1/* 2 * Copyright (c) 2018-2021 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 "helpers.h" 25 26#if defined(DATA_TYPE) && defined(AXIS) 27 28/** Performs the Gather operation along the chosen axis 29 * @note Datatype should be given as a preprocessor argument using -DDATA_TYPE=type. e.g. -DDATA_TYPE=short 30 * @note Axis should be given as a preprocessor argument using -DAXIS=axis. e.g. -DAXIS=1 31 * @attention Output tensor depth should be given as a preprocessor argument using -DOUTPUT_DIM_Z=size. e.g. -DOUTPUT_DIM_Z=16 32 * @attention Input tensor depth should be given as a preprocessor argument using -DINPUT_DIM_Z=size. e.g. -DINPUT_DIM_Z=16 33 * 34 * 35 * @param[in] input_ptr Pointer to the source tensor. Supported data types: All 36 * @param[in] input_stride_x Stride of the source tensor in X dimension (in bytes) 37 * @param[in] input_step_x input_stride_x * number of elements along X processed per work item (in bytes) 38 * @param[in] input_stride_y Stride of the source tensor in Y dimension (in bytes) 39 * @param[in] input_step_y input_stride_y * number of elements along Y processed per work item (in bytes) 40 * @param[in] input_stride_z Stride of the source tensor in Y dimension (in bytes) 41 * @param[in] input_step_z input_stride_z * number of elements along Z processed per work item (in bytes) 42 * @param[in] input_stride_w Stride of the source tensor in Z dimension (in bytes) 43 * @param[in] input_step_w input_stride_w * number of elements along W processed per work item (in bytes) 44 * @param[in] input_offset_first_element_in_bytes Offset of the first element in the source tensor 45 * @param[in] indices_ptr Pointer to the indices vector. Supported data types: S32/U32. 46 * @param[in] indices_stride_x Stride of the indices vector in X dimension (in bytes) 47 * @param[in] indices_step_x input_stride_x * number of elements along X processed per work item (in bytes) 48 * @param[in] indices_offset_first_element_in_bytes Offset of the first element in the indices vector 49 * @param[out] output_ptr Pointer to the destination tensor. Supported data types: same as @p input_ptr 50 * @param[in] output_stride_x Stride of the destination tensor in X dimension (in bytes) 51 * @param[in] output_step_x output_stride_x * number of elements along X processed per work item (in bytes) 52 * @param[in] output_stride_y Stride of the destination tensor in Y dimension (in bytes) 53 * @param[in] output_step_y output_stride_y * number of elements along Y processed per work item (in bytes) 54 * @param[in] output_stride_z Stride of the destination tensor in Z dimension (in bytes) 55 * @param[in] output_step_z output_stride_z * number of elements along Z processed per work item (in bytes) 56 * @param[in] output_stride_w Stride of the destination tensor in W dimension (in bytes) 57 * @param[in] output_step_w output_stride_w * number of elements along W processed per work item (in bytes) 58 * @param[in] output_offset_first_element_in_bytes Offset of the first element in the destination tensor 59 */ 60__kernel void gather( 61 TENSOR4D_DECLARATION(input), 62 VECTOR_DECLARATION(indices), 63 TENSOR4D_DECLARATION(output)) 64{ 65 const int px = get_global_id(0); 66 const int py = get_global_id(1); 67 const int pz = get_global_id(2) % OUTPUT_DIM_Z; 68 const int pw = get_global_id(2) / OUTPUT_DIM_Z; 69 70 const Tensor4D input = CONVERT_TO_TENSOR4D_STRUCT_NO_STEP(input, INPUT_DIM_Z); 71 const Vector indices = CONVERT_TO_VECTOR_STRUCT_NO_STEP(indices); 72 Tensor4D output = CONVERT_TO_TENSOR4D_STRUCT(output, OUTPUT_DIM_Z); 73 74#if AXIS == 0 75 const uint index = *(__global const uint *)vector_offset(&indices, px); 76 __global const uchar *input_addr = tensor4D_offset(&input, index, py, pz, pw); 77#elif AXIS == 1 78 const uint index = *(__global const uint *)vector_offset(&indices, py); 79 __global const uchar *input_addr = tensor4D_offset(&input, px, index, pz, pw); 80#elif AXIS == 2 81 const uint index = *(__global const uint *)vector_offset(&indices, pz); 82 __global const uchar *input_addr = tensor4D_offset(&input, px, py, index, pw); 83#elif AXIS == 3 84 const uint index = *(__global const uint *)vector_offset(&indices, pw); 85 __global const uchar *input_addr = tensor4D_offset(&input, px, py, pz, index); 86#endif //AXIS 87 88 *(__global DATA_TYPE *)output.ptr = *((__global const DATA_TYPE *)input_addr); 89} 90 91#endif //defined(DATA_TYPE) && defined(AXIS)