1/*M///////////////////////////////////////////////////////////////////////////////////////
2//
3//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
4//
5//  By downloading, copying, installing or using the software you agree to this license.
6//  If you do not agree to this license, do not download, install,
7//  copy or use the software.
8//
9//
10//                           License Agreement
11//                For Open Source Computer Vision Library
12//
13// Copyright (C) 2010-2012, Institute Of Software Chinese Academy Of Science, all rights reserved.
14// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
15// Third party copyrights are property of their respective owners.
16//
17// @Authors
18//    Zhang Ying, zhangying913@gmail.com
19//
20// Redistribution and use in source and binary forms, with or without modification,
21// are permitted provided that the following conditions are met:
22//
23//   * Redistribution's of source code must retain the above copyright notice,
24//     this list of conditions and the following disclaimer.
25//
26//   * Redistribution's in binary form must reproduce the above copyright notice,
27//     this list of conditions and the following disclaimer in the documentation
28//     and/or other materials provided with the distribution.
29//
30//   * The name of the copyright holders may not be used to endorse or promote products
31//     derived from this software without specific prior written permission.
32//
33// This software is provided by the copyright holders and contributors as is and
34// any express or implied warranties, including, but not limited to, the implied
35// warranties of merchantability and fitness for a particular purpose are disclaimed.
36// In no event shall the Intel Corporation or contributors be liable for any direct,
37// indirect, incidental, special, exemplary, or consequential damages
38// (including, but not limited to, procurement of substitute goods or services;
39// loss of use, data, or profits; or business interruption) however caused
40// and on any theory of liability, whether in contract, strict liability,
41// or tort (including negligence or otherwise) arising in any way out of
42// the use of this software, even if advised of the possibility of such damage.
43//
44//M*/
45
46#ifdef DOUBLE_SUPPORT
47#ifdef cl_amd_fp64
48#pragma OPENCL EXTENSION cl_amd_fp64:enable
49#elif defined (cl_khr_fp64)
50#pragma OPENCL EXTENSION cl_khr_fp64:enable
51#endif
52#define CT double
53#else
54#define CT float
55#endif
56
57#define INTER_BITS 5
58#define INTER_TAB_SIZE (1 << INTER_BITS)
59#define INTER_SCALE 1.f / INTER_TAB_SIZE
60#define AB_BITS max(10, (int)INTER_BITS)
61#define AB_SCALE (1 << AB_BITS)
62#define INTER_REMAP_COEF_BITS 15
63#define INTER_REMAP_COEF_SCALE (1 << INTER_REMAP_COEF_BITS)
64
65#define noconvert
66
67#ifndef ST
68#define ST T
69#endif
70
71#if cn != 3
72#define loadpix(addr)  *(__global const T*)(addr)
73#define storepix(val, addr)  *(__global T*)(addr) = val
74#define scalar scalar_
75#define pixsize (int)sizeof(T)
76#else
77#define loadpix(addr)  vload3(0, (__global const T1*)(addr))
78#define storepix(val, addr) vstore3(val, 0, (__global T1*)(addr))
79#ifdef INTER_NEAREST
80#define scalar (T)(scalar_.x, scalar_.y, scalar_.z)
81#else
82#define scalar (WT)(scalar_.x, scalar_.y, scalar_.z)
83#endif
84#define pixsize ((int)sizeof(T1)*3)
85#endif
86
87#ifdef INTER_NEAREST
88
89__kernel void warpPerspective(__global const uchar * srcptr, int src_step, int src_offset, int src_rows, int src_cols,
90                              __global uchar * dstptr, int dst_step, int dst_offset, int dst_rows, int dst_cols,
91                              __constant CT * M, ST scalar_)
92{
93    int dx = get_global_id(0);
94    int dy = get_global_id(1);
95
96    if (dx < dst_cols && dy < dst_rows)
97    {
98        CT X0 = M[0] * dx + M[1] * dy + M[2];
99        CT Y0 = M[3] * dx + M[4] * dy + M[5];
100        CT W = M[6] * dx + M[7] * dy + M[8];
101        W = W != 0.0f ? 1.f / W : 0.0f;
102        short sx = convert_short_sat_rte(X0*W);
103        short sy = convert_short_sat_rte(Y0*W);
104
105        int dst_index = mad24(dy, dst_step, dx * pixsize + dst_offset);
106
107        if (sx >= 0 && sx < src_cols && sy >= 0 && sy < src_rows)
108        {
109            int src_index = mad24(sy, src_step, sx * pixsize + src_offset);
110            storepix(loadpix(srcptr + src_index), dstptr + dst_index);
111        }
112        else
113            storepix(scalar, dstptr + dst_index);
114    }
115}
116
117#elif defined INTER_LINEAR
118
119__kernel void warpPerspective(__global const uchar * srcptr, int src_step, int src_offset, int src_rows, int src_cols,
120                              __global uchar * dstptr, int dst_step, int dst_offset, int dst_rows, int dst_cols,
121                              __constant CT * M, ST scalar_)
122{
123    int dx = get_global_id(0);
124    int dy = get_global_id(1);
125
126    if (dx < dst_cols && dy < dst_rows)
127    {
128        CT X0 = M[0] * dx + M[1] * dy + M[2];
129        CT Y0 = M[3] * dx + M[4] * dy + M[5];
130        CT W = M[6] * dx + M[7] * dy + M[8];
131        W = W != 0.0f ? INTER_TAB_SIZE / W : 0.0f;
132        int X = rint(X0 * W), Y = rint(Y0 * W);
133
134        short sx = convert_short_sat(X >> INTER_BITS);
135        short sy = convert_short_sat(Y >> INTER_BITS);
136        short ay = (short)(Y & (INTER_TAB_SIZE - 1));
137        short ax = (short)(X & (INTER_TAB_SIZE - 1));
138
139        WT v0 = (sx >= 0 && sx < src_cols && sy >= 0 && sy < src_rows) ?
140            convertToWT(loadpix(srcptr + mad24(sy, src_step, src_offset + sx * pixsize))) : scalar;
141        WT v1 = (sx+1 >= 0 && sx+1 < src_cols && sy >= 0 && sy < src_rows) ?
142            convertToWT(loadpix(srcptr + mad24(sy, src_step, src_offset + (sx+1) * pixsize))) : scalar;
143        WT v2 = (sx >= 0 && sx < src_cols && sy+1 >= 0 && sy+1 < src_rows) ?
144            convertToWT(loadpix(srcptr + mad24(sy+1, src_step, src_offset + sx * pixsize))) : scalar;
145        WT v3 = (sx+1 >= 0 && sx+1 < src_cols && sy+1 >= 0 && sy+1 < src_rows) ?
146            convertToWT(loadpix(srcptr + mad24(sy+1, src_step, src_offset + (sx+1) * pixsize))) : scalar;
147
148        float taby = 1.f/INTER_TAB_SIZE*ay;
149        float tabx = 1.f/INTER_TAB_SIZE*ax;
150
151        int dst_index = mad24(dy, dst_step, dst_offset + dx * pixsize);
152
153#if depth <= 4
154        int itab0 = convert_short_sat_rte( (1.0f-taby)*(1.0f-tabx) * INTER_REMAP_COEF_SCALE );
155        int itab1 = convert_short_sat_rte( (1.0f-taby)*tabx * INTER_REMAP_COEF_SCALE );
156        int itab2 = convert_short_sat_rte( taby*(1.0f-tabx) * INTER_REMAP_COEF_SCALE );
157        int itab3 = convert_short_sat_rte( taby*tabx * INTER_REMAP_COEF_SCALE );
158
159        WT val = v0 * itab0 +  v1 * itab1 + v2 * itab2 + v3 * itab3;
160        storepix(convertToT((val + (1 << (INTER_REMAP_COEF_BITS-1))) >> INTER_REMAP_COEF_BITS), dstptr + dst_index);
161#else
162        float tabx2 = 1.0f - tabx, taby2 = 1.0f - taby;
163        WT val = v0 * tabx2 * taby2 +  v1 * tabx * taby2 + v2 * tabx2 * taby + v3 * tabx * taby;
164        storepix(convertToT(val), dstptr + dst_index);
165#endif
166    }
167}
168
169#elif defined INTER_CUBIC
170
171inline void interpolateCubic( float x, float* coeffs )
172{
173    const float A = -0.75f;
174
175    coeffs[0] = ((A*(x + 1.f) - 5.0f*A)*(x + 1.f) + 8.0f*A)*(x + 1.f) - 4.0f*A;
176    coeffs[1] = ((A + 2.f)*x - (A + 3.f))*x*x + 1.f;
177    coeffs[2] = ((A + 2.f)*(1.f - x) - (A + 3.f))*(1.f - x)*(1.f - x) + 1.f;
178    coeffs[3] = 1.f - coeffs[0] - coeffs[1] - coeffs[2];
179}
180
181__kernel void warpPerspective(__global const uchar * srcptr, int src_step, int src_offset, int src_rows, int src_cols,
182                              __global uchar * dstptr, int dst_step, int dst_offset, int dst_rows, int dst_cols,
183                              __constant CT * M, ST scalar_)
184{
185    int dx = get_global_id(0);
186    int dy = get_global_id(1);
187
188    if (dx < dst_cols && dy < dst_rows)
189    {
190        CT X0 = M[0] * dx + M[1] * dy + M[2];
191        CT Y0 = M[3] * dx + M[4] * dy + M[5];
192        CT W = M[6] * dx + M[7] * dy + M[8];
193        W = W != 0.0f ? INTER_TAB_SIZE / W : 0.0f;
194        int X = rint(X0 * W), Y = rint(Y0 * W);
195
196        short sx = convert_short_sat(X >> INTER_BITS) - 1;
197        short sy = convert_short_sat(Y >> INTER_BITS) - 1;
198        short ay = (short)(Y & (INTER_TAB_SIZE-1));
199        short ax = (short)(X & (INTER_TAB_SIZE-1));
200
201        WT v[16];
202        #pragma unroll
203        for (int y = 0; y < 4; y++)
204            #pragma unroll
205            for (int x = 0; x < 4; x++)
206                v[mad24(y, 4, x)] = (sx+x >= 0 && sx+x < src_cols && sy+y >= 0 && sy+y < src_rows) ?
207                    convertToWT(loadpix(srcptr + mad24(sy+y, src_step, src_offset + (sx+x) * pixsize))) : scalar;
208
209        float tab1y[4], tab1x[4];
210
211        float ayy = INTER_SCALE * ay;
212        float axx = INTER_SCALE * ax;
213        interpolateCubic(ayy, tab1y);
214        interpolateCubic(axx, tab1x);
215
216        int dst_index = mad24(dy, dst_step, dst_offset + dx * pixsize);
217
218        WT sum = (WT)(0);
219#if depth <= 4
220        int itab[16];
221
222        #pragma unroll
223        for (int i = 0; i < 16; i++)
224            itab[i] = rint(tab1y[(i>>2)] * tab1x[(i&3)] * INTER_REMAP_COEF_SCALE);
225
226        #pragma unroll
227        for (int i = 0; i < 16; i++)
228            sum += v[i] * itab[i];
229        storepix(convertToT( (sum + (1 << (INTER_REMAP_COEF_BITS-1))) >> INTER_REMAP_COEF_BITS ), dstptr + dst_index);
230#else
231        #pragma unroll
232        for (int i = 0; i < 16; i++)
233            sum += v[i] * tab1y[(i>>2)] * tab1x[(i&3)];
234        storepix(convertToT( sum ), dstptr + dst_index);
235#endif
236    }
237}
238
239#endif
240