2 * Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
4 * This file is part of FFmpeg.
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 #define _SVID_SOURCE // needed for MAP_ANONYMOUS
24 #define _DARWIN_C_SOURCE // needed for MAP_ANON
31 #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
32 #define MAP_ANONYMOUS MAP_ANON
36 #define WIN32_LEAN_AND_MEAN
40 #include "libavutil/attributes.h"
41 #include "libavutil/avassert.h"
42 #include "libavutil/avutil.h"
43 #include "libavutil/bswap.h"
44 #include "libavutil/cpu.h"
45 #include "libavutil/intreadwrite.h"
46 #include "libavutil/mathematics.h"
47 #include "libavutil/opt.h"
48 #include "libavutil/pixdesc.h"
49 #include "libavutil/x86/asm.h"
50 #include "libavutil/x86/cpu.h"
53 #include "swscale_internal.h"
55 unsigned swscale_version(void)
57 av_assert0(LIBSWSCALE_VERSION_MICRO >= 100);
58 return LIBSWSCALE_VERSION_INT;
61 const char *swscale_configuration(void)
63 return FFMPEG_CONFIGURATION;
66 const char *swscale_license(void)
68 #define LICENSE_PREFIX "libswscale license: "
69 return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
72 #define RET 0xC3 // near return opcode for x86
74 typedef struct FormatEntry {
75 int is_supported_in, is_supported_out;
78 static const FormatEntry format_entries[AV_PIX_FMT_NB] = {
79 [AV_PIX_FMT_YUV420P] = { 1, 1 },
80 [AV_PIX_FMT_YUYV422] = { 1, 1 },
81 [AV_PIX_FMT_RGB24] = { 1, 1 },
82 [AV_PIX_FMT_BGR24] = { 1, 1 },
83 [AV_PIX_FMT_YUV422P] = { 1, 1 },
84 [AV_PIX_FMT_YUV444P] = { 1, 1 },
85 [AV_PIX_FMT_YUV410P] = { 1, 1 },
86 [AV_PIX_FMT_YUV411P] = { 1, 1 },
87 [AV_PIX_FMT_GRAY8] = { 1, 1 },
88 [AV_PIX_FMT_MONOWHITE] = { 1, 1 },
89 [AV_PIX_FMT_MONOBLACK] = { 1, 1 },
90 [AV_PIX_FMT_PAL8] = { 1, 0 },
91 [AV_PIX_FMT_YUVJ420P] = { 1, 1 },
92 [AV_PIX_FMT_YUVJ422P] = { 1, 1 },
93 [AV_PIX_FMT_YUVJ444P] = { 1, 1 },
94 [AV_PIX_FMT_UYVY422] = { 1, 1 },
95 [AV_PIX_FMT_UYYVYY411] = { 0, 0 },
96 [AV_PIX_FMT_BGR8] = { 1, 1 },
97 [AV_PIX_FMT_BGR4] = { 0, 1 },
98 [AV_PIX_FMT_BGR4_BYTE] = { 1, 1 },
99 [AV_PIX_FMT_RGB8] = { 1, 1 },
100 [AV_PIX_FMT_RGB4] = { 0, 1 },
101 [AV_PIX_FMT_RGB4_BYTE] = { 1, 1 },
102 [AV_PIX_FMT_NV12] = { 1, 1 },
103 [AV_PIX_FMT_NV21] = { 1, 1 },
104 [AV_PIX_FMT_ARGB] = { 1, 1 },
105 [AV_PIX_FMT_RGBA] = { 1, 1 },
106 [AV_PIX_FMT_ABGR] = { 1, 1 },
107 [AV_PIX_FMT_BGRA] = { 1, 1 },
108 [AV_PIX_FMT_0RGB] = { 1, 1 },
109 [AV_PIX_FMT_RGB0] = { 1, 1 },
110 [AV_PIX_FMT_0BGR] = { 1, 1 },
111 [AV_PIX_FMT_BGR0] = { 1, 1 },
112 [AV_PIX_FMT_GRAY16BE] = { 1, 1 },
113 [AV_PIX_FMT_GRAY16LE] = { 1, 1 },
114 [AV_PIX_FMT_YUV440P] = { 1, 1 },
115 [AV_PIX_FMT_YUVJ440P] = { 1, 1 },
116 [AV_PIX_FMT_YUVA420P] = { 1, 1 },
117 [AV_PIX_FMT_YUVA422P] = { 1, 1 },
118 [AV_PIX_FMT_YUVA444P] = { 1, 1 },
119 [AV_PIX_FMT_YUVA420P9BE] = { 1, 1 },
120 [AV_PIX_FMT_YUVA420P9LE] = { 1, 1 },
121 [AV_PIX_FMT_YUVA422P9BE] = { 1, 1 },
122 [AV_PIX_FMT_YUVA422P9LE] = { 1, 1 },
123 [AV_PIX_FMT_YUVA444P9BE] = { 1, 1 },
124 [AV_PIX_FMT_YUVA444P9LE] = { 1, 1 },
125 [AV_PIX_FMT_YUVA420P10BE]= { 1, 1 },
126 [AV_PIX_FMT_YUVA420P10LE]= { 1, 1 },
127 [AV_PIX_FMT_YUVA422P10BE]= { 1, 1 },
128 [AV_PIX_FMT_YUVA422P10LE]= { 1, 1 },
129 [AV_PIX_FMT_YUVA444P10BE]= { 1, 1 },
130 [AV_PIX_FMT_YUVA444P10LE]= { 1, 1 },
131 [AV_PIX_FMT_YUVA420P16BE]= { 1, 1 },
132 [AV_PIX_FMT_YUVA420P16LE]= { 1, 1 },
133 [AV_PIX_FMT_YUVA422P16BE]= { 1, 1 },
134 [AV_PIX_FMT_YUVA422P16LE]= { 1, 1 },
135 [AV_PIX_FMT_YUVA444P16BE]= { 1, 1 },
136 [AV_PIX_FMT_YUVA444P16LE]= { 1, 1 },
137 [AV_PIX_FMT_RGB48BE] = { 1, 1 },
138 [AV_PIX_FMT_RGB48LE] = { 1, 1 },
139 [AV_PIX_FMT_RGBA64BE] = { 1, 0 },
140 [AV_PIX_FMT_RGBA64LE] = { 1, 0 },
141 [AV_PIX_FMT_RGB565BE] = { 1, 1 },
142 [AV_PIX_FMT_RGB565LE] = { 1, 1 },
143 [AV_PIX_FMT_RGB555BE] = { 1, 1 },
144 [AV_PIX_FMT_RGB555LE] = { 1, 1 },
145 [AV_PIX_FMT_BGR565BE] = { 1, 1 },
146 [AV_PIX_FMT_BGR565LE] = { 1, 1 },
147 [AV_PIX_FMT_BGR555BE] = { 1, 1 },
148 [AV_PIX_FMT_BGR555LE] = { 1, 1 },
149 [AV_PIX_FMT_YUV420P16LE] = { 1, 1 },
150 [AV_PIX_FMT_YUV420P16BE] = { 1, 1 },
151 [AV_PIX_FMT_YUV422P16LE] = { 1, 1 },
152 [AV_PIX_FMT_YUV422P16BE] = { 1, 1 },
153 [AV_PIX_FMT_YUV444P16LE] = { 1, 1 },
154 [AV_PIX_FMT_YUV444P16BE] = { 1, 1 },
155 [AV_PIX_FMT_RGB444LE] = { 1, 1 },
156 [AV_PIX_FMT_RGB444BE] = { 1, 1 },
157 [AV_PIX_FMT_BGR444LE] = { 1, 1 },
158 [AV_PIX_FMT_BGR444BE] = { 1, 1 },
159 [AV_PIX_FMT_Y400A] = { 1, 0 },
160 [AV_PIX_FMT_BGR48BE] = { 1, 1 },
161 [AV_PIX_FMT_BGR48LE] = { 1, 1 },
162 [AV_PIX_FMT_BGRA64BE] = { 0, 0 },
163 [AV_PIX_FMT_BGRA64LE] = { 0, 0 },
164 [AV_PIX_FMT_YUV420P9BE] = { 1, 1 },
165 [AV_PIX_FMT_YUV420P9LE] = { 1, 1 },
166 [AV_PIX_FMT_YUV420P10BE] = { 1, 1 },
167 [AV_PIX_FMT_YUV420P10LE] = { 1, 1 },
168 [AV_PIX_FMT_YUV420P12BE] = { 1, 1 },
169 [AV_PIX_FMT_YUV420P12LE] = { 1, 1 },
170 [AV_PIX_FMT_YUV420P14BE] = { 1, 1 },
171 [AV_PIX_FMT_YUV420P14LE] = { 1, 1 },
172 [AV_PIX_FMT_YUV422P9BE] = { 1, 1 },
173 [AV_PIX_FMT_YUV422P9LE] = { 1, 1 },
174 [AV_PIX_FMT_YUV422P10BE] = { 1, 1 },
175 [AV_PIX_FMT_YUV422P10LE] = { 1, 1 },
176 [AV_PIX_FMT_YUV422P12BE] = { 1, 1 },
177 [AV_PIX_FMT_YUV422P12LE] = { 1, 1 },
178 [AV_PIX_FMT_YUV422P14BE] = { 1, 1 },
179 [AV_PIX_FMT_YUV422P14LE] = { 1, 1 },
180 [AV_PIX_FMT_YUV444P9BE] = { 1, 1 },
181 [AV_PIX_FMT_YUV444P9LE] = { 1, 1 },
182 [AV_PIX_FMT_YUV444P10BE] = { 1, 1 },
183 [AV_PIX_FMT_YUV444P10LE] = { 1, 1 },
184 [AV_PIX_FMT_YUV444P12BE] = { 1, 1 },
185 [AV_PIX_FMT_YUV444P12LE] = { 1, 1 },
186 [AV_PIX_FMT_YUV444P14BE] = { 1, 1 },
187 [AV_PIX_FMT_YUV444P14LE] = { 1, 1 },
188 [AV_PIX_FMT_GBRP] = { 1, 1 },
189 [AV_PIX_FMT_GBRP9LE] = { 1, 1 },
190 [AV_PIX_FMT_GBRP9BE] = { 1, 1 },
191 [AV_PIX_FMT_GBRP10LE] = { 1, 1 },
192 [AV_PIX_FMT_GBRP10BE] = { 1, 1 },
193 [AV_PIX_FMT_GBRP12LE] = { 1, 1 },
194 [AV_PIX_FMT_GBRP12BE] = { 1, 1 },
195 [AV_PIX_FMT_GBRP14LE] = { 1, 1 },
196 [AV_PIX_FMT_GBRP14BE] = { 1, 1 },
197 [AV_PIX_FMT_GBRP16LE] = { 1, 0 },
198 [AV_PIX_FMT_GBRP16BE] = { 1, 0 },
199 [AV_PIX_FMT_XYZ12BE] = { 1, 0 },
200 [AV_PIX_FMT_XYZ12LE] = { 1, 0 },
203 int sws_isSupportedInput(enum AVPixelFormat pix_fmt)
205 return (unsigned)pix_fmt < AV_PIX_FMT_NB ?
206 format_entries[pix_fmt].is_supported_in : 0;
209 int sws_isSupportedOutput(enum AVPixelFormat pix_fmt)
211 return (unsigned)pix_fmt < AV_PIX_FMT_NB ?
212 format_entries[pix_fmt].is_supported_out : 0;
215 extern const int32_t ff_yuv2rgb_coeffs[8][4];
217 #if FF_API_SWS_FORMAT_NAME
218 const char *sws_format_name(enum AVPixelFormat format)
220 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(format);
224 return "Unknown format";
228 static double getSplineCoeff(double a, double b, double c, double d,
232 return ((d * dist + c) * dist + b) * dist + a;
234 return getSplineCoeff(0.0,
235 b + 2.0 * c + 3.0 * d,
237 -b - 3.0 * c - 6.0 * d,
241 static int initFilter(int16_t **outFilter, int32_t **filterPos,
242 int *outFilterSize, int xInc, int srcW, int dstW,
243 int filterAlign, int one, int flags, int cpu_flags,
244 SwsVector *srcFilter, SwsVector *dstFilter,
251 int64_t *filter = NULL;
252 int64_t *filter2 = NULL;
253 const int64_t fone = 1LL << (54 - FFMIN(av_log2(srcW/dstW), 8));
256 emms_c(); // FIXME should not be required but IS (even for non-MMX versions)
258 // NOTE: the +3 is for the MMX(+1) / SSE(+3) scaler which reads over the end
259 FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW + 3) * sizeof(**filterPos), fail);
261 if (FFABS(xInc - 0x10000) < 10) { // unscaled
264 FF_ALLOCZ_OR_GOTO(NULL, filter,
265 dstW * sizeof(*filter) * filterSize, fail);
267 for (i = 0; i < dstW; i++) {
268 filter[i * filterSize] = fone;
271 } else if (flags & SWS_POINT) { // lame looking point sampling mode
275 FF_ALLOC_OR_GOTO(NULL, filter,
276 dstW * sizeof(*filter) * filterSize, fail);
278 xDstInSrc = xInc / 2 - 0x8000;
279 for (i = 0; i < dstW; i++) {
280 int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;
282 (*filterPos)[i] = xx;
286 } else if ((xInc <= (1 << 16) && (flags & SWS_AREA)) ||
287 (flags & SWS_FAST_BILINEAR)) { // bilinear upscale
291 FF_ALLOC_OR_GOTO(NULL, filter,
292 dstW * sizeof(*filter) * filterSize, fail);
294 xDstInSrc = xInc / 2 - 0x8000;
295 for (i = 0; i < dstW; i++) {
296 int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;
299 (*filterPos)[i] = xx;
300 // bilinear upscale / linear interpolate / area averaging
301 for (j = 0; j < filterSize; j++) {
302 int64_t coeff= fone - FFABS(((int64_t)xx<<16) - xDstInSrc)*(fone>>16);
305 filter[i * filterSize + j] = coeff;
314 if (flags & SWS_BICUBIC)
316 else if (flags & SWS_X)
318 else if (flags & SWS_AREA)
319 sizeFactor = 1; // downscale only, for upscale it is bilinear
320 else if (flags & SWS_GAUSS)
321 sizeFactor = 8; // infinite ;)
322 else if (flags & SWS_LANCZOS)
323 sizeFactor = param[0] != SWS_PARAM_DEFAULT ? ceil(2 * param[0]) : 6;
324 else if (flags & SWS_SINC)
325 sizeFactor = 20; // infinite ;)
326 else if (flags & SWS_SPLINE)
327 sizeFactor = 20; // infinite ;)
328 else if (flags & SWS_BILINEAR)
335 filterSize = 1 + sizeFactor; // upscale
337 filterSize = 1 + (sizeFactor * srcW + dstW - 1) / dstW;
339 filterSize = FFMIN(filterSize, srcW - 2);
340 filterSize = FFMAX(filterSize, 1);
342 FF_ALLOC_OR_GOTO(NULL, filter,
343 dstW * sizeof(*filter) * filterSize, fail);
345 xDstInSrc = xInc - 0x10000;
346 for (i = 0; i < dstW; i++) {
347 int xx = (xDstInSrc - ((filterSize - 2) << 16)) / (1 << 17);
349 (*filterPos)[i] = xx;
350 for (j = 0; j < filterSize; j++) {
351 int64_t d = (FFABS(((int64_t)xx << 17) - xDstInSrc)) << 13;
357 floatd = d * (1.0 / (1 << 30));
359 if (flags & SWS_BICUBIC) {
360 int64_t B = (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1 << 24);
361 int64_t C = (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1 << 24);
363 if (d >= 1LL << 31) {
366 int64_t dd = (d * d) >> 30;
367 int64_t ddd = (dd * d) >> 30;
370 coeff = (12 * (1 << 24) - 9 * B - 6 * C) * ddd +
371 (-18 * (1 << 24) + 12 * B + 6 * C) * dd +
372 (6 * (1 << 24) - 2 * B) * (1 << 30);
374 coeff = (-B - 6 * C) * ddd +
375 (6 * B + 30 * C) * dd +
376 (-12 * B - 48 * C) * d +
377 (8 * B + 24 * C) * (1 << 30);
379 coeff /= (1LL<<54)/fone;
382 else if (flags & SWS_X) {
383 double p = param ? param * 0.01 : 0.3;
384 coeff = d ? sin(d * M_PI) / (d * M_PI) : 1.0;
385 coeff *= pow(2.0, -p * d * d);
388 else if (flags & SWS_X) {
389 double A = param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
393 c = cos(floatd * M_PI);
400 coeff = (c * 0.5 + 0.5) * fone;
401 } else if (flags & SWS_AREA) {
402 int64_t d2 = d - (1 << 29);
403 if (d2 * xInc < -(1LL << (29 + 16)))
404 coeff = 1.0 * (1LL << (30 + 16));
405 else if (d2 * xInc < (1LL << (29 + 16)))
406 coeff = -d2 * xInc + (1LL << (29 + 16));
409 coeff *= fone >> (30 + 16);
410 } else if (flags & SWS_GAUSS) {
411 double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
412 coeff = (pow(2.0, -p * floatd * floatd)) * fone;
413 } else if (flags & SWS_SINC) {
414 coeff = (d ? sin(floatd * M_PI) / (floatd * M_PI) : 1.0) * fone;
415 } else if (flags & SWS_LANCZOS) {
416 double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
417 coeff = (d ? sin(floatd * M_PI) * sin(floatd * M_PI / p) /
418 (floatd * floatd * M_PI * M_PI / p) : 1.0) * fone;
421 } else if (flags & SWS_BILINEAR) {
422 coeff = (1 << 30) - d;
426 } else if (flags & SWS_SPLINE) {
427 double p = -2.196152422706632;
428 coeff = getSplineCoeff(1.0, 0.0, p, -p - 1.0, floatd) * fone;
433 filter[i * filterSize + j] = coeff;
436 xDstInSrc += 2 * xInc;
440 /* apply src & dst Filter to filter -> filter2
443 av_assert0(filterSize > 0);
444 filter2Size = filterSize;
446 filter2Size += srcFilter->length - 1;
448 filter2Size += dstFilter->length - 1;
449 av_assert0(filter2Size > 0);
450 FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size * dstW * sizeof(*filter2), fail);
452 for (i = 0; i < dstW; i++) {
456 for (k = 0; k < srcFilter->length; k++) {
457 for (j = 0; j < filterSize; j++)
458 filter2[i * filter2Size + k + j] +=
459 srcFilter->coeff[k] * filter[i * filterSize + j];
462 for (j = 0; j < filterSize; j++)
463 filter2[i * filter2Size + j] = filter[i * filterSize + j];
467 (*filterPos)[i] += (filterSize - 1) / 2 - (filter2Size - 1) / 2;
471 /* try to reduce the filter-size (step1 find size and shift left) */
472 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
474 for (i = dstW - 1; i >= 0; i--) {
475 int min = filter2Size;
477 int64_t cutOff = 0.0;
479 /* get rid of near zero elements on the left by shifting left */
480 for (j = 0; j < filter2Size; j++) {
482 cutOff += FFABS(filter2[i * filter2Size]);
484 if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)
487 /* preserve monotonicity because the core can't handle the
488 * filter otherwise */
489 if (i < dstW - 1 && (*filterPos)[i] >= (*filterPos)[i + 1])
492 // move filter coefficients left
493 for (k = 1; k < filter2Size; k++)
494 filter2[i * filter2Size + k - 1] = filter2[i * filter2Size + k];
495 filter2[i * filter2Size + k - 1] = 0;
500 /* count near zeros on the right */
501 for (j = filter2Size - 1; j > 0; j--) {
502 cutOff += FFABS(filter2[i * filter2Size + j]);
504 if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)
509 if (min > minFilterSize)
513 if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) {
514 // we can handle the special case 4, so we don't want to go the full 8
515 if (minFilterSize < 5)
518 /* We really don't want to waste our time doing useless computation, so
519 * fall back on the scalar C code for very small filters.
520 * Vectorizing is worth it only if you have a decent-sized vector. */
521 if (minFilterSize < 3)
525 if (INLINE_MMX(cpu_flags)) {
526 // special case for unscaled vertical filtering
527 if (minFilterSize == 1 && filterAlign == 2)
531 av_assert0(minFilterSize > 0);
532 filterSize = (minFilterSize + (filterAlign - 1)) & (~(filterAlign - 1));
533 av_assert0(filterSize > 0);
534 filter = av_malloc(filterSize * dstW * sizeof(*filter));
535 if (filterSize >= MAX_FILTER_SIZE * 16 /
536 ((flags & SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter) {
537 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);
540 *outFilterSize = filterSize;
542 if (flags & SWS_PRINT_INFO)
543 av_log(NULL, AV_LOG_VERBOSE,
544 "SwScaler: reducing / aligning filtersize %d -> %d\n",
545 filter2Size, filterSize);
546 /* try to reduce the filter-size (step2 reduce it) */
547 for (i = 0; i < dstW; i++) {
550 for (j = 0; j < filterSize; j++) {
551 if (j >= filter2Size)
552 filter[i * filterSize + j] = 0;
554 filter[i * filterSize + j] = filter2[i * filter2Size + j];
555 if ((flags & SWS_BITEXACT) && j >= minFilterSize)
556 filter[i * filterSize + j] = 0;
560 // FIXME try to align filterPos if possible
563 for (i = 0; i < dstW; i++) {
565 if ((*filterPos)[i] < 0) {
566 // move filter coefficients left to compensate for filterPos
567 for (j = 1; j < filterSize; j++) {
568 int left = FFMAX(j + (*filterPos)[i], 0);
569 filter[i * filterSize + left] += filter[i * filterSize + j];
570 filter[i * filterSize + j] = 0;
575 if ((*filterPos)[i] + filterSize > srcW) {
576 int shift = (*filterPos)[i] + filterSize - srcW;
577 // move filter coefficients right to compensate for filterPos
578 for (j = filterSize - 2; j >= 0; j--) {
579 int right = FFMIN(j + shift, filterSize - 1);
580 filter[i * filterSize + right] += filter[i * filterSize + j];
581 filter[i * filterSize + j] = 0;
583 (*filterPos)[i]= srcW - filterSize;
587 // Note the +1 is for the MMX scaler which reads over the end
588 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
589 FF_ALLOCZ_OR_GOTO(NULL, *outFilter,
590 *outFilterSize * (dstW + 3) * sizeof(int16_t), fail);
592 /* normalize & store in outFilter */
593 for (i = 0; i < dstW; i++) {
598 for (j = 0; j < filterSize; j++) {
599 sum += filter[i * filterSize + j];
601 sum = (sum + one / 2) / one;
602 for (j = 0; j < *outFilterSize; j++) {
603 int64_t v = filter[i * filterSize + j] + error;
604 int intV = ROUNDED_DIV(v, sum);
605 (*outFilter)[i * (*outFilterSize) + j] = intV;
606 error = v - intV * sum;
610 (*filterPos)[dstW + 0] =
611 (*filterPos)[dstW + 1] =
612 (*filterPos)[dstW + 2] = (*filterPos)[dstW - 1]; /* the MMX/SSE scaler will
613 * read over the end */
614 for (i = 0; i < *outFilterSize; i++) {
615 int k = (dstW - 1) * (*outFilterSize) + i;
616 (*outFilter)[k + 1 * (*outFilterSize)] =
617 (*outFilter)[k + 2 * (*outFilterSize)] =
618 (*outFilter)[k + 3 * (*outFilterSize)] = (*outFilter)[k];
625 av_log(NULL, AV_LOG_ERROR, "sws: initFilter failed\n");
631 #if HAVE_MMXEXT_INLINE
632 static int init_hscaler_mmxext(int dstW, int xInc, uint8_t *filterCode,
633 int16_t *filter, int32_t *filterPos,
637 x86_reg imm8OfPShufW1A;
638 x86_reg imm8OfPShufW2A;
639 x86_reg fragmentLengthA;
641 x86_reg imm8OfPShufW1B;
642 x86_reg imm8OfPShufW2B;
643 x86_reg fragmentLengthB;
648 // create an optimized horizontal scaling routine
649 /* This scaler is made of runtime-generated MMXEXT code using specially tuned
650 * pshufw instructions. For every four output pixels, if four input pixels
651 * are enough for the fast bilinear scaling, then a chunk of fragmentB is
652 * used. If five input pixels are needed, then a chunk of fragmentA is used.
661 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
662 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
663 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
664 "punpcklbw %%mm7, %%mm1 \n\t"
665 "punpcklbw %%mm7, %%mm0 \n\t"
666 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
668 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
670 "psubw %%mm1, %%mm0 \n\t"
671 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
672 "pmullw %%mm3, %%mm0 \n\t"
673 "psllw $7, %%mm1 \n\t"
674 "paddw %%mm1, %%mm0 \n\t"
676 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
678 "add $8, %%"REG_a" \n\t"
682 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
683 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
684 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
689 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
693 : "=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
694 "=r" (fragmentLengthA)
701 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
702 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
703 "punpcklbw %%mm7, %%mm0 \n\t"
704 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
706 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
708 "psubw %%mm1, %%mm0 \n\t"
709 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
710 "pmullw %%mm3, %%mm0 \n\t"
711 "psllw $7, %%mm1 \n\t"
712 "paddw %%mm1, %%mm0 \n\t"
714 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
716 "add $8, %%"REG_a" \n\t"
720 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
721 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
722 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
727 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
731 : "=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
732 "=r" (fragmentLengthB)
735 xpos = 0; // lumXInc/2 - 0x8000; // difference between pixel centers
738 for (i = 0; i < dstW / numSplits; i++) {
743 int b = ((xpos + xInc) >> 16) - xx;
744 int c = ((xpos + xInc * 2) >> 16) - xx;
745 int d = ((xpos + xInc * 3) >> 16) - xx;
746 int inc = (d + 1 < 4);
747 uint8_t *fragment = (d + 1 < 4) ? fragmentB : fragmentA;
748 x86_reg imm8OfPShufW1 = (d + 1 < 4) ? imm8OfPShufW1B : imm8OfPShufW1A;
749 x86_reg imm8OfPShufW2 = (d + 1 < 4) ? imm8OfPShufW2B : imm8OfPShufW2A;
750 x86_reg fragmentLength = (d + 1 < 4) ? fragmentLengthB : fragmentLengthA;
751 int maxShift = 3 - (d + inc);
755 filter[i] = ((xpos & 0xFFFF) ^ 0xFFFF) >> 9;
756 filter[i + 1] = (((xpos + xInc) & 0xFFFF) ^ 0xFFFF) >> 9;
757 filter[i + 2] = (((xpos + xInc * 2) & 0xFFFF) ^ 0xFFFF) >> 9;
758 filter[i + 3] = (((xpos + xInc * 3) & 0xFFFF) ^ 0xFFFF) >> 9;
759 filterPos[i / 2] = xx;
761 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
763 filterCode[fragmentPos + imm8OfPShufW1] = (a + inc) |
767 filterCode[fragmentPos + imm8OfPShufW2] = a | (b << 2) |
771 if (i + 4 - inc >= dstW)
772 shift = maxShift; // avoid overread
773 else if ((filterPos[i / 2] & 3) <= maxShift)
774 shift = filterPos[i / 2] & 3; // align
776 if (shift && i >= shift) {
777 filterCode[fragmentPos + imm8OfPShufW1] += 0x55 * shift;
778 filterCode[fragmentPos + imm8OfPShufW2] += 0x55 * shift;
779 filterPos[i / 2] -= shift;
783 fragmentPos += fragmentLength;
786 filterCode[fragmentPos] = RET;
791 filterPos[((i / 2) + 1) & (~1)] = xpos >> 16; // needed to jump to the next part
793 return fragmentPos + 1;
795 #endif /* HAVE_MMXEXT_INLINE */
797 static void getSubSampleFactors(int *h, int *v, enum AVPixelFormat format)
799 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(format);
800 *h = desc->log2_chroma_w;
801 *v = desc->log2_chroma_h;
804 static void fill_rgb2yuv_table(SwsContext *c, const int table[4], int dstRange)
806 int64_t W, V, Z, Cy, Cu, Cv;
807 int64_t vr = table[0];
808 int64_t ub = table[1];
809 int64_t ug = -table[2];
810 int64_t vg = -table[3];
813 uint8_t *p = (uint8_t*)c->input_rgb2yuv_table;
815 static const int8_t map[] = {
816 BY_IDX, GY_IDX, -1 , BY_IDX, BY_IDX, GY_IDX, -1 , BY_IDX,
817 RY_IDX, -1 , GY_IDX, RY_IDX, RY_IDX, -1 , GY_IDX, RY_IDX,
818 RY_IDX, GY_IDX, -1 , RY_IDX, RY_IDX, GY_IDX, -1 , RY_IDX,
819 BY_IDX, -1 , GY_IDX, BY_IDX, BY_IDX, -1 , GY_IDX, BY_IDX,
820 BU_IDX, GU_IDX, -1 , BU_IDX, BU_IDX, GU_IDX, -1 , BU_IDX,
821 RU_IDX, -1 , GU_IDX, RU_IDX, RU_IDX, -1 , GU_IDX, RU_IDX,
822 RU_IDX, GU_IDX, -1 , RU_IDX, RU_IDX, GU_IDX, -1 , RU_IDX,
823 BU_IDX, -1 , GU_IDX, BU_IDX, BU_IDX, -1 , GU_IDX, BU_IDX,
824 BV_IDX, GV_IDX, -1 , BV_IDX, BV_IDX, GV_IDX, -1 , BV_IDX,
825 RV_IDX, -1 , GV_IDX, RV_IDX, RV_IDX, -1 , GV_IDX, RV_IDX,
826 RV_IDX, GV_IDX, -1 , RV_IDX, RV_IDX, GV_IDX, -1 , RV_IDX,
827 BV_IDX, -1 , GV_IDX, BV_IDX, BV_IDX, -1 , GV_IDX, BV_IDX,
828 RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX,
829 BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX,
830 GY_IDX, -1 , GY_IDX, -1 , GY_IDX, -1 , GY_IDX, -1 ,
831 -1 , GY_IDX, -1 , GY_IDX, -1 , GY_IDX, -1 , GY_IDX,
832 RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX,
833 BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX,
834 GU_IDX, -1 , GU_IDX, -1 , GU_IDX, -1 , GU_IDX, -1 ,
835 -1 , GU_IDX, -1 , GU_IDX, -1 , GU_IDX, -1 , GU_IDX,
836 RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX,
837 BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX,
838 GV_IDX, -1 , GV_IDX, -1 , GV_IDX, -1 , GV_IDX, -1 ,
839 -1 , GV_IDX, -1 , GV_IDX, -1 , GV_IDX, -1 , GV_IDX, //23
840 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //24
841 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //25
842 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //26
843 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //27
844 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //28
845 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //29
846 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //30
847 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //31
848 BY_IDX, GY_IDX, RY_IDX, -1 , -1 , -1 , -1 , -1 , //32
849 BU_IDX, GU_IDX, RU_IDX, -1 , -1 , -1 , -1 , -1 , //33
850 BV_IDX, GV_IDX, RV_IDX, -1 , -1 , -1 , -1 , -1 , //34
853 dstRange = 0; //FIXME range = 1 is handled elsewhere
863 W = ROUNDED_DIV(ONE*ONE*ug, ub);
864 V = ROUNDED_DIV(ONE*ONE*vg, vr);
867 Cy = ROUNDED_DIV(cy*Z, ONE);
868 Cu = ROUNDED_DIV(ub*Z, ONE);
869 Cv = ROUNDED_DIV(vr*Z, ONE);
871 c->input_rgb2yuv_table[RY_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*V , Cy);
872 c->input_rgb2yuv_table[GY_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE , Cy);
873 c->input_rgb2yuv_table[BY_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*W , Cy);
875 c->input_rgb2yuv_table[RU_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*V , Cu);
876 c->input_rgb2yuv_table[GU_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE , Cu);
877 c->input_rgb2yuv_table[BU_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*(Z+W) , Cu);
879 c->input_rgb2yuv_table[RV_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*(V+Z) , Cv);
880 c->input_rgb2yuv_table[GV_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE , Cv);
881 c->input_rgb2yuv_table[BV_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*W , Cv);
883 if(/*!dstRange && */table == ff_yuv2rgb_coeffs[SWS_CS_DEFAULT]) {
884 c->input_rgb2yuv_table[BY_IDX] = ((int)(0.114 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
885 c->input_rgb2yuv_table[BV_IDX] = (-(int)(0.081 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
886 c->input_rgb2yuv_table[BU_IDX] = ((int)(0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
887 c->input_rgb2yuv_table[GY_IDX] = ((int)(0.587 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
888 c->input_rgb2yuv_table[GV_IDX] = (-(int)(0.419 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
889 c->input_rgb2yuv_table[GU_IDX] = (-(int)(0.331 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
890 c->input_rgb2yuv_table[RY_IDX] = ((int)(0.299 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
891 c->input_rgb2yuv_table[RV_IDX] = ((int)(0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
892 c->input_rgb2yuv_table[RU_IDX] = (-(int)(0.169 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
894 for(i=0; i<FF_ARRAY_ELEMS(map); i++)
895 AV_WL16(p + 16*4 + 2*i, map[i] >= 0 ? c->input_rgb2yuv_table[map[i]] : 0);
898 static void fill_xyztables(struct SwsContext *c)
901 double xyzgamma = XYZ_GAMMA;
902 double rgbgamma = 1.0 / RGB_GAMMA;
903 static const int16_t xyz2rgb_matrix[3][4] = {
904 {13270, -6295, -2041},
906 { 228, -835, 4329} };
907 static int16_t xyzgamma_tab[4096], rgbgamma_tab[4096];
909 memcpy(c->xyz2rgb_matrix, xyz2rgb_matrix, sizeof(c->xyz2rgb_matrix));
910 c->xyzgamma = xyzgamma_tab;
911 c->rgbgamma = rgbgamma_tab;
913 if (rgbgamma_tab[4095])
916 /* set gamma vectors */
917 for (i = 0; i < 4096; i++) {
918 xyzgamma_tab[i] = lrint(pow(i / 4095.0, xyzgamma) * 4095.0);
919 rgbgamma_tab[i] = lrint(pow(i / 4095.0, rgbgamma) * 4095.0);
923 int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],
924 int srcRange, const int table[4], int dstRange,
925 int brightness, int contrast, int saturation)
927 const AVPixFmtDescriptor *desc_dst = av_pix_fmt_desc_get(c->dstFormat);
928 const AVPixFmtDescriptor *desc_src = av_pix_fmt_desc_get(c->srcFormat);
929 memcpy(c->srcColorspaceTable, inv_table, sizeof(int) * 4);
930 memcpy(c->dstColorspaceTable, table, sizeof(int) * 4);
932 if(!isYUV(c->dstFormat) && !isGray(c->dstFormat))
934 if(!isYUV(c->srcFormat) && !isGray(c->srcFormat))
937 c->brightness = brightness;
938 c->contrast = contrast;
939 c->saturation = saturation;
940 c->srcRange = srcRange;
941 c->dstRange = dstRange;
945 if ((isYUV(c->dstFormat) || isGray(c->dstFormat)) && (isYUV(c->srcFormat) || isGray(c->srcFormat)))
948 c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);
949 c->srcFormatBpp = av_get_bits_per_pixel(desc_src);
951 if (!isYUV(c->dstFormat) && !isGray(c->dstFormat)) {
952 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness,
953 contrast, saturation);
956 if (HAVE_ALTIVEC && av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)
957 ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness,
958 contrast, saturation);
961 fill_rgb2yuv_table(c, table, dstRange);
966 int sws_getColorspaceDetails(struct SwsContext *c, int **inv_table,
967 int *srcRange, int **table, int *dstRange,
968 int *brightness, int *contrast, int *saturation)
970 if (!c || isYUV(c->dstFormat) || isGray(c->dstFormat))
973 *inv_table = c->srcColorspaceTable;
974 *table = c->dstColorspaceTable;
975 *srcRange = c->srcRange;
976 *dstRange = c->dstRange;
977 *brightness = c->brightness;
978 *contrast = c->contrast;
979 *saturation = c->saturation;
984 static int handle_jpeg(enum AVPixelFormat *format)
987 case AV_PIX_FMT_YUVJ420P:
988 *format = AV_PIX_FMT_YUV420P;
990 case AV_PIX_FMT_YUVJ422P:
991 *format = AV_PIX_FMT_YUV422P;
993 case AV_PIX_FMT_YUVJ444P:
994 *format = AV_PIX_FMT_YUV444P;
996 case AV_PIX_FMT_YUVJ440P:
997 *format = AV_PIX_FMT_YUV440P;
1004 static int handle_0alpha(enum AVPixelFormat *format)
1007 case AV_PIX_FMT_0BGR : *format = AV_PIX_FMT_ABGR ; return 1;
1008 case AV_PIX_FMT_BGR0 : *format = AV_PIX_FMT_BGRA ; return 4;
1009 case AV_PIX_FMT_0RGB : *format = AV_PIX_FMT_ARGB ; return 1;
1010 case AV_PIX_FMT_RGB0 : *format = AV_PIX_FMT_RGBA ; return 4;
1015 static int handle_xyz(enum AVPixelFormat *format)
1018 case AV_PIX_FMT_XYZ12BE : *format = AV_PIX_FMT_RGB48BE; return 1;
1019 case AV_PIX_FMT_XYZ12LE : *format = AV_PIX_FMT_RGB48LE; return 1;
1024 SwsContext *sws_alloc_context(void)
1026 SwsContext *c = av_mallocz(sizeof(SwsContext));
1029 c->av_class = &sws_context_class;
1030 av_opt_set_defaults(c);
1036 av_cold int sws_init_context(SwsContext *c, SwsFilter *srcFilter,
1037 SwsFilter *dstFilter)
1040 int usesVFilter, usesHFilter;
1042 SwsFilter dummyFilter = { NULL, NULL, NULL, NULL };
1047 int dst_stride = FFALIGN(dstW * sizeof(int16_t) + 66, 16);
1048 int flags, cpu_flags;
1049 enum AVPixelFormat srcFormat = c->srcFormat;
1050 enum AVPixelFormat dstFormat = c->dstFormat;
1051 const AVPixFmtDescriptor *desc_src = av_pix_fmt_desc_get(srcFormat);
1052 const AVPixFmtDescriptor *desc_dst = av_pix_fmt_desc_get(dstFormat);
1054 cpu_flags = av_get_cpu_flags();
1060 unscaled = (srcW == dstW && srcH == dstH);
1062 handle_jpeg(&srcFormat);
1063 handle_jpeg(&dstFormat);
1064 handle_0alpha(&srcFormat);
1065 handle_0alpha(&dstFormat);
1066 handle_xyz(&srcFormat);
1067 handle_xyz(&dstFormat);
1069 if(srcFormat!=c->srcFormat || dstFormat!=c->dstFormat){
1070 av_log(c, AV_LOG_WARNING, "deprecated pixel format used, make sure you did set range correctly\n");
1071 c->srcFormat= srcFormat;
1072 c->dstFormat= dstFormat;
1075 if (!sws_isSupportedInput(srcFormat)) {
1076 av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n",
1077 av_get_pix_fmt_name(srcFormat));
1078 return AVERROR(EINVAL);
1080 if (!sws_isSupportedOutput(dstFormat)) {
1081 av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n",
1082 av_get_pix_fmt_name(dstFormat));
1083 return AVERROR(EINVAL);
1086 i = flags & (SWS_POINT |
1097 if (!i || (i & (i - 1))) {
1098 av_log(c, AV_LOG_ERROR, "Exactly one scaler algorithm must be chosen, got %X\n", i);
1099 return AVERROR(EINVAL);
1102 if (srcW < 1 || srcH < 1 || dstW < 1 || dstH < 1) {
1103 /* FIXME check if these are enough and try to lower them after
1104 * fixing the relevant parts of the code */
1105 av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",
1106 srcW, srcH, dstW, dstH);
1107 return AVERROR(EINVAL);
1111 dstFilter = &dummyFilter;
1113 srcFilter = &dummyFilter;
1115 c->lumXInc = (((int64_t)srcW << 16) + (dstW >> 1)) / dstW;
1116 c->lumYInc = (((int64_t)srcH << 16) + (dstH >> 1)) / dstH;
1117 c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);
1118 c->srcFormatBpp = av_get_bits_per_pixel(desc_src);
1119 c->vRounder = 4 * 0x0001000100010001ULL;
1121 usesVFilter = (srcFilter->lumV && srcFilter->lumV->length > 1) ||
1122 (srcFilter->chrV && srcFilter->chrV->length > 1) ||
1123 (dstFilter->lumV && dstFilter->lumV->length > 1) ||
1124 (dstFilter->chrV && dstFilter->chrV->length > 1);
1125 usesHFilter = (srcFilter->lumH && srcFilter->lumH->length > 1) ||
1126 (srcFilter->chrH && srcFilter->chrH->length > 1) ||
1127 (dstFilter->lumH && dstFilter->lumH->length > 1) ||
1128 (dstFilter->chrH && dstFilter->chrH->length > 1);
1130 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
1131 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
1133 if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) {
1135 av_log(c, AV_LOG_DEBUG, "Forcing full internal H chroma due to odd output size\n");
1136 flags |= SWS_FULL_CHR_H_INT;
1141 if(dstFormat == AV_PIX_FMT_BGR4_BYTE ||
1142 dstFormat == AV_PIX_FMT_RGB4_BYTE ||
1143 dstFormat == AV_PIX_FMT_BGR8 ||
1144 dstFormat == AV_PIX_FMT_RGB8) {
1145 if (flags & SWS_ERROR_DIFFUSION && !(flags & SWS_FULL_CHR_H_INT)) {
1146 av_log(c, AV_LOG_DEBUG,
1147 "Error diffusion dither is only supported in full chroma interpolation for destination format '%s'\n",
1148 av_get_pix_fmt_name(dstFormat));
1149 flags |= SWS_FULL_CHR_H_INT;
1152 if (!(flags & SWS_ERROR_DIFFUSION) && (flags & SWS_FULL_CHR_H_INT)) {
1153 av_log(c, AV_LOG_DEBUG,
1154 "Ordered dither is not supported in full chroma interpolation for destination format '%s'\n",
1155 av_get_pix_fmt_name(dstFormat));
1156 flags |= SWS_ERROR_DIFFUSION;
1160 if (isPlanarRGB(dstFormat)) {
1161 if (!(flags & SWS_FULL_CHR_H_INT)) {
1162 av_log(c, AV_LOG_DEBUG,
1163 "%s output is not supported with half chroma resolution, switching to full\n",
1164 av_get_pix_fmt_name(dstFormat));
1165 flags |= SWS_FULL_CHR_H_INT;
1170 /* reuse chroma for 2 pixels RGB/BGR unless user wants full
1171 * chroma interpolation */
1172 if (flags & SWS_FULL_CHR_H_INT &&
1173 isAnyRGB(dstFormat) &&
1174 !isPlanarRGB(dstFormat) &&
1175 dstFormat != AV_PIX_FMT_RGBA &&
1176 dstFormat != AV_PIX_FMT_ARGB &&
1177 dstFormat != AV_PIX_FMT_BGRA &&
1178 dstFormat != AV_PIX_FMT_ABGR &&
1179 dstFormat != AV_PIX_FMT_RGB24 &&
1180 dstFormat != AV_PIX_FMT_BGR24 &&
1181 dstFormat != AV_PIX_FMT_BGR4_BYTE &&
1182 dstFormat != AV_PIX_FMT_RGB4_BYTE &&
1183 dstFormat != AV_PIX_FMT_BGR8 &&
1184 dstFormat != AV_PIX_FMT_RGB8
1186 av_log(c, AV_LOG_WARNING,
1187 "full chroma interpolation for destination format '%s' not yet implemented\n",
1188 av_get_pix_fmt_name(dstFormat));
1189 flags &= ~SWS_FULL_CHR_H_INT;
1192 if (isAnyRGB(dstFormat) && !(flags & SWS_FULL_CHR_H_INT))
1193 c->chrDstHSubSample = 1;
1195 // drop some chroma lines if the user wants it
1196 c->vChrDrop = (flags & SWS_SRC_V_CHR_DROP_MASK) >>
1197 SWS_SRC_V_CHR_DROP_SHIFT;
1198 c->chrSrcVSubSample += c->vChrDrop;
1200 /* drop every other pixel for chroma calculation unless user
1201 * wants full chroma */
1202 if (isAnyRGB(srcFormat) && !(flags & SWS_FULL_CHR_H_INP) &&
1203 srcFormat != AV_PIX_FMT_RGB8 && srcFormat != AV_PIX_FMT_BGR8 &&
1204 srcFormat != AV_PIX_FMT_RGB4 && srcFormat != AV_PIX_FMT_BGR4 &&
1205 srcFormat != AV_PIX_FMT_RGB4_BYTE && srcFormat != AV_PIX_FMT_BGR4_BYTE &&
1206 srcFormat != AV_PIX_FMT_GBRP9BE && srcFormat != AV_PIX_FMT_GBRP9LE &&
1207 srcFormat != AV_PIX_FMT_GBRP10BE && srcFormat != AV_PIX_FMT_GBRP10LE &&
1208 srcFormat != AV_PIX_FMT_GBRP12BE && srcFormat != AV_PIX_FMT_GBRP12LE &&
1209 srcFormat != AV_PIX_FMT_GBRP14BE && srcFormat != AV_PIX_FMT_GBRP14LE &&
1210 srcFormat != AV_PIX_FMT_GBRP16BE && srcFormat != AV_PIX_FMT_GBRP16LE &&
1211 ((dstW >> c->chrDstHSubSample) <= (srcW >> 1) ||
1212 (flags & SWS_FAST_BILINEAR)))
1213 c->chrSrcHSubSample = 1;
1215 // Note the -((-x)>>y) is so that we always round toward +inf.
1216 c->chrSrcW = -((-srcW) >> c->chrSrcHSubSample);
1217 c->chrSrcH = -((-srcH) >> c->chrSrcVSubSample);
1218 c->chrDstW = -((-dstW) >> c->chrDstHSubSample);
1219 c->chrDstH = -((-dstH) >> c->chrDstVSubSample);
1221 FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW*2+78, 16) * 2, fail);
1223 /* unscaled special cases */
1224 if (unscaled && !usesHFilter && !usesVFilter &&
1225 (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
1226 ff_get_unscaled_swscale(c);
1229 if (flags & SWS_PRINT_INFO)
1230 av_log(c, AV_LOG_INFO,
1231 "using unscaled %s -> %s special converter\n",
1232 av_get_pix_fmt_name(srcFormat), av_get_pix_fmt_name(dstFormat));
1237 c->srcBpc = 1 + desc_src->comp[0].depth_minus1;
1240 c->dstBpc = 1 + desc_dst->comp[0].depth_minus1;
1243 if (isAnyRGB(srcFormat) || srcFormat == AV_PIX_FMT_PAL8)
1245 if (c->dstBpc == 16)
1248 if (INLINE_MMXEXT(cpu_flags) && c->srcBpc == 8 && c->dstBpc <= 14) {
1249 c->canMMXEXTBeUsed = (dstW >= srcW && (dstW & 31) == 0 &&
1250 (srcW & 15) == 0) ? 1 : 0;
1251 if (!c->canMMXEXTBeUsed && dstW >= srcW && (srcW & 15) == 0
1253 && (flags & SWS_FAST_BILINEAR)) {
1254 if (flags & SWS_PRINT_INFO)
1255 av_log(c, AV_LOG_INFO,
1256 "output width is not a multiple of 32 -> no MMXEXT scaler\n");
1258 if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat))
1259 c->canMMXEXTBeUsed = 0;
1261 c->canMMXEXTBeUsed = 0;
1263 c->chrXInc = (((int64_t)c->chrSrcW << 16) + (c->chrDstW >> 1)) / c->chrDstW;
1264 c->chrYInc = (((int64_t)c->chrSrcH << 16) + (c->chrDstH >> 1)) / c->chrDstH;
1266 /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src
1267 * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do
1269 * n-2 is the last chrominance sample available.
1270 * This is not perfect, but no one should notice the difference, the more
1271 * correct variant would be like the vertical one, but that would require
1272 * some special code for the first and last pixel */
1273 if (flags & SWS_FAST_BILINEAR) {
1274 if (c->canMMXEXTBeUsed) {
1278 // we don't use the x86 asm scaler if MMX is available
1279 else if (INLINE_MMX(cpu_flags) && c->dstBpc <= 14) {
1280 c->lumXInc = ((int64_t)(srcW - 2) << 16) / (dstW - 2) - 20;
1281 c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;
1285 #define USE_MMAP (HAVE_MMAP && HAVE_MPROTECT && defined MAP_ANONYMOUS)
1287 /* precalculate horizontal scaler filter coefficients */
1289 #if HAVE_MMXEXT_INLINE
1290 // can't downscale !!!
1291 if (c->canMMXEXTBeUsed && (flags & SWS_FAST_BILINEAR)) {
1292 c->lumMmxextFilterCodeSize = init_hscaler_mmxext(dstW, c->lumXInc, NULL,
1294 c->chrMmxextFilterCodeSize = init_hscaler_mmxext(c->chrDstW, c->chrXInc,
1295 NULL, NULL, NULL, 4);
1298 c->lumMmxextFilterCode = mmap(NULL, c->lumMmxextFilterCodeSize,
1299 PROT_READ | PROT_WRITE,
1300 MAP_PRIVATE | MAP_ANONYMOUS,
1302 c->chrMmxextFilterCode = mmap(NULL, c->chrMmxextFilterCodeSize,
1303 PROT_READ | PROT_WRITE,
1304 MAP_PRIVATE | MAP_ANONYMOUS,
1306 #elif HAVE_VIRTUALALLOC
1307 c->lumMmxextFilterCode = VirtualAlloc(NULL,
1308 c->lumMmxextFilterCodeSize,
1310 PAGE_EXECUTE_READWRITE);
1311 c->chrMmxextFilterCode = VirtualAlloc(NULL,
1312 c->chrMmxextFilterCodeSize,
1314 PAGE_EXECUTE_READWRITE);
1316 c->lumMmxextFilterCode = av_malloc(c->lumMmxextFilterCodeSize);
1317 c->chrMmxextFilterCode = av_malloc(c->chrMmxextFilterCodeSize);
1320 #ifdef MAP_ANONYMOUS
1321 if (c->lumMmxextFilterCode == MAP_FAILED || c->chrMmxextFilterCode == MAP_FAILED)
1323 if (!c->lumMmxextFilterCode || !c->chrMmxextFilterCode)
1326 av_log(c, AV_LOG_ERROR, "Failed to allocate MMX2FilterCode\n");
1327 return AVERROR(ENOMEM);
1330 FF_ALLOCZ_OR_GOTO(c, c->hLumFilter, (dstW / 8 + 8) * sizeof(int16_t), fail);
1331 FF_ALLOCZ_OR_GOTO(c, c->hChrFilter, (c->chrDstW / 4 + 8) * sizeof(int16_t), fail);
1332 FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW / 2 / 8 + 8) * sizeof(int32_t), fail);
1333 FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);
1335 init_hscaler_mmxext( dstW, c->lumXInc, c->lumMmxextFilterCode,
1336 c->hLumFilter, (uint32_t*)c->hLumFilterPos, 8);
1337 init_hscaler_mmxext(c->chrDstW, c->chrXInc, c->chrMmxextFilterCode,
1338 c->hChrFilter, (uint32_t*)c->hChrFilterPos, 4);
1341 mprotect(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize, PROT_EXEC | PROT_READ);
1342 mprotect(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize, PROT_EXEC | PROT_READ);
1345 #endif /* HAVE_MMXEXT_INLINE */
1347 const int filterAlign =
1348 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :
1349 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
1352 if (initFilter(&c->hLumFilter, &c->hLumFilterPos,
1353 &c->hLumFilterSize, c->lumXInc,
1354 srcW, dstW, filterAlign, 1 << 14,
1355 (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
1356 cpu_flags, srcFilter->lumH, dstFilter->lumH,
1359 if (initFilter(&c->hChrFilter, &c->hChrFilterPos,
1360 &c->hChrFilterSize, c->chrXInc,
1361 c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,
1362 (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
1363 cpu_flags, srcFilter->chrH, dstFilter->chrH,
1367 } // initialize horizontal stuff
1369 /* precalculate vertical scaler filter coefficients */
1371 const int filterAlign =
1372 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 2 :
1373 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
1376 if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,
1377 c->lumYInc, srcH, dstH, filterAlign, (1 << 12),
1378 (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
1379 cpu_flags, srcFilter->lumV, dstFilter->lumV,
1382 if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,
1383 c->chrYInc, c->chrSrcH, c->chrDstH,
1384 filterAlign, (1 << 12),
1385 (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
1386 cpu_flags, srcFilter->chrV, dstFilter->chrV,
1391 FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH, fail);
1392 FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);
1394 for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {
1396 short *p = (short *)&c->vYCoeffsBank[i];
1397 for (j = 0; j < 8; j++)
1398 p[j] = c->vLumFilter[i];
1401 for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {
1403 short *p = (short *)&c->vCCoeffsBank[i];
1404 for (j = 0; j < 8; j++)
1405 p[j] = c->vChrFilter[i];
1410 // calculate buffer sizes so that they won't run out while handling these damn slices
1411 c->vLumBufSize = c->vLumFilterSize;
1412 c->vChrBufSize = c->vChrFilterSize;
1413 for (i = 0; i < dstH; i++) {
1414 int chrI = (int64_t)i * c->chrDstH / dstH;
1415 int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,
1416 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)
1417 << c->chrSrcVSubSample));
1419 nextSlice >>= c->chrSrcVSubSample;
1420 nextSlice <<= c->chrSrcVSubSample;
1421 if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)
1422 c->vLumBufSize = nextSlice - c->vLumFilterPos[i];
1423 if (c->vChrFilterPos[chrI] + c->vChrBufSize <
1424 (nextSlice >> c->chrSrcVSubSample))
1425 c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -
1426 c->vChrFilterPos[chrI];
1429 for (i = 0; i < 4; i++)
1430 FF_ALLOCZ_OR_GOTO(c, c->dither_error[i], (c->dstW+2) * sizeof(int), fail);
1432 /* Allocate pixbufs (we use dynamic allocation because otherwise we would
1433 * need to allocate several megabytes to handle all possible cases) */
1434 FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);
1435 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
1436 FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
1437 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
1438 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);
1439 /* Note we need at least one pixel more at the end because of the MMX code
1440 * (just in case someone wants to replace the 4000/8000). */
1441 /* align at 16 bytes for AltiVec */
1442 for (i = 0; i < c->vLumBufSize; i++) {
1443 FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],
1444 dst_stride + 16, fail);
1445 c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];
1447 // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)
1448 c->uv_off = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);
1449 c->uv_offx2 = dst_stride + 16;
1450 for (i = 0; i < c->vChrBufSize; i++) {
1451 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],
1452 dst_stride * 2 + 32, fail);
1453 c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];
1454 c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]
1455 = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;
1457 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
1458 for (i = 0; i < c->vLumBufSize; i++) {
1459 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i + c->vLumBufSize],
1460 dst_stride + 16, fail);
1461 c->alpPixBuf[i] = c->alpPixBuf[i + c->vLumBufSize];
1464 // try to avoid drawing green stuff between the right end and the stride end
1465 for (i = 0; i < c->vChrBufSize; i++)
1466 if(desc_dst->comp[0].depth_minus1 == 15){
1467 av_assert0(c->dstBpc > 14);
1468 for(j=0; j<dst_stride/2+1; j++)
1469 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;
1471 for(j=0; j<dst_stride+1; j++)
1472 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;
1474 av_assert0(c->chrDstH <= dstH);
1476 if (flags & SWS_PRINT_INFO) {
1477 if (flags & SWS_FAST_BILINEAR)
1478 av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
1479 else if (flags & SWS_BILINEAR)
1480 av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
1481 else if (flags & SWS_BICUBIC)
1482 av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
1483 else if (flags & SWS_X)
1484 av_log(c, AV_LOG_INFO, "Experimental scaler, ");
1485 else if (flags & SWS_POINT)
1486 av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
1487 else if (flags & SWS_AREA)
1488 av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
1489 else if (flags & SWS_BICUBLIN)
1490 av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
1491 else if (flags & SWS_GAUSS)
1492 av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
1493 else if (flags & SWS_SINC)
1494 av_log(c, AV_LOG_INFO, "Sinc scaler, ");
1495 else if (flags & SWS_LANCZOS)
1496 av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
1497 else if (flags & SWS_SPLINE)
1498 av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
1500 av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
1502 av_log(c, AV_LOG_INFO, "from %s to %s%s ",
1503 av_get_pix_fmt_name(srcFormat),
1505 dstFormat == AV_PIX_FMT_BGR555 || dstFormat == AV_PIX_FMT_BGR565 ||
1506 dstFormat == AV_PIX_FMT_RGB444BE || dstFormat == AV_PIX_FMT_RGB444LE ||
1507 dstFormat == AV_PIX_FMT_BGR444BE || dstFormat == AV_PIX_FMT_BGR444LE ?
1512 av_get_pix_fmt_name(dstFormat));
1514 if (INLINE_MMXEXT(cpu_flags))
1515 av_log(c, AV_LOG_INFO, "using MMXEXT\n");
1516 else if (INLINE_AMD3DNOW(cpu_flags))
1517 av_log(c, AV_LOG_INFO, "using 3DNOW\n");
1518 else if (INLINE_MMX(cpu_flags))
1519 av_log(c, AV_LOG_INFO, "using MMX\n");
1520 else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC)
1521 av_log(c, AV_LOG_INFO, "using AltiVec\n");
1523 av_log(c, AV_LOG_INFO, "using C\n");
1525 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
1526 av_log(c, AV_LOG_DEBUG,
1527 "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1528 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
1529 av_log(c, AV_LOG_DEBUG,
1530 "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1531 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,
1532 c->chrXInc, c->chrYInc);
1535 c->swScale = ff_getSwsFunc(c);
1537 fail: // FIXME replace things by appropriate error codes
1541 #if FF_API_SWS_GETCONTEXT
1542 SwsContext *sws_getContext(int srcW, int srcH, enum AVPixelFormat srcFormat,
1543 int dstW, int dstH, enum AVPixelFormat dstFormat,
1544 int flags, SwsFilter *srcFilter,
1545 SwsFilter *dstFilter, const double *param)
1549 if (!(c = sws_alloc_context()))
1557 c->srcRange = handle_jpeg(&srcFormat);
1558 c->dstRange = handle_jpeg(&dstFormat);
1559 c->src0Alpha = handle_0alpha(&srcFormat);
1560 c->dst0Alpha = handle_0alpha(&dstFormat);
1561 c->srcXYZ = handle_xyz(&srcFormat);
1562 c->dstXYZ = handle_xyz(&dstFormat);
1563 c->srcFormat = srcFormat;
1564 c->dstFormat = dstFormat;
1567 c->param[0] = param[0];
1568 c->param[1] = param[1];
1570 sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,
1571 ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,
1572 c->dstRange, 0, 1 << 16, 1 << 16);
1574 if (sws_init_context(c, srcFilter, dstFilter) < 0) {
1583 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
1584 float lumaSharpen, float chromaSharpen,
1585 float chromaHShift, float chromaVShift,
1588 SwsFilter *filter = av_malloc(sizeof(SwsFilter));
1592 if (lumaGBlur != 0.0) {
1593 filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);
1594 filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);
1596 filter->lumH = sws_getIdentityVec();
1597 filter->lumV = sws_getIdentityVec();
1600 if (chromaGBlur != 0.0) {
1601 filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);
1602 filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);
1604 filter->chrH = sws_getIdentityVec();
1605 filter->chrV = sws_getIdentityVec();
1608 if (chromaSharpen != 0.0) {
1609 SwsVector *id = sws_getIdentityVec();
1610 sws_scaleVec(filter->chrH, -chromaSharpen);
1611 sws_scaleVec(filter->chrV, -chromaSharpen);
1612 sws_addVec(filter->chrH, id);
1613 sws_addVec(filter->chrV, id);
1617 if (lumaSharpen != 0.0) {
1618 SwsVector *id = sws_getIdentityVec();
1619 sws_scaleVec(filter->lumH, -lumaSharpen);
1620 sws_scaleVec(filter->lumV, -lumaSharpen);
1621 sws_addVec(filter->lumH, id);
1622 sws_addVec(filter->lumV, id);
1626 if (chromaHShift != 0.0)
1627 sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));
1629 if (chromaVShift != 0.0)
1630 sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));
1632 sws_normalizeVec(filter->chrH, 1.0);
1633 sws_normalizeVec(filter->chrV, 1.0);
1634 sws_normalizeVec(filter->lumH, 1.0);
1635 sws_normalizeVec(filter->lumV, 1.0);
1638 sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
1640 sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
1645 SwsVector *sws_allocVec(int length)
1649 if(length <= 0 || length > INT_MAX/ sizeof(double))
1652 vec = av_malloc(sizeof(SwsVector));
1655 vec->length = length;
1656 vec->coeff = av_malloc(sizeof(double) * length);
1662 SwsVector *sws_getGaussianVec(double variance, double quality)
1664 const int length = (int)(variance * quality + 0.5) | 1;
1666 double middle = (length - 1) * 0.5;
1669 if(variance < 0 || quality < 0)
1672 vec = sws_allocVec(length);
1677 for (i = 0; i < length; i++) {
1678 double dist = i - middle;
1679 vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /
1680 sqrt(2 * variance * M_PI);
1683 sws_normalizeVec(vec, 1.0);
1688 SwsVector *sws_getConstVec(double c, int length)
1691 SwsVector *vec = sws_allocVec(length);
1696 for (i = 0; i < length; i++)
1702 SwsVector *sws_getIdentityVec(void)
1704 return sws_getConstVec(1.0, 1);
1707 static double sws_dcVec(SwsVector *a)
1712 for (i = 0; i < a->length; i++)
1718 void sws_scaleVec(SwsVector *a, double scalar)
1722 for (i = 0; i < a->length; i++)
1723 a->coeff[i] *= scalar;
1726 void sws_normalizeVec(SwsVector *a, double height)
1728 sws_scaleVec(a, height / sws_dcVec(a));
1731 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
1733 int length = a->length + b->length - 1;
1735 SwsVector *vec = sws_getConstVec(0.0, length);
1740 for (i = 0; i < a->length; i++) {
1741 for (j = 0; j < b->length; j++) {
1742 vec->coeff[i + j] += a->coeff[i] * b->coeff[j];
1749 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
1751 int length = FFMAX(a->length, b->length);
1753 SwsVector *vec = sws_getConstVec(0.0, length);
1758 for (i = 0; i < a->length; i++)
1759 vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
1760 for (i = 0; i < b->length; i++)
1761 vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] += b->coeff[i];
1766 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
1768 int length = FFMAX(a->length, b->length);
1770 SwsVector *vec = sws_getConstVec(0.0, length);
1775 for (i = 0; i < a->length; i++)
1776 vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
1777 for (i = 0; i < b->length; i++)
1778 vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] -= b->coeff[i];
1783 /* shift left / or right if "shift" is negative */
1784 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
1786 int length = a->length + FFABS(shift) * 2;
1788 SwsVector *vec = sws_getConstVec(0.0, length);
1793 for (i = 0; i < a->length; i++) {
1794 vec->coeff[i + (length - 1) / 2 -
1795 (a->length - 1) / 2 - shift] = a->coeff[i];
1801 void sws_shiftVec(SwsVector *a, int shift)
1803 SwsVector *shifted = sws_getShiftedVec(a, shift);
1805 a->coeff = shifted->coeff;
1806 a->length = shifted->length;
1810 void sws_addVec(SwsVector *a, SwsVector *b)
1812 SwsVector *sum = sws_sumVec(a, b);
1814 a->coeff = sum->coeff;
1815 a->length = sum->length;
1819 void sws_subVec(SwsVector *a, SwsVector *b)
1821 SwsVector *diff = sws_diffVec(a, b);
1823 a->coeff = diff->coeff;
1824 a->length = diff->length;
1828 void sws_convVec(SwsVector *a, SwsVector *b)
1830 SwsVector *conv = sws_getConvVec(a, b);
1832 a->coeff = conv->coeff;
1833 a->length = conv->length;
1837 SwsVector *sws_cloneVec(SwsVector *a)
1840 SwsVector *vec = sws_allocVec(a->length);
1845 for (i = 0; i < a->length; i++)
1846 vec->coeff[i] = a->coeff[i];
1851 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
1858 for (i = 0; i < a->length; i++)
1859 if (a->coeff[i] > max)
1862 for (i = 0; i < a->length; i++)
1863 if (a->coeff[i] < min)
1868 for (i = 0; i < a->length; i++) {
1869 int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);
1870 av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
1872 av_log(log_ctx, log_level, " ");
1873 av_log(log_ctx, log_level, "|\n");
1877 void sws_freeVec(SwsVector *a)
1881 av_freep(&a->coeff);
1886 void sws_freeFilter(SwsFilter *filter)
1892 sws_freeVec(filter->lumH);
1894 sws_freeVec(filter->lumV);
1896 sws_freeVec(filter->chrH);
1898 sws_freeVec(filter->chrV);
1902 void sws_freeContext(SwsContext *c)
1909 for (i = 0; i < c->vLumBufSize; i++)
1910 av_freep(&c->lumPixBuf[i]);
1911 av_freep(&c->lumPixBuf);
1914 if (c->chrUPixBuf) {
1915 for (i = 0; i < c->vChrBufSize; i++)
1916 av_freep(&c->chrUPixBuf[i]);
1917 av_freep(&c->chrUPixBuf);
1918 av_freep(&c->chrVPixBuf);
1921 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1922 for (i = 0; i < c->vLumBufSize; i++)
1923 av_freep(&c->alpPixBuf[i]);
1924 av_freep(&c->alpPixBuf);
1927 for (i = 0; i < 4; i++)
1928 av_freep(&c->dither_error[i]);
1930 av_freep(&c->vLumFilter);
1931 av_freep(&c->vChrFilter);
1932 av_freep(&c->hLumFilter);
1933 av_freep(&c->hChrFilter);
1935 av_freep(&c->vYCoeffsBank);
1936 av_freep(&c->vCCoeffsBank);
1939 av_freep(&c->vLumFilterPos);
1940 av_freep(&c->vChrFilterPos);
1941 av_freep(&c->hLumFilterPos);
1942 av_freep(&c->hChrFilterPos);
1946 if (c->lumMmxextFilterCode)
1947 munmap(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize);
1948 if (c->chrMmxextFilterCode)
1949 munmap(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize);
1950 #elif HAVE_VIRTUALALLOC
1951 if (c->lumMmxextFilterCode)
1952 VirtualFree(c->lumMmxextFilterCode, 0, MEM_RELEASE);
1953 if (c->chrMmxextFilterCode)
1954 VirtualFree(c->chrMmxextFilterCode, 0, MEM_RELEASE);
1956 av_free(c->lumMmxextFilterCode);
1957 av_free(c->chrMmxextFilterCode);
1959 c->lumMmxextFilterCode = NULL;
1960 c->chrMmxextFilterCode = NULL;
1961 #endif /* HAVE_MMX_INLINE */
1963 av_freep(&c->yuvTable);
1964 av_freep(&c->formatConvBuffer);
1969 struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,
1970 int srcH, enum AVPixelFormat srcFormat,
1972 enum AVPixelFormat dstFormat, int flags,
1973 SwsFilter *srcFilter,
1974 SwsFilter *dstFilter,
1975 const double *param)
1977 static const double default_param[2] = { SWS_PARAM_DEFAULT,
1978 SWS_PARAM_DEFAULT };
1981 param = default_param;
1984 (context->srcW != srcW ||
1985 context->srcH != srcH ||
1986 context->srcFormat != srcFormat ||
1987 context->dstW != dstW ||
1988 context->dstH != dstH ||
1989 context->dstFormat != dstFormat ||
1990 context->flags != flags ||
1991 context->param[0] != param[0] ||
1992 context->param[1] != param[1])) {
1993 sws_freeContext(context);
1998 if (!(context = sws_alloc_context()))
2000 context->srcW = srcW;
2001 context->srcH = srcH;
2002 context->srcRange = handle_jpeg(&srcFormat);
2003 context->src0Alpha = handle_0alpha(&srcFormat);
2004 context->srcXYZ = handle_xyz(&srcFormat);
2005 context->srcFormat = srcFormat;
2006 context->dstW = dstW;
2007 context->dstH = dstH;
2008 context->dstRange = handle_jpeg(&dstFormat);
2009 context->dst0Alpha = handle_0alpha(&dstFormat);
2010 context->dstXYZ = handle_xyz(&dstFormat);
2011 context->dstFormat = dstFormat;
2012 context->flags = flags;
2013 context->param[0] = param[0];
2014 context->param[1] = param[1];
2015 sws_setColorspaceDetails(context, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],
2017 ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,
2018 context->dstRange, 0, 1 << 16, 1 << 16);
2019 if (sws_init_context(context, srcFilter, dstFilter) < 0) {
2020 sws_freeContext(context);