Gian Marco Iodice | fd68311 | 2018-04-17 09:52:44 +0100 | [diff] [blame] | 319 | { |
Gian Marco Iodice | e52a300 | 2018-04-11 15:59:10 +0100 | [diff] [blame] | 320 | kernel_name += "_" + lower_string(string_from_data_type(data_type)) + "_bifrost"; |
Vidhya Sudhan Loganathan | 38d93bd | 2018-11-20 15:38:13 +0000 | [diff] [blame] | 321 | if(fp_mixed_precision && data_type == DataType::F16) |
| 322 | { |
| 323 | // currently wider accumulator is only supported for fp16 kernels. |
| 324 | kernel_name += "_acc32"; |
| 325 | } |
Gian Marco Iodice | e52a300 | 2018-04-11 15:59:10 +0100 | [diff] [blame] | 326 | } |
| 327 | else if(input1->info()->dimension(0) <= 1000 && data_type == DataType::F32) |
| 328 | { |
| 329 | // The first kernel is optimized for the case of 1000 or less output elements (e.g. FC8 of AlexNet and VGG-16, and |
| 330 | // FC1 of Inception v3). The second kernel is optimized for the case of greater than 1000 output elements (e.g. |
| 331 | // FC6 and FC7 of AlexNet and VGG-16). |
| 332 | kernel_name += "_" + lower_string(string_from_data_type(data_type)) + "_bifrost_1000"; |
Gian Marco Iodice | fd68311 | 2018-04-17 09:52:44 +0100 | [diff] [blame] | 333 | } |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 334 | |
Anton Lokhmotov | 3e80c7f | 2017-11-20 11:02:10 +0000 | [diff] [blame] | 335 | // The work-group size equal to the Bifrost quad size has been proved to be optimal for these kernels |
| 336 | // via exhaustive autotuning over a range of representative layer configurations. |
Anthony Barbier | b6eb353 | 2018-08-08 13:20:04 +0100 | [diff] [blame] | 337 | set_lws_hint(cl::NDRange(4)); |
Gian Marco Iodice | edfa9f4 | 2017-08-15 11:45:22 +0100 | [diff] [blame] | 338 | } |
Anton Lokhmotov | 3e80c7f | 2017-11-20 11:02:10 +0000 | [diff] [blame] | 339 | else // (MIDGARD and F32) or (F16) |
| 340 | { |
Anton Lokhmotov | 3e80c7f | 2017-11-20 11:02:10 +0000 | [diff] [blame] | 341 | kernel_name = "gemm_mm_floating_point"; |
| 342 | } |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 343 | build_opts.add_option("-DNUM_ELEMS_PROCESSED_PER_THREAD_Y=" + support::cpp11::to_string(num_elements_processed.y())); |
| 344 | build_opts.add_option("-DNUM_ELEMS_PROCESSED_PER_THREAD_X=" + support::cpp11::to_string(num_elements_processed.x())); |
Gian Marco Iodice | edfa9f4 | 2017-08-15 11:45:22 +0100 | [diff] [blame] | 345 | } |
Anton Lokhmotov | 3e80c7f | 2017-11-20 11:02:10 +0000 | [diff] [blame] | 346 | |
| 347 | // Create kernel |
| 348 | _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel(kernel_name, build_opts.options())); |
| 349 | |
| 350 | // Set config_id for enabling LWS tuning |
| 351 | _config_id = "gemm_"; |
| 352 | _config_id += (is_interleaved_transposed ? "reshaped_" : ""); |
Vidhya Sudhan Loganathan | a25d16c | 2018-11-16 11:33:12 +0000 | [diff] [blame] | 353 | _config_id += (fp_mixed_precision ? "fp_mixed_" : ""); |
Gian Marco Iodice | 68a3f56 | 2018-07-26 11:44:03 +0100 | [diff] [blame] | 354 | _config_id += (_reinterpret_input_as_3d ? "3di_" : ""); |
| 355 | _config_id += (_reinterpret_output_as_3d ? "3do_" : ""); |
Anton Lokhmotov | 3e80c7f | 2017-11-20 11:02:10 +0000 | [diff] [blame] | 356 | _config_id += lower_string(string_from_data_type(input0->info()->data_type())); |
| 357 | _config_id += "_"; |
| 358 | _config_id += support::cpp11::to_string(output->info()->dimension(1)); |
| 359 | _config_id += "_"; |
| 360 | _config_id += support::cpp11::to_string(output->info()->dimension(0)); |
| 361 | _config_id += "_"; |
Gian Marco | ae2af74 | 2018-02-15 12:35:44 +0000 | [diff] [blame] | 362 | _config_id += support::cpp11::to_string(output->info()->dimension(2)); |
| 363 | _config_id += "_"; |
| 364 | _config_id += support::cpp11::to_string(output->info()->dimension(3)); |
| 365 | _config_id += "_"; |
Anton Lokhmotov | 3e80c7f | 2017-11-20 11:02:10 +0000 | [diff] [blame] | 366 | _config_id += (is_interleaved_transposed ? support::cpp11::to_string(input1->info()->dimension(0)) : support::cpp11::to_string(input1->info()->dimension(1))); |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 367 | } |
| 368 | |
Gian Marco | 36a0a46 | 2018-01-12 10:21:40 +0000 | [diff] [blame] | 369 | Status CLGEMMMatrixMultiplyKernel::validate(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *output, float alpha, bool is_interleaved_transposed, |
Vidhya Sudhan Loganathan | a25d16c | 2018-11-16 11:33:12 +0000 | [diff] [blame] | 370 | const GEMMReshapeInfo &reshape_info, GPUTarget gpu_target, bool fp_mixed_precision) |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 371 | { |
Gian Marco | 36a0a46 | 2018-01-12 10:21:40 +0000 | [diff] [blame] | 372 | // Note: num_elements_processed will be set in validate_and_configure_window() |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 373 | ElementsProcessed num_elements_processed{}; |
| 374 | ARM_COMPUTE_UNUSED(alpha); |
Vidhya Sudhan Loganathan | a25d16c | 2018-11-16 11:33:12 +0000 | [diff] [blame] | 375 | ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input0, input1, output, is_interleaved_transposed, reshape_info, fp_mixed_precision)); |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 376 | ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input0->clone().get(), |
| 377 | input1->clone().get(), |
| 378 | output->clone().get(), |
| 379 | is_interleaved_transposed, |
Gian Marco Iodice | 750641d | 2018-05-08 12:01:57 +0100 | [diff] [blame] | 380 | reshape_info, |
Georgios Pinitas | 358ca20 | 2017-12-07 16:47:52 +0000 | [diff] [blame] | 381 | gpu_target, |
| 382 | num_elements_processed) |
| 383 | .first); |
| 384 | |
| 385 | return Status{}; |
| 386 | } |
| 387 | |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 388 | void CLGEMMMatrixMultiplyKernel::run(const Window &window, cl::CommandQueue &queue) |
| 389 | { |
| 390 | ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this); |
| 391 | ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window); |
| 392 | |
Gian Marco | ae2af74 | 2018-02-15 12:35:44 +0000 | [diff] [blame] | 393 | if(_input1->info()->num_dimensions() < 3) |
| 394 | { |
| 395 | // The stride_z for matrix B must be zero if we do not slice |
| 396 | ARM_COMPUTE_ERROR_ON(_input1->info()->strides_in_bytes()[3] != 0); |
| 397 | } |
| 398 | |
| 399 | Window slice = window.first_slice_window_3D(); |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 400 | Window slice_matrix_b = slice; |
Gian Marco Iodice | edfa9f4 | 2017-08-15 11:45:22 +0100 | [diff] [blame] | 401 | |
| 402 | slice_matrix_b.set(Window::DimX, Window::Dimension(0, 1, 1)); |
| 403 | slice_matrix_b.set(Window::DimY, Window::Dimension(0, 1, 1)); |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 404 | |
Gian Marco Iodice | 68a3f56 | 2018-07-26 11:44:03 +0100 | [diff] [blame] | 405 | if(_reinterpret_input_as_3d) |
Isabella Gottardi | 8e74f44 | 2018-03-01 16:42:00 +0000 | [diff] [blame] | 406 | { |
Isabella Gottardi | b92805b | 2018-09-28 18:24:27 +0100 | [diff] [blame] | 407 | // Pass bottom paddings to the kernel if the input has to be reinterpreted as 3D tensor |
Georgios Pinitas | e8bd2c7 | 2018-07-11 15:54:56 +0100 | [diff] [blame] | 408 | const unsigned int idx0 = 3 * num_arguments_per_2D_tensor() + 3; |
Gian Marco Iodice | 68a3f56 | 2018-07-26 11:44:03 +0100 | [diff] [blame] | 409 | const unsigned int total_cross_plane_pad = _input0->info()->padding().top + _input0->info()->padding().bottom; |
| 410 | _kernel.setArg<cl_uint>(idx0, static_cast<unsigned int>(total_cross_plane_pad)); |
| 411 | } |
| 412 | |
| 413 | if(_reinterpret_output_as_3d) |
| 414 | { |
| 415 | // Pass bottom paddings to the kernel if the output has to be reinterpreted as 3D tensor |
| 416 | const unsigned int idx0 = 3 * num_arguments_per_2D_tensor() + 3 + (_reinterpret_input_as_3d ? 1 : 0); |
Georgios Pinitas | e8bd2c7 | 2018-07-11 15:54:56 +0100 | [diff] [blame] | 417 | const unsigned int total_cross_plane_pad = _output->info()->padding().top + _output->info()->padding().bottom; |
| 418 | _kernel.setArg<cl_uint>(idx0, static_cast<unsigned int>(total_cross_plane_pad)); |
Isabella Gottardi | 8e74f44 | 2018-03-01 16:42:00 +0000 | [diff] [blame] | 419 | } |
| 420 | |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 421 | do |
| 422 | { |
| 423 | Window slice_b = slice; |
| 424 | // Don't slice matrix B along the z dimension if matrix B has just 2 dimensions and matrix A more than 2 |
Gian Marco | ae2af74 | 2018-02-15 12:35:44 +0000 | [diff] [blame] | 425 | // This scenario can happen when the matrix multiplication is used to perform a convolution operation |
Gian Marco Iodice | d2fab73 | 2018-03-02 11:18:12 +0000 | [diff] [blame] | 426 | if(!_slide_matrix_b) |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 427 | { |
| 428 | slice_b = slice_matrix_b; |
| 429 | } |
| 430 | |
| 431 | unsigned int idx = 0; |
| 432 | add_2D_tensor_argument(idx, _input0, slice); |
| 433 | add_2D_tensor_argument(idx, _input1, slice_b); |
| 434 | add_2D_tensor_argument(idx, _output, slice); |
Gian Marco Iodice | d2fab73 | 2018-03-02 11:18:12 +0000 | [diff] [blame] | 435 | _kernel.setArg<cl_uint>(idx++, static_cast<unsigned int>(_input0->info()->strides_in_bytes()[2])); |
| 436 | _kernel.setArg<cl_uint>(idx++, static_cast<unsigned int>(_input1->info()->strides_in_bytes()[2])); |
| 437 | _kernel.setArg<cl_uint>(idx++, static_cast<unsigned int>(_output->info()->strides_in_bytes()[2])); |
Anthony Barbier | b6eb353 | 2018-08-08 13:20:04 +0100 | [diff] [blame] | 438 | enqueue(queue, *this, slice, lws_hint()); |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 439 | } |
Gian Marco | ae2af74 | 2018-02-15 12:35:44 +0000 | [diff] [blame] | 440 | while(window.slide_window_slice_3D(slice)); |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 441 | } |