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, 0, 1 },
205 [AV_PIX_FMT_XYZ12LE] = { 1, 0, 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 extern const int32_t ff_yuv2rgb_coeffs[8][4];
231 #if FF_API_SWS_FORMAT_NAME
232 const char *sws_format_name(enum AVPixelFormat format)
234 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(format);
238 return "Unknown format";
242 static double getSplineCoeff(double a, double b, double c, double d,
246 return ((d * dist + c) * dist + b) * dist + a;
248 return getSplineCoeff(0.0,
249 b + 2.0 * c + 3.0 * d,
251 -b - 3.0 * c - 6.0 * d,
255 static av_cold int get_local_pos(SwsContext *s, int chr_subsample, int pos, int dir)
258 pos = (128 << chr_subsample) - 128;
260 pos += 128; // relative to ideal left edge
261 return pos >> chr_subsample;
264 static av_cold int initFilter(int16_t **outFilter, int32_t **filterPos,
265 int *outFilterSize, int xInc, int srcW,
266 int dstW, int filterAlign, int one,
267 int flags, int cpu_flags,
268 SwsVector *srcFilter, SwsVector *dstFilter,
269 double param[2], int srcPos, int dstPos)
275 int64_t *filter = NULL;
276 int64_t *filter2 = NULL;
277 const int64_t fone = 1LL << (54 - FFMIN(av_log2(srcW/dstW), 8));
280 emms_c(); // FIXME should not be required but IS (even for non-MMX versions)
282 // NOTE: the +3 is for the MMX(+1) / SSE(+3) scaler which reads over the end
283 FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW + 3) * sizeof(**filterPos), fail);
285 if (FFABS(xInc - 0x10000) < 10 && srcPos == dstPos) { // unscaled
288 FF_ALLOCZ_OR_GOTO(NULL, filter,
289 dstW * sizeof(*filter) * filterSize, fail);
291 for (i = 0; i < dstW; i++) {
292 filter[i * filterSize] = fone;
295 } else if (flags & SWS_POINT) { // lame looking point sampling mode
299 FF_ALLOC_OR_GOTO(NULL, filter,
300 dstW * sizeof(*filter) * filterSize, fail);
302 xDstInSrc = ((srcPos*xInc)>>8) - ((dstPos*0x8000)>>7);
303 for (i = 0; i < dstW; i++) {
304 int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;
306 (*filterPos)[i] = xx;
310 } else if ((xInc <= (1 << 16) && (flags & SWS_AREA)) ||
311 (flags & SWS_FAST_BILINEAR)) { // bilinear upscale
315 FF_ALLOC_OR_GOTO(NULL, filter,
316 dstW * sizeof(*filter) * filterSize, fail);
318 xDstInSrc = ((srcPos*xInc)>>8) - ((dstPos*0x8000)>>7);
319 for (i = 0; i < dstW; i++) {
320 int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;
323 (*filterPos)[i] = xx;
324 // bilinear upscale / linear interpolate / area averaging
325 for (j = 0; j < filterSize; j++) {
326 int64_t coeff= fone - FFABS(((int64_t)xx<<16) - xDstInSrc)*(fone>>16);
329 filter[i * filterSize + j] = coeff;
338 if (flags & SWS_BICUBIC)
340 else if (flags & SWS_X)
342 else if (flags & SWS_AREA)
343 sizeFactor = 1; // downscale only, for upscale it is bilinear
344 else if (flags & SWS_GAUSS)
345 sizeFactor = 8; // infinite ;)
346 else if (flags & SWS_LANCZOS)
347 sizeFactor = param[0] != SWS_PARAM_DEFAULT ? ceil(2 * param[0]) : 6;
348 else if (flags & SWS_SINC)
349 sizeFactor = 20; // infinite ;)
350 else if (flags & SWS_SPLINE)
351 sizeFactor = 20; // infinite ;)
352 else if (flags & SWS_BILINEAR)
359 filterSize = 1 + sizeFactor; // upscale
361 filterSize = 1 + (sizeFactor * srcW + dstW - 1) / dstW;
363 filterSize = FFMIN(filterSize, srcW - 2);
364 filterSize = FFMAX(filterSize, 1);
366 FF_ALLOC_OR_GOTO(NULL, filter,
367 dstW * sizeof(*filter) * filterSize, fail);
369 xDstInSrc = ((srcPos*xInc)>>7) - ((dstPos*0x10000)>>7);
370 for (i = 0; i < dstW; i++) {
371 int xx = (xDstInSrc - ((filterSize - 2) << 16)) / (1 << 17);
373 (*filterPos)[i] = xx;
374 for (j = 0; j < filterSize; j++) {
375 int64_t d = (FFABS(((int64_t)xx << 17) - xDstInSrc)) << 13;
381 floatd = d * (1.0 / (1 << 30));
383 if (flags & SWS_BICUBIC) {
384 int64_t B = (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1 << 24);
385 int64_t C = (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1 << 24);
387 if (d >= 1LL << 31) {
390 int64_t dd = (d * d) >> 30;
391 int64_t ddd = (dd * d) >> 30;
394 coeff = (12 * (1 << 24) - 9 * B - 6 * C) * ddd +
395 (-18 * (1 << 24) + 12 * B + 6 * C) * dd +
396 (6 * (1 << 24) - 2 * B) * (1 << 30);
398 coeff = (-B - 6 * C) * ddd +
399 (6 * B + 30 * C) * dd +
400 (-12 * B - 48 * C) * d +
401 (8 * B + 24 * C) * (1 << 30);
403 coeff /= (1LL<<54)/fone;
406 else if (flags & SWS_X) {
407 double p = param ? param * 0.01 : 0.3;
408 coeff = d ? sin(d * M_PI) / (d * M_PI) : 1.0;
409 coeff *= pow(2.0, -p * d * d);
412 else if (flags & SWS_X) {
413 double A = param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
417 c = cos(floatd * M_PI);
424 coeff = (c * 0.5 + 0.5) * fone;
425 } else if (flags & SWS_AREA) {
426 int64_t d2 = d - (1 << 29);
427 if (d2 * xInc < -(1LL << (29 + 16)))
428 coeff = 1.0 * (1LL << (30 + 16));
429 else if (d2 * xInc < (1LL << (29 + 16)))
430 coeff = -d2 * xInc + (1LL << (29 + 16));
433 coeff *= fone >> (30 + 16);
434 } else if (flags & SWS_GAUSS) {
435 double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
436 coeff = (pow(2.0, -p * floatd * floatd)) * fone;
437 } else if (flags & SWS_SINC) {
438 coeff = (d ? sin(floatd * M_PI) / (floatd * M_PI) : 1.0) * fone;
439 } else if (flags & SWS_LANCZOS) {
440 double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
441 coeff = (d ? sin(floatd * M_PI) * sin(floatd * M_PI / p) /
442 (floatd * floatd * M_PI * M_PI / p) : 1.0) * fone;
445 } else if (flags & SWS_BILINEAR) {
446 coeff = (1 << 30) - d;
450 } else if (flags & SWS_SPLINE) {
451 double p = -2.196152422706632;
452 coeff = getSplineCoeff(1.0, 0.0, p, -p - 1.0, floatd) * fone;
457 filter[i * filterSize + j] = coeff;
460 xDstInSrc += 2 * xInc;
464 /* apply src & dst Filter to filter -> filter2
467 av_assert0(filterSize > 0);
468 filter2Size = filterSize;
470 filter2Size += srcFilter->length - 1;
472 filter2Size += dstFilter->length - 1;
473 av_assert0(filter2Size > 0);
474 FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size * dstW * sizeof(*filter2), fail);
476 for (i = 0; i < dstW; i++) {
480 for (k = 0; k < srcFilter->length; k++) {
481 for (j = 0; j < filterSize; j++)
482 filter2[i * filter2Size + k + j] +=
483 srcFilter->coeff[k] * filter[i * filterSize + j];
486 for (j = 0; j < filterSize; j++)
487 filter2[i * filter2Size + j] = filter[i * filterSize + j];
491 (*filterPos)[i] += (filterSize - 1) / 2 - (filter2Size - 1) / 2;
495 /* try to reduce the filter-size (step1 find size and shift left) */
496 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
498 for (i = dstW - 1; i >= 0; i--) {
499 int min = filter2Size;
501 int64_t cutOff = 0.0;
503 /* get rid of near zero elements on the left by shifting left */
504 for (j = 0; j < filter2Size; j++) {
506 cutOff += FFABS(filter2[i * filter2Size]);
508 if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)
511 /* preserve monotonicity because the core can't handle the
512 * filter otherwise */
513 if (i < dstW - 1 && (*filterPos)[i] >= (*filterPos)[i + 1])
516 // move filter coefficients left
517 for (k = 1; k < filter2Size; k++)
518 filter2[i * filter2Size + k - 1] = filter2[i * filter2Size + k];
519 filter2[i * filter2Size + k - 1] = 0;
524 /* count near zeros on the right */
525 for (j = filter2Size - 1; j > 0; j--) {
526 cutOff += FFABS(filter2[i * filter2Size + j]);
528 if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)
533 if (min > minFilterSize)
537 if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) {
538 // we can handle the special case 4, so we don't want to go the full 8
539 if (minFilterSize < 5)
542 /* We really don't want to waste our time doing useless computation, so
543 * fall back on the scalar C code for very small filters.
544 * Vectorizing is worth it only if you have a decent-sized vector. */
545 if (minFilterSize < 3)
549 if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
550 // special case for unscaled vertical filtering
551 if (minFilterSize == 1 && filterAlign == 2)
555 av_assert0(minFilterSize > 0);
556 filterSize = (minFilterSize + (filterAlign - 1)) & (~(filterAlign - 1));
557 av_assert0(filterSize > 0);
558 filter = av_malloc(filterSize * dstW * sizeof(*filter));
559 if (filterSize >= MAX_FILTER_SIZE * 16 /
560 ((flags & SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter) {
561 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);
564 *outFilterSize = filterSize;
566 if (flags & SWS_PRINT_INFO)
567 av_log(NULL, AV_LOG_VERBOSE,
568 "SwScaler: reducing / aligning filtersize %d -> %d\n",
569 filter2Size, filterSize);
570 /* try to reduce the filter-size (step2 reduce it) */
571 for (i = 0; i < dstW; i++) {
574 for (j = 0; j < filterSize; j++) {
575 if (j >= filter2Size)
576 filter[i * filterSize + j] = 0;
578 filter[i * filterSize + j] = filter2[i * filter2Size + j];
579 if ((flags & SWS_BITEXACT) && j >= minFilterSize)
580 filter[i * filterSize + j] = 0;
584 // FIXME try to align filterPos if possible
587 for (i = 0; i < dstW; i++) {
589 if ((*filterPos)[i] < 0) {
590 // move filter coefficients left to compensate for filterPos
591 for (j = 1; j < filterSize; j++) {
592 int left = FFMAX(j + (*filterPos)[i], 0);
593 filter[i * filterSize + left] += filter[i * filterSize + j];
594 filter[i * filterSize + j] = 0;
599 if ((*filterPos)[i] + filterSize > srcW) {
600 int shift = (*filterPos)[i] + filterSize - srcW;
601 // move filter coefficients right to compensate for filterPos
602 for (j = filterSize - 2; j >= 0; j--) {
603 int right = FFMIN(j + shift, filterSize - 1);
604 filter[i * filterSize + right] += filter[i * filterSize + j];
605 filter[i * filterSize + j] = 0;
607 (*filterPos)[i]= srcW - filterSize;
611 // Note the +1 is for the MMX scaler which reads over the end
612 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
613 FF_ALLOCZ_OR_GOTO(NULL, *outFilter,
614 *outFilterSize * (dstW + 3) * sizeof(int16_t), fail);
616 /* normalize & store in outFilter */
617 for (i = 0; i < dstW; i++) {
622 for (j = 0; j < filterSize; j++) {
623 sum += filter[i * filterSize + j];
625 sum = (sum + one / 2) / one;
626 for (j = 0; j < *outFilterSize; j++) {
627 int64_t v = filter[i * filterSize + j] + error;
628 int intV = ROUNDED_DIV(v, sum);
629 (*outFilter)[i * (*outFilterSize) + j] = intV;
630 error = v - intV * sum;
634 (*filterPos)[dstW + 0] =
635 (*filterPos)[dstW + 1] =
636 (*filterPos)[dstW + 2] = (*filterPos)[dstW - 1]; /* the MMX/SSE scaler will
637 * read over the end */
638 for (i = 0; i < *outFilterSize; i++) {
639 int k = (dstW - 1) * (*outFilterSize) + i;
640 (*outFilter)[k + 1 * (*outFilterSize)] =
641 (*outFilter)[k + 2 * (*outFilterSize)] =
642 (*outFilter)[k + 3 * (*outFilterSize)] = (*outFilter)[k];
649 av_log(NULL, AV_LOG_ERROR, "sws: initFilter failed\n");
655 #if HAVE_MMXEXT_INLINE
656 static av_cold int init_hscaler_mmxext(int dstW, int xInc, uint8_t *filterCode,
657 int16_t *filter, int32_t *filterPos,
661 x86_reg imm8OfPShufW1A;
662 x86_reg imm8OfPShufW2A;
663 x86_reg fragmentLengthA;
665 x86_reg imm8OfPShufW1B;
666 x86_reg imm8OfPShufW2B;
667 x86_reg fragmentLengthB;
672 // create an optimized horizontal scaling routine
673 /* This scaler is made of runtime-generated MMXEXT code using specially tuned
674 * pshufw instructions. For every four output pixels, if four input pixels
675 * are enough for the fast bilinear scaling, then a chunk of fragmentB is
676 * used. If five input pixels are needed, then a chunk of fragmentA is used.
685 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
686 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
687 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
688 "punpcklbw %%mm7, %%mm1 \n\t"
689 "punpcklbw %%mm7, %%mm0 \n\t"
690 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
692 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
694 "psubw %%mm1, %%mm0 \n\t"
695 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
696 "pmullw %%mm3, %%mm0 \n\t"
697 "psllw $7, %%mm1 \n\t"
698 "paddw %%mm1, %%mm0 \n\t"
700 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
702 "add $8, %%"REG_a" \n\t"
706 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
707 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
708 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
713 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
717 : "=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
718 "=r" (fragmentLengthA)
725 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
726 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
727 "punpcklbw %%mm7, %%mm0 \n\t"
728 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
730 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
732 "psubw %%mm1, %%mm0 \n\t"
733 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
734 "pmullw %%mm3, %%mm0 \n\t"
735 "psllw $7, %%mm1 \n\t"
736 "paddw %%mm1, %%mm0 \n\t"
738 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
740 "add $8, %%"REG_a" \n\t"
744 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
745 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
746 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
751 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
755 : "=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
756 "=r" (fragmentLengthB)
759 xpos = 0; // lumXInc/2 - 0x8000; // difference between pixel centers
762 for (i = 0; i < dstW / numSplits; i++) {
767 int b = ((xpos + xInc) >> 16) - xx;
768 int c = ((xpos + xInc * 2) >> 16) - xx;
769 int d = ((xpos + xInc * 3) >> 16) - xx;
770 int inc = (d + 1 < 4);
771 uint8_t *fragment = (d + 1 < 4) ? fragmentB : fragmentA;
772 x86_reg imm8OfPShufW1 = (d + 1 < 4) ? imm8OfPShufW1B : imm8OfPShufW1A;
773 x86_reg imm8OfPShufW2 = (d + 1 < 4) ? imm8OfPShufW2B : imm8OfPShufW2A;
774 x86_reg fragmentLength = (d + 1 < 4) ? fragmentLengthB : fragmentLengthA;
775 int maxShift = 3 - (d + inc);
779 filter[i] = ((xpos & 0xFFFF) ^ 0xFFFF) >> 9;
780 filter[i + 1] = (((xpos + xInc) & 0xFFFF) ^ 0xFFFF) >> 9;
781 filter[i + 2] = (((xpos + xInc * 2) & 0xFFFF) ^ 0xFFFF) >> 9;
782 filter[i + 3] = (((xpos + xInc * 3) & 0xFFFF) ^ 0xFFFF) >> 9;
783 filterPos[i / 2] = xx;
785 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
787 filterCode[fragmentPos + imm8OfPShufW1] = (a + inc) |
791 filterCode[fragmentPos + imm8OfPShufW2] = a | (b << 2) |
795 if (i + 4 - inc >= dstW)
796 shift = maxShift; // avoid overread
797 else if ((filterPos[i / 2] & 3) <= maxShift)
798 shift = filterPos[i / 2] & 3; // align
800 if (shift && i >= shift) {
801 filterCode[fragmentPos + imm8OfPShufW1] += 0x55 * shift;
802 filterCode[fragmentPos + imm8OfPShufW2] += 0x55 * shift;
803 filterPos[i / 2] -= shift;
807 fragmentPos += fragmentLength;
810 filterCode[fragmentPos] = RET;
815 filterPos[((i / 2) + 1) & (~1)] = xpos >> 16; // needed to jump to the next part
817 return fragmentPos + 1;
819 #endif /* HAVE_MMXEXT_INLINE */
821 static void getSubSampleFactors(int *h, int *v, enum AVPixelFormat format)
823 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(format);
824 *h = desc->log2_chroma_w;
825 *v = desc->log2_chroma_h;
828 static void fill_rgb2yuv_table(SwsContext *c, const int table[4], int dstRange)
830 int64_t W, V, Z, Cy, Cu, Cv;
831 int64_t vr = table[0];
832 int64_t ub = table[1];
833 int64_t ug = -table[2];
834 int64_t vg = -table[3];
837 uint8_t *p = (uint8_t*)c->input_rgb2yuv_table;
839 static const int8_t map[] = {
840 BY_IDX, GY_IDX, -1 , BY_IDX, BY_IDX, GY_IDX, -1 , BY_IDX,
841 RY_IDX, -1 , GY_IDX, RY_IDX, RY_IDX, -1 , GY_IDX, RY_IDX,
842 RY_IDX, GY_IDX, -1 , RY_IDX, RY_IDX, GY_IDX, -1 , RY_IDX,
843 BY_IDX, -1 , GY_IDX, BY_IDX, BY_IDX, -1 , GY_IDX, BY_IDX,
844 BU_IDX, GU_IDX, -1 , BU_IDX, BU_IDX, GU_IDX, -1 , BU_IDX,
845 RU_IDX, -1 , GU_IDX, RU_IDX, RU_IDX, -1 , GU_IDX, RU_IDX,
846 RU_IDX, GU_IDX, -1 , RU_IDX, RU_IDX, GU_IDX, -1 , RU_IDX,
847 BU_IDX, -1 , GU_IDX, BU_IDX, BU_IDX, -1 , GU_IDX, BU_IDX,
848 BV_IDX, GV_IDX, -1 , BV_IDX, BV_IDX, GV_IDX, -1 , BV_IDX,
849 RV_IDX, -1 , GV_IDX, RV_IDX, RV_IDX, -1 , GV_IDX, RV_IDX,
850 RV_IDX, GV_IDX, -1 , RV_IDX, RV_IDX, GV_IDX, -1 , RV_IDX,
851 BV_IDX, -1 , GV_IDX, BV_IDX, BV_IDX, -1 , GV_IDX, BV_IDX,
852 RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX,
853 BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX,
854 GY_IDX, -1 , GY_IDX, -1 , GY_IDX, -1 , GY_IDX, -1 ,
855 -1 , GY_IDX, -1 , GY_IDX, -1 , GY_IDX, -1 , GY_IDX,
856 RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX,
857 BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX,
858 GU_IDX, -1 , GU_IDX, -1 , GU_IDX, -1 , GU_IDX, -1 ,
859 -1 , GU_IDX, -1 , GU_IDX, -1 , GU_IDX, -1 , GU_IDX,
860 RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX,
861 BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX,
862 GV_IDX, -1 , GV_IDX, -1 , GV_IDX, -1 , GV_IDX, -1 ,
863 -1 , GV_IDX, -1 , GV_IDX, -1 , GV_IDX, -1 , GV_IDX, //23
864 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //24
865 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //25
866 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //26
867 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //27
868 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //28
869 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //29
870 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //30
871 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //31
872 BY_IDX, GY_IDX, RY_IDX, -1 , -1 , -1 , -1 , -1 , //32
873 BU_IDX, GU_IDX, RU_IDX, -1 , -1 , -1 , -1 , -1 , //33
874 BV_IDX, GV_IDX, RV_IDX, -1 , -1 , -1 , -1 , -1 , //34
877 dstRange = 0; //FIXME range = 1 is handled elsewhere
887 W = ROUNDED_DIV(ONE*ONE*ug, ub);
888 V = ROUNDED_DIV(ONE*ONE*vg, vr);
891 Cy = ROUNDED_DIV(cy*Z, ONE);
892 Cu = ROUNDED_DIV(ub*Z, ONE);
893 Cv = ROUNDED_DIV(vr*Z, ONE);
895 c->input_rgb2yuv_table[RY_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*V , Cy);
896 c->input_rgb2yuv_table[GY_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE , Cy);
897 c->input_rgb2yuv_table[BY_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*W , Cy);
899 c->input_rgb2yuv_table[RU_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*V , Cu);
900 c->input_rgb2yuv_table[GU_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE , Cu);
901 c->input_rgb2yuv_table[BU_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*(Z+W) , Cu);
903 c->input_rgb2yuv_table[RV_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*(V+Z) , Cv);
904 c->input_rgb2yuv_table[GV_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE , Cv);
905 c->input_rgb2yuv_table[BV_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*W , Cv);
907 if(/*!dstRange && */table == ff_yuv2rgb_coeffs[SWS_CS_DEFAULT]) {
908 c->input_rgb2yuv_table[BY_IDX] = ((int)(0.114 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
909 c->input_rgb2yuv_table[BV_IDX] = (-(int)(0.081 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
910 c->input_rgb2yuv_table[BU_IDX] = ((int)(0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
911 c->input_rgb2yuv_table[GY_IDX] = ((int)(0.587 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
912 c->input_rgb2yuv_table[GV_IDX] = (-(int)(0.419 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
913 c->input_rgb2yuv_table[GU_IDX] = (-(int)(0.331 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
914 c->input_rgb2yuv_table[RY_IDX] = ((int)(0.299 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
915 c->input_rgb2yuv_table[RV_IDX] = ((int)(0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
916 c->input_rgb2yuv_table[RU_IDX] = (-(int)(0.169 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
918 for(i=0; i<FF_ARRAY_ELEMS(map); i++)
919 AV_WL16(p + 16*4 + 2*i, map[i] >= 0 ? c->input_rgb2yuv_table[map[i]] : 0);
922 static void fill_xyztables(struct SwsContext *c)
925 double xyzgamma = XYZ_GAMMA;
926 double rgbgamma = 1.0 / RGB_GAMMA;
927 static const int16_t xyz2rgb_matrix[3][4] = {
928 {13270, -6295, -2041},
930 { 228, -835, 4329} };
931 static int16_t xyzgamma_tab[4096], rgbgamma_tab[4096];
933 memcpy(c->xyz2rgb_matrix, xyz2rgb_matrix, sizeof(c->xyz2rgb_matrix));
934 c->xyzgamma = xyzgamma_tab;
935 c->rgbgamma = rgbgamma_tab;
937 if (rgbgamma_tab[4095])
940 /* set gamma vectors */
941 for (i = 0; i < 4096; i++) {
942 xyzgamma_tab[i] = lrint(pow(i / 4095.0, xyzgamma) * 4095.0);
943 rgbgamma_tab[i] = lrint(pow(i / 4095.0, rgbgamma) * 4095.0);
947 int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],
948 int srcRange, const int table[4], int dstRange,
949 int brightness, int contrast, int saturation)
951 const AVPixFmtDescriptor *desc_dst;
952 const AVPixFmtDescriptor *desc_src;
953 memcpy(c->srcColorspaceTable, inv_table, sizeof(int) * 4);
954 memcpy(c->dstColorspaceTable, table, sizeof(int) * 4);
957 desc_dst = av_pix_fmt_desc_get(c->dstFormat);
958 desc_src = av_pix_fmt_desc_get(c->srcFormat);
960 if(!isYUV(c->dstFormat) && !isGray(c->dstFormat))
962 if(!isYUV(c->srcFormat) && !isGray(c->srcFormat))
965 c->brightness = brightness;
966 c->contrast = contrast;
967 c->saturation = saturation;
968 c->srcRange = srcRange;
969 c->dstRange = dstRange;
973 if ((isYUV(c->dstFormat) || isGray(c->dstFormat)) && (isYUV(c->srcFormat) || isGray(c->srcFormat)))
976 c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);
977 c->srcFormatBpp = av_get_bits_per_pixel(desc_src);
979 if (!isYUV(c->dstFormat) && !isGray(c->dstFormat)) {
980 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness,
981 contrast, saturation);
984 if (HAVE_ALTIVEC && av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)
985 ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness,
986 contrast, saturation);
989 fill_rgb2yuv_table(c, table, dstRange);
994 int sws_getColorspaceDetails(struct SwsContext *c, int **inv_table,
995 int *srcRange, int **table, int *dstRange,
996 int *brightness, int *contrast, int *saturation)
998 if (!c || isYUV(c->dstFormat) || isGray(c->dstFormat))
1001 *inv_table = c->srcColorspaceTable;
1002 *table = c->dstColorspaceTable;
1003 *srcRange = c->srcRange;
1004 *dstRange = c->dstRange;
1005 *brightness = c->brightness;
1006 *contrast = c->contrast;
1007 *saturation = c->saturation;
1012 static int handle_jpeg(enum AVPixelFormat *format)
1015 case AV_PIX_FMT_YUVJ420P:
1016 *format = AV_PIX_FMT_YUV420P;
1018 case AV_PIX_FMT_YUVJ411P:
1019 *format = AV_PIX_FMT_YUV411P;
1021 case AV_PIX_FMT_YUVJ422P:
1022 *format = AV_PIX_FMT_YUV422P;
1024 case AV_PIX_FMT_YUVJ444P:
1025 *format = AV_PIX_FMT_YUV444P;
1027 case AV_PIX_FMT_YUVJ440P:
1028 *format = AV_PIX_FMT_YUV440P;
1030 case AV_PIX_FMT_GRAY8:
1037 static int handle_0alpha(enum AVPixelFormat *format)
1040 case AV_PIX_FMT_0BGR : *format = AV_PIX_FMT_ABGR ; return 1;
1041 case AV_PIX_FMT_BGR0 : *format = AV_PIX_FMT_BGRA ; return 4;
1042 case AV_PIX_FMT_0RGB : *format = AV_PIX_FMT_ARGB ; return 1;
1043 case AV_PIX_FMT_RGB0 : *format = AV_PIX_FMT_RGBA ; return 4;
1048 static int handle_xyz(enum AVPixelFormat *format)
1051 case AV_PIX_FMT_XYZ12BE : *format = AV_PIX_FMT_RGB48BE; return 1;
1052 case AV_PIX_FMT_XYZ12LE : *format = AV_PIX_FMT_RGB48LE; return 1;
1057 static void handle_formats(SwsContext *c)
1059 c->src0Alpha |= handle_0alpha(&c->srcFormat);
1060 c->dst0Alpha |= handle_0alpha(&c->dstFormat);
1061 c->srcXYZ |= handle_xyz(&c->srcFormat);
1062 c->dstXYZ |= handle_xyz(&c->dstFormat);
1065 SwsContext *sws_alloc_context(void)
1067 SwsContext *c = av_mallocz(sizeof(SwsContext));
1070 c->av_class = &sws_context_class;
1071 av_opt_set_defaults(c);
1077 av_cold int sws_init_context(SwsContext *c, SwsFilter *srcFilter,
1078 SwsFilter *dstFilter)
1081 int usesVFilter, usesHFilter;
1083 SwsFilter dummyFilter = { NULL, NULL, NULL, NULL };
1088 int dst_stride = FFALIGN(dstW * sizeof(int16_t) + 66, 16);
1089 int flags, cpu_flags;
1090 enum AVPixelFormat srcFormat = c->srcFormat;
1091 enum AVPixelFormat dstFormat = c->dstFormat;
1092 const AVPixFmtDescriptor *desc_src;
1093 const AVPixFmtDescriptor *desc_dst;
1095 cpu_flags = av_get_cpu_flags();
1101 unscaled = (srcW == dstW && srcH == dstH);
1103 c->srcRange |= handle_jpeg(&c->srcFormat);
1104 c->dstRange |= handle_jpeg(&c->dstFormat);
1106 if (!c->contrast && !c->saturation && !c->dstFormatBpp)
1107 sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,
1108 ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],
1109 c->dstRange, 0, 1 << 16, 1 << 16);
1111 if(srcFormat!=c->srcFormat || dstFormat!=c->dstFormat)
1112 av_log(c, AV_LOG_WARNING, "deprecated pixel format used, make sure you did set range correctly\n");
1114 srcFormat = c->srcFormat;
1115 dstFormat = c->dstFormat;
1116 desc_src = av_pix_fmt_desc_get(srcFormat);
1117 desc_dst = av_pix_fmt_desc_get(dstFormat);
1119 if (!(unscaled && sws_isSupportedEndiannessConversion(srcFormat) &&
1120 av_pix_fmt_swap_endianness(srcFormat) == dstFormat)) {
1121 if (!sws_isSupportedInput(srcFormat)) {
1122 av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n",
1123 av_get_pix_fmt_name(srcFormat));
1124 return AVERROR(EINVAL);
1126 if (!sws_isSupportedOutput(dstFormat)) {
1127 av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n",
1128 av_get_pix_fmt_name(dstFormat));
1129 return AVERROR(EINVAL);
1133 i = flags & (SWS_POINT |
1144 if (!i || (i & (i - 1))) {
1145 av_log(c, AV_LOG_ERROR, "Exactly one scaler algorithm must be chosen, got %X\n", i);
1146 return AVERROR(EINVAL);
1149 if (srcW < 1 || srcH < 1 || dstW < 1 || dstH < 1) {
1150 /* FIXME check if these are enough and try to lower them after
1151 * fixing the relevant parts of the code */
1152 av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",
1153 srcW, srcH, dstW, dstH);
1154 return AVERROR(EINVAL);
1158 dstFilter = &dummyFilter;
1160 srcFilter = &dummyFilter;
1162 c->lumXInc = (((int64_t)srcW << 16) + (dstW >> 1)) / dstW;
1163 c->lumYInc = (((int64_t)srcH << 16) + (dstH >> 1)) / dstH;
1164 c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);
1165 c->srcFormatBpp = av_get_bits_per_pixel(desc_src);
1166 c->vRounder = 4 * 0x0001000100010001ULL;
1168 usesVFilter = (srcFilter->lumV && srcFilter->lumV->length > 1) ||
1169 (srcFilter->chrV && srcFilter->chrV->length > 1) ||
1170 (dstFilter->lumV && dstFilter->lumV->length > 1) ||
1171 (dstFilter->chrV && dstFilter->chrV->length > 1);
1172 usesHFilter = (srcFilter->lumH && srcFilter->lumH->length > 1) ||
1173 (srcFilter->chrH && srcFilter->chrH->length > 1) ||
1174 (dstFilter->lumH && dstFilter->lumH->length > 1) ||
1175 (dstFilter->chrH && dstFilter->chrH->length > 1);
1177 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
1178 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
1180 if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) {
1182 av_log(c, AV_LOG_DEBUG, "Forcing full internal H chroma due to odd output size\n");
1183 flags |= SWS_FULL_CHR_H_INT;
1188 if(dstFormat == AV_PIX_FMT_BGR4_BYTE ||
1189 dstFormat == AV_PIX_FMT_RGB4_BYTE ||
1190 dstFormat == AV_PIX_FMT_BGR8 ||
1191 dstFormat == AV_PIX_FMT_RGB8) {
1192 if (flags & SWS_ERROR_DIFFUSION && !(flags & SWS_FULL_CHR_H_INT)) {
1193 av_log(c, AV_LOG_DEBUG,
1194 "Error diffusion dither is only supported in full chroma interpolation for destination format '%s'\n",
1195 av_get_pix_fmt_name(dstFormat));
1196 flags |= SWS_FULL_CHR_H_INT;
1199 if (!(flags & SWS_ERROR_DIFFUSION) && (flags & SWS_FULL_CHR_H_INT)) {
1200 av_log(c, AV_LOG_DEBUG,
1201 "Ordered dither is not supported in full chroma interpolation for destination format '%s'\n",
1202 av_get_pix_fmt_name(dstFormat));
1203 flags |= SWS_ERROR_DIFFUSION;
1207 if (isPlanarRGB(dstFormat)) {
1208 if (!(flags & SWS_FULL_CHR_H_INT)) {
1209 av_log(c, AV_LOG_DEBUG,
1210 "%s output is not supported with half chroma resolution, switching to full\n",
1211 av_get_pix_fmt_name(dstFormat));
1212 flags |= SWS_FULL_CHR_H_INT;
1217 /* reuse chroma for 2 pixels RGB/BGR unless user wants full
1218 * chroma interpolation */
1219 if (flags & SWS_FULL_CHR_H_INT &&
1220 isAnyRGB(dstFormat) &&
1221 !isPlanarRGB(dstFormat) &&
1222 dstFormat != AV_PIX_FMT_RGBA &&
1223 dstFormat != AV_PIX_FMT_ARGB &&
1224 dstFormat != AV_PIX_FMT_BGRA &&
1225 dstFormat != AV_PIX_FMT_ABGR &&
1226 dstFormat != AV_PIX_FMT_RGB24 &&
1227 dstFormat != AV_PIX_FMT_BGR24 &&
1228 dstFormat != AV_PIX_FMT_BGR4_BYTE &&
1229 dstFormat != AV_PIX_FMT_RGB4_BYTE &&
1230 dstFormat != AV_PIX_FMT_BGR8 &&
1231 dstFormat != AV_PIX_FMT_RGB8
1233 av_log(c, AV_LOG_WARNING,
1234 "full chroma interpolation for destination format '%s' not yet implemented\n",
1235 av_get_pix_fmt_name(dstFormat));
1236 flags &= ~SWS_FULL_CHR_H_INT;
1239 if (isAnyRGB(dstFormat) && !(flags & SWS_FULL_CHR_H_INT))
1240 c->chrDstHSubSample = 1;
1242 // drop some chroma lines if the user wants it
1243 c->vChrDrop = (flags & SWS_SRC_V_CHR_DROP_MASK) >>
1244 SWS_SRC_V_CHR_DROP_SHIFT;
1245 c->chrSrcVSubSample += c->vChrDrop;
1247 /* drop every other pixel for chroma calculation unless user
1248 * wants full chroma */
1249 if (isAnyRGB(srcFormat) && !(flags & SWS_FULL_CHR_H_INP) &&
1250 srcFormat != AV_PIX_FMT_RGB8 && srcFormat != AV_PIX_FMT_BGR8 &&
1251 srcFormat != AV_PIX_FMT_RGB4 && srcFormat != AV_PIX_FMT_BGR4 &&
1252 srcFormat != AV_PIX_FMT_RGB4_BYTE && srcFormat != AV_PIX_FMT_BGR4_BYTE &&
1253 srcFormat != AV_PIX_FMT_GBRP9BE && srcFormat != AV_PIX_FMT_GBRP9LE &&
1254 srcFormat != AV_PIX_FMT_GBRP10BE && srcFormat != AV_PIX_FMT_GBRP10LE &&
1255 srcFormat != AV_PIX_FMT_GBRP12BE && srcFormat != AV_PIX_FMT_GBRP12LE &&
1256 srcFormat != AV_PIX_FMT_GBRP14BE && srcFormat != AV_PIX_FMT_GBRP14LE &&
1257 srcFormat != AV_PIX_FMT_GBRP16BE && srcFormat != AV_PIX_FMT_GBRP16LE &&
1258 ((dstW >> c->chrDstHSubSample) <= (srcW >> 1) ||
1259 (flags & SWS_FAST_BILINEAR)))
1260 c->chrSrcHSubSample = 1;
1262 // Note the FF_CEIL_RSHIFT is so that we always round toward +inf.
1263 c->chrSrcW = FF_CEIL_RSHIFT(srcW, c->chrSrcHSubSample);
1264 c->chrSrcH = FF_CEIL_RSHIFT(srcH, c->chrSrcVSubSample);
1265 c->chrDstW = FF_CEIL_RSHIFT(dstW, c->chrDstHSubSample);
1266 c->chrDstH = FF_CEIL_RSHIFT(dstH, c->chrDstVSubSample);
1268 FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW*2+78, 16) * 2, fail);
1270 /* unscaled special cases */
1271 if (unscaled && !usesHFilter && !usesVFilter &&
1272 (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
1273 ff_get_unscaled_swscale(c);
1276 if (flags & SWS_PRINT_INFO)
1277 av_log(c, AV_LOG_INFO,
1278 "using unscaled %s -> %s special converter\n",
1279 av_get_pix_fmt_name(srcFormat), av_get_pix_fmt_name(dstFormat));
1284 c->srcBpc = 1 + desc_src->comp[0].depth_minus1;
1287 c->dstBpc = 1 + desc_dst->comp[0].depth_minus1;
1290 if (isAnyRGB(srcFormat) || srcFormat == AV_PIX_FMT_PAL8)
1292 if (c->dstBpc == 16)
1295 if (INLINE_MMXEXT(cpu_flags) && c->srcBpc == 8 && c->dstBpc <= 14) {
1296 c->canMMXEXTBeUsed = (dstW >= srcW && (dstW & 31) == 0 &&
1297 (srcW & 15) == 0) ? 1 : 0;
1298 if (!c->canMMXEXTBeUsed && dstW >= srcW && (srcW & 15) == 0
1300 && (flags & SWS_FAST_BILINEAR)) {
1301 if (flags & SWS_PRINT_INFO)
1302 av_log(c, AV_LOG_INFO,
1303 "output width is not a multiple of 32 -> no MMXEXT scaler\n");
1305 if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat))
1306 c->canMMXEXTBeUsed = 0;
1308 c->canMMXEXTBeUsed = 0;
1310 c->chrXInc = (((int64_t)c->chrSrcW << 16) + (c->chrDstW >> 1)) / c->chrDstW;
1311 c->chrYInc = (((int64_t)c->chrSrcH << 16) + (c->chrDstH >> 1)) / c->chrDstH;
1313 /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src
1314 * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do
1316 * n-2 is the last chrominance sample available.
1317 * This is not perfect, but no one should notice the difference, the more
1318 * correct variant would be like the vertical one, but that would require
1319 * some special code for the first and last pixel */
1320 if (flags & SWS_FAST_BILINEAR) {
1321 if (c->canMMXEXTBeUsed) {
1325 // we don't use the x86 asm scaler if MMX is available
1326 else if (INLINE_MMX(cpu_flags) && c->dstBpc <= 14) {
1327 c->lumXInc = ((int64_t)(srcW - 2) << 16) / (dstW - 2) - 20;
1328 c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;
1332 #define USE_MMAP (HAVE_MMAP && HAVE_MPROTECT && defined MAP_ANONYMOUS)
1334 /* precalculate horizontal scaler filter coefficients */
1336 #if HAVE_MMXEXT_INLINE
1337 // can't downscale !!!
1338 if (c->canMMXEXTBeUsed && (flags & SWS_FAST_BILINEAR)) {
1339 c->lumMmxextFilterCodeSize = init_hscaler_mmxext(dstW, c->lumXInc, NULL,
1341 c->chrMmxextFilterCodeSize = init_hscaler_mmxext(c->chrDstW, c->chrXInc,
1342 NULL, NULL, NULL, 4);
1345 c->lumMmxextFilterCode = mmap(NULL, c->lumMmxextFilterCodeSize,
1346 PROT_READ | PROT_WRITE,
1347 MAP_PRIVATE | MAP_ANONYMOUS,
1349 c->chrMmxextFilterCode = mmap(NULL, c->chrMmxextFilterCodeSize,
1350 PROT_READ | PROT_WRITE,
1351 MAP_PRIVATE | MAP_ANONYMOUS,
1353 #elif HAVE_VIRTUALALLOC
1354 c->lumMmxextFilterCode = VirtualAlloc(NULL,
1355 c->lumMmxextFilterCodeSize,
1357 PAGE_EXECUTE_READWRITE);
1358 c->chrMmxextFilterCode = VirtualAlloc(NULL,
1359 c->chrMmxextFilterCodeSize,
1361 PAGE_EXECUTE_READWRITE);
1363 c->lumMmxextFilterCode = av_malloc(c->lumMmxextFilterCodeSize);
1364 c->chrMmxextFilterCode = av_malloc(c->chrMmxextFilterCodeSize);
1367 #ifdef MAP_ANONYMOUS
1368 if (c->lumMmxextFilterCode == MAP_FAILED || c->chrMmxextFilterCode == MAP_FAILED)
1370 if (!c->lumMmxextFilterCode || !c->chrMmxextFilterCode)
1373 av_log(c, AV_LOG_ERROR, "Failed to allocate MMX2FilterCode\n");
1374 return AVERROR(ENOMEM);
1377 FF_ALLOCZ_OR_GOTO(c, c->hLumFilter, (dstW / 8 + 8) * sizeof(int16_t), fail);
1378 FF_ALLOCZ_OR_GOTO(c, c->hChrFilter, (c->chrDstW / 4 + 8) * sizeof(int16_t), fail);
1379 FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW / 2 / 8 + 8) * sizeof(int32_t), fail);
1380 FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);
1382 init_hscaler_mmxext( dstW, c->lumXInc, c->lumMmxextFilterCode,
1383 c->hLumFilter, (uint32_t*)c->hLumFilterPos, 8);
1384 init_hscaler_mmxext(c->chrDstW, c->chrXInc, c->chrMmxextFilterCode,
1385 c->hChrFilter, (uint32_t*)c->hChrFilterPos, 4);
1388 mprotect(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize, PROT_EXEC | PROT_READ);
1389 mprotect(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize, PROT_EXEC | PROT_READ);
1392 #endif /* HAVE_MMXEXT_INLINE */
1394 const int filterAlign =
1395 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :
1396 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
1399 if (initFilter(&c->hLumFilter, &c->hLumFilterPos,
1400 &c->hLumFilterSize, c->lumXInc,
1401 srcW, dstW, filterAlign, 1 << 14,
1402 (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
1403 cpu_flags, srcFilter->lumH, dstFilter->lumH,
1405 get_local_pos(c, 0, 0, 0),
1406 get_local_pos(c, 0, 0, 0)) < 0)
1408 if (initFilter(&c->hChrFilter, &c->hChrFilterPos,
1409 &c->hChrFilterSize, c->chrXInc,
1410 c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,
1411 (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
1412 cpu_flags, srcFilter->chrH, dstFilter->chrH,
1414 get_local_pos(c, c->chrSrcHSubSample, c->src_h_chr_pos, 0),
1415 get_local_pos(c, c->chrDstHSubSample, c->dst_h_chr_pos, 0)) < 0)
1418 } // initialize horizontal stuff
1420 /* precalculate vertical scaler filter coefficients */
1422 const int filterAlign =
1423 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 2 :
1424 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
1427 if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,
1428 c->lumYInc, srcH, dstH, filterAlign, (1 << 12),
1429 (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
1430 cpu_flags, srcFilter->lumV, dstFilter->lumV,
1432 get_local_pos(c, 0, 0, 1),
1433 get_local_pos(c, 0, 0, 1)) < 0)
1435 if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,
1436 c->chrYInc, c->chrSrcH, c->chrDstH,
1437 filterAlign, (1 << 12),
1438 (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
1439 cpu_flags, srcFilter->chrV, dstFilter->chrV,
1441 get_local_pos(c, c->chrSrcVSubSample, c->src_v_chr_pos, 1),
1442 get_local_pos(c, c->chrDstVSubSample, c->dst_v_chr_pos, 1)) < 0)
1447 FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH, fail);
1448 FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);
1450 for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {
1452 short *p = (short *)&c->vYCoeffsBank[i];
1453 for (j = 0; j < 8; j++)
1454 p[j] = c->vLumFilter[i];
1457 for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {
1459 short *p = (short *)&c->vCCoeffsBank[i];
1460 for (j = 0; j < 8; j++)
1461 p[j] = c->vChrFilter[i];
1466 // calculate buffer sizes so that they won't run out while handling these damn slices
1467 c->vLumBufSize = c->vLumFilterSize;
1468 c->vChrBufSize = c->vChrFilterSize;
1469 for (i = 0; i < dstH; i++) {
1470 int chrI = (int64_t)i * c->chrDstH / dstH;
1471 int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,
1472 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)
1473 << c->chrSrcVSubSample));
1475 nextSlice >>= c->chrSrcVSubSample;
1476 nextSlice <<= c->chrSrcVSubSample;
1477 if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)
1478 c->vLumBufSize = nextSlice - c->vLumFilterPos[i];
1479 if (c->vChrFilterPos[chrI] + c->vChrBufSize <
1480 (nextSlice >> c->chrSrcVSubSample))
1481 c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -
1482 c->vChrFilterPos[chrI];
1485 for (i = 0; i < 4; i++)
1486 FF_ALLOCZ_OR_GOTO(c, c->dither_error[i], (c->dstW+2) * sizeof(int), fail);
1488 /* Allocate pixbufs (we use dynamic allocation because otherwise we would
1489 * need to allocate several megabytes to handle all possible cases) */
1490 FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);
1491 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
1492 FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
1493 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
1494 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);
1495 /* Note we need at least one pixel more at the end because of the MMX code
1496 * (just in case someone wants to replace the 4000/8000). */
1497 /* align at 16 bytes for AltiVec */
1498 for (i = 0; i < c->vLumBufSize; i++) {
1499 FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],
1500 dst_stride + 16, fail);
1501 c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];
1503 // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)
1504 c->uv_off = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);
1505 c->uv_offx2 = dst_stride + 16;
1506 for (i = 0; i < c->vChrBufSize; i++) {
1507 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],
1508 dst_stride * 2 + 32, fail);
1509 c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];
1510 c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]
1511 = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;
1513 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
1514 for (i = 0; i < c->vLumBufSize; i++) {
1515 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i + c->vLumBufSize],
1516 dst_stride + 16, fail);
1517 c->alpPixBuf[i] = c->alpPixBuf[i + c->vLumBufSize];
1520 // try to avoid drawing green stuff between the right end and the stride end
1521 for (i = 0; i < c->vChrBufSize; i++)
1522 if(desc_dst->comp[0].depth_minus1 == 15){
1523 av_assert0(c->dstBpc > 14);
1524 for(j=0; j<dst_stride/2+1; j++)
1525 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;
1527 for(j=0; j<dst_stride+1; j++)
1528 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;
1530 av_assert0(c->chrDstH <= dstH);
1532 if (flags & SWS_PRINT_INFO) {
1533 if (flags & SWS_FAST_BILINEAR)
1534 av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
1535 else if (flags & SWS_BILINEAR)
1536 av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
1537 else if (flags & SWS_BICUBIC)
1538 av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
1539 else if (flags & SWS_X)
1540 av_log(c, AV_LOG_INFO, "Experimental scaler, ");
1541 else if (flags & SWS_POINT)
1542 av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
1543 else if (flags & SWS_AREA)
1544 av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
1545 else if (flags & SWS_BICUBLIN)
1546 av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
1547 else if (flags & SWS_GAUSS)
1548 av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
1549 else if (flags & SWS_SINC)
1550 av_log(c, AV_LOG_INFO, "Sinc scaler, ");
1551 else if (flags & SWS_LANCZOS)
1552 av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
1553 else if (flags & SWS_SPLINE)
1554 av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
1556 av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
1558 av_log(c, AV_LOG_INFO, "from %s to %s%s ",
1559 av_get_pix_fmt_name(srcFormat),
1561 dstFormat == AV_PIX_FMT_BGR555 || dstFormat == AV_PIX_FMT_BGR565 ||
1562 dstFormat == AV_PIX_FMT_RGB444BE || dstFormat == AV_PIX_FMT_RGB444LE ||
1563 dstFormat == AV_PIX_FMT_BGR444BE || dstFormat == AV_PIX_FMT_BGR444LE ?
1568 av_get_pix_fmt_name(dstFormat));
1570 if (INLINE_MMXEXT(cpu_flags))
1571 av_log(c, AV_LOG_INFO, "using MMXEXT\n");
1572 else if (INLINE_AMD3DNOW(cpu_flags))
1573 av_log(c, AV_LOG_INFO, "using 3DNOW\n");
1574 else if (INLINE_MMX(cpu_flags))
1575 av_log(c, AV_LOG_INFO, "using MMX\n");
1576 else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC)
1577 av_log(c, AV_LOG_INFO, "using AltiVec\n");
1579 av_log(c, AV_LOG_INFO, "using C\n");
1581 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
1582 av_log(c, AV_LOG_DEBUG,
1583 "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1584 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
1585 av_log(c, AV_LOG_DEBUG,
1586 "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1587 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,
1588 c->chrXInc, c->chrYInc);
1591 c->swScale = ff_getSwsFunc(c);
1593 fail: // FIXME replace things by appropriate error codes
1597 #if FF_API_SWS_GETCONTEXT
1598 SwsContext *sws_getContext(int srcW, int srcH, enum AVPixelFormat srcFormat,
1599 int dstW, int dstH, enum AVPixelFormat dstFormat,
1600 int flags, SwsFilter *srcFilter,
1601 SwsFilter *dstFilter, const double *param)
1605 if (!(c = sws_alloc_context()))
1613 c->srcFormat = srcFormat;
1614 c->dstFormat = dstFormat;
1617 c->param[0] = param[0];
1618 c->param[1] = param[1];
1621 if (sws_init_context(c, srcFilter, dstFilter) < 0) {
1630 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
1631 float lumaSharpen, float chromaSharpen,
1632 float chromaHShift, float chromaVShift,
1635 SwsFilter *filter = av_malloc(sizeof(SwsFilter));
1639 if (lumaGBlur != 0.0) {
1640 filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);
1641 filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);
1643 filter->lumH = sws_getIdentityVec();
1644 filter->lumV = sws_getIdentityVec();
1647 if (chromaGBlur != 0.0) {
1648 filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);
1649 filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);
1651 filter->chrH = sws_getIdentityVec();
1652 filter->chrV = sws_getIdentityVec();
1655 if (chromaSharpen != 0.0) {
1656 SwsVector *id = sws_getIdentityVec();
1657 sws_scaleVec(filter->chrH, -chromaSharpen);
1658 sws_scaleVec(filter->chrV, -chromaSharpen);
1659 sws_addVec(filter->chrH, id);
1660 sws_addVec(filter->chrV, id);
1664 if (lumaSharpen != 0.0) {
1665 SwsVector *id = sws_getIdentityVec();
1666 sws_scaleVec(filter->lumH, -lumaSharpen);
1667 sws_scaleVec(filter->lumV, -lumaSharpen);
1668 sws_addVec(filter->lumH, id);
1669 sws_addVec(filter->lumV, id);
1673 if (chromaHShift != 0.0)
1674 sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));
1676 if (chromaVShift != 0.0)
1677 sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));
1679 sws_normalizeVec(filter->chrH, 1.0);
1680 sws_normalizeVec(filter->chrV, 1.0);
1681 sws_normalizeVec(filter->lumH, 1.0);
1682 sws_normalizeVec(filter->lumV, 1.0);
1685 sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
1687 sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
1692 SwsVector *sws_allocVec(int length)
1696 if(length <= 0 || length > INT_MAX/ sizeof(double))
1699 vec = av_malloc(sizeof(SwsVector));
1702 vec->length = length;
1703 vec->coeff = av_malloc(sizeof(double) * length);
1709 SwsVector *sws_getGaussianVec(double variance, double quality)
1711 const int length = (int)(variance * quality + 0.5) | 1;
1713 double middle = (length - 1) * 0.5;
1716 if(variance < 0 || quality < 0)
1719 vec = sws_allocVec(length);
1724 for (i = 0; i < length; i++) {
1725 double dist = i - middle;
1726 vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /
1727 sqrt(2 * variance * M_PI);
1730 sws_normalizeVec(vec, 1.0);
1735 SwsVector *sws_getConstVec(double c, int length)
1738 SwsVector *vec = sws_allocVec(length);
1743 for (i = 0; i < length; i++)
1749 SwsVector *sws_getIdentityVec(void)
1751 return sws_getConstVec(1.0, 1);
1754 static double sws_dcVec(SwsVector *a)
1759 for (i = 0; i < a->length; i++)
1765 void sws_scaleVec(SwsVector *a, double scalar)
1769 for (i = 0; i < a->length; i++)
1770 a->coeff[i] *= scalar;
1773 void sws_normalizeVec(SwsVector *a, double height)
1775 sws_scaleVec(a, height / sws_dcVec(a));
1778 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
1780 int length = a->length + b->length - 1;
1782 SwsVector *vec = sws_getConstVec(0.0, length);
1787 for (i = 0; i < a->length; i++) {
1788 for (j = 0; j < b->length; j++) {
1789 vec->coeff[i + j] += a->coeff[i] * b->coeff[j];
1796 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
1798 int length = FFMAX(a->length, b->length);
1800 SwsVector *vec = sws_getConstVec(0.0, length);
1805 for (i = 0; i < a->length; i++)
1806 vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
1807 for (i = 0; i < b->length; i++)
1808 vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] += b->coeff[i];
1813 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
1815 int length = FFMAX(a->length, b->length);
1817 SwsVector *vec = sws_getConstVec(0.0, length);
1822 for (i = 0; i < a->length; i++)
1823 vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
1824 for (i = 0; i < b->length; i++)
1825 vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] -= b->coeff[i];
1830 /* shift left / or right if "shift" is negative */
1831 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
1833 int length = a->length + FFABS(shift) * 2;
1835 SwsVector *vec = sws_getConstVec(0.0, length);
1840 for (i = 0; i < a->length; i++) {
1841 vec->coeff[i + (length - 1) / 2 -
1842 (a->length - 1) / 2 - shift] = a->coeff[i];
1848 void sws_shiftVec(SwsVector *a, int shift)
1850 SwsVector *shifted = sws_getShiftedVec(a, shift);
1852 a->coeff = shifted->coeff;
1853 a->length = shifted->length;
1857 void sws_addVec(SwsVector *a, SwsVector *b)
1859 SwsVector *sum = sws_sumVec(a, b);
1861 a->coeff = sum->coeff;
1862 a->length = sum->length;
1866 void sws_subVec(SwsVector *a, SwsVector *b)
1868 SwsVector *diff = sws_diffVec(a, b);
1870 a->coeff = diff->coeff;
1871 a->length = diff->length;
1875 void sws_convVec(SwsVector *a, SwsVector *b)
1877 SwsVector *conv = sws_getConvVec(a, b);
1879 a->coeff = conv->coeff;
1880 a->length = conv->length;
1884 SwsVector *sws_cloneVec(SwsVector *a)
1887 SwsVector *vec = sws_allocVec(a->length);
1892 for (i = 0; i < a->length; i++)
1893 vec->coeff[i] = a->coeff[i];
1898 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
1905 for (i = 0; i < a->length; i++)
1906 if (a->coeff[i] > max)
1909 for (i = 0; i < a->length; i++)
1910 if (a->coeff[i] < min)
1915 for (i = 0; i < a->length; i++) {
1916 int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);
1917 av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
1919 av_log(log_ctx, log_level, " ");
1920 av_log(log_ctx, log_level, "|\n");
1924 void sws_freeVec(SwsVector *a)
1928 av_freep(&a->coeff);
1933 void sws_freeFilter(SwsFilter *filter)
1939 sws_freeVec(filter->lumH);
1941 sws_freeVec(filter->lumV);
1943 sws_freeVec(filter->chrH);
1945 sws_freeVec(filter->chrV);
1949 void sws_freeContext(SwsContext *c)
1956 for (i = 0; i < c->vLumBufSize; i++)
1957 av_freep(&c->lumPixBuf[i]);
1958 av_freep(&c->lumPixBuf);
1961 if (c->chrUPixBuf) {
1962 for (i = 0; i < c->vChrBufSize; i++)
1963 av_freep(&c->chrUPixBuf[i]);
1964 av_freep(&c->chrUPixBuf);
1965 av_freep(&c->chrVPixBuf);
1968 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1969 for (i = 0; i < c->vLumBufSize; i++)
1970 av_freep(&c->alpPixBuf[i]);
1971 av_freep(&c->alpPixBuf);
1974 for (i = 0; i < 4; i++)
1975 av_freep(&c->dither_error[i]);
1977 av_freep(&c->vLumFilter);
1978 av_freep(&c->vChrFilter);
1979 av_freep(&c->hLumFilter);
1980 av_freep(&c->hChrFilter);
1982 av_freep(&c->vYCoeffsBank);
1983 av_freep(&c->vCCoeffsBank);
1986 av_freep(&c->vLumFilterPos);
1987 av_freep(&c->vChrFilterPos);
1988 av_freep(&c->hLumFilterPos);
1989 av_freep(&c->hChrFilterPos);
1993 if (c->lumMmxextFilterCode)
1994 munmap(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize);
1995 if (c->chrMmxextFilterCode)
1996 munmap(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize);
1997 #elif HAVE_VIRTUALALLOC
1998 if (c->lumMmxextFilterCode)
1999 VirtualFree(c->lumMmxextFilterCode, 0, MEM_RELEASE);
2000 if (c->chrMmxextFilterCode)
2001 VirtualFree(c->chrMmxextFilterCode, 0, MEM_RELEASE);
2003 av_free(c->lumMmxextFilterCode);
2004 av_free(c->chrMmxextFilterCode);
2006 c->lumMmxextFilterCode = NULL;
2007 c->chrMmxextFilterCode = NULL;
2008 #endif /* HAVE_MMX_INLINE */
2010 av_freep(&c->yuvTable);
2011 av_freep(&c->formatConvBuffer);
2016 struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,
2017 int srcH, enum AVPixelFormat srcFormat,
2019 enum AVPixelFormat dstFormat, int flags,
2020 SwsFilter *srcFilter,
2021 SwsFilter *dstFilter,
2022 const double *param)
2024 static const double default_param[2] = { SWS_PARAM_DEFAULT,
2025 SWS_PARAM_DEFAULT };
2028 param = default_param;
2031 (context->srcW != srcW ||
2032 context->srcH != srcH ||
2033 context->srcFormat != srcFormat ||
2034 context->dstW != dstW ||
2035 context->dstH != dstH ||
2036 context->dstFormat != dstFormat ||
2037 context->flags != flags ||
2038 context->param[0] != param[0] ||
2039 context->param[1] != param[1])) {
2040 sws_freeContext(context);
2045 if (!(context = sws_alloc_context()))
2047 context->srcW = srcW;
2048 context->srcH = srcH;
2049 context->srcFormat = srcFormat;
2050 context->dstW = dstW;
2051 context->dstH = dstH;
2052 context->dstFormat = dstFormat;
2053 context->flags = flags;
2054 context->param[0] = param[0];
2055 context->param[1] = param[1];
2056 if (sws_init_context(context, srcFilter, dstFilter) < 0) {
2057 sws_freeContext(context);