2 * Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
4 * This file is part of FFmpeg.
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 #define _SVID_SOURCE // needed for MAP_ANONYMOUS
24 #define _DARWIN_C_SOURCE // needed for MAP_ANON
31 #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
32 #define MAP_ANONYMOUS MAP_ANON
36 #define WIN32_LEAN_AND_MEAN
40 #include "libavutil/attributes.h"
41 #include "libavutil/avassert.h"
42 #include "libavutil/avutil.h"
43 #include "libavutil/bswap.h"
44 #include "libavutil/cpu.h"
45 #include "libavutil/intreadwrite.h"
46 #include "libavutil/mathematics.h"
47 #include "libavutil/opt.h"
48 #include "libavutil/pixdesc.h"
49 #include "libavutil/x86/asm.h"
50 #include "libavutil/x86/cpu.h"
53 #include "swscale_internal.h"
55 static void handle_formats(SwsContext *c);
57 unsigned swscale_version(void)
59 av_assert0(LIBSWSCALE_VERSION_MICRO >= 100);
60 return LIBSWSCALE_VERSION_INT;
63 const char *swscale_configuration(void)
65 return FFMPEG_CONFIGURATION;
68 const char *swscale_license(void)
70 #define LICENSE_PREFIX "libswscale license: "
71 return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
74 #define RET 0xC3 // near return opcode for x86
76 typedef struct FormatEntry {
77 uint8_t is_supported_in :1;
78 uint8_t is_supported_out :1;
79 uint8_t is_supported_endianness :1;
82 static const FormatEntry format_entries[AV_PIX_FMT_NB] = {
83 [AV_PIX_FMT_YUV420P] = { 1, 1 },
84 [AV_PIX_FMT_YUYV422] = { 1, 1 },
85 [AV_PIX_FMT_RGB24] = { 1, 1 },
86 [AV_PIX_FMT_BGR24] = { 1, 1 },
87 [AV_PIX_FMT_YUV422P] = { 1, 1 },
88 [AV_PIX_FMT_YUV444P] = { 1, 1 },
89 [AV_PIX_FMT_YUV410P] = { 1, 1 },
90 [AV_PIX_FMT_YUV411P] = { 1, 1 },
91 [AV_PIX_FMT_GRAY8] = { 1, 1 },
92 [AV_PIX_FMT_MONOWHITE] = { 1, 1 },
93 [AV_PIX_FMT_MONOBLACK] = { 1, 1 },
94 [AV_PIX_FMT_PAL8] = { 1, 0 },
95 [AV_PIX_FMT_YUVJ420P] = { 1, 1 },
96 [AV_PIX_FMT_YUVJ411P] = { 1, 1 },
97 [AV_PIX_FMT_YUVJ422P] = { 1, 1 },
98 [AV_PIX_FMT_YUVJ444P] = { 1, 1 },
99 [AV_PIX_FMT_UYVY422] = { 1, 1 },
100 [AV_PIX_FMT_UYYVYY411] = { 0, 0 },
101 [AV_PIX_FMT_BGR8] = { 1, 1 },
102 [AV_PIX_FMT_BGR4] = { 0, 1 },
103 [AV_PIX_FMT_BGR4_BYTE] = { 1, 1 },
104 [AV_PIX_FMT_RGB8] = { 1, 1 },
105 [AV_PIX_FMT_RGB4] = { 0, 1 },
106 [AV_PIX_FMT_RGB4_BYTE] = { 1, 1 },
107 [AV_PIX_FMT_NV12] = { 1, 1 },
108 [AV_PIX_FMT_NV21] = { 1, 1 },
109 [AV_PIX_FMT_ARGB] = { 1, 1 },
110 [AV_PIX_FMT_RGBA] = { 1, 1 },
111 [AV_PIX_FMT_ABGR] = { 1, 1 },
112 [AV_PIX_FMT_BGRA] = { 1, 1 },
113 [AV_PIX_FMT_0RGB] = { 1, 1 },
114 [AV_PIX_FMT_RGB0] = { 1, 1 },
115 [AV_PIX_FMT_0BGR] = { 1, 1 },
116 [AV_PIX_FMT_BGR0] = { 1, 1 },
117 [AV_PIX_FMT_GRAY16BE] = { 1, 1 },
118 [AV_PIX_FMT_GRAY16LE] = { 1, 1 },
119 [AV_PIX_FMT_YUV440P] = { 1, 1 },
120 [AV_PIX_FMT_YUVJ440P] = { 1, 1 },
121 [AV_PIX_FMT_YUVA420P] = { 1, 1 },
122 [AV_PIX_FMT_YUVA422P] = { 1, 1 },
123 [AV_PIX_FMT_YUVA444P] = { 1, 1 },
124 [AV_PIX_FMT_YUVA420P9BE] = { 1, 1 },
125 [AV_PIX_FMT_YUVA420P9LE] = { 1, 1 },
126 [AV_PIX_FMT_YUVA422P9BE] = { 1, 1 },
127 [AV_PIX_FMT_YUVA422P9LE] = { 1, 1 },
128 [AV_PIX_FMT_YUVA444P9BE] = { 1, 1 },
129 [AV_PIX_FMT_YUVA444P9LE] = { 1, 1 },
130 [AV_PIX_FMT_YUVA420P10BE]= { 1, 1 },
131 [AV_PIX_FMT_YUVA420P10LE]= { 1, 1 },
132 [AV_PIX_FMT_YUVA422P10BE]= { 1, 1 },
133 [AV_PIX_FMT_YUVA422P10LE]= { 1, 1 },
134 [AV_PIX_FMT_YUVA444P10BE]= { 1, 1 },
135 [AV_PIX_FMT_YUVA444P10LE]= { 1, 1 },
136 [AV_PIX_FMT_YUVA420P16BE]= { 1, 1 },
137 [AV_PIX_FMT_YUVA420P16LE]= { 1, 1 },
138 [AV_PIX_FMT_YUVA422P16BE]= { 1, 1 },
139 [AV_PIX_FMT_YUVA422P16LE]= { 1, 1 },
140 [AV_PIX_FMT_YUVA444P16BE]= { 1, 1 },
141 [AV_PIX_FMT_YUVA444P16LE]= { 1, 1 },
142 [AV_PIX_FMT_RGB48BE] = { 1, 1 },
143 [AV_PIX_FMT_RGB48LE] = { 1, 1 },
144 [AV_PIX_FMT_RGBA64BE] = { 1, 1 },
145 [AV_PIX_FMT_RGBA64LE] = { 1, 1 },
146 [AV_PIX_FMT_RGB565BE] = { 1, 1 },
147 [AV_PIX_FMT_RGB565LE] = { 1, 1 },
148 [AV_PIX_FMT_RGB555BE] = { 1, 1 },
149 [AV_PIX_FMT_RGB555LE] = { 1, 1 },
150 [AV_PIX_FMT_BGR565BE] = { 1, 1 },
151 [AV_PIX_FMT_BGR565LE] = { 1, 1 },
152 [AV_PIX_FMT_BGR555BE] = { 1, 1 },
153 [AV_PIX_FMT_BGR555LE] = { 1, 1 },
154 [AV_PIX_FMT_YUV420P16LE] = { 1, 1 },
155 [AV_PIX_FMT_YUV420P16BE] = { 1, 1 },
156 [AV_PIX_FMT_YUV422P16LE] = { 1, 1 },
157 [AV_PIX_FMT_YUV422P16BE] = { 1, 1 },
158 [AV_PIX_FMT_YUV444P16LE] = { 1, 1 },
159 [AV_PIX_FMT_YUV444P16BE] = { 1, 1 },
160 [AV_PIX_FMT_RGB444LE] = { 1, 1 },
161 [AV_PIX_FMT_RGB444BE] = { 1, 1 },
162 [AV_PIX_FMT_BGR444LE] = { 1, 1 },
163 [AV_PIX_FMT_BGR444BE] = { 1, 1 },
164 [AV_PIX_FMT_Y400A] = { 1, 0 },
165 [AV_PIX_FMT_BGR48BE] = { 1, 1 },
166 [AV_PIX_FMT_BGR48LE] = { 1, 1 },
167 [AV_PIX_FMT_BGRA64BE] = { 0, 0 },
168 [AV_PIX_FMT_BGRA64LE] = { 0, 0 },
169 [AV_PIX_FMT_YUV420P9BE] = { 1, 1 },
170 [AV_PIX_FMT_YUV420P9LE] = { 1, 1 },
171 [AV_PIX_FMT_YUV420P10BE] = { 1, 1 },
172 [AV_PIX_FMT_YUV420P10LE] = { 1, 1 },
173 [AV_PIX_FMT_YUV420P12BE] = { 1, 1 },
174 [AV_PIX_FMT_YUV420P12LE] = { 1, 1 },
175 [AV_PIX_FMT_YUV420P14BE] = { 1, 1 },
176 [AV_PIX_FMT_YUV420P14LE] = { 1, 1 },
177 [AV_PIX_FMT_YUV422P9BE] = { 1, 1 },
178 [AV_PIX_FMT_YUV422P9LE] = { 1, 1 },
179 [AV_PIX_FMT_YUV422P10BE] = { 1, 1 },
180 [AV_PIX_FMT_YUV422P10LE] = { 1, 1 },
181 [AV_PIX_FMT_YUV422P12BE] = { 1, 1 },
182 [AV_PIX_FMT_YUV422P12LE] = { 1, 1 },
183 [AV_PIX_FMT_YUV422P14BE] = { 1, 1 },
184 [AV_PIX_FMT_YUV422P14LE] = { 1, 1 },
185 [AV_PIX_FMT_YUV444P9BE] = { 1, 1 },
186 [AV_PIX_FMT_YUV444P9LE] = { 1, 1 },
187 [AV_PIX_FMT_YUV444P10BE] = { 1, 1 },
188 [AV_PIX_FMT_YUV444P10LE] = { 1, 1 },
189 [AV_PIX_FMT_YUV444P12BE] = { 1, 1 },
190 [AV_PIX_FMT_YUV444P12LE] = { 1, 1 },
191 [AV_PIX_FMT_YUV444P14BE] = { 1, 1 },
192 [AV_PIX_FMT_YUV444P14LE] = { 1, 1 },
193 [AV_PIX_FMT_GBRP] = { 1, 1 },
194 [AV_PIX_FMT_GBRP9LE] = { 1, 1 },
195 [AV_PIX_FMT_GBRP9BE] = { 1, 1 },
196 [AV_PIX_FMT_GBRP10LE] = { 1, 1 },
197 [AV_PIX_FMT_GBRP10BE] = { 1, 1 },
198 [AV_PIX_FMT_GBRP12LE] = { 1, 1 },
199 [AV_PIX_FMT_GBRP12BE] = { 1, 1 },
200 [AV_PIX_FMT_GBRP14LE] = { 1, 1 },
201 [AV_PIX_FMT_GBRP14BE] = { 1, 1 },
202 [AV_PIX_FMT_GBRP16LE] = { 1, 0 },
203 [AV_PIX_FMT_GBRP16BE] = { 1, 0 },
204 [AV_PIX_FMT_XYZ12BE] = { 1, 0, 1 },
205 [AV_PIX_FMT_XYZ12LE] = { 1, 0, 1 },
206 [AV_PIX_FMT_GBRAP] = { 1, 1 },
207 [AV_PIX_FMT_GBRAP16LE] = { 1, 0 },
208 [AV_PIX_FMT_GBRAP16BE] = { 1, 0 },
211 int sws_isSupportedInput(enum AVPixelFormat pix_fmt)
213 return (unsigned)pix_fmt < AV_PIX_FMT_NB ?
214 format_entries[pix_fmt].is_supported_in : 0;
217 int sws_isSupportedOutput(enum AVPixelFormat pix_fmt)
219 return (unsigned)pix_fmt < AV_PIX_FMT_NB ?
220 format_entries[pix_fmt].is_supported_out : 0;
223 int sws_isSupportedEndiannessConversion(enum AVPixelFormat pix_fmt)
225 return (unsigned)pix_fmt < AV_PIX_FMT_NB ?
226 format_entries[pix_fmt].is_supported_endianness : 0;
229 extern const int32_t ff_yuv2rgb_coeffs[8][4];
231 #if FF_API_SWS_FORMAT_NAME
232 const char *sws_format_name(enum AVPixelFormat format)
234 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(format);
238 return "Unknown format";
242 static double getSplineCoeff(double a, double b, double c, double d,
246 return ((d * dist + c) * dist + b) * dist + a;
248 return getSplineCoeff(0.0,
249 b + 2.0 * c + 3.0 * d,
251 -b - 3.0 * c - 6.0 * d,
255 static av_cold int initFilter(int16_t **outFilter, int32_t **filterPos,
256 int *outFilterSize, int xInc, int srcW,
257 int dstW, int filterAlign, int one,
258 int flags, int cpu_flags,
259 SwsVector *srcFilter, SwsVector *dstFilter,
266 int64_t *filter = NULL;
267 int64_t *filter2 = NULL;
268 const int64_t fone = 1LL << (54 - FFMIN(av_log2(srcW/dstW), 8));
271 emms_c(); // FIXME should not be required but IS (even for non-MMX versions)
273 // NOTE: the +3 is for the MMX(+1) / SSE(+3) scaler which reads over the end
274 FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW + 3) * sizeof(**filterPos), fail);
276 if (FFABS(xInc - 0x10000) < 10) { // unscaled
279 FF_ALLOCZ_OR_GOTO(NULL, filter,
280 dstW * sizeof(*filter) * filterSize, fail);
282 for (i = 0; i < dstW; i++) {
283 filter[i * filterSize] = fone;
286 } else if (flags & SWS_POINT) { // lame looking point sampling mode
290 FF_ALLOC_OR_GOTO(NULL, filter,
291 dstW * sizeof(*filter) * filterSize, fail);
293 xDstInSrc = xInc / 2 - 0x8000;
294 for (i = 0; i < dstW; i++) {
295 int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;
297 (*filterPos)[i] = xx;
301 } else if ((xInc <= (1 << 16) && (flags & SWS_AREA)) ||
302 (flags & SWS_FAST_BILINEAR)) { // bilinear upscale
306 FF_ALLOC_OR_GOTO(NULL, filter,
307 dstW * sizeof(*filter) * filterSize, fail);
309 xDstInSrc = xInc / 2 - 0x8000;
310 for (i = 0; i < dstW; i++) {
311 int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;
314 (*filterPos)[i] = xx;
315 // bilinear upscale / linear interpolate / area averaging
316 for (j = 0; j < filterSize; j++) {
317 int64_t coeff= fone - FFABS(((int64_t)xx<<16) - xDstInSrc)*(fone>>16);
320 filter[i * filterSize + j] = coeff;
329 if (flags & SWS_BICUBIC)
331 else if (flags & SWS_X)
333 else if (flags & SWS_AREA)
334 sizeFactor = 1; // downscale only, for upscale it is bilinear
335 else if (flags & SWS_GAUSS)
336 sizeFactor = 8; // infinite ;)
337 else if (flags & SWS_LANCZOS)
338 sizeFactor = param[0] != SWS_PARAM_DEFAULT ? ceil(2 * param[0]) : 6;
339 else if (flags & SWS_SINC)
340 sizeFactor = 20; // infinite ;)
341 else if (flags & SWS_SPLINE)
342 sizeFactor = 20; // infinite ;)
343 else if (flags & SWS_BILINEAR)
350 filterSize = 1 + sizeFactor; // upscale
352 filterSize = 1 + (sizeFactor * srcW + dstW - 1) / dstW;
354 filterSize = FFMIN(filterSize, srcW - 2);
355 filterSize = FFMAX(filterSize, 1);
357 FF_ALLOC_OR_GOTO(NULL, filter,
358 dstW * sizeof(*filter) * filterSize, fail);
360 xDstInSrc = xInc - 0x10000;
361 for (i = 0; i < dstW; i++) {
362 int xx = (xDstInSrc - ((filterSize - 2) << 16)) / (1 << 17);
364 (*filterPos)[i] = xx;
365 for (j = 0; j < filterSize; j++) {
366 int64_t d = (FFABS(((int64_t)xx << 17) - xDstInSrc)) << 13;
372 floatd = d * (1.0 / (1 << 30));
374 if (flags & SWS_BICUBIC) {
375 int64_t B = (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1 << 24);
376 int64_t C = (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1 << 24);
378 if (d >= 1LL << 31) {
381 int64_t dd = (d * d) >> 30;
382 int64_t ddd = (dd * d) >> 30;
385 coeff = (12 * (1 << 24) - 9 * B - 6 * C) * ddd +
386 (-18 * (1 << 24) + 12 * B + 6 * C) * dd +
387 (6 * (1 << 24) - 2 * B) * (1 << 30);
389 coeff = (-B - 6 * C) * ddd +
390 (6 * B + 30 * C) * dd +
391 (-12 * B - 48 * C) * d +
392 (8 * B + 24 * C) * (1 << 30);
394 coeff /= (1LL<<54)/fone;
397 else if (flags & SWS_X) {
398 double p = param ? param * 0.01 : 0.3;
399 coeff = d ? sin(d * M_PI) / (d * M_PI) : 1.0;
400 coeff *= pow(2.0, -p * d * d);
403 else if (flags & SWS_X) {
404 double A = param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
408 c = cos(floatd * M_PI);
415 coeff = (c * 0.5 + 0.5) * fone;
416 } else if (flags & SWS_AREA) {
417 int64_t d2 = d - (1 << 29);
418 if (d2 * xInc < -(1LL << (29 + 16)))
419 coeff = 1.0 * (1LL << (30 + 16));
420 else if (d2 * xInc < (1LL << (29 + 16)))
421 coeff = -d2 * xInc + (1LL << (29 + 16));
424 coeff *= fone >> (30 + 16);
425 } else if (flags & SWS_GAUSS) {
426 double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
427 coeff = (pow(2.0, -p * floatd * floatd)) * fone;
428 } else if (flags & SWS_SINC) {
429 coeff = (d ? sin(floatd * M_PI) / (floatd * M_PI) : 1.0) * fone;
430 } else if (flags & SWS_LANCZOS) {
431 double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
432 coeff = (d ? sin(floatd * M_PI) * sin(floatd * M_PI / p) /
433 (floatd * floatd * M_PI * M_PI / p) : 1.0) * fone;
436 } else if (flags & SWS_BILINEAR) {
437 coeff = (1 << 30) - d;
441 } else if (flags & SWS_SPLINE) {
442 double p = -2.196152422706632;
443 coeff = getSplineCoeff(1.0, 0.0, p, -p - 1.0, floatd) * fone;
448 filter[i * filterSize + j] = coeff;
451 xDstInSrc += 2 * xInc;
455 /* apply src & dst Filter to filter -> filter2
458 av_assert0(filterSize > 0);
459 filter2Size = filterSize;
461 filter2Size += srcFilter->length - 1;
463 filter2Size += dstFilter->length - 1;
464 av_assert0(filter2Size > 0);
465 FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size * dstW * sizeof(*filter2), fail);
467 for (i = 0; i < dstW; i++) {
471 for (k = 0; k < srcFilter->length; k++) {
472 for (j = 0; j < filterSize; j++)
473 filter2[i * filter2Size + k + j] +=
474 srcFilter->coeff[k] * filter[i * filterSize + j];
477 for (j = 0; j < filterSize; j++)
478 filter2[i * filter2Size + j] = filter[i * filterSize + j];
482 (*filterPos)[i] += (filterSize - 1) / 2 - (filter2Size - 1) / 2;
486 /* try to reduce the filter-size (step1 find size and shift left) */
487 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
489 for (i = dstW - 1; i >= 0; i--) {
490 int min = filter2Size;
492 int64_t cutOff = 0.0;
494 /* get rid of near zero elements on the left by shifting left */
495 for (j = 0; j < filter2Size; j++) {
497 cutOff += FFABS(filter2[i * filter2Size]);
499 if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)
502 /* preserve monotonicity because the core can't handle the
503 * filter otherwise */
504 if (i < dstW - 1 && (*filterPos)[i] >= (*filterPos)[i + 1])
507 // move filter coefficients left
508 for (k = 1; k < filter2Size; k++)
509 filter2[i * filter2Size + k - 1] = filter2[i * filter2Size + k];
510 filter2[i * filter2Size + k - 1] = 0;
515 /* count near zeros on the right */
516 for (j = filter2Size - 1; j > 0; j--) {
517 cutOff += FFABS(filter2[i * filter2Size + j]);
519 if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)
524 if (min > minFilterSize)
528 if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) {
529 // we can handle the special case 4, so we don't want to go the full 8
530 if (minFilterSize < 5)
533 /* We really don't want to waste our time doing useless computation, so
534 * fall back on the scalar C code for very small filters.
535 * Vectorizing is worth it only if you have a decent-sized vector. */
536 if (minFilterSize < 3)
540 if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
541 // special case for unscaled vertical filtering
542 if (minFilterSize == 1 && filterAlign == 2)
546 av_assert0(minFilterSize > 0);
547 filterSize = (minFilterSize + (filterAlign - 1)) & (~(filterAlign - 1));
548 av_assert0(filterSize > 0);
549 filter = av_malloc(filterSize * dstW * sizeof(*filter));
550 if (filterSize >= MAX_FILTER_SIZE * 16 /
551 ((flags & SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter) {
552 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);
555 *outFilterSize = filterSize;
557 if (flags & SWS_PRINT_INFO)
558 av_log(NULL, AV_LOG_VERBOSE,
559 "SwScaler: reducing / aligning filtersize %d -> %d\n",
560 filter2Size, filterSize);
561 /* try to reduce the filter-size (step2 reduce it) */
562 for (i = 0; i < dstW; i++) {
565 for (j = 0; j < filterSize; j++) {
566 if (j >= filter2Size)
567 filter[i * filterSize + j] = 0;
569 filter[i * filterSize + j] = filter2[i * filter2Size + j];
570 if ((flags & SWS_BITEXACT) && j >= minFilterSize)
571 filter[i * filterSize + j] = 0;
575 // FIXME try to align filterPos if possible
578 for (i = 0; i < dstW; i++) {
580 if ((*filterPos)[i] < 0) {
581 // move filter coefficients left to compensate for filterPos
582 for (j = 1; j < filterSize; j++) {
583 int left = FFMAX(j + (*filterPos)[i], 0);
584 filter[i * filterSize + left] += filter[i * filterSize + j];
585 filter[i * filterSize + j] = 0;
590 if ((*filterPos)[i] + filterSize > srcW) {
591 int shift = (*filterPos)[i] + filterSize - srcW;
592 // move filter coefficients right to compensate for filterPos
593 for (j = filterSize - 2; j >= 0; j--) {
594 int right = FFMIN(j + shift, filterSize - 1);
595 filter[i * filterSize + right] += filter[i * filterSize + j];
596 filter[i * filterSize + j] = 0;
598 (*filterPos)[i]= srcW - filterSize;
602 // Note the +1 is for the MMX scaler which reads over the end
603 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
604 FF_ALLOCZ_OR_GOTO(NULL, *outFilter,
605 *outFilterSize * (dstW + 3) * sizeof(int16_t), fail);
607 /* normalize & store in outFilter */
608 for (i = 0; i < dstW; i++) {
613 for (j = 0; j < filterSize; j++) {
614 sum += filter[i * filterSize + j];
616 sum = (sum + one / 2) / one;
617 for (j = 0; j < *outFilterSize; j++) {
618 int64_t v = filter[i * filterSize + j] + error;
619 int intV = ROUNDED_DIV(v, sum);
620 (*outFilter)[i * (*outFilterSize) + j] = intV;
621 error = v - intV * sum;
625 (*filterPos)[dstW + 0] =
626 (*filterPos)[dstW + 1] =
627 (*filterPos)[dstW + 2] = (*filterPos)[dstW - 1]; /* the MMX/SSE scaler will
628 * read over the end */
629 for (i = 0; i < *outFilterSize; i++) {
630 int k = (dstW - 1) * (*outFilterSize) + i;
631 (*outFilter)[k + 1 * (*outFilterSize)] =
632 (*outFilter)[k + 2 * (*outFilterSize)] =
633 (*outFilter)[k + 3 * (*outFilterSize)] = (*outFilter)[k];
640 av_log(NULL, AV_LOG_ERROR, "sws: initFilter failed\n");
646 #if HAVE_MMXEXT_INLINE
647 static av_cold int init_hscaler_mmxext(int dstW, int xInc, uint8_t *filterCode,
648 int16_t *filter, int32_t *filterPos,
652 x86_reg imm8OfPShufW1A;
653 x86_reg imm8OfPShufW2A;
654 x86_reg fragmentLengthA;
656 x86_reg imm8OfPShufW1B;
657 x86_reg imm8OfPShufW2B;
658 x86_reg fragmentLengthB;
663 // create an optimized horizontal scaling routine
664 /* This scaler is made of runtime-generated MMXEXT code using specially tuned
665 * pshufw instructions. For every four output pixels, if four input pixels
666 * are enough for the fast bilinear scaling, then a chunk of fragmentB is
667 * used. If five input pixels are needed, then a chunk of fragmentA is used.
676 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
677 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
678 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
679 "punpcklbw %%mm7, %%mm1 \n\t"
680 "punpcklbw %%mm7, %%mm0 \n\t"
681 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
683 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
685 "psubw %%mm1, %%mm0 \n\t"
686 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
687 "pmullw %%mm3, %%mm0 \n\t"
688 "psllw $7, %%mm1 \n\t"
689 "paddw %%mm1, %%mm0 \n\t"
691 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
693 "add $8, %%"REG_a" \n\t"
697 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
698 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
699 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
704 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
708 : "=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
709 "=r" (fragmentLengthA)
716 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
717 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
718 "punpcklbw %%mm7, %%mm0 \n\t"
719 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
721 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
723 "psubw %%mm1, %%mm0 \n\t"
724 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
725 "pmullw %%mm3, %%mm0 \n\t"
726 "psllw $7, %%mm1 \n\t"
727 "paddw %%mm1, %%mm0 \n\t"
729 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
731 "add $8, %%"REG_a" \n\t"
735 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
736 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
737 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
742 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
746 : "=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
747 "=r" (fragmentLengthB)
750 xpos = 0; // lumXInc/2 - 0x8000; // difference between pixel centers
753 for (i = 0; i < dstW / numSplits; i++) {
758 int b = ((xpos + xInc) >> 16) - xx;
759 int c = ((xpos + xInc * 2) >> 16) - xx;
760 int d = ((xpos + xInc * 3) >> 16) - xx;
761 int inc = (d + 1 < 4);
762 uint8_t *fragment = (d + 1 < 4) ? fragmentB : fragmentA;
763 x86_reg imm8OfPShufW1 = (d + 1 < 4) ? imm8OfPShufW1B : imm8OfPShufW1A;
764 x86_reg imm8OfPShufW2 = (d + 1 < 4) ? imm8OfPShufW2B : imm8OfPShufW2A;
765 x86_reg fragmentLength = (d + 1 < 4) ? fragmentLengthB : fragmentLengthA;
766 int maxShift = 3 - (d + inc);
770 filter[i] = ((xpos & 0xFFFF) ^ 0xFFFF) >> 9;
771 filter[i + 1] = (((xpos + xInc) & 0xFFFF) ^ 0xFFFF) >> 9;
772 filter[i + 2] = (((xpos + xInc * 2) & 0xFFFF) ^ 0xFFFF) >> 9;
773 filter[i + 3] = (((xpos + xInc * 3) & 0xFFFF) ^ 0xFFFF) >> 9;
774 filterPos[i / 2] = xx;
776 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
778 filterCode[fragmentPos + imm8OfPShufW1] = (a + inc) |
782 filterCode[fragmentPos + imm8OfPShufW2] = a | (b << 2) |
786 if (i + 4 - inc >= dstW)
787 shift = maxShift; // avoid overread
788 else if ((filterPos[i / 2] & 3) <= maxShift)
789 shift = filterPos[i / 2] & 3; // align
791 if (shift && i >= shift) {
792 filterCode[fragmentPos + imm8OfPShufW1] += 0x55 * shift;
793 filterCode[fragmentPos + imm8OfPShufW2] += 0x55 * shift;
794 filterPos[i / 2] -= shift;
798 fragmentPos += fragmentLength;
801 filterCode[fragmentPos] = RET;
806 filterPos[((i / 2) + 1) & (~1)] = xpos >> 16; // needed to jump to the next part
808 return fragmentPos + 1;
810 #endif /* HAVE_MMXEXT_INLINE */
812 static void getSubSampleFactors(int *h, int *v, enum AVPixelFormat format)
814 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(format);
815 *h = desc->log2_chroma_w;
816 *v = desc->log2_chroma_h;
819 static void fill_rgb2yuv_table(SwsContext *c, const int table[4], int dstRange)
821 int64_t W, V, Z, Cy, Cu, Cv;
822 int64_t vr = table[0];
823 int64_t ub = table[1];
824 int64_t ug = -table[2];
825 int64_t vg = -table[3];
828 uint8_t *p = (uint8_t*)c->input_rgb2yuv_table;
830 static const int8_t map[] = {
831 BY_IDX, GY_IDX, -1 , BY_IDX, BY_IDX, GY_IDX, -1 , BY_IDX,
832 RY_IDX, -1 , GY_IDX, RY_IDX, RY_IDX, -1 , GY_IDX, RY_IDX,
833 RY_IDX, GY_IDX, -1 , RY_IDX, RY_IDX, GY_IDX, -1 , RY_IDX,
834 BY_IDX, -1 , GY_IDX, BY_IDX, BY_IDX, -1 , GY_IDX, BY_IDX,
835 BU_IDX, GU_IDX, -1 , BU_IDX, BU_IDX, GU_IDX, -1 , BU_IDX,
836 RU_IDX, -1 , GU_IDX, RU_IDX, RU_IDX, -1 , GU_IDX, RU_IDX,
837 RU_IDX, GU_IDX, -1 , RU_IDX, RU_IDX, GU_IDX, -1 , RU_IDX,
838 BU_IDX, -1 , GU_IDX, BU_IDX, BU_IDX, -1 , GU_IDX, BU_IDX,
839 BV_IDX, GV_IDX, -1 , BV_IDX, BV_IDX, GV_IDX, -1 , BV_IDX,
840 RV_IDX, -1 , GV_IDX, RV_IDX, RV_IDX, -1 , GV_IDX, RV_IDX,
841 RV_IDX, GV_IDX, -1 , RV_IDX, RV_IDX, GV_IDX, -1 , RV_IDX,
842 BV_IDX, -1 , GV_IDX, BV_IDX, BV_IDX, -1 , GV_IDX, BV_IDX,
843 RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX,
844 BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX,
845 GY_IDX, -1 , GY_IDX, -1 , GY_IDX, -1 , GY_IDX, -1 ,
846 -1 , GY_IDX, -1 , GY_IDX, -1 , GY_IDX, -1 , GY_IDX,
847 RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX,
848 BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX,
849 GU_IDX, -1 , GU_IDX, -1 , GU_IDX, -1 , GU_IDX, -1 ,
850 -1 , GU_IDX, -1 , GU_IDX, -1 , GU_IDX, -1 , GU_IDX,
851 RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX,
852 BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX,
853 GV_IDX, -1 , GV_IDX, -1 , GV_IDX, -1 , GV_IDX, -1 ,
854 -1 , GV_IDX, -1 , GV_IDX, -1 , GV_IDX, -1 , GV_IDX, //23
855 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //24
856 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //25
857 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //26
858 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //27
859 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //28
860 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //29
861 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //30
862 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //31
863 BY_IDX, GY_IDX, RY_IDX, -1 , -1 , -1 , -1 , -1 , //32
864 BU_IDX, GU_IDX, RU_IDX, -1 , -1 , -1 , -1 , -1 , //33
865 BV_IDX, GV_IDX, RV_IDX, -1 , -1 , -1 , -1 , -1 , //34
868 dstRange = 0; //FIXME range = 1 is handled elsewhere
878 W = ROUNDED_DIV(ONE*ONE*ug, ub);
879 V = ROUNDED_DIV(ONE*ONE*vg, vr);
882 Cy = ROUNDED_DIV(cy*Z, ONE);
883 Cu = ROUNDED_DIV(ub*Z, ONE);
884 Cv = ROUNDED_DIV(vr*Z, ONE);
886 c->input_rgb2yuv_table[RY_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*V , Cy);
887 c->input_rgb2yuv_table[GY_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE , Cy);
888 c->input_rgb2yuv_table[BY_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*W , Cy);
890 c->input_rgb2yuv_table[RU_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*V , Cu);
891 c->input_rgb2yuv_table[GU_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE , Cu);
892 c->input_rgb2yuv_table[BU_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*(Z+W) , Cu);
894 c->input_rgb2yuv_table[RV_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*(V+Z) , Cv);
895 c->input_rgb2yuv_table[GV_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE , Cv);
896 c->input_rgb2yuv_table[BV_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*W , Cv);
898 if(/*!dstRange && */table == ff_yuv2rgb_coeffs[SWS_CS_DEFAULT]) {
899 c->input_rgb2yuv_table[BY_IDX] = ((int)(0.114 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
900 c->input_rgb2yuv_table[BV_IDX] = (-(int)(0.081 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
901 c->input_rgb2yuv_table[BU_IDX] = ((int)(0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
902 c->input_rgb2yuv_table[GY_IDX] = ((int)(0.587 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
903 c->input_rgb2yuv_table[GV_IDX] = (-(int)(0.419 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
904 c->input_rgb2yuv_table[GU_IDX] = (-(int)(0.331 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
905 c->input_rgb2yuv_table[RY_IDX] = ((int)(0.299 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
906 c->input_rgb2yuv_table[RV_IDX] = ((int)(0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
907 c->input_rgb2yuv_table[RU_IDX] = (-(int)(0.169 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
909 for(i=0; i<FF_ARRAY_ELEMS(map); i++)
910 AV_WL16(p + 16*4 + 2*i, map[i] >= 0 ? c->input_rgb2yuv_table[map[i]] : 0);
913 static void fill_xyztables(struct SwsContext *c)
916 double xyzgamma = XYZ_GAMMA;
917 double rgbgamma = 1.0 / RGB_GAMMA;
918 static const int16_t xyz2rgb_matrix[3][4] = {
919 {13270, -6295, -2041},
921 { 228, -835, 4329} };
922 static int16_t xyzgamma_tab[4096], rgbgamma_tab[4096];
924 memcpy(c->xyz2rgb_matrix, xyz2rgb_matrix, sizeof(c->xyz2rgb_matrix));
925 c->xyzgamma = xyzgamma_tab;
926 c->rgbgamma = rgbgamma_tab;
928 if (rgbgamma_tab[4095])
931 /* set gamma vectors */
932 for (i = 0; i < 4096; i++) {
933 xyzgamma_tab[i] = lrint(pow(i / 4095.0, xyzgamma) * 4095.0);
934 rgbgamma_tab[i] = lrint(pow(i / 4095.0, rgbgamma) * 4095.0);
938 int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],
939 int srcRange, const int table[4], int dstRange,
940 int brightness, int contrast, int saturation)
942 const AVPixFmtDescriptor *desc_dst;
943 const AVPixFmtDescriptor *desc_src;
944 memcpy(c->srcColorspaceTable, inv_table, sizeof(int) * 4);
945 memcpy(c->dstColorspaceTable, table, sizeof(int) * 4);
948 desc_dst = av_pix_fmt_desc_get(c->dstFormat);
949 desc_src = av_pix_fmt_desc_get(c->srcFormat);
951 if(!isYUV(c->dstFormat) && !isGray(c->dstFormat))
953 if(!isYUV(c->srcFormat) && !isGray(c->srcFormat))
956 c->brightness = brightness;
957 c->contrast = contrast;
958 c->saturation = saturation;
959 c->srcRange = srcRange;
960 c->dstRange = dstRange;
964 if ((isYUV(c->dstFormat) || isGray(c->dstFormat)) && (isYUV(c->srcFormat) || isGray(c->srcFormat)))
967 c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);
968 c->srcFormatBpp = av_get_bits_per_pixel(desc_src);
970 if (!isYUV(c->dstFormat) && !isGray(c->dstFormat)) {
971 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness,
972 contrast, saturation);
975 if (HAVE_ALTIVEC && av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)
976 ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness,
977 contrast, saturation);
980 fill_rgb2yuv_table(c, table, dstRange);
985 int sws_getColorspaceDetails(struct SwsContext *c, int **inv_table,
986 int *srcRange, int **table, int *dstRange,
987 int *brightness, int *contrast, int *saturation)
989 if (!c || isYUV(c->dstFormat) || isGray(c->dstFormat))
992 *inv_table = c->srcColorspaceTable;
993 *table = c->dstColorspaceTable;
994 *srcRange = c->srcRange;
995 *dstRange = c->dstRange;
996 *brightness = c->brightness;
997 *contrast = c->contrast;
998 *saturation = c->saturation;
1003 static int handle_jpeg(enum AVPixelFormat *format)
1006 case AV_PIX_FMT_YUVJ420P:
1007 *format = AV_PIX_FMT_YUV420P;
1009 case AV_PIX_FMT_YUVJ411P:
1010 *format = AV_PIX_FMT_YUV411P;
1012 case AV_PIX_FMT_YUVJ422P:
1013 *format = AV_PIX_FMT_YUV422P;
1015 case AV_PIX_FMT_YUVJ444P:
1016 *format = AV_PIX_FMT_YUV444P;
1018 case AV_PIX_FMT_YUVJ440P:
1019 *format = AV_PIX_FMT_YUV440P;
1021 case AV_PIX_FMT_GRAY8:
1028 static int handle_0alpha(enum AVPixelFormat *format)
1031 case AV_PIX_FMT_0BGR : *format = AV_PIX_FMT_ABGR ; return 1;
1032 case AV_PIX_FMT_BGR0 : *format = AV_PIX_FMT_BGRA ; return 4;
1033 case AV_PIX_FMT_0RGB : *format = AV_PIX_FMT_ARGB ; return 1;
1034 case AV_PIX_FMT_RGB0 : *format = AV_PIX_FMT_RGBA ; return 4;
1039 static int handle_xyz(enum AVPixelFormat *format)
1042 case AV_PIX_FMT_XYZ12BE : *format = AV_PIX_FMT_RGB48BE; return 1;
1043 case AV_PIX_FMT_XYZ12LE : *format = AV_PIX_FMT_RGB48LE; return 1;
1048 static void handle_formats(SwsContext *c)
1050 c->src0Alpha |= handle_0alpha(&c->srcFormat);
1051 c->dst0Alpha |= handle_0alpha(&c->dstFormat);
1052 c->srcXYZ |= handle_xyz(&c->srcFormat);
1053 c->dstXYZ |= handle_xyz(&c->dstFormat);
1056 SwsContext *sws_alloc_context(void)
1058 SwsContext *c = av_mallocz(sizeof(SwsContext));
1061 c->av_class = &sws_context_class;
1062 av_opt_set_defaults(c);
1068 av_cold int sws_init_context(SwsContext *c, SwsFilter *srcFilter,
1069 SwsFilter *dstFilter)
1072 int usesVFilter, usesHFilter;
1074 SwsFilter dummyFilter = { NULL, NULL, NULL, NULL };
1079 int dst_stride = FFALIGN(dstW * sizeof(int16_t) + 66, 16);
1080 int flags, cpu_flags;
1081 enum AVPixelFormat srcFormat = c->srcFormat;
1082 enum AVPixelFormat dstFormat = c->dstFormat;
1083 const AVPixFmtDescriptor *desc_src;
1084 const AVPixFmtDescriptor *desc_dst;
1086 cpu_flags = av_get_cpu_flags();
1092 unscaled = (srcW == dstW && srcH == dstH);
1094 handle_jpeg(&c->srcFormat);
1095 handle_jpeg(&c->dstFormat);
1096 if(srcFormat!=c->srcFormat || dstFormat!=c->dstFormat)
1097 av_log(c, AV_LOG_WARNING, "deprecated pixel format used, make sure you did set range correctly\n");
1099 srcFormat = c->srcFormat;
1100 dstFormat = c->dstFormat;
1101 desc_src = av_pix_fmt_desc_get(srcFormat);
1102 desc_dst = av_pix_fmt_desc_get(dstFormat);
1104 if (!(unscaled && sws_isSupportedEndiannessConversion(srcFormat) &&
1105 av_pix_fmt_swap_endianness(srcFormat) == dstFormat)) {
1106 if (!sws_isSupportedInput(srcFormat)) {
1107 av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n",
1108 av_get_pix_fmt_name(srcFormat));
1109 return AVERROR(EINVAL);
1111 if (!sws_isSupportedOutput(dstFormat)) {
1112 av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n",
1113 av_get_pix_fmt_name(dstFormat));
1114 return AVERROR(EINVAL);
1118 i = flags & (SWS_POINT |
1129 if (!i || (i & (i - 1))) {
1130 av_log(c, AV_LOG_ERROR, "Exactly one scaler algorithm must be chosen, got %X\n", i);
1131 return AVERROR(EINVAL);
1134 if (srcW < 1 || srcH < 1 || dstW < 1 || dstH < 1) {
1135 /* FIXME check if these are enough and try to lower them after
1136 * fixing the relevant parts of the code */
1137 av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",
1138 srcW, srcH, dstW, dstH);
1139 return AVERROR(EINVAL);
1143 dstFilter = &dummyFilter;
1145 srcFilter = &dummyFilter;
1147 c->lumXInc = (((int64_t)srcW << 16) + (dstW >> 1)) / dstW;
1148 c->lumYInc = (((int64_t)srcH << 16) + (dstH >> 1)) / dstH;
1149 c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);
1150 c->srcFormatBpp = av_get_bits_per_pixel(desc_src);
1151 c->vRounder = 4 * 0x0001000100010001ULL;
1153 usesVFilter = (srcFilter->lumV && srcFilter->lumV->length > 1) ||
1154 (srcFilter->chrV && srcFilter->chrV->length > 1) ||
1155 (dstFilter->lumV && dstFilter->lumV->length > 1) ||
1156 (dstFilter->chrV && dstFilter->chrV->length > 1);
1157 usesHFilter = (srcFilter->lumH && srcFilter->lumH->length > 1) ||
1158 (srcFilter->chrH && srcFilter->chrH->length > 1) ||
1159 (dstFilter->lumH && dstFilter->lumH->length > 1) ||
1160 (dstFilter->chrH && dstFilter->chrH->length > 1);
1162 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
1163 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
1165 if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) {
1167 av_log(c, AV_LOG_DEBUG, "Forcing full internal H chroma due to odd output size\n");
1168 flags |= SWS_FULL_CHR_H_INT;
1173 if(dstFormat == AV_PIX_FMT_BGR4_BYTE ||
1174 dstFormat == AV_PIX_FMT_RGB4_BYTE ||
1175 dstFormat == AV_PIX_FMT_BGR8 ||
1176 dstFormat == AV_PIX_FMT_RGB8) {
1177 if (flags & SWS_ERROR_DIFFUSION && !(flags & SWS_FULL_CHR_H_INT)) {
1178 av_log(c, AV_LOG_DEBUG,
1179 "Error diffusion dither is only supported in full chroma interpolation for destination format '%s'\n",
1180 av_get_pix_fmt_name(dstFormat));
1181 flags |= SWS_FULL_CHR_H_INT;
1184 if (!(flags & SWS_ERROR_DIFFUSION) && (flags & SWS_FULL_CHR_H_INT)) {
1185 av_log(c, AV_LOG_DEBUG,
1186 "Ordered dither is not supported in full chroma interpolation for destination format '%s'\n",
1187 av_get_pix_fmt_name(dstFormat));
1188 flags |= SWS_ERROR_DIFFUSION;
1192 if (isPlanarRGB(dstFormat)) {
1193 if (!(flags & SWS_FULL_CHR_H_INT)) {
1194 av_log(c, AV_LOG_DEBUG,
1195 "%s output is not supported with half chroma resolution, switching to full\n",
1196 av_get_pix_fmt_name(dstFormat));
1197 flags |= SWS_FULL_CHR_H_INT;
1202 /* reuse chroma for 2 pixels RGB/BGR unless user wants full
1203 * chroma interpolation */
1204 if (flags & SWS_FULL_CHR_H_INT &&
1205 isAnyRGB(dstFormat) &&
1206 !isPlanarRGB(dstFormat) &&
1207 dstFormat != AV_PIX_FMT_RGBA &&
1208 dstFormat != AV_PIX_FMT_ARGB &&
1209 dstFormat != AV_PIX_FMT_BGRA &&
1210 dstFormat != AV_PIX_FMT_ABGR &&
1211 dstFormat != AV_PIX_FMT_RGB24 &&
1212 dstFormat != AV_PIX_FMT_BGR24 &&
1213 dstFormat != AV_PIX_FMT_BGR4_BYTE &&
1214 dstFormat != AV_PIX_FMT_RGB4_BYTE &&
1215 dstFormat != AV_PIX_FMT_BGR8 &&
1216 dstFormat != AV_PIX_FMT_RGB8
1218 av_log(c, AV_LOG_WARNING,
1219 "full chroma interpolation for destination format '%s' not yet implemented\n",
1220 av_get_pix_fmt_name(dstFormat));
1221 flags &= ~SWS_FULL_CHR_H_INT;
1224 if (isAnyRGB(dstFormat) && !(flags & SWS_FULL_CHR_H_INT))
1225 c->chrDstHSubSample = 1;
1227 // drop some chroma lines if the user wants it
1228 c->vChrDrop = (flags & SWS_SRC_V_CHR_DROP_MASK) >>
1229 SWS_SRC_V_CHR_DROP_SHIFT;
1230 c->chrSrcVSubSample += c->vChrDrop;
1232 /* drop every other pixel for chroma calculation unless user
1233 * wants full chroma */
1234 if (isAnyRGB(srcFormat) && !(flags & SWS_FULL_CHR_H_INP) &&
1235 srcFormat != AV_PIX_FMT_RGB8 && srcFormat != AV_PIX_FMT_BGR8 &&
1236 srcFormat != AV_PIX_FMT_RGB4 && srcFormat != AV_PIX_FMT_BGR4 &&
1237 srcFormat != AV_PIX_FMT_RGB4_BYTE && srcFormat != AV_PIX_FMT_BGR4_BYTE &&
1238 srcFormat != AV_PIX_FMT_GBRP9BE && srcFormat != AV_PIX_FMT_GBRP9LE &&
1239 srcFormat != AV_PIX_FMT_GBRP10BE && srcFormat != AV_PIX_FMT_GBRP10LE &&
1240 srcFormat != AV_PIX_FMT_GBRP12BE && srcFormat != AV_PIX_FMT_GBRP12LE &&
1241 srcFormat != AV_PIX_FMT_GBRP14BE && srcFormat != AV_PIX_FMT_GBRP14LE &&
1242 srcFormat != AV_PIX_FMT_GBRP16BE && srcFormat != AV_PIX_FMT_GBRP16LE &&
1243 ((dstW >> c->chrDstHSubSample) <= (srcW >> 1) ||
1244 (flags & SWS_FAST_BILINEAR)))
1245 c->chrSrcHSubSample = 1;
1247 // Note the FF_CEIL_RSHIFT is so that we always round toward +inf.
1248 c->chrSrcW = FF_CEIL_RSHIFT(srcW, c->chrSrcHSubSample);
1249 c->chrSrcH = FF_CEIL_RSHIFT(srcH, c->chrSrcVSubSample);
1250 c->chrDstW = FF_CEIL_RSHIFT(dstW, c->chrDstHSubSample);
1251 c->chrDstH = FF_CEIL_RSHIFT(dstH, c->chrDstVSubSample);
1253 FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW*2+78, 16) * 2, fail);
1255 /* unscaled special cases */
1256 if (unscaled && !usesHFilter && !usesVFilter &&
1257 (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
1258 ff_get_unscaled_swscale(c);
1261 if (flags & SWS_PRINT_INFO)
1262 av_log(c, AV_LOG_INFO,
1263 "using unscaled %s -> %s special converter\n",
1264 av_get_pix_fmt_name(srcFormat), av_get_pix_fmt_name(dstFormat));
1269 c->srcBpc = 1 + desc_src->comp[0].depth_minus1;
1272 c->dstBpc = 1 + desc_dst->comp[0].depth_minus1;
1275 if (isAnyRGB(srcFormat) || srcFormat == AV_PIX_FMT_PAL8)
1277 if (c->dstBpc == 16)
1280 if (INLINE_MMXEXT(cpu_flags) && c->srcBpc == 8 && c->dstBpc <= 14) {
1281 c->canMMXEXTBeUsed = (dstW >= srcW && (dstW & 31) == 0 &&
1282 (srcW & 15) == 0) ? 1 : 0;
1283 if (!c->canMMXEXTBeUsed && dstW >= srcW && (srcW & 15) == 0
1285 && (flags & SWS_FAST_BILINEAR)) {
1286 if (flags & SWS_PRINT_INFO)
1287 av_log(c, AV_LOG_INFO,
1288 "output width is not a multiple of 32 -> no MMXEXT scaler\n");
1290 if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat))
1291 c->canMMXEXTBeUsed = 0;
1293 c->canMMXEXTBeUsed = 0;
1295 c->chrXInc = (((int64_t)c->chrSrcW << 16) + (c->chrDstW >> 1)) / c->chrDstW;
1296 c->chrYInc = (((int64_t)c->chrSrcH << 16) + (c->chrDstH >> 1)) / c->chrDstH;
1298 /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src
1299 * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do
1301 * n-2 is the last chrominance sample available.
1302 * This is not perfect, but no one should notice the difference, the more
1303 * correct variant would be like the vertical one, but that would require
1304 * some special code for the first and last pixel */
1305 if (flags & SWS_FAST_BILINEAR) {
1306 if (c->canMMXEXTBeUsed) {
1310 // we don't use the x86 asm scaler if MMX is available
1311 else if (INLINE_MMX(cpu_flags) && c->dstBpc <= 14) {
1312 c->lumXInc = ((int64_t)(srcW - 2) << 16) / (dstW - 2) - 20;
1313 c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;
1317 #define USE_MMAP (HAVE_MMAP && HAVE_MPROTECT && defined MAP_ANONYMOUS)
1319 /* precalculate horizontal scaler filter coefficients */
1321 #if HAVE_MMXEXT_INLINE
1322 // can't downscale !!!
1323 if (c->canMMXEXTBeUsed && (flags & SWS_FAST_BILINEAR)) {
1324 c->lumMmxextFilterCodeSize = init_hscaler_mmxext(dstW, c->lumXInc, NULL,
1326 c->chrMmxextFilterCodeSize = init_hscaler_mmxext(c->chrDstW, c->chrXInc,
1327 NULL, NULL, NULL, 4);
1330 c->lumMmxextFilterCode = mmap(NULL, c->lumMmxextFilterCodeSize,
1331 PROT_READ | PROT_WRITE,
1332 MAP_PRIVATE | MAP_ANONYMOUS,
1334 c->chrMmxextFilterCode = mmap(NULL, c->chrMmxextFilterCodeSize,
1335 PROT_READ | PROT_WRITE,
1336 MAP_PRIVATE | MAP_ANONYMOUS,
1338 #elif HAVE_VIRTUALALLOC
1339 c->lumMmxextFilterCode = VirtualAlloc(NULL,
1340 c->lumMmxextFilterCodeSize,
1342 PAGE_EXECUTE_READWRITE);
1343 c->chrMmxextFilterCode = VirtualAlloc(NULL,
1344 c->chrMmxextFilterCodeSize,
1346 PAGE_EXECUTE_READWRITE);
1348 c->lumMmxextFilterCode = av_malloc(c->lumMmxextFilterCodeSize);
1349 c->chrMmxextFilterCode = av_malloc(c->chrMmxextFilterCodeSize);
1352 #ifdef MAP_ANONYMOUS
1353 if (c->lumMmxextFilterCode == MAP_FAILED || c->chrMmxextFilterCode == MAP_FAILED)
1355 if (!c->lumMmxextFilterCode || !c->chrMmxextFilterCode)
1358 av_log(c, AV_LOG_ERROR, "Failed to allocate MMX2FilterCode\n");
1359 return AVERROR(ENOMEM);
1362 FF_ALLOCZ_OR_GOTO(c, c->hLumFilter, (dstW / 8 + 8) * sizeof(int16_t), fail);
1363 FF_ALLOCZ_OR_GOTO(c, c->hChrFilter, (c->chrDstW / 4 + 8) * sizeof(int16_t), fail);
1364 FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW / 2 / 8 + 8) * sizeof(int32_t), fail);
1365 FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);
1367 init_hscaler_mmxext( dstW, c->lumXInc, c->lumMmxextFilterCode,
1368 c->hLumFilter, (uint32_t*)c->hLumFilterPos, 8);
1369 init_hscaler_mmxext(c->chrDstW, c->chrXInc, c->chrMmxextFilterCode,
1370 c->hChrFilter, (uint32_t*)c->hChrFilterPos, 4);
1373 mprotect(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize, PROT_EXEC | PROT_READ);
1374 mprotect(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize, PROT_EXEC | PROT_READ);
1377 #endif /* HAVE_MMXEXT_INLINE */
1379 const int filterAlign =
1380 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :
1381 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
1384 if (initFilter(&c->hLumFilter, &c->hLumFilterPos,
1385 &c->hLumFilterSize, c->lumXInc,
1386 srcW, dstW, filterAlign, 1 << 14,
1387 (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
1388 cpu_flags, srcFilter->lumH, dstFilter->lumH,
1391 if (initFilter(&c->hChrFilter, &c->hChrFilterPos,
1392 &c->hChrFilterSize, c->chrXInc,
1393 c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,
1394 (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
1395 cpu_flags, srcFilter->chrH, dstFilter->chrH,
1399 } // initialize horizontal stuff
1401 /* precalculate vertical scaler filter coefficients */
1403 const int filterAlign =
1404 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 2 :
1405 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
1408 if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,
1409 c->lumYInc, srcH, dstH, filterAlign, (1 << 12),
1410 (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
1411 cpu_flags, srcFilter->lumV, dstFilter->lumV,
1414 if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,
1415 c->chrYInc, c->chrSrcH, c->chrDstH,
1416 filterAlign, (1 << 12),
1417 (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
1418 cpu_flags, srcFilter->chrV, dstFilter->chrV,
1423 FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH, fail);
1424 FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);
1426 for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {
1428 short *p = (short *)&c->vYCoeffsBank[i];
1429 for (j = 0; j < 8; j++)
1430 p[j] = c->vLumFilter[i];
1433 for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {
1435 short *p = (short *)&c->vCCoeffsBank[i];
1436 for (j = 0; j < 8; j++)
1437 p[j] = c->vChrFilter[i];
1442 // calculate buffer sizes so that they won't run out while handling these damn slices
1443 c->vLumBufSize = c->vLumFilterSize;
1444 c->vChrBufSize = c->vChrFilterSize;
1445 for (i = 0; i < dstH; i++) {
1446 int chrI = (int64_t)i * c->chrDstH / dstH;
1447 int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,
1448 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)
1449 << c->chrSrcVSubSample));
1451 nextSlice >>= c->chrSrcVSubSample;
1452 nextSlice <<= c->chrSrcVSubSample;
1453 if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)
1454 c->vLumBufSize = nextSlice - c->vLumFilterPos[i];
1455 if (c->vChrFilterPos[chrI] + c->vChrBufSize <
1456 (nextSlice >> c->chrSrcVSubSample))
1457 c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -
1458 c->vChrFilterPos[chrI];
1461 for (i = 0; i < 4; i++)
1462 FF_ALLOCZ_OR_GOTO(c, c->dither_error[i], (c->dstW+2) * sizeof(int), fail);
1464 /* Allocate pixbufs (we use dynamic allocation because otherwise we would
1465 * need to allocate several megabytes to handle all possible cases) */
1466 FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);
1467 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
1468 FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
1469 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
1470 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);
1471 /* Note we need at least one pixel more at the end because of the MMX code
1472 * (just in case someone wants to replace the 4000/8000). */
1473 /* align at 16 bytes for AltiVec */
1474 for (i = 0; i < c->vLumBufSize; i++) {
1475 FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],
1476 dst_stride + 16, fail);
1477 c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];
1479 // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)
1480 c->uv_off = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);
1481 c->uv_offx2 = dst_stride + 16;
1482 for (i = 0; i < c->vChrBufSize; i++) {
1483 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],
1484 dst_stride * 2 + 32, fail);
1485 c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];
1486 c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]
1487 = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;
1489 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
1490 for (i = 0; i < c->vLumBufSize; i++) {
1491 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i + c->vLumBufSize],
1492 dst_stride + 16, fail);
1493 c->alpPixBuf[i] = c->alpPixBuf[i + c->vLumBufSize];
1496 // try to avoid drawing green stuff between the right end and the stride end
1497 for (i = 0; i < c->vChrBufSize; i++)
1498 if(desc_dst->comp[0].depth_minus1 == 15){
1499 av_assert0(c->dstBpc > 14);
1500 for(j=0; j<dst_stride/2+1; j++)
1501 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;
1503 for(j=0; j<dst_stride+1; j++)
1504 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;
1506 av_assert0(c->chrDstH <= dstH);
1508 if (flags & SWS_PRINT_INFO) {
1509 if (flags & SWS_FAST_BILINEAR)
1510 av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
1511 else if (flags & SWS_BILINEAR)
1512 av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
1513 else if (flags & SWS_BICUBIC)
1514 av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
1515 else if (flags & SWS_X)
1516 av_log(c, AV_LOG_INFO, "Experimental scaler, ");
1517 else if (flags & SWS_POINT)
1518 av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
1519 else if (flags & SWS_AREA)
1520 av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
1521 else if (flags & SWS_BICUBLIN)
1522 av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
1523 else if (flags & SWS_GAUSS)
1524 av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
1525 else if (flags & SWS_SINC)
1526 av_log(c, AV_LOG_INFO, "Sinc scaler, ");
1527 else if (flags & SWS_LANCZOS)
1528 av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
1529 else if (flags & SWS_SPLINE)
1530 av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
1532 av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
1534 av_log(c, AV_LOG_INFO, "from %s to %s%s ",
1535 av_get_pix_fmt_name(srcFormat),
1537 dstFormat == AV_PIX_FMT_BGR555 || dstFormat == AV_PIX_FMT_BGR565 ||
1538 dstFormat == AV_PIX_FMT_RGB444BE || dstFormat == AV_PIX_FMT_RGB444LE ||
1539 dstFormat == AV_PIX_FMT_BGR444BE || dstFormat == AV_PIX_FMT_BGR444LE ?
1544 av_get_pix_fmt_name(dstFormat));
1546 if (INLINE_MMXEXT(cpu_flags))
1547 av_log(c, AV_LOG_INFO, "using MMXEXT\n");
1548 else if (INLINE_AMD3DNOW(cpu_flags))
1549 av_log(c, AV_LOG_INFO, "using 3DNOW\n");
1550 else if (INLINE_MMX(cpu_flags))
1551 av_log(c, AV_LOG_INFO, "using MMX\n");
1552 else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC)
1553 av_log(c, AV_LOG_INFO, "using AltiVec\n");
1555 av_log(c, AV_LOG_INFO, "using C\n");
1557 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
1558 av_log(c, AV_LOG_DEBUG,
1559 "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1560 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
1561 av_log(c, AV_LOG_DEBUG,
1562 "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1563 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,
1564 c->chrXInc, c->chrYInc);
1567 c->swScale = ff_getSwsFunc(c);
1569 fail: // FIXME replace things by appropriate error codes
1573 #if FF_API_SWS_GETCONTEXT
1574 SwsContext *sws_getContext(int srcW, int srcH, enum AVPixelFormat srcFormat,
1575 int dstW, int dstH, enum AVPixelFormat dstFormat,
1576 int flags, SwsFilter *srcFilter,
1577 SwsFilter *dstFilter, const double *param)
1581 if (!(c = sws_alloc_context()))
1589 c->srcRange = handle_jpeg(&srcFormat);
1590 c->dstRange = handle_jpeg(&dstFormat);
1591 c->srcFormat = srcFormat;
1592 c->dstFormat = dstFormat;
1595 c->param[0] = param[0];
1596 c->param[1] = param[1];
1598 sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,
1599 ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,
1600 c->dstRange, 0, 1 << 16, 1 << 16);
1602 if (sws_init_context(c, srcFilter, dstFilter) < 0) {
1611 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
1612 float lumaSharpen, float chromaSharpen,
1613 float chromaHShift, float chromaVShift,
1616 SwsFilter *filter = av_malloc(sizeof(SwsFilter));
1620 if (lumaGBlur != 0.0) {
1621 filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);
1622 filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);
1624 filter->lumH = sws_getIdentityVec();
1625 filter->lumV = sws_getIdentityVec();
1628 if (chromaGBlur != 0.0) {
1629 filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);
1630 filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);
1632 filter->chrH = sws_getIdentityVec();
1633 filter->chrV = sws_getIdentityVec();
1636 if (chromaSharpen != 0.0) {
1637 SwsVector *id = sws_getIdentityVec();
1638 sws_scaleVec(filter->chrH, -chromaSharpen);
1639 sws_scaleVec(filter->chrV, -chromaSharpen);
1640 sws_addVec(filter->chrH, id);
1641 sws_addVec(filter->chrV, id);
1645 if (lumaSharpen != 0.0) {
1646 SwsVector *id = sws_getIdentityVec();
1647 sws_scaleVec(filter->lumH, -lumaSharpen);
1648 sws_scaleVec(filter->lumV, -lumaSharpen);
1649 sws_addVec(filter->lumH, id);
1650 sws_addVec(filter->lumV, id);
1654 if (chromaHShift != 0.0)
1655 sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));
1657 if (chromaVShift != 0.0)
1658 sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));
1660 sws_normalizeVec(filter->chrH, 1.0);
1661 sws_normalizeVec(filter->chrV, 1.0);
1662 sws_normalizeVec(filter->lumH, 1.0);
1663 sws_normalizeVec(filter->lumV, 1.0);
1666 sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
1668 sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
1673 SwsVector *sws_allocVec(int length)
1677 if(length <= 0 || length > INT_MAX/ sizeof(double))
1680 vec = av_malloc(sizeof(SwsVector));
1683 vec->length = length;
1684 vec->coeff = av_malloc(sizeof(double) * length);
1690 SwsVector *sws_getGaussianVec(double variance, double quality)
1692 const int length = (int)(variance * quality + 0.5) | 1;
1694 double middle = (length - 1) * 0.5;
1697 if(variance < 0 || quality < 0)
1700 vec = sws_allocVec(length);
1705 for (i = 0; i < length; i++) {
1706 double dist = i - middle;
1707 vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /
1708 sqrt(2 * variance * M_PI);
1711 sws_normalizeVec(vec, 1.0);
1716 SwsVector *sws_getConstVec(double c, int length)
1719 SwsVector *vec = sws_allocVec(length);
1724 for (i = 0; i < length; i++)
1730 SwsVector *sws_getIdentityVec(void)
1732 return sws_getConstVec(1.0, 1);
1735 static double sws_dcVec(SwsVector *a)
1740 for (i = 0; i < a->length; i++)
1746 void sws_scaleVec(SwsVector *a, double scalar)
1750 for (i = 0; i < a->length; i++)
1751 a->coeff[i] *= scalar;
1754 void sws_normalizeVec(SwsVector *a, double height)
1756 sws_scaleVec(a, height / sws_dcVec(a));
1759 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
1761 int length = a->length + b->length - 1;
1763 SwsVector *vec = sws_getConstVec(0.0, length);
1768 for (i = 0; i < a->length; i++) {
1769 for (j = 0; j < b->length; j++) {
1770 vec->coeff[i + j] += a->coeff[i] * b->coeff[j];
1777 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
1779 int length = FFMAX(a->length, b->length);
1781 SwsVector *vec = sws_getConstVec(0.0, length);
1786 for (i = 0; i < a->length; i++)
1787 vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
1788 for (i = 0; i < b->length; i++)
1789 vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] += b->coeff[i];
1794 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
1796 int length = FFMAX(a->length, b->length);
1798 SwsVector *vec = sws_getConstVec(0.0, length);
1803 for (i = 0; i < a->length; i++)
1804 vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
1805 for (i = 0; i < b->length; i++)
1806 vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] -= b->coeff[i];
1811 /* shift left / or right if "shift" is negative */
1812 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
1814 int length = a->length + FFABS(shift) * 2;
1816 SwsVector *vec = sws_getConstVec(0.0, length);
1821 for (i = 0; i < a->length; i++) {
1822 vec->coeff[i + (length - 1) / 2 -
1823 (a->length - 1) / 2 - shift] = a->coeff[i];
1829 void sws_shiftVec(SwsVector *a, int shift)
1831 SwsVector *shifted = sws_getShiftedVec(a, shift);
1833 a->coeff = shifted->coeff;
1834 a->length = shifted->length;
1838 void sws_addVec(SwsVector *a, SwsVector *b)
1840 SwsVector *sum = sws_sumVec(a, b);
1842 a->coeff = sum->coeff;
1843 a->length = sum->length;
1847 void sws_subVec(SwsVector *a, SwsVector *b)
1849 SwsVector *diff = sws_diffVec(a, b);
1851 a->coeff = diff->coeff;
1852 a->length = diff->length;
1856 void sws_convVec(SwsVector *a, SwsVector *b)
1858 SwsVector *conv = sws_getConvVec(a, b);
1860 a->coeff = conv->coeff;
1861 a->length = conv->length;
1865 SwsVector *sws_cloneVec(SwsVector *a)
1868 SwsVector *vec = sws_allocVec(a->length);
1873 for (i = 0; i < a->length; i++)
1874 vec->coeff[i] = a->coeff[i];
1879 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
1886 for (i = 0; i < a->length; i++)
1887 if (a->coeff[i] > max)
1890 for (i = 0; i < a->length; i++)
1891 if (a->coeff[i] < min)
1896 for (i = 0; i < a->length; i++) {
1897 int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);
1898 av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
1900 av_log(log_ctx, log_level, " ");
1901 av_log(log_ctx, log_level, "|\n");
1905 void sws_freeVec(SwsVector *a)
1909 av_freep(&a->coeff);
1914 void sws_freeFilter(SwsFilter *filter)
1920 sws_freeVec(filter->lumH);
1922 sws_freeVec(filter->lumV);
1924 sws_freeVec(filter->chrH);
1926 sws_freeVec(filter->chrV);
1930 void sws_freeContext(SwsContext *c)
1937 for (i = 0; i < c->vLumBufSize; i++)
1938 av_freep(&c->lumPixBuf[i]);
1939 av_freep(&c->lumPixBuf);
1942 if (c->chrUPixBuf) {
1943 for (i = 0; i < c->vChrBufSize; i++)
1944 av_freep(&c->chrUPixBuf[i]);
1945 av_freep(&c->chrUPixBuf);
1946 av_freep(&c->chrVPixBuf);
1949 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1950 for (i = 0; i < c->vLumBufSize; i++)
1951 av_freep(&c->alpPixBuf[i]);
1952 av_freep(&c->alpPixBuf);
1955 for (i = 0; i < 4; i++)
1956 av_freep(&c->dither_error[i]);
1958 av_freep(&c->vLumFilter);
1959 av_freep(&c->vChrFilter);
1960 av_freep(&c->hLumFilter);
1961 av_freep(&c->hChrFilter);
1963 av_freep(&c->vYCoeffsBank);
1964 av_freep(&c->vCCoeffsBank);
1967 av_freep(&c->vLumFilterPos);
1968 av_freep(&c->vChrFilterPos);
1969 av_freep(&c->hLumFilterPos);
1970 av_freep(&c->hChrFilterPos);
1974 if (c->lumMmxextFilterCode)
1975 munmap(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize);
1976 if (c->chrMmxextFilterCode)
1977 munmap(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize);
1978 #elif HAVE_VIRTUALALLOC
1979 if (c->lumMmxextFilterCode)
1980 VirtualFree(c->lumMmxextFilterCode, 0, MEM_RELEASE);
1981 if (c->chrMmxextFilterCode)
1982 VirtualFree(c->chrMmxextFilterCode, 0, MEM_RELEASE);
1984 av_free(c->lumMmxextFilterCode);
1985 av_free(c->chrMmxextFilterCode);
1987 c->lumMmxextFilterCode = NULL;
1988 c->chrMmxextFilterCode = NULL;
1989 #endif /* HAVE_MMX_INLINE */
1991 av_freep(&c->yuvTable);
1992 av_freep(&c->formatConvBuffer);
1997 struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,
1998 int srcH, enum AVPixelFormat srcFormat,
2000 enum AVPixelFormat dstFormat, int flags,
2001 SwsFilter *srcFilter,
2002 SwsFilter *dstFilter,
2003 const double *param)
2005 static const double default_param[2] = { SWS_PARAM_DEFAULT,
2006 SWS_PARAM_DEFAULT };
2009 param = default_param;
2012 (context->srcW != srcW ||
2013 context->srcH != srcH ||
2014 context->srcFormat != srcFormat ||
2015 context->dstW != dstW ||
2016 context->dstH != dstH ||
2017 context->dstFormat != dstFormat ||
2018 context->flags != flags ||
2019 context->param[0] != param[0] ||
2020 context->param[1] != param[1])) {
2021 sws_freeContext(context);
2026 if (!(context = sws_alloc_context()))
2028 context->srcW = srcW;
2029 context->srcH = srcH;
2030 context->srcRange = handle_jpeg(&srcFormat);
2031 context->srcFormat = srcFormat;
2032 context->dstW = dstW;
2033 context->dstH = dstH;
2034 context->dstRange = handle_jpeg(&dstFormat);
2035 context->dstFormat = dstFormat;
2036 context->flags = flags;
2037 context->param[0] = param[0];
2038 context->param[1] = param[1];
2039 sws_setColorspaceDetails(context, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],
2041 ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,
2042 context->dstRange, 0, 1 << 16, 1 << 16);
2043 if (sws_init_context(context, srcFilter, dstFilter) < 0) {
2044 sws_freeContext(context);