2 * Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
4 * This file is part of FFmpeg.
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 #define _SVID_SOURCE // needed for MAP_ANONYMOUS
24 #define _DARWIN_C_SOURCE // needed for MAP_ANON
31 #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
32 #define MAP_ANONYMOUS MAP_ANON
36 #define WIN32_LEAN_AND_MEAN
40 #include "libavutil/attributes.h"
41 #include "libavutil/avassert.h"
42 #include "libavutil/avutil.h"
43 #include "libavutil/bswap.h"
44 #include "libavutil/cpu.h"
45 #include "libavutil/intreadwrite.h"
46 #include "libavutil/mathematics.h"
47 #include "libavutil/opt.h"
48 #include "libavutil/pixdesc.h"
49 #include "libavutil/x86/asm.h"
50 #include "libavutil/x86/cpu.h"
53 #include "swscale_internal.h"
55 unsigned swscale_version(void)
57 av_assert0(LIBSWSCALE_VERSION_MICRO >= 100);
58 return LIBSWSCALE_VERSION_INT;
61 const char *swscale_configuration(void)
63 return FFMPEG_CONFIGURATION;
66 const char *swscale_license(void)
68 #define LICENSE_PREFIX "libswscale license: "
69 return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
72 #define RET 0xC3 // near return opcode for x86
74 typedef struct FormatEntry {
75 uint8_t is_supported_in :1;
76 uint8_t is_supported_out :1;
77 uint8_t is_supported_endianness :1;
80 static const FormatEntry format_entries[AV_PIX_FMT_NB] = {
81 [AV_PIX_FMT_YUV420P] = { 1, 1 },
82 [AV_PIX_FMT_YUYV422] = { 1, 1 },
83 [AV_PIX_FMT_RGB24] = { 1, 1 },
84 [AV_PIX_FMT_BGR24] = { 1, 1 },
85 [AV_PIX_FMT_YUV422P] = { 1, 1 },
86 [AV_PIX_FMT_YUV444P] = { 1, 1 },
87 [AV_PIX_FMT_YUV410P] = { 1, 1 },
88 [AV_PIX_FMT_YUV411P] = { 1, 1 },
89 [AV_PIX_FMT_GRAY8] = { 1, 1 },
90 [AV_PIX_FMT_MONOWHITE] = { 1, 1 },
91 [AV_PIX_FMT_MONOBLACK] = { 1, 1 },
92 [AV_PIX_FMT_PAL8] = { 1, 0 },
93 [AV_PIX_FMT_YUVJ420P] = { 1, 1 },
94 [AV_PIX_FMT_YUVJ422P] = { 1, 1 },
95 [AV_PIX_FMT_YUVJ444P] = { 1, 1 },
96 [AV_PIX_FMT_UYVY422] = { 1, 1 },
97 [AV_PIX_FMT_UYYVYY411] = { 0, 0 },
98 [AV_PIX_FMT_BGR8] = { 1, 1 },
99 [AV_PIX_FMT_BGR4] = { 0, 1 },
100 [AV_PIX_FMT_BGR4_BYTE] = { 1, 1 },
101 [AV_PIX_FMT_RGB8] = { 1, 1 },
102 [AV_PIX_FMT_RGB4] = { 0, 1 },
103 [AV_PIX_FMT_RGB4_BYTE] = { 1, 1 },
104 [AV_PIX_FMT_NV12] = { 1, 1 },
105 [AV_PIX_FMT_NV21] = { 1, 1 },
106 [AV_PIX_FMT_ARGB] = { 1, 1 },
107 [AV_PIX_FMT_RGBA] = { 1, 1 },
108 [AV_PIX_FMT_ABGR] = { 1, 1 },
109 [AV_PIX_FMT_BGRA] = { 1, 1 },
110 [AV_PIX_FMT_0RGB] = { 1, 1 },
111 [AV_PIX_FMT_RGB0] = { 1, 1 },
112 [AV_PIX_FMT_0BGR] = { 1, 1 },
113 [AV_PIX_FMT_BGR0] = { 1, 1 },
114 [AV_PIX_FMT_GRAY16BE] = { 1, 1 },
115 [AV_PIX_FMT_GRAY16LE] = { 1, 1 },
116 [AV_PIX_FMT_YUV440P] = { 1, 1 },
117 [AV_PIX_FMT_YUVJ440P] = { 1, 1 },
118 [AV_PIX_FMT_YUVA420P] = { 1, 1 },
119 [AV_PIX_FMT_YUVA422P] = { 1, 1 },
120 [AV_PIX_FMT_YUVA444P] = { 1, 1 },
121 [AV_PIX_FMT_YUVA420P9BE] = { 1, 1 },
122 [AV_PIX_FMT_YUVA420P9LE] = { 1, 1 },
123 [AV_PIX_FMT_YUVA422P9BE] = { 1, 1 },
124 [AV_PIX_FMT_YUVA422P9LE] = { 1, 1 },
125 [AV_PIX_FMT_YUVA444P9BE] = { 1, 1 },
126 [AV_PIX_FMT_YUVA444P9LE] = { 1, 1 },
127 [AV_PIX_FMT_YUVA420P10BE]= { 1, 1 },
128 [AV_PIX_FMT_YUVA420P10LE]= { 1, 1 },
129 [AV_PIX_FMT_YUVA422P10BE]= { 1, 1 },
130 [AV_PIX_FMT_YUVA422P10LE]= { 1, 1 },
131 [AV_PIX_FMT_YUVA444P10BE]= { 1, 1 },
132 [AV_PIX_FMT_YUVA444P10LE]= { 1, 1 },
133 [AV_PIX_FMT_YUVA420P16BE]= { 1, 1 },
134 [AV_PIX_FMT_YUVA420P16LE]= { 1, 1 },
135 [AV_PIX_FMT_YUVA422P16BE]= { 1, 1 },
136 [AV_PIX_FMT_YUVA422P16LE]= { 1, 1 },
137 [AV_PIX_FMT_YUVA444P16BE]= { 1, 1 },
138 [AV_PIX_FMT_YUVA444P16LE]= { 1, 1 },
139 [AV_PIX_FMT_RGB48BE] = { 1, 1 },
140 [AV_PIX_FMT_RGB48LE] = { 1, 1 },
141 [AV_PIX_FMT_RGBA64BE] = { 1, 1 },
142 [AV_PIX_FMT_RGBA64LE] = { 1, 1 },
143 [AV_PIX_FMT_RGB565BE] = { 1, 1 },
144 [AV_PIX_FMT_RGB565LE] = { 1, 1 },
145 [AV_PIX_FMT_RGB555BE] = { 1, 1 },
146 [AV_PIX_FMT_RGB555LE] = { 1, 1 },
147 [AV_PIX_FMT_BGR565BE] = { 1, 1 },
148 [AV_PIX_FMT_BGR565LE] = { 1, 1 },
149 [AV_PIX_FMT_BGR555BE] = { 1, 1 },
150 [AV_PIX_FMT_BGR555LE] = { 1, 1 },
151 [AV_PIX_FMT_YUV420P16LE] = { 1, 1 },
152 [AV_PIX_FMT_YUV420P16BE] = { 1, 1 },
153 [AV_PIX_FMT_YUV422P16LE] = { 1, 1 },
154 [AV_PIX_FMT_YUV422P16BE] = { 1, 1 },
155 [AV_PIX_FMT_YUV444P16LE] = { 1, 1 },
156 [AV_PIX_FMT_YUV444P16BE] = { 1, 1 },
157 [AV_PIX_FMT_RGB444LE] = { 1, 1 },
158 [AV_PIX_FMT_RGB444BE] = { 1, 1 },
159 [AV_PIX_FMT_BGR444LE] = { 1, 1 },
160 [AV_PIX_FMT_BGR444BE] = { 1, 1 },
161 [AV_PIX_FMT_Y400A] = { 1, 0 },
162 [AV_PIX_FMT_BGR48BE] = { 1, 1 },
163 [AV_PIX_FMT_BGR48LE] = { 1, 1 },
164 [AV_PIX_FMT_BGRA64BE] = { 0, 0 },
165 [AV_PIX_FMT_BGRA64LE] = { 0, 0 },
166 [AV_PIX_FMT_YUV420P9BE] = { 1, 1 },
167 [AV_PIX_FMT_YUV420P9LE] = { 1, 1 },
168 [AV_PIX_FMT_YUV420P10BE] = { 1, 1 },
169 [AV_PIX_FMT_YUV420P10LE] = { 1, 1 },
170 [AV_PIX_FMT_YUV420P12BE] = { 1, 1 },
171 [AV_PIX_FMT_YUV420P12LE] = { 1, 1 },
172 [AV_PIX_FMT_YUV420P14BE] = { 1, 1 },
173 [AV_PIX_FMT_YUV420P14LE] = { 1, 1 },
174 [AV_PIX_FMT_YUV422P9BE] = { 1, 1 },
175 [AV_PIX_FMT_YUV422P9LE] = { 1, 1 },
176 [AV_PIX_FMT_YUV422P10BE] = { 1, 1 },
177 [AV_PIX_FMT_YUV422P10LE] = { 1, 1 },
178 [AV_PIX_FMT_YUV422P12BE] = { 1, 1 },
179 [AV_PIX_FMT_YUV422P12LE] = { 1, 1 },
180 [AV_PIX_FMT_YUV422P14BE] = { 1, 1 },
181 [AV_PIX_FMT_YUV422P14LE] = { 1, 1 },
182 [AV_PIX_FMT_YUV444P9BE] = { 1, 1 },
183 [AV_PIX_FMT_YUV444P9LE] = { 1, 1 },
184 [AV_PIX_FMT_YUV444P10BE] = { 1, 1 },
185 [AV_PIX_FMT_YUV444P10LE] = { 1, 1 },
186 [AV_PIX_FMT_YUV444P12BE] = { 1, 1 },
187 [AV_PIX_FMT_YUV444P12LE] = { 1, 1 },
188 [AV_PIX_FMT_YUV444P14BE] = { 1, 1 },
189 [AV_PIX_FMT_YUV444P14LE] = { 1, 1 },
190 [AV_PIX_FMT_GBRP] = { 1, 1 },
191 [AV_PIX_FMT_GBRP9LE] = { 1, 1 },
192 [AV_PIX_FMT_GBRP9BE] = { 1, 1 },
193 [AV_PIX_FMT_GBRP10LE] = { 1, 1 },
194 [AV_PIX_FMT_GBRP10BE] = { 1, 1 },
195 [AV_PIX_FMT_GBRP12LE] = { 1, 1 },
196 [AV_PIX_FMT_GBRP12BE] = { 1, 1 },
197 [AV_PIX_FMT_GBRP14LE] = { 1, 1 },
198 [AV_PIX_FMT_GBRP14BE] = { 1, 1 },
199 [AV_PIX_FMT_GBRP16LE] = { 1, 0 },
200 [AV_PIX_FMT_GBRP16BE] = { 1, 0 },
201 [AV_PIX_FMT_XYZ12BE] = { 1, 0, 1 },
202 [AV_PIX_FMT_XYZ12LE] = { 1, 0, 1 },
203 [AV_PIX_FMT_GBRAP] = { 1, 1 },
204 [AV_PIX_FMT_GBRAP16LE] = { 1, 0 },
205 [AV_PIX_FMT_GBRAP16BE] = { 1, 0 },
208 int sws_isSupportedInput(enum AVPixelFormat pix_fmt)
210 return (unsigned)pix_fmt < AV_PIX_FMT_NB ?
211 format_entries[pix_fmt].is_supported_in : 0;
214 int sws_isSupportedOutput(enum AVPixelFormat pix_fmt)
216 return (unsigned)pix_fmt < AV_PIX_FMT_NB ?
217 format_entries[pix_fmt].is_supported_out : 0;
220 int sws_isSupportedEndiannessConversion(enum AVPixelFormat pix_fmt)
222 return (unsigned)pix_fmt < AV_PIX_FMT_NB ?
223 format_entries[pix_fmt].is_supported_endianness : 0;
226 extern const int32_t ff_yuv2rgb_coeffs[8][4];
228 #if FF_API_SWS_FORMAT_NAME
229 const char *sws_format_name(enum AVPixelFormat format)
231 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(format);
235 return "Unknown format";
239 static double getSplineCoeff(double a, double b, double c, double d,
243 return ((d * dist + c) * dist + b) * dist + a;
245 return getSplineCoeff(0.0,
246 b + 2.0 * c + 3.0 * d,
248 -b - 3.0 * c - 6.0 * d,
252 static av_cold int initFilter(int16_t **outFilter, int32_t **filterPos,
253 int *outFilterSize, int xInc, int srcW,
254 int dstW, int filterAlign, int one,
255 int flags, int cpu_flags,
256 SwsVector *srcFilter, SwsVector *dstFilter,
263 int64_t *filter = NULL;
264 int64_t *filter2 = NULL;
265 const int64_t fone = 1LL << (54 - FFMIN(av_log2(srcW/dstW), 8));
268 emms_c(); // FIXME should not be required but IS (even for non-MMX versions)
270 // NOTE: the +3 is for the MMX(+1) / SSE(+3) scaler which reads over the end
271 FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW + 3) * sizeof(**filterPos), fail);
273 if (FFABS(xInc - 0x10000) < 10) { // unscaled
276 FF_ALLOCZ_OR_GOTO(NULL, filter,
277 dstW * sizeof(*filter) * filterSize, fail);
279 for (i = 0; i < dstW; i++) {
280 filter[i * filterSize] = fone;
283 } else if (flags & SWS_POINT) { // lame looking point sampling mode
287 FF_ALLOC_OR_GOTO(NULL, filter,
288 dstW * sizeof(*filter) * filterSize, fail);
290 xDstInSrc = xInc / 2 - 0x8000;
291 for (i = 0; i < dstW; i++) {
292 int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;
294 (*filterPos)[i] = xx;
298 } else if ((xInc <= (1 << 16) && (flags & SWS_AREA)) ||
299 (flags & SWS_FAST_BILINEAR)) { // bilinear upscale
303 FF_ALLOC_OR_GOTO(NULL, filter,
304 dstW * sizeof(*filter) * filterSize, fail);
306 xDstInSrc = xInc / 2 - 0x8000;
307 for (i = 0; i < dstW; i++) {
308 int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;
311 (*filterPos)[i] = xx;
312 // bilinear upscale / linear interpolate / area averaging
313 for (j = 0; j < filterSize; j++) {
314 int64_t coeff= fone - FFABS(((int64_t)xx<<16) - xDstInSrc)*(fone>>16);
317 filter[i * filterSize + j] = coeff;
326 if (flags & SWS_BICUBIC)
328 else if (flags & SWS_X)
330 else if (flags & SWS_AREA)
331 sizeFactor = 1; // downscale only, for upscale it is bilinear
332 else if (flags & SWS_GAUSS)
333 sizeFactor = 8; // infinite ;)
334 else if (flags & SWS_LANCZOS)
335 sizeFactor = param[0] != SWS_PARAM_DEFAULT ? ceil(2 * param[0]) : 6;
336 else if (flags & SWS_SINC)
337 sizeFactor = 20; // infinite ;)
338 else if (flags & SWS_SPLINE)
339 sizeFactor = 20; // infinite ;)
340 else if (flags & SWS_BILINEAR)
347 filterSize = 1 + sizeFactor; // upscale
349 filterSize = 1 + (sizeFactor * srcW + dstW - 1) / dstW;
351 filterSize = FFMIN(filterSize, srcW - 2);
352 filterSize = FFMAX(filterSize, 1);
354 FF_ALLOC_OR_GOTO(NULL, filter,
355 dstW * sizeof(*filter) * filterSize, fail);
357 xDstInSrc = xInc - 0x10000;
358 for (i = 0; i < dstW; i++) {
359 int xx = (xDstInSrc - ((filterSize - 2) << 16)) / (1 << 17);
361 (*filterPos)[i] = xx;
362 for (j = 0; j < filterSize; j++) {
363 int64_t d = (FFABS(((int64_t)xx << 17) - xDstInSrc)) << 13;
369 floatd = d * (1.0 / (1 << 30));
371 if (flags & SWS_BICUBIC) {
372 int64_t B = (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1 << 24);
373 int64_t C = (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1 << 24);
375 if (d >= 1LL << 31) {
378 int64_t dd = (d * d) >> 30;
379 int64_t ddd = (dd * d) >> 30;
382 coeff = (12 * (1 << 24) - 9 * B - 6 * C) * ddd +
383 (-18 * (1 << 24) + 12 * B + 6 * C) * dd +
384 (6 * (1 << 24) - 2 * B) * (1 << 30);
386 coeff = (-B - 6 * C) * ddd +
387 (6 * B + 30 * C) * dd +
388 (-12 * B - 48 * C) * d +
389 (8 * B + 24 * C) * (1 << 30);
391 coeff /= (1LL<<54)/fone;
394 else if (flags & SWS_X) {
395 double p = param ? param * 0.01 : 0.3;
396 coeff = d ? sin(d * M_PI) / (d * M_PI) : 1.0;
397 coeff *= pow(2.0, -p * d * d);
400 else if (flags & SWS_X) {
401 double A = param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
405 c = cos(floatd * M_PI);
412 coeff = (c * 0.5 + 0.5) * fone;
413 } else if (flags & SWS_AREA) {
414 int64_t d2 = d - (1 << 29);
415 if (d2 * xInc < -(1LL << (29 + 16)))
416 coeff = 1.0 * (1LL << (30 + 16));
417 else if (d2 * xInc < (1LL << (29 + 16)))
418 coeff = -d2 * xInc + (1LL << (29 + 16));
421 coeff *= fone >> (30 + 16);
422 } else if (flags & SWS_GAUSS) {
423 double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
424 coeff = (pow(2.0, -p * floatd * floatd)) * fone;
425 } else if (flags & SWS_SINC) {
426 coeff = (d ? sin(floatd * M_PI) / (floatd * M_PI) : 1.0) * fone;
427 } else if (flags & SWS_LANCZOS) {
428 double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
429 coeff = (d ? sin(floatd * M_PI) * sin(floatd * M_PI / p) /
430 (floatd * floatd * M_PI * M_PI / p) : 1.0) * fone;
433 } else if (flags & SWS_BILINEAR) {
434 coeff = (1 << 30) - d;
438 } else if (flags & SWS_SPLINE) {
439 double p = -2.196152422706632;
440 coeff = getSplineCoeff(1.0, 0.0, p, -p - 1.0, floatd) * fone;
445 filter[i * filterSize + j] = coeff;
448 xDstInSrc += 2 * xInc;
452 /* apply src & dst Filter to filter -> filter2
455 av_assert0(filterSize > 0);
456 filter2Size = filterSize;
458 filter2Size += srcFilter->length - 1;
460 filter2Size += dstFilter->length - 1;
461 av_assert0(filter2Size > 0);
462 FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size * dstW * sizeof(*filter2), fail);
464 for (i = 0; i < dstW; i++) {
468 for (k = 0; k < srcFilter->length; k++) {
469 for (j = 0; j < filterSize; j++)
470 filter2[i * filter2Size + k + j] +=
471 srcFilter->coeff[k] * filter[i * filterSize + j];
474 for (j = 0; j < filterSize; j++)
475 filter2[i * filter2Size + j] = filter[i * filterSize + j];
479 (*filterPos)[i] += (filterSize - 1) / 2 - (filter2Size - 1) / 2;
483 /* try to reduce the filter-size (step1 find size and shift left) */
484 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
486 for (i = dstW - 1; i >= 0; i--) {
487 int min = filter2Size;
489 int64_t cutOff = 0.0;
491 /* get rid of near zero elements on the left by shifting left */
492 for (j = 0; j < filter2Size; j++) {
494 cutOff += FFABS(filter2[i * filter2Size]);
496 if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)
499 /* preserve monotonicity because the core can't handle the
500 * filter otherwise */
501 if (i < dstW - 1 && (*filterPos)[i] >= (*filterPos)[i + 1])
504 // move filter coefficients left
505 for (k = 1; k < filter2Size; k++)
506 filter2[i * filter2Size + k - 1] = filter2[i * filter2Size + k];
507 filter2[i * filter2Size + k - 1] = 0;
512 /* count near zeros on the right */
513 for (j = filter2Size - 1; j > 0; j--) {
514 cutOff += FFABS(filter2[i * filter2Size + j]);
516 if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)
521 if (min > minFilterSize)
525 if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) {
526 // we can handle the special case 4, so we don't want to go the full 8
527 if (minFilterSize < 5)
530 /* We really don't want to waste our time doing useless computation, so
531 * fall back on the scalar C code for very small filters.
532 * Vectorizing is worth it only if you have a decent-sized vector. */
533 if (minFilterSize < 3)
537 if (INLINE_MMX(cpu_flags)) {
538 // special case for unscaled vertical filtering
539 if (minFilterSize == 1 && filterAlign == 2)
543 av_assert0(minFilterSize > 0);
544 filterSize = (minFilterSize + (filterAlign - 1)) & (~(filterAlign - 1));
545 av_assert0(filterSize > 0);
546 filter = av_malloc(filterSize * dstW * sizeof(*filter));
547 if (filterSize >= MAX_FILTER_SIZE * 16 /
548 ((flags & SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter) {
549 av_log(NULL, AV_LOG_ERROR, "sws: filterSize %d is too large, try less extreem scaling or increase MAX_FILTER_SIZE and recompile\n", filterSize);
552 *outFilterSize = filterSize;
554 if (flags & SWS_PRINT_INFO)
555 av_log(NULL, AV_LOG_VERBOSE,
556 "SwScaler: reducing / aligning filtersize %d -> %d\n",
557 filter2Size, filterSize);
558 /* try to reduce the filter-size (step2 reduce it) */
559 for (i = 0; i < dstW; i++) {
562 for (j = 0; j < filterSize; j++) {
563 if (j >= filter2Size)
564 filter[i * filterSize + j] = 0;
566 filter[i * filterSize + j] = filter2[i * filter2Size + j];
567 if ((flags & SWS_BITEXACT) && j >= minFilterSize)
568 filter[i * filterSize + j] = 0;
572 // FIXME try to align filterPos if possible
575 for (i = 0; i < dstW; i++) {
577 if ((*filterPos)[i] < 0) {
578 // move filter coefficients left to compensate for filterPos
579 for (j = 1; j < filterSize; j++) {
580 int left = FFMAX(j + (*filterPos)[i], 0);
581 filter[i * filterSize + left] += filter[i * filterSize + j];
582 filter[i * filterSize + j] = 0;
587 if ((*filterPos)[i] + filterSize > srcW) {
588 int shift = (*filterPos)[i] + filterSize - srcW;
589 // move filter coefficients right to compensate for filterPos
590 for (j = filterSize - 2; j >= 0; j--) {
591 int right = FFMIN(j + shift, filterSize - 1);
592 filter[i * filterSize + right] += filter[i * filterSize + j];
593 filter[i * filterSize + j] = 0;
595 (*filterPos)[i]= srcW - filterSize;
599 // Note the +1 is for the MMX scaler which reads over the end
600 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
601 FF_ALLOCZ_OR_GOTO(NULL, *outFilter,
602 *outFilterSize * (dstW + 3) * sizeof(int16_t), fail);
604 /* normalize & store in outFilter */
605 for (i = 0; i < dstW; i++) {
610 for (j = 0; j < filterSize; j++) {
611 sum += filter[i * filterSize + j];
613 sum = (sum + one / 2) / one;
614 for (j = 0; j < *outFilterSize; j++) {
615 int64_t v = filter[i * filterSize + j] + error;
616 int intV = ROUNDED_DIV(v, sum);
617 (*outFilter)[i * (*outFilterSize) + j] = intV;
618 error = v - intV * sum;
622 (*filterPos)[dstW + 0] =
623 (*filterPos)[dstW + 1] =
624 (*filterPos)[dstW + 2] = (*filterPos)[dstW - 1]; /* the MMX/SSE scaler will
625 * read over the end */
626 for (i = 0; i < *outFilterSize; i++) {
627 int k = (dstW - 1) * (*outFilterSize) + i;
628 (*outFilter)[k + 1 * (*outFilterSize)] =
629 (*outFilter)[k + 2 * (*outFilterSize)] =
630 (*outFilter)[k + 3 * (*outFilterSize)] = (*outFilter)[k];
637 av_log(NULL, AV_LOG_ERROR, "sws: initFilter failed\n");
643 #if HAVE_MMXEXT_INLINE
644 static av_cold int init_hscaler_mmxext(int dstW, int xInc, uint8_t *filterCode,
645 int16_t *filter, int32_t *filterPos,
649 x86_reg imm8OfPShufW1A;
650 x86_reg imm8OfPShufW2A;
651 x86_reg fragmentLengthA;
653 x86_reg imm8OfPShufW1B;
654 x86_reg imm8OfPShufW2B;
655 x86_reg fragmentLengthB;
660 // create an optimized horizontal scaling routine
661 /* This scaler is made of runtime-generated MMXEXT code using specially tuned
662 * pshufw instructions. For every four output pixels, if four input pixels
663 * are enough for the fast bilinear scaling, then a chunk of fragmentB is
664 * used. If five input pixels are needed, then a chunk of fragmentA is used.
673 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
674 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
675 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
676 "punpcklbw %%mm7, %%mm1 \n\t"
677 "punpcklbw %%mm7, %%mm0 \n\t"
678 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
680 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
682 "psubw %%mm1, %%mm0 \n\t"
683 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
684 "pmullw %%mm3, %%mm0 \n\t"
685 "psllw $7, %%mm1 \n\t"
686 "paddw %%mm1, %%mm0 \n\t"
688 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
690 "add $8, %%"REG_a" \n\t"
694 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
695 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
696 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
701 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
705 : "=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
706 "=r" (fragmentLengthA)
713 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
714 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
715 "punpcklbw %%mm7, %%mm0 \n\t"
716 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
718 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
720 "psubw %%mm1, %%mm0 \n\t"
721 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
722 "pmullw %%mm3, %%mm0 \n\t"
723 "psllw $7, %%mm1 \n\t"
724 "paddw %%mm1, %%mm0 \n\t"
726 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
728 "add $8, %%"REG_a" \n\t"
732 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
733 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
734 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
739 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
743 : "=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
744 "=r" (fragmentLengthB)
747 xpos = 0; // lumXInc/2 - 0x8000; // difference between pixel centers
750 for (i = 0; i < dstW / numSplits; i++) {
755 int b = ((xpos + xInc) >> 16) - xx;
756 int c = ((xpos + xInc * 2) >> 16) - xx;
757 int d = ((xpos + xInc * 3) >> 16) - xx;
758 int inc = (d + 1 < 4);
759 uint8_t *fragment = (d + 1 < 4) ? fragmentB : fragmentA;
760 x86_reg imm8OfPShufW1 = (d + 1 < 4) ? imm8OfPShufW1B : imm8OfPShufW1A;
761 x86_reg imm8OfPShufW2 = (d + 1 < 4) ? imm8OfPShufW2B : imm8OfPShufW2A;
762 x86_reg fragmentLength = (d + 1 < 4) ? fragmentLengthB : fragmentLengthA;
763 int maxShift = 3 - (d + inc);
767 filter[i] = ((xpos & 0xFFFF) ^ 0xFFFF) >> 9;
768 filter[i + 1] = (((xpos + xInc) & 0xFFFF) ^ 0xFFFF) >> 9;
769 filter[i + 2] = (((xpos + xInc * 2) & 0xFFFF) ^ 0xFFFF) >> 9;
770 filter[i + 3] = (((xpos + xInc * 3) & 0xFFFF) ^ 0xFFFF) >> 9;
771 filterPos[i / 2] = xx;
773 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
775 filterCode[fragmentPos + imm8OfPShufW1] = (a + inc) |
779 filterCode[fragmentPos + imm8OfPShufW2] = a | (b << 2) |
783 if (i + 4 - inc >= dstW)
784 shift = maxShift; // avoid overread
785 else if ((filterPos[i / 2] & 3) <= maxShift)
786 shift = filterPos[i / 2] & 3; // align
788 if (shift && i >= shift) {
789 filterCode[fragmentPos + imm8OfPShufW1] += 0x55 * shift;
790 filterCode[fragmentPos + imm8OfPShufW2] += 0x55 * shift;
791 filterPos[i / 2] -= shift;
795 fragmentPos += fragmentLength;
798 filterCode[fragmentPos] = RET;
803 filterPos[((i / 2) + 1) & (~1)] = xpos >> 16; // needed to jump to the next part
805 return fragmentPos + 1;
807 #endif /* HAVE_MMXEXT_INLINE */
809 static void getSubSampleFactors(int *h, int *v, enum AVPixelFormat format)
811 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(format);
812 *h = desc->log2_chroma_w;
813 *v = desc->log2_chroma_h;
816 static void fill_rgb2yuv_table(SwsContext *c, const int table[4], int dstRange)
818 int64_t W, V, Z, Cy, Cu, Cv;
819 int64_t vr = table[0];
820 int64_t ub = table[1];
821 int64_t ug = -table[2];
822 int64_t vg = -table[3];
825 uint8_t *p = (uint8_t*)c->input_rgb2yuv_table;
827 static const int8_t map[] = {
828 BY_IDX, GY_IDX, -1 , BY_IDX, BY_IDX, GY_IDX, -1 , BY_IDX,
829 RY_IDX, -1 , GY_IDX, RY_IDX, RY_IDX, -1 , GY_IDX, RY_IDX,
830 RY_IDX, GY_IDX, -1 , RY_IDX, RY_IDX, GY_IDX, -1 , RY_IDX,
831 BY_IDX, -1 , GY_IDX, BY_IDX, BY_IDX, -1 , GY_IDX, BY_IDX,
832 BU_IDX, GU_IDX, -1 , BU_IDX, BU_IDX, GU_IDX, -1 , BU_IDX,
833 RU_IDX, -1 , GU_IDX, RU_IDX, RU_IDX, -1 , GU_IDX, RU_IDX,
834 RU_IDX, GU_IDX, -1 , RU_IDX, RU_IDX, GU_IDX, -1 , RU_IDX,
835 BU_IDX, -1 , GU_IDX, BU_IDX, BU_IDX, -1 , GU_IDX, BU_IDX,
836 BV_IDX, GV_IDX, -1 , BV_IDX, BV_IDX, GV_IDX, -1 , BV_IDX,
837 RV_IDX, -1 , GV_IDX, RV_IDX, RV_IDX, -1 , GV_IDX, RV_IDX,
838 RV_IDX, GV_IDX, -1 , RV_IDX, RV_IDX, GV_IDX, -1 , RV_IDX,
839 BV_IDX, -1 , GV_IDX, BV_IDX, BV_IDX, -1 , GV_IDX, BV_IDX,
840 RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX,
841 BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX,
842 GY_IDX, -1 , GY_IDX, -1 , GY_IDX, -1 , GY_IDX, -1 ,
843 -1 , GY_IDX, -1 , GY_IDX, -1 , GY_IDX, -1 , GY_IDX,
844 RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX,
845 BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX,
846 GU_IDX, -1 , GU_IDX, -1 , GU_IDX, -1 , GU_IDX, -1 ,
847 -1 , GU_IDX, -1 , GU_IDX, -1 , GU_IDX, -1 , GU_IDX,
848 RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX,
849 BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX,
850 GV_IDX, -1 , GV_IDX, -1 , GV_IDX, -1 , GV_IDX, -1 ,
851 -1 , GV_IDX, -1 , GV_IDX, -1 , GV_IDX, -1 , GV_IDX, //23
852 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //24
853 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //25
854 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //26
855 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //27
856 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //28
857 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //29
858 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //30
859 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //31
860 BY_IDX, GY_IDX, RY_IDX, -1 , -1 , -1 , -1 , -1 , //32
861 BU_IDX, GU_IDX, RU_IDX, -1 , -1 , -1 , -1 , -1 , //33
862 BV_IDX, GV_IDX, RV_IDX, -1 , -1 , -1 , -1 , -1 , //34
865 dstRange = 0; //FIXME range = 1 is handled elsewhere
875 W = ROUNDED_DIV(ONE*ONE*ug, ub);
876 V = ROUNDED_DIV(ONE*ONE*vg, vr);
879 Cy = ROUNDED_DIV(cy*Z, ONE);
880 Cu = ROUNDED_DIV(ub*Z, ONE);
881 Cv = ROUNDED_DIV(vr*Z, ONE);
883 c->input_rgb2yuv_table[RY_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*V , Cy);
884 c->input_rgb2yuv_table[GY_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE , Cy);
885 c->input_rgb2yuv_table[BY_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*W , Cy);
887 c->input_rgb2yuv_table[RU_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*V , Cu);
888 c->input_rgb2yuv_table[GU_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE , Cu);
889 c->input_rgb2yuv_table[BU_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*(Z+W) , Cu);
891 c->input_rgb2yuv_table[RV_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*(V+Z) , Cv);
892 c->input_rgb2yuv_table[GV_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE , Cv);
893 c->input_rgb2yuv_table[BV_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*W , Cv);
895 if(/*!dstRange && */table == ff_yuv2rgb_coeffs[SWS_CS_DEFAULT]) {
896 c->input_rgb2yuv_table[BY_IDX] = ((int)(0.114 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
897 c->input_rgb2yuv_table[BV_IDX] = (-(int)(0.081 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
898 c->input_rgb2yuv_table[BU_IDX] = ((int)(0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
899 c->input_rgb2yuv_table[GY_IDX] = ((int)(0.587 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
900 c->input_rgb2yuv_table[GV_IDX] = (-(int)(0.419 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
901 c->input_rgb2yuv_table[GU_IDX] = (-(int)(0.331 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
902 c->input_rgb2yuv_table[RY_IDX] = ((int)(0.299 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
903 c->input_rgb2yuv_table[RV_IDX] = ((int)(0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
904 c->input_rgb2yuv_table[RU_IDX] = (-(int)(0.169 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
906 for(i=0; i<FF_ARRAY_ELEMS(map); i++)
907 AV_WL16(p + 16*4 + 2*i, map[i] >= 0 ? c->input_rgb2yuv_table[map[i]] : 0);
910 static void fill_xyztables(struct SwsContext *c)
913 double xyzgamma = XYZ_GAMMA;
914 double rgbgamma = 1.0 / RGB_GAMMA;
915 static const int16_t xyz2rgb_matrix[3][4] = {
916 {13270, -6295, -2041},
918 { 228, -835, 4329} };
919 static int16_t xyzgamma_tab[4096], rgbgamma_tab[4096];
921 memcpy(c->xyz2rgb_matrix, xyz2rgb_matrix, sizeof(c->xyz2rgb_matrix));
922 c->xyzgamma = xyzgamma_tab;
923 c->rgbgamma = rgbgamma_tab;
925 if (rgbgamma_tab[4095])
928 /* set gamma vectors */
929 for (i = 0; i < 4096; i++) {
930 xyzgamma_tab[i] = lrint(pow(i / 4095.0, xyzgamma) * 4095.0);
931 rgbgamma_tab[i] = lrint(pow(i / 4095.0, rgbgamma) * 4095.0);
935 int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],
936 int srcRange, const int table[4], int dstRange,
937 int brightness, int contrast, int saturation)
939 const AVPixFmtDescriptor *desc_dst = av_pix_fmt_desc_get(c->dstFormat);
940 const AVPixFmtDescriptor *desc_src = av_pix_fmt_desc_get(c->srcFormat);
941 memcpy(c->srcColorspaceTable, inv_table, sizeof(int) * 4);
942 memcpy(c->dstColorspaceTable, table, sizeof(int) * 4);
944 if(!isYUV(c->dstFormat) && !isGray(c->dstFormat))
946 if(!isYUV(c->srcFormat) && !isGray(c->srcFormat))
949 c->brightness = brightness;
950 c->contrast = contrast;
951 c->saturation = saturation;
952 c->srcRange = srcRange;
953 c->dstRange = dstRange;
957 if ((isYUV(c->dstFormat) || isGray(c->dstFormat)) && (isYUV(c->srcFormat) || isGray(c->srcFormat)))
960 c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);
961 c->srcFormatBpp = av_get_bits_per_pixel(desc_src);
963 if (!isYUV(c->dstFormat) && !isGray(c->dstFormat)) {
964 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness,
965 contrast, saturation);
968 if (HAVE_ALTIVEC && av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)
969 ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness,
970 contrast, saturation);
973 fill_rgb2yuv_table(c, table, dstRange);
978 int sws_getColorspaceDetails(struct SwsContext *c, int **inv_table,
979 int *srcRange, int **table, int *dstRange,
980 int *brightness, int *contrast, int *saturation)
982 if (!c || isYUV(c->dstFormat) || isGray(c->dstFormat))
985 *inv_table = c->srcColorspaceTable;
986 *table = c->dstColorspaceTable;
987 *srcRange = c->srcRange;
988 *dstRange = c->dstRange;
989 *brightness = c->brightness;
990 *contrast = c->contrast;
991 *saturation = c->saturation;
996 static int handle_jpeg(enum AVPixelFormat *format)
999 case AV_PIX_FMT_YUVJ420P:
1000 *format = AV_PIX_FMT_YUV420P;
1002 case AV_PIX_FMT_YUVJ422P:
1003 *format = AV_PIX_FMT_YUV422P;
1005 case AV_PIX_FMT_YUVJ444P:
1006 *format = AV_PIX_FMT_YUV444P;
1008 case AV_PIX_FMT_YUVJ440P:
1009 *format = AV_PIX_FMT_YUV440P;
1016 static int handle_0alpha(enum AVPixelFormat *format)
1019 case AV_PIX_FMT_0BGR : *format = AV_PIX_FMT_ABGR ; return 1;
1020 case AV_PIX_FMT_BGR0 : *format = AV_PIX_FMT_BGRA ; return 4;
1021 case AV_PIX_FMT_0RGB : *format = AV_PIX_FMT_ARGB ; return 1;
1022 case AV_PIX_FMT_RGB0 : *format = AV_PIX_FMT_RGBA ; return 4;
1027 static int handle_xyz(enum AVPixelFormat *format)
1030 case AV_PIX_FMT_XYZ12BE : *format = AV_PIX_FMT_RGB48BE; return 1;
1031 case AV_PIX_FMT_XYZ12LE : *format = AV_PIX_FMT_RGB48LE; return 1;
1036 SwsContext *sws_alloc_context(void)
1038 SwsContext *c = av_mallocz(sizeof(SwsContext));
1041 c->av_class = &sws_context_class;
1042 av_opt_set_defaults(c);
1048 av_cold int sws_init_context(SwsContext *c, SwsFilter *srcFilter,
1049 SwsFilter *dstFilter)
1052 int usesVFilter, usesHFilter;
1054 SwsFilter dummyFilter = { NULL, NULL, NULL, NULL };
1059 int dst_stride = FFALIGN(dstW * sizeof(int16_t) + 66, 16);
1060 int flags, cpu_flags;
1061 enum AVPixelFormat srcFormat = c->srcFormat;
1062 enum AVPixelFormat dstFormat = c->dstFormat;
1063 const AVPixFmtDescriptor *desc_src = av_pix_fmt_desc_get(srcFormat);
1064 const AVPixFmtDescriptor *desc_dst = av_pix_fmt_desc_get(dstFormat);
1066 cpu_flags = av_get_cpu_flags();
1072 unscaled = (srcW == dstW && srcH == dstH);
1074 handle_jpeg(&srcFormat);
1075 handle_jpeg(&dstFormat);
1076 if(srcFormat!=c->srcFormat || dstFormat!=c->dstFormat)
1077 av_log(c, AV_LOG_WARNING, "deprecated pixel format used, make sure you did set range correctly\n");
1078 handle_0alpha(&srcFormat);
1079 handle_0alpha(&dstFormat);
1080 handle_xyz(&srcFormat);
1081 handle_xyz(&dstFormat);
1083 if(srcFormat!=c->srcFormat || dstFormat!=c->dstFormat){
1084 c->srcFormat= srcFormat;
1085 c->dstFormat= dstFormat;
1088 if (!(unscaled && sws_isSupportedEndiannessConversion(srcFormat) &&
1089 av_pix_fmt_swap_endianness(srcFormat) == dstFormat)) {
1090 if (!sws_isSupportedInput(srcFormat)) {
1091 av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n",
1092 av_get_pix_fmt_name(srcFormat));
1093 return AVERROR(EINVAL);
1095 if (!sws_isSupportedOutput(dstFormat)) {
1096 av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n",
1097 av_get_pix_fmt_name(dstFormat));
1098 return AVERROR(EINVAL);
1102 i = flags & (SWS_POINT |
1113 if (!i || (i & (i - 1))) {
1114 av_log(c, AV_LOG_ERROR, "Exactly one scaler algorithm must be chosen, got %X\n", i);
1115 return AVERROR(EINVAL);
1118 if (srcW < 1 || srcH < 1 || dstW < 1 || dstH < 1) {
1119 /* FIXME check if these are enough and try to lower them after
1120 * fixing the relevant parts of the code */
1121 av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",
1122 srcW, srcH, dstW, dstH);
1123 return AVERROR(EINVAL);
1127 dstFilter = &dummyFilter;
1129 srcFilter = &dummyFilter;
1131 c->lumXInc = (((int64_t)srcW << 16) + (dstW >> 1)) / dstW;
1132 c->lumYInc = (((int64_t)srcH << 16) + (dstH >> 1)) / dstH;
1133 c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);
1134 c->srcFormatBpp = av_get_bits_per_pixel(desc_src);
1135 c->vRounder = 4 * 0x0001000100010001ULL;
1137 usesVFilter = (srcFilter->lumV && srcFilter->lumV->length > 1) ||
1138 (srcFilter->chrV && srcFilter->chrV->length > 1) ||
1139 (dstFilter->lumV && dstFilter->lumV->length > 1) ||
1140 (dstFilter->chrV && dstFilter->chrV->length > 1);
1141 usesHFilter = (srcFilter->lumH && srcFilter->lumH->length > 1) ||
1142 (srcFilter->chrH && srcFilter->chrH->length > 1) ||
1143 (dstFilter->lumH && dstFilter->lumH->length > 1) ||
1144 (dstFilter->chrH && dstFilter->chrH->length > 1);
1146 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
1147 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
1149 if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) {
1151 av_log(c, AV_LOG_DEBUG, "Forcing full internal H chroma due to odd output size\n");
1152 flags |= SWS_FULL_CHR_H_INT;
1157 if(dstFormat == AV_PIX_FMT_BGR4_BYTE ||
1158 dstFormat == AV_PIX_FMT_RGB4_BYTE ||
1159 dstFormat == AV_PIX_FMT_BGR8 ||
1160 dstFormat == AV_PIX_FMT_RGB8) {
1161 if (flags & SWS_ERROR_DIFFUSION && !(flags & SWS_FULL_CHR_H_INT)) {
1162 av_log(c, AV_LOG_DEBUG,
1163 "Error diffusion dither is only supported in full chroma interpolation for destination format '%s'\n",
1164 av_get_pix_fmt_name(dstFormat));
1165 flags |= SWS_FULL_CHR_H_INT;
1168 if (!(flags & SWS_ERROR_DIFFUSION) && (flags & SWS_FULL_CHR_H_INT)) {
1169 av_log(c, AV_LOG_DEBUG,
1170 "Ordered dither is not supported in full chroma interpolation for destination format '%s'\n",
1171 av_get_pix_fmt_name(dstFormat));
1172 flags |= SWS_ERROR_DIFFUSION;
1176 if (isPlanarRGB(dstFormat)) {
1177 if (!(flags & SWS_FULL_CHR_H_INT)) {
1178 av_log(c, AV_LOG_DEBUG,
1179 "%s output is not supported with half chroma resolution, switching to full\n",
1180 av_get_pix_fmt_name(dstFormat));
1181 flags |= SWS_FULL_CHR_H_INT;
1186 /* reuse chroma for 2 pixels RGB/BGR unless user wants full
1187 * chroma interpolation */
1188 if (flags & SWS_FULL_CHR_H_INT &&
1189 isAnyRGB(dstFormat) &&
1190 !isPlanarRGB(dstFormat) &&
1191 dstFormat != AV_PIX_FMT_RGBA &&
1192 dstFormat != AV_PIX_FMT_ARGB &&
1193 dstFormat != AV_PIX_FMT_BGRA &&
1194 dstFormat != AV_PIX_FMT_ABGR &&
1195 dstFormat != AV_PIX_FMT_RGB24 &&
1196 dstFormat != AV_PIX_FMT_BGR24 &&
1197 dstFormat != AV_PIX_FMT_BGR4_BYTE &&
1198 dstFormat != AV_PIX_FMT_RGB4_BYTE &&
1199 dstFormat != AV_PIX_FMT_BGR8 &&
1200 dstFormat != AV_PIX_FMT_RGB8
1202 av_log(c, AV_LOG_WARNING,
1203 "full chroma interpolation for destination format '%s' not yet implemented\n",
1204 av_get_pix_fmt_name(dstFormat));
1205 flags &= ~SWS_FULL_CHR_H_INT;
1208 if (isAnyRGB(dstFormat) && !(flags & SWS_FULL_CHR_H_INT))
1209 c->chrDstHSubSample = 1;
1211 // drop some chroma lines if the user wants it
1212 c->vChrDrop = (flags & SWS_SRC_V_CHR_DROP_MASK) >>
1213 SWS_SRC_V_CHR_DROP_SHIFT;
1214 c->chrSrcVSubSample += c->vChrDrop;
1216 /* drop every other pixel for chroma calculation unless user
1217 * wants full chroma */
1218 if (isAnyRGB(srcFormat) && !(flags & SWS_FULL_CHR_H_INP) &&
1219 srcFormat != AV_PIX_FMT_RGB8 && srcFormat != AV_PIX_FMT_BGR8 &&
1220 srcFormat != AV_PIX_FMT_RGB4 && srcFormat != AV_PIX_FMT_BGR4 &&
1221 srcFormat != AV_PIX_FMT_RGB4_BYTE && srcFormat != AV_PIX_FMT_BGR4_BYTE &&
1222 srcFormat != AV_PIX_FMT_GBRP9BE && srcFormat != AV_PIX_FMT_GBRP9LE &&
1223 srcFormat != AV_PIX_FMT_GBRP10BE && srcFormat != AV_PIX_FMT_GBRP10LE &&
1224 srcFormat != AV_PIX_FMT_GBRP12BE && srcFormat != AV_PIX_FMT_GBRP12LE &&
1225 srcFormat != AV_PIX_FMT_GBRP14BE && srcFormat != AV_PIX_FMT_GBRP14LE &&
1226 srcFormat != AV_PIX_FMT_GBRP16BE && srcFormat != AV_PIX_FMT_GBRP16LE &&
1227 ((dstW >> c->chrDstHSubSample) <= (srcW >> 1) ||
1228 (flags & SWS_FAST_BILINEAR)))
1229 c->chrSrcHSubSample = 1;
1231 // Note the FF_CEIL_RSHIFT is so that we always round toward +inf.
1232 c->chrSrcW = FF_CEIL_RSHIFT(srcW, c->chrSrcHSubSample);
1233 c->chrSrcH = FF_CEIL_RSHIFT(srcH, c->chrSrcVSubSample);
1234 c->chrDstW = FF_CEIL_RSHIFT(dstW, c->chrDstHSubSample);
1235 c->chrDstH = FF_CEIL_RSHIFT(dstH, c->chrDstVSubSample);
1237 FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW*2+78, 16) * 2, fail);
1239 /* unscaled special cases */
1240 if (unscaled && !usesHFilter && !usesVFilter &&
1241 (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
1242 ff_get_unscaled_swscale(c);
1245 if (flags & SWS_PRINT_INFO)
1246 av_log(c, AV_LOG_INFO,
1247 "using unscaled %s -> %s special converter\n",
1248 av_get_pix_fmt_name(srcFormat), av_get_pix_fmt_name(dstFormat));
1253 c->srcBpc = 1 + desc_src->comp[0].depth_minus1;
1256 c->dstBpc = 1 + desc_dst->comp[0].depth_minus1;
1259 if (isAnyRGB(srcFormat) || srcFormat == AV_PIX_FMT_PAL8)
1261 if (c->dstBpc == 16)
1264 if (INLINE_MMXEXT(cpu_flags) && c->srcBpc == 8 && c->dstBpc <= 14) {
1265 c->canMMXEXTBeUsed = (dstW >= srcW && (dstW & 31) == 0 &&
1266 (srcW & 15) == 0) ? 1 : 0;
1267 if (!c->canMMXEXTBeUsed && dstW >= srcW && (srcW & 15) == 0
1269 && (flags & SWS_FAST_BILINEAR)) {
1270 if (flags & SWS_PRINT_INFO)
1271 av_log(c, AV_LOG_INFO,
1272 "output width is not a multiple of 32 -> no MMXEXT scaler\n");
1274 if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat))
1275 c->canMMXEXTBeUsed = 0;
1277 c->canMMXEXTBeUsed = 0;
1279 c->chrXInc = (((int64_t)c->chrSrcW << 16) + (c->chrDstW >> 1)) / c->chrDstW;
1280 c->chrYInc = (((int64_t)c->chrSrcH << 16) + (c->chrDstH >> 1)) / c->chrDstH;
1282 /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src
1283 * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do
1285 * n-2 is the last chrominance sample available.
1286 * This is not perfect, but no one should notice the difference, the more
1287 * correct variant would be like the vertical one, but that would require
1288 * some special code for the first and last pixel */
1289 if (flags & SWS_FAST_BILINEAR) {
1290 if (c->canMMXEXTBeUsed) {
1294 // we don't use the x86 asm scaler if MMX is available
1295 else if (INLINE_MMX(cpu_flags) && c->dstBpc <= 14) {
1296 c->lumXInc = ((int64_t)(srcW - 2) << 16) / (dstW - 2) - 20;
1297 c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;
1301 #define USE_MMAP (HAVE_MMAP && HAVE_MPROTECT && defined MAP_ANONYMOUS)
1303 /* precalculate horizontal scaler filter coefficients */
1305 #if HAVE_MMXEXT_INLINE
1306 // can't downscale !!!
1307 if (c->canMMXEXTBeUsed && (flags & SWS_FAST_BILINEAR)) {
1308 c->lumMmxextFilterCodeSize = init_hscaler_mmxext(dstW, c->lumXInc, NULL,
1310 c->chrMmxextFilterCodeSize = init_hscaler_mmxext(c->chrDstW, c->chrXInc,
1311 NULL, NULL, NULL, 4);
1314 c->lumMmxextFilterCode = mmap(NULL, c->lumMmxextFilterCodeSize,
1315 PROT_READ | PROT_WRITE,
1316 MAP_PRIVATE | MAP_ANONYMOUS,
1318 c->chrMmxextFilterCode = mmap(NULL, c->chrMmxextFilterCodeSize,
1319 PROT_READ | PROT_WRITE,
1320 MAP_PRIVATE | MAP_ANONYMOUS,
1322 #elif HAVE_VIRTUALALLOC
1323 c->lumMmxextFilterCode = VirtualAlloc(NULL,
1324 c->lumMmxextFilterCodeSize,
1326 PAGE_EXECUTE_READWRITE);
1327 c->chrMmxextFilterCode = VirtualAlloc(NULL,
1328 c->chrMmxextFilterCodeSize,
1330 PAGE_EXECUTE_READWRITE);
1332 c->lumMmxextFilterCode = av_malloc(c->lumMmxextFilterCodeSize);
1333 c->chrMmxextFilterCode = av_malloc(c->chrMmxextFilterCodeSize);
1336 #ifdef MAP_ANONYMOUS
1337 if (c->lumMmxextFilterCode == MAP_FAILED || c->chrMmxextFilterCode == MAP_FAILED)
1339 if (!c->lumMmxextFilterCode || !c->chrMmxextFilterCode)
1342 av_log(c, AV_LOG_ERROR, "Failed to allocate MMX2FilterCode\n");
1343 return AVERROR(ENOMEM);
1346 FF_ALLOCZ_OR_GOTO(c, c->hLumFilter, (dstW / 8 + 8) * sizeof(int16_t), fail);
1347 FF_ALLOCZ_OR_GOTO(c, c->hChrFilter, (c->chrDstW / 4 + 8) * sizeof(int16_t), fail);
1348 FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW / 2 / 8 + 8) * sizeof(int32_t), fail);
1349 FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);
1351 init_hscaler_mmxext( dstW, c->lumXInc, c->lumMmxextFilterCode,
1352 c->hLumFilter, (uint32_t*)c->hLumFilterPos, 8);
1353 init_hscaler_mmxext(c->chrDstW, c->chrXInc, c->chrMmxextFilterCode,
1354 c->hChrFilter, (uint32_t*)c->hChrFilterPos, 4);
1357 mprotect(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize, PROT_EXEC | PROT_READ);
1358 mprotect(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize, PROT_EXEC | PROT_READ);
1361 #endif /* HAVE_MMXEXT_INLINE */
1363 const int filterAlign =
1364 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :
1365 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
1368 if (initFilter(&c->hLumFilter, &c->hLumFilterPos,
1369 &c->hLumFilterSize, c->lumXInc,
1370 srcW, dstW, filterAlign, 1 << 14,
1371 (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
1372 cpu_flags, srcFilter->lumH, dstFilter->lumH,
1375 if (initFilter(&c->hChrFilter, &c->hChrFilterPos,
1376 &c->hChrFilterSize, c->chrXInc,
1377 c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,
1378 (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
1379 cpu_flags, srcFilter->chrH, dstFilter->chrH,
1383 } // initialize horizontal stuff
1385 /* precalculate vertical scaler filter coefficients */
1387 const int filterAlign =
1388 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 2 :
1389 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
1392 if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,
1393 c->lumYInc, srcH, dstH, filterAlign, (1 << 12),
1394 (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
1395 cpu_flags, srcFilter->lumV, dstFilter->lumV,
1398 if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,
1399 c->chrYInc, c->chrSrcH, c->chrDstH,
1400 filterAlign, (1 << 12),
1401 (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
1402 cpu_flags, srcFilter->chrV, dstFilter->chrV,
1407 FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH, fail);
1408 FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);
1410 for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {
1412 short *p = (short *)&c->vYCoeffsBank[i];
1413 for (j = 0; j < 8; j++)
1414 p[j] = c->vLumFilter[i];
1417 for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {
1419 short *p = (short *)&c->vCCoeffsBank[i];
1420 for (j = 0; j < 8; j++)
1421 p[j] = c->vChrFilter[i];
1426 // calculate buffer sizes so that they won't run out while handling these damn slices
1427 c->vLumBufSize = c->vLumFilterSize;
1428 c->vChrBufSize = c->vChrFilterSize;
1429 for (i = 0; i < dstH; i++) {
1430 int chrI = (int64_t)i * c->chrDstH / dstH;
1431 int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,
1432 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)
1433 << c->chrSrcVSubSample));
1435 nextSlice >>= c->chrSrcVSubSample;
1436 nextSlice <<= c->chrSrcVSubSample;
1437 if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)
1438 c->vLumBufSize = nextSlice - c->vLumFilterPos[i];
1439 if (c->vChrFilterPos[chrI] + c->vChrBufSize <
1440 (nextSlice >> c->chrSrcVSubSample))
1441 c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -
1442 c->vChrFilterPos[chrI];
1445 for (i = 0; i < 4; i++)
1446 FF_ALLOCZ_OR_GOTO(c, c->dither_error[i], (c->dstW+2) * sizeof(int), fail);
1448 /* Allocate pixbufs (we use dynamic allocation because otherwise we would
1449 * need to allocate several megabytes to handle all possible cases) */
1450 FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);
1451 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
1452 FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
1453 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
1454 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);
1455 /* Note we need at least one pixel more at the end because of the MMX code
1456 * (just in case someone wants to replace the 4000/8000). */
1457 /* align at 16 bytes for AltiVec */
1458 for (i = 0; i < c->vLumBufSize; i++) {
1459 FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],
1460 dst_stride + 16, fail);
1461 c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];
1463 // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)
1464 c->uv_off = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);
1465 c->uv_offx2 = dst_stride + 16;
1466 for (i = 0; i < c->vChrBufSize; i++) {
1467 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],
1468 dst_stride * 2 + 32, fail);
1469 c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];
1470 c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]
1471 = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;
1473 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
1474 for (i = 0; i < c->vLumBufSize; i++) {
1475 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i + c->vLumBufSize],
1476 dst_stride + 16, fail);
1477 c->alpPixBuf[i] = c->alpPixBuf[i + c->vLumBufSize];
1480 // try to avoid drawing green stuff between the right end and the stride end
1481 for (i = 0; i < c->vChrBufSize; i++)
1482 if(desc_dst->comp[0].depth_minus1 == 15){
1483 av_assert0(c->dstBpc > 14);
1484 for(j=0; j<dst_stride/2+1; j++)
1485 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;
1487 for(j=0; j<dst_stride+1; j++)
1488 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;
1490 av_assert0(c->chrDstH <= dstH);
1492 if (flags & SWS_PRINT_INFO) {
1493 if (flags & SWS_FAST_BILINEAR)
1494 av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
1495 else if (flags & SWS_BILINEAR)
1496 av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
1497 else if (flags & SWS_BICUBIC)
1498 av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
1499 else if (flags & SWS_X)
1500 av_log(c, AV_LOG_INFO, "Experimental scaler, ");
1501 else if (flags & SWS_POINT)
1502 av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
1503 else if (flags & SWS_AREA)
1504 av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
1505 else if (flags & SWS_BICUBLIN)
1506 av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
1507 else if (flags & SWS_GAUSS)
1508 av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
1509 else if (flags & SWS_SINC)
1510 av_log(c, AV_LOG_INFO, "Sinc scaler, ");
1511 else if (flags & SWS_LANCZOS)
1512 av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
1513 else if (flags & SWS_SPLINE)
1514 av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
1516 av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
1518 av_log(c, AV_LOG_INFO, "from %s to %s%s ",
1519 av_get_pix_fmt_name(srcFormat),
1521 dstFormat == AV_PIX_FMT_BGR555 || dstFormat == AV_PIX_FMT_BGR565 ||
1522 dstFormat == AV_PIX_FMT_RGB444BE || dstFormat == AV_PIX_FMT_RGB444LE ||
1523 dstFormat == AV_PIX_FMT_BGR444BE || dstFormat == AV_PIX_FMT_BGR444LE ?
1528 av_get_pix_fmt_name(dstFormat));
1530 if (INLINE_MMXEXT(cpu_flags))
1531 av_log(c, AV_LOG_INFO, "using MMXEXT\n");
1532 else if (INLINE_AMD3DNOW(cpu_flags))
1533 av_log(c, AV_LOG_INFO, "using 3DNOW\n");
1534 else if (INLINE_MMX(cpu_flags))
1535 av_log(c, AV_LOG_INFO, "using MMX\n");
1536 else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC)
1537 av_log(c, AV_LOG_INFO, "using AltiVec\n");
1539 av_log(c, AV_LOG_INFO, "using C\n");
1541 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
1542 av_log(c, AV_LOG_DEBUG,
1543 "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1544 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
1545 av_log(c, AV_LOG_DEBUG,
1546 "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1547 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,
1548 c->chrXInc, c->chrYInc);
1551 c->swScale = ff_getSwsFunc(c);
1553 fail: // FIXME replace things by appropriate error codes
1557 #if FF_API_SWS_GETCONTEXT
1558 SwsContext *sws_getContext(int srcW, int srcH, enum AVPixelFormat srcFormat,
1559 int dstW, int dstH, enum AVPixelFormat dstFormat,
1560 int flags, SwsFilter *srcFilter,
1561 SwsFilter *dstFilter, const double *param)
1565 if (!(c = sws_alloc_context()))
1573 c->srcRange = handle_jpeg(&srcFormat);
1574 c->dstRange = handle_jpeg(&dstFormat);
1575 c->src0Alpha = handle_0alpha(&srcFormat);
1576 c->dst0Alpha = handle_0alpha(&dstFormat);
1577 c->srcXYZ = handle_xyz(&srcFormat);
1578 c->dstXYZ = handle_xyz(&dstFormat);
1579 c->srcFormat = srcFormat;
1580 c->dstFormat = dstFormat;
1583 c->param[0] = param[0];
1584 c->param[1] = param[1];
1586 sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,
1587 ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,
1588 c->dstRange, 0, 1 << 16, 1 << 16);
1590 if (sws_init_context(c, srcFilter, dstFilter) < 0) {
1599 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
1600 float lumaSharpen, float chromaSharpen,
1601 float chromaHShift, float chromaVShift,
1604 SwsFilter *filter = av_malloc(sizeof(SwsFilter));
1608 if (lumaGBlur != 0.0) {
1609 filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);
1610 filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);
1612 filter->lumH = sws_getIdentityVec();
1613 filter->lumV = sws_getIdentityVec();
1616 if (chromaGBlur != 0.0) {
1617 filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);
1618 filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);
1620 filter->chrH = sws_getIdentityVec();
1621 filter->chrV = sws_getIdentityVec();
1624 if (chromaSharpen != 0.0) {
1625 SwsVector *id = sws_getIdentityVec();
1626 sws_scaleVec(filter->chrH, -chromaSharpen);
1627 sws_scaleVec(filter->chrV, -chromaSharpen);
1628 sws_addVec(filter->chrH, id);
1629 sws_addVec(filter->chrV, id);
1633 if (lumaSharpen != 0.0) {
1634 SwsVector *id = sws_getIdentityVec();
1635 sws_scaleVec(filter->lumH, -lumaSharpen);
1636 sws_scaleVec(filter->lumV, -lumaSharpen);
1637 sws_addVec(filter->lumH, id);
1638 sws_addVec(filter->lumV, id);
1642 if (chromaHShift != 0.0)
1643 sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));
1645 if (chromaVShift != 0.0)
1646 sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));
1648 sws_normalizeVec(filter->chrH, 1.0);
1649 sws_normalizeVec(filter->chrV, 1.0);
1650 sws_normalizeVec(filter->lumH, 1.0);
1651 sws_normalizeVec(filter->lumV, 1.0);
1654 sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
1656 sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
1661 SwsVector *sws_allocVec(int length)
1665 if(length <= 0 || length > INT_MAX/ sizeof(double))
1668 vec = av_malloc(sizeof(SwsVector));
1671 vec->length = length;
1672 vec->coeff = av_malloc(sizeof(double) * length);
1678 SwsVector *sws_getGaussianVec(double variance, double quality)
1680 const int length = (int)(variance * quality + 0.5) | 1;
1682 double middle = (length - 1) * 0.5;
1685 if(variance < 0 || quality < 0)
1688 vec = sws_allocVec(length);
1693 for (i = 0; i < length; i++) {
1694 double dist = i - middle;
1695 vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /
1696 sqrt(2 * variance * M_PI);
1699 sws_normalizeVec(vec, 1.0);
1704 SwsVector *sws_getConstVec(double c, int length)
1707 SwsVector *vec = sws_allocVec(length);
1712 for (i = 0; i < length; i++)
1718 SwsVector *sws_getIdentityVec(void)
1720 return sws_getConstVec(1.0, 1);
1723 static double sws_dcVec(SwsVector *a)
1728 for (i = 0; i < a->length; i++)
1734 void sws_scaleVec(SwsVector *a, double scalar)
1738 for (i = 0; i < a->length; i++)
1739 a->coeff[i] *= scalar;
1742 void sws_normalizeVec(SwsVector *a, double height)
1744 sws_scaleVec(a, height / sws_dcVec(a));
1747 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
1749 int length = a->length + b->length - 1;
1751 SwsVector *vec = sws_getConstVec(0.0, length);
1756 for (i = 0; i < a->length; i++) {
1757 for (j = 0; j < b->length; j++) {
1758 vec->coeff[i + j] += a->coeff[i] * b->coeff[j];
1765 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
1767 int length = FFMAX(a->length, b->length);
1769 SwsVector *vec = sws_getConstVec(0.0, length);
1774 for (i = 0; i < a->length; i++)
1775 vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
1776 for (i = 0; i < b->length; i++)
1777 vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] += b->coeff[i];
1782 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
1784 int length = FFMAX(a->length, b->length);
1786 SwsVector *vec = sws_getConstVec(0.0, length);
1791 for (i = 0; i < a->length; i++)
1792 vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
1793 for (i = 0; i < b->length; i++)
1794 vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] -= b->coeff[i];
1799 /* shift left / or right if "shift" is negative */
1800 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
1802 int length = a->length + FFABS(shift) * 2;
1804 SwsVector *vec = sws_getConstVec(0.0, length);
1809 for (i = 0; i < a->length; i++) {
1810 vec->coeff[i + (length - 1) / 2 -
1811 (a->length - 1) / 2 - shift] = a->coeff[i];
1817 void sws_shiftVec(SwsVector *a, int shift)
1819 SwsVector *shifted = sws_getShiftedVec(a, shift);
1821 a->coeff = shifted->coeff;
1822 a->length = shifted->length;
1826 void sws_addVec(SwsVector *a, SwsVector *b)
1828 SwsVector *sum = sws_sumVec(a, b);
1830 a->coeff = sum->coeff;
1831 a->length = sum->length;
1835 void sws_subVec(SwsVector *a, SwsVector *b)
1837 SwsVector *diff = sws_diffVec(a, b);
1839 a->coeff = diff->coeff;
1840 a->length = diff->length;
1844 void sws_convVec(SwsVector *a, SwsVector *b)
1846 SwsVector *conv = sws_getConvVec(a, b);
1848 a->coeff = conv->coeff;
1849 a->length = conv->length;
1853 SwsVector *sws_cloneVec(SwsVector *a)
1856 SwsVector *vec = sws_allocVec(a->length);
1861 for (i = 0; i < a->length; i++)
1862 vec->coeff[i] = a->coeff[i];
1867 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
1874 for (i = 0; i < a->length; i++)
1875 if (a->coeff[i] > max)
1878 for (i = 0; i < a->length; i++)
1879 if (a->coeff[i] < min)
1884 for (i = 0; i < a->length; i++) {
1885 int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);
1886 av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
1888 av_log(log_ctx, log_level, " ");
1889 av_log(log_ctx, log_level, "|\n");
1893 void sws_freeVec(SwsVector *a)
1897 av_freep(&a->coeff);
1902 void sws_freeFilter(SwsFilter *filter)
1908 sws_freeVec(filter->lumH);
1910 sws_freeVec(filter->lumV);
1912 sws_freeVec(filter->chrH);
1914 sws_freeVec(filter->chrV);
1918 void sws_freeContext(SwsContext *c)
1925 for (i = 0; i < c->vLumBufSize; i++)
1926 av_freep(&c->lumPixBuf[i]);
1927 av_freep(&c->lumPixBuf);
1930 if (c->chrUPixBuf) {
1931 for (i = 0; i < c->vChrBufSize; i++)
1932 av_freep(&c->chrUPixBuf[i]);
1933 av_freep(&c->chrUPixBuf);
1934 av_freep(&c->chrVPixBuf);
1937 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1938 for (i = 0; i < c->vLumBufSize; i++)
1939 av_freep(&c->alpPixBuf[i]);
1940 av_freep(&c->alpPixBuf);
1943 for (i = 0; i < 4; i++)
1944 av_freep(&c->dither_error[i]);
1946 av_freep(&c->vLumFilter);
1947 av_freep(&c->vChrFilter);
1948 av_freep(&c->hLumFilter);
1949 av_freep(&c->hChrFilter);
1951 av_freep(&c->vYCoeffsBank);
1952 av_freep(&c->vCCoeffsBank);
1955 av_freep(&c->vLumFilterPos);
1956 av_freep(&c->vChrFilterPos);
1957 av_freep(&c->hLumFilterPos);
1958 av_freep(&c->hChrFilterPos);
1962 if (c->lumMmxextFilterCode)
1963 munmap(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize);
1964 if (c->chrMmxextFilterCode)
1965 munmap(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize);
1966 #elif HAVE_VIRTUALALLOC
1967 if (c->lumMmxextFilterCode)
1968 VirtualFree(c->lumMmxextFilterCode, 0, MEM_RELEASE);
1969 if (c->chrMmxextFilterCode)
1970 VirtualFree(c->chrMmxextFilterCode, 0, MEM_RELEASE);
1972 av_free(c->lumMmxextFilterCode);
1973 av_free(c->chrMmxextFilterCode);
1975 c->lumMmxextFilterCode = NULL;
1976 c->chrMmxextFilterCode = NULL;
1977 #endif /* HAVE_MMX_INLINE */
1979 av_freep(&c->yuvTable);
1980 av_freep(&c->formatConvBuffer);
1985 struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,
1986 int srcH, enum AVPixelFormat srcFormat,
1988 enum AVPixelFormat dstFormat, int flags,
1989 SwsFilter *srcFilter,
1990 SwsFilter *dstFilter,
1991 const double *param)
1993 static const double default_param[2] = { SWS_PARAM_DEFAULT,
1994 SWS_PARAM_DEFAULT };
1997 param = default_param;
2000 (context->srcW != srcW ||
2001 context->srcH != srcH ||
2002 context->srcFormat != srcFormat ||
2003 context->dstW != dstW ||
2004 context->dstH != dstH ||
2005 context->dstFormat != dstFormat ||
2006 context->flags != flags ||
2007 context->param[0] != param[0] ||
2008 context->param[1] != param[1])) {
2009 sws_freeContext(context);
2014 if (!(context = sws_alloc_context()))
2016 context->srcW = srcW;
2017 context->srcH = srcH;
2018 context->srcRange = handle_jpeg(&srcFormat);
2019 context->src0Alpha = handle_0alpha(&srcFormat);
2020 context->srcXYZ = handle_xyz(&srcFormat);
2021 context->srcFormat = srcFormat;
2022 context->dstW = dstW;
2023 context->dstH = dstH;
2024 context->dstRange = handle_jpeg(&dstFormat);
2025 context->dst0Alpha = handle_0alpha(&dstFormat);
2026 context->dstXYZ = handle_xyz(&dstFormat);
2027 context->dstFormat = dstFormat;
2028 context->flags = flags;
2029 context->param[0] = param[0];
2030 context->param[1] = param[1];
2031 sws_setColorspaceDetails(context, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],
2033 ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,
2034 context->dstRange, 0, 1 << 16, 1 << 16);
2035 if (sws_init_context(context, srcFilter, dstFilter) < 0) {
2036 sws_freeContext(context);