xref: /aosp_15_r20/external/ComputeLibrary/src/core/CL/cl_kernels/common/reverse.cl (revision c217d954acce2dbc11938adb493fc0abd69584f3)
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(NUM_REVERSE_DIMS)
27
28#if NUM_REVERSE_DIMS > 4
29#error("Reversing more than 4 dimensions is not currently supported")
30#endif /* NUM_REVERSE_DIMS > 4 */
31
32/** Performs reverse along the specified axis.
33 *
34 * @note The data type must be given as a preprocessor argument using -DDATA_TYPE=num. e.g. -DDATA_TYPE=uint
35 * @note The number of dimensions to reverse must be given as a preprocessor argument using -DNUM_REVERSE_DIMS=num, e.g. -DNUM_REVERSE_DIMS=3
36 *
37 * @param[in]  src_ptr                            Pointer to the source tensor. Supported data types: All
38 * @param[in]  src_stride_x                       Stride of the first source tensor in X dimension (in bytes)
39 * @param[in]  src_step_x                         src_stride_x * number of elements along X processed per workitem(in bytes)
40 * @param[in]  src_stride_y                       Stride of the first source tensor in Y dimension (in bytes)
41 * @param[in]  src_step_y                         src_stride_y * number of elements along Y processed per workitem(in bytes)
42 * @param[in]  src_stride_z                       Stride of the first source tensor in Z dimension (in bytes)
43 * @param[in]  src_step_z                         src_stride_z * number of elements along Z processed per workitem(in bytes)
44 * @param[in]  src_stride_w                       Stride of the first source tensor in Z dimension (in bytes)
45 * @param[in]  src_step_w                         src_stride_z * number of elements along Z processed per workitem(in bytes)
46 * @param[in]  src_offset_first_element_in_bytes  The offset of the first element in the first source tensor
47 * @param[in]  axis_ptr                           Pointer to the source vector. Supported data types: U32
48 * @param[in]  axis_stride_x                      Stride of the first source tensor in X dimension (in bytes)
49 * @param[in]  axis_step_x                        src_stride_x * number of elements along X processed per workitem(in bytes)
50 * @param[in]  axis_offset_first_element_in_bytes The offset of the first element in the first source tensor
51 * @param[out] dst_ptr                            Pointer to the destination tensor. Supported data types: same as @p src_ptr
52 * @param[in]  dst_stride_x                       Stride of the destination tensor in X dimension (in bytes)
53 * @param[in]  dst_step_x                         output_stride_x * number of elements along X processed per workitem(in bytes)
54 * @param[in]  dst_stride_y                       Stride of the destination tensor in Y dimension (in bytes)
55 * @param[in]  dst_step_y                         output_stride_y * number of elements along Y processed per workitem(in bytes)
56 * @param[in]  dst_stride_z                       Stride of the destination tensor in Z dimension (in bytes)
57 * @param[in]  dst_step_z                         output_stride_z * number of elements along Z processed per workitem(in bytes)
58 * @param[in]  dst_stride_w                       Stride of the destination tensor in Z dimension (in bytes)
59 * @param[in]  dst_step_w                         output_stride_z * number of elements along Z processed per workitem(in bytes)
60 * @param[in]  dst_offset_first_element_in_bytes  The offset of the first element in the destination tensor
61 */
62__kernel void reverse(TENSOR4D_DECLARATION(src),
63                      VECTOR_DECLARATION(axis),
64                      TENSOR4D_DECLARATION(dst),
65                      const uint width,
66                      const uint height,
67                      const uint depth,
68                      const uint batches)
69{
70    Tensor4D src  = CONVERT_TO_TENSOR4D_STRUCT(src, depth);
71    Vector   axis = CONVERT_TO_VECTOR_STRUCT_NO_STEP(axis);
72    Tensor4D dst  = CONVERT_TO_TENSOR4D_STRUCT_NO_STEP(dst, depth);
73
74    const uint x_in = get_global_id(0);
75    const uint y_in = get_global_id(1);
76    const uint z_in = get_global_id(2) % depth;
77    const uint w_in = get_global_id(2) / depth;
78
79    const uint4 dims       = (uint4)(0, 1, 2, 3);
80    int4        to_reverse = (int4)(0, 0, 0, 0);
81#if NUM_REVERSE_DIMS == 1
82    const uint index = *((__global uint *)axis.ptr);
83    to_reverse       = (uint4)index == dims;
84#elif NUM_REVERSE_DIMS == 2
85    const uint2 indices = vload2(0, (__global uint *)axis.ptr);
86    to_reverse          = ((uint4)indices.s0 == dims) || ((uint4)indices.s1 == dims);
87#elif NUM_REVERSE_DIMS == 3
88    const uint2 indices01 = vload2(0, (__global uint *)axis.ptr);
89    const uint index2     = *((__global uint *)axis.ptr + 2);
90    to_reverse            = ((uint4)indices01.s0 == dims) || ((uint4)indices01.s1 == dims) || ((uint4)index2 == dims);
91#else  /* NUM_REVERSE_DIMS == 3 */
92    const uint4 indices = vload4(0, (__global uint *)axis.ptr);
93    to_reverse          = ((uint4)indices.s0 == dims) || ((uint4)indices.s1 == dims) || ((uint4)indices.s2 == dims) || ((uint4)indices.s3 == dims);
94#endif /* NUM_REVERSE_DIMS == 1 */
95    const uint x_out = to_reverse.s0 ? width - x_in - 1 : x_in;
96    const uint y_out = to_reverse.s1 ? height - y_in - 1 : y_in;
97    const uint z_out = to_reverse.s2 ? depth - z_in - 1 : z_in;
98    const uint w_out = to_reverse.s3 ? batches - w_in - 1 : w_in;
99
100    *((__global DATA_TYPE *)tensor4D_offset(&dst, x_out, y_out, z_out, w_out)) = *((__global DATA_TYPE *)src.ptr);
101}
102#endif // defined(DATA_TYPE) && defined(NUM_REVERSE_DIMS)
103