xref: /aosp_15_r20/external/ComputeLibrary/src/core/CL/cl_kernels/nhwc/depth_to_space.cl (revision c217d954acce2dbc11938adb493fc0abd69584f3)
1/*
2 * Copyright (c) 2019-2021 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(BLOCK_SHAPE) && defined(CHANNEL_SIZE)
27/** Depth to space transformation. (NHWC)
28 *
29 * @note Datatype should be given as a preprocessor argument using -DDATA_TYPE=type. e.g. -DDATA_TYPE=float
30 * @note The input tensor depth size must be passed at compile time using -DCHANNEL_SIZE. e.g. -DCHANNEL_SIZE=2
31 * @note The block shape must be passed at compile time using -DBLOCK_SHAPE. e.g. -DBLOCK_SHAPE=2
32 *
33 * @param[in]  input_ptr                            Pointer to the source tensor. Supported data types: All.
34 * @param[in]  input_stride_x                       Stride of the source tensor in X dimension (in bytes)
35 * @param[in]  input_step_x                         input_stride_x * number of elements along X processed per workitem(in bytes)
36 * @param[in]  input_stride_y                       Stride of the source tensor in Y dimension (in bytes)
37 * @param[in]  input_step_y                         input_stride_y * number of elements along Y processed per workitem(in bytes)
38 * @param[in]  input_stride_z                       Stride of the source tensor in Z dimension (in bytes)
39 * @param[in]  input_step_z                         input_stride_z * number of elements along Z processed per workitem(in bytes)
40 * @param[in]  input_offset_first_element_in_bytes  The offset of the first element in the first source tensor
41 * @param[in]  batch_id                             The input tensor batch id
42 * @param[out] output_ptr                           Pointer to the destination tensor. Supported data types: same as @p input_ptr
43 * @param[in]  output_stride_x                      Stride of the destination tensor in X dimension (in bytes)
44 * @param[in]  output_step_x                        output_stride_x * number of elements along X processed per workitem(in bytes)
45 * @param[in]  output_stride_y                      Stride of the destination tensor in Y dimension (in bytes)
46 * @param[in]  output_step_y                        output_stride_y * number of elements along Y processed per workitem(in bytes)
47 * @param[in]  output_stride_z                      Stride of the source tensor in Z dimension (in bytes)
48 * @param[in]  output_step_z                        output_stride_z * number of elements along Z processed per workitem(in bytes)
49 * @param[in]  output_offset_first_element_in_bytes The offset of the first element in the destination tensor
50 */
51__kernel void depth_to_space_nhwc(
52    TENSOR3D_DECLARATION(input),
53    const int batch_id,
54    TENSOR4D_DECLARATION(output))
55{
56    Tensor3D in  = CONVERT_TO_TENSOR3D_STRUCT(input);
57    Tensor4D out = CONVERT_TO_TENSOR4D_STRUCT_NO_STEP(output, 0);
58
59    const int r = (CHANNEL_SIZE / (BLOCK_SHAPE * BLOCK_SHAPE));
60    const int x = get_global_id(1);
61    const int y = get_global_id(2);
62    const int z = get_global_id(0) % r;
63
64    const int out_x = x * BLOCK_SHAPE + (get_global_id(0) / r) % BLOCK_SHAPE;
65    const int out_y = y * BLOCK_SHAPE + (get_global_id(0) / r) / BLOCK_SHAPE;
66
67    *((__global DATA_TYPE *)tensor4D_offset(&out, z, out_x, out_y, batch_id)) = *((__global DATA_TYPE *)in.ptr);
68}
69#endif // defined(DATA_TYPE) && defined(BLOCK_SHAPE) && defined(CHANNEL_SIZE)