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 int is_supported_in, is_supported_out;
78 static const FormatEntry format_entries[AV_PIX_FMT_NB] = {
79 [AV_PIX_FMT_YUV420P] = { 1, 1 },
80 [AV_PIX_FMT_YUYV422] = { 1, 1 },
81 [AV_PIX_FMT_RGB24] = { 1, 1 },
82 [AV_PIX_FMT_BGR24] = { 1, 1 },
83 [AV_PIX_FMT_YUV422P] = { 1, 1 },
84 [AV_PIX_FMT_YUV444P] = { 1, 1 },
85 [AV_PIX_FMT_YUV410P] = { 1, 1 },
86 [AV_PIX_FMT_YUV411P] = { 1, 1 },
87 [AV_PIX_FMT_GRAY8] = { 1, 1 },
88 [AV_PIX_FMT_MONOWHITE] = { 1, 1 },
89 [AV_PIX_FMT_MONOBLACK] = { 1, 1 },
90 [AV_PIX_FMT_PAL8] = { 1, 0 },
91 [AV_PIX_FMT_YUVJ420P] = { 1, 1 },
92 [AV_PIX_FMT_YUVJ422P] = { 1, 1 },
93 [AV_PIX_FMT_YUVJ444P] = { 1, 1 },
94 [AV_PIX_FMT_UYVY422] = { 1, 1 },
95 [AV_PIX_FMT_UYYVYY411] = { 0, 0 },
96 [AV_PIX_FMT_BGR8] = { 1, 1 },
97 [AV_PIX_FMT_BGR4] = { 0, 1 },
98 [AV_PIX_FMT_BGR4_BYTE] = { 1, 1 },
99 [AV_PIX_FMT_RGB8] = { 1, 1 },
100 [AV_PIX_FMT_RGB4] = { 0, 1 },
101 [AV_PIX_FMT_RGB4_BYTE] = { 1, 1 },
102 [AV_PIX_FMT_NV12] = { 1, 1 },
103 [AV_PIX_FMT_NV21] = { 1, 1 },
104 [AV_PIX_FMT_ARGB] = { 1, 1 },
105 [AV_PIX_FMT_RGBA] = { 1, 1 },
106 [AV_PIX_FMT_ABGR] = { 1, 1 },
107 [AV_PIX_FMT_BGRA] = { 1, 1 },
108 [AV_PIX_FMT_0RGB] = { 1, 1 },
109 [AV_PIX_FMT_RGB0] = { 1, 1 },
110 [AV_PIX_FMT_0BGR] = { 1, 1 },
111 [AV_PIX_FMT_BGR0] = { 1, 1 },
112 [AV_PIX_FMT_GRAY16BE] = { 1, 1 },
113 [AV_PIX_FMT_GRAY16LE] = { 1, 1 },
114 [AV_PIX_FMT_YUV440P] = { 1, 1 },
115 [AV_PIX_FMT_YUVJ440P] = { 1, 1 },
116 [AV_PIX_FMT_YUVA420P] = { 1, 1 },
117 [AV_PIX_FMT_YUVA422P] = { 1, 1 },
118 [AV_PIX_FMT_YUVA444P] = { 1, 1 },
119 [AV_PIX_FMT_YUVA420P9BE] = { 1, 1 },
120 [AV_PIX_FMT_YUVA420P9LE] = { 1, 1 },
121 [AV_PIX_FMT_YUVA422P9BE] = { 1, 1 },
122 [AV_PIX_FMT_YUVA422P9LE] = { 1, 1 },
123 [AV_PIX_FMT_YUVA444P9BE] = { 1, 1 },
124 [AV_PIX_FMT_YUVA444P9LE] = { 1, 1 },
125 [AV_PIX_FMT_YUVA420P10BE]= { 1, 1 },
126 [AV_PIX_FMT_YUVA420P10LE]= { 1, 1 },
127 [AV_PIX_FMT_YUVA422P10BE]= { 1, 1 },
128 [AV_PIX_FMT_YUVA422P10LE]= { 1, 1 },
129 [AV_PIX_FMT_YUVA444P10BE]= { 1, 1 },
130 [AV_PIX_FMT_YUVA444P10LE]= { 1, 1 },
131 [AV_PIX_FMT_YUVA420P16BE]= { 1, 1 },
132 [AV_PIX_FMT_YUVA420P16LE]= { 1, 1 },
133 [AV_PIX_FMT_YUVA422P16BE]= { 1, 1 },
134 [AV_PIX_FMT_YUVA422P16LE]= { 1, 1 },
135 [AV_PIX_FMT_YUVA444P16BE]= { 1, 1 },
136 [AV_PIX_FMT_YUVA444P16LE]= { 1, 1 },
137 [AV_PIX_FMT_RGB48BE] = { 1, 1 },
138 [AV_PIX_FMT_RGB48LE] = { 1, 1 },
139 [AV_PIX_FMT_RGBA64BE] = { 1, 0 },
140 [AV_PIX_FMT_RGBA64LE] = { 1, 0 },
141 [AV_PIX_FMT_RGB565BE] = { 1, 1 },
142 [AV_PIX_FMT_RGB565LE] = { 1, 1 },
143 [AV_PIX_FMT_RGB555BE] = { 1, 1 },
144 [AV_PIX_FMT_RGB555LE] = { 1, 1 },
145 [AV_PIX_FMT_BGR565BE] = { 1, 1 },
146 [AV_PIX_FMT_BGR565LE] = { 1, 1 },
147 [AV_PIX_FMT_BGR555BE] = { 1, 1 },
148 [AV_PIX_FMT_BGR555LE] = { 1, 1 },
149 [AV_PIX_FMT_YUV420P16LE] = { 1, 1 },
150 [AV_PIX_FMT_YUV420P16BE] = { 1, 1 },
151 [AV_PIX_FMT_YUV422P16LE] = { 1, 1 },
152 [AV_PIX_FMT_YUV422P16BE] = { 1, 1 },
153 [AV_PIX_FMT_YUV444P16LE] = { 1, 1 },
154 [AV_PIX_FMT_YUV444P16BE] = { 1, 1 },
155 [AV_PIX_FMT_RGB444LE] = { 1, 1 },
156 [AV_PIX_FMT_RGB444BE] = { 1, 1 },
157 [AV_PIX_FMT_BGR444LE] = { 1, 1 },
158 [AV_PIX_FMT_BGR444BE] = { 1, 1 },
159 [AV_PIX_FMT_Y400A] = { 1, 0 },
160 [AV_PIX_FMT_BGR48BE] = { 1, 1 },
161 [AV_PIX_FMT_BGR48LE] = { 1, 1 },
162 [AV_PIX_FMT_BGRA64BE] = { 0, 0 },
163 [AV_PIX_FMT_BGRA64LE] = { 0, 0 },
164 [AV_PIX_FMT_YUV420P9BE] = { 1, 1 },
165 [AV_PIX_FMT_YUV420P9LE] = { 1, 1 },
166 [AV_PIX_FMT_YUV420P10BE] = { 1, 1 },
167 [AV_PIX_FMT_YUV420P10LE] = { 1, 1 },
168 [AV_PIX_FMT_YUV420P12BE] = { 1, 1 },
169 [AV_PIX_FMT_YUV420P12LE] = { 1, 1 },
170 [AV_PIX_FMT_YUV420P14BE] = { 1, 1 },
171 [AV_PIX_FMT_YUV420P14LE] = { 1, 1 },
172 [AV_PIX_FMT_YUV422P9BE] = { 1, 1 },
173 [AV_PIX_FMT_YUV422P9LE] = { 1, 1 },
174 [AV_PIX_FMT_YUV422P10BE] = { 1, 1 },
175 [AV_PIX_FMT_YUV422P10LE] = { 1, 1 },
176 [AV_PIX_FMT_YUV422P12BE] = { 1, 1 },
177 [AV_PIX_FMT_YUV422P12LE] = { 1, 1 },
178 [AV_PIX_FMT_YUV422P14BE] = { 1, 1 },
179 [AV_PIX_FMT_YUV422P14LE] = { 1, 1 },
180 [AV_PIX_FMT_YUV444P9BE] = { 1, 1 },
181 [AV_PIX_FMT_YUV444P9LE] = { 1, 1 },
182 [AV_PIX_FMT_YUV444P10BE] = { 1, 1 },
183 [AV_PIX_FMT_YUV444P10LE] = { 1, 1 },
184 [AV_PIX_FMT_YUV444P12BE] = { 1, 1 },
185 [AV_PIX_FMT_YUV444P12LE] = { 1, 1 },
186 [AV_PIX_FMT_YUV444P14BE] = { 1, 1 },
187 [AV_PIX_FMT_YUV444P14LE] = { 1, 1 },
188 [AV_PIX_FMT_GBRP] = { 1, 1 },
189 [AV_PIX_FMT_GBRP9LE] = { 1, 1 },
190 [AV_PIX_FMT_GBRP9BE] = { 1, 1 },
191 [AV_PIX_FMT_GBRP10LE] = { 1, 1 },
192 [AV_PIX_FMT_GBRP10BE] = { 1, 1 },
193 [AV_PIX_FMT_GBRP12LE] = { 1, 1 },
194 [AV_PIX_FMT_GBRP12BE] = { 1, 1 },
195 [AV_PIX_FMT_GBRP14LE] = { 1, 1 },
196 [AV_PIX_FMT_GBRP14BE] = { 1, 1 },
197 [AV_PIX_FMT_GBRP16LE] = { 1, 0 },
198 [AV_PIX_FMT_GBRP16BE] = { 1, 0 },
199 [AV_PIX_FMT_XYZ12BE] = { 1, 0 },
200 [AV_PIX_FMT_XYZ12LE] = { 1, 0 },
201 [AV_PIX_FMT_GBRAP] = { 0, 0 },
202 [AV_PIX_FMT_GBRAP16LE] = { 0, 0 },
203 [AV_PIX_FMT_GBRAP16BE] = { 0, 0 },
206 int sws_isSupportedInput(enum AVPixelFormat pix_fmt)
208 return (unsigned)pix_fmt < AV_PIX_FMT_NB ?
209 format_entries[pix_fmt].is_supported_in : 0;
212 int sws_isSupportedOutput(enum AVPixelFormat pix_fmt)
214 return (unsigned)pix_fmt < AV_PIX_FMT_NB ?
215 format_entries[pix_fmt].is_supported_out : 0;
218 extern const int32_t ff_yuv2rgb_coeffs[8][4];
220 #if FF_API_SWS_FORMAT_NAME
221 const char *sws_format_name(enum AVPixelFormat format)
223 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(format);
227 return "Unknown format";
231 static double getSplineCoeff(double a, double b, double c, double d,
235 return ((d * dist + c) * dist + b) * dist + a;
237 return getSplineCoeff(0.0,
238 b + 2.0 * c + 3.0 * d,
240 -b - 3.0 * c - 6.0 * d,
244 static av_cold int initFilter(int16_t **outFilter, int32_t **filterPos,
245 int *outFilterSize, int xInc, int srcW,
246 int dstW, int filterAlign, int one,
247 int flags, int cpu_flags,
248 SwsVector *srcFilter, SwsVector *dstFilter,
255 int64_t *filter = NULL;
256 int64_t *filter2 = NULL;
257 const int64_t fone = 1LL << (54 - FFMIN(av_log2(srcW/dstW), 8));
260 emms_c(); // FIXME should not be required but IS (even for non-MMX versions)
262 // NOTE: the +3 is for the MMX(+1) / SSE(+3) scaler which reads over the end
263 FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW + 3) * sizeof(**filterPos), fail);
265 if (FFABS(xInc - 0x10000) < 10) { // unscaled
268 FF_ALLOCZ_OR_GOTO(NULL, filter,
269 dstW * sizeof(*filter) * filterSize, fail);
271 for (i = 0; i < dstW; i++) {
272 filter[i * filterSize] = fone;
275 } else if (flags & SWS_POINT) { // lame looking point sampling mode
279 FF_ALLOC_OR_GOTO(NULL, filter,
280 dstW * sizeof(*filter) * filterSize, fail);
282 xDstInSrc = xInc / 2 - 0x8000;
283 for (i = 0; i < dstW; i++) {
284 int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;
286 (*filterPos)[i] = xx;
290 } else if ((xInc <= (1 << 16) && (flags & SWS_AREA)) ||
291 (flags & SWS_FAST_BILINEAR)) { // bilinear upscale
295 FF_ALLOC_OR_GOTO(NULL, filter,
296 dstW * sizeof(*filter) * filterSize, fail);
298 xDstInSrc = xInc / 2 - 0x8000;
299 for (i = 0; i < dstW; i++) {
300 int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;
303 (*filterPos)[i] = xx;
304 // bilinear upscale / linear interpolate / area averaging
305 for (j = 0; j < filterSize; j++) {
306 int64_t coeff= fone - FFABS(((int64_t)xx<<16) - xDstInSrc)*(fone>>16);
309 filter[i * filterSize + j] = coeff;
318 if (flags & SWS_BICUBIC)
320 else if (flags & SWS_X)
322 else if (flags & SWS_AREA)
323 sizeFactor = 1; // downscale only, for upscale it is bilinear
324 else if (flags & SWS_GAUSS)
325 sizeFactor = 8; // infinite ;)
326 else if (flags & SWS_LANCZOS)
327 sizeFactor = param[0] != SWS_PARAM_DEFAULT ? ceil(2 * param[0]) : 6;
328 else if (flags & SWS_SINC)
329 sizeFactor = 20; // infinite ;)
330 else if (flags & SWS_SPLINE)
331 sizeFactor = 20; // infinite ;)
332 else if (flags & SWS_BILINEAR)
339 filterSize = 1 + sizeFactor; // upscale
341 filterSize = 1 + (sizeFactor * srcW + dstW - 1) / dstW;
343 filterSize = FFMIN(filterSize, srcW - 2);
344 filterSize = FFMAX(filterSize, 1);
346 FF_ALLOC_OR_GOTO(NULL, filter,
347 dstW * sizeof(*filter) * filterSize, fail);
349 xDstInSrc = xInc - 0x10000;
350 for (i = 0; i < dstW; i++) {
351 int xx = (xDstInSrc - ((filterSize - 2) << 16)) / (1 << 17);
353 (*filterPos)[i] = xx;
354 for (j = 0; j < filterSize; j++) {
355 int64_t d = (FFABS(((int64_t)xx << 17) - xDstInSrc)) << 13;
361 floatd = d * (1.0 / (1 << 30));
363 if (flags & SWS_BICUBIC) {
364 int64_t B = (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1 << 24);
365 int64_t C = (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1 << 24);
367 if (d >= 1LL << 31) {
370 int64_t dd = (d * d) >> 30;
371 int64_t ddd = (dd * d) >> 30;
374 coeff = (12 * (1 << 24) - 9 * B - 6 * C) * ddd +
375 (-18 * (1 << 24) + 12 * B + 6 * C) * dd +
376 (6 * (1 << 24) - 2 * B) * (1 << 30);
378 coeff = (-B - 6 * C) * ddd +
379 (6 * B + 30 * C) * dd +
380 (-12 * B - 48 * C) * d +
381 (8 * B + 24 * C) * (1 << 30);
383 coeff /= (1LL<<54)/fone;
386 else if (flags & SWS_X) {
387 double p = param ? param * 0.01 : 0.3;
388 coeff = d ? sin(d * M_PI) / (d * M_PI) : 1.0;
389 coeff *= pow(2.0, -p * d * d);
392 else if (flags & SWS_X) {
393 double A = param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
397 c = cos(floatd * M_PI);
404 coeff = (c * 0.5 + 0.5) * fone;
405 } else if (flags & SWS_AREA) {
406 int64_t d2 = d - (1 << 29);
407 if (d2 * xInc < -(1LL << (29 + 16)))
408 coeff = 1.0 * (1LL << (30 + 16));
409 else if (d2 * xInc < (1LL << (29 + 16)))
410 coeff = -d2 * xInc + (1LL << (29 + 16));
413 coeff *= fone >> (30 + 16);
414 } else if (flags & SWS_GAUSS) {
415 double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
416 coeff = (pow(2.0, -p * floatd * floatd)) * fone;
417 } else if (flags & SWS_SINC) {
418 coeff = (d ? sin(floatd * M_PI) / (floatd * M_PI) : 1.0) * fone;
419 } else if (flags & SWS_LANCZOS) {
420 double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
421 coeff = (d ? sin(floatd * M_PI) * sin(floatd * M_PI / p) /
422 (floatd * floatd * M_PI * M_PI / p) : 1.0) * fone;
425 } else if (flags & SWS_BILINEAR) {
426 coeff = (1 << 30) - d;
430 } else if (flags & SWS_SPLINE) {
431 double p = -2.196152422706632;
432 coeff = getSplineCoeff(1.0, 0.0, p, -p - 1.0, floatd) * fone;
437 filter[i * filterSize + j] = coeff;
440 xDstInSrc += 2 * xInc;
444 /* apply src & dst Filter to filter -> filter2
447 av_assert0(filterSize > 0);
448 filter2Size = filterSize;
450 filter2Size += srcFilter->length - 1;
452 filter2Size += dstFilter->length - 1;
453 av_assert0(filter2Size > 0);
454 FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size * dstW * sizeof(*filter2), fail);
456 for (i = 0; i < dstW; i++) {
460 for (k = 0; k < srcFilter->length; k++) {
461 for (j = 0; j < filterSize; j++)
462 filter2[i * filter2Size + k + j] +=
463 srcFilter->coeff[k] * filter[i * filterSize + j];
466 for (j = 0; j < filterSize; j++)
467 filter2[i * filter2Size + j] = filter[i * filterSize + j];
471 (*filterPos)[i] += (filterSize - 1) / 2 - (filter2Size - 1) / 2;
475 /* try to reduce the filter-size (step1 find size and shift left) */
476 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
478 for (i = dstW - 1; i >= 0; i--) {
479 int min = filter2Size;
481 int64_t cutOff = 0.0;
483 /* get rid of near zero elements on the left by shifting left */
484 for (j = 0; j < filter2Size; j++) {
486 cutOff += FFABS(filter2[i * filter2Size]);
488 if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)
491 /* preserve monotonicity because the core can't handle the
492 * filter otherwise */
493 if (i < dstW - 1 && (*filterPos)[i] >= (*filterPos)[i + 1])
496 // move filter coefficients left
497 for (k = 1; k < filter2Size; k++)
498 filter2[i * filter2Size + k - 1] = filter2[i * filter2Size + k];
499 filter2[i * filter2Size + k - 1] = 0;
504 /* count near zeros on the right */
505 for (j = filter2Size - 1; j > 0; j--) {
506 cutOff += FFABS(filter2[i * filter2Size + j]);
508 if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)
513 if (min > minFilterSize)
517 if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) {
518 // we can handle the special case 4, so we don't want to go the full 8
519 if (minFilterSize < 5)
522 /* We really don't want to waste our time doing useless computation, so
523 * fall back on the scalar C code for very small filters.
524 * Vectorizing is worth it only if you have a decent-sized vector. */
525 if (minFilterSize < 3)
529 if (INLINE_MMX(cpu_flags)) {
530 // special case for unscaled vertical filtering
531 if (minFilterSize == 1 && filterAlign == 2)
535 av_assert0(minFilterSize > 0);
536 filterSize = (minFilterSize + (filterAlign - 1)) & (~(filterAlign - 1));
537 av_assert0(filterSize > 0);
538 filter = av_malloc(filterSize * dstW * sizeof(*filter));
539 if (filterSize >= MAX_FILTER_SIZE * 16 /
540 ((flags & SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter) {
541 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);
544 *outFilterSize = filterSize;
546 if (flags & SWS_PRINT_INFO)
547 av_log(NULL, AV_LOG_VERBOSE,
548 "SwScaler: reducing / aligning filtersize %d -> %d\n",
549 filter2Size, filterSize);
550 /* try to reduce the filter-size (step2 reduce it) */
551 for (i = 0; i < dstW; i++) {
554 for (j = 0; j < filterSize; j++) {
555 if (j >= filter2Size)
556 filter[i * filterSize + j] = 0;
558 filter[i * filterSize + j] = filter2[i * filter2Size + j];
559 if ((flags & SWS_BITEXACT) && j >= minFilterSize)
560 filter[i * filterSize + j] = 0;
564 // FIXME try to align filterPos if possible
567 for (i = 0; i < dstW; i++) {
569 if ((*filterPos)[i] < 0) {
570 // move filter coefficients left to compensate for filterPos
571 for (j = 1; j < filterSize; j++) {
572 int left = FFMAX(j + (*filterPos)[i], 0);
573 filter[i * filterSize + left] += filter[i * filterSize + j];
574 filter[i * filterSize + j] = 0;
579 if ((*filterPos)[i] + filterSize > srcW) {
580 int shift = (*filterPos)[i] + filterSize - srcW;
581 // move filter coefficients right to compensate for filterPos
582 for (j = filterSize - 2; j >= 0; j--) {
583 int right = FFMIN(j + shift, filterSize - 1);
584 filter[i * filterSize + right] += filter[i * filterSize + j];
585 filter[i * filterSize + j] = 0;
587 (*filterPos)[i]= srcW - filterSize;
591 // Note the +1 is for the MMX scaler which reads over the end
592 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
593 FF_ALLOCZ_OR_GOTO(NULL, *outFilter,
594 *outFilterSize * (dstW + 3) * sizeof(int16_t), fail);
596 /* normalize & store in outFilter */
597 for (i = 0; i < dstW; i++) {
602 for (j = 0; j < filterSize; j++) {
603 sum += filter[i * filterSize + j];
605 sum = (sum + one / 2) / one;
606 for (j = 0; j < *outFilterSize; j++) {
607 int64_t v = filter[i * filterSize + j] + error;
608 int intV = ROUNDED_DIV(v, sum);
609 (*outFilter)[i * (*outFilterSize) + j] = intV;
610 error = v - intV * sum;
614 (*filterPos)[dstW + 0] =
615 (*filterPos)[dstW + 1] =
616 (*filterPos)[dstW + 2] = (*filterPos)[dstW - 1]; /* the MMX/SSE scaler will
617 * read over the end */
618 for (i = 0; i < *outFilterSize; i++) {
619 int k = (dstW - 1) * (*outFilterSize) + i;
620 (*outFilter)[k + 1 * (*outFilterSize)] =
621 (*outFilter)[k + 2 * (*outFilterSize)] =
622 (*outFilter)[k + 3 * (*outFilterSize)] = (*outFilter)[k];
629 av_log(NULL, AV_LOG_ERROR, "sws: initFilter failed\n");
635 #if HAVE_MMXEXT_INLINE
636 static av_cold int init_hscaler_mmxext(int dstW, int xInc, uint8_t *filterCode,
637 int16_t *filter, int32_t *filterPos,
641 x86_reg imm8OfPShufW1A;
642 x86_reg imm8OfPShufW2A;
643 x86_reg fragmentLengthA;
645 x86_reg imm8OfPShufW1B;
646 x86_reg imm8OfPShufW2B;
647 x86_reg fragmentLengthB;
652 // create an optimized horizontal scaling routine
653 /* This scaler is made of runtime-generated MMXEXT code using specially tuned
654 * pshufw instructions. For every four output pixels, if four input pixels
655 * are enough for the fast bilinear scaling, then a chunk of fragmentB is
656 * used. If five input pixels are needed, then a chunk of fragmentA is used.
665 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
666 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
667 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
668 "punpcklbw %%mm7, %%mm1 \n\t"
669 "punpcklbw %%mm7, %%mm0 \n\t"
670 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
672 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
674 "psubw %%mm1, %%mm0 \n\t"
675 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
676 "pmullw %%mm3, %%mm0 \n\t"
677 "psllw $7, %%mm1 \n\t"
678 "paddw %%mm1, %%mm0 \n\t"
680 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
682 "add $8, %%"REG_a" \n\t"
686 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
687 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
688 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
693 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
697 : "=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
698 "=r" (fragmentLengthA)
705 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
706 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
707 "punpcklbw %%mm7, %%mm0 \n\t"
708 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
710 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
712 "psubw %%mm1, %%mm0 \n\t"
713 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
714 "pmullw %%mm3, %%mm0 \n\t"
715 "psllw $7, %%mm1 \n\t"
716 "paddw %%mm1, %%mm0 \n\t"
718 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
720 "add $8, %%"REG_a" \n\t"
724 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
725 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
726 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
731 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
735 : "=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
736 "=r" (fragmentLengthB)
739 xpos = 0; // lumXInc/2 - 0x8000; // difference between pixel centers
742 for (i = 0; i < dstW / numSplits; i++) {
747 int b = ((xpos + xInc) >> 16) - xx;
748 int c = ((xpos + xInc * 2) >> 16) - xx;
749 int d = ((xpos + xInc * 3) >> 16) - xx;
750 int inc = (d + 1 < 4);
751 uint8_t *fragment = (d + 1 < 4) ? fragmentB : fragmentA;
752 x86_reg imm8OfPShufW1 = (d + 1 < 4) ? imm8OfPShufW1B : imm8OfPShufW1A;
753 x86_reg imm8OfPShufW2 = (d + 1 < 4) ? imm8OfPShufW2B : imm8OfPShufW2A;
754 x86_reg fragmentLength = (d + 1 < 4) ? fragmentLengthB : fragmentLengthA;
755 int maxShift = 3 - (d + inc);
759 filter[i] = ((xpos & 0xFFFF) ^ 0xFFFF) >> 9;
760 filter[i + 1] = (((xpos + xInc) & 0xFFFF) ^ 0xFFFF) >> 9;
761 filter[i + 2] = (((xpos + xInc * 2) & 0xFFFF) ^ 0xFFFF) >> 9;
762 filter[i + 3] = (((xpos + xInc * 3) & 0xFFFF) ^ 0xFFFF) >> 9;
763 filterPos[i / 2] = xx;
765 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
767 filterCode[fragmentPos + imm8OfPShufW1] = (a + inc) |
771 filterCode[fragmentPos + imm8OfPShufW2] = a | (b << 2) |
775 if (i + 4 - inc >= dstW)
776 shift = maxShift; // avoid overread
777 else if ((filterPos[i / 2] & 3) <= maxShift)
778 shift = filterPos[i / 2] & 3; // align
780 if (shift && i >= shift) {
781 filterCode[fragmentPos + imm8OfPShufW1] += 0x55 * shift;
782 filterCode[fragmentPos + imm8OfPShufW2] += 0x55 * shift;
783 filterPos[i / 2] -= shift;
787 fragmentPos += fragmentLength;
790 filterCode[fragmentPos] = RET;
795 filterPos[((i / 2) + 1) & (~1)] = xpos >> 16; // needed to jump to the next part
797 return fragmentPos + 1;
799 #endif /* HAVE_MMXEXT_INLINE */
801 static void getSubSampleFactors(int *h, int *v, enum AVPixelFormat format)
803 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(format);
804 *h = desc->log2_chroma_w;
805 *v = desc->log2_chroma_h;
808 static void fill_rgb2yuv_table(SwsContext *c, const int table[4], int dstRange)
810 int64_t W, V, Z, Cy, Cu, Cv;
811 int64_t vr = table[0];
812 int64_t ub = table[1];
813 int64_t ug = -table[2];
814 int64_t vg = -table[3];
817 uint8_t *p = (uint8_t*)c->input_rgb2yuv_table;
819 static const int8_t map[] = {
820 BY_IDX, GY_IDX, -1 , BY_IDX, BY_IDX, GY_IDX, -1 , BY_IDX,
821 RY_IDX, -1 , GY_IDX, RY_IDX, RY_IDX, -1 , GY_IDX, RY_IDX,
822 RY_IDX, GY_IDX, -1 , RY_IDX, RY_IDX, GY_IDX, -1 , RY_IDX,
823 BY_IDX, -1 , GY_IDX, BY_IDX, BY_IDX, -1 , GY_IDX, BY_IDX,
824 BU_IDX, GU_IDX, -1 , BU_IDX, BU_IDX, GU_IDX, -1 , BU_IDX,
825 RU_IDX, -1 , GU_IDX, RU_IDX, RU_IDX, -1 , GU_IDX, RU_IDX,
826 RU_IDX, GU_IDX, -1 , RU_IDX, RU_IDX, GU_IDX, -1 , RU_IDX,
827 BU_IDX, -1 , GU_IDX, BU_IDX, BU_IDX, -1 , GU_IDX, BU_IDX,
828 BV_IDX, GV_IDX, -1 , BV_IDX, BV_IDX, GV_IDX, -1 , BV_IDX,
829 RV_IDX, -1 , GV_IDX, RV_IDX, RV_IDX, -1 , GV_IDX, RV_IDX,
830 RV_IDX, GV_IDX, -1 , RV_IDX, RV_IDX, GV_IDX, -1 , RV_IDX,
831 BV_IDX, -1 , GV_IDX, BV_IDX, BV_IDX, -1 , GV_IDX, BV_IDX,
832 RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX,
833 BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX,
834 GY_IDX, -1 , GY_IDX, -1 , GY_IDX, -1 , GY_IDX, -1 ,
835 -1 , GY_IDX, -1 , GY_IDX, -1 , GY_IDX, -1 , GY_IDX,
836 RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX,
837 BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX,
838 GU_IDX, -1 , GU_IDX, -1 , GU_IDX, -1 , GU_IDX, -1 ,
839 -1 , GU_IDX, -1 , GU_IDX, -1 , GU_IDX, -1 , GU_IDX,
840 RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX,
841 BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX,
842 GV_IDX, -1 , GV_IDX, -1 , GV_IDX, -1 , GV_IDX, -1 ,
843 -1 , GV_IDX, -1 , GV_IDX, -1 , GV_IDX, -1 , GV_IDX, //23
844 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //24
845 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //25
846 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //26
847 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //27
848 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //28
849 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //29
850 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //30
851 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //31
852 BY_IDX, GY_IDX, RY_IDX, -1 , -1 , -1 , -1 , -1 , //32
853 BU_IDX, GU_IDX, RU_IDX, -1 , -1 , -1 , -1 , -1 , //33
854 BV_IDX, GV_IDX, RV_IDX, -1 , -1 , -1 , -1 , -1 , //34
857 dstRange = 0; //FIXME range = 1 is handled elsewhere
867 W = ROUNDED_DIV(ONE*ONE*ug, ub);
868 V = ROUNDED_DIV(ONE*ONE*vg, vr);
871 Cy = ROUNDED_DIV(cy*Z, ONE);
872 Cu = ROUNDED_DIV(ub*Z, ONE);
873 Cv = ROUNDED_DIV(vr*Z, ONE);
875 c->input_rgb2yuv_table[RY_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*V , Cy);
876 c->input_rgb2yuv_table[GY_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE , Cy);
877 c->input_rgb2yuv_table[BY_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*W , Cy);
879 c->input_rgb2yuv_table[RU_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*V , Cu);
880 c->input_rgb2yuv_table[GU_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE , Cu);
881 c->input_rgb2yuv_table[BU_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*(Z+W) , Cu);
883 c->input_rgb2yuv_table[RV_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*(V+Z) , Cv);
884 c->input_rgb2yuv_table[GV_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE , Cv);
885 c->input_rgb2yuv_table[BV_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*W , Cv);
887 if(/*!dstRange && */table == ff_yuv2rgb_coeffs[SWS_CS_DEFAULT]) {
888 c->input_rgb2yuv_table[BY_IDX] = ((int)(0.114 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
889 c->input_rgb2yuv_table[BV_IDX] = (-(int)(0.081 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
890 c->input_rgb2yuv_table[BU_IDX] = ((int)(0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
891 c->input_rgb2yuv_table[GY_IDX] = ((int)(0.587 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
892 c->input_rgb2yuv_table[GV_IDX] = (-(int)(0.419 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
893 c->input_rgb2yuv_table[GU_IDX] = (-(int)(0.331 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
894 c->input_rgb2yuv_table[RY_IDX] = ((int)(0.299 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
895 c->input_rgb2yuv_table[RV_IDX] = ((int)(0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
896 c->input_rgb2yuv_table[RU_IDX] = (-(int)(0.169 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
898 for(i=0; i<FF_ARRAY_ELEMS(map); i++)
899 AV_WL16(p + 16*4 + 2*i, map[i] >= 0 ? c->input_rgb2yuv_table[map[i]] : 0);
902 static void fill_xyztables(struct SwsContext *c)
905 double xyzgamma = XYZ_GAMMA;
906 double rgbgamma = 1.0 / RGB_GAMMA;
907 static const int16_t xyz2rgb_matrix[3][4] = {
908 {13270, -6295, -2041},
910 { 228, -835, 4329} };
911 static int16_t xyzgamma_tab[4096], rgbgamma_tab[4096];
913 memcpy(c->xyz2rgb_matrix, xyz2rgb_matrix, sizeof(c->xyz2rgb_matrix));
914 c->xyzgamma = xyzgamma_tab;
915 c->rgbgamma = rgbgamma_tab;
917 if (rgbgamma_tab[4095])
920 /* set gamma vectors */
921 for (i = 0; i < 4096; i++) {
922 xyzgamma_tab[i] = lrint(pow(i / 4095.0, xyzgamma) * 4095.0);
923 rgbgamma_tab[i] = lrint(pow(i / 4095.0, rgbgamma) * 4095.0);
927 int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],
928 int srcRange, const int table[4], int dstRange,
929 int brightness, int contrast, int saturation)
931 const AVPixFmtDescriptor *desc_dst = av_pix_fmt_desc_get(c->dstFormat);
932 const AVPixFmtDescriptor *desc_src = av_pix_fmt_desc_get(c->srcFormat);
933 memcpy(c->srcColorspaceTable, inv_table, sizeof(int) * 4);
934 memcpy(c->dstColorspaceTable, table, sizeof(int) * 4);
936 if(!isYUV(c->dstFormat) && !isGray(c->dstFormat))
938 if(!isYUV(c->srcFormat) && !isGray(c->srcFormat))
941 c->brightness = brightness;
942 c->contrast = contrast;
943 c->saturation = saturation;
944 c->srcRange = srcRange;
945 c->dstRange = dstRange;
949 if ((isYUV(c->dstFormat) || isGray(c->dstFormat)) && (isYUV(c->srcFormat) || isGray(c->srcFormat)))
952 c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);
953 c->srcFormatBpp = av_get_bits_per_pixel(desc_src);
955 if (!isYUV(c->dstFormat) && !isGray(c->dstFormat)) {
956 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness,
957 contrast, saturation);
960 if (HAVE_ALTIVEC && av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)
961 ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness,
962 contrast, saturation);
965 fill_rgb2yuv_table(c, table, dstRange);
970 int sws_getColorspaceDetails(struct SwsContext *c, int **inv_table,
971 int *srcRange, int **table, int *dstRange,
972 int *brightness, int *contrast, int *saturation)
974 if (!c || isYUV(c->dstFormat) || isGray(c->dstFormat))
977 *inv_table = c->srcColorspaceTable;
978 *table = c->dstColorspaceTable;
979 *srcRange = c->srcRange;
980 *dstRange = c->dstRange;
981 *brightness = c->brightness;
982 *contrast = c->contrast;
983 *saturation = c->saturation;
988 static int handle_jpeg(enum AVPixelFormat *format)
991 case AV_PIX_FMT_YUVJ420P:
992 *format = AV_PIX_FMT_YUV420P;
994 case AV_PIX_FMT_YUVJ422P:
995 *format = AV_PIX_FMT_YUV422P;
997 case AV_PIX_FMT_YUVJ444P:
998 *format = AV_PIX_FMT_YUV444P;
1000 case AV_PIX_FMT_YUVJ440P:
1001 *format = AV_PIX_FMT_YUV440P;
1008 static int handle_0alpha(enum AVPixelFormat *format)
1011 case AV_PIX_FMT_0BGR : *format = AV_PIX_FMT_ABGR ; return 1;
1012 case AV_PIX_FMT_BGR0 : *format = AV_PIX_FMT_BGRA ; return 4;
1013 case AV_PIX_FMT_0RGB : *format = AV_PIX_FMT_ARGB ; return 1;
1014 case AV_PIX_FMT_RGB0 : *format = AV_PIX_FMT_RGBA ; return 4;
1019 static int handle_xyz(enum AVPixelFormat *format)
1022 case AV_PIX_FMT_XYZ12BE : *format = AV_PIX_FMT_RGB48BE; return 1;
1023 case AV_PIX_FMT_XYZ12LE : *format = AV_PIX_FMT_RGB48LE; return 1;
1028 SwsContext *sws_alloc_context(void)
1030 SwsContext *c = av_mallocz(sizeof(SwsContext));
1033 c->av_class = &sws_context_class;
1034 av_opt_set_defaults(c);
1040 av_cold int sws_init_context(SwsContext *c, SwsFilter *srcFilter,
1041 SwsFilter *dstFilter)
1044 int usesVFilter, usesHFilter;
1046 SwsFilter dummyFilter = { NULL, NULL, NULL, NULL };
1051 int dst_stride = FFALIGN(dstW * sizeof(int16_t) + 66, 16);
1052 int flags, cpu_flags;
1053 enum AVPixelFormat srcFormat = c->srcFormat;
1054 enum AVPixelFormat dstFormat = c->dstFormat;
1055 const AVPixFmtDescriptor *desc_src = av_pix_fmt_desc_get(srcFormat);
1056 const AVPixFmtDescriptor *desc_dst = av_pix_fmt_desc_get(dstFormat);
1058 cpu_flags = av_get_cpu_flags();
1064 unscaled = (srcW == dstW && srcH == dstH);
1066 handle_jpeg(&srcFormat);
1067 handle_jpeg(&dstFormat);
1068 if(srcFormat!=c->srcFormat || dstFormat!=c->dstFormat)
1069 av_log(c, AV_LOG_WARNING, "deprecated pixel format used, make sure you did set range correctly\n");
1070 handle_0alpha(&srcFormat);
1071 handle_0alpha(&dstFormat);
1072 handle_xyz(&srcFormat);
1073 handle_xyz(&dstFormat);
1075 if(srcFormat!=c->srcFormat || dstFormat!=c->dstFormat){
1076 c->srcFormat= srcFormat;
1077 c->dstFormat= dstFormat;
1080 if (!sws_isSupportedInput(srcFormat)) {
1081 av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n",
1082 av_get_pix_fmt_name(srcFormat));
1083 return AVERROR(EINVAL);
1085 if (!sws_isSupportedOutput(dstFormat)) {
1086 av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n",
1087 av_get_pix_fmt_name(dstFormat));
1088 return AVERROR(EINVAL);
1091 i = flags & (SWS_POINT |
1102 if (!i || (i & (i - 1))) {
1103 av_log(c, AV_LOG_ERROR, "Exactly one scaler algorithm must be chosen, got %X\n", i);
1104 return AVERROR(EINVAL);
1107 if (srcW < 1 || srcH < 1 || dstW < 1 || dstH < 1) {
1108 /* FIXME check if these are enough and try to lower them after
1109 * fixing the relevant parts of the code */
1110 av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",
1111 srcW, srcH, dstW, dstH);
1112 return AVERROR(EINVAL);
1116 dstFilter = &dummyFilter;
1118 srcFilter = &dummyFilter;
1120 c->lumXInc = (((int64_t)srcW << 16) + (dstW >> 1)) / dstW;
1121 c->lumYInc = (((int64_t)srcH << 16) + (dstH >> 1)) / dstH;
1122 c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);
1123 c->srcFormatBpp = av_get_bits_per_pixel(desc_src);
1124 c->vRounder = 4 * 0x0001000100010001ULL;
1126 usesVFilter = (srcFilter->lumV && srcFilter->lumV->length > 1) ||
1127 (srcFilter->chrV && srcFilter->chrV->length > 1) ||
1128 (dstFilter->lumV && dstFilter->lumV->length > 1) ||
1129 (dstFilter->chrV && dstFilter->chrV->length > 1);
1130 usesHFilter = (srcFilter->lumH && srcFilter->lumH->length > 1) ||
1131 (srcFilter->chrH && srcFilter->chrH->length > 1) ||
1132 (dstFilter->lumH && dstFilter->lumH->length > 1) ||
1133 (dstFilter->chrH && dstFilter->chrH->length > 1);
1135 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
1136 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
1138 if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) {
1140 av_log(c, AV_LOG_DEBUG, "Forcing full internal H chroma due to odd output size\n");
1141 flags |= SWS_FULL_CHR_H_INT;
1146 if(dstFormat == AV_PIX_FMT_BGR4_BYTE ||
1147 dstFormat == AV_PIX_FMT_RGB4_BYTE ||
1148 dstFormat == AV_PIX_FMT_BGR8 ||
1149 dstFormat == AV_PIX_FMT_RGB8) {
1150 if (flags & SWS_ERROR_DIFFUSION && !(flags & SWS_FULL_CHR_H_INT)) {
1151 av_log(c, AV_LOG_DEBUG,
1152 "Error diffusion dither is only supported in full chroma interpolation for destination format '%s'\n",
1153 av_get_pix_fmt_name(dstFormat));
1154 flags |= SWS_FULL_CHR_H_INT;
1157 if (!(flags & SWS_ERROR_DIFFUSION) && (flags & SWS_FULL_CHR_H_INT)) {
1158 av_log(c, AV_LOG_DEBUG,
1159 "Ordered dither is not supported in full chroma interpolation for destination format '%s'\n",
1160 av_get_pix_fmt_name(dstFormat));
1161 flags |= SWS_ERROR_DIFFUSION;
1165 if (isPlanarRGB(dstFormat)) {
1166 if (!(flags & SWS_FULL_CHR_H_INT)) {
1167 av_log(c, AV_LOG_DEBUG,
1168 "%s output is not supported with half chroma resolution, switching to full\n",
1169 av_get_pix_fmt_name(dstFormat));
1170 flags |= SWS_FULL_CHR_H_INT;
1175 /* reuse chroma for 2 pixels RGB/BGR unless user wants full
1176 * chroma interpolation */
1177 if (flags & SWS_FULL_CHR_H_INT &&
1178 isAnyRGB(dstFormat) &&
1179 !isPlanarRGB(dstFormat) &&
1180 dstFormat != AV_PIX_FMT_RGBA &&
1181 dstFormat != AV_PIX_FMT_ARGB &&
1182 dstFormat != AV_PIX_FMT_BGRA &&
1183 dstFormat != AV_PIX_FMT_ABGR &&
1184 dstFormat != AV_PIX_FMT_RGB24 &&
1185 dstFormat != AV_PIX_FMT_BGR24 &&
1186 dstFormat != AV_PIX_FMT_BGR4_BYTE &&
1187 dstFormat != AV_PIX_FMT_RGB4_BYTE &&
1188 dstFormat != AV_PIX_FMT_BGR8 &&
1189 dstFormat != AV_PIX_FMT_RGB8
1191 av_log(c, AV_LOG_WARNING,
1192 "full chroma interpolation for destination format '%s' not yet implemented\n",
1193 av_get_pix_fmt_name(dstFormat));
1194 flags &= ~SWS_FULL_CHR_H_INT;
1197 if (isAnyRGB(dstFormat) && !(flags & SWS_FULL_CHR_H_INT))
1198 c->chrDstHSubSample = 1;
1200 // drop some chroma lines if the user wants it
1201 c->vChrDrop = (flags & SWS_SRC_V_CHR_DROP_MASK) >>
1202 SWS_SRC_V_CHR_DROP_SHIFT;
1203 c->chrSrcVSubSample += c->vChrDrop;
1205 /* drop every other pixel for chroma calculation unless user
1206 * wants full chroma */
1207 if (isAnyRGB(srcFormat) && !(flags & SWS_FULL_CHR_H_INP) &&
1208 srcFormat != AV_PIX_FMT_RGB8 && srcFormat != AV_PIX_FMT_BGR8 &&
1209 srcFormat != AV_PIX_FMT_RGB4 && srcFormat != AV_PIX_FMT_BGR4 &&
1210 srcFormat != AV_PIX_FMT_RGB4_BYTE && srcFormat != AV_PIX_FMT_BGR4_BYTE &&
1211 srcFormat != AV_PIX_FMT_GBRP9BE && srcFormat != AV_PIX_FMT_GBRP9LE &&
1212 srcFormat != AV_PIX_FMT_GBRP10BE && srcFormat != AV_PIX_FMT_GBRP10LE &&
1213 srcFormat != AV_PIX_FMT_GBRP12BE && srcFormat != AV_PIX_FMT_GBRP12LE &&
1214 srcFormat != AV_PIX_FMT_GBRP14BE && srcFormat != AV_PIX_FMT_GBRP14LE &&
1215 srcFormat != AV_PIX_FMT_GBRP16BE && srcFormat != AV_PIX_FMT_GBRP16LE &&
1216 ((dstW >> c->chrDstHSubSample) <= (srcW >> 1) ||
1217 (flags & SWS_FAST_BILINEAR)))
1218 c->chrSrcHSubSample = 1;
1220 // Note the -((-x)>>y) is so that we always round toward +inf.
1221 c->chrSrcW = -((-srcW) >> c->chrSrcHSubSample);
1222 c->chrSrcH = -((-srcH) >> c->chrSrcVSubSample);
1223 c->chrDstW = -((-dstW) >> c->chrDstHSubSample);
1224 c->chrDstH = -((-dstH) >> c->chrDstVSubSample);
1226 FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW*2+78, 16) * 2, fail);
1228 /* unscaled special cases */
1229 if (unscaled && !usesHFilter && !usesVFilter &&
1230 (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
1231 ff_get_unscaled_swscale(c);
1234 if (flags & SWS_PRINT_INFO)
1235 av_log(c, AV_LOG_INFO,
1236 "using unscaled %s -> %s special converter\n",
1237 av_get_pix_fmt_name(srcFormat), av_get_pix_fmt_name(dstFormat));
1242 c->srcBpc = 1 + desc_src->comp[0].depth_minus1;
1245 c->dstBpc = 1 + desc_dst->comp[0].depth_minus1;
1248 if (isAnyRGB(srcFormat) || srcFormat == AV_PIX_FMT_PAL8)
1250 if (c->dstBpc == 16)
1253 if (INLINE_MMXEXT(cpu_flags) && c->srcBpc == 8 && c->dstBpc <= 14) {
1254 c->canMMXEXTBeUsed = (dstW >= srcW && (dstW & 31) == 0 &&
1255 (srcW & 15) == 0) ? 1 : 0;
1256 if (!c->canMMXEXTBeUsed && dstW >= srcW && (srcW & 15) == 0
1258 && (flags & SWS_FAST_BILINEAR)) {
1259 if (flags & SWS_PRINT_INFO)
1260 av_log(c, AV_LOG_INFO,
1261 "output width is not a multiple of 32 -> no MMXEXT scaler\n");
1263 if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat))
1264 c->canMMXEXTBeUsed = 0;
1266 c->canMMXEXTBeUsed = 0;
1268 c->chrXInc = (((int64_t)c->chrSrcW << 16) + (c->chrDstW >> 1)) / c->chrDstW;
1269 c->chrYInc = (((int64_t)c->chrSrcH << 16) + (c->chrDstH >> 1)) / c->chrDstH;
1271 /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src
1272 * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do
1274 * n-2 is the last chrominance sample available.
1275 * This is not perfect, but no one should notice the difference, the more
1276 * correct variant would be like the vertical one, but that would require
1277 * some special code for the first and last pixel */
1278 if (flags & SWS_FAST_BILINEAR) {
1279 if (c->canMMXEXTBeUsed) {
1283 // we don't use the x86 asm scaler if MMX is available
1284 else if (INLINE_MMX(cpu_flags) && c->dstBpc <= 14) {
1285 c->lumXInc = ((int64_t)(srcW - 2) << 16) / (dstW - 2) - 20;
1286 c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;
1290 #define USE_MMAP (HAVE_MMAP && HAVE_MPROTECT && defined MAP_ANONYMOUS)
1292 /* precalculate horizontal scaler filter coefficients */
1294 #if HAVE_MMXEXT_INLINE
1295 // can't downscale !!!
1296 if (c->canMMXEXTBeUsed && (flags & SWS_FAST_BILINEAR)) {
1297 c->lumMmxextFilterCodeSize = init_hscaler_mmxext(dstW, c->lumXInc, NULL,
1299 c->chrMmxextFilterCodeSize = init_hscaler_mmxext(c->chrDstW, c->chrXInc,
1300 NULL, NULL, NULL, 4);
1303 c->lumMmxextFilterCode = mmap(NULL, c->lumMmxextFilterCodeSize,
1304 PROT_READ | PROT_WRITE,
1305 MAP_PRIVATE | MAP_ANONYMOUS,
1307 c->chrMmxextFilterCode = mmap(NULL, c->chrMmxextFilterCodeSize,
1308 PROT_READ | PROT_WRITE,
1309 MAP_PRIVATE | MAP_ANONYMOUS,
1311 #elif HAVE_VIRTUALALLOC
1312 c->lumMmxextFilterCode = VirtualAlloc(NULL,
1313 c->lumMmxextFilterCodeSize,
1315 PAGE_EXECUTE_READWRITE);
1316 c->chrMmxextFilterCode = VirtualAlloc(NULL,
1317 c->chrMmxextFilterCodeSize,
1319 PAGE_EXECUTE_READWRITE);
1321 c->lumMmxextFilterCode = av_malloc(c->lumMmxextFilterCodeSize);
1322 c->chrMmxextFilterCode = av_malloc(c->chrMmxextFilterCodeSize);
1325 #ifdef MAP_ANONYMOUS
1326 if (c->lumMmxextFilterCode == MAP_FAILED || c->chrMmxextFilterCode == MAP_FAILED)
1328 if (!c->lumMmxextFilterCode || !c->chrMmxextFilterCode)
1331 av_log(c, AV_LOG_ERROR, "Failed to allocate MMX2FilterCode\n");
1332 return AVERROR(ENOMEM);
1335 FF_ALLOCZ_OR_GOTO(c, c->hLumFilter, (dstW / 8 + 8) * sizeof(int16_t), fail);
1336 FF_ALLOCZ_OR_GOTO(c, c->hChrFilter, (c->chrDstW / 4 + 8) * sizeof(int16_t), fail);
1337 FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW / 2 / 8 + 8) * sizeof(int32_t), fail);
1338 FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);
1340 init_hscaler_mmxext( dstW, c->lumXInc, c->lumMmxextFilterCode,
1341 c->hLumFilter, (uint32_t*)c->hLumFilterPos, 8);
1342 init_hscaler_mmxext(c->chrDstW, c->chrXInc, c->chrMmxextFilterCode,
1343 c->hChrFilter, (uint32_t*)c->hChrFilterPos, 4);
1346 mprotect(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize, PROT_EXEC | PROT_READ);
1347 mprotect(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize, PROT_EXEC | PROT_READ);
1350 #endif /* HAVE_MMXEXT_INLINE */
1352 const int filterAlign =
1353 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :
1354 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
1357 if (initFilter(&c->hLumFilter, &c->hLumFilterPos,
1358 &c->hLumFilterSize, c->lumXInc,
1359 srcW, dstW, filterAlign, 1 << 14,
1360 (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
1361 cpu_flags, srcFilter->lumH, dstFilter->lumH,
1364 if (initFilter(&c->hChrFilter, &c->hChrFilterPos,
1365 &c->hChrFilterSize, c->chrXInc,
1366 c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,
1367 (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
1368 cpu_flags, srcFilter->chrH, dstFilter->chrH,
1372 } // initialize horizontal stuff
1374 /* precalculate vertical scaler filter coefficients */
1376 const int filterAlign =
1377 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 2 :
1378 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
1381 if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,
1382 c->lumYInc, srcH, dstH, filterAlign, (1 << 12),
1383 (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
1384 cpu_flags, srcFilter->lumV, dstFilter->lumV,
1387 if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,
1388 c->chrYInc, c->chrSrcH, c->chrDstH,
1389 filterAlign, (1 << 12),
1390 (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
1391 cpu_flags, srcFilter->chrV, dstFilter->chrV,
1396 FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH, fail);
1397 FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);
1399 for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {
1401 short *p = (short *)&c->vYCoeffsBank[i];
1402 for (j = 0; j < 8; j++)
1403 p[j] = c->vLumFilter[i];
1406 for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {
1408 short *p = (short *)&c->vCCoeffsBank[i];
1409 for (j = 0; j < 8; j++)
1410 p[j] = c->vChrFilter[i];
1415 // calculate buffer sizes so that they won't run out while handling these damn slices
1416 c->vLumBufSize = c->vLumFilterSize;
1417 c->vChrBufSize = c->vChrFilterSize;
1418 for (i = 0; i < dstH; i++) {
1419 int chrI = (int64_t)i * c->chrDstH / dstH;
1420 int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,
1421 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)
1422 << c->chrSrcVSubSample));
1424 nextSlice >>= c->chrSrcVSubSample;
1425 nextSlice <<= c->chrSrcVSubSample;
1426 if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)
1427 c->vLumBufSize = nextSlice - c->vLumFilterPos[i];
1428 if (c->vChrFilterPos[chrI] + c->vChrBufSize <
1429 (nextSlice >> c->chrSrcVSubSample))
1430 c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -
1431 c->vChrFilterPos[chrI];
1434 for (i = 0; i < 4; i++)
1435 FF_ALLOCZ_OR_GOTO(c, c->dither_error[i], (c->dstW+2) * sizeof(int), fail);
1437 /* Allocate pixbufs (we use dynamic allocation because otherwise we would
1438 * need to allocate several megabytes to handle all possible cases) */
1439 FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);
1440 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
1441 FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
1442 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
1443 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);
1444 /* Note we need at least one pixel more at the end because of the MMX code
1445 * (just in case someone wants to replace the 4000/8000). */
1446 /* align at 16 bytes for AltiVec */
1447 for (i = 0; i < c->vLumBufSize; i++) {
1448 FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],
1449 dst_stride + 16, fail);
1450 c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];
1452 // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)
1453 c->uv_off = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);
1454 c->uv_offx2 = dst_stride + 16;
1455 for (i = 0; i < c->vChrBufSize; i++) {
1456 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],
1457 dst_stride * 2 + 32, fail);
1458 c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];
1459 c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]
1460 = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;
1462 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
1463 for (i = 0; i < c->vLumBufSize; i++) {
1464 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i + c->vLumBufSize],
1465 dst_stride + 16, fail);
1466 c->alpPixBuf[i] = c->alpPixBuf[i + c->vLumBufSize];
1469 // try to avoid drawing green stuff between the right end and the stride end
1470 for (i = 0; i < c->vChrBufSize; i++)
1471 if(desc_dst->comp[0].depth_minus1 == 15){
1472 av_assert0(c->dstBpc > 14);
1473 for(j=0; j<dst_stride/2+1; j++)
1474 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;
1476 for(j=0; j<dst_stride+1; j++)
1477 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;
1479 av_assert0(c->chrDstH <= dstH);
1481 if (flags & SWS_PRINT_INFO) {
1482 if (flags & SWS_FAST_BILINEAR)
1483 av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
1484 else if (flags & SWS_BILINEAR)
1485 av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
1486 else if (flags & SWS_BICUBIC)
1487 av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
1488 else if (flags & SWS_X)
1489 av_log(c, AV_LOG_INFO, "Experimental scaler, ");
1490 else if (flags & SWS_POINT)
1491 av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
1492 else if (flags & SWS_AREA)
1493 av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
1494 else if (flags & SWS_BICUBLIN)
1495 av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
1496 else if (flags & SWS_GAUSS)
1497 av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
1498 else if (flags & SWS_SINC)
1499 av_log(c, AV_LOG_INFO, "Sinc scaler, ");
1500 else if (flags & SWS_LANCZOS)
1501 av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
1502 else if (flags & SWS_SPLINE)
1503 av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
1505 av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
1507 av_log(c, AV_LOG_INFO, "from %s to %s%s ",
1508 av_get_pix_fmt_name(srcFormat),
1510 dstFormat == AV_PIX_FMT_BGR555 || dstFormat == AV_PIX_FMT_BGR565 ||
1511 dstFormat == AV_PIX_FMT_RGB444BE || dstFormat == AV_PIX_FMT_RGB444LE ||
1512 dstFormat == AV_PIX_FMT_BGR444BE || dstFormat == AV_PIX_FMT_BGR444LE ?
1517 av_get_pix_fmt_name(dstFormat));
1519 if (INLINE_MMXEXT(cpu_flags))
1520 av_log(c, AV_LOG_INFO, "using MMXEXT\n");
1521 else if (INLINE_AMD3DNOW(cpu_flags))
1522 av_log(c, AV_LOG_INFO, "using 3DNOW\n");
1523 else if (INLINE_MMX(cpu_flags))
1524 av_log(c, AV_LOG_INFO, "using MMX\n");
1525 else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC)
1526 av_log(c, AV_LOG_INFO, "using AltiVec\n");
1528 av_log(c, AV_LOG_INFO, "using C\n");
1530 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
1531 av_log(c, AV_LOG_DEBUG,
1532 "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1533 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
1534 av_log(c, AV_LOG_DEBUG,
1535 "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1536 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,
1537 c->chrXInc, c->chrYInc);
1540 c->swScale = ff_getSwsFunc(c);
1542 fail: // FIXME replace things by appropriate error codes
1546 #if FF_API_SWS_GETCONTEXT
1547 SwsContext *sws_getContext(int srcW, int srcH, enum AVPixelFormat srcFormat,
1548 int dstW, int dstH, enum AVPixelFormat dstFormat,
1549 int flags, SwsFilter *srcFilter,
1550 SwsFilter *dstFilter, const double *param)
1554 if (!(c = sws_alloc_context()))
1562 c->srcRange = handle_jpeg(&srcFormat);
1563 c->dstRange = handle_jpeg(&dstFormat);
1564 c->src0Alpha = handle_0alpha(&srcFormat);
1565 c->dst0Alpha = handle_0alpha(&dstFormat);
1566 c->srcXYZ = handle_xyz(&srcFormat);
1567 c->dstXYZ = handle_xyz(&dstFormat);
1568 c->srcFormat = srcFormat;
1569 c->dstFormat = dstFormat;
1572 c->param[0] = param[0];
1573 c->param[1] = param[1];
1575 sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,
1576 ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,
1577 c->dstRange, 0, 1 << 16, 1 << 16);
1579 if (sws_init_context(c, srcFilter, dstFilter) < 0) {
1588 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
1589 float lumaSharpen, float chromaSharpen,
1590 float chromaHShift, float chromaVShift,
1593 SwsFilter *filter = av_malloc(sizeof(SwsFilter));
1597 if (lumaGBlur != 0.0) {
1598 filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);
1599 filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);
1601 filter->lumH = sws_getIdentityVec();
1602 filter->lumV = sws_getIdentityVec();
1605 if (chromaGBlur != 0.0) {
1606 filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);
1607 filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);
1609 filter->chrH = sws_getIdentityVec();
1610 filter->chrV = sws_getIdentityVec();
1613 if (chromaSharpen != 0.0) {
1614 SwsVector *id = sws_getIdentityVec();
1615 sws_scaleVec(filter->chrH, -chromaSharpen);
1616 sws_scaleVec(filter->chrV, -chromaSharpen);
1617 sws_addVec(filter->chrH, id);
1618 sws_addVec(filter->chrV, id);
1622 if (lumaSharpen != 0.0) {
1623 SwsVector *id = sws_getIdentityVec();
1624 sws_scaleVec(filter->lumH, -lumaSharpen);
1625 sws_scaleVec(filter->lumV, -lumaSharpen);
1626 sws_addVec(filter->lumH, id);
1627 sws_addVec(filter->lumV, id);
1631 if (chromaHShift != 0.0)
1632 sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));
1634 if (chromaVShift != 0.0)
1635 sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));
1637 sws_normalizeVec(filter->chrH, 1.0);
1638 sws_normalizeVec(filter->chrV, 1.0);
1639 sws_normalizeVec(filter->lumH, 1.0);
1640 sws_normalizeVec(filter->lumV, 1.0);
1643 sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
1645 sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
1650 SwsVector *sws_allocVec(int length)
1654 if(length <= 0 || length > INT_MAX/ sizeof(double))
1657 vec = av_malloc(sizeof(SwsVector));
1660 vec->length = length;
1661 vec->coeff = av_malloc(sizeof(double) * length);
1667 SwsVector *sws_getGaussianVec(double variance, double quality)
1669 const int length = (int)(variance * quality + 0.5) | 1;
1671 double middle = (length - 1) * 0.5;
1674 if(variance < 0 || quality < 0)
1677 vec = sws_allocVec(length);
1682 for (i = 0; i < length; i++) {
1683 double dist = i - middle;
1684 vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /
1685 sqrt(2 * variance * M_PI);
1688 sws_normalizeVec(vec, 1.0);
1693 SwsVector *sws_getConstVec(double c, int length)
1696 SwsVector *vec = sws_allocVec(length);
1701 for (i = 0; i < length; i++)
1707 SwsVector *sws_getIdentityVec(void)
1709 return sws_getConstVec(1.0, 1);
1712 static double sws_dcVec(SwsVector *a)
1717 for (i = 0; i < a->length; i++)
1723 void sws_scaleVec(SwsVector *a, double scalar)
1727 for (i = 0; i < a->length; i++)
1728 a->coeff[i] *= scalar;
1731 void sws_normalizeVec(SwsVector *a, double height)
1733 sws_scaleVec(a, height / sws_dcVec(a));
1736 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
1738 int length = a->length + b->length - 1;
1740 SwsVector *vec = sws_getConstVec(0.0, length);
1745 for (i = 0; i < a->length; i++) {
1746 for (j = 0; j < b->length; j++) {
1747 vec->coeff[i + j] += a->coeff[i] * b->coeff[j];
1754 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
1756 int length = FFMAX(a->length, b->length);
1758 SwsVector *vec = sws_getConstVec(0.0, length);
1763 for (i = 0; i < a->length; i++)
1764 vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
1765 for (i = 0; i < b->length; i++)
1766 vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] += b->coeff[i];
1771 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
1773 int length = FFMAX(a->length, b->length);
1775 SwsVector *vec = sws_getConstVec(0.0, length);
1780 for (i = 0; i < a->length; i++)
1781 vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
1782 for (i = 0; i < b->length; i++)
1783 vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] -= b->coeff[i];
1788 /* shift left / or right if "shift" is negative */
1789 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
1791 int length = a->length + FFABS(shift) * 2;
1793 SwsVector *vec = sws_getConstVec(0.0, length);
1798 for (i = 0; i < a->length; i++) {
1799 vec->coeff[i + (length - 1) / 2 -
1800 (a->length - 1) / 2 - shift] = a->coeff[i];
1806 void sws_shiftVec(SwsVector *a, int shift)
1808 SwsVector *shifted = sws_getShiftedVec(a, shift);
1810 a->coeff = shifted->coeff;
1811 a->length = shifted->length;
1815 void sws_addVec(SwsVector *a, SwsVector *b)
1817 SwsVector *sum = sws_sumVec(a, b);
1819 a->coeff = sum->coeff;
1820 a->length = sum->length;
1824 void sws_subVec(SwsVector *a, SwsVector *b)
1826 SwsVector *diff = sws_diffVec(a, b);
1828 a->coeff = diff->coeff;
1829 a->length = diff->length;
1833 void sws_convVec(SwsVector *a, SwsVector *b)
1835 SwsVector *conv = sws_getConvVec(a, b);
1837 a->coeff = conv->coeff;
1838 a->length = conv->length;
1842 SwsVector *sws_cloneVec(SwsVector *a)
1845 SwsVector *vec = sws_allocVec(a->length);
1850 for (i = 0; i < a->length; i++)
1851 vec->coeff[i] = a->coeff[i];
1856 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
1863 for (i = 0; i < a->length; i++)
1864 if (a->coeff[i] > max)
1867 for (i = 0; i < a->length; i++)
1868 if (a->coeff[i] < min)
1873 for (i = 0; i < a->length; i++) {
1874 int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);
1875 av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
1877 av_log(log_ctx, log_level, " ");
1878 av_log(log_ctx, log_level, "|\n");
1882 void sws_freeVec(SwsVector *a)
1886 av_freep(&a->coeff);
1891 void sws_freeFilter(SwsFilter *filter)
1897 sws_freeVec(filter->lumH);
1899 sws_freeVec(filter->lumV);
1901 sws_freeVec(filter->chrH);
1903 sws_freeVec(filter->chrV);
1907 void sws_freeContext(SwsContext *c)
1914 for (i = 0; i < c->vLumBufSize; i++)
1915 av_freep(&c->lumPixBuf[i]);
1916 av_freep(&c->lumPixBuf);
1919 if (c->chrUPixBuf) {
1920 for (i = 0; i < c->vChrBufSize; i++)
1921 av_freep(&c->chrUPixBuf[i]);
1922 av_freep(&c->chrUPixBuf);
1923 av_freep(&c->chrVPixBuf);
1926 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1927 for (i = 0; i < c->vLumBufSize; i++)
1928 av_freep(&c->alpPixBuf[i]);
1929 av_freep(&c->alpPixBuf);
1932 for (i = 0; i < 4; i++)
1933 av_freep(&c->dither_error[i]);
1935 av_freep(&c->vLumFilter);
1936 av_freep(&c->vChrFilter);
1937 av_freep(&c->hLumFilter);
1938 av_freep(&c->hChrFilter);
1940 av_freep(&c->vYCoeffsBank);
1941 av_freep(&c->vCCoeffsBank);
1944 av_freep(&c->vLumFilterPos);
1945 av_freep(&c->vChrFilterPos);
1946 av_freep(&c->hLumFilterPos);
1947 av_freep(&c->hChrFilterPos);
1951 if (c->lumMmxextFilterCode)
1952 munmap(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize);
1953 if (c->chrMmxextFilterCode)
1954 munmap(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize);
1955 #elif HAVE_VIRTUALALLOC
1956 if (c->lumMmxextFilterCode)
1957 VirtualFree(c->lumMmxextFilterCode, 0, MEM_RELEASE);
1958 if (c->chrMmxextFilterCode)
1959 VirtualFree(c->chrMmxextFilterCode, 0, MEM_RELEASE);
1961 av_free(c->lumMmxextFilterCode);
1962 av_free(c->chrMmxextFilterCode);
1964 c->lumMmxextFilterCode = NULL;
1965 c->chrMmxextFilterCode = NULL;
1966 #endif /* HAVE_MMX_INLINE */
1968 av_freep(&c->yuvTable);
1969 av_freep(&c->formatConvBuffer);
1974 struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,
1975 int srcH, enum AVPixelFormat srcFormat,
1977 enum AVPixelFormat dstFormat, int flags,
1978 SwsFilter *srcFilter,
1979 SwsFilter *dstFilter,
1980 const double *param)
1982 static const double default_param[2] = { SWS_PARAM_DEFAULT,
1983 SWS_PARAM_DEFAULT };
1986 param = default_param;
1989 (context->srcW != srcW ||
1990 context->srcH != srcH ||
1991 context->srcFormat != srcFormat ||
1992 context->dstW != dstW ||
1993 context->dstH != dstH ||
1994 context->dstFormat != dstFormat ||
1995 context->flags != flags ||
1996 context->param[0] != param[0] ||
1997 context->param[1] != param[1])) {
1998 sws_freeContext(context);
2003 if (!(context = sws_alloc_context()))
2005 context->srcW = srcW;
2006 context->srcH = srcH;
2007 context->srcRange = handle_jpeg(&srcFormat);
2008 context->src0Alpha = handle_0alpha(&srcFormat);
2009 context->srcXYZ = handle_xyz(&srcFormat);
2010 context->srcFormat = srcFormat;
2011 context->dstW = dstW;
2012 context->dstH = dstH;
2013 context->dstRange = handle_jpeg(&dstFormat);
2014 context->dst0Alpha = handle_0alpha(&dstFormat);
2015 context->dstXYZ = handle_xyz(&dstFormat);
2016 context->dstFormat = dstFormat;
2017 context->flags = flags;
2018 context->param[0] = param[0];
2019 context->param[1] = param[1];
2020 sws_setColorspaceDetails(context, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],
2022 ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,
2023 context->dstRange, 0, 1 << 16, 1 << 16);
2024 if (sws_init_context(context, srcFilter, dstFilter) < 0) {
2025 sws_freeContext(context);