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 static void handle_formats(SwsContext *c);
57 unsigned swscale_version(void)
59 av_assert0(LIBSWSCALE_VERSION_MICRO >= 100);
60 return LIBSWSCALE_VERSION_INT;
63 const char *swscale_configuration(void)
65 return FFMPEG_CONFIGURATION;
68 const char *swscale_license(void)
70 #define LICENSE_PREFIX "libswscale license: "
71 return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
74 #define RET 0xC3 // near return opcode for x86
76 typedef struct FormatEntry {
77 uint8_t is_supported_in :1;
78 uint8_t is_supported_out :1;
79 uint8_t is_supported_endianness :1;
82 static const FormatEntry format_entries[AV_PIX_FMT_NB] = {
83 [AV_PIX_FMT_YUV420P] = { 1, 1 },
84 [AV_PIX_FMT_YUYV422] = { 1, 1 },
85 [AV_PIX_FMT_RGB24] = { 1, 1 },
86 [AV_PIX_FMT_BGR24] = { 1, 1 },
87 [AV_PIX_FMT_YUV422P] = { 1, 1 },
88 [AV_PIX_FMT_YUV444P] = { 1, 1 },
89 [AV_PIX_FMT_YUV410P] = { 1, 1 },
90 [AV_PIX_FMT_YUV411P] = { 1, 1 },
91 [AV_PIX_FMT_GRAY8] = { 1, 1 },
92 [AV_PIX_FMT_MONOWHITE] = { 1, 1 },
93 [AV_PIX_FMT_MONOBLACK] = { 1, 1 },
94 [AV_PIX_FMT_PAL8] = { 1, 0 },
95 [AV_PIX_FMT_YUVJ420P] = { 1, 1 },
96 [AV_PIX_FMT_YUVJ411P] = { 1, 1 },
97 [AV_PIX_FMT_YUVJ422P] = { 1, 1 },
98 [AV_PIX_FMT_YUVJ444P] = { 1, 1 },
99 [AV_PIX_FMT_UYVY422] = { 1, 1 },
100 [AV_PIX_FMT_UYYVYY411] = { 0, 0 },
101 [AV_PIX_FMT_BGR8] = { 1, 1 },
102 [AV_PIX_FMT_BGR4] = { 0, 1 },
103 [AV_PIX_FMT_BGR4_BYTE] = { 1, 1 },
104 [AV_PIX_FMT_RGB8] = { 1, 1 },
105 [AV_PIX_FMT_RGB4] = { 0, 1 },
106 [AV_PIX_FMT_RGB4_BYTE] = { 1, 1 },
107 [AV_PIX_FMT_NV12] = { 1, 1 },
108 [AV_PIX_FMT_NV21] = { 1, 1 },
109 [AV_PIX_FMT_ARGB] = { 1, 1 },
110 [AV_PIX_FMT_RGBA] = { 1, 1 },
111 [AV_PIX_FMT_ABGR] = { 1, 1 },
112 [AV_PIX_FMT_BGRA] = { 1, 1 },
113 [AV_PIX_FMT_0RGB] = { 1, 1 },
114 [AV_PIX_FMT_RGB0] = { 1, 1 },
115 [AV_PIX_FMT_0BGR] = { 1, 1 },
116 [AV_PIX_FMT_BGR0] = { 1, 1 },
117 [AV_PIX_FMT_GRAY16BE] = { 1, 1 },
118 [AV_PIX_FMT_GRAY16LE] = { 1, 1 },
119 [AV_PIX_FMT_YUV440P] = { 1, 1 },
120 [AV_PIX_FMT_YUVJ440P] = { 1, 1 },
121 [AV_PIX_FMT_YUVA420P] = { 1, 1 },
122 [AV_PIX_FMT_YUVA422P] = { 1, 1 },
123 [AV_PIX_FMT_YUVA444P] = { 1, 1 },
124 [AV_PIX_FMT_YUVA420P9BE] = { 1, 1 },
125 [AV_PIX_FMT_YUVA420P9LE] = { 1, 1 },
126 [AV_PIX_FMT_YUVA422P9BE] = { 1, 1 },
127 [AV_PIX_FMT_YUVA422P9LE] = { 1, 1 },
128 [AV_PIX_FMT_YUVA444P9BE] = { 1, 1 },
129 [AV_PIX_FMT_YUVA444P9LE] = { 1, 1 },
130 [AV_PIX_FMT_YUVA420P10BE]= { 1, 1 },
131 [AV_PIX_FMT_YUVA420P10LE]= { 1, 1 },
132 [AV_PIX_FMT_YUVA422P10BE]= { 1, 1 },
133 [AV_PIX_FMT_YUVA422P10LE]= { 1, 1 },
134 [AV_PIX_FMT_YUVA444P10BE]= { 1, 1 },
135 [AV_PIX_FMT_YUVA444P10LE]= { 1, 1 },
136 [AV_PIX_FMT_YUVA420P16BE]= { 1, 1 },
137 [AV_PIX_FMT_YUVA420P16LE]= { 1, 1 },
138 [AV_PIX_FMT_YUVA422P16BE]= { 1, 1 },
139 [AV_PIX_FMT_YUVA422P16LE]= { 1, 1 },
140 [AV_PIX_FMT_YUVA444P16BE]= { 1, 1 },
141 [AV_PIX_FMT_YUVA444P16LE]= { 1, 1 },
142 [AV_PIX_FMT_RGB48BE] = { 1, 1 },
143 [AV_PIX_FMT_RGB48LE] = { 1, 1 },
144 [AV_PIX_FMT_RGBA64BE] = { 1, 1 },
145 [AV_PIX_FMT_RGBA64LE] = { 1, 1 },
146 [AV_PIX_FMT_RGB565BE] = { 1, 1 },
147 [AV_PIX_FMT_RGB565LE] = { 1, 1 },
148 [AV_PIX_FMT_RGB555BE] = { 1, 1 },
149 [AV_PIX_FMT_RGB555LE] = { 1, 1 },
150 [AV_PIX_FMT_BGR565BE] = { 1, 1 },
151 [AV_PIX_FMT_BGR565LE] = { 1, 1 },
152 [AV_PIX_FMT_BGR555BE] = { 1, 1 },
153 [AV_PIX_FMT_BGR555LE] = { 1, 1 },
154 [AV_PIX_FMT_YUV420P16LE] = { 1, 1 },
155 [AV_PIX_FMT_YUV420P16BE] = { 1, 1 },
156 [AV_PIX_FMT_YUV422P16LE] = { 1, 1 },
157 [AV_PIX_FMT_YUV422P16BE] = { 1, 1 },
158 [AV_PIX_FMT_YUV444P16LE] = { 1, 1 },
159 [AV_PIX_FMT_YUV444P16BE] = { 1, 1 },
160 [AV_PIX_FMT_RGB444LE] = { 1, 1 },
161 [AV_PIX_FMT_RGB444BE] = { 1, 1 },
162 [AV_PIX_FMT_BGR444LE] = { 1, 1 },
163 [AV_PIX_FMT_BGR444BE] = { 1, 1 },
164 [AV_PIX_FMT_Y400A] = { 1, 0 },
165 [AV_PIX_FMT_BGR48BE] = { 1, 1 },
166 [AV_PIX_FMT_BGR48LE] = { 1, 1 },
167 [AV_PIX_FMT_BGRA64BE] = { 0, 0 },
168 [AV_PIX_FMT_BGRA64LE] = { 0, 0 },
169 [AV_PIX_FMT_YUV420P9BE] = { 1, 1 },
170 [AV_PIX_FMT_YUV420P9LE] = { 1, 1 },
171 [AV_PIX_FMT_YUV420P10BE] = { 1, 1 },
172 [AV_PIX_FMT_YUV420P10LE] = { 1, 1 },
173 [AV_PIX_FMT_YUV420P12BE] = { 1, 1 },
174 [AV_PIX_FMT_YUV420P12LE] = { 1, 1 },
175 [AV_PIX_FMT_YUV420P14BE] = { 1, 1 },
176 [AV_PIX_FMT_YUV420P14LE] = { 1, 1 },
177 [AV_PIX_FMT_YUV422P9BE] = { 1, 1 },
178 [AV_PIX_FMT_YUV422P9LE] = { 1, 1 },
179 [AV_PIX_FMT_YUV422P10BE] = { 1, 1 },
180 [AV_PIX_FMT_YUV422P10LE] = { 1, 1 },
181 [AV_PIX_FMT_YUV422P12BE] = { 1, 1 },
182 [AV_PIX_FMT_YUV422P12LE] = { 1, 1 },
183 [AV_PIX_FMT_YUV422P14BE] = { 1, 1 },
184 [AV_PIX_FMT_YUV422P14LE] = { 1, 1 },
185 [AV_PIX_FMT_YUV444P9BE] = { 1, 1 },
186 [AV_PIX_FMT_YUV444P9LE] = { 1, 1 },
187 [AV_PIX_FMT_YUV444P10BE] = { 1, 1 },
188 [AV_PIX_FMT_YUV444P10LE] = { 1, 1 },
189 [AV_PIX_FMT_YUV444P12BE] = { 1, 1 },
190 [AV_PIX_FMT_YUV444P12LE] = { 1, 1 },
191 [AV_PIX_FMT_YUV444P14BE] = { 1, 1 },
192 [AV_PIX_FMT_YUV444P14LE] = { 1, 1 },
193 [AV_PIX_FMT_GBRP] = { 1, 1 },
194 [AV_PIX_FMT_GBRP9LE] = { 1, 1 },
195 [AV_PIX_FMT_GBRP9BE] = { 1, 1 },
196 [AV_PIX_FMT_GBRP10LE] = { 1, 1 },
197 [AV_PIX_FMT_GBRP10BE] = { 1, 1 },
198 [AV_PIX_FMT_GBRP12LE] = { 1, 1 },
199 [AV_PIX_FMT_GBRP12BE] = { 1, 1 },
200 [AV_PIX_FMT_GBRP14LE] = { 1, 1 },
201 [AV_PIX_FMT_GBRP14BE] = { 1, 1 },
202 [AV_PIX_FMT_GBRP16LE] = { 1, 0 },
203 [AV_PIX_FMT_GBRP16BE] = { 1, 0 },
204 [AV_PIX_FMT_XYZ12BE] = { 1, 1, 1 },
205 [AV_PIX_FMT_XYZ12LE] = { 1, 1, 1 },
206 [AV_PIX_FMT_GBRAP] = { 1, 1 },
207 [AV_PIX_FMT_GBRAP16LE] = { 1, 0 },
208 [AV_PIX_FMT_GBRAP16BE] = { 1, 0 },
211 int sws_isSupportedInput(enum AVPixelFormat pix_fmt)
213 return (unsigned)pix_fmt < AV_PIX_FMT_NB ?
214 format_entries[pix_fmt].is_supported_in : 0;
217 int sws_isSupportedOutput(enum AVPixelFormat pix_fmt)
219 return (unsigned)pix_fmt < AV_PIX_FMT_NB ?
220 format_entries[pix_fmt].is_supported_out : 0;
223 int sws_isSupportedEndiannessConversion(enum AVPixelFormat pix_fmt)
225 return (unsigned)pix_fmt < AV_PIX_FMT_NB ?
226 format_entries[pix_fmt].is_supported_endianness : 0;
229 #if FF_API_SWS_FORMAT_NAME
230 const char *sws_format_name(enum AVPixelFormat format)
232 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(format);
236 return "Unknown format";
240 static double getSplineCoeff(double a, double b, double c, double d,
244 return ((d * dist + c) * dist + b) * dist + a;
246 return getSplineCoeff(0.0,
247 b + 2.0 * c + 3.0 * d,
249 -b - 3.0 * c - 6.0 * d,
253 static av_cold int get_local_pos(SwsContext *s, int chr_subsample, int pos, int dir)
256 pos = (128 << chr_subsample) - 128;
258 pos += 128; // relative to ideal left edge
259 return pos >> chr_subsample;
262 static av_cold int initFilter(int16_t **outFilter, int32_t **filterPos,
263 int *outFilterSize, int xInc, int srcW,
264 int dstW, int filterAlign, int one,
265 int flags, int cpu_flags,
266 SwsVector *srcFilter, SwsVector *dstFilter,
267 double param[2], int srcPos, int dstPos)
273 int64_t *filter = NULL;
274 int64_t *filter2 = NULL;
275 const int64_t fone = 1LL << (54 - FFMIN(av_log2(srcW/dstW), 8));
278 emms_c(); // FIXME should not be required but IS (even for non-MMX versions)
280 // NOTE: the +3 is for the MMX(+1) / SSE(+3) scaler which reads over the end
281 FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW + 3) * sizeof(**filterPos), fail);
283 if (FFABS(xInc - 0x10000) < 10 && srcPos == dstPos) { // unscaled
286 FF_ALLOCZ_OR_GOTO(NULL, filter,
287 dstW * sizeof(*filter) * filterSize, fail);
289 for (i = 0; i < dstW; i++) {
290 filter[i * filterSize] = fone;
293 } else if (flags & SWS_POINT) { // lame looking point sampling mode
297 FF_ALLOC_OR_GOTO(NULL, filter,
298 dstW * sizeof(*filter) * filterSize, fail);
300 xDstInSrc = ((dstPos*(int64_t)xInc)>>8) - ((srcPos*0x8000LL)>>7);
301 for (i = 0; i < dstW; i++) {
302 int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;
304 (*filterPos)[i] = xx;
308 } else if ((xInc <= (1 << 16) && (flags & SWS_AREA)) ||
309 (flags & SWS_FAST_BILINEAR)) { // bilinear upscale
313 FF_ALLOC_OR_GOTO(NULL, filter,
314 dstW * sizeof(*filter) * filterSize, fail);
316 xDstInSrc = ((dstPos*(int64_t)xInc)>>8) - ((srcPos*0x8000LL)>>7);
317 for (i = 0; i < dstW; i++) {
318 int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;
321 (*filterPos)[i] = xx;
322 // bilinear upscale / linear interpolate / area averaging
323 for (j = 0; j < filterSize; j++) {
324 int64_t coeff= fone - FFABS(((int64_t)xx<<16) - xDstInSrc)*(fone>>16);
327 filter[i * filterSize + j] = coeff;
336 if (flags & SWS_BICUBIC)
338 else if (flags & SWS_X)
340 else if (flags & SWS_AREA)
341 sizeFactor = 1; // downscale only, for upscale it is bilinear
342 else if (flags & SWS_GAUSS)
343 sizeFactor = 8; // infinite ;)
344 else if (flags & SWS_LANCZOS)
345 sizeFactor = param[0] != SWS_PARAM_DEFAULT ? ceil(2 * param[0]) : 6;
346 else if (flags & SWS_SINC)
347 sizeFactor = 20; // infinite ;)
348 else if (flags & SWS_SPLINE)
349 sizeFactor = 20; // infinite ;)
350 else if (flags & SWS_BILINEAR)
357 filterSize = 1 + sizeFactor; // upscale
359 filterSize = 1 + (sizeFactor * srcW + dstW - 1) / dstW;
361 filterSize = FFMIN(filterSize, srcW - 2);
362 filterSize = FFMAX(filterSize, 1);
364 FF_ALLOC_OR_GOTO(NULL, filter,
365 dstW * sizeof(*filter) * filterSize, fail);
367 xDstInSrc = ((dstPos*(int64_t)xInc)>>7) - ((srcPos*0x10000LL)>>7);
368 for (i = 0; i < dstW; i++) {
369 int xx = (xDstInSrc - ((filterSize - 2) << 16)) / (1 << 17);
371 (*filterPos)[i] = xx;
372 for (j = 0; j < filterSize; j++) {
373 int64_t d = (FFABS(((int64_t)xx << 17) - xDstInSrc)) << 13;
379 floatd = d * (1.0 / (1 << 30));
381 if (flags & SWS_BICUBIC) {
382 int64_t B = (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1 << 24);
383 int64_t C = (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1 << 24);
385 if (d >= 1LL << 31) {
388 int64_t dd = (d * d) >> 30;
389 int64_t ddd = (dd * d) >> 30;
392 coeff = (12 * (1 << 24) - 9 * B - 6 * C) * ddd +
393 (-18 * (1 << 24) + 12 * B + 6 * C) * dd +
394 (6 * (1 << 24) - 2 * B) * (1 << 30);
396 coeff = (-B - 6 * C) * ddd +
397 (6 * B + 30 * C) * dd +
398 (-12 * B - 48 * C) * d +
399 (8 * B + 24 * C) * (1 << 30);
401 coeff /= (1LL<<54)/fone;
404 else if (flags & SWS_X) {
405 double p = param ? param * 0.01 : 0.3;
406 coeff = d ? sin(d * M_PI) / (d * M_PI) : 1.0;
407 coeff *= pow(2.0, -p * d * d);
410 else if (flags & SWS_X) {
411 double A = param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
415 c = cos(floatd * M_PI);
422 coeff = (c * 0.5 + 0.5) * fone;
423 } else if (flags & SWS_AREA) {
424 int64_t d2 = d - (1 << 29);
425 if (d2 * xInc < -(1LL << (29 + 16)))
426 coeff = 1.0 * (1LL << (30 + 16));
427 else if (d2 * xInc < (1LL << (29 + 16)))
428 coeff = -d2 * xInc + (1LL << (29 + 16));
431 coeff *= fone >> (30 + 16);
432 } else if (flags & SWS_GAUSS) {
433 double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
434 coeff = (pow(2.0, -p * floatd * floatd)) * fone;
435 } else if (flags & SWS_SINC) {
436 coeff = (d ? sin(floatd * M_PI) / (floatd * M_PI) : 1.0) * fone;
437 } else if (flags & SWS_LANCZOS) {
438 double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
439 coeff = (d ? sin(floatd * M_PI) * sin(floatd * M_PI / p) /
440 (floatd * floatd * M_PI * M_PI / p) : 1.0) * fone;
443 } else if (flags & SWS_BILINEAR) {
444 coeff = (1 << 30) - d;
448 } else if (flags & SWS_SPLINE) {
449 double p = -2.196152422706632;
450 coeff = getSplineCoeff(1.0, 0.0, p, -p - 1.0, floatd) * fone;
455 filter[i * filterSize + j] = coeff;
458 xDstInSrc += 2 * xInc;
462 /* apply src & dst Filter to filter -> filter2
465 av_assert0(filterSize > 0);
466 filter2Size = filterSize;
468 filter2Size += srcFilter->length - 1;
470 filter2Size += dstFilter->length - 1;
471 av_assert0(filter2Size > 0);
472 FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size * dstW * sizeof(*filter2), fail);
474 for (i = 0; i < dstW; i++) {
478 for (k = 0; k < srcFilter->length; k++) {
479 for (j = 0; j < filterSize; j++)
480 filter2[i * filter2Size + k + j] +=
481 srcFilter->coeff[k] * filter[i * filterSize + j];
484 for (j = 0; j < filterSize; j++)
485 filter2[i * filter2Size + j] = filter[i * filterSize + j];
489 (*filterPos)[i] += (filterSize - 1) / 2 - (filter2Size - 1) / 2;
493 /* try to reduce the filter-size (step1 find size and shift left) */
494 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
496 for (i = dstW - 1; i >= 0; i--) {
497 int min = filter2Size;
499 int64_t cutOff = 0.0;
501 /* get rid of near zero elements on the left by shifting left */
502 for (j = 0; j < filter2Size; j++) {
504 cutOff += FFABS(filter2[i * filter2Size]);
506 if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)
509 /* preserve monotonicity because the core can't handle the
510 * filter otherwise */
511 if (i < dstW - 1 && (*filterPos)[i] >= (*filterPos)[i + 1])
514 // move filter coefficients left
515 for (k = 1; k < filter2Size; k++)
516 filter2[i * filter2Size + k - 1] = filter2[i * filter2Size + k];
517 filter2[i * filter2Size + k - 1] = 0;
522 /* count near zeros on the right */
523 for (j = filter2Size - 1; j > 0; j--) {
524 cutOff += FFABS(filter2[i * filter2Size + j]);
526 if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)
531 if (min > minFilterSize)
535 if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) {
536 // we can handle the special case 4, so we don't want to go the full 8
537 if (minFilterSize < 5)
540 /* We really don't want to waste our time doing useless computation, so
541 * fall back on the scalar C code for very small filters.
542 * Vectorizing is worth it only if you have a decent-sized vector. */
543 if (minFilterSize < 3)
547 if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
548 // special case for unscaled vertical filtering
549 if (minFilterSize == 1 && filterAlign == 2)
553 av_assert0(minFilterSize > 0);
554 filterSize = (minFilterSize + (filterAlign - 1)) & (~(filterAlign - 1));
555 av_assert0(filterSize > 0);
556 filter = av_malloc(filterSize * dstW * sizeof(*filter));
557 if (filterSize >= MAX_FILTER_SIZE * 16 /
558 ((flags & SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter) {
559 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);
562 *outFilterSize = filterSize;
564 if (flags & SWS_PRINT_INFO)
565 av_log(NULL, AV_LOG_VERBOSE,
566 "SwScaler: reducing / aligning filtersize %d -> %d\n",
567 filter2Size, filterSize);
568 /* try to reduce the filter-size (step2 reduce it) */
569 for (i = 0; i < dstW; i++) {
572 for (j = 0; j < filterSize; j++) {
573 if (j >= filter2Size)
574 filter[i * filterSize + j] = 0;
576 filter[i * filterSize + j] = filter2[i * filter2Size + j];
577 if ((flags & SWS_BITEXACT) && j >= minFilterSize)
578 filter[i * filterSize + j] = 0;
582 // FIXME try to align filterPos if possible
585 for (i = 0; i < dstW; i++) {
587 if ((*filterPos)[i] < 0) {
588 // move filter coefficients left to compensate for filterPos
589 for (j = 1; j < filterSize; j++) {
590 int left = FFMAX(j + (*filterPos)[i], 0);
591 filter[i * filterSize + left] += filter[i * filterSize + j];
592 filter[i * filterSize + j] = 0;
597 if ((*filterPos)[i] + filterSize > srcW) {
598 int shift = (*filterPos)[i] + filterSize - srcW;
599 // move filter coefficients right to compensate for filterPos
600 for (j = filterSize - 2; j >= 0; j--) {
601 int right = FFMIN(j + shift, filterSize - 1);
602 filter[i * filterSize + right] += filter[i * filterSize + j];
603 filter[i * filterSize + j] = 0;
605 (*filterPos)[i]= srcW - filterSize;
609 // Note the +1 is for the MMX scaler which reads over the end
610 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
611 FF_ALLOCZ_OR_GOTO(NULL, *outFilter,
612 *outFilterSize * (dstW + 3) * sizeof(int16_t), fail);
614 /* normalize & store in outFilter */
615 for (i = 0; i < dstW; i++) {
620 for (j = 0; j < filterSize; j++) {
621 sum += filter[i * filterSize + j];
623 sum = (sum + one / 2) / one;
624 for (j = 0; j < *outFilterSize; j++) {
625 int64_t v = filter[i * filterSize + j] + error;
626 int intV = ROUNDED_DIV(v, sum);
627 (*outFilter)[i * (*outFilterSize) + j] = intV;
628 error = v - intV * sum;
632 (*filterPos)[dstW + 0] =
633 (*filterPos)[dstW + 1] =
634 (*filterPos)[dstW + 2] = (*filterPos)[dstW - 1]; /* the MMX/SSE scaler will
635 * read over the end */
636 for (i = 0; i < *outFilterSize; i++) {
637 int k = (dstW - 1) * (*outFilterSize) + i;
638 (*outFilter)[k + 1 * (*outFilterSize)] =
639 (*outFilter)[k + 2 * (*outFilterSize)] =
640 (*outFilter)[k + 3 * (*outFilterSize)] = (*outFilter)[k];
647 av_log(NULL, AV_LOG_ERROR, "sws: initFilter failed\n");
653 #if HAVE_MMXEXT_INLINE
654 static av_cold int init_hscaler_mmxext(int dstW, int xInc, uint8_t *filterCode,
655 int16_t *filter, int32_t *filterPos,
659 x86_reg imm8OfPShufW1A;
660 x86_reg imm8OfPShufW2A;
661 x86_reg fragmentLengthA;
663 x86_reg imm8OfPShufW1B;
664 x86_reg imm8OfPShufW2B;
665 x86_reg fragmentLengthB;
670 // create an optimized horizontal scaling routine
671 /* This scaler is made of runtime-generated MMXEXT code using specially tuned
672 * pshufw instructions. For every four output pixels, if four input pixels
673 * are enough for the fast bilinear scaling, then a chunk of fragmentB is
674 * used. If five input pixels are needed, then a chunk of fragmentA is used.
683 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
684 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
685 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
686 "punpcklbw %%mm7, %%mm1 \n\t"
687 "punpcklbw %%mm7, %%mm0 \n\t"
688 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
690 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
692 "psubw %%mm1, %%mm0 \n\t"
693 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
694 "pmullw %%mm3, %%mm0 \n\t"
695 "psllw $7, %%mm1 \n\t"
696 "paddw %%mm1, %%mm0 \n\t"
698 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
700 "add $8, %%"REG_a" \n\t"
704 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
705 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
706 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
711 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
715 : "=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
716 "=r" (fragmentLengthA)
723 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
724 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
725 "punpcklbw %%mm7, %%mm0 \n\t"
726 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
728 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
730 "psubw %%mm1, %%mm0 \n\t"
731 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
732 "pmullw %%mm3, %%mm0 \n\t"
733 "psllw $7, %%mm1 \n\t"
734 "paddw %%mm1, %%mm0 \n\t"
736 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
738 "add $8, %%"REG_a" \n\t"
742 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
743 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
744 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
749 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
753 : "=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
754 "=r" (fragmentLengthB)
757 xpos = 0; // lumXInc/2 - 0x8000; // difference between pixel centers
760 for (i = 0; i < dstW / numSplits; i++) {
765 int b = ((xpos + xInc) >> 16) - xx;
766 int c = ((xpos + xInc * 2) >> 16) - xx;
767 int d = ((xpos + xInc * 3) >> 16) - xx;
768 int inc = (d + 1 < 4);
769 uint8_t *fragment = (d + 1 < 4) ? fragmentB : fragmentA;
770 x86_reg imm8OfPShufW1 = (d + 1 < 4) ? imm8OfPShufW1B : imm8OfPShufW1A;
771 x86_reg imm8OfPShufW2 = (d + 1 < 4) ? imm8OfPShufW2B : imm8OfPShufW2A;
772 x86_reg fragmentLength = (d + 1 < 4) ? fragmentLengthB : fragmentLengthA;
773 int maxShift = 3 - (d + inc);
777 filter[i] = ((xpos & 0xFFFF) ^ 0xFFFF) >> 9;
778 filter[i + 1] = (((xpos + xInc) & 0xFFFF) ^ 0xFFFF) >> 9;
779 filter[i + 2] = (((xpos + xInc * 2) & 0xFFFF) ^ 0xFFFF) >> 9;
780 filter[i + 3] = (((xpos + xInc * 3) & 0xFFFF) ^ 0xFFFF) >> 9;
781 filterPos[i / 2] = xx;
783 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
785 filterCode[fragmentPos + imm8OfPShufW1] = (a + inc) |
789 filterCode[fragmentPos + imm8OfPShufW2] = a | (b << 2) |
793 if (i + 4 - inc >= dstW)
794 shift = maxShift; // avoid overread
795 else if ((filterPos[i / 2] & 3) <= maxShift)
796 shift = filterPos[i / 2] & 3; // align
798 if (shift && i >= shift) {
799 filterCode[fragmentPos + imm8OfPShufW1] += 0x55 * shift;
800 filterCode[fragmentPos + imm8OfPShufW2] += 0x55 * shift;
801 filterPos[i / 2] -= shift;
805 fragmentPos += fragmentLength;
808 filterCode[fragmentPos] = RET;
813 filterPos[((i / 2) + 1) & (~1)] = xpos >> 16; // needed to jump to the next part
815 return fragmentPos + 1;
817 #endif /* HAVE_MMXEXT_INLINE */
819 static void getSubSampleFactors(int *h, int *v, enum AVPixelFormat format)
821 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(format);
822 *h = desc->log2_chroma_w;
823 *v = desc->log2_chroma_h;
826 static void fill_rgb2yuv_table(SwsContext *c, const int table[4], int dstRange)
828 int64_t W, V, Z, Cy, Cu, Cv;
829 int64_t vr = table[0];
830 int64_t ub = table[1];
831 int64_t ug = -table[2];
832 int64_t vg = -table[3];
835 uint8_t *p = (uint8_t*)c->input_rgb2yuv_table;
837 static const int8_t map[] = {
838 BY_IDX, GY_IDX, -1 , BY_IDX, BY_IDX, GY_IDX, -1 , BY_IDX,
839 RY_IDX, -1 , GY_IDX, RY_IDX, RY_IDX, -1 , GY_IDX, RY_IDX,
840 RY_IDX, GY_IDX, -1 , RY_IDX, RY_IDX, GY_IDX, -1 , RY_IDX,
841 BY_IDX, -1 , GY_IDX, BY_IDX, BY_IDX, -1 , GY_IDX, BY_IDX,
842 BU_IDX, GU_IDX, -1 , BU_IDX, BU_IDX, GU_IDX, -1 , BU_IDX,
843 RU_IDX, -1 , GU_IDX, RU_IDX, RU_IDX, -1 , GU_IDX, RU_IDX,
844 RU_IDX, GU_IDX, -1 , RU_IDX, RU_IDX, GU_IDX, -1 , RU_IDX,
845 BU_IDX, -1 , GU_IDX, BU_IDX, BU_IDX, -1 , GU_IDX, BU_IDX,
846 BV_IDX, GV_IDX, -1 , BV_IDX, BV_IDX, GV_IDX, -1 , BV_IDX,
847 RV_IDX, -1 , GV_IDX, RV_IDX, RV_IDX, -1 , GV_IDX, RV_IDX,
848 RV_IDX, GV_IDX, -1 , RV_IDX, RV_IDX, GV_IDX, -1 , RV_IDX,
849 BV_IDX, -1 , GV_IDX, BV_IDX, BV_IDX, -1 , GV_IDX, BV_IDX,
850 RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX,
851 BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX,
852 GY_IDX, -1 , GY_IDX, -1 , GY_IDX, -1 , GY_IDX, -1 ,
853 -1 , GY_IDX, -1 , GY_IDX, -1 , GY_IDX, -1 , GY_IDX,
854 RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX,
855 BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX,
856 GU_IDX, -1 , GU_IDX, -1 , GU_IDX, -1 , GU_IDX, -1 ,
857 -1 , GU_IDX, -1 , GU_IDX, -1 , GU_IDX, -1 , GU_IDX,
858 RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX,
859 BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX,
860 GV_IDX, -1 , GV_IDX, -1 , GV_IDX, -1 , GV_IDX, -1 ,
861 -1 , GV_IDX, -1 , GV_IDX, -1 , GV_IDX, -1 , GV_IDX, //23
862 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //24
863 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //25
864 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //26
865 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //27
866 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //28
867 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //29
868 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //30
869 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //31
870 BY_IDX, GY_IDX, RY_IDX, -1 , -1 , -1 , -1 , -1 , //32
871 BU_IDX, GU_IDX, RU_IDX, -1 , -1 , -1 , -1 , -1 , //33
872 BV_IDX, GV_IDX, RV_IDX, -1 , -1 , -1 , -1 , -1 , //34
875 dstRange = 0; //FIXME range = 1 is handled elsewhere
885 W = ROUNDED_DIV(ONE*ONE*ug, ub);
886 V = ROUNDED_DIV(ONE*ONE*vg, vr);
889 Cy = ROUNDED_DIV(cy*Z, ONE);
890 Cu = ROUNDED_DIV(ub*Z, ONE);
891 Cv = ROUNDED_DIV(vr*Z, ONE);
893 c->input_rgb2yuv_table[RY_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*V , Cy);
894 c->input_rgb2yuv_table[GY_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE , Cy);
895 c->input_rgb2yuv_table[BY_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*W , Cy);
897 c->input_rgb2yuv_table[RU_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*V , Cu);
898 c->input_rgb2yuv_table[GU_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE , Cu);
899 c->input_rgb2yuv_table[BU_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*(Z+W) , Cu);
901 c->input_rgb2yuv_table[RV_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*(V+Z) , Cv);
902 c->input_rgb2yuv_table[GV_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE , Cv);
903 c->input_rgb2yuv_table[BV_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*W , Cv);
905 if(/*!dstRange && */!memcmp(table, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], sizeof(ff_yuv2rgb_coeffs[SWS_CS_DEFAULT]))) {
906 c->input_rgb2yuv_table[BY_IDX] = ((int)(0.114 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
907 c->input_rgb2yuv_table[BV_IDX] = (-(int)(0.081 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
908 c->input_rgb2yuv_table[BU_IDX] = ((int)(0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
909 c->input_rgb2yuv_table[GY_IDX] = ((int)(0.587 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
910 c->input_rgb2yuv_table[GV_IDX] = (-(int)(0.419 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
911 c->input_rgb2yuv_table[GU_IDX] = (-(int)(0.331 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
912 c->input_rgb2yuv_table[RY_IDX] = ((int)(0.299 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
913 c->input_rgb2yuv_table[RV_IDX] = ((int)(0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
914 c->input_rgb2yuv_table[RU_IDX] = (-(int)(0.169 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
916 for(i=0; i<FF_ARRAY_ELEMS(map); i++)
917 AV_WL16(p + 16*4 + 2*i, map[i] >= 0 ? c->input_rgb2yuv_table[map[i]] : 0);
920 static void fill_xyztables(struct SwsContext *c)
923 double xyzgamma = XYZ_GAMMA;
924 double rgbgamma = 1.0 / RGB_GAMMA;
925 double xyzgammainv = 1.0 / XYZ_GAMMA;
926 double rgbgammainv = RGB_GAMMA;
927 static const int16_t xyz2rgb_matrix[3][4] = {
928 {13270, -6295, -2041},
930 { 228, -835, 4329} };
931 static const int16_t rgb2xyz_matrix[3][4] = {
935 static int16_t xyzgamma_tab[4096], rgbgamma_tab[4096], xyzgammainv_tab[4096], rgbgammainv_tab[4096];
937 memcpy(c->xyz2rgb_matrix, xyz2rgb_matrix, sizeof(c->xyz2rgb_matrix));
938 memcpy(c->rgb2xyz_matrix, rgb2xyz_matrix, sizeof(c->rgb2xyz_matrix));
939 c->xyzgamma = xyzgamma_tab;
940 c->rgbgamma = rgbgamma_tab;
941 c->xyzgammainv = xyzgammainv_tab;
942 c->rgbgammainv = rgbgammainv_tab;
944 if (rgbgamma_tab[4095])
947 /* set gamma vectors */
948 for (i = 0; i < 4096; i++) {
949 xyzgamma_tab[i] = lrint(pow(i / 4095.0, xyzgamma) * 4095.0);
950 rgbgamma_tab[i] = lrint(pow(i / 4095.0, rgbgamma) * 4095.0);
951 xyzgammainv_tab[i] = lrint(pow(i / 4095.0, xyzgammainv) * 4095.0);
952 rgbgammainv_tab[i] = lrint(pow(i / 4095.0, rgbgammainv) * 4095.0);
956 int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],
957 int srcRange, const int table[4], int dstRange,
958 int brightness, int contrast, int saturation)
960 const AVPixFmtDescriptor *desc_dst;
961 const AVPixFmtDescriptor *desc_src;
962 memcpy(c->srcColorspaceTable, inv_table, sizeof(int) * 4);
963 memcpy(c->dstColorspaceTable, table, sizeof(int) * 4);
966 desc_dst = av_pix_fmt_desc_get(c->dstFormat);
967 desc_src = av_pix_fmt_desc_get(c->srcFormat);
969 if(!isYUV(c->dstFormat) && !isGray(c->dstFormat))
971 if(!isYUV(c->srcFormat) && !isGray(c->srcFormat))
974 c->brightness = brightness;
975 c->contrast = contrast;
976 c->saturation = saturation;
977 c->srcRange = srcRange;
978 c->dstRange = dstRange;
982 if ((isYUV(c->dstFormat) || isGray(c->dstFormat)) && (isYUV(c->srcFormat) || isGray(c->srcFormat)))
985 c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);
986 c->srcFormatBpp = av_get_bits_per_pixel(desc_src);
988 if (!isYUV(c->dstFormat) && !isGray(c->dstFormat)) {
989 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness,
990 contrast, saturation);
993 if (HAVE_ALTIVEC && av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)
994 ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness,
995 contrast, saturation);
998 fill_rgb2yuv_table(c, table, dstRange);
1003 int sws_getColorspaceDetails(struct SwsContext *c, int **inv_table,
1004 int *srcRange, int **table, int *dstRange,
1005 int *brightness, int *contrast, int *saturation)
1010 *inv_table = c->srcColorspaceTable;
1011 *table = c->dstColorspaceTable;
1012 *srcRange = c->srcRange;
1013 *dstRange = c->dstRange;
1014 *brightness = c->brightness;
1015 *contrast = c->contrast;
1016 *saturation = c->saturation;
1021 static int handle_jpeg(enum AVPixelFormat *format)
1024 case AV_PIX_FMT_YUVJ420P:
1025 *format = AV_PIX_FMT_YUV420P;
1027 case AV_PIX_FMT_YUVJ411P:
1028 *format = AV_PIX_FMT_YUV411P;
1030 case AV_PIX_FMT_YUVJ422P:
1031 *format = AV_PIX_FMT_YUV422P;
1033 case AV_PIX_FMT_YUVJ444P:
1034 *format = AV_PIX_FMT_YUV444P;
1036 case AV_PIX_FMT_YUVJ440P:
1037 *format = AV_PIX_FMT_YUV440P;
1039 case AV_PIX_FMT_GRAY8:
1046 static int handle_0alpha(enum AVPixelFormat *format)
1049 case AV_PIX_FMT_0BGR : *format = AV_PIX_FMT_ABGR ; return 1;
1050 case AV_PIX_FMT_BGR0 : *format = AV_PIX_FMT_BGRA ; return 4;
1051 case AV_PIX_FMT_0RGB : *format = AV_PIX_FMT_ARGB ; return 1;
1052 case AV_PIX_FMT_RGB0 : *format = AV_PIX_FMT_RGBA ; return 4;
1057 static int handle_xyz(enum AVPixelFormat *format)
1060 case AV_PIX_FMT_XYZ12BE : *format = AV_PIX_FMT_RGB48BE; return 1;
1061 case AV_PIX_FMT_XYZ12LE : *format = AV_PIX_FMT_RGB48LE; return 1;
1066 static void handle_formats(SwsContext *c)
1068 c->src0Alpha |= handle_0alpha(&c->srcFormat);
1069 c->dst0Alpha |= handle_0alpha(&c->dstFormat);
1070 c->srcXYZ |= handle_xyz(&c->srcFormat);
1071 c->dstXYZ |= handle_xyz(&c->dstFormat);
1074 SwsContext *sws_alloc_context(void)
1076 SwsContext *c = av_mallocz(sizeof(SwsContext));
1079 c->av_class = &sws_context_class;
1080 av_opt_set_defaults(c);
1086 av_cold int sws_init_context(SwsContext *c, SwsFilter *srcFilter,
1087 SwsFilter *dstFilter)
1090 int usesVFilter, usesHFilter;
1092 SwsFilter dummyFilter = { NULL, NULL, NULL, NULL };
1097 int dst_stride = FFALIGN(dstW * sizeof(int16_t) + 66, 16);
1098 int flags, cpu_flags;
1099 enum AVPixelFormat srcFormat = c->srcFormat;
1100 enum AVPixelFormat dstFormat = c->dstFormat;
1101 const AVPixFmtDescriptor *desc_src;
1102 const AVPixFmtDescriptor *desc_dst;
1104 cpu_flags = av_get_cpu_flags();
1110 unscaled = (srcW == dstW && srcH == dstH);
1112 c->srcRange |= handle_jpeg(&c->srcFormat);
1113 c->dstRange |= handle_jpeg(&c->dstFormat);
1115 if (!c->contrast && !c->saturation && !c->dstFormatBpp)
1116 sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,
1117 ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],
1118 c->dstRange, 0, 1 << 16, 1 << 16);
1120 if(srcFormat!=c->srcFormat || dstFormat!=c->dstFormat)
1121 av_log(c, AV_LOG_WARNING, "deprecated pixel format used, make sure you did set range correctly\n");
1123 srcFormat = c->srcFormat;
1124 dstFormat = c->dstFormat;
1125 desc_src = av_pix_fmt_desc_get(srcFormat);
1126 desc_dst = av_pix_fmt_desc_get(dstFormat);
1128 if (!(unscaled && sws_isSupportedEndiannessConversion(srcFormat) &&
1129 av_pix_fmt_swap_endianness(srcFormat) == dstFormat)) {
1130 if (!sws_isSupportedInput(srcFormat)) {
1131 av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n",
1132 av_get_pix_fmt_name(srcFormat));
1133 return AVERROR(EINVAL);
1135 if (!sws_isSupportedOutput(dstFormat)) {
1136 av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n",
1137 av_get_pix_fmt_name(dstFormat));
1138 return AVERROR(EINVAL);
1142 i = flags & (SWS_POINT |
1153 if (!i || (i & (i - 1))) {
1154 av_log(c, AV_LOG_ERROR, "Exactly one scaler algorithm must be chosen, got %X\n", i);
1155 return AVERROR(EINVAL);
1158 if (srcW < 1 || srcH < 1 || dstW < 1 || dstH < 1) {
1159 /* FIXME check if these are enough and try to lower them after
1160 * fixing the relevant parts of the code */
1161 av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",
1162 srcW, srcH, dstW, dstH);
1163 return AVERROR(EINVAL);
1167 dstFilter = &dummyFilter;
1169 srcFilter = &dummyFilter;
1171 c->lumXInc = (((int64_t)srcW << 16) + (dstW >> 1)) / dstW;
1172 c->lumYInc = (((int64_t)srcH << 16) + (dstH >> 1)) / dstH;
1173 c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);
1174 c->srcFormatBpp = av_get_bits_per_pixel(desc_src);
1175 c->vRounder = 4 * 0x0001000100010001ULL;
1177 usesVFilter = (srcFilter->lumV && srcFilter->lumV->length > 1) ||
1178 (srcFilter->chrV && srcFilter->chrV->length > 1) ||
1179 (dstFilter->lumV && dstFilter->lumV->length > 1) ||
1180 (dstFilter->chrV && dstFilter->chrV->length > 1);
1181 usesHFilter = (srcFilter->lumH && srcFilter->lumH->length > 1) ||
1182 (srcFilter->chrH && srcFilter->chrH->length > 1) ||
1183 (dstFilter->lumH && dstFilter->lumH->length > 1) ||
1184 (dstFilter->chrH && dstFilter->chrH->length > 1);
1186 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
1187 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
1189 if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) {
1191 av_log(c, AV_LOG_DEBUG, "Forcing full internal H chroma due to odd output size\n");
1192 flags |= SWS_FULL_CHR_H_INT;
1197 if (c->dither == SWS_DITHER_AUTO) {
1198 if (flags & SWS_ERROR_DIFFUSION)
1199 c->dither = SWS_DITHER_ED;
1202 if(dstFormat == AV_PIX_FMT_BGR4_BYTE ||
1203 dstFormat == AV_PIX_FMT_RGB4_BYTE ||
1204 dstFormat == AV_PIX_FMT_BGR8 ||
1205 dstFormat == AV_PIX_FMT_RGB8) {
1206 if (c->dither == SWS_DITHER_AUTO)
1207 c->dither = (flags & SWS_FULL_CHR_H_INT) ? SWS_DITHER_ED : SWS_DITHER_BAYER;
1208 if (!(flags & SWS_FULL_CHR_H_INT)) {
1209 if (c->dither == SWS_DITHER_ED) {
1210 av_log(c, AV_LOG_DEBUG,
1211 "Desired dithering only supported in full chroma interpolation for destination format '%s'\n",
1212 av_get_pix_fmt_name(dstFormat));
1213 flags |= SWS_FULL_CHR_H_INT;
1217 if (flags & SWS_FULL_CHR_H_INT) {
1218 if (c->dither == SWS_DITHER_BAYER) {
1219 av_log(c, AV_LOG_DEBUG,
1220 "Ordered dither is not supported in full chroma interpolation for destination format '%s'\n",
1221 av_get_pix_fmt_name(dstFormat));
1222 c->dither = SWS_DITHER_ED;
1226 if (isPlanarRGB(dstFormat)) {
1227 if (!(flags & SWS_FULL_CHR_H_INT)) {
1228 av_log(c, AV_LOG_DEBUG,
1229 "%s output is not supported with half chroma resolution, switching to full\n",
1230 av_get_pix_fmt_name(dstFormat));
1231 flags |= SWS_FULL_CHR_H_INT;
1236 /* reuse chroma for 2 pixels RGB/BGR unless user wants full
1237 * chroma interpolation */
1238 if (flags & SWS_FULL_CHR_H_INT &&
1239 isAnyRGB(dstFormat) &&
1240 !isPlanarRGB(dstFormat) &&
1241 dstFormat != AV_PIX_FMT_RGBA &&
1242 dstFormat != AV_PIX_FMT_ARGB &&
1243 dstFormat != AV_PIX_FMT_BGRA &&
1244 dstFormat != AV_PIX_FMT_ABGR &&
1245 dstFormat != AV_PIX_FMT_RGB24 &&
1246 dstFormat != AV_PIX_FMT_BGR24 &&
1247 dstFormat != AV_PIX_FMT_BGR4_BYTE &&
1248 dstFormat != AV_PIX_FMT_RGB4_BYTE &&
1249 dstFormat != AV_PIX_FMT_BGR8 &&
1250 dstFormat != AV_PIX_FMT_RGB8
1252 av_log(c, AV_LOG_WARNING,
1253 "full chroma interpolation for destination format '%s' not yet implemented\n",
1254 av_get_pix_fmt_name(dstFormat));
1255 flags &= ~SWS_FULL_CHR_H_INT;
1258 if (isAnyRGB(dstFormat) && !(flags & SWS_FULL_CHR_H_INT))
1259 c->chrDstHSubSample = 1;
1261 // drop some chroma lines if the user wants it
1262 c->vChrDrop = (flags & SWS_SRC_V_CHR_DROP_MASK) >>
1263 SWS_SRC_V_CHR_DROP_SHIFT;
1264 c->chrSrcVSubSample += c->vChrDrop;
1266 /* drop every other pixel for chroma calculation unless user
1267 * wants full chroma */
1268 if (isAnyRGB(srcFormat) && !(flags & SWS_FULL_CHR_H_INP) &&
1269 srcFormat != AV_PIX_FMT_RGB8 && srcFormat != AV_PIX_FMT_BGR8 &&
1270 srcFormat != AV_PIX_FMT_RGB4 && srcFormat != AV_PIX_FMT_BGR4 &&
1271 srcFormat != AV_PIX_FMT_RGB4_BYTE && srcFormat != AV_PIX_FMT_BGR4_BYTE &&
1272 srcFormat != AV_PIX_FMT_GBRP9BE && srcFormat != AV_PIX_FMT_GBRP9LE &&
1273 srcFormat != AV_PIX_FMT_GBRP10BE && srcFormat != AV_PIX_FMT_GBRP10LE &&
1274 srcFormat != AV_PIX_FMT_GBRP12BE && srcFormat != AV_PIX_FMT_GBRP12LE &&
1275 srcFormat != AV_PIX_FMT_GBRP14BE && srcFormat != AV_PIX_FMT_GBRP14LE &&
1276 srcFormat != AV_PIX_FMT_GBRP16BE && srcFormat != AV_PIX_FMT_GBRP16LE &&
1277 ((dstW >> c->chrDstHSubSample) <= (srcW >> 1) ||
1278 (flags & SWS_FAST_BILINEAR)))
1279 c->chrSrcHSubSample = 1;
1281 // Note the FF_CEIL_RSHIFT is so that we always round toward +inf.
1282 c->chrSrcW = FF_CEIL_RSHIFT(srcW, c->chrSrcHSubSample);
1283 c->chrSrcH = FF_CEIL_RSHIFT(srcH, c->chrSrcVSubSample);
1284 c->chrDstW = FF_CEIL_RSHIFT(dstW, c->chrDstHSubSample);
1285 c->chrDstH = FF_CEIL_RSHIFT(dstH, c->chrDstVSubSample);
1287 FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW*2+78, 16) * 2, fail);
1289 /* unscaled special cases */
1290 if (unscaled && !usesHFilter && !usesVFilter &&
1291 (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
1292 ff_get_unscaled_swscale(c);
1295 if (flags & SWS_PRINT_INFO)
1296 av_log(c, AV_LOG_INFO,
1297 "using unscaled %s -> %s special converter\n",
1298 av_get_pix_fmt_name(srcFormat), av_get_pix_fmt_name(dstFormat));
1303 c->srcBpc = 1 + desc_src->comp[0].depth_minus1;
1306 c->dstBpc = 1 + desc_dst->comp[0].depth_minus1;
1309 if (isAnyRGB(srcFormat) || srcFormat == AV_PIX_FMT_PAL8)
1311 if (c->dstBpc == 16)
1314 if (INLINE_MMXEXT(cpu_flags) && c->srcBpc == 8 && c->dstBpc <= 14) {
1315 c->canMMXEXTBeUsed = (dstW >= srcW && (dstW & 31) == 0 &&
1316 (srcW & 15) == 0) ? 1 : 0;
1317 if (!c->canMMXEXTBeUsed && dstW >= srcW && (srcW & 15) == 0
1319 && (flags & SWS_FAST_BILINEAR)) {
1320 if (flags & SWS_PRINT_INFO)
1321 av_log(c, AV_LOG_INFO,
1322 "output width is not a multiple of 32 -> no MMXEXT scaler\n");
1324 if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat))
1325 c->canMMXEXTBeUsed = 0;
1327 c->canMMXEXTBeUsed = 0;
1329 c->chrXInc = (((int64_t)c->chrSrcW << 16) + (c->chrDstW >> 1)) / c->chrDstW;
1330 c->chrYInc = (((int64_t)c->chrSrcH << 16) + (c->chrDstH >> 1)) / c->chrDstH;
1332 /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src
1333 * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do
1335 * n-2 is the last chrominance sample available.
1336 * This is not perfect, but no one should notice the difference, the more
1337 * correct variant would be like the vertical one, but that would require
1338 * some special code for the first and last pixel */
1339 if (flags & SWS_FAST_BILINEAR) {
1340 if (c->canMMXEXTBeUsed) {
1344 // we don't use the x86 asm scaler if MMX is available
1345 else if (INLINE_MMX(cpu_flags) && c->dstBpc <= 14) {
1346 c->lumXInc = ((int64_t)(srcW - 2) << 16) / (dstW - 2) - 20;
1347 c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;
1351 #define USE_MMAP (HAVE_MMAP && HAVE_MPROTECT && defined MAP_ANONYMOUS)
1353 /* precalculate horizontal scaler filter coefficients */
1355 #if HAVE_MMXEXT_INLINE
1356 // can't downscale !!!
1357 if (c->canMMXEXTBeUsed && (flags & SWS_FAST_BILINEAR)) {
1358 c->lumMmxextFilterCodeSize = init_hscaler_mmxext(dstW, c->lumXInc, NULL,
1360 c->chrMmxextFilterCodeSize = init_hscaler_mmxext(c->chrDstW, c->chrXInc,
1361 NULL, NULL, NULL, 4);
1364 c->lumMmxextFilterCode = mmap(NULL, c->lumMmxextFilterCodeSize,
1365 PROT_READ | PROT_WRITE,
1366 MAP_PRIVATE | MAP_ANONYMOUS,
1368 c->chrMmxextFilterCode = mmap(NULL, c->chrMmxextFilterCodeSize,
1369 PROT_READ | PROT_WRITE,
1370 MAP_PRIVATE | MAP_ANONYMOUS,
1372 #elif HAVE_VIRTUALALLOC
1373 c->lumMmxextFilterCode = VirtualAlloc(NULL,
1374 c->lumMmxextFilterCodeSize,
1376 PAGE_EXECUTE_READWRITE);
1377 c->chrMmxextFilterCode = VirtualAlloc(NULL,
1378 c->chrMmxextFilterCodeSize,
1380 PAGE_EXECUTE_READWRITE);
1382 c->lumMmxextFilterCode = av_malloc(c->lumMmxextFilterCodeSize);
1383 c->chrMmxextFilterCode = av_malloc(c->chrMmxextFilterCodeSize);
1386 #ifdef MAP_ANONYMOUS
1387 if (c->lumMmxextFilterCode == MAP_FAILED || c->chrMmxextFilterCode == MAP_FAILED)
1389 if (!c->lumMmxextFilterCode || !c->chrMmxextFilterCode)
1392 av_log(c, AV_LOG_ERROR, "Failed to allocate MMX2FilterCode\n");
1393 return AVERROR(ENOMEM);
1396 FF_ALLOCZ_OR_GOTO(c, c->hLumFilter, (dstW / 8 + 8) * sizeof(int16_t), fail);
1397 FF_ALLOCZ_OR_GOTO(c, c->hChrFilter, (c->chrDstW / 4 + 8) * sizeof(int16_t), fail);
1398 FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW / 2 / 8 + 8) * sizeof(int32_t), fail);
1399 FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);
1401 init_hscaler_mmxext( dstW, c->lumXInc, c->lumMmxextFilterCode,
1402 c->hLumFilter, (uint32_t*)c->hLumFilterPos, 8);
1403 init_hscaler_mmxext(c->chrDstW, c->chrXInc, c->chrMmxextFilterCode,
1404 c->hChrFilter, (uint32_t*)c->hChrFilterPos, 4);
1407 mprotect(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize, PROT_EXEC | PROT_READ);
1408 mprotect(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize, PROT_EXEC | PROT_READ);
1411 #endif /* HAVE_MMXEXT_INLINE */
1413 const int filterAlign =
1414 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :
1415 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
1418 if (initFilter(&c->hLumFilter, &c->hLumFilterPos,
1419 &c->hLumFilterSize, c->lumXInc,
1420 srcW, dstW, filterAlign, 1 << 14,
1421 (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
1422 cpu_flags, srcFilter->lumH, dstFilter->lumH,
1424 get_local_pos(c, 0, 0, 0),
1425 get_local_pos(c, 0, 0, 0)) < 0)
1427 if (initFilter(&c->hChrFilter, &c->hChrFilterPos,
1428 &c->hChrFilterSize, c->chrXInc,
1429 c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,
1430 (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
1431 cpu_flags, srcFilter->chrH, dstFilter->chrH,
1433 get_local_pos(c, c->chrSrcHSubSample, c->src_h_chr_pos, 0),
1434 get_local_pos(c, c->chrDstHSubSample, c->dst_h_chr_pos, 0)) < 0)
1437 } // initialize horizontal stuff
1439 /* precalculate vertical scaler filter coefficients */
1441 const int filterAlign =
1442 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 2 :
1443 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
1446 if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,
1447 c->lumYInc, srcH, dstH, filterAlign, (1 << 12),
1448 (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
1449 cpu_flags, srcFilter->lumV, dstFilter->lumV,
1451 get_local_pos(c, 0, 0, 1),
1452 get_local_pos(c, 0, 0, 1)) < 0)
1454 if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,
1455 c->chrYInc, c->chrSrcH, c->chrDstH,
1456 filterAlign, (1 << 12),
1457 (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
1458 cpu_flags, srcFilter->chrV, dstFilter->chrV,
1460 get_local_pos(c, c->chrSrcVSubSample, c->src_v_chr_pos, 1),
1461 get_local_pos(c, c->chrDstVSubSample, c->dst_v_chr_pos, 1)) < 0)
1466 FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH, fail);
1467 FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);
1469 for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {
1471 short *p = (short *)&c->vYCoeffsBank[i];
1472 for (j = 0; j < 8; j++)
1473 p[j] = c->vLumFilter[i];
1476 for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {
1478 short *p = (short *)&c->vCCoeffsBank[i];
1479 for (j = 0; j < 8; j++)
1480 p[j] = c->vChrFilter[i];
1485 // calculate buffer sizes so that they won't run out while handling these damn slices
1486 c->vLumBufSize = c->vLumFilterSize;
1487 c->vChrBufSize = c->vChrFilterSize;
1488 for (i = 0; i < dstH; i++) {
1489 int chrI = (int64_t)i * c->chrDstH / dstH;
1490 int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,
1491 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)
1492 << c->chrSrcVSubSample));
1494 nextSlice >>= c->chrSrcVSubSample;
1495 nextSlice <<= c->chrSrcVSubSample;
1496 if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)
1497 c->vLumBufSize = nextSlice - c->vLumFilterPos[i];
1498 if (c->vChrFilterPos[chrI] + c->vChrBufSize <
1499 (nextSlice >> c->chrSrcVSubSample))
1500 c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -
1501 c->vChrFilterPos[chrI];
1504 for (i = 0; i < 4; i++)
1505 FF_ALLOCZ_OR_GOTO(c, c->dither_error[i], (c->dstW+2) * sizeof(int), fail);
1507 /* Allocate pixbufs (we use dynamic allocation because otherwise we would
1508 * need to allocate several megabytes to handle all possible cases) */
1509 FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);
1510 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
1511 FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
1512 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
1513 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);
1514 /* Note we need at least one pixel more at the end because of the MMX code
1515 * (just in case someone wants to replace the 4000/8000). */
1516 /* align at 16 bytes for AltiVec */
1517 for (i = 0; i < c->vLumBufSize; i++) {
1518 FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],
1519 dst_stride + 16, fail);
1520 c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];
1522 // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)
1523 c->uv_off = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);
1524 c->uv_offx2 = dst_stride + 16;
1525 for (i = 0; i < c->vChrBufSize; i++) {
1526 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],
1527 dst_stride * 2 + 32, fail);
1528 c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];
1529 c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]
1530 = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;
1532 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
1533 for (i = 0; i < c->vLumBufSize; i++) {
1534 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i + c->vLumBufSize],
1535 dst_stride + 16, fail);
1536 c->alpPixBuf[i] = c->alpPixBuf[i + c->vLumBufSize];
1539 // try to avoid drawing green stuff between the right end and the stride end
1540 for (i = 0; i < c->vChrBufSize; i++)
1541 if(desc_dst->comp[0].depth_minus1 == 15){
1542 av_assert0(c->dstBpc > 14);
1543 for(j=0; j<dst_stride/2+1; j++)
1544 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;
1546 for(j=0; j<dst_stride+1; j++)
1547 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;
1549 av_assert0(c->chrDstH <= dstH);
1551 if (flags & SWS_PRINT_INFO) {
1552 if (flags & SWS_FAST_BILINEAR)
1553 av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
1554 else if (flags & SWS_BILINEAR)
1555 av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
1556 else if (flags & SWS_BICUBIC)
1557 av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
1558 else if (flags & SWS_X)
1559 av_log(c, AV_LOG_INFO, "Experimental scaler, ");
1560 else if (flags & SWS_POINT)
1561 av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
1562 else if (flags & SWS_AREA)
1563 av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
1564 else if (flags & SWS_BICUBLIN)
1565 av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
1566 else if (flags & SWS_GAUSS)
1567 av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
1568 else if (flags & SWS_SINC)
1569 av_log(c, AV_LOG_INFO, "Sinc scaler, ");
1570 else if (flags & SWS_LANCZOS)
1571 av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
1572 else if (flags & SWS_SPLINE)
1573 av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
1575 av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
1577 av_log(c, AV_LOG_INFO, "from %s to %s%s ",
1578 av_get_pix_fmt_name(srcFormat),
1580 dstFormat == AV_PIX_FMT_BGR555 || dstFormat == AV_PIX_FMT_BGR565 ||
1581 dstFormat == AV_PIX_FMT_RGB444BE || dstFormat == AV_PIX_FMT_RGB444LE ||
1582 dstFormat == AV_PIX_FMT_BGR444BE || dstFormat == AV_PIX_FMT_BGR444LE ?
1587 av_get_pix_fmt_name(dstFormat));
1589 if (INLINE_MMXEXT(cpu_flags))
1590 av_log(c, AV_LOG_INFO, "using MMXEXT\n");
1591 else if (INLINE_AMD3DNOW(cpu_flags))
1592 av_log(c, AV_LOG_INFO, "using 3DNOW\n");
1593 else if (INLINE_MMX(cpu_flags))
1594 av_log(c, AV_LOG_INFO, "using MMX\n");
1595 else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC)
1596 av_log(c, AV_LOG_INFO, "using AltiVec\n");
1598 av_log(c, AV_LOG_INFO, "using C\n");
1600 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
1601 av_log(c, AV_LOG_DEBUG,
1602 "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1603 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
1604 av_log(c, AV_LOG_DEBUG,
1605 "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1606 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,
1607 c->chrXInc, c->chrYInc);
1610 c->swScale = ff_getSwsFunc(c);
1612 fail: // FIXME replace things by appropriate error codes
1616 #if FF_API_SWS_GETCONTEXT
1617 SwsContext *sws_getContext(int srcW, int srcH, enum AVPixelFormat srcFormat,
1618 int dstW, int dstH, enum AVPixelFormat dstFormat,
1619 int flags, SwsFilter *srcFilter,
1620 SwsFilter *dstFilter, const double *param)
1624 if (!(c = sws_alloc_context()))
1632 c->srcFormat = srcFormat;
1633 c->dstFormat = dstFormat;
1636 c->param[0] = param[0];
1637 c->param[1] = param[1];
1640 if (sws_init_context(c, srcFilter, dstFilter) < 0) {
1649 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
1650 float lumaSharpen, float chromaSharpen,
1651 float chromaHShift, float chromaVShift,
1654 SwsFilter *filter = av_malloc(sizeof(SwsFilter));
1658 if (lumaGBlur != 0.0) {
1659 filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);
1660 filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);
1662 filter->lumH = sws_getIdentityVec();
1663 filter->lumV = sws_getIdentityVec();
1666 if (chromaGBlur != 0.0) {
1667 filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);
1668 filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);
1670 filter->chrH = sws_getIdentityVec();
1671 filter->chrV = sws_getIdentityVec();
1674 if (chromaSharpen != 0.0) {
1675 SwsVector *id = sws_getIdentityVec();
1676 sws_scaleVec(filter->chrH, -chromaSharpen);
1677 sws_scaleVec(filter->chrV, -chromaSharpen);
1678 sws_addVec(filter->chrH, id);
1679 sws_addVec(filter->chrV, id);
1683 if (lumaSharpen != 0.0) {
1684 SwsVector *id = sws_getIdentityVec();
1685 sws_scaleVec(filter->lumH, -lumaSharpen);
1686 sws_scaleVec(filter->lumV, -lumaSharpen);
1687 sws_addVec(filter->lumH, id);
1688 sws_addVec(filter->lumV, id);
1692 if (chromaHShift != 0.0)
1693 sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));
1695 if (chromaVShift != 0.0)
1696 sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));
1698 sws_normalizeVec(filter->chrH, 1.0);
1699 sws_normalizeVec(filter->chrV, 1.0);
1700 sws_normalizeVec(filter->lumH, 1.0);
1701 sws_normalizeVec(filter->lumV, 1.0);
1704 sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
1706 sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
1711 SwsVector *sws_allocVec(int length)
1715 if(length <= 0 || length > INT_MAX/ sizeof(double))
1718 vec = av_malloc(sizeof(SwsVector));
1721 vec->length = length;
1722 vec->coeff = av_malloc(sizeof(double) * length);
1728 SwsVector *sws_getGaussianVec(double variance, double quality)
1730 const int length = (int)(variance * quality + 0.5) | 1;
1732 double middle = (length - 1) * 0.5;
1735 if(variance < 0 || quality < 0)
1738 vec = sws_allocVec(length);
1743 for (i = 0; i < length; i++) {
1744 double dist = i - middle;
1745 vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /
1746 sqrt(2 * variance * M_PI);
1749 sws_normalizeVec(vec, 1.0);
1754 SwsVector *sws_getConstVec(double c, int length)
1757 SwsVector *vec = sws_allocVec(length);
1762 for (i = 0; i < length; i++)
1768 SwsVector *sws_getIdentityVec(void)
1770 return sws_getConstVec(1.0, 1);
1773 static double sws_dcVec(SwsVector *a)
1778 for (i = 0; i < a->length; i++)
1784 void sws_scaleVec(SwsVector *a, double scalar)
1788 for (i = 0; i < a->length; i++)
1789 a->coeff[i] *= scalar;
1792 void sws_normalizeVec(SwsVector *a, double height)
1794 sws_scaleVec(a, height / sws_dcVec(a));
1797 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
1799 int length = a->length + b->length - 1;
1801 SwsVector *vec = sws_getConstVec(0.0, length);
1806 for (i = 0; i < a->length; i++) {
1807 for (j = 0; j < b->length; j++) {
1808 vec->coeff[i + j] += a->coeff[i] * b->coeff[j];
1815 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
1817 int length = FFMAX(a->length, b->length);
1819 SwsVector *vec = sws_getConstVec(0.0, length);
1824 for (i = 0; i < a->length; i++)
1825 vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
1826 for (i = 0; i < b->length; i++)
1827 vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] += b->coeff[i];
1832 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
1834 int length = FFMAX(a->length, b->length);
1836 SwsVector *vec = sws_getConstVec(0.0, length);
1841 for (i = 0; i < a->length; i++)
1842 vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
1843 for (i = 0; i < b->length; i++)
1844 vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] -= b->coeff[i];
1849 /* shift left / or right if "shift" is negative */
1850 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
1852 int length = a->length + FFABS(shift) * 2;
1854 SwsVector *vec = sws_getConstVec(0.0, length);
1859 for (i = 0; i < a->length; i++) {
1860 vec->coeff[i + (length - 1) / 2 -
1861 (a->length - 1) / 2 - shift] = a->coeff[i];
1867 void sws_shiftVec(SwsVector *a, int shift)
1869 SwsVector *shifted = sws_getShiftedVec(a, shift);
1871 a->coeff = shifted->coeff;
1872 a->length = shifted->length;
1876 void sws_addVec(SwsVector *a, SwsVector *b)
1878 SwsVector *sum = sws_sumVec(a, b);
1880 a->coeff = sum->coeff;
1881 a->length = sum->length;
1885 void sws_subVec(SwsVector *a, SwsVector *b)
1887 SwsVector *diff = sws_diffVec(a, b);
1889 a->coeff = diff->coeff;
1890 a->length = diff->length;
1894 void sws_convVec(SwsVector *a, SwsVector *b)
1896 SwsVector *conv = sws_getConvVec(a, b);
1898 a->coeff = conv->coeff;
1899 a->length = conv->length;
1903 SwsVector *sws_cloneVec(SwsVector *a)
1906 SwsVector *vec = sws_allocVec(a->length);
1911 for (i = 0; i < a->length; i++)
1912 vec->coeff[i] = a->coeff[i];
1917 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
1924 for (i = 0; i < a->length; i++)
1925 if (a->coeff[i] > max)
1928 for (i = 0; i < a->length; i++)
1929 if (a->coeff[i] < min)
1934 for (i = 0; i < a->length; i++) {
1935 int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);
1936 av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
1938 av_log(log_ctx, log_level, " ");
1939 av_log(log_ctx, log_level, "|\n");
1943 void sws_freeVec(SwsVector *a)
1947 av_freep(&a->coeff);
1952 void sws_freeFilter(SwsFilter *filter)
1958 sws_freeVec(filter->lumH);
1960 sws_freeVec(filter->lumV);
1962 sws_freeVec(filter->chrH);
1964 sws_freeVec(filter->chrV);
1968 void sws_freeContext(SwsContext *c)
1975 for (i = 0; i < c->vLumBufSize; i++)
1976 av_freep(&c->lumPixBuf[i]);
1977 av_freep(&c->lumPixBuf);
1980 if (c->chrUPixBuf) {
1981 for (i = 0; i < c->vChrBufSize; i++)
1982 av_freep(&c->chrUPixBuf[i]);
1983 av_freep(&c->chrUPixBuf);
1984 av_freep(&c->chrVPixBuf);
1987 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1988 for (i = 0; i < c->vLumBufSize; i++)
1989 av_freep(&c->alpPixBuf[i]);
1990 av_freep(&c->alpPixBuf);
1993 for (i = 0; i < 4; i++)
1994 av_freep(&c->dither_error[i]);
1996 av_freep(&c->vLumFilter);
1997 av_freep(&c->vChrFilter);
1998 av_freep(&c->hLumFilter);
1999 av_freep(&c->hChrFilter);
2001 av_freep(&c->vYCoeffsBank);
2002 av_freep(&c->vCCoeffsBank);
2005 av_freep(&c->vLumFilterPos);
2006 av_freep(&c->vChrFilterPos);
2007 av_freep(&c->hLumFilterPos);
2008 av_freep(&c->hChrFilterPos);
2012 if (c->lumMmxextFilterCode)
2013 munmap(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize);
2014 if (c->chrMmxextFilterCode)
2015 munmap(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize);
2016 #elif HAVE_VIRTUALALLOC
2017 if (c->lumMmxextFilterCode)
2018 VirtualFree(c->lumMmxextFilterCode, 0, MEM_RELEASE);
2019 if (c->chrMmxextFilterCode)
2020 VirtualFree(c->chrMmxextFilterCode, 0, MEM_RELEASE);
2022 av_free(c->lumMmxextFilterCode);
2023 av_free(c->chrMmxextFilterCode);
2025 c->lumMmxextFilterCode = NULL;
2026 c->chrMmxextFilterCode = NULL;
2027 #endif /* HAVE_MMX_INLINE */
2029 av_freep(&c->yuvTable);
2030 av_freep(&c->formatConvBuffer);
2035 struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,
2036 int srcH, enum AVPixelFormat srcFormat,
2038 enum AVPixelFormat dstFormat, int flags,
2039 SwsFilter *srcFilter,
2040 SwsFilter *dstFilter,
2041 const double *param)
2043 static const double default_param[2] = { SWS_PARAM_DEFAULT,
2044 SWS_PARAM_DEFAULT };
2047 param = default_param;
2050 (context->srcW != srcW ||
2051 context->srcH != srcH ||
2052 context->srcFormat != srcFormat ||
2053 context->dstW != dstW ||
2054 context->dstH != dstH ||
2055 context->dstFormat != dstFormat ||
2056 context->flags != flags ||
2057 context->param[0] != param[0] ||
2058 context->param[1] != param[1])) {
2059 sws_freeContext(context);
2064 if (!(context = sws_alloc_context()))
2066 context->srcW = srcW;
2067 context->srcH = srcH;
2068 context->srcFormat = srcFormat;
2069 context->dstW = dstW;
2070 context->dstH = dstH;
2071 context->dstFormat = dstFormat;
2072 context->flags = flags;
2073 context->param[0] = param[0];
2074 context->param[1] = param[1];
2075 if (sws_init_context(context, srcFilter, dstFilter) < 0) {
2076 sws_freeContext(context);