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/ppc/cpu.h"
50 #include "libavutil/x86/asm.h"
51 #include "libavutil/x86/cpu.h"
54 #include "swscale_internal.h"
56 static void handle_formats(SwsContext *c);
58 unsigned swscale_version(void)
60 av_assert0(LIBSWSCALE_VERSION_MICRO >= 100);
61 return LIBSWSCALE_VERSION_INT;
64 const char *swscale_configuration(void)
66 return FFMPEG_CONFIGURATION;
69 const char *swscale_license(void)
71 #define LICENSE_PREFIX "libswscale license: "
72 return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
75 #define RET 0xC3 // near return opcode for x86
77 typedef struct FormatEntry {
78 uint8_t is_supported_in :1;
79 uint8_t is_supported_out :1;
80 uint8_t is_supported_endianness :1;
83 static const FormatEntry format_entries[AV_PIX_FMT_NB] = {
84 [AV_PIX_FMT_YUV420P] = { 1, 1 },
85 [AV_PIX_FMT_YUYV422] = { 1, 1 },
86 [AV_PIX_FMT_RGB24] = { 1, 1 },
87 [AV_PIX_FMT_BGR24] = { 1, 1 },
88 [AV_PIX_FMT_YUV422P] = { 1, 1 },
89 [AV_PIX_FMT_YUV444P] = { 1, 1 },
90 [AV_PIX_FMT_YUV410P] = { 1, 1 },
91 [AV_PIX_FMT_YUV411P] = { 1, 1 },
92 [AV_PIX_FMT_GRAY8] = { 1, 1 },
93 [AV_PIX_FMT_MONOWHITE] = { 1, 1 },
94 [AV_PIX_FMT_MONOBLACK] = { 1, 1 },
95 [AV_PIX_FMT_PAL8] = { 1, 0 },
96 [AV_PIX_FMT_YUVJ420P] = { 1, 1 },
97 [AV_PIX_FMT_YUVJ411P] = { 1, 1 },
98 [AV_PIX_FMT_YUVJ422P] = { 1, 1 },
99 [AV_PIX_FMT_YUVJ444P] = { 1, 1 },
100 [AV_PIX_FMT_UYVY422] = { 1, 1 },
101 [AV_PIX_FMT_UYYVYY411] = { 0, 0 },
102 [AV_PIX_FMT_BGR8] = { 1, 1 },
103 [AV_PIX_FMT_BGR4] = { 0, 1 },
104 [AV_PIX_FMT_BGR4_BYTE] = { 1, 1 },
105 [AV_PIX_FMT_RGB8] = { 1, 1 },
106 [AV_PIX_FMT_RGB4] = { 0, 1 },
107 [AV_PIX_FMT_RGB4_BYTE] = { 1, 1 },
108 [AV_PIX_FMT_NV12] = { 1, 1 },
109 [AV_PIX_FMT_NV21] = { 1, 1 },
110 [AV_PIX_FMT_ARGB] = { 1, 1 },
111 [AV_PIX_FMT_RGBA] = { 1, 1 },
112 [AV_PIX_FMT_ABGR] = { 1, 1 },
113 [AV_PIX_FMT_BGRA] = { 1, 1 },
114 [AV_PIX_FMT_0RGB] = { 1, 1 },
115 [AV_PIX_FMT_RGB0] = { 1, 1 },
116 [AV_PIX_FMT_0BGR] = { 1, 1 },
117 [AV_PIX_FMT_BGR0] = { 1, 1 },
118 [AV_PIX_FMT_GRAY16BE] = { 1, 1 },
119 [AV_PIX_FMT_GRAY16LE] = { 1, 1 },
120 [AV_PIX_FMT_YUV440P] = { 1, 1 },
121 [AV_PIX_FMT_YUVJ440P] = { 1, 1 },
122 [AV_PIX_FMT_YUVA420P] = { 1, 1 },
123 [AV_PIX_FMT_YUVA422P] = { 1, 1 },
124 [AV_PIX_FMT_YUVA444P] = { 1, 1 },
125 [AV_PIX_FMT_YUVA420P9BE] = { 1, 1 },
126 [AV_PIX_FMT_YUVA420P9LE] = { 1, 1 },
127 [AV_PIX_FMT_YUVA422P9BE] = { 1, 1 },
128 [AV_PIX_FMT_YUVA422P9LE] = { 1, 1 },
129 [AV_PIX_FMT_YUVA444P9BE] = { 1, 1 },
130 [AV_PIX_FMT_YUVA444P9LE] = { 1, 1 },
131 [AV_PIX_FMT_YUVA420P10BE]= { 1, 1 },
132 [AV_PIX_FMT_YUVA420P10LE]= { 1, 1 },
133 [AV_PIX_FMT_YUVA422P10BE]= { 1, 1 },
134 [AV_PIX_FMT_YUVA422P10LE]= { 1, 1 },
135 [AV_PIX_FMT_YUVA444P10BE]= { 1, 1 },
136 [AV_PIX_FMT_YUVA444P10LE]= { 1, 1 },
137 [AV_PIX_FMT_YUVA420P16BE]= { 1, 1 },
138 [AV_PIX_FMT_YUVA420P16LE]= { 1, 1 },
139 [AV_PIX_FMT_YUVA422P16BE]= { 1, 1 },
140 [AV_PIX_FMT_YUVA422P16LE]= { 1, 1 },
141 [AV_PIX_FMT_YUVA444P16BE]= { 1, 1 },
142 [AV_PIX_FMT_YUVA444P16LE]= { 1, 1 },
143 [AV_PIX_FMT_RGB48BE] = { 1, 1 },
144 [AV_PIX_FMT_RGB48LE] = { 1, 1 },
145 [AV_PIX_FMT_RGBA64BE] = { 1, 1 },
146 [AV_PIX_FMT_RGBA64LE] = { 1, 1 },
147 [AV_PIX_FMT_RGB565BE] = { 1, 1 },
148 [AV_PIX_FMT_RGB565LE] = { 1, 1 },
149 [AV_PIX_FMT_RGB555BE] = { 1, 1 },
150 [AV_PIX_FMT_RGB555LE] = { 1, 1 },
151 [AV_PIX_FMT_BGR565BE] = { 1, 1 },
152 [AV_PIX_FMT_BGR565LE] = { 1, 1 },
153 [AV_PIX_FMT_BGR555BE] = { 1, 1 },
154 [AV_PIX_FMT_BGR555LE] = { 1, 1 },
155 [AV_PIX_FMT_YUV420P16LE] = { 1, 1 },
156 [AV_PIX_FMT_YUV420P16BE] = { 1, 1 },
157 [AV_PIX_FMT_YUV422P16LE] = { 1, 1 },
158 [AV_PIX_FMT_YUV422P16BE] = { 1, 1 },
159 [AV_PIX_FMT_YUV444P16LE] = { 1, 1 },
160 [AV_PIX_FMT_YUV444P16BE] = { 1, 1 },
161 [AV_PIX_FMT_RGB444LE] = { 1, 1 },
162 [AV_PIX_FMT_RGB444BE] = { 1, 1 },
163 [AV_PIX_FMT_BGR444LE] = { 1, 1 },
164 [AV_PIX_FMT_BGR444BE] = { 1, 1 },
165 [AV_PIX_FMT_Y400A] = { 1, 0 },
166 [AV_PIX_FMT_BGR48BE] = { 1, 1 },
167 [AV_PIX_FMT_BGR48LE] = { 1, 1 },
168 [AV_PIX_FMT_BGRA64BE] = { 0, 0 },
169 [AV_PIX_FMT_BGRA64LE] = { 0, 0 },
170 [AV_PIX_FMT_YUV420P9BE] = { 1, 1 },
171 [AV_PIX_FMT_YUV420P9LE] = { 1, 1 },
172 [AV_PIX_FMT_YUV420P10BE] = { 1, 1 },
173 [AV_PIX_FMT_YUV420P10LE] = { 1, 1 },
174 [AV_PIX_FMT_YUV420P12BE] = { 1, 1 },
175 [AV_PIX_FMT_YUV420P12LE] = { 1, 1 },
176 [AV_PIX_FMT_YUV420P14BE] = { 1, 1 },
177 [AV_PIX_FMT_YUV420P14LE] = { 1, 1 },
178 [AV_PIX_FMT_YUV422P9BE] = { 1, 1 },
179 [AV_PIX_FMT_YUV422P9LE] = { 1, 1 },
180 [AV_PIX_FMT_YUV422P10BE] = { 1, 1 },
181 [AV_PIX_FMT_YUV422P10LE] = { 1, 1 },
182 [AV_PIX_FMT_YUV422P12BE] = { 1, 1 },
183 [AV_PIX_FMT_YUV422P12LE] = { 1, 1 },
184 [AV_PIX_FMT_YUV422P14BE] = { 1, 1 },
185 [AV_PIX_FMT_YUV422P14LE] = { 1, 1 },
186 [AV_PIX_FMT_YUV444P9BE] = { 1, 1 },
187 [AV_PIX_FMT_YUV444P9LE] = { 1, 1 },
188 [AV_PIX_FMT_YUV444P10BE] = { 1, 1 },
189 [AV_PIX_FMT_YUV444P10LE] = { 1, 1 },
190 [AV_PIX_FMT_YUV444P12BE] = { 1, 1 },
191 [AV_PIX_FMT_YUV444P12LE] = { 1, 1 },
192 [AV_PIX_FMT_YUV444P14BE] = { 1, 1 },
193 [AV_PIX_FMT_YUV444P14LE] = { 1, 1 },
194 [AV_PIX_FMT_GBRP] = { 1, 1 },
195 [AV_PIX_FMT_GBRP9LE] = { 1, 1 },
196 [AV_PIX_FMT_GBRP9BE] = { 1, 1 },
197 [AV_PIX_FMT_GBRP10LE] = { 1, 1 },
198 [AV_PIX_FMT_GBRP10BE] = { 1, 1 },
199 [AV_PIX_FMT_GBRP12LE] = { 1, 1 },
200 [AV_PIX_FMT_GBRP12BE] = { 1, 1 },
201 [AV_PIX_FMT_GBRP14LE] = { 1, 1 },
202 [AV_PIX_FMT_GBRP14BE] = { 1, 1 },
203 [AV_PIX_FMT_GBRP16LE] = { 1, 0 },
204 [AV_PIX_FMT_GBRP16BE] = { 1, 0 },
205 [AV_PIX_FMT_XYZ12BE] = { 1, 1, 1 },
206 [AV_PIX_FMT_XYZ12LE] = { 1, 1, 1 },
207 [AV_PIX_FMT_GBRAP] = { 1, 1 },
208 [AV_PIX_FMT_GBRAP16LE] = { 1, 0 },
209 [AV_PIX_FMT_GBRAP16BE] = { 1, 0 },
212 int sws_isSupportedInput(enum AVPixelFormat pix_fmt)
214 return (unsigned)pix_fmt < AV_PIX_FMT_NB ?
215 format_entries[pix_fmt].is_supported_in : 0;
218 int sws_isSupportedOutput(enum AVPixelFormat pix_fmt)
220 return (unsigned)pix_fmt < AV_PIX_FMT_NB ?
221 format_entries[pix_fmt].is_supported_out : 0;
224 int sws_isSupportedEndiannessConversion(enum AVPixelFormat pix_fmt)
226 return (unsigned)pix_fmt < AV_PIX_FMT_NB ?
227 format_entries[pix_fmt].is_supported_endianness : 0;
230 #if FF_API_SWS_FORMAT_NAME
231 const char *sws_format_name(enum AVPixelFormat format)
233 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(format);
237 return "Unknown format";
241 static double getSplineCoeff(double a, double b, double c, double d,
245 return ((d * dist + c) * dist + b) * dist + a;
247 return getSplineCoeff(0.0,
248 b + 2.0 * c + 3.0 * d,
250 -b - 3.0 * c - 6.0 * d,
254 static av_cold int get_local_pos(SwsContext *s, int chr_subsample, int pos, int dir)
257 pos = (128 << chr_subsample) - 128;
259 pos += 128; // relative to ideal left edge
260 return pos >> chr_subsample;
263 static av_cold int initFilter(int16_t **outFilter, int32_t **filterPos,
264 int *outFilterSize, int xInc, int srcW,
265 int dstW, int filterAlign, int one,
266 int flags, int cpu_flags,
267 SwsVector *srcFilter, SwsVector *dstFilter,
268 double param[2], int srcPos, int dstPos)
274 int64_t *filter = NULL;
275 int64_t *filter2 = NULL;
276 const int64_t fone = 1LL << (54 - FFMIN(av_log2(srcW/dstW), 8));
279 emms_c(); // FIXME should not be required but IS (even for non-MMX versions)
281 // NOTE: the +3 is for the MMX(+1) / SSE(+3) scaler which reads over the end
282 FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW + 3) * sizeof(**filterPos), fail);
284 if (FFABS(xInc - 0x10000) < 10 && srcPos == dstPos) { // unscaled
287 FF_ALLOCZ_OR_GOTO(NULL, filter,
288 dstW * sizeof(*filter) * filterSize, fail);
290 for (i = 0; i < dstW; i++) {
291 filter[i * filterSize] = fone;
294 } else if (flags & SWS_POINT) { // lame looking point sampling mode
298 FF_ALLOC_OR_GOTO(NULL, filter,
299 dstW * sizeof(*filter) * filterSize, fail);
301 xDstInSrc = ((dstPos*(int64_t)xInc)>>8) - ((srcPos*0x8000LL)>>7);
302 for (i = 0; i < dstW; i++) {
303 int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;
305 (*filterPos)[i] = xx;
309 } else if ((xInc <= (1 << 16) && (flags & SWS_AREA)) ||
310 (flags & SWS_FAST_BILINEAR)) { // bilinear upscale
314 FF_ALLOC_OR_GOTO(NULL, filter,
315 dstW * sizeof(*filter) * filterSize, fail);
317 xDstInSrc = ((dstPos*(int64_t)xInc)>>8) - ((srcPos*0x8000LL)>>7);
318 for (i = 0; i < dstW; i++) {
319 int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;
322 (*filterPos)[i] = xx;
323 // bilinear upscale / linear interpolate / area averaging
324 for (j = 0; j < filterSize; j++) {
325 int64_t coeff= fone - FFABS(((int64_t)xx<<16) - xDstInSrc)*(fone>>16);
328 filter[i * filterSize + j] = coeff;
337 if (flags & SWS_BICUBIC)
339 else if (flags & SWS_X)
341 else if (flags & SWS_AREA)
342 sizeFactor = 1; // downscale only, for upscale it is bilinear
343 else if (flags & SWS_GAUSS)
344 sizeFactor = 8; // infinite ;)
345 else if (flags & SWS_LANCZOS)
346 sizeFactor = param[0] != SWS_PARAM_DEFAULT ? ceil(2 * param[0]) : 6;
347 else if (flags & SWS_SINC)
348 sizeFactor = 20; // infinite ;)
349 else if (flags & SWS_SPLINE)
350 sizeFactor = 20; // infinite ;)
351 else if (flags & SWS_BILINEAR)
358 filterSize = 1 + sizeFactor; // upscale
360 filterSize = 1 + (sizeFactor * srcW + dstW - 1) / dstW;
362 filterSize = FFMIN(filterSize, srcW - 2);
363 filterSize = FFMAX(filterSize, 1);
365 FF_ALLOC_OR_GOTO(NULL, filter,
366 dstW * sizeof(*filter) * filterSize, fail);
368 xDstInSrc = ((dstPos*(int64_t)xInc)>>7) - ((srcPos*0x10000LL)>>7);
369 for (i = 0; i < dstW; i++) {
370 int xx = (xDstInSrc - ((filterSize - 2) << 16)) / (1 << 17);
372 (*filterPos)[i] = xx;
373 for (j = 0; j < filterSize; j++) {
374 int64_t d = (FFABS(((int64_t)xx << 17) - xDstInSrc)) << 13;
380 floatd = d * (1.0 / (1 << 30));
382 if (flags & SWS_BICUBIC) {
383 int64_t B = (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1 << 24);
384 int64_t C = (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1 << 24);
386 if (d >= 1LL << 31) {
389 int64_t dd = (d * d) >> 30;
390 int64_t ddd = (dd * d) >> 30;
393 coeff = (12 * (1 << 24) - 9 * B - 6 * C) * ddd +
394 (-18 * (1 << 24) + 12 * B + 6 * C) * dd +
395 (6 * (1 << 24) - 2 * B) * (1 << 30);
397 coeff = (-B - 6 * C) * ddd +
398 (6 * B + 30 * C) * dd +
399 (-12 * B - 48 * C) * d +
400 (8 * B + 24 * C) * (1 << 30);
402 coeff /= (1LL<<54)/fone;
405 else if (flags & SWS_X) {
406 double p = param ? param * 0.01 : 0.3;
407 coeff = d ? sin(d * M_PI) / (d * M_PI) : 1.0;
408 coeff *= pow(2.0, -p * d * d);
411 else if (flags & SWS_X) {
412 double A = param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
416 c = cos(floatd * M_PI);
423 coeff = (c * 0.5 + 0.5) * fone;
424 } else if (flags & SWS_AREA) {
425 int64_t d2 = d - (1 << 29);
426 if (d2 * xInc < -(1LL << (29 + 16)))
427 coeff = 1.0 * (1LL << (30 + 16));
428 else if (d2 * xInc < (1LL << (29 + 16)))
429 coeff = -d2 * xInc + (1LL << (29 + 16));
432 coeff *= fone >> (30 + 16);
433 } else if (flags & SWS_GAUSS) {
434 double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
435 coeff = (pow(2.0, -p * floatd * floatd)) * fone;
436 } else if (flags & SWS_SINC) {
437 coeff = (d ? sin(floatd * M_PI) / (floatd * M_PI) : 1.0) * fone;
438 } else if (flags & SWS_LANCZOS) {
439 double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
440 coeff = (d ? sin(floatd * M_PI) * sin(floatd * M_PI / p) /
441 (floatd * floatd * M_PI * M_PI / p) : 1.0) * fone;
444 } else if (flags & SWS_BILINEAR) {
445 coeff = (1 << 30) - d;
449 } else if (flags & SWS_SPLINE) {
450 double p = -2.196152422706632;
451 coeff = getSplineCoeff(1.0, 0.0, p, -p - 1.0, floatd) * fone;
456 filter[i * filterSize + j] = coeff;
459 xDstInSrc += 2 * xInc;
463 /* apply src & dst Filter to filter -> filter2
466 av_assert0(filterSize > 0);
467 filter2Size = filterSize;
469 filter2Size += srcFilter->length - 1;
471 filter2Size += dstFilter->length - 1;
472 av_assert0(filter2Size > 0);
473 FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size * dstW * sizeof(*filter2), fail);
475 for (i = 0; i < dstW; i++) {
479 for (k = 0; k < srcFilter->length; k++) {
480 for (j = 0; j < filterSize; j++)
481 filter2[i * filter2Size + k + j] +=
482 srcFilter->coeff[k] * filter[i * filterSize + j];
485 for (j = 0; j < filterSize; j++)
486 filter2[i * filter2Size + j] = filter[i * filterSize + j];
490 (*filterPos)[i] += (filterSize - 1) / 2 - (filter2Size - 1) / 2;
494 /* try to reduce the filter-size (step1 find size and shift left) */
495 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
497 for (i = dstW - 1; i >= 0; i--) {
498 int min = filter2Size;
500 int64_t cutOff = 0.0;
502 /* get rid of near zero elements on the left by shifting left */
503 for (j = 0; j < filter2Size; j++) {
505 cutOff += FFABS(filter2[i * filter2Size]);
507 if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)
510 /* preserve monotonicity because the core can't handle the
511 * filter otherwise */
512 if (i < dstW - 1 && (*filterPos)[i] >= (*filterPos)[i + 1])
515 // move filter coefficients left
516 for (k = 1; k < filter2Size; k++)
517 filter2[i * filter2Size + k - 1] = filter2[i * filter2Size + k];
518 filter2[i * filter2Size + k - 1] = 0;
523 /* count near zeros on the right */
524 for (j = filter2Size - 1; j > 0; j--) {
525 cutOff += FFABS(filter2[i * filter2Size + j]);
527 if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)
532 if (min > minFilterSize)
536 if (PPC_ALTIVEC(cpu_flags)) {
537 // we can handle the special case 4, so we don't want to go the full 8
538 if (minFilterSize < 5)
541 /* We really don't want to waste our time doing useless computation, so
542 * fall back on the scalar C code for very small filters.
543 * Vectorizing is worth it only if you have a decent-sized vector. */
544 if (minFilterSize < 3)
548 if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
549 // special case for unscaled vertical filtering
550 if (minFilterSize == 1 && filterAlign == 2)
554 av_assert0(minFilterSize > 0);
555 filterSize = (minFilterSize + (filterAlign - 1)) & (~(filterAlign - 1));
556 av_assert0(filterSize > 0);
557 filter = av_malloc(filterSize * dstW * sizeof(*filter));
558 if (filterSize >= MAX_FILTER_SIZE * 16 /
559 ((flags & SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter) {
560 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);
563 *outFilterSize = filterSize;
565 if (flags & SWS_PRINT_INFO)
566 av_log(NULL, AV_LOG_VERBOSE,
567 "SwScaler: reducing / aligning filtersize %d -> %d\n",
568 filter2Size, filterSize);
569 /* try to reduce the filter-size (step2 reduce it) */
570 for (i = 0; i < dstW; i++) {
573 for (j = 0; j < filterSize; j++) {
574 if (j >= filter2Size)
575 filter[i * filterSize + j] = 0;
577 filter[i * filterSize + j] = filter2[i * filter2Size + j];
578 if ((flags & SWS_BITEXACT) && j >= minFilterSize)
579 filter[i * filterSize + j] = 0;
583 // FIXME try to align filterPos if possible
586 for (i = 0; i < dstW; i++) {
588 if ((*filterPos)[i] < 0) {
589 // move filter coefficients left to compensate for filterPos
590 for (j = 1; j < filterSize; j++) {
591 int left = FFMAX(j + (*filterPos)[i], 0);
592 filter[i * filterSize + left] += filter[i * filterSize + j];
593 filter[i * filterSize + j] = 0;
598 if ((*filterPos)[i] + filterSize > srcW) {
599 int shift = (*filterPos)[i] + filterSize - srcW;
600 // move filter coefficients right to compensate for filterPos
601 for (j = filterSize - 2; j >= 0; j--) {
602 int right = FFMIN(j + shift, filterSize - 1);
603 filter[i * filterSize + right] += filter[i * filterSize + j];
604 filter[i * filterSize + j] = 0;
606 (*filterPos)[i]= srcW - filterSize;
610 // Note the +1 is for the MMX scaler which reads over the end
611 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
612 FF_ALLOCZ_OR_GOTO(NULL, *outFilter,
613 *outFilterSize * (dstW + 3) * sizeof(int16_t), fail);
615 /* normalize & store in outFilter */
616 for (i = 0; i < dstW; i++) {
621 for (j = 0; j < filterSize; j++) {
622 sum += filter[i * filterSize + j];
624 sum = (sum + one / 2) / one;
625 for (j = 0; j < *outFilterSize; j++) {
626 int64_t v = filter[i * filterSize + j] + error;
627 int intV = ROUNDED_DIV(v, sum);
628 (*outFilter)[i * (*outFilterSize) + j] = intV;
629 error = v - intV * sum;
633 (*filterPos)[dstW + 0] =
634 (*filterPos)[dstW + 1] =
635 (*filterPos)[dstW + 2] = (*filterPos)[dstW - 1]; /* the MMX/SSE scaler will
636 * read over the end */
637 for (i = 0; i < *outFilterSize; i++) {
638 int k = (dstW - 1) * (*outFilterSize) + i;
639 (*outFilter)[k + 1 * (*outFilterSize)] =
640 (*outFilter)[k + 2 * (*outFilterSize)] =
641 (*outFilter)[k + 3 * (*outFilterSize)] = (*outFilter)[k];
648 av_log(NULL, AV_LOG_ERROR, "sws: initFilter failed\n");
654 #if HAVE_MMXEXT_INLINE
655 static av_cold int init_hscaler_mmxext(int dstW, int xInc, uint8_t *filterCode,
656 int16_t *filter, int32_t *filterPos,
660 x86_reg imm8OfPShufW1A;
661 x86_reg imm8OfPShufW2A;
662 x86_reg fragmentLengthA;
664 x86_reg imm8OfPShufW1B;
665 x86_reg imm8OfPShufW2B;
666 x86_reg fragmentLengthB;
671 // create an optimized horizontal scaling routine
672 /* This scaler is made of runtime-generated MMXEXT code using specially tuned
673 * pshufw instructions. For every four output pixels, if four input pixels
674 * are enough for the fast bilinear scaling, then a chunk of fragmentB is
675 * used. If five input pixels are needed, then a chunk of fragmentA is used.
684 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
685 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
686 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
687 "punpcklbw %%mm7, %%mm1 \n\t"
688 "punpcklbw %%mm7, %%mm0 \n\t"
689 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
691 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
693 "psubw %%mm1, %%mm0 \n\t"
694 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
695 "pmullw %%mm3, %%mm0 \n\t"
696 "psllw $7, %%mm1 \n\t"
697 "paddw %%mm1, %%mm0 \n\t"
699 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
701 "add $8, %%"REG_a" \n\t"
705 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
706 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
707 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
712 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
716 : "=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
717 "=r" (fragmentLengthA)
724 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
725 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
726 "punpcklbw %%mm7, %%mm0 \n\t"
727 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
729 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
731 "psubw %%mm1, %%mm0 \n\t"
732 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
733 "pmullw %%mm3, %%mm0 \n\t"
734 "psllw $7, %%mm1 \n\t"
735 "paddw %%mm1, %%mm0 \n\t"
737 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
739 "add $8, %%"REG_a" \n\t"
743 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
744 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
745 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
750 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
754 : "=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
755 "=r" (fragmentLengthB)
758 xpos = 0; // lumXInc/2 - 0x8000; // difference between pixel centers
761 for (i = 0; i < dstW / numSplits; i++) {
766 int b = ((xpos + xInc) >> 16) - xx;
767 int c = ((xpos + xInc * 2) >> 16) - xx;
768 int d = ((xpos + xInc * 3) >> 16) - xx;
769 int inc = (d + 1 < 4);
770 uint8_t *fragment = (d + 1 < 4) ? fragmentB : fragmentA;
771 x86_reg imm8OfPShufW1 = (d + 1 < 4) ? imm8OfPShufW1B : imm8OfPShufW1A;
772 x86_reg imm8OfPShufW2 = (d + 1 < 4) ? imm8OfPShufW2B : imm8OfPShufW2A;
773 x86_reg fragmentLength = (d + 1 < 4) ? fragmentLengthB : fragmentLengthA;
774 int maxShift = 3 - (d + inc);
778 filter[i] = ((xpos & 0xFFFF) ^ 0xFFFF) >> 9;
779 filter[i + 1] = (((xpos + xInc) & 0xFFFF) ^ 0xFFFF) >> 9;
780 filter[i + 2] = (((xpos + xInc * 2) & 0xFFFF) ^ 0xFFFF) >> 9;
781 filter[i + 3] = (((xpos + xInc * 3) & 0xFFFF) ^ 0xFFFF) >> 9;
782 filterPos[i / 2] = xx;
784 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
786 filterCode[fragmentPos + imm8OfPShufW1] = (a + inc) |
790 filterCode[fragmentPos + imm8OfPShufW2] = a | (b << 2) |
794 if (i + 4 - inc >= dstW)
795 shift = maxShift; // avoid overread
796 else if ((filterPos[i / 2] & 3) <= maxShift)
797 shift = filterPos[i / 2] & 3; // align
799 if (shift && i >= shift) {
800 filterCode[fragmentPos + imm8OfPShufW1] += 0x55 * shift;
801 filterCode[fragmentPos + imm8OfPShufW2] += 0x55 * shift;
802 filterPos[i / 2] -= shift;
806 fragmentPos += fragmentLength;
809 filterCode[fragmentPos] = RET;
814 filterPos[((i / 2) + 1) & (~1)] = xpos >> 16; // needed to jump to the next part
816 return fragmentPos + 1;
818 #endif /* HAVE_MMXEXT_INLINE */
820 static void getSubSampleFactors(int *h, int *v, enum AVPixelFormat format)
822 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(format);
823 *h = desc->log2_chroma_w;
824 *v = desc->log2_chroma_h;
827 static void fill_rgb2yuv_table(SwsContext *c, const int table[4], int dstRange)
829 int64_t W, V, Z, Cy, Cu, Cv;
830 int64_t vr = table[0];
831 int64_t ub = table[1];
832 int64_t ug = -table[2];
833 int64_t vg = -table[3];
836 uint8_t *p = (uint8_t*)c->input_rgb2yuv_table;
838 static const int8_t map[] = {
839 BY_IDX, GY_IDX, -1 , BY_IDX, BY_IDX, GY_IDX, -1 , BY_IDX,
840 RY_IDX, -1 , GY_IDX, RY_IDX, RY_IDX, -1 , GY_IDX, RY_IDX,
841 RY_IDX, GY_IDX, -1 , RY_IDX, RY_IDX, GY_IDX, -1 , RY_IDX,
842 BY_IDX, -1 , GY_IDX, BY_IDX, BY_IDX, -1 , GY_IDX, BY_IDX,
843 BU_IDX, GU_IDX, -1 , BU_IDX, BU_IDX, GU_IDX, -1 , BU_IDX,
844 RU_IDX, -1 , GU_IDX, RU_IDX, RU_IDX, -1 , GU_IDX, RU_IDX,
845 RU_IDX, GU_IDX, -1 , RU_IDX, RU_IDX, GU_IDX, -1 , RU_IDX,
846 BU_IDX, -1 , GU_IDX, BU_IDX, BU_IDX, -1 , GU_IDX, BU_IDX,
847 BV_IDX, GV_IDX, -1 , BV_IDX, BV_IDX, GV_IDX, -1 , BV_IDX,
848 RV_IDX, -1 , GV_IDX, RV_IDX, RV_IDX, -1 , GV_IDX, RV_IDX,
849 RV_IDX, GV_IDX, -1 , RV_IDX, RV_IDX, GV_IDX, -1 , RV_IDX,
850 BV_IDX, -1 , GV_IDX, BV_IDX, BV_IDX, -1 , GV_IDX, BV_IDX,
851 RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX,
852 BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX,
853 GY_IDX, -1 , GY_IDX, -1 , GY_IDX, -1 , GY_IDX, -1 ,
854 -1 , GY_IDX, -1 , GY_IDX, -1 , GY_IDX, -1 , GY_IDX,
855 RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX,
856 BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX,
857 GU_IDX, -1 , GU_IDX, -1 , GU_IDX, -1 , GU_IDX, -1 ,
858 -1 , GU_IDX, -1 , GU_IDX, -1 , GU_IDX, -1 , GU_IDX,
859 RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX,
860 BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX,
861 GV_IDX, -1 , GV_IDX, -1 , GV_IDX, -1 , GV_IDX, -1 ,
862 -1 , GV_IDX, -1 , GV_IDX, -1 , GV_IDX, -1 , GV_IDX, //23
863 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //24
864 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //25
865 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //26
866 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //27
867 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //28
868 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //29
869 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //30
870 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //31
871 BY_IDX, GY_IDX, RY_IDX, -1 , -1 , -1 , -1 , -1 , //32
872 BU_IDX, GU_IDX, RU_IDX, -1 , -1 , -1 , -1 , -1 , //33
873 BV_IDX, GV_IDX, RV_IDX, -1 , -1 , -1 , -1 , -1 , //34
876 dstRange = 0; //FIXME range = 1 is handled elsewhere
886 W = ROUNDED_DIV(ONE*ONE*ug, ub);
887 V = ROUNDED_DIV(ONE*ONE*vg, vr);
890 Cy = ROUNDED_DIV(cy*Z, ONE);
891 Cu = ROUNDED_DIV(ub*Z, ONE);
892 Cv = ROUNDED_DIV(vr*Z, ONE);
894 c->input_rgb2yuv_table[RY_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*V , Cy);
895 c->input_rgb2yuv_table[GY_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE , Cy);
896 c->input_rgb2yuv_table[BY_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*W , Cy);
898 c->input_rgb2yuv_table[RU_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*V , Cu);
899 c->input_rgb2yuv_table[GU_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE , Cu);
900 c->input_rgb2yuv_table[BU_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*(Z+W) , Cu);
902 c->input_rgb2yuv_table[RV_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*(V+Z) , Cv);
903 c->input_rgb2yuv_table[GV_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE , Cv);
904 c->input_rgb2yuv_table[BV_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*W , Cv);
906 if(/*!dstRange && */!memcmp(table, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], sizeof(ff_yuv2rgb_coeffs[SWS_CS_DEFAULT]))) {
907 c->input_rgb2yuv_table[BY_IDX] = ((int)(0.114 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
908 c->input_rgb2yuv_table[BV_IDX] = (-(int)(0.081 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
909 c->input_rgb2yuv_table[BU_IDX] = ((int)(0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
910 c->input_rgb2yuv_table[GY_IDX] = ((int)(0.587 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
911 c->input_rgb2yuv_table[GV_IDX] = (-(int)(0.419 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
912 c->input_rgb2yuv_table[GU_IDX] = (-(int)(0.331 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
913 c->input_rgb2yuv_table[RY_IDX] = ((int)(0.299 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
914 c->input_rgb2yuv_table[RV_IDX] = ((int)(0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
915 c->input_rgb2yuv_table[RU_IDX] = (-(int)(0.169 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
917 for(i=0; i<FF_ARRAY_ELEMS(map); i++)
918 AV_WL16(p + 16*4 + 2*i, map[i] >= 0 ? c->input_rgb2yuv_table[map[i]] : 0);
921 static void fill_xyztables(struct SwsContext *c)
924 double xyzgamma = XYZ_GAMMA;
925 double rgbgamma = 1.0 / RGB_GAMMA;
926 double xyzgammainv = 1.0 / XYZ_GAMMA;
927 double rgbgammainv = RGB_GAMMA;
928 static const int16_t xyz2rgb_matrix[3][4] = {
929 {13270, -6295, -2041},
931 { 228, -835, 4329} };
932 static const int16_t rgb2xyz_matrix[3][4] = {
936 static int16_t xyzgamma_tab[4096], rgbgamma_tab[4096], xyzgammainv_tab[4096], rgbgammainv_tab[4096];
938 memcpy(c->xyz2rgb_matrix, xyz2rgb_matrix, sizeof(c->xyz2rgb_matrix));
939 memcpy(c->rgb2xyz_matrix, rgb2xyz_matrix, sizeof(c->rgb2xyz_matrix));
940 c->xyzgamma = xyzgamma_tab;
941 c->rgbgamma = rgbgamma_tab;
942 c->xyzgammainv = xyzgammainv_tab;
943 c->rgbgammainv = rgbgammainv_tab;
945 if (rgbgamma_tab[4095])
948 /* set gamma vectors */
949 for (i = 0; i < 4096; i++) {
950 xyzgamma_tab[i] = lrint(pow(i / 4095.0, xyzgamma) * 4095.0);
951 rgbgamma_tab[i] = lrint(pow(i / 4095.0, rgbgamma) * 4095.0);
952 xyzgammainv_tab[i] = lrint(pow(i / 4095.0, xyzgammainv) * 4095.0);
953 rgbgammainv_tab[i] = lrint(pow(i / 4095.0, rgbgammainv) * 4095.0);
957 int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],
958 int srcRange, const int table[4], int dstRange,
959 int brightness, int contrast, int saturation)
961 const AVPixFmtDescriptor *desc_dst;
962 const AVPixFmtDescriptor *desc_src;
963 memcpy(c->srcColorspaceTable, inv_table, sizeof(int) * 4);
964 memcpy(c->dstColorspaceTable, table, sizeof(int) * 4);
967 desc_dst = av_pix_fmt_desc_get(c->dstFormat);
968 desc_src = av_pix_fmt_desc_get(c->srcFormat);
970 if(!isYUV(c->dstFormat) && !isGray(c->dstFormat))
972 if(!isYUV(c->srcFormat) && !isGray(c->srcFormat))
975 c->brightness = brightness;
976 c->contrast = contrast;
977 c->saturation = saturation;
978 c->srcRange = srcRange;
979 c->dstRange = dstRange;
983 if ((isYUV(c->dstFormat) || isGray(c->dstFormat)) && (isYUV(c->srcFormat) || isGray(c->srcFormat)))
986 c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);
987 c->srcFormatBpp = av_get_bits_per_pixel(desc_src);
989 if (!isYUV(c->dstFormat) && !isGray(c->dstFormat)) {
990 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness,
991 contrast, saturation);
994 if (PPC_ALTIVEC(av_get_cpu_flags()))
995 ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness,
996 contrast, saturation);
999 fill_rgb2yuv_table(c, table, dstRange);
1004 int sws_getColorspaceDetails(struct SwsContext *c, int **inv_table,
1005 int *srcRange, int **table, int *dstRange,
1006 int *brightness, int *contrast, int *saturation)
1011 *inv_table = c->srcColorspaceTable;
1012 *table = c->dstColorspaceTable;
1013 *srcRange = c->srcRange;
1014 *dstRange = c->dstRange;
1015 *brightness = c->brightness;
1016 *contrast = c->contrast;
1017 *saturation = c->saturation;
1022 static int handle_jpeg(enum AVPixelFormat *format)
1025 case AV_PIX_FMT_YUVJ420P:
1026 *format = AV_PIX_FMT_YUV420P;
1028 case AV_PIX_FMT_YUVJ411P:
1029 *format = AV_PIX_FMT_YUV411P;
1031 case AV_PIX_FMT_YUVJ422P:
1032 *format = AV_PIX_FMT_YUV422P;
1034 case AV_PIX_FMT_YUVJ444P:
1035 *format = AV_PIX_FMT_YUV444P;
1037 case AV_PIX_FMT_YUVJ440P:
1038 *format = AV_PIX_FMT_YUV440P;
1040 case AV_PIX_FMT_GRAY8:
1047 static int handle_0alpha(enum AVPixelFormat *format)
1050 case AV_PIX_FMT_0BGR : *format = AV_PIX_FMT_ABGR ; return 1;
1051 case AV_PIX_FMT_BGR0 : *format = AV_PIX_FMT_BGRA ; return 4;
1052 case AV_PIX_FMT_0RGB : *format = AV_PIX_FMT_ARGB ; return 1;
1053 case AV_PIX_FMT_RGB0 : *format = AV_PIX_FMT_RGBA ; return 4;
1058 static int handle_xyz(enum AVPixelFormat *format)
1061 case AV_PIX_FMT_XYZ12BE : *format = AV_PIX_FMT_RGB48BE; return 1;
1062 case AV_PIX_FMT_XYZ12LE : *format = AV_PIX_FMT_RGB48LE; return 1;
1067 static void handle_formats(SwsContext *c)
1069 c->src0Alpha |= handle_0alpha(&c->srcFormat);
1070 c->dst0Alpha |= handle_0alpha(&c->dstFormat);
1071 c->srcXYZ |= handle_xyz(&c->srcFormat);
1072 c->dstXYZ |= handle_xyz(&c->dstFormat);
1075 SwsContext *sws_alloc_context(void)
1077 SwsContext *c = av_mallocz(sizeof(SwsContext));
1080 c->av_class = &sws_context_class;
1081 av_opt_set_defaults(c);
1087 av_cold int sws_init_context(SwsContext *c, SwsFilter *srcFilter,
1088 SwsFilter *dstFilter)
1091 int usesVFilter, usesHFilter;
1093 SwsFilter dummyFilter = { NULL, NULL, NULL, NULL };
1098 int dst_stride = FFALIGN(dstW * sizeof(int16_t) + 66, 16);
1099 int flags, cpu_flags;
1100 enum AVPixelFormat srcFormat = c->srcFormat;
1101 enum AVPixelFormat dstFormat = c->dstFormat;
1102 const AVPixFmtDescriptor *desc_src;
1103 const AVPixFmtDescriptor *desc_dst;
1105 cpu_flags = av_get_cpu_flags();
1111 unscaled = (srcW == dstW && srcH == dstH);
1113 c->srcRange |= handle_jpeg(&c->srcFormat);
1114 c->dstRange |= handle_jpeg(&c->dstFormat);
1116 if (!c->contrast && !c->saturation && !c->dstFormatBpp)
1117 sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,
1118 ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],
1119 c->dstRange, 0, 1 << 16, 1 << 16);
1121 if(srcFormat!=c->srcFormat || dstFormat!=c->dstFormat)
1122 av_log(c, AV_LOG_WARNING, "deprecated pixel format used, make sure you did set range correctly\n");
1124 srcFormat = c->srcFormat;
1125 dstFormat = c->dstFormat;
1126 desc_src = av_pix_fmt_desc_get(srcFormat);
1127 desc_dst = av_pix_fmt_desc_get(dstFormat);
1129 if (!(unscaled && sws_isSupportedEndiannessConversion(srcFormat) &&
1130 av_pix_fmt_swap_endianness(srcFormat) == dstFormat)) {
1131 if (!sws_isSupportedInput(srcFormat)) {
1132 av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n",
1133 av_get_pix_fmt_name(srcFormat));
1134 return AVERROR(EINVAL);
1136 if (!sws_isSupportedOutput(dstFormat)) {
1137 av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n",
1138 av_get_pix_fmt_name(dstFormat));
1139 return AVERROR(EINVAL);
1143 i = flags & (SWS_POINT |
1154 if (!i || (i & (i - 1))) {
1155 av_log(c, AV_LOG_ERROR, "Exactly one scaler algorithm must be chosen, got %X\n", i);
1156 return AVERROR(EINVAL);
1159 if (srcW < 1 || srcH < 1 || dstW < 1 || dstH < 1) {
1160 /* FIXME check if these are enough and try to lower them after
1161 * fixing the relevant parts of the code */
1162 av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",
1163 srcW, srcH, dstW, dstH);
1164 return AVERROR(EINVAL);
1168 dstFilter = &dummyFilter;
1170 srcFilter = &dummyFilter;
1172 c->lumXInc = (((int64_t)srcW << 16) + (dstW >> 1)) / dstW;
1173 c->lumYInc = (((int64_t)srcH << 16) + (dstH >> 1)) / dstH;
1174 c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);
1175 c->srcFormatBpp = av_get_bits_per_pixel(desc_src);
1176 c->vRounder = 4 * 0x0001000100010001ULL;
1178 usesVFilter = (srcFilter->lumV && srcFilter->lumV->length > 1) ||
1179 (srcFilter->chrV && srcFilter->chrV->length > 1) ||
1180 (dstFilter->lumV && dstFilter->lumV->length > 1) ||
1181 (dstFilter->chrV && dstFilter->chrV->length > 1);
1182 usesHFilter = (srcFilter->lumH && srcFilter->lumH->length > 1) ||
1183 (srcFilter->chrH && srcFilter->chrH->length > 1) ||
1184 (dstFilter->lumH && dstFilter->lumH->length > 1) ||
1185 (dstFilter->chrH && dstFilter->chrH->length > 1);
1187 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
1188 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
1190 if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) {
1192 av_log(c, AV_LOG_DEBUG, "Forcing full internal H chroma due to odd output size\n");
1193 flags |= SWS_FULL_CHR_H_INT;
1198 if (c->dither == SWS_DITHER_AUTO) {
1199 if (flags & SWS_ERROR_DIFFUSION)
1200 c->dither = SWS_DITHER_ED;
1203 if(dstFormat == AV_PIX_FMT_BGR4_BYTE ||
1204 dstFormat == AV_PIX_FMT_RGB4_BYTE ||
1205 dstFormat == AV_PIX_FMT_BGR8 ||
1206 dstFormat == AV_PIX_FMT_RGB8) {
1207 if (c->dither == SWS_DITHER_AUTO)
1208 c->dither = (flags & SWS_FULL_CHR_H_INT) ? SWS_DITHER_ED : SWS_DITHER_BAYER;
1209 if (!(flags & SWS_FULL_CHR_H_INT)) {
1210 if (c->dither == SWS_DITHER_ED) {
1211 av_log(c, AV_LOG_DEBUG,
1212 "Desired dithering only supported in full chroma interpolation for destination format '%s'\n",
1213 av_get_pix_fmt_name(dstFormat));
1214 flags |= SWS_FULL_CHR_H_INT;
1218 if (flags & SWS_FULL_CHR_H_INT) {
1219 if (c->dither == SWS_DITHER_BAYER) {
1220 av_log(c, AV_LOG_DEBUG,
1221 "Ordered dither is not supported in full chroma interpolation for destination format '%s'\n",
1222 av_get_pix_fmt_name(dstFormat));
1223 c->dither = SWS_DITHER_ED;
1227 if (isPlanarRGB(dstFormat)) {
1228 if (!(flags & SWS_FULL_CHR_H_INT)) {
1229 av_log(c, AV_LOG_DEBUG,
1230 "%s output is not supported with half chroma resolution, switching to full\n",
1231 av_get_pix_fmt_name(dstFormat));
1232 flags |= SWS_FULL_CHR_H_INT;
1237 /* reuse chroma for 2 pixels RGB/BGR unless user wants full
1238 * chroma interpolation */
1239 if (flags & SWS_FULL_CHR_H_INT &&
1240 isAnyRGB(dstFormat) &&
1241 !isPlanarRGB(dstFormat) &&
1242 dstFormat != AV_PIX_FMT_RGBA &&
1243 dstFormat != AV_PIX_FMT_ARGB &&
1244 dstFormat != AV_PIX_FMT_BGRA &&
1245 dstFormat != AV_PIX_FMT_ABGR &&
1246 dstFormat != AV_PIX_FMT_RGB24 &&
1247 dstFormat != AV_PIX_FMT_BGR24 &&
1248 dstFormat != AV_PIX_FMT_BGR4_BYTE &&
1249 dstFormat != AV_PIX_FMT_RGB4_BYTE &&
1250 dstFormat != AV_PIX_FMT_BGR8 &&
1251 dstFormat != AV_PIX_FMT_RGB8
1253 av_log(c, AV_LOG_WARNING,
1254 "full chroma interpolation for destination format '%s' not yet implemented\n",
1255 av_get_pix_fmt_name(dstFormat));
1256 flags &= ~SWS_FULL_CHR_H_INT;
1259 if (isAnyRGB(dstFormat) && !(flags & SWS_FULL_CHR_H_INT))
1260 c->chrDstHSubSample = 1;
1262 // drop some chroma lines if the user wants it
1263 c->vChrDrop = (flags & SWS_SRC_V_CHR_DROP_MASK) >>
1264 SWS_SRC_V_CHR_DROP_SHIFT;
1265 c->chrSrcVSubSample += c->vChrDrop;
1267 /* drop every other pixel for chroma calculation unless user
1268 * wants full chroma */
1269 if (isAnyRGB(srcFormat) && !(flags & SWS_FULL_CHR_H_INP) &&
1270 srcFormat != AV_PIX_FMT_RGB8 && srcFormat != AV_PIX_FMT_BGR8 &&
1271 srcFormat != AV_PIX_FMT_RGB4 && srcFormat != AV_PIX_FMT_BGR4 &&
1272 srcFormat != AV_PIX_FMT_RGB4_BYTE && srcFormat != AV_PIX_FMT_BGR4_BYTE &&
1273 srcFormat != AV_PIX_FMT_GBRP9BE && srcFormat != AV_PIX_FMT_GBRP9LE &&
1274 srcFormat != AV_PIX_FMT_GBRP10BE && srcFormat != AV_PIX_FMT_GBRP10LE &&
1275 srcFormat != AV_PIX_FMT_GBRP12BE && srcFormat != AV_PIX_FMT_GBRP12LE &&
1276 srcFormat != AV_PIX_FMT_GBRP14BE && srcFormat != AV_PIX_FMT_GBRP14LE &&
1277 srcFormat != AV_PIX_FMT_GBRP16BE && srcFormat != AV_PIX_FMT_GBRP16LE &&
1278 ((dstW >> c->chrDstHSubSample) <= (srcW >> 1) ||
1279 (flags & SWS_FAST_BILINEAR)))
1280 c->chrSrcHSubSample = 1;
1282 // Note the FF_CEIL_RSHIFT is so that we always round toward +inf.
1283 c->chrSrcW = FF_CEIL_RSHIFT(srcW, c->chrSrcHSubSample);
1284 c->chrSrcH = FF_CEIL_RSHIFT(srcH, c->chrSrcVSubSample);
1285 c->chrDstW = FF_CEIL_RSHIFT(dstW, c->chrDstHSubSample);
1286 c->chrDstH = FF_CEIL_RSHIFT(dstH, c->chrDstVSubSample);
1288 FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW*2+78, 16) * 2, fail);
1290 /* unscaled special cases */
1291 if (unscaled && !usesHFilter && !usesVFilter &&
1292 (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
1293 ff_get_unscaled_swscale(c);
1296 if (flags & SWS_PRINT_INFO)
1297 av_log(c, AV_LOG_INFO,
1298 "using unscaled %s -> %s special converter\n",
1299 av_get_pix_fmt_name(srcFormat), av_get_pix_fmt_name(dstFormat));
1304 c->srcBpc = 1 + desc_src->comp[0].depth_minus1;
1307 c->dstBpc = 1 + desc_dst->comp[0].depth_minus1;
1310 if (isAnyRGB(srcFormat) || srcFormat == AV_PIX_FMT_PAL8)
1312 if (c->dstBpc == 16)
1315 if (INLINE_MMXEXT(cpu_flags) && c->srcBpc == 8 && c->dstBpc <= 14) {
1316 c->canMMXEXTBeUsed = (dstW >= srcW && (dstW & 31) == 0 &&
1317 (srcW & 15) == 0) ? 1 : 0;
1318 if (!c->canMMXEXTBeUsed && dstW >= srcW && (srcW & 15) == 0
1320 && (flags & SWS_FAST_BILINEAR)) {
1321 if (flags & SWS_PRINT_INFO)
1322 av_log(c, AV_LOG_INFO,
1323 "output width is not a multiple of 32 -> no MMXEXT scaler\n");
1325 if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat))
1326 c->canMMXEXTBeUsed = 0;
1328 c->canMMXEXTBeUsed = 0;
1330 c->chrXInc = (((int64_t)c->chrSrcW << 16) + (c->chrDstW >> 1)) / c->chrDstW;
1331 c->chrYInc = (((int64_t)c->chrSrcH << 16) + (c->chrDstH >> 1)) / c->chrDstH;
1333 /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src
1334 * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do
1336 * n-2 is the last chrominance sample available.
1337 * This is not perfect, but no one should notice the difference, the more
1338 * correct variant would be like the vertical one, but that would require
1339 * some special code for the first and last pixel */
1340 if (flags & SWS_FAST_BILINEAR) {
1341 if (c->canMMXEXTBeUsed) {
1345 // we don't use the x86 asm scaler if MMX is available
1346 else if (INLINE_MMX(cpu_flags) && c->dstBpc <= 14) {
1347 c->lumXInc = ((int64_t)(srcW - 2) << 16) / (dstW - 2) - 20;
1348 c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;
1352 #define USE_MMAP (HAVE_MMAP && HAVE_MPROTECT && defined MAP_ANONYMOUS)
1354 /* precalculate horizontal scaler filter coefficients */
1356 #if HAVE_MMXEXT_INLINE
1357 // can't downscale !!!
1358 if (c->canMMXEXTBeUsed && (flags & SWS_FAST_BILINEAR)) {
1359 c->lumMmxextFilterCodeSize = init_hscaler_mmxext(dstW, c->lumXInc, NULL,
1361 c->chrMmxextFilterCodeSize = init_hscaler_mmxext(c->chrDstW, c->chrXInc,
1362 NULL, NULL, NULL, 4);
1365 c->lumMmxextFilterCode = mmap(NULL, c->lumMmxextFilterCodeSize,
1366 PROT_READ | PROT_WRITE,
1367 MAP_PRIVATE | MAP_ANONYMOUS,
1369 c->chrMmxextFilterCode = mmap(NULL, c->chrMmxextFilterCodeSize,
1370 PROT_READ | PROT_WRITE,
1371 MAP_PRIVATE | MAP_ANONYMOUS,
1373 #elif HAVE_VIRTUALALLOC
1374 c->lumMmxextFilterCode = VirtualAlloc(NULL,
1375 c->lumMmxextFilterCodeSize,
1377 PAGE_EXECUTE_READWRITE);
1378 c->chrMmxextFilterCode = VirtualAlloc(NULL,
1379 c->chrMmxextFilterCodeSize,
1381 PAGE_EXECUTE_READWRITE);
1383 c->lumMmxextFilterCode = av_malloc(c->lumMmxextFilterCodeSize);
1384 c->chrMmxextFilterCode = av_malloc(c->chrMmxextFilterCodeSize);
1387 #ifdef MAP_ANONYMOUS
1388 if (c->lumMmxextFilterCode == MAP_FAILED || c->chrMmxextFilterCode == MAP_FAILED)
1390 if (!c->lumMmxextFilterCode || !c->chrMmxextFilterCode)
1393 av_log(c, AV_LOG_ERROR, "Failed to allocate MMX2FilterCode\n");
1394 return AVERROR(ENOMEM);
1397 FF_ALLOCZ_OR_GOTO(c, c->hLumFilter, (dstW / 8 + 8) * sizeof(int16_t), fail);
1398 FF_ALLOCZ_OR_GOTO(c, c->hChrFilter, (c->chrDstW / 4 + 8) * sizeof(int16_t), fail);
1399 FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW / 2 / 8 + 8) * sizeof(int32_t), fail);
1400 FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);
1402 init_hscaler_mmxext( dstW, c->lumXInc, c->lumMmxextFilterCode,
1403 c->hLumFilter, (uint32_t*)c->hLumFilterPos, 8);
1404 init_hscaler_mmxext(c->chrDstW, c->chrXInc, c->chrMmxextFilterCode,
1405 c->hChrFilter, (uint32_t*)c->hChrFilterPos, 4);
1408 mprotect(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize, PROT_EXEC | PROT_READ);
1409 mprotect(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize, PROT_EXEC | PROT_READ);
1412 #endif /* HAVE_MMXEXT_INLINE */
1414 const int filterAlign =
1415 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :
1416 PPC_ALTIVEC(cpu_flags) ? 8 : 1;
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 PPC_ALTIVEC(cpu_flags) ? 8 : 1;
1445 if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,
1446 c->lumYInc, srcH, dstH, filterAlign, (1 << 12),
1447 (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
1448 cpu_flags, srcFilter->lumV, dstFilter->lumV,
1450 get_local_pos(c, 0, 0, 1),
1451 get_local_pos(c, 0, 0, 1)) < 0)
1453 if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,
1454 c->chrYInc, c->chrSrcH, c->chrDstH,
1455 filterAlign, (1 << 12),
1456 (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
1457 cpu_flags, srcFilter->chrV, dstFilter->chrV,
1459 get_local_pos(c, c->chrSrcVSubSample, c->src_v_chr_pos, 1),
1460 get_local_pos(c, c->chrDstVSubSample, c->dst_v_chr_pos, 1)) < 0)
1465 FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH, fail);
1466 FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);
1468 for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {
1470 short *p = (short *)&c->vYCoeffsBank[i];
1471 for (j = 0; j < 8; j++)
1472 p[j] = c->vLumFilter[i];
1475 for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {
1477 short *p = (short *)&c->vCCoeffsBank[i];
1478 for (j = 0; j < 8; j++)
1479 p[j] = c->vChrFilter[i];
1484 // calculate buffer sizes so that they won't run out while handling these damn slices
1485 c->vLumBufSize = c->vLumFilterSize;
1486 c->vChrBufSize = c->vChrFilterSize;
1487 for (i = 0; i < dstH; i++) {
1488 int chrI = (int64_t)i * c->chrDstH / dstH;
1489 int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,
1490 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)
1491 << c->chrSrcVSubSample));
1493 nextSlice >>= c->chrSrcVSubSample;
1494 nextSlice <<= c->chrSrcVSubSample;
1495 if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)
1496 c->vLumBufSize = nextSlice - c->vLumFilterPos[i];
1497 if (c->vChrFilterPos[chrI] + c->vChrBufSize <
1498 (nextSlice >> c->chrSrcVSubSample))
1499 c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -
1500 c->vChrFilterPos[chrI];
1503 for (i = 0; i < 4; i++)
1504 FF_ALLOCZ_OR_GOTO(c, c->dither_error[i], (c->dstW+2) * sizeof(int), fail);
1506 /* Allocate pixbufs (we use dynamic allocation because otherwise we would
1507 * need to allocate several megabytes to handle all possible cases) */
1508 FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);
1509 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
1510 FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
1511 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
1512 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);
1513 /* Note we need at least one pixel more at the end because of the MMX code
1514 * (just in case someone wants to replace the 4000/8000). */
1515 /* align at 16 bytes for AltiVec */
1516 for (i = 0; i < c->vLumBufSize; i++) {
1517 FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],
1518 dst_stride + 16, fail);
1519 c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];
1521 // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)
1522 c->uv_off = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);
1523 c->uv_offx2 = dst_stride + 16;
1524 for (i = 0; i < c->vChrBufSize; i++) {
1525 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],
1526 dst_stride * 2 + 32, fail);
1527 c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];
1528 c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]
1529 = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;
1531 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
1532 for (i = 0; i < c->vLumBufSize; i++) {
1533 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i + c->vLumBufSize],
1534 dst_stride + 16, fail);
1535 c->alpPixBuf[i] = c->alpPixBuf[i + c->vLumBufSize];
1538 // try to avoid drawing green stuff between the right end and the stride end
1539 for (i = 0; i < c->vChrBufSize; i++)
1540 if(desc_dst->comp[0].depth_minus1 == 15){
1541 av_assert0(c->dstBpc > 14);
1542 for(j=0; j<dst_stride/2+1; j++)
1543 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;
1545 for(j=0; j<dst_stride+1; j++)
1546 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;
1548 av_assert0(c->chrDstH <= dstH);
1550 if (flags & SWS_PRINT_INFO) {
1551 if (flags & SWS_FAST_BILINEAR)
1552 av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
1553 else if (flags & SWS_BILINEAR)
1554 av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
1555 else if (flags & SWS_BICUBIC)
1556 av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
1557 else if (flags & SWS_X)
1558 av_log(c, AV_LOG_INFO, "Experimental scaler, ");
1559 else if (flags & SWS_POINT)
1560 av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
1561 else if (flags & SWS_AREA)
1562 av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
1563 else if (flags & SWS_BICUBLIN)
1564 av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
1565 else if (flags & SWS_GAUSS)
1566 av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
1567 else if (flags & SWS_SINC)
1568 av_log(c, AV_LOG_INFO, "Sinc scaler, ");
1569 else if (flags & SWS_LANCZOS)
1570 av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
1571 else if (flags & SWS_SPLINE)
1572 av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
1574 av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
1576 av_log(c, AV_LOG_INFO, "from %s to %s%s ",
1577 av_get_pix_fmt_name(srcFormat),
1579 dstFormat == AV_PIX_FMT_BGR555 || dstFormat == AV_PIX_FMT_BGR565 ||
1580 dstFormat == AV_PIX_FMT_RGB444BE || dstFormat == AV_PIX_FMT_RGB444LE ||
1581 dstFormat == AV_PIX_FMT_BGR444BE || dstFormat == AV_PIX_FMT_BGR444LE ?
1586 av_get_pix_fmt_name(dstFormat));
1588 if (INLINE_MMXEXT(cpu_flags))
1589 av_log(c, AV_LOG_INFO, "using MMXEXT\n");
1590 else if (INLINE_AMD3DNOW(cpu_flags))
1591 av_log(c, AV_LOG_INFO, "using 3DNOW\n");
1592 else if (INLINE_MMX(cpu_flags))
1593 av_log(c, AV_LOG_INFO, "using MMX\n");
1594 else if (PPC_ALTIVEC(cpu_flags))
1595 av_log(c, AV_LOG_INFO, "using AltiVec\n");
1597 av_log(c, AV_LOG_INFO, "using C\n");
1599 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
1600 av_log(c, AV_LOG_DEBUG,
1601 "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1602 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
1603 av_log(c, AV_LOG_DEBUG,
1604 "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1605 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,
1606 c->chrXInc, c->chrYInc);
1609 c->swScale = ff_getSwsFunc(c);
1611 fail: // FIXME replace things by appropriate error codes
1615 #if FF_API_SWS_GETCONTEXT
1616 SwsContext *sws_getContext(int srcW, int srcH, enum AVPixelFormat srcFormat,
1617 int dstW, int dstH, enum AVPixelFormat dstFormat,
1618 int flags, SwsFilter *srcFilter,
1619 SwsFilter *dstFilter, const double *param)
1623 if (!(c = sws_alloc_context()))
1631 c->srcFormat = srcFormat;
1632 c->dstFormat = dstFormat;
1635 c->param[0] = param[0];
1636 c->param[1] = param[1];
1639 if (sws_init_context(c, srcFilter, dstFilter) < 0) {
1648 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
1649 float lumaSharpen, float chromaSharpen,
1650 float chromaHShift, float chromaVShift,
1653 SwsFilter *filter = av_malloc(sizeof(SwsFilter));
1657 if (lumaGBlur != 0.0) {
1658 filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);
1659 filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);
1661 filter->lumH = sws_getIdentityVec();
1662 filter->lumV = sws_getIdentityVec();
1665 if (chromaGBlur != 0.0) {
1666 filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);
1667 filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);
1669 filter->chrH = sws_getIdentityVec();
1670 filter->chrV = sws_getIdentityVec();
1673 if (chromaSharpen != 0.0) {
1674 SwsVector *id = sws_getIdentityVec();
1675 sws_scaleVec(filter->chrH, -chromaSharpen);
1676 sws_scaleVec(filter->chrV, -chromaSharpen);
1677 sws_addVec(filter->chrH, id);
1678 sws_addVec(filter->chrV, id);
1682 if (lumaSharpen != 0.0) {
1683 SwsVector *id = sws_getIdentityVec();
1684 sws_scaleVec(filter->lumH, -lumaSharpen);
1685 sws_scaleVec(filter->lumV, -lumaSharpen);
1686 sws_addVec(filter->lumH, id);
1687 sws_addVec(filter->lumV, id);
1691 if (chromaHShift != 0.0)
1692 sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));
1694 if (chromaVShift != 0.0)
1695 sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));
1697 sws_normalizeVec(filter->chrH, 1.0);
1698 sws_normalizeVec(filter->chrV, 1.0);
1699 sws_normalizeVec(filter->lumH, 1.0);
1700 sws_normalizeVec(filter->lumV, 1.0);
1703 sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
1705 sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
1710 SwsVector *sws_allocVec(int length)
1714 if(length <= 0 || length > INT_MAX/ sizeof(double))
1717 vec = av_malloc(sizeof(SwsVector));
1720 vec->length = length;
1721 vec->coeff = av_malloc(sizeof(double) * length);
1727 SwsVector *sws_getGaussianVec(double variance, double quality)
1729 const int length = (int)(variance * quality + 0.5) | 1;
1731 double middle = (length - 1) * 0.5;
1734 if(variance < 0 || quality < 0)
1737 vec = sws_allocVec(length);
1742 for (i = 0; i < length; i++) {
1743 double dist = i - middle;
1744 vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /
1745 sqrt(2 * variance * M_PI);
1748 sws_normalizeVec(vec, 1.0);
1753 SwsVector *sws_getConstVec(double c, int length)
1756 SwsVector *vec = sws_allocVec(length);
1761 for (i = 0; i < length; i++)
1767 SwsVector *sws_getIdentityVec(void)
1769 return sws_getConstVec(1.0, 1);
1772 static double sws_dcVec(SwsVector *a)
1777 for (i = 0; i < a->length; i++)
1783 void sws_scaleVec(SwsVector *a, double scalar)
1787 for (i = 0; i < a->length; i++)
1788 a->coeff[i] *= scalar;
1791 void sws_normalizeVec(SwsVector *a, double height)
1793 sws_scaleVec(a, height / sws_dcVec(a));
1796 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
1798 int length = a->length + b->length - 1;
1800 SwsVector *vec = sws_getConstVec(0.0, length);
1805 for (i = 0; i < a->length; i++) {
1806 for (j = 0; j < b->length; j++) {
1807 vec->coeff[i + j] += a->coeff[i] * b->coeff[j];
1814 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
1816 int length = FFMAX(a->length, b->length);
1818 SwsVector *vec = sws_getConstVec(0.0, length);
1823 for (i = 0; i < a->length; i++)
1824 vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
1825 for (i = 0; i < b->length; i++)
1826 vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] += b->coeff[i];
1831 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
1833 int length = FFMAX(a->length, b->length);
1835 SwsVector *vec = sws_getConstVec(0.0, length);
1840 for (i = 0; i < a->length; i++)
1841 vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
1842 for (i = 0; i < b->length; i++)
1843 vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] -= b->coeff[i];
1848 /* shift left / or right if "shift" is negative */
1849 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
1851 int length = a->length + FFABS(shift) * 2;
1853 SwsVector *vec = sws_getConstVec(0.0, length);
1858 for (i = 0; i < a->length; i++) {
1859 vec->coeff[i + (length - 1) / 2 -
1860 (a->length - 1) / 2 - shift] = a->coeff[i];
1866 void sws_shiftVec(SwsVector *a, int shift)
1868 SwsVector *shifted = sws_getShiftedVec(a, shift);
1870 a->coeff = shifted->coeff;
1871 a->length = shifted->length;
1875 void sws_addVec(SwsVector *a, SwsVector *b)
1877 SwsVector *sum = sws_sumVec(a, b);
1879 a->coeff = sum->coeff;
1880 a->length = sum->length;
1884 void sws_subVec(SwsVector *a, SwsVector *b)
1886 SwsVector *diff = sws_diffVec(a, b);
1888 a->coeff = diff->coeff;
1889 a->length = diff->length;
1893 void sws_convVec(SwsVector *a, SwsVector *b)
1895 SwsVector *conv = sws_getConvVec(a, b);
1897 a->coeff = conv->coeff;
1898 a->length = conv->length;
1902 SwsVector *sws_cloneVec(SwsVector *a)
1905 SwsVector *vec = sws_allocVec(a->length);
1910 for (i = 0; i < a->length; i++)
1911 vec->coeff[i] = a->coeff[i];
1916 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
1923 for (i = 0; i < a->length; i++)
1924 if (a->coeff[i] > max)
1927 for (i = 0; i < a->length; i++)
1928 if (a->coeff[i] < min)
1933 for (i = 0; i < a->length; i++) {
1934 int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);
1935 av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
1937 av_log(log_ctx, log_level, " ");
1938 av_log(log_ctx, log_level, "|\n");
1942 void sws_freeVec(SwsVector *a)
1946 av_freep(&a->coeff);
1951 void sws_freeFilter(SwsFilter *filter)
1957 sws_freeVec(filter->lumH);
1959 sws_freeVec(filter->lumV);
1961 sws_freeVec(filter->chrH);
1963 sws_freeVec(filter->chrV);
1967 void sws_freeContext(SwsContext *c)
1974 for (i = 0; i < c->vLumBufSize; i++)
1975 av_freep(&c->lumPixBuf[i]);
1976 av_freep(&c->lumPixBuf);
1979 if (c->chrUPixBuf) {
1980 for (i = 0; i < c->vChrBufSize; i++)
1981 av_freep(&c->chrUPixBuf[i]);
1982 av_freep(&c->chrUPixBuf);
1983 av_freep(&c->chrVPixBuf);
1986 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1987 for (i = 0; i < c->vLumBufSize; i++)
1988 av_freep(&c->alpPixBuf[i]);
1989 av_freep(&c->alpPixBuf);
1992 for (i = 0; i < 4; i++)
1993 av_freep(&c->dither_error[i]);
1995 av_freep(&c->vLumFilter);
1996 av_freep(&c->vChrFilter);
1997 av_freep(&c->hLumFilter);
1998 av_freep(&c->hChrFilter);
2000 av_freep(&c->vYCoeffsBank);
2001 av_freep(&c->vCCoeffsBank);
2004 av_freep(&c->vLumFilterPos);
2005 av_freep(&c->vChrFilterPos);
2006 av_freep(&c->hLumFilterPos);
2007 av_freep(&c->hChrFilterPos);
2011 if (c->lumMmxextFilterCode)
2012 munmap(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize);
2013 if (c->chrMmxextFilterCode)
2014 munmap(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize);
2015 #elif HAVE_VIRTUALALLOC
2016 if (c->lumMmxextFilterCode)
2017 VirtualFree(c->lumMmxextFilterCode, 0, MEM_RELEASE);
2018 if (c->chrMmxextFilterCode)
2019 VirtualFree(c->chrMmxextFilterCode, 0, MEM_RELEASE);
2021 av_free(c->lumMmxextFilterCode);
2022 av_free(c->chrMmxextFilterCode);
2024 c->lumMmxextFilterCode = NULL;
2025 c->chrMmxextFilterCode = NULL;
2026 #endif /* HAVE_MMX_INLINE */
2028 av_freep(&c->yuvTable);
2029 av_freep(&c->formatConvBuffer);
2034 struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,
2035 int srcH, enum AVPixelFormat srcFormat,
2037 enum AVPixelFormat dstFormat, int flags,
2038 SwsFilter *srcFilter,
2039 SwsFilter *dstFilter,
2040 const double *param)
2042 static const double default_param[2] = { SWS_PARAM_DEFAULT,
2043 SWS_PARAM_DEFAULT };
2046 param = default_param;
2049 (context->srcW != srcW ||
2050 context->srcH != srcH ||
2051 context->srcFormat != srcFormat ||
2052 context->dstW != dstW ||
2053 context->dstH != dstH ||
2054 context->dstFormat != dstFormat ||
2055 context->flags != flags ||
2056 context->param[0] != param[0] ||
2057 context->param[1] != param[1])) {
2058 sws_freeContext(context);
2063 if (!(context = sws_alloc_context()))
2065 context->srcW = srcW;
2066 context->srcH = srcH;
2067 context->srcFormat = srcFormat;
2068 context->dstW = dstW;
2069 context->dstH = dstH;
2070 context->dstFormat = dstFormat;
2071 context->flags = flags;
2072 context->param[0] = param[0];
2073 context->param[1] = param[1];
2074 if (sws_init_context(context, srcFilter, dstFilter) < 0) {
2075 sws_freeContext(context);