• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1/*
2 * Copyright (c) 2018-2020 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(SRC_WIDTH) && defined(SRC_HEIGHT) && defined(SRC_DEPTH)
27
28/** This opencl kernel flattens the first 3 dimensions of the input tensor
29 *
30 * @note Datatype should be given as a preprocessor argument using -DDATA_TYPE=type. e.g. -DDATA_TYPE=float
31 * @note The width, height and depth of the input tensor must be passed at compile time using -DSRC_WIDTH, -DSRC_HEIGHT and -DSRC_DEPTH. e.g. -DSRC_WIDTH=24, -DSRC_HEIGHT=24, -DSRC_DEPTH=16
32 * @note If the output has 3 dimensions, the 2nd dimension of the output tensor must be passed at compile time using -DDST_DIM1. e.g -DDST_DIM1=3
33 *
34 * @param[in]  src_ptr                           Pointer to the source tensor. Supported data types: All
35 * @param[in]  src_stride_x                      Stride of the source tensor in X dimension (in bytes)
36 * @param[in]  src_step_x                        src_stride_x * number of elements along X processed per workitem(in bytes)
37 * @param[in]  src_stride_y                      Stride of the source tensor in Y dimension (in bytes)
38 * @param[in]  src_step_y                        src_stride_y * number of elements along Y processed per workitem(in bytes)
39 * @param[in]  src_stride_z                      Stride of the source tensor in Z dimension (in bytes)
40 * @param[in]  src_step_z                        src_stride_z * number of elements along Y processed per workitem(in bytes)
41 * @param[in]  src_stride_w                      Stride of the source tensor in W dimension (in bytes)
42 * @param[in]  src_step_w                        src_stride_w * number of elements along Y processed per workitem(in bytes)
43 * @param[in]  src_offset_first_element_in_bytes The offset of the first element in the source tensor
44 * @param[out] dst_ptr                           Pointer to the destination tensor. Same as @p src_ptr
45 * @param[in]  dst_stride_x                      Stride of the destination tensor in X dimension (in bytes)
46 * @param[in]  dst_step_x                        dst_stride_x * number of elements along X processed per workitem(in bytes)
47 * @param[in]  dst_stride_y                      Stride of the destination tensor in Y dimension (in bytes)
48 * @param[in]  dst_step_y                        dst_stride_y * number of elements along Y processed per workitem(in bytes)
49 * @param[in]  dst_stride_z                      Stride of the destination tensor in Z dimension (in bytes)
50 * @param[in]  dst_step_z                        dst_stride_z * number of elements along Z processed per workitem(in bytes)
51 * @param[in]  dst_offset_first_element_in_bytes The offset of the first element in the destination tensor
52 */
53__kernel void flatten(
54    TENSOR4D_DECLARATION(src),
55    TENSOR3D_DECLARATION(dst))
56{
57    Tensor4D src = CONVERT_TO_TENSOR4D_STRUCT(src, SRC_DEPTH);
58
59    uint c  = get_global_id(2) % SRC_DEPTH; // input feature map
60    uint b0 = get_global_id(2) / SRC_DEPTH; // batch id
61    uint b1 = 0;
62
63#if defined(DST_DIM1)
64    uint b_tmp = b0;
65    b0         = b_tmp % DST_DIM1; // batch id0
66    b1         = b_tmp / DST_DIM1; // batch id1
67#endif                             // defined(DST_DIM1)
68
69    __global uchar *output_ptr = dst_ptr + dst_offset_first_element_in_bytes + (get_global_id(0) + get_global_id(1) * (uint)SRC_WIDTH + c * (uint)(SRC_WIDTH * SRC_HEIGHT)) * sizeof(
70                                     DATA_TYPE) + b0 * dst_stride_y + b1 * dst_stride_z;
71
72    *((__global DATA_TYPE *)output_ptr) = *((__global DATA_TYPE *)src.ptr);
73}
74#endif // defined(DATA_TYPE) && defined(SRC_WIDTH) && defined(SRC_HEIGHT)