xref: /aosp_15_r20/external/ComputeLibrary/src/core/CL/cl_kernels/nchw/scale.cl (revision c217d954acce2dbc11938adb493fc0abd69584f3)
1/*
2 * Copyright (c) 2016-2021 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#include "tile_helpers.h"
26
27/** Transforms four 2D coordinates. This is used to map the output coordinates to the input coordinates.
28 *
29 * @param[in] coord 2D coordinates to transform.
30 * @param[in] scale input/output scale ratio
31 *
32 * @return a float8 containing 4 2D transformed values in the input image.
33 */
34inline const float8 transform_nearest(const float2 coord, const float2 scale)
35{
36#ifdef SAMPLING_POLICY_TOP_LEFT
37    const float4 in_x_coords = (float4)(coord.s0, 1 + coord.s0, 2 + coord.s0, 3 + coord.s0);
38    const float4 new_x       = in_x_coords * (float4)(scale.s0);
39    const float4 new_y       = (float4)(coord.s1 * scale.s1);
40    return (float8)(new_x.s0, new_y.s0, new_x.s1, new_y.s1, new_x.s2, new_y.s2, new_x.s3, new_y.s3);
41#elif SAMPLING_POLICY_CENTER
42    const float4 in_x_coords = (float4)(coord.s0, 1 + coord.s0, 2 + coord.s0, 3 + coord.s0);
43    const float4 new_x       = (in_x_coords + ((float4)(0.5f))) * (float4)(scale.s0);
44    const float4 new_y       = (float4)((coord.s1 + 0.5f) * scale.s1);
45    return (float8)(new_x.s0, new_y.s0, new_x.s1, new_y.s1, new_x.s2, new_y.s2, new_x.s3, new_y.s3);
46#else /* SAMPLING_POLICY */
47#error("Unsupported sampling policy");
48#endif /* SAMPLING_POLICY */
49}
50
51/** Transforms four 2D coordinates. This is used to map the output coordinates to the input coordinates.
52 *
53 * @param[in] coord 2D coordinates to transform.
54 * @param[in] scale input/output scale ratio
55 *
56 * @return a float8 containing 4 2D transformed values in the input image.
57 */
58inline const float8 transform_bilinear(const float2 coord, const float2 scale)
59{
60    const float4 in_x_coords = (float4)(coord.s0, 1 + coord.s0, 2 + coord.s0, 3 + coord.s0);
61#ifdef SAMPLING_POLICY_TOP_LEFT
62    const float4 new_x = in_x_coords * (float4)(scale.s0);
63    const float4 new_y = (float4)(coord.s1 * scale.s1);
64    return (float8)(new_x.s0, new_y.s0, new_x.s1, new_y.s1, new_x.s2, new_y.s2, new_x.s3, new_y.s3);
65#elif SAMPLING_POLICY_CENTER
66    const float4 new_x = (in_x_coords + ((float4)(0.5f))) * (float4)(scale.s0) - (float4)(0.5f);
67    const float4 new_y = (float4)((coord.s1 + 0.5f) * scale.s1 - 0.5f);
68    return (float8)(new_x.s0, new_y.s0, new_x.s1, new_y.s1, new_x.s2, new_y.s2, new_x.s3, new_y.s3);
69#else /* SAMPLING_POLICY */
70#error("Unsupported sampling policy");
71#endif /* SAMPLING_POLICY */
72}
73
74/** Performs an affine transformation on an image interpolating with the NEAREAST NEIGHBOUR method. Input and output are single channel U8 or S16.
75 *
76 * @note Sampling policy to used is passed as -DSAMPLING_POLICY_(TYPE) e.g. -DSAMPLING_POLICY_TOP_LEFT
77 *
78 * @param[in]  in_ptr                            Pointer to the source image. Supported data types: U8, S16.
79 * @param[in]  in_stride_x                       Stride of the source image in X dimension (in bytes)
80 * @param[in]  in_step_x                         src_stride_x * number of elements along X processed per workitem(in bytes)
81 * @param[in]  in_stride_y                       Stride of the source image in Y dimension (in bytes)
82 * @param[in]  in_step_y                         src_stride_y * number of elements along Y processed per workitem(in bytes)
83 * @param[in]  in_offset_first_element_in_bytes  The offset of the first element in the source image
84 * @param[out] out_ptr                           Pointer to the destination image. Supported data types: U8, S16. (Must be the same as the input)
85 * @param[in]  out_stride_x                      Stride of the destination image in X dimension (in bytes)
86 * @param[in]  out_step_x                        dst_stride_x * number of elements along X processed per workitem(in bytes)
87 * @param[in]  out_stride_y                      Stride of the destination image in Y dimension (in bytes)
88 * @param[in]  out_step_y                        dst_stride_y * number of elements along Y processed per workitem(in bytes)
89 * @param[in]  out_offset_first_element_in_bytes The offset of the first element in the destination image
90 */
91__kernel void scale_nearest_neighbour_nchw(
92    IMAGE_DECLARATION(in),
93    IMAGE_DECLARATION(out))
94{
95    const int x = get_global_id(0);
96    const int y = get_global_id(1);
97
98    float8 transformed = transform_nearest((float2)(x * VEC_SIZE, y), (float2)(SCALE_X, SCALE_Y));
99#ifdef ALIGN_CORNERS
100    transformed = round(transformed);
101#endif // ALIGN_CORNERS
102
103    TILE(SELECT_DATA_TYPE(DATA_TYPE), 1, 4, cond);
104    cond[0].v = CONVERT(((transformed.even < 0) || (transformed.even >= (int)SRC_WIDTH)) || ((transformed.odd < 0) || (transformed.odd >= (int)SRC_HEIGHT)), SELECT_VEC_DATA_TYPE(DATA_TYPE, 4));
105
106    TILE(int, 1, 4, in_x);
107    TILE(int, 1, 4, in_y);
108    in_x[0].v = convert_int4(clamp(transformed.even, 0.f, SRC_WIDTH - 1.f));
109    in_y[0].v = convert_int4(clamp(transformed.odd, 0.f, SRC_HEIGHT - 1.f));
110
111    TILE(DATA_TYPE, 1, VEC_SIZE, out_vals);
112    LOOP_UNROLLING(int, i, 0, 1, VEC_SIZE,
113    {
114        out_vals[0].s[i] = select(*((__global DATA_TYPE *)(in_ptr + in_offset_first_element_in_bytes + in_x[0].s[i] * sizeof(DATA_TYPE) + in_y[0].s[i] * in_stride_y)), (DATA_TYPE)CONSTANT_VALUE, cond[0].s[i]);
115    })
116
117    __global uchar *out_addr = out_ptr + out_offset_first_element_in_bytes + x * out_step_x + y * out_stride_y;
118
119    if(x == get_global_size(0) - 1)
120    {
121#if VEC_SIZE == 1
122        VSTORE_PARTIAL(VEC_SIZE, VEC_SIZE_LEFTOVER)
123        (out_vals[0].s[0], 0, (__global DATA_TYPE *)out_addr);
124#else  // VEC_SIZE == 1
125        VSTORE_PARTIAL(VEC_SIZE, VEC_SIZE_LEFTOVER)
126        (out_vals[0].v, 0, (__global DATA_TYPE *)out_addr);
127#endif // VEC_SIZE == 1
128    }
129    else
130    {
131#if VEC_SIZE == 1
132        VSTORE(VEC_SIZE)
133        (out_vals[0].s[0], 0, (__global DATA_TYPE *)out_addr);
134#else  // VEC_SIZE == 1
135        VSTORE(VEC_SIZE)
136        (out_vals[0].v, 0, (__global DATA_TYPE *)out_addr);
137#endif // VEC_SIZE == 1
138    }
139}
140
141/** Performs an affine transformation on an image interpolating with the BILINEAR method.
142 *
143 * @note Sampling policy to used is passed as -DSAMPLING_POLICY_(TYPE) e.g. -DSAMPLING_POLICY_TOP_LEFT
144 *
145 * @param[in]  in_ptr                            Pointer to the source image. Supported data types: U8, S16.
146 * @param[in]  in_stride_x                       Stride of the source image in X dimension (in bytes)
147 * @param[in]  in_step_x                         src_stride_x * number of elements along X processed per workitem(in bytes)
148 * @param[in]  in_stride_y                       Stride of the source image in Y dimension (in bytes)
149 * @param[in]  in_step_y                         src_stride_y * number of elements along Y processed per workitem(in bytes)
150 * @param[in]  in_offset_first_element_in_bytes  The offset of the first element in the source image
151 * @param[out] out_ptr                           Pointer to the destination image. Supported data types: U8, S16. (Must be the same as the input)
152 * @param[in]  out_stride_x                      Stride of the destination image in X dimension (in bytes)
153 * @param[in]  out_step_x                        dst_stride_x * number of elements along X processed per workitem(in bytes)
154 * @param[in]  out_stride_y                      Stride of the destination image in Y dimension (in bytes)
155 * @param[in]  out_step_y                        dst_stride_y * number of elements along Y processed per workitem(in bytes)
156 * @param[in]  out_offset_first_element_in_bytes The offset of the first element in the destination image
157 */
158__kernel void scale_bilinear_nchw(
159    IMAGE_DECLARATION(in),
160    IMAGE_DECLARATION(out))
161{
162    const int x = get_global_id(0);
163    const int y = get_global_id(1);
164
165    TILE(float, 1, 8, trans_coords);
166    TILE(float, 1, 8, floor_coords);
167    TILE(int, 1, 16, in_x);
168    TILE(int, 1, 16, in_y);
169
170    trans_coords[0].v = transform_bilinear((float2)(x * VEC_SIZE, y), (float2)(SCALE_X, SCALE_Y));
171    floor_coords[0].v = floor(trans_coords[0].v);
172
173    LOOP_UNROLLING(int, i, 0, 1, 4,
174    {
175        LOOP_UNROLLING(int, j, 0, 1, 4,
176        {
177            in_x[0].s[i * 4 + j] = floor_coords[0].s[i * 2 + 0] + (j % 2);
178            in_y[0].s[i * 4 + j] = floor_coords[0].s[i * 2 + 1] + (j > 1);
179        })
180    })
181
182#if defined(BORDER_MODE_CONSTANT)
183    TILE(SELECT_DATA_TYPE(DATA_TYPE), 1, 16, cond);
184    cond[0].v = CONVERT(((in_x[0].v < 0) || (in_x[0].v >= (int)SRC_WIDTH)) || ((in_y[0].v < 0) || (in_y[0].v >= (int)SRC_HEIGHT)), SELECT_VEC_DATA_TYPE(DATA_TYPE, 16));
185#endif // defined(BORDER_MODE_CONSTANT)
186
187    in_x[0].v = clamp(in_x[0].v, 0, (int16)((int)SRC_WIDTH - 1));
188    in_y[0].v = clamp(in_y[0].v, 0, (int16)((int)SRC_HEIGHT - 1));
189
190    TILE(DATA_TYPE, 1, 16, in_vals);
191
192    // Loads the values from the input image
193#if defined(BORDER_MODE_CONSTANT)
194    LOOP_UNROLLING(int, i, 0, 1, 16,
195    {
196        in_vals[0].s[i] = select(*((__global DATA_TYPE *)(in_ptr + in_offset_first_element_in_bytes + in_x[0].s[i] * sizeof(DATA_TYPE) + in_y[0].s[i] * (int)in_stride_y)), (DATA_TYPE)CONSTANT_VALUE, cond[0].s[i]);
197    })
198#else  // defined(BORDER_MODE_CONSTANT)
199    LOOP_UNROLLING(int, i, 0, 1, 16,
200    {
201        in_vals[0].s[i] = *((__global DATA_TYPE *)(in_ptr + in_offset_first_element_in_bytes + in_x[0].s[i] * sizeof(DATA_TYPE) + in_y[0].s[i] * (int)in_stride_y));
202    })
203#endif // defined(BORDER_MODE_CONSTANT)
204
205    TILE(float, 1, 8, a);
206    TILE(float, 1, 8, b);
207
208    a[0].v = trans_coords[0].v - floor_coords[0].v;
209    b[0].v = ((float8)(1.f)) - a[0].v;
210
211#if defined(OFFSET) && defined(SCALE)
212    TILE(float, 1, 16, in_vals_f32);
213    TILE(float, 1, 4, out_vals_f32);
214
215    in_vals_f32[0].v = convert_float16(convert_int16(in_vals[0].v) - (int16)OFFSET) * (float16)SCALE;
216
217    // Bilinear interpolation: (in0  * b0 * b1) + (in1  * a0 * b1) + (in2  * b0 * a1) + (in3  * a0 * a1)
218    //                         (in4  * b2 * b3) + (in5  * a2 * b3) + (in6  * b2 * a3) + (in7  * a2 * a3)
219    //                         (in8  * b4 * b5) + (in9  * a4 * b5) + (in10 * b4 * a5) + (in11 * a4 * a5)
220    //                         (in12 * b6 * b7) + (in13 * a6 * b7) + (in14 * b6 * a7) + (in15 * a6 * a7)
221    LOOP_UNROLLING(int, i, 0, 1, 4,
222    {
223        out_vals_f32[0].s[i] = (in_vals_f32[0].s[i * 4 + 0] * b[0].s[i * 2] * b[0].s[i * 2 + 1]) + (in_vals_f32[0].s[i * 4 + 1] * a[0].s[i * 2] * b[0].s[i * 2 + 1]) + (in_vals_f32[0].s[i * 4 + 2] * b[0].s[i * 2] * a[0].s[i * 2 + 1]) + (in_vals_f32[0].s[i * 4 + 3] * a[0].s[i * 2] * a[0].s[i * 2 + 1]);
224    })
225
226    TILE(DATA_TYPE, 1, 4, out_vals_4);
227    TILE(DATA_TYPE, 1, VEC_SIZE, out_vals);
228
229    out_vals_4[0].v = CONVERT_SAT(convert_int4_sat_rtp(out_vals_f32[0].v / (float)SCALE) + OFFSET, VEC_DATA_TYPE(DATA_TYPE, 4));
230
231    LOOP_UNROLLING(int, i, 0, 1, VEC_SIZE,
232    {
233        out_vals[0].s[i] = out_vals_4[0].s[i];
234    })
235#else  // defined(OFFSET) && defined(SCALE)
236
237    TILE(DATA_TYPE, 1, VEC_SIZE, out_vals);
238
239    // Bilinear interpolation: (in0  * b0 * b1) + (in1  * a0 * b1) + (in2  * b0 * a1) + (in3  * a0 * a1)
240    //                         (in4  * b2 * b3) + (in5  * a2 * b3) + (in6  * b2 * a3) + (in7  * a2 * a3)
241    //                         (in8  * b4 * b5) + (in9  * a4 * b5) + (in10 * b4 * a5) + (in11 * a4 * a5)
242    //                         (in12 * b6 * b7) + (in13 * a6 * b7) + (in14 * b6 * a7) + (in15 * a6 * a7)
243    LOOP_UNROLLING(int, i, 0, 1, VEC_SIZE,
244    {
245        out_vals[0].s[i] = (in_vals[0].s[i * 4 + 0] * b[0].s[i * 2] * b[0].s[i * 2 + 1]) + (in_vals[0].s[i * 4 + 1] * a[0].s[i * 2] * b[0].s[i * 2 + 1]) + (in_vals[0].s[i * 4 + 2] * b[0].s[i * 2] * a[0].s[i * 2 + 1]) + (in_vals[0].s[i * 4 + 3] * a[0].s[i * 2] * a[0].s[i * 2 + 1]);
246    })
247#endif // defined(OFFSET) && defined(SCALE)
248
249    __global uchar *out_addr = out_ptr + out_offset_first_element_in_bytes + x * out_step_x + y * out_stride_y;
250
251    if(x == get_global_size(0) - 1)
252    {
253#if VEC_SIZE == 1
254        VSTORE_PARTIAL(VEC_SIZE, VEC_SIZE_LEFTOVER)
255        (out_vals[0].s[0], 0, (__global DATA_TYPE *)out_addr);
256#else  // VEC_SIZE == 1
257        VSTORE_PARTIAL(VEC_SIZE, VEC_SIZE_LEFTOVER)
258        (out_vals[0].v, 0, (__global DATA_TYPE *)out_addr);
259#endif // VEC_SIZE == 1
260    }
261    else
262    {
263#if VEC_SIZE == 1
264        VSTORE(VEC_SIZE)
265        (out_vals[0].s[0], 0, (__global DATA_TYPE *)out_addr);
266#else  // VEC_SIZE == 1
267        VSTORE(VEC_SIZE)
268        (out_vals[0].v, 0, (__global DATA_TYPE *)out_addr);
269#endif // VEC_SIZE == 1
270    }
271}