/src/Simd/src/Simd/SimdBaseShiftBilinear.cpp
Line | Count | Source |
1 | | /* |
2 | | * Simd Library (http://ermig1979.github.io/Simd). |
3 | | * |
4 | | * Copyright (c) 2011-2017 Yermalayeu Ihar. |
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 deal |
8 | | * in the Software without restriction, including without limitation the rights |
9 | | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell |
10 | | * 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 |
14 | | * all 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 "Simd/SimdMath.h" |
25 | | #include "Simd/SimdBase.h" |
26 | | |
27 | | namespace Simd |
28 | | { |
29 | | namespace Base |
30 | | { |
31 | | SIMD_INLINE int Interpolate(int s[2][2], int k[2][2]) |
32 | 0 | { |
33 | 0 | return (s[0][0] * k[0][0] + s[0][1] * k[0][1] + |
34 | 0 | s[1][0] * k[1][0] + s[1][1] * k[1][1] + BILINEAR_ROUND_TERM) >> BILINEAR_SHIFT; |
35 | 0 | } |
36 | | |
37 | | SIMD_INLINE int Interpolate(const unsigned char *src, size_t dx, size_t dy, int k[2][2]) |
38 | 0 | { |
39 | 0 | return (src[0] * k[0][0] + src[dx] * k[0][1] + |
40 | 0 | src[dy] * k[1][0] + src[dx + dy] * k[1][1] + BILINEAR_ROUND_TERM) >> BILINEAR_SHIFT; |
41 | 0 | } |
42 | | |
43 | | SIMD_INLINE int Interpolate(const unsigned char *src, size_t dr, int k[2]) |
44 | 0 | { |
45 | 0 | return (src[0] * k[0] + src[dr] * k[1] + LINEAR_ROUND_TERM) >> LINEAR_SHIFT; |
46 | 0 | } |
47 | | |
48 | | void MixBorder(const uint8_t * src, size_t srcStride, size_t width, size_t height, size_t channelCount, |
49 | | const uint8_t * bkg, size_t bkgStride, ptrdiff_t iDx, ptrdiff_t iDy, int fDx, int fDy, uint8_t * dst, size_t dstStride) |
50 | 0 | { |
51 | 0 | size_t bkgWidth = Abs(iDx) - (iDx < 0 && fDx ? 1 : 0); |
52 | 0 | size_t bkgHeight = Abs(iDy) - (iDy < 0 && fDy ? 1 : 0); |
53 | |
|
54 | 0 | size_t mainWidth = width - bkgWidth - (fDx ? 1 : 0); |
55 | 0 | size_t mainHeight = height - bkgHeight - (fDy ? 1 : 0); |
56 | |
|
57 | 0 | int k[2][2]; |
58 | 0 | k[0][0] = (FRACTION_RANGE - fDx)*(FRACTION_RANGE - fDy); |
59 | 0 | k[0][1] = fDx*(FRACTION_RANGE - fDy); |
60 | 0 | k[1][0] = (FRACTION_RANGE - fDx)*fDy; |
61 | 0 | k[1][1] = fDx*fDy; |
62 | |
|
63 | 0 | if (fDx) |
64 | 0 | { |
65 | 0 | const uint8_t * ps[2][2]; |
66 | 0 | size_t xOffset = (iDx >= 0 ? width - 1 - iDx : -iDx - 1)*channelCount; |
67 | 0 | size_t bkgOffset = (iDy > 0 ? 0 : -iDy)*bkgStride + xOffset; |
68 | 0 | size_t dstOffset = (iDy > 0 ? 0 : -iDy)*dstStride + xOffset; |
69 | |
|
70 | 0 | if (iDx < 0) |
71 | 0 | { |
72 | 0 | ps[0][0] = bkg + bkgOffset; |
73 | 0 | ps[0][1] = src + (iDy < 0 ? 0 : iDy)*srcStride; |
74 | 0 | ps[1][0] = bkg + bkgOffset; |
75 | 0 | ps[1][1] = src + ((iDy < 0 ? 0 : iDy) + (fDy ? 1 : 0))*srcStride; |
76 | 0 | } |
77 | 0 | else |
78 | 0 | { |
79 | 0 | ps[0][0] = src + (iDy < 0 ? 0 : iDy)*srcStride + (width - 1)*channelCount; |
80 | 0 | ps[0][1] = bkg + bkgOffset; |
81 | 0 | ps[1][0] = src + ((iDy < 0 ? 0 : iDy) + (fDy ? 1 : 0))*srcStride + (width - 1)*channelCount; |
82 | 0 | ps[1][1] = bkg + bkgOffset; |
83 | 0 | } |
84 | |
|
85 | 0 | for (size_t row = 0; row < mainHeight; ++row) |
86 | 0 | { |
87 | 0 | for (size_t channel = 0; channel < channelCount; channel++) |
88 | 0 | { |
89 | 0 | int s[2][2]; |
90 | 0 | s[0][0] = ps[0][0][channel]; |
91 | 0 | s[0][1] = ps[0][1][channel]; |
92 | 0 | s[1][0] = ps[1][0][channel]; |
93 | 0 | s[1][1] = ps[1][1][channel]; |
94 | 0 | dst[dstOffset + channel] = Interpolate(s, k); |
95 | 0 | } |
96 | 0 | ps[0][0] += srcStride; |
97 | 0 | ps[0][1] += bkgStride; |
98 | 0 | ps[1][0] += srcStride; |
99 | 0 | ps[1][1] += bkgStride; |
100 | 0 | dstOffset += dstStride; |
101 | 0 | } |
102 | 0 | } |
103 | |
|
104 | 0 | if (fDy) |
105 | 0 | { |
106 | 0 | const uint8_t * ps[2][2]; |
107 | 0 | size_t bkgOffset = (iDy >= 0 ? height - 1 - iDy : -iDy - 1)*bkgStride + (iDx > 0 ? 0 : -iDx)*channelCount; |
108 | 0 | size_t dstOffset = (iDy >= 0 ? height - 1 - iDy : -iDy - 1)*dstStride + (iDx > 0 ? 0 : -iDx)*channelCount; |
109 | |
|
110 | 0 | if (iDy < 0) |
111 | 0 | { |
112 | 0 | ps[0][0] = bkg + bkgOffset; |
113 | 0 | ps[0][1] = bkg + bkgOffset; |
114 | 0 | ps[1][0] = src + (iDx < 0 ? 0 : iDx)*channelCount; |
115 | 0 | ps[1][1] = src + ((iDx < 0 ? 0 : iDx) + (fDx ? 1 : 0))*channelCount; |
116 | 0 | } |
117 | 0 | else |
118 | 0 | { |
119 | 0 | ps[0][0] = src + (height - 1)*srcStride + (iDx < 0 ? 0 : iDx)*channelCount; |
120 | 0 | ps[0][1] = src + (height - 1)*srcStride + ((iDx < 0 ? 0 : iDx) + (fDx ? 1 : 0))*channelCount; |
121 | 0 | ps[1][0] = bkg + bkgOffset; |
122 | 0 | ps[1][1] = bkg + bkgOffset; |
123 | 0 | } |
124 | |
|
125 | 0 | for (size_t col = 0; col < mainWidth; ++col) |
126 | 0 | { |
127 | 0 | for (size_t channel = 0; channel < channelCount; channel++) |
128 | 0 | { |
129 | 0 | int s[2][2]; |
130 | 0 | s[0][0] = ps[0][0][channel]; |
131 | 0 | s[0][1] = ps[0][1][channel]; |
132 | 0 | s[1][0] = ps[1][0][channel]; |
133 | 0 | s[1][1] = ps[1][1][channel]; |
134 | 0 | dst[dstOffset + channel] = Interpolate(s, k); |
135 | 0 | } |
136 | 0 | ps[0][0] += channelCount; |
137 | 0 | ps[0][1] += channelCount; |
138 | 0 | ps[1][0] += channelCount; |
139 | 0 | ps[1][1] += channelCount; |
140 | 0 | dstOffset += channelCount; |
141 | 0 | } |
142 | 0 | } |
143 | |
|
144 | 0 | if (fDx && fDy) |
145 | 0 | { |
146 | 0 | const uint8_t * ps[2][2]; |
147 | 0 | size_t xOffset = (iDx >= 0 ? width - 1 - iDx : -iDx - 1)*channelCount; |
148 | 0 | size_t bkgOffset = (iDy >= 0 ? height - 1 - iDy : -iDy - 1)*bkgStride + xOffset; |
149 | 0 | size_t dstOffset = (iDy >= 0 ? height - 1 - iDy : -iDy - 1)*dstStride + xOffset; |
150 | |
|
151 | 0 | ps[0][0] = (iDx >= 0 && iDy >= 0) ? (src + (height - 1)*srcStride + (width - 1)*channelCount) : bkg + bkgOffset; |
152 | 0 | ps[0][1] = (iDx < 0 && iDy >= 0) ? (src + (height - 1)*srcStride) : bkg + bkgOffset; |
153 | 0 | ps[1][0] = (iDx >= 0 && iDy < 0) ? (src + (width - 1)*channelCount) : bkg + bkgOffset; |
154 | 0 | ps[1][1] = (iDx < 0 && iDy < 0) ? (src) : bkg + bkgOffset; |
155 | |
|
156 | 0 | for (size_t channel = 0; channel < channelCount; channel++) |
157 | 0 | { |
158 | 0 | int s[2][2]; |
159 | 0 | s[0][0] = ps[0][0][channel]; |
160 | 0 | s[0][1] = ps[0][1][channel]; |
161 | 0 | s[1][0] = ps[1][0][channel]; |
162 | 0 | s[1][1] = ps[1][1][channel]; |
163 | 0 | dst[dstOffset + channel] = Interpolate(s, k); |
164 | 0 | } |
165 | 0 | } |
166 | 0 | } |
167 | | |
168 | | void CommonShiftAction( |
169 | | const uint8_t * & src, size_t srcStride, size_t & width, size_t & height, size_t channelCount, |
170 | | const uint8_t * bkg, size_t bkgStride, const double * shiftX, const double * shiftY, |
171 | | size_t cropLeft, size_t cropTop, size_t cropRight, size_t cropBottom, uint8_t * & dst, size_t dstStride, |
172 | | int & fDx, int & fDy) |
173 | 0 | { |
174 | 0 | assert(cropLeft <= cropRight && cropTop <= cropBottom && cropRight <= width && cropBottom <= height); |
175 | 0 | assert(*shiftX < cropRight - cropLeft && *shiftY < cropBottom - cropTop); |
176 | |
|
177 | 0 | Base::CopyFrame(src, srcStride, width, height, channelCount, cropLeft, cropTop, cropRight, cropBottom, dst, dstStride); |
178 | |
|
179 | 0 | dst += dstStride*cropTop + cropLeft*channelCount; |
180 | 0 | src += srcStride*cropTop + cropLeft*channelCount; |
181 | 0 | bkg += bkgStride*cropTop + cropLeft*channelCount; |
182 | 0 | width = cropRight - cropLeft; |
183 | 0 | height = cropBottom - cropTop; |
184 | |
|
185 | 0 | ptrdiff_t iDx = (ptrdiff_t)floor(*shiftX + FRACTION_ROUND_TERM); |
186 | 0 | ptrdiff_t iDy = (ptrdiff_t)floor(*shiftY + FRACTION_ROUND_TERM); |
187 | 0 | fDx = (int)floor((*shiftX + FRACTION_ROUND_TERM - iDx)*FRACTION_RANGE); |
188 | 0 | fDy = (int)floor((*shiftY + FRACTION_ROUND_TERM - iDy)*FRACTION_RANGE); |
189 | |
|
190 | 0 | ptrdiff_t left = (iDx < 0 ? (-iDx - (fDx ? 1 : 0)) : 0); |
191 | 0 | ptrdiff_t top = (iDy < 0 ? (-iDy - (fDy ? 1 : 0)) : 0); |
192 | 0 | ptrdiff_t right = (iDx < 0 ? width : width - iDx); |
193 | 0 | ptrdiff_t bottom = (iDy < 0 ? height : height - iDy); |
194 | |
|
195 | 0 | Base::CopyFrame(bkg, bkgStride, width, height, channelCount, left, top, right, bottom, dst, dstStride); |
196 | |
|
197 | 0 | MixBorder(src, srcStride, width, height, channelCount, bkg, bkgStride, iDx, iDy, fDx, fDy, dst, dstStride); |
198 | |
|
199 | 0 | src += Simd::Max((ptrdiff_t)0, iDy)*srcStride + Simd::Max((ptrdiff_t)0, iDx)*channelCount; |
200 | 0 | dst += Simd::Max((ptrdiff_t)0, -iDy)*dstStride + Simd::Max((ptrdiff_t)0, -iDx)*channelCount; |
201 | |
|
202 | 0 | width = width - Abs(iDx) + (iDx < 0 && fDx ? 1 : 0) - (fDx ? 1 : 0); |
203 | 0 | height = height - Abs(iDy) + (iDy < 0 && fDy ? 1 : 0) - (fDy ? 1 : 0); |
204 | 0 | } |
205 | | |
206 | | void ShiftBilinear(const uint8_t * src, size_t srcStride, size_t width, size_t height, size_t channelCount, |
207 | | int fDx, int fDy, uint8_t * dst, size_t dstStride) |
208 | 0 | { |
209 | 0 | size_t size = width*channelCount; |
210 | 0 | if (fDy) |
211 | 0 | { |
212 | 0 | if (fDx) |
213 | 0 | { |
214 | 0 | int k[2][2]; |
215 | 0 | k[0][0] = (FRACTION_RANGE - fDx)*(FRACTION_RANGE - fDy); |
216 | 0 | k[0][1] = fDx*(FRACTION_RANGE - fDy); |
217 | 0 | k[1][0] = (FRACTION_RANGE - fDx)*fDy; |
218 | 0 | k[1][1] = fDx*fDy; |
219 | 0 | for (size_t row = 0; row < height; ++row) |
220 | 0 | { |
221 | 0 | for (size_t col = 0; col < size; col++) |
222 | 0 | { |
223 | 0 | dst[col] = Interpolate(src + col, channelCount, srcStride, k); |
224 | 0 | } |
225 | 0 | src += srcStride; |
226 | 0 | dst += dstStride; |
227 | 0 | } |
228 | 0 | } |
229 | 0 | else |
230 | 0 | { |
231 | 0 | int k[2]; |
232 | 0 | k[0] = FRACTION_RANGE - fDy; |
233 | 0 | k[1] = fDy; |
234 | 0 | for (size_t row = 0; row < height; ++row) |
235 | 0 | { |
236 | 0 | for (size_t col = 0; col < size; col++) |
237 | 0 | { |
238 | 0 | dst[col] = Interpolate(src + col, srcStride, k); |
239 | 0 | } |
240 | 0 | src += srcStride; |
241 | 0 | dst += dstStride; |
242 | 0 | } |
243 | 0 | } |
244 | 0 | } |
245 | 0 | else |
246 | 0 | { |
247 | 0 | if (fDx) |
248 | 0 | { |
249 | 0 | int k[2]; |
250 | 0 | k[0] = FRACTION_RANGE - fDx; |
251 | 0 | k[1] = fDx; |
252 | 0 | for (size_t row = 0; row < height; ++row) |
253 | 0 | { |
254 | 0 | for (size_t col = 0; col < size; col++) |
255 | 0 | { |
256 | 0 | dst[col] = Interpolate(src + col, channelCount, k); |
257 | 0 | } |
258 | 0 | src += srcStride; |
259 | 0 | dst += dstStride; |
260 | 0 | } |
261 | 0 | } |
262 | 0 | else |
263 | 0 | { |
264 | 0 | for (size_t row = 0; row < height; ++row) |
265 | 0 | { |
266 | 0 | memcpy(dst, src, size); |
267 | 0 | src += srcStride; |
268 | 0 | dst += dstStride; |
269 | 0 | } |
270 | 0 | } |
271 | 0 | } |
272 | 0 | } |
273 | | |
274 | | void ShiftBilinear( |
275 | | const uint8_t * src, size_t srcStride, size_t width, size_t height, size_t channelCount, |
276 | | const uint8_t * bkg, size_t bkgStride, const double * shiftX, const double * shiftY, |
277 | | size_t cropLeft, size_t cropTop, size_t cropRight, size_t cropBottom, uint8_t * dst, size_t dstStride) |
278 | 0 | { |
279 | 0 | int fDx, fDy; |
280 | 0 | CommonShiftAction(src, srcStride, width, height, channelCount, bkg, bkgStride, shiftX, shiftY, |
281 | 0 | cropLeft, cropTop, cropRight, cropBottom, dst, dstStride, fDx, fDy); |
282 | |
|
283 | 0 | ShiftBilinear(src, srcStride, width, height, channelCount, fDx, fDy, dst, dstStride); |
284 | 0 | } |
285 | | } |
286 | | } |
287 | | |