xref: /aosp_15_r20/external/ComputeLibrary/src/core/CL/cl_kernels/nchw/normalize_planar_yuv_layer.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(VEC_SIZE)
27
28#define TYPE VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE)
29
30/** Apply normalize_planar_yuv layer on tensors with NCHW data layout.
31 *
32 * @note Data type should be given as a preprocessor argument using -DDATA_TYPE=type. e.g. -DDATA_TYPE=float
33 * @note Vector size should be given as a preprocessor argument using -DVEC_SIZE e.g. -DVEC_SIZE=8
34 * @note The depth of the input tensor should be given as a preprocessor argument using -DNUM_CHANNELS e.g. -DNUM_CHANNELS=8
35 *
36 * @param[in]  src_ptr                            Pointer to the first source tensor. Supported data types: F16/F32
37 * @param[in]  src_stride_x                       Stride of the first source tensor in X dimension (in bytes)
38 * @param[in]  src_step_x                         input_stride_x * number of elements along X processed per workitem(in bytes)
39 * @param[in]  src_stride_y                       Stride of the first source tensor in Y dimension (in bytes)
40 * @param[in]  src_step_y                         input_stride_y * number of elements along Y processed per workitem(in bytes)
41 * @param[in]  src_stride_z                       Stride of the first source tensor in Z dimension (in bytes)
42 * @param[in]  src_step_z                         input_stride_z * number of elements along Z processed per workitem(in bytes)
43 * @param[in]  src_offset_first_element_in_bytes  The offset of the first element in the first source tensor
44 * @param[out] dst_ptr                            Pointer to the destination tensor. Supported data types: 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                         output_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                         output_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                         output_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 * @param[in]  mean_ptr                           Pointer to the mean source tensor. Supported data types: same as @p src_ptr
53 * @param[in]  mean_stride_x                      Stride of the mean source tensor in X dimension (in bytes)
54 * @param[in]  mean_step_x                        mean_stride_x * number of elements along X processed per workitem(in bytes)
55 * @param[in]  mean_offset_first_element_in_bytes The offset of the first element in the mean source tensor
56 * @param[in]  std_ptr                            Pointer to the std tensor. Supported data types: same as @p src_ptr
57 * @param[in]  std_stride_x                       Stride of the std tensor in X dimension (in bytes)
58 * @param[in]  std_step_x                         std_stride_x * number of elements along X processed per workitem(in bytes)
59 * @param[in]  std_offset_first_element_in_bytes  The offset of the first element in the var source tensor
60 */
61__kernel void normalize_planar_yuv_layer_nchw(TENSOR3D_DECLARATION(src),
62                                              TENSOR3D_DECLARATION(dst),
63                                              VECTOR_DECLARATION(mean),
64                                              VECTOR_DECLARATION(std))
65{
66    Tensor3D src  = CONVERT_TO_TENSOR3D_STRUCT(src);
67    Tensor3D dst  = CONVERT_TO_TENSOR3D_STRUCT(dst);
68    Vector   mean = CONVERT_TO_VECTOR_STRUCT(mean);
69    Vector   std  = CONVERT_TO_VECTOR_STRUCT(std);
70
71    const uint current_slice = get_global_id(2) % NUM_CHANNELS;
72
73    const DATA_TYPE curr_mean = *((__global DATA_TYPE *)(mean.ptr + current_slice * sizeof(DATA_TYPE)));
74    const DATA_TYPE curr_std  = *((__global DATA_TYPE *)(std.ptr + current_slice * sizeof(DATA_TYPE)));
75
76    TYPE data = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)src.ptr);
77    TYPE res  = (data - curr_mean) / curr_std;
78
79    VSTORE(VEC_SIZE)
80    (res, 0, (__global DATA_TYPE *)dst.ptr);
81}
82#endif // defined(DATA_TYPE) && defined(VEC_SIZE)