xref: /aosp_15_r20/external/ComputeLibrary/src/core/CL/cl_kernels/common/crop_tensor.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) // Compile time constants
27
28/** Performs a tensor cropping.
29 *
30 * @param[in]  in_ptr                            Pointer to the source tensor. Supported data types: All
31 * @param[in]  in_stride_x                       Stride of the source tensor in X dimension (in bytes)
32 * @param[in]  in_step_x                         input_stride_x * number of elements along X processed per workitem(in bytes)
33 * @param[in]  in_stride_y                       Stride of the source tensor in Y dimension (in bytes)
34 * @param[in]  in_step_y                         input_stride_y * number of elements along Y processed per workitem(in bytes)
35 * @param[in]  in_stride_z                       Stride of the source tensor in Z dimension (in bytes)
36 * @param[in]  in_step_z                         input_stride_z * number of elements along Z processed per workitem(in bytes)
37 * @param[in]  in_offset_first_element_in_bytes  The offset of the first element in the source tensor
38 * @param[out] out_ptr                           Pointer to the destination tensor. Supported data types: F32
39 * @param[in]  out_stride_x                      Stride of the destination tensor in X dimension (in bytes)
40 * @param[in]  out_step_x                        output_stride_x * number of elements along X processed per workitem(in bytes)
41 * @param[in]  out_stride_y                      Stride of the destination tensor in Y dimension (in bytes)
42 * @param[in]  out_step_y                        output_stride_y * number of elements along Y processed per workitem(in bytes)
43 * @param[in]  out_stride_z                      Stride of the source tensor in Z dimension (in bytes)
44 * @param[in]  out_step_z                        output_stride_z * number of elements along Z processed per workitem(in bytes)
45 * @param[in]  out_offset_first_element_in_bytes The offset of the first element in the destination tensor
46 * @param[in]  in_offset_y                       The initial offset of the input address along Y.
47 * @param[in]  in_offset_z                       The initial offset of the input address along Z.
48 */
49__kernel void crop_tensor(
50    TENSOR3D_DECLARATION(in),
51    TENSOR3D_DECLARATION(out),
52    int in_offset_y,
53    int in_offset_z)
54{
55    Tensor3D in  = CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(in);
56    Tensor3D out = CONVERT_TO_TENSOR3D_STRUCT(out);
57
58    const int in_x = get_global_id(0) * (in_step_x / in_stride_x);
59
60#if defined(WIDTH_FLIPPED)
61    const int in_y = in_offset_y - get_global_id(1);
62#else  // defined(WIDTH_FLIPPED)
63    const int in_y                 = in_offset_y + get_global_id(1);
64#endif // defined(WIDTH_FLIPPED)
65
66#if defined(HEIGHT_FLIPPED)
67    const int in_z = in_offset_z - get_global_id(2);
68#else  // defined(HEIGHT_FLIPPED)
69    const int in_z                 = in_offset_z + get_global_id(2);
70#endif // defined(HEIGHT_FLIPPED)
71
72#if defined(VEC_SIZE)
73
74#if defined(LAST_ACCESSED_X)
75    // Check if access on width gets out of bounds
76    // If it does then shift access vector to access elements within bounds
77    const int shift = max((int)(get_global_id(0) * VEC_SIZE) - (int)LAST_ACCESSED_X, 0);
78    in.ptr -= shift * in.stride_x;
79    out.ptr -= shift * out.stride_x;
80#endif // defined(LAST_ACCESSED_X)
81
82    __global const uchar *input_addr = tensor3D_offset(&in, in_x, in_y, in_z);
83
84    // Load data
85    VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE)
86    data = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)input_addr);
87
88    // Store result
89    VSTORE(VEC_SIZE)
90    (CONVERT(data, VEC_DATA_TYPE(float, VEC_SIZE)), 0, (__global float *)out.ptr);
91#else  // defined(VEC_SIZE)
92    *((__global float *)(out.ptr)) = CONVERT(*((__global DATA_TYPE *)tensor3D_offset(&in, in_x, in_y, in_z)), float);
93#endif // defined(VEC_SIZE)
94}
95
96#endif // defined(DATA_TYPE) && defined(LAST_ACCESSED_X)