2 * Copyright (c) 2016, 2017 ARM Limited.
4 * SPDX-License-Identifier: MIT
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:
13 * The above copyright notice and this permission notice shall be included in all
14 * copies or substantial portions of the Software.
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
25 #include "warp_helpers.h"
27 /** Transforms four 2D coordinates. This is used to map the output coordinates to the input coordinates.
29 * @param[in] coord 2D coordinates to transform.
30 * @param[in] scale input/output scale ratio
32 * @return a float8 containing 4 2D transformed values in the input image.
34 inline const float8 transform_nearest(const float2 coord, const float2 scale)
36 const float4 in_x_coords = (float4)(coord.s0, 1 + coord.s0, 2 + coord.s0, 3 + coord.s0);
37 const float4 new_x = (in_x_coords + ((float4)(0.5f))) * (float4)(scale.s0);
38 const float4 new_y = (float4)((coord.s1 + 0.5f) * scale.s1);
39 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);
42 /** Transforms four 2D coordinates. This is used to map the output coordinates to the input coordinates.
44 * @param[in] coord 2D coordinates to transform.
45 * @param[in] scale input/output scale ratio
47 * @return a float8 containing 4 2D transformed values in the input image.
49 inline const float8 transform_bilinear(const float2 coord, const float2 scale)
51 const float4 in_x_coords = (float4)(coord.s0, 1 + coord.s0, 2 + coord.s0, 3 + coord.s0);
52 #ifdef SAMPLING_POLICY_TOP_LEFT
53 const float4 new_x = in_x_coords * (float4)(scale.s0);
54 const float4 new_y = (float4)(coord.s1 * scale.s1);
55 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);
56 #elif SAMPLING_POLICY_CENTER
57 const float4 new_x = (in_x_coords + ((float4)(0.5f))) * (float4)(scale.s0) - (float4)(0.5f);
58 const float4 new_y = (float4)((coord.s1 + 0.5f) * scale.s1 - 0.5f);
59 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);
60 #else /* SAMPLING_POLICY */
61 #error("Unsupported sampling policy");
62 #endif /* SAMPLING_POLICY */
65 /** Performs an affine transformation on an image interpolating with the NEAREAST NEIGHBOUR method. Input and output are single channel U8 or S16.
67 * @note Sampling policy to used is passed as -DSAMPLING_POLICY_(TYPE) e.g. -DSAMPLING_POLICY_TOP_LEFT
69 * @param[in] in_ptr Pointer to the source image. Supported data types: U8, S16.
70 * @param[in] in_stride_x Stride of the source image in X dimension (in bytes)
71 * @param[in] in_step_x src_stride_x * number of elements along X processed per workitem(in bytes)
72 * @param[in] in_stride_y Stride of the source image in Y dimension (in bytes)
73 * @param[in] in_step_y src_stride_y * number of elements along Y processed per workitem(in bytes)
74 * @param[in] in_offset_first_element_in_bytes The offset of the first element in the source image
75 * @param[out] out_ptr Pointer to the destination image. Supported data types: U8, S16. (Must be the same as the input)
76 * @param[in] out_stride_x Stride of the destination image in X dimension (in bytes)
77 * @param[in] out_step_x dst_stride_x * number of elements along X processed per workitem(in bytes)
78 * @param[in] out_stride_y Stride of the destination image in Y dimension (in bytes)
79 * @param[in] out_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes)
80 * @param[in] out_offset_first_element_in_bytes The offset of the first element in the destination image
81 * @param[in] input_width Input image width
82 * @param[in] input_height Input image height
83 * @param[in] scale_x The scale factor along x dimension
84 * @param[in] scale_y The scale factor along y dimension
86 __kernel void scale_nearest_neighbour(
87 IMAGE_DECLARATION(in),
88 IMAGE_DECLARATION(out),
89 const float input_width,
90 const float input_height,
94 Image in = CONVERT_TO_IMAGE_STRUCT_NO_STEP(in);
95 Image out = CONVERT_TO_IMAGE_STRUCT(out);
96 const float2 r = (float2)(scale_x, scale_y);
97 const float8 tc = clamp_to_border_with_size(transform_nearest(get_current_coords(), r), input_width, input_height, BORDER_SIZE);
98 vstore4(read_texels4(&in, convert_int8(tc)), 0, (__global DATA_TYPE *)out.ptr);
101 /** Performs an affine transformation on an image interpolating with the BILINEAR method.
103 * @note Sampling policy to used is passed as -DSAMPLING_POLICY_(TYPE) e.g. -DSAMPLING_POLICY_TOP_LEFT
105 * @param[in] in_ptr Pointer to the source image. Supported data types: U8, S16.
106 * @param[in] in_stride_x Stride of the source image in X dimension (in bytes)
107 * @param[in] in_step_x src_stride_x * number of elements along X processed per workitem(in bytes)
108 * @param[in] in_stride_y Stride of the source image in Y dimension (in bytes)
109 * @param[in] in_step_y src_stride_y * number of elements along Y processed per workitem(in bytes)
110 * @param[in] in_offset_first_element_in_bytes The offset of the first element in the source image
111 * @param[out] out_ptr Pointer to the destination image. Supported data types: U8, S16. (Must be the same as the input)
112 * @param[in] out_stride_x Stride of the destination image in X dimension (in bytes)
113 * @param[in] out_step_x dst_stride_x * number of elements along X processed per workitem(in bytes)
114 * @param[in] out_stride_y Stride of the destination image in Y dimension (in bytes)
115 * @param[in] out_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes)
116 * @param[in] out_offset_first_element_in_bytes The offset of the first element in the destination image
117 * @param[in] input_width Input image width
118 * @param[in] input_height Input image height
119 * @param[in] scale_x The scale factor along x dimension
120 * @param[in] scale_y The scale factor along y dimension
122 __kernel void scale_bilinear(
123 IMAGE_DECLARATION(in),
124 IMAGE_DECLARATION(out),
125 const float input_width,
126 const float input_height,
130 Image in = CONVERT_TO_IMAGE_STRUCT_NO_STEP(in);
131 Image out = CONVERT_TO_IMAGE_STRUCT(out);
132 const float2 r = (float2)(scale_x, scale_y);
133 const float8 tc = transform_bilinear(get_current_coords(), r);
134 vstore4(bilinear_interpolate_with_border(&in, tc, input_width, input_height, BORDER_SIZE), 0, (__global DATA_TYPE *)out.ptr);