2 * Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
4 * This file is part of FFmpeg.
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 #define _SVID_SOURCE // needed for MAP_ANONYMOUS
24 #define _DARWIN_C_SOURCE // needed for MAP_ANON
31 #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
32 #define MAP_ANONYMOUS MAP_ANON
36 #define WIN32_LEAN_AND_MEAN
40 #include "libavutil/attributes.h"
41 #include "libavutil/avassert.h"
42 #include "libavutil/avutil.h"
43 #include "libavutil/bswap.h"
44 #include "libavutil/cpu.h"
45 #include "libavutil/intreadwrite.h"
46 #include "libavutil/mathematics.h"
47 #include "libavutil/opt.h"
48 #include "libavutil/pixdesc.h"
49 #include "libavutil/ppc/cpu.h"
50 #include "libavutil/x86/asm.h"
51 #include "libavutil/x86/cpu.h"
54 #include "swscale_internal.h"
56 static void handle_formats(SwsContext *c);
58 unsigned swscale_version(void)
60 av_assert0(LIBSWSCALE_VERSION_MICRO >= 100);
61 return LIBSWSCALE_VERSION_INT;
64 const char *swscale_configuration(void)
66 return FFMPEG_CONFIGURATION;
69 const char *swscale_license(void)
71 #define LICENSE_PREFIX "libswscale license: "
72 return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
75 #define RET 0xC3 // near return opcode for x86
77 typedef struct FormatEntry {
78 uint8_t is_supported_in :1;
79 uint8_t is_supported_out :1;
80 uint8_t is_supported_endianness :1;
83 static const FormatEntry format_entries[AV_PIX_FMT_NB] = {
84 [AV_PIX_FMT_YUV420P] = { 1, 1 },
85 [AV_PIX_FMT_YUYV422] = { 1, 1 },
86 [AV_PIX_FMT_RGB24] = { 1, 1 },
87 [AV_PIX_FMT_BGR24] = { 1, 1 },
88 [AV_PIX_FMT_YUV422P] = { 1, 1 },
89 [AV_PIX_FMT_YUV444P] = { 1, 1 },
90 [AV_PIX_FMT_YUV410P] = { 1, 1 },
91 [AV_PIX_FMT_YUV411P] = { 1, 1 },
92 [AV_PIX_FMT_GRAY8] = { 1, 1 },
93 [AV_PIX_FMT_MONOWHITE] = { 1, 1 },
94 [AV_PIX_FMT_MONOBLACK] = { 1, 1 },
95 [AV_PIX_FMT_PAL8] = { 1, 0 },
96 [AV_PIX_FMT_YUVJ420P] = { 1, 1 },
97 [AV_PIX_FMT_YUVJ411P] = { 1, 1 },
98 [AV_PIX_FMT_YUVJ422P] = { 1, 1 },
99 [AV_PIX_FMT_YUVJ444P] = { 1, 1 },
100 [AV_PIX_FMT_UYVY422] = { 1, 1 },
101 [AV_PIX_FMT_UYYVYY411] = { 0, 0 },
102 [AV_PIX_FMT_BGR8] = { 1, 1 },
103 [AV_PIX_FMT_BGR4] = { 0, 1 },
104 [AV_PIX_FMT_BGR4_BYTE] = { 1, 1 },
105 [AV_PIX_FMT_RGB8] = { 1, 1 },
106 [AV_PIX_FMT_RGB4] = { 0, 1 },
107 [AV_PIX_FMT_RGB4_BYTE] = { 1, 1 },
108 [AV_PIX_FMT_NV12] = { 1, 1 },
109 [AV_PIX_FMT_NV21] = { 1, 1 },
110 [AV_PIX_FMT_ARGB] = { 1, 1 },
111 [AV_PIX_FMT_RGBA] = { 1, 1 },
112 [AV_PIX_FMT_ABGR] = { 1, 1 },
113 [AV_PIX_FMT_BGRA] = { 1, 1 },
114 [AV_PIX_FMT_0RGB] = { 1, 1 },
115 [AV_PIX_FMT_RGB0] = { 1, 1 },
116 [AV_PIX_FMT_0BGR] = { 1, 1 },
117 [AV_PIX_FMT_BGR0] = { 1, 1 },
118 [AV_PIX_FMT_GRAY16BE] = { 1, 1 },
119 [AV_PIX_FMT_GRAY16LE] = { 1, 1 },
120 [AV_PIX_FMT_YUV440P] = { 1, 1 },
121 [AV_PIX_FMT_YUVJ440P] = { 1, 1 },
122 [AV_PIX_FMT_YUVA420P] = { 1, 1 },
123 [AV_PIX_FMT_YUVA422P] = { 1, 1 },
124 [AV_PIX_FMT_YUVA444P] = { 1, 1 },
125 [AV_PIX_FMT_YUVA420P9BE] = { 1, 1 },
126 [AV_PIX_FMT_YUVA420P9LE] = { 1, 1 },
127 [AV_PIX_FMT_YUVA422P9BE] = { 1, 1 },
128 [AV_PIX_FMT_YUVA422P9LE] = { 1, 1 },
129 [AV_PIX_FMT_YUVA444P9BE] = { 1, 1 },
130 [AV_PIX_FMT_YUVA444P9LE] = { 1, 1 },
131 [AV_PIX_FMT_YUVA420P10BE]= { 1, 1 },
132 [AV_PIX_FMT_YUVA420P10LE]= { 1, 1 },
133 [AV_PIX_FMT_YUVA422P10BE]= { 1, 1 },
134 [AV_PIX_FMT_YUVA422P10LE]= { 1, 1 },
135 [AV_PIX_FMT_YUVA444P10BE]= { 1, 1 },
136 [AV_PIX_FMT_YUVA444P10LE]= { 1, 1 },
137 [AV_PIX_FMT_YUVA420P16BE]= { 1, 1 },
138 [AV_PIX_FMT_YUVA420P16LE]= { 1, 1 },
139 [AV_PIX_FMT_YUVA422P16BE]= { 1, 1 },
140 [AV_PIX_FMT_YUVA422P16LE]= { 1, 1 },
141 [AV_PIX_FMT_YUVA444P16BE]= { 1, 1 },
142 [AV_PIX_FMT_YUVA444P16LE]= { 1, 1 },
143 [AV_PIX_FMT_RGB48BE] = { 1, 1 },
144 [AV_PIX_FMT_RGB48LE] = { 1, 1 },
145 [AV_PIX_FMT_RGBA64BE] = { 1, 1 },
146 [AV_PIX_FMT_RGBA64LE] = { 1, 1 },
147 [AV_PIX_FMT_RGB565BE] = { 1, 1 },
148 [AV_PIX_FMT_RGB565LE] = { 1, 1 },
149 [AV_PIX_FMT_RGB555BE] = { 1, 1 },
150 [AV_PIX_FMT_RGB555LE] = { 1, 1 },
151 [AV_PIX_FMT_BGR565BE] = { 1, 1 },
152 [AV_PIX_FMT_BGR565LE] = { 1, 1 },
153 [AV_PIX_FMT_BGR555BE] = { 1, 1 },
154 [AV_PIX_FMT_BGR555LE] = { 1, 1 },
155 [AV_PIX_FMT_YUV420P16LE] = { 1, 1 },
156 [AV_PIX_FMT_YUV420P16BE] = { 1, 1 },
157 [AV_PIX_FMT_YUV422P16LE] = { 1, 1 },
158 [AV_PIX_FMT_YUV422P16BE] = { 1, 1 },
159 [AV_PIX_FMT_YUV444P16LE] = { 1, 1 },
160 [AV_PIX_FMT_YUV444P16BE] = { 1, 1 },
161 [AV_PIX_FMT_RGB444LE] = { 1, 1 },
162 [AV_PIX_FMT_RGB444BE] = { 1, 1 },
163 [AV_PIX_FMT_BGR444LE] = { 1, 1 },
164 [AV_PIX_FMT_BGR444BE] = { 1, 1 },
165 [AV_PIX_FMT_Y400A] = { 1, 0 },
166 [AV_PIX_FMT_BGR48BE] = { 1, 1 },
167 [AV_PIX_FMT_BGR48LE] = { 1, 1 },
168 [AV_PIX_FMT_BGRA64BE] = { 0, 0 },
169 [AV_PIX_FMT_BGRA64LE] = { 0, 0 },
170 [AV_PIX_FMT_YUV420P9BE] = { 1, 1 },
171 [AV_PIX_FMT_YUV420P9LE] = { 1, 1 },
172 [AV_PIX_FMT_YUV420P10BE] = { 1, 1 },
173 [AV_PIX_FMT_YUV420P10LE] = { 1, 1 },
174 [AV_PIX_FMT_YUV420P12BE] = { 1, 1 },
175 [AV_PIX_FMT_YUV420P12LE] = { 1, 1 },
176 [AV_PIX_FMT_YUV420P14BE] = { 1, 1 },
177 [AV_PIX_FMT_YUV420P14LE] = { 1, 1 },
178 [AV_PIX_FMT_YUV422P9BE] = { 1, 1 },
179 [AV_PIX_FMT_YUV422P9LE] = { 1, 1 },
180 [AV_PIX_FMT_YUV422P10BE] = { 1, 1 },
181 [AV_PIX_FMT_YUV422P10LE] = { 1, 1 },
182 [AV_PIX_FMT_YUV422P12BE] = { 1, 1 },
183 [AV_PIX_FMT_YUV422P12LE] = { 1, 1 },
184 [AV_PIX_FMT_YUV422P14BE] = { 1, 1 },
185 [AV_PIX_FMT_YUV422P14LE] = { 1, 1 },
186 [AV_PIX_FMT_YUV444P9BE] = { 1, 1 },
187 [AV_PIX_FMT_YUV444P9LE] = { 1, 1 },
188 [AV_PIX_FMT_YUV444P10BE] = { 1, 1 },
189 [AV_PIX_FMT_YUV444P10LE] = { 1, 1 },
190 [AV_PIX_FMT_YUV444P12BE] = { 1, 1 },
191 [AV_PIX_FMT_YUV444P12LE] = { 1, 1 },
192 [AV_PIX_FMT_YUV444P14BE] = { 1, 1 },
193 [AV_PIX_FMT_YUV444P14LE] = { 1, 1 },
194 [AV_PIX_FMT_GBRP] = { 1, 1 },
195 [AV_PIX_FMT_GBRP9LE] = { 1, 1 },
196 [AV_PIX_FMT_GBRP9BE] = { 1, 1 },
197 [AV_PIX_FMT_GBRP10LE] = { 1, 1 },
198 [AV_PIX_FMT_GBRP10BE] = { 1, 1 },
199 [AV_PIX_FMT_GBRP12LE] = { 1, 1 },
200 [AV_PIX_FMT_GBRP12BE] = { 1, 1 },
201 [AV_PIX_FMT_GBRP14LE] = { 1, 1 },
202 [AV_PIX_FMT_GBRP14BE] = { 1, 1 },
203 [AV_PIX_FMT_GBRP16LE] = { 1, 0 },
204 [AV_PIX_FMT_GBRP16BE] = { 1, 0 },
205 [AV_PIX_FMT_XYZ12BE] = { 1, 1, 1 },
206 [AV_PIX_FMT_XYZ12LE] = { 1, 1, 1 },
207 [AV_PIX_FMT_GBRAP] = { 1, 1 },
208 [AV_PIX_FMT_GBRAP16LE] = { 1, 0 },
209 [AV_PIX_FMT_GBRAP16BE] = { 1, 0 },
212 int sws_isSupportedInput(enum AVPixelFormat pix_fmt)
214 return (unsigned)pix_fmt < AV_PIX_FMT_NB ?
215 format_entries[pix_fmt].is_supported_in : 0;
218 int sws_isSupportedOutput(enum AVPixelFormat pix_fmt)
220 return (unsigned)pix_fmt < AV_PIX_FMT_NB ?
221 format_entries[pix_fmt].is_supported_out : 0;
224 int sws_isSupportedEndiannessConversion(enum AVPixelFormat pix_fmt)
226 return (unsigned)pix_fmt < AV_PIX_FMT_NB ?
227 format_entries[pix_fmt].is_supported_endianness : 0;
230 #if FF_API_SWS_FORMAT_NAME
231 const char *sws_format_name(enum AVPixelFormat format)
233 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(format);
237 return "Unknown format";
241 static double getSplineCoeff(double a, double b, double c, double d,
245 return ((d * dist + c) * dist + b) * dist + a;
247 return getSplineCoeff(0.0,
248 b + 2.0 * c + 3.0 * d,
250 -b - 3.0 * c - 6.0 * d,
254 static av_cold int get_local_pos(SwsContext *s, int chr_subsample, int pos, int dir)
257 pos = (128 << chr_subsample) - 128;
259 pos += 128; // relative to ideal left edge
260 return pos >> chr_subsample;
264 int flag; ///< flag associated to the algorithm
265 const char *description; ///< human-readable description
266 int size_factor; ///< size factor used when initing the filters
269 static const ScaleAlgorithm scale_algorithms[] = {
270 { SWS_AREA, "area averaging", 1 /* downscale only, for upscale it is bilinear */ },
271 { SWS_BICUBIC, "bicubic", 4 },
272 { SWS_BICUBLIN, "luma bicubic / chroma bilinear", -1 },
273 { SWS_BILINEAR, "bilinear", 2 },
274 { SWS_FAST_BILINEAR, "fast bilinear", -1 },
275 { SWS_GAUSS, "Gaussian", 8 /* infinite ;) */ },
276 { SWS_LANCZOS, "Lanczos", -1 /* custom */ },
277 { SWS_POINT, "nearest neighbor / point", -1 },
278 { SWS_SINC, "sinc", 20 /* infinite ;) */ },
279 { SWS_SPLINE, "bicubic spline", 20 /* infinite :)*/ },
280 { SWS_X, "experimental", 8 },
283 static av_cold int initFilter(int16_t **outFilter, int32_t **filterPos,
284 int *outFilterSize, int xInc, int srcW,
285 int dstW, int filterAlign, int one,
286 int flags, int cpu_flags,
287 SwsVector *srcFilter, SwsVector *dstFilter,
288 double param[2], int srcPos, int dstPos)
294 int64_t *filter = NULL;
295 int64_t *filter2 = NULL;
296 const int64_t fone = 1LL << (54 - FFMIN(av_log2(srcW/dstW), 8));
299 emms_c(); // FIXME should not be required but IS (even for non-MMX versions)
301 // NOTE: the +3 is for the MMX(+1) / SSE(+3) scaler which reads over the end
302 FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW + 3) * sizeof(**filterPos), fail);
304 if (FFABS(xInc - 0x10000) < 10 && srcPos == dstPos) { // unscaled
307 FF_ALLOCZ_OR_GOTO(NULL, filter,
308 dstW * sizeof(*filter) * filterSize, fail);
310 for (i = 0; i < dstW; i++) {
311 filter[i * filterSize] = fone;
314 } else if (flags & SWS_POINT) { // lame looking point sampling mode
318 FF_ALLOC_OR_GOTO(NULL, filter,
319 dstW * sizeof(*filter) * filterSize, fail);
321 xDstInSrc = ((dstPos*(int64_t)xInc)>>8) - ((srcPos*0x8000LL)>>7);
322 for (i = 0; i < dstW; i++) {
323 int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;
325 (*filterPos)[i] = xx;
329 } else if ((xInc <= (1 << 16) && (flags & SWS_AREA)) ||
330 (flags & SWS_FAST_BILINEAR)) { // bilinear upscale
334 FF_ALLOC_OR_GOTO(NULL, filter,
335 dstW * sizeof(*filter) * filterSize, fail);
337 xDstInSrc = ((dstPos*(int64_t)xInc)>>8) - ((srcPos*0x8000LL)>>7);
338 for (i = 0; i < dstW; i++) {
339 int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;
342 (*filterPos)[i] = xx;
343 // bilinear upscale / linear interpolate / area averaging
344 for (j = 0; j < filterSize; j++) {
345 int64_t coeff= fone - FFABS(((int64_t)xx<<16) - xDstInSrc)*(fone>>16);
348 filter[i * filterSize + j] = coeff;
357 for (i = 0; i < FF_ARRAY_ELEMS(scale_algorithms); i++) {
358 if (flags & scale_algorithms[i].flag) {
359 sizeFactor = scale_algorithms[i].size_factor;
363 if (flags & SWS_LANCZOS)
364 sizeFactor = param[0] != SWS_PARAM_DEFAULT ? ceil(2 * param[0]) : 6;
365 av_assert0(sizeFactor > 0);
368 filterSize = 1 + sizeFactor; // upscale
370 filterSize = 1 + (sizeFactor * srcW + dstW - 1) / dstW;
372 filterSize = FFMIN(filterSize, srcW - 2);
373 filterSize = FFMAX(filterSize, 1);
375 FF_ALLOC_OR_GOTO(NULL, filter,
376 dstW * sizeof(*filter) * filterSize, fail);
378 xDstInSrc = ((dstPos*(int64_t)xInc)>>7) - ((srcPos*0x10000LL)>>7);
379 for (i = 0; i < dstW; i++) {
380 int xx = (xDstInSrc - ((filterSize - 2) << 16)) / (1 << 17);
382 (*filterPos)[i] = xx;
383 for (j = 0; j < filterSize; j++) {
384 int64_t d = (FFABS(((int64_t)xx << 17) - xDstInSrc)) << 13;
390 floatd = d * (1.0 / (1 << 30));
392 if (flags & SWS_BICUBIC) {
393 int64_t B = (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1 << 24);
394 int64_t C = (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1 << 24);
396 if (d >= 1LL << 31) {
399 int64_t dd = (d * d) >> 30;
400 int64_t ddd = (dd * d) >> 30;
403 coeff = (12 * (1 << 24) - 9 * B - 6 * C) * ddd +
404 (-18 * (1 << 24) + 12 * B + 6 * C) * dd +
405 (6 * (1 << 24) - 2 * B) * (1 << 30);
407 coeff = (-B - 6 * C) * ddd +
408 (6 * B + 30 * C) * dd +
409 (-12 * B - 48 * C) * d +
410 (8 * B + 24 * C) * (1 << 30);
412 coeff /= (1LL<<54)/fone;
415 else if (flags & SWS_X) {
416 double p = param ? param * 0.01 : 0.3;
417 coeff = d ? sin(d * M_PI) / (d * M_PI) : 1.0;
418 coeff *= pow(2.0, -p * d * d);
421 else if (flags & SWS_X) {
422 double A = param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
426 c = cos(floatd * M_PI);
433 coeff = (c * 0.5 + 0.5) * fone;
434 } else if (flags & SWS_AREA) {
435 int64_t d2 = d - (1 << 29);
436 if (d2 * xInc < -(1LL << (29 + 16)))
437 coeff = 1.0 * (1LL << (30 + 16));
438 else if (d2 * xInc < (1LL << (29 + 16)))
439 coeff = -d2 * xInc + (1LL << (29 + 16));
442 coeff *= fone >> (30 + 16);
443 } else if (flags & SWS_GAUSS) {
444 double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
445 coeff = (pow(2.0, -p * floatd * floatd)) * fone;
446 } else if (flags & SWS_SINC) {
447 coeff = (d ? sin(floatd * M_PI) / (floatd * M_PI) : 1.0) * fone;
448 } else if (flags & SWS_LANCZOS) {
449 double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
450 coeff = (d ? sin(floatd * M_PI) * sin(floatd * M_PI / p) /
451 (floatd * floatd * M_PI * M_PI / p) : 1.0) * fone;
454 } else if (flags & SWS_BILINEAR) {
455 coeff = (1 << 30) - d;
459 } else if (flags & SWS_SPLINE) {
460 double p = -2.196152422706632;
461 coeff = getSplineCoeff(1.0, 0.0, p, -p - 1.0, floatd) * fone;
466 filter[i * filterSize + j] = coeff;
469 xDstInSrc += 2 * xInc;
473 /* apply src & dst Filter to filter -> filter2
476 av_assert0(filterSize > 0);
477 filter2Size = filterSize;
479 filter2Size += srcFilter->length - 1;
481 filter2Size += dstFilter->length - 1;
482 av_assert0(filter2Size > 0);
483 FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size * dstW * sizeof(*filter2), fail);
485 for (i = 0; i < dstW; i++) {
489 for (k = 0; k < srcFilter->length; k++) {
490 for (j = 0; j < filterSize; j++)
491 filter2[i * filter2Size + k + j] +=
492 srcFilter->coeff[k] * filter[i * filterSize + j];
495 for (j = 0; j < filterSize; j++)
496 filter2[i * filter2Size + j] = filter[i * filterSize + j];
500 (*filterPos)[i] += (filterSize - 1) / 2 - (filter2Size - 1) / 2;
504 /* try to reduce the filter-size (step1 find size and shift left) */
505 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
507 for (i = dstW - 1; i >= 0; i--) {
508 int min = filter2Size;
510 int64_t cutOff = 0.0;
512 /* get rid of near zero elements on the left by shifting left */
513 for (j = 0; j < filter2Size; j++) {
515 cutOff += FFABS(filter2[i * filter2Size]);
517 if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)
520 /* preserve monotonicity because the core can't handle the
521 * filter otherwise */
522 if (i < dstW - 1 && (*filterPos)[i] >= (*filterPos)[i + 1])
525 // move filter coefficients left
526 for (k = 1; k < filter2Size; k++)
527 filter2[i * filter2Size + k - 1] = filter2[i * filter2Size + k];
528 filter2[i * filter2Size + k - 1] = 0;
533 /* count near zeros on the right */
534 for (j = filter2Size - 1; j > 0; j--) {
535 cutOff += FFABS(filter2[i * filter2Size + j]);
537 if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)
542 if (min > minFilterSize)
546 if (PPC_ALTIVEC(cpu_flags)) {
547 // we can handle the special case 4, so we don't want to go the full 8
548 if (minFilterSize < 5)
551 /* We really don't want to waste our time doing useless computation, so
552 * fall back on the scalar C code for very small filters.
553 * Vectorizing is worth it only if you have a decent-sized vector. */
554 if (minFilterSize < 3)
558 if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
559 // special case for unscaled vertical filtering
560 if (minFilterSize == 1 && filterAlign == 2)
564 av_assert0(minFilterSize > 0);
565 filterSize = (minFilterSize + (filterAlign - 1)) & (~(filterAlign - 1));
566 av_assert0(filterSize > 0);
567 filter = av_malloc(filterSize * dstW * sizeof(*filter));
568 if (filterSize >= MAX_FILTER_SIZE * 16 /
569 ((flags & SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter) {
570 av_log(NULL, AV_LOG_ERROR, "sws: filterSize %d is too large, try less extreme scaling or increase MAX_FILTER_SIZE and recompile\n", filterSize);
573 *outFilterSize = filterSize;
575 if (flags & SWS_PRINT_INFO)
576 av_log(NULL, AV_LOG_VERBOSE,
577 "SwScaler: reducing / aligning filtersize %d -> %d\n",
578 filter2Size, filterSize);
579 /* try to reduce the filter-size (step2 reduce it) */
580 for (i = 0; i < dstW; i++) {
583 for (j = 0; j < filterSize; j++) {
584 if (j >= filter2Size)
585 filter[i * filterSize + j] = 0;
587 filter[i * filterSize + j] = filter2[i * filter2Size + j];
588 if ((flags & SWS_BITEXACT) && j >= minFilterSize)
589 filter[i * filterSize + j] = 0;
593 // FIXME try to align filterPos if possible
596 for (i = 0; i < dstW; i++) {
598 if ((*filterPos)[i] < 0) {
599 // move filter coefficients left to compensate for filterPos
600 for (j = 1; j < filterSize; j++) {
601 int left = FFMAX(j + (*filterPos)[i], 0);
602 filter[i * filterSize + left] += filter[i * filterSize + j];
603 filter[i * filterSize + j] = 0;
608 if ((*filterPos)[i] + filterSize > srcW) {
609 int shift = (*filterPos)[i] + filterSize - srcW;
610 // move filter coefficients right to compensate for filterPos
611 for (j = filterSize - 2; j >= 0; j--) {
612 int right = FFMIN(j + shift, filterSize - 1);
613 filter[i * filterSize + right] += filter[i * filterSize + j];
614 filter[i * filterSize + j] = 0;
616 (*filterPos)[i]= srcW - filterSize;
620 // Note the +1 is for the MMX scaler which reads over the end
621 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
622 FF_ALLOCZ_OR_GOTO(NULL, *outFilter,
623 *outFilterSize * (dstW + 3) * sizeof(int16_t), fail);
625 /* normalize & store in outFilter */
626 for (i = 0; i < dstW; i++) {
631 for (j = 0; j < filterSize; j++) {
632 sum += filter[i * filterSize + j];
634 sum = (sum + one / 2) / one;
636 av_log(NULL, AV_LOG_WARNING, "SwScaler: zero vector in scaling\n");
639 for (j = 0; j < *outFilterSize; j++) {
640 int64_t v = filter[i * filterSize + j] + error;
641 int intV = ROUNDED_DIV(v, sum);
642 (*outFilter)[i * (*outFilterSize) + j] = intV;
643 error = v - intV * sum;
647 (*filterPos)[dstW + 0] =
648 (*filterPos)[dstW + 1] =
649 (*filterPos)[dstW + 2] = (*filterPos)[dstW - 1]; /* the MMX/SSE scaler will
650 * read over the end */
651 for (i = 0; i < *outFilterSize; i++) {
652 int k = (dstW - 1) * (*outFilterSize) + i;
653 (*outFilter)[k + 1 * (*outFilterSize)] =
654 (*outFilter)[k + 2 * (*outFilterSize)] =
655 (*outFilter)[k + 3 * (*outFilterSize)] = (*outFilter)[k];
662 av_log(NULL, AV_LOG_ERROR, "sws: initFilter failed\n");
668 #if HAVE_MMXEXT_INLINE
669 static av_cold int init_hscaler_mmxext(int dstW, int xInc, uint8_t *filterCode,
670 int16_t *filter, int32_t *filterPos,
674 x86_reg imm8OfPShufW1A;
675 x86_reg imm8OfPShufW2A;
676 x86_reg fragmentLengthA;
678 x86_reg imm8OfPShufW1B;
679 x86_reg imm8OfPShufW2B;
680 x86_reg fragmentLengthB;
685 // create an optimized horizontal scaling routine
686 /* This scaler is made of runtime-generated MMXEXT code using specially tuned
687 * pshufw instructions. For every four output pixels, if four input pixels
688 * are enough for the fast bilinear scaling, then a chunk of fragmentB is
689 * used. If five input pixels are needed, then a chunk of fragmentA is used.
698 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
699 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
700 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
701 "punpcklbw %%mm7, %%mm1 \n\t"
702 "punpcklbw %%mm7, %%mm0 \n\t"
703 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
705 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
707 "psubw %%mm1, %%mm0 \n\t"
708 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
709 "pmullw %%mm3, %%mm0 \n\t"
710 "psllw $7, %%mm1 \n\t"
711 "paddw %%mm1, %%mm0 \n\t"
713 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
715 "add $8, %%"REG_a" \n\t"
719 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
720 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
721 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
726 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
730 : "=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
731 "=r" (fragmentLengthA)
738 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
739 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
740 "punpcklbw %%mm7, %%mm0 \n\t"
741 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
743 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
745 "psubw %%mm1, %%mm0 \n\t"
746 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
747 "pmullw %%mm3, %%mm0 \n\t"
748 "psllw $7, %%mm1 \n\t"
749 "paddw %%mm1, %%mm0 \n\t"
751 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
753 "add $8, %%"REG_a" \n\t"
757 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
758 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
759 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
764 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
768 : "=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
769 "=r" (fragmentLengthB)
772 xpos = 0; // lumXInc/2 - 0x8000; // difference between pixel centers
775 for (i = 0; i < dstW / numSplits; i++) {
780 int b = ((xpos + xInc) >> 16) - xx;
781 int c = ((xpos + xInc * 2) >> 16) - xx;
782 int d = ((xpos + xInc * 3) >> 16) - xx;
783 int inc = (d + 1 < 4);
784 uint8_t *fragment = inc ? fragmentB : fragmentA;
785 x86_reg imm8OfPShufW1 = inc ? imm8OfPShufW1B : imm8OfPShufW1A;
786 x86_reg imm8OfPShufW2 = inc ? imm8OfPShufW2B : imm8OfPShufW2A;
787 x86_reg fragmentLength = inc ? fragmentLengthB : fragmentLengthA;
788 int maxShift = 3 - (d + inc);
792 filter[i] = ((xpos & 0xFFFF) ^ 0xFFFF) >> 9;
793 filter[i + 1] = (((xpos + xInc) & 0xFFFF) ^ 0xFFFF) >> 9;
794 filter[i + 2] = (((xpos + xInc * 2) & 0xFFFF) ^ 0xFFFF) >> 9;
795 filter[i + 3] = (((xpos + xInc * 3) & 0xFFFF) ^ 0xFFFF) >> 9;
796 filterPos[i / 2] = xx;
798 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
800 filterCode[fragmentPos + imm8OfPShufW1] = (a + inc) |
804 filterCode[fragmentPos + imm8OfPShufW2] = a | (b << 2) |
808 if (i + 4 - inc >= dstW)
809 shift = maxShift; // avoid overread
810 else if ((filterPos[i / 2] & 3) <= maxShift)
811 shift = filterPos[i / 2] & 3; // align
813 if (shift && i >= shift) {
814 filterCode[fragmentPos + imm8OfPShufW1] += 0x55 * shift;
815 filterCode[fragmentPos + imm8OfPShufW2] += 0x55 * shift;
816 filterPos[i / 2] -= shift;
820 fragmentPos += fragmentLength;
823 filterCode[fragmentPos] = RET;
828 filterPos[((i / 2) + 1) & (~1)] = xpos >> 16; // needed to jump to the next part
830 return fragmentPos + 1;
832 #endif /* HAVE_MMXEXT_INLINE */
834 static void fill_rgb2yuv_table(SwsContext *c, const int table[4], int dstRange)
836 int64_t W, V, Z, Cy, Cu, Cv;
837 int64_t vr = table[0];
838 int64_t ub = table[1];
839 int64_t ug = -table[2];
840 int64_t vg = -table[3];
843 uint8_t *p = (uint8_t*)c->input_rgb2yuv_table;
845 static const int8_t map[] = {
846 BY_IDX, GY_IDX, -1 , BY_IDX, BY_IDX, GY_IDX, -1 , BY_IDX,
847 RY_IDX, -1 , GY_IDX, RY_IDX, RY_IDX, -1 , GY_IDX, RY_IDX,
848 RY_IDX, GY_IDX, -1 , RY_IDX, RY_IDX, GY_IDX, -1 , RY_IDX,
849 BY_IDX, -1 , GY_IDX, BY_IDX, BY_IDX, -1 , GY_IDX, BY_IDX,
850 BU_IDX, GU_IDX, -1 , BU_IDX, BU_IDX, GU_IDX, -1 , BU_IDX,
851 RU_IDX, -1 , GU_IDX, RU_IDX, RU_IDX, -1 , GU_IDX, RU_IDX,
852 RU_IDX, GU_IDX, -1 , RU_IDX, RU_IDX, GU_IDX, -1 , RU_IDX,
853 BU_IDX, -1 , GU_IDX, BU_IDX, BU_IDX, -1 , GU_IDX, BU_IDX,
854 BV_IDX, GV_IDX, -1 , BV_IDX, BV_IDX, GV_IDX, -1 , BV_IDX,
855 RV_IDX, -1 , GV_IDX, RV_IDX, RV_IDX, -1 , GV_IDX, RV_IDX,
856 RV_IDX, GV_IDX, -1 , RV_IDX, RV_IDX, GV_IDX, -1 , RV_IDX,
857 BV_IDX, -1 , GV_IDX, BV_IDX, BV_IDX, -1 , GV_IDX, BV_IDX,
858 RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX,
859 BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX,
860 GY_IDX, -1 , GY_IDX, -1 , GY_IDX, -1 , GY_IDX, -1 ,
861 -1 , GY_IDX, -1 , GY_IDX, -1 , GY_IDX, -1 , GY_IDX,
862 RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX,
863 BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX,
864 GU_IDX, -1 , GU_IDX, -1 , GU_IDX, -1 , GU_IDX, -1 ,
865 -1 , GU_IDX, -1 , GU_IDX, -1 , GU_IDX, -1 , GU_IDX,
866 RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX,
867 BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX,
868 GV_IDX, -1 , GV_IDX, -1 , GV_IDX, -1 , GV_IDX, -1 ,
869 -1 , GV_IDX, -1 , GV_IDX, -1 , GV_IDX, -1 , GV_IDX, //23
870 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //24
871 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //25
872 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //26
873 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //27
874 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //28
875 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //29
876 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //30
877 -1 , -1 , -1 , -1 , -1 , -1 , -1 , -1 , //31
878 BY_IDX, GY_IDX, RY_IDX, -1 , -1 , -1 , -1 , -1 , //32
879 BU_IDX, GU_IDX, RU_IDX, -1 , -1 , -1 , -1 , -1 , //33
880 BV_IDX, GV_IDX, RV_IDX, -1 , -1 , -1 , -1 , -1 , //34
883 dstRange = 0; //FIXME range = 1 is handled elsewhere
893 W = ROUNDED_DIV(ONE*ONE*ug, ub);
894 V = ROUNDED_DIV(ONE*ONE*vg, vr);
897 Cy = ROUNDED_DIV(cy*Z, ONE);
898 Cu = ROUNDED_DIV(ub*Z, ONE);
899 Cv = ROUNDED_DIV(vr*Z, ONE);
901 c->input_rgb2yuv_table[RY_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*V , Cy);
902 c->input_rgb2yuv_table[GY_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE , Cy);
903 c->input_rgb2yuv_table[BY_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*W , Cy);
905 c->input_rgb2yuv_table[RU_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*V , Cu);
906 c->input_rgb2yuv_table[GU_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE , Cu);
907 c->input_rgb2yuv_table[BU_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*(Z+W) , Cu);
909 c->input_rgb2yuv_table[RV_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*(V+Z) , Cv);
910 c->input_rgb2yuv_table[GV_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE , Cv);
911 c->input_rgb2yuv_table[BV_IDX] = ROUNDED_DIV((1 << RGB2YUV_SHIFT)*W , Cv);
913 if(/*!dstRange && */!memcmp(table, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], sizeof(ff_yuv2rgb_coeffs[SWS_CS_DEFAULT]))) {
914 c->input_rgb2yuv_table[BY_IDX] = ((int)(0.114 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
915 c->input_rgb2yuv_table[BV_IDX] = (-(int)(0.081 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
916 c->input_rgb2yuv_table[BU_IDX] = ((int)(0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
917 c->input_rgb2yuv_table[GY_IDX] = ((int)(0.587 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
918 c->input_rgb2yuv_table[GV_IDX] = (-(int)(0.419 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
919 c->input_rgb2yuv_table[GU_IDX] = (-(int)(0.331 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
920 c->input_rgb2yuv_table[RY_IDX] = ((int)(0.299 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
921 c->input_rgb2yuv_table[RV_IDX] = ((int)(0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
922 c->input_rgb2yuv_table[RU_IDX] = (-(int)(0.169 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
924 for(i=0; i<FF_ARRAY_ELEMS(map); i++)
925 AV_WL16(p + 16*4 + 2*i, map[i] >= 0 ? c->input_rgb2yuv_table[map[i]] : 0);
928 static void fill_xyztables(struct SwsContext *c)
931 double xyzgamma = XYZ_GAMMA;
932 double rgbgamma = 1.0 / RGB_GAMMA;
933 double xyzgammainv = 1.0 / XYZ_GAMMA;
934 double rgbgammainv = RGB_GAMMA;
935 static const int16_t xyz2rgb_matrix[3][4] = {
936 {13270, -6295, -2041},
938 { 228, -835, 4329} };
939 static const int16_t rgb2xyz_matrix[3][4] = {
943 static int16_t xyzgamma_tab[4096], rgbgamma_tab[4096], xyzgammainv_tab[4096], rgbgammainv_tab[4096];
945 memcpy(c->xyz2rgb_matrix, xyz2rgb_matrix, sizeof(c->xyz2rgb_matrix));
946 memcpy(c->rgb2xyz_matrix, rgb2xyz_matrix, sizeof(c->rgb2xyz_matrix));
947 c->xyzgamma = xyzgamma_tab;
948 c->rgbgamma = rgbgamma_tab;
949 c->xyzgammainv = xyzgammainv_tab;
950 c->rgbgammainv = rgbgammainv_tab;
952 if (rgbgamma_tab[4095])
955 /* set gamma vectors */
956 for (i = 0; i < 4096; i++) {
957 xyzgamma_tab[i] = lrint(pow(i / 4095.0, xyzgamma) * 4095.0);
958 rgbgamma_tab[i] = lrint(pow(i / 4095.0, rgbgamma) * 4095.0);
959 xyzgammainv_tab[i] = lrint(pow(i / 4095.0, xyzgammainv) * 4095.0);
960 rgbgammainv_tab[i] = lrint(pow(i / 4095.0, rgbgammainv) * 4095.0);
964 int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],
965 int srcRange, const int table[4], int dstRange,
966 int brightness, int contrast, int saturation)
968 const AVPixFmtDescriptor *desc_dst;
969 const AVPixFmtDescriptor *desc_src;
970 memmove(c->srcColorspaceTable, inv_table, sizeof(int) * 4);
971 memmove(c->dstColorspaceTable, table, sizeof(int) * 4);
974 desc_dst = av_pix_fmt_desc_get(c->dstFormat);
975 desc_src = av_pix_fmt_desc_get(c->srcFormat);
977 if(!isYUV(c->dstFormat) && !isGray(c->dstFormat))
979 if(!isYUV(c->srcFormat) && !isGray(c->srcFormat))
982 c->brightness = brightness;
983 c->contrast = contrast;
984 c->saturation = saturation;
985 c->srcRange = srcRange;
986 c->dstRange = dstRange;
988 if ((isYUV(c->dstFormat) || isGray(c->dstFormat)) && (isYUV(c->srcFormat) || isGray(c->srcFormat)))
991 c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);
992 c->srcFormatBpp = av_get_bits_per_pixel(desc_src);
994 if (!isYUV(c->dstFormat) && !isGray(c->dstFormat)) {
995 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness,
996 contrast, saturation);
1000 ff_yuv2rgb_init_tables_ppc(c, inv_table, brightness,
1001 contrast, saturation);
1004 fill_rgb2yuv_table(c, table, dstRange);
1009 int sws_getColorspaceDetails(struct SwsContext *c, int **inv_table,
1010 int *srcRange, int **table, int *dstRange,
1011 int *brightness, int *contrast, int *saturation)
1016 *inv_table = c->srcColorspaceTable;
1017 *table = c->dstColorspaceTable;
1018 *srcRange = c->srcRange;
1019 *dstRange = c->dstRange;
1020 *brightness = c->brightness;
1021 *contrast = c->contrast;
1022 *saturation = c->saturation;
1027 static int handle_jpeg(enum AVPixelFormat *format)
1030 case AV_PIX_FMT_YUVJ420P:
1031 *format = AV_PIX_FMT_YUV420P;
1033 case AV_PIX_FMT_YUVJ411P:
1034 *format = AV_PIX_FMT_YUV411P;
1036 case AV_PIX_FMT_YUVJ422P:
1037 *format = AV_PIX_FMT_YUV422P;
1039 case AV_PIX_FMT_YUVJ444P:
1040 *format = AV_PIX_FMT_YUV444P;
1042 case AV_PIX_FMT_YUVJ440P:
1043 *format = AV_PIX_FMT_YUV440P;
1045 case AV_PIX_FMT_GRAY8:
1052 static int handle_0alpha(enum AVPixelFormat *format)
1055 case AV_PIX_FMT_0BGR : *format = AV_PIX_FMT_ABGR ; return 1;
1056 case AV_PIX_FMT_BGR0 : *format = AV_PIX_FMT_BGRA ; return 4;
1057 case AV_PIX_FMT_0RGB : *format = AV_PIX_FMT_ARGB ; return 1;
1058 case AV_PIX_FMT_RGB0 : *format = AV_PIX_FMT_RGBA ; return 4;
1063 static int handle_xyz(enum AVPixelFormat *format)
1066 case AV_PIX_FMT_XYZ12BE : *format = AV_PIX_FMT_RGB48BE; return 1;
1067 case AV_PIX_FMT_XYZ12LE : *format = AV_PIX_FMT_RGB48LE; return 1;
1072 static void handle_formats(SwsContext *c)
1074 c->src0Alpha |= handle_0alpha(&c->srcFormat);
1075 c->dst0Alpha |= handle_0alpha(&c->dstFormat);
1076 c->srcXYZ |= handle_xyz(&c->srcFormat);
1077 c->dstXYZ |= handle_xyz(&c->dstFormat);
1078 if (c->srcXYZ || c->dstXYZ)
1082 SwsContext *sws_alloc_context(void)
1084 SwsContext *c = av_mallocz(sizeof(SwsContext));
1087 c->av_class = &sws_context_class;
1088 av_opt_set_defaults(c);
1094 av_cold int sws_init_context(SwsContext *c, SwsFilter *srcFilter,
1095 SwsFilter *dstFilter)
1098 int usesVFilter, usesHFilter;
1100 SwsFilter dummyFilter = { NULL, NULL, NULL, NULL };
1105 int dst_stride = FFALIGN(dstW * sizeof(int16_t) + 66, 16);
1106 int flags, cpu_flags;
1107 enum AVPixelFormat srcFormat = c->srcFormat;
1108 enum AVPixelFormat dstFormat = c->dstFormat;
1109 const AVPixFmtDescriptor *desc_src;
1110 const AVPixFmtDescriptor *desc_dst;
1112 cpu_flags = av_get_cpu_flags();
1118 unscaled = (srcW == dstW && srcH == dstH);
1120 c->srcRange |= handle_jpeg(&c->srcFormat);
1121 c->dstRange |= handle_jpeg(&c->dstFormat);
1123 if(srcFormat!=c->srcFormat || dstFormat!=c->dstFormat)
1124 av_log(c, AV_LOG_WARNING, "deprecated pixel format used, make sure you did set range correctly\n");
1126 if (!c->contrast && !c->saturation && !c->dstFormatBpp)
1127 sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,
1128 ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],
1129 c->dstRange, 0, 1 << 16, 1 << 16);
1132 srcFormat = c->srcFormat;
1133 dstFormat = c->dstFormat;
1134 desc_src = av_pix_fmt_desc_get(srcFormat);
1135 desc_dst = av_pix_fmt_desc_get(dstFormat);
1137 if (!(unscaled && sws_isSupportedEndiannessConversion(srcFormat) &&
1138 av_pix_fmt_swap_endianness(srcFormat) == dstFormat)) {
1139 if (!sws_isSupportedInput(srcFormat)) {
1140 av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n",
1141 av_get_pix_fmt_name(srcFormat));
1142 return AVERROR(EINVAL);
1144 if (!sws_isSupportedOutput(dstFormat)) {
1145 av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n",
1146 av_get_pix_fmt_name(dstFormat));
1147 return AVERROR(EINVAL);
1151 i = flags & (SWS_POINT |
1163 /* provide a default scaler if not set by caller */
1165 if (dstW < srcW && dstH < srcH)
1166 flags |= SWS_BICUBIC;
1167 else if (dstW > srcW && dstH > srcH)
1168 flags |= SWS_BICUBIC;
1170 flags |= SWS_BICUBIC;
1172 } else if (i & (i - 1)) {
1173 av_log(c, AV_LOG_ERROR,
1174 "Exactly one scaler algorithm must be chosen, got %X\n", i);
1175 return AVERROR(EINVAL);
1178 if (srcW < 1 || srcH < 1 || dstW < 1 || dstH < 1) {
1179 /* FIXME check if these are enough and try to lower them after
1180 * fixing the relevant parts of the code */
1181 av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",
1182 srcW, srcH, dstW, dstH);
1183 return AVERROR(EINVAL);
1187 dstFilter = &dummyFilter;
1189 srcFilter = &dummyFilter;
1191 c->lumXInc = (((int64_t)srcW << 16) + (dstW >> 1)) / dstW;
1192 c->lumYInc = (((int64_t)srcH << 16) + (dstH >> 1)) / dstH;
1193 c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);
1194 c->srcFormatBpp = av_get_bits_per_pixel(desc_src);
1195 c->vRounder = 4 * 0x0001000100010001ULL;
1197 usesVFilter = (srcFilter->lumV && srcFilter->lumV->length > 1) ||
1198 (srcFilter->chrV && srcFilter->chrV->length > 1) ||
1199 (dstFilter->lumV && dstFilter->lumV->length > 1) ||
1200 (dstFilter->chrV && dstFilter->chrV->length > 1);
1201 usesHFilter = (srcFilter->lumH && srcFilter->lumH->length > 1) ||
1202 (srcFilter->chrH && srcFilter->chrH->length > 1) ||
1203 (dstFilter->lumH && dstFilter->lumH->length > 1) ||
1204 (dstFilter->chrH && dstFilter->chrH->length > 1);
1206 av_pix_fmt_get_chroma_sub_sample(srcFormat, &c->chrSrcHSubSample, &c->chrSrcVSubSample);
1207 av_pix_fmt_get_chroma_sub_sample(dstFormat, &c->chrDstHSubSample, &c->chrDstVSubSample);
1209 if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) {
1211 av_log(c, AV_LOG_DEBUG, "Forcing full internal H chroma due to odd output size\n");
1212 flags |= SWS_FULL_CHR_H_INT;
1216 if ( c->chrSrcHSubSample == 0
1217 && c->chrSrcVSubSample == 0
1218 && c->dither != SWS_DITHER_BAYER //SWS_FULL_CHR_H_INT is currently not supported with SWS_DITHER_BAYER
1219 && !(c->flags & SWS_FAST_BILINEAR)
1221 av_log(c, AV_LOG_DEBUG, "Forcing full internal H chroma due to input having non subsampled chroma\n");
1222 flags |= SWS_FULL_CHR_H_INT;
1227 if (c->dither == SWS_DITHER_AUTO) {
1228 if (flags & SWS_ERROR_DIFFUSION)
1229 c->dither = SWS_DITHER_ED;
1232 if(dstFormat == AV_PIX_FMT_BGR4_BYTE ||
1233 dstFormat == AV_PIX_FMT_RGB4_BYTE ||
1234 dstFormat == AV_PIX_FMT_BGR8 ||
1235 dstFormat == AV_PIX_FMT_RGB8) {
1236 if (c->dither == SWS_DITHER_AUTO)
1237 c->dither = (flags & SWS_FULL_CHR_H_INT) ? SWS_DITHER_ED : SWS_DITHER_BAYER;
1238 if (!(flags & SWS_FULL_CHR_H_INT)) {
1239 if (c->dither == SWS_DITHER_ED) {
1240 av_log(c, AV_LOG_DEBUG,
1241 "Desired dithering only supported in full chroma interpolation for destination format '%s'\n",
1242 av_get_pix_fmt_name(dstFormat));
1243 flags |= SWS_FULL_CHR_H_INT;
1247 if (flags & SWS_FULL_CHR_H_INT) {
1248 if (c->dither == SWS_DITHER_BAYER) {
1249 av_log(c, AV_LOG_DEBUG,
1250 "Ordered dither is not supported in full chroma interpolation for destination format '%s'\n",
1251 av_get_pix_fmt_name(dstFormat));
1252 c->dither = SWS_DITHER_ED;
1256 if (isPlanarRGB(dstFormat)) {
1257 if (!(flags & SWS_FULL_CHR_H_INT)) {
1258 av_log(c, AV_LOG_DEBUG,
1259 "%s output is not supported with half chroma resolution, switching to full\n",
1260 av_get_pix_fmt_name(dstFormat));
1261 flags |= SWS_FULL_CHR_H_INT;
1266 /* reuse chroma for 2 pixels RGB/BGR unless user wants full
1267 * chroma interpolation */
1268 if (flags & SWS_FULL_CHR_H_INT &&
1269 isAnyRGB(dstFormat) &&
1270 !isPlanarRGB(dstFormat) &&
1271 dstFormat != AV_PIX_FMT_RGBA &&
1272 dstFormat != AV_PIX_FMT_ARGB &&
1273 dstFormat != AV_PIX_FMT_BGRA &&
1274 dstFormat != AV_PIX_FMT_ABGR &&
1275 dstFormat != AV_PIX_FMT_RGB24 &&
1276 dstFormat != AV_PIX_FMT_BGR24 &&
1277 dstFormat != AV_PIX_FMT_BGR4_BYTE &&
1278 dstFormat != AV_PIX_FMT_RGB4_BYTE &&
1279 dstFormat != AV_PIX_FMT_BGR8 &&
1280 dstFormat != AV_PIX_FMT_RGB8
1282 av_log(c, AV_LOG_WARNING,
1283 "full chroma interpolation for destination format '%s' not yet implemented\n",
1284 av_get_pix_fmt_name(dstFormat));
1285 flags &= ~SWS_FULL_CHR_H_INT;
1288 if (isAnyRGB(dstFormat) && !(flags & SWS_FULL_CHR_H_INT))
1289 c->chrDstHSubSample = 1;
1291 // drop some chroma lines if the user wants it
1292 c->vChrDrop = (flags & SWS_SRC_V_CHR_DROP_MASK) >>
1293 SWS_SRC_V_CHR_DROP_SHIFT;
1294 c->chrSrcVSubSample += c->vChrDrop;
1296 /* drop every other pixel for chroma calculation unless user
1297 * wants full chroma */
1298 if (isAnyRGB(srcFormat) && !(flags & SWS_FULL_CHR_H_INP) &&
1299 srcFormat != AV_PIX_FMT_RGB8 && srcFormat != AV_PIX_FMT_BGR8 &&
1300 srcFormat != AV_PIX_FMT_RGB4 && srcFormat != AV_PIX_FMT_BGR4 &&
1301 srcFormat != AV_PIX_FMT_RGB4_BYTE && srcFormat != AV_PIX_FMT_BGR4_BYTE &&
1302 srcFormat != AV_PIX_FMT_GBRP9BE && srcFormat != AV_PIX_FMT_GBRP9LE &&
1303 srcFormat != AV_PIX_FMT_GBRP10BE && srcFormat != AV_PIX_FMT_GBRP10LE &&
1304 srcFormat != AV_PIX_FMT_GBRP12BE && srcFormat != AV_PIX_FMT_GBRP12LE &&
1305 srcFormat != AV_PIX_FMT_GBRP14BE && srcFormat != AV_PIX_FMT_GBRP14LE &&
1306 srcFormat != AV_PIX_FMT_GBRP16BE && srcFormat != AV_PIX_FMT_GBRP16LE &&
1307 ((dstW >> c->chrDstHSubSample) <= (srcW >> 1) ||
1308 (flags & SWS_FAST_BILINEAR)))
1309 c->chrSrcHSubSample = 1;
1311 // Note the FF_CEIL_RSHIFT is so that we always round toward +inf.
1312 c->chrSrcW = FF_CEIL_RSHIFT(srcW, c->chrSrcHSubSample);
1313 c->chrSrcH = FF_CEIL_RSHIFT(srcH, c->chrSrcVSubSample);
1314 c->chrDstW = FF_CEIL_RSHIFT(dstW, c->chrDstHSubSample);
1315 c->chrDstH = FF_CEIL_RSHIFT(dstH, c->chrDstVSubSample);
1317 FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW*2+78, 16) * 2, fail);
1319 /* unscaled special cases */
1320 if (unscaled && !usesHFilter && !usesVFilter &&
1321 (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
1322 ff_get_unscaled_swscale(c);
1325 if (flags & SWS_PRINT_INFO)
1326 av_log(c, AV_LOG_INFO,
1327 "using unscaled %s -> %s special converter\n",
1328 av_get_pix_fmt_name(srcFormat), av_get_pix_fmt_name(dstFormat));
1333 c->srcBpc = 1 + desc_src->comp[0].depth_minus1;
1336 c->dstBpc = 1 + desc_dst->comp[0].depth_minus1;
1339 if (isAnyRGB(srcFormat) || srcFormat == AV_PIX_FMT_PAL8)
1341 if (c->dstBpc == 16)
1344 if (INLINE_MMXEXT(cpu_flags) && c->srcBpc == 8 && c->dstBpc <= 14) {
1345 c->canMMXEXTBeUsed = dstW >= srcW && (dstW & 31) == 0 &&
1346 c->chrDstW >= c->chrSrcW &&
1348 if (!c->canMMXEXTBeUsed && dstW >= srcW && c->chrDstW >= c->chrSrcW && (srcW & 15) == 0
1350 && (flags & SWS_FAST_BILINEAR)) {
1351 if (flags & SWS_PRINT_INFO)
1352 av_log(c, AV_LOG_INFO,
1353 "output width is not a multiple of 32 -> no MMXEXT scaler\n");
1355 if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat))
1356 c->canMMXEXTBeUsed = 0;
1358 c->canMMXEXTBeUsed = 0;
1360 c->chrXInc = (((int64_t)c->chrSrcW << 16) + (c->chrDstW >> 1)) / c->chrDstW;
1361 c->chrYInc = (((int64_t)c->chrSrcH << 16) + (c->chrDstH >> 1)) / c->chrDstH;
1363 /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src
1364 * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do
1366 * n-2 is the last chrominance sample available.
1367 * This is not perfect, but no one should notice the difference, the more
1368 * correct variant would be like the vertical one, but that would require
1369 * some special code for the first and last pixel */
1370 if (flags & SWS_FAST_BILINEAR) {
1371 if (c->canMMXEXTBeUsed) {
1375 // we don't use the x86 asm scaler if MMX is available
1376 else if (INLINE_MMX(cpu_flags) && c->dstBpc <= 14) {
1377 c->lumXInc = ((int64_t)(srcW - 2) << 16) / (dstW - 2) - 20;
1378 c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;
1382 #define USE_MMAP (HAVE_MMAP && HAVE_MPROTECT && defined MAP_ANONYMOUS)
1384 /* precalculate horizontal scaler filter coefficients */
1386 #if HAVE_MMXEXT_INLINE
1387 // can't downscale !!!
1388 if (c->canMMXEXTBeUsed && (flags & SWS_FAST_BILINEAR)) {
1389 c->lumMmxextFilterCodeSize = init_hscaler_mmxext(dstW, c->lumXInc, NULL,
1391 c->chrMmxextFilterCodeSize = init_hscaler_mmxext(c->chrDstW, c->chrXInc,
1392 NULL, NULL, NULL, 4);
1395 c->lumMmxextFilterCode = mmap(NULL, c->lumMmxextFilterCodeSize,
1396 PROT_READ | PROT_WRITE,
1397 MAP_PRIVATE | MAP_ANONYMOUS,
1399 c->chrMmxextFilterCode = mmap(NULL, c->chrMmxextFilterCodeSize,
1400 PROT_READ | PROT_WRITE,
1401 MAP_PRIVATE | MAP_ANONYMOUS,
1403 #elif HAVE_VIRTUALALLOC
1404 c->lumMmxextFilterCode = VirtualAlloc(NULL,
1405 c->lumMmxextFilterCodeSize,
1407 PAGE_EXECUTE_READWRITE);
1408 c->chrMmxextFilterCode = VirtualAlloc(NULL,
1409 c->chrMmxextFilterCodeSize,
1411 PAGE_EXECUTE_READWRITE);
1413 c->lumMmxextFilterCode = av_malloc(c->lumMmxextFilterCodeSize);
1414 c->chrMmxextFilterCode = av_malloc(c->chrMmxextFilterCodeSize);
1417 #ifdef MAP_ANONYMOUS
1418 if (c->lumMmxextFilterCode == MAP_FAILED || c->chrMmxextFilterCode == MAP_FAILED)
1420 if (!c->lumMmxextFilterCode || !c->chrMmxextFilterCode)
1423 av_log(c, AV_LOG_ERROR, "Failed to allocate MMX2FilterCode\n");
1424 return AVERROR(ENOMEM);
1427 FF_ALLOCZ_OR_GOTO(c, c->hLumFilter, (dstW / 8 + 8) * sizeof(int16_t), fail);
1428 FF_ALLOCZ_OR_GOTO(c, c->hChrFilter, (c->chrDstW / 4 + 8) * sizeof(int16_t), fail);
1429 FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW / 2 / 8 + 8) * sizeof(int32_t), fail);
1430 FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);
1432 init_hscaler_mmxext( dstW, c->lumXInc, c->lumMmxextFilterCode,
1433 c->hLumFilter, (uint32_t*)c->hLumFilterPos, 8);
1434 init_hscaler_mmxext(c->chrDstW, c->chrXInc, c->chrMmxextFilterCode,
1435 c->hChrFilter, (uint32_t*)c->hChrFilterPos, 4);
1438 if ( mprotect(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize, PROT_EXEC | PROT_READ) == -1
1439 || mprotect(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize, PROT_EXEC | PROT_READ) == -1) {
1440 av_log(c, AV_LOG_ERROR, "mprotect failed, cannot use fast bilinear scaler\n");
1445 #endif /* HAVE_MMXEXT_INLINE */
1447 const int filterAlign = X86_MMX(cpu_flags) ? 4 :
1448 PPC_ALTIVEC(cpu_flags) ? 8 : 1;
1450 if (initFilter(&c->hLumFilter, &c->hLumFilterPos,
1451 &c->hLumFilterSize, c->lumXInc,
1452 srcW, dstW, filterAlign, 1 << 14,
1453 (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
1454 cpu_flags, srcFilter->lumH, dstFilter->lumH,
1456 get_local_pos(c, 0, 0, 0),
1457 get_local_pos(c, 0, 0, 0)) < 0)
1459 if (initFilter(&c->hChrFilter, &c->hChrFilterPos,
1460 &c->hChrFilterSize, c->chrXInc,
1461 c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,
1462 (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
1463 cpu_flags, srcFilter->chrH, dstFilter->chrH,
1465 get_local_pos(c, c->chrSrcHSubSample, c->src_h_chr_pos, 0),
1466 get_local_pos(c, c->chrDstHSubSample, c->dst_h_chr_pos, 0)) < 0)
1469 } // initialize horizontal stuff
1471 /* precalculate vertical scaler filter coefficients */
1473 const int filterAlign = X86_MMX(cpu_flags) ? 2 :
1474 PPC_ALTIVEC(cpu_flags) ? 8 : 1;
1476 if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,
1477 c->lumYInc, srcH, dstH, filterAlign, (1 << 12),
1478 (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
1479 cpu_flags, srcFilter->lumV, dstFilter->lumV,
1481 get_local_pos(c, 0, 0, 1),
1482 get_local_pos(c, 0, 0, 1)) < 0)
1484 if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,
1485 c->chrYInc, c->chrSrcH, c->chrDstH,
1486 filterAlign, (1 << 12),
1487 (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
1488 cpu_flags, srcFilter->chrV, dstFilter->chrV,
1490 get_local_pos(c, c->chrSrcVSubSample, c->src_v_chr_pos, 1),
1491 get_local_pos(c, c->chrDstVSubSample, c->dst_v_chr_pos, 1)) < 0)
1496 FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH, fail);
1497 FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);
1499 for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {
1501 short *p = (short *)&c->vYCoeffsBank[i];
1502 for (j = 0; j < 8; j++)
1503 p[j] = c->vLumFilter[i];
1506 for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {
1508 short *p = (short *)&c->vCCoeffsBank[i];
1509 for (j = 0; j < 8; j++)
1510 p[j] = c->vChrFilter[i];
1515 // calculate buffer sizes so that they won't run out while handling these damn slices
1516 c->vLumBufSize = c->vLumFilterSize;
1517 c->vChrBufSize = c->vChrFilterSize;
1518 for (i = 0; i < dstH; i++) {
1519 int chrI = (int64_t)i * c->chrDstH / dstH;
1520 int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,
1521 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)
1522 << c->chrSrcVSubSample));
1524 nextSlice >>= c->chrSrcVSubSample;
1525 nextSlice <<= c->chrSrcVSubSample;
1526 if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)
1527 c->vLumBufSize = nextSlice - c->vLumFilterPos[i];
1528 if (c->vChrFilterPos[chrI] + c->vChrBufSize <
1529 (nextSlice >> c->chrSrcVSubSample))
1530 c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -
1531 c->vChrFilterPos[chrI];
1534 for (i = 0; i < 4; i++)
1535 FF_ALLOCZ_OR_GOTO(c, c->dither_error[i], (c->dstW+2) * sizeof(int), fail);
1537 /* Allocate pixbufs (we use dynamic allocation because otherwise we would
1538 * need to allocate several megabytes to handle all possible cases) */
1539 FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);
1540 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
1541 FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
1542 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
1543 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);
1544 /* Note we need at least one pixel more at the end because of the MMX code
1545 * (just in case someone wants to replace the 4000/8000). */
1546 /* align at 16 bytes for AltiVec */
1547 for (i = 0; i < c->vLumBufSize; i++) {
1548 FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],
1549 dst_stride + 16, fail);
1550 c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];
1552 // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)
1553 c->uv_off = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);
1554 c->uv_offx2 = dst_stride + 16;
1555 for (i = 0; i < c->vChrBufSize; i++) {
1556 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],
1557 dst_stride * 2 + 32, fail);
1558 c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];
1559 c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]
1560 = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;
1562 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
1563 for (i = 0; i < c->vLumBufSize; i++) {
1564 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i + c->vLumBufSize],
1565 dst_stride + 16, fail);
1566 c->alpPixBuf[i] = c->alpPixBuf[i + c->vLumBufSize];
1569 // try to avoid drawing green stuff between the right end and the stride end
1570 for (i = 0; i < c->vChrBufSize; i++)
1571 if(desc_dst->comp[0].depth_minus1 == 15){
1572 av_assert0(c->dstBpc > 14);
1573 for(j=0; j<dst_stride/2+1; j++)
1574 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;
1576 for(j=0; j<dst_stride+1; j++)
1577 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;
1579 av_assert0(c->chrDstH <= dstH);
1581 if (flags & SWS_PRINT_INFO) {
1582 const char *scaler = NULL, *cpucaps;
1584 for (i = 0; i < FF_ARRAY_ELEMS(scale_algorithms); i++) {
1585 if (flags & scale_algorithms[i].flag) {
1586 scaler = scale_algorithms[i].description;
1591 scaler = "ehh flags invalid?!";
1592 av_log(c, AV_LOG_INFO, "%s scaler, from %s to %s%s ",
1594 av_get_pix_fmt_name(srcFormat),
1596 dstFormat == AV_PIX_FMT_BGR555 || dstFormat == AV_PIX_FMT_BGR565 ||
1597 dstFormat == AV_PIX_FMT_RGB444BE || dstFormat == AV_PIX_FMT_RGB444LE ||
1598 dstFormat == AV_PIX_FMT_BGR444BE || dstFormat == AV_PIX_FMT_BGR444LE ?
1603 av_get_pix_fmt_name(dstFormat));
1605 if (INLINE_MMXEXT(cpu_flags))
1607 else if (INLINE_AMD3DNOW(cpu_flags))
1609 else if (INLINE_MMX(cpu_flags))
1611 else if (PPC_ALTIVEC(cpu_flags))
1612 cpucaps = "AltiVec";
1616 av_log(c, AV_LOG_INFO, "using %s\n", cpucaps);
1618 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
1619 av_log(c, AV_LOG_DEBUG,
1620 "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1621 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
1622 av_log(c, AV_LOG_DEBUG,
1623 "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1624 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,
1625 c->chrXInc, c->chrYInc);
1628 c->swscale = ff_getSwsFunc(c);
1630 fail: // FIXME replace things by appropriate error codes
1634 #if FF_API_SWS_GETCONTEXT
1635 SwsContext *sws_getContext(int srcW, int srcH, enum AVPixelFormat srcFormat,
1636 int dstW, int dstH, enum AVPixelFormat dstFormat,
1637 int flags, SwsFilter *srcFilter,
1638 SwsFilter *dstFilter, const double *param)
1642 if (!(c = sws_alloc_context()))
1650 c->srcFormat = srcFormat;
1651 c->dstFormat = dstFormat;
1654 c->param[0] = param[0];
1655 c->param[1] = param[1];
1658 if (sws_init_context(c, srcFilter, dstFilter) < 0) {
1667 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
1668 float lumaSharpen, float chromaSharpen,
1669 float chromaHShift, float chromaVShift,
1672 SwsFilter *filter = av_malloc(sizeof(SwsFilter));
1676 if (lumaGBlur != 0.0) {
1677 filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);
1678 filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);
1680 filter->lumH = sws_getIdentityVec();
1681 filter->lumV = sws_getIdentityVec();
1684 if (chromaGBlur != 0.0) {
1685 filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);
1686 filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);
1688 filter->chrH = sws_getIdentityVec();
1689 filter->chrV = sws_getIdentityVec();
1692 if (chromaSharpen != 0.0) {
1693 SwsVector *id = sws_getIdentityVec();
1694 sws_scaleVec(filter->chrH, -chromaSharpen);
1695 sws_scaleVec(filter->chrV, -chromaSharpen);
1696 sws_addVec(filter->chrH, id);
1697 sws_addVec(filter->chrV, id);
1701 if (lumaSharpen != 0.0) {
1702 SwsVector *id = sws_getIdentityVec();
1703 sws_scaleVec(filter->lumH, -lumaSharpen);
1704 sws_scaleVec(filter->lumV, -lumaSharpen);
1705 sws_addVec(filter->lumH, id);
1706 sws_addVec(filter->lumV, id);
1710 if (chromaHShift != 0.0)
1711 sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));
1713 if (chromaVShift != 0.0)
1714 sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));
1716 sws_normalizeVec(filter->chrH, 1.0);
1717 sws_normalizeVec(filter->chrV, 1.0);
1718 sws_normalizeVec(filter->lumH, 1.0);
1719 sws_normalizeVec(filter->lumV, 1.0);
1722 sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
1724 sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
1729 SwsVector *sws_allocVec(int length)
1733 if(length <= 0 || length > INT_MAX/ sizeof(double))
1736 vec = av_malloc(sizeof(SwsVector));
1739 vec->length = length;
1740 vec->coeff = av_malloc(sizeof(double) * length);
1746 SwsVector *sws_getGaussianVec(double variance, double quality)
1748 const int length = (int)(variance * quality + 0.5) | 1;
1750 double middle = (length - 1) * 0.5;
1753 if(variance < 0 || quality < 0)
1756 vec = sws_allocVec(length);
1761 for (i = 0; i < length; i++) {
1762 double dist = i - middle;
1763 vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /
1764 sqrt(2 * variance * M_PI);
1767 sws_normalizeVec(vec, 1.0);
1772 SwsVector *sws_getConstVec(double c, int length)
1775 SwsVector *vec = sws_allocVec(length);
1780 for (i = 0; i < length; i++)
1786 SwsVector *sws_getIdentityVec(void)
1788 return sws_getConstVec(1.0, 1);
1791 static double sws_dcVec(SwsVector *a)
1796 for (i = 0; i < a->length; i++)
1802 void sws_scaleVec(SwsVector *a, double scalar)
1806 for (i = 0; i < a->length; i++)
1807 a->coeff[i] *= scalar;
1810 void sws_normalizeVec(SwsVector *a, double height)
1812 sws_scaleVec(a, height / sws_dcVec(a));
1815 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
1817 int length = a->length + b->length - 1;
1819 SwsVector *vec = sws_getConstVec(0.0, length);
1824 for (i = 0; i < a->length; i++) {
1825 for (j = 0; j < b->length; j++) {
1826 vec->coeff[i + j] += a->coeff[i] * b->coeff[j];
1833 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
1835 int length = FFMAX(a->length, b->length);
1837 SwsVector *vec = sws_getConstVec(0.0, length);
1842 for (i = 0; i < a->length; i++)
1843 vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
1844 for (i = 0; i < b->length; i++)
1845 vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] += b->coeff[i];
1850 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
1852 int length = FFMAX(a->length, b->length);
1854 SwsVector *vec = sws_getConstVec(0.0, length);
1859 for (i = 0; i < a->length; i++)
1860 vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
1861 for (i = 0; i < b->length; i++)
1862 vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] -= b->coeff[i];
1867 /* shift left / or right if "shift" is negative */
1868 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
1870 int length = a->length + FFABS(shift) * 2;
1872 SwsVector *vec = sws_getConstVec(0.0, length);
1877 for (i = 0; i < a->length; i++) {
1878 vec->coeff[i + (length - 1) / 2 -
1879 (a->length - 1) / 2 - shift] = a->coeff[i];
1885 void sws_shiftVec(SwsVector *a, int shift)
1887 SwsVector *shifted = sws_getShiftedVec(a, shift);
1889 a->coeff = shifted->coeff;
1890 a->length = shifted->length;
1894 void sws_addVec(SwsVector *a, SwsVector *b)
1896 SwsVector *sum = sws_sumVec(a, b);
1898 a->coeff = sum->coeff;
1899 a->length = sum->length;
1903 void sws_subVec(SwsVector *a, SwsVector *b)
1905 SwsVector *diff = sws_diffVec(a, b);
1907 a->coeff = diff->coeff;
1908 a->length = diff->length;
1912 void sws_convVec(SwsVector *a, SwsVector *b)
1914 SwsVector *conv = sws_getConvVec(a, b);
1916 a->coeff = conv->coeff;
1917 a->length = conv->length;
1921 SwsVector *sws_cloneVec(SwsVector *a)
1923 SwsVector *vec = sws_allocVec(a->length);
1928 memcpy(vec->coeff, a->coeff, a->length * sizeof(*a->coeff));
1933 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
1940 for (i = 0; i < a->length; i++)
1941 if (a->coeff[i] > max)
1944 for (i = 0; i < a->length; i++)
1945 if (a->coeff[i] < min)
1950 for (i = 0; i < a->length; i++) {
1951 int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);
1952 av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
1954 av_log(log_ctx, log_level, " ");
1955 av_log(log_ctx, log_level, "|\n");
1959 void sws_freeVec(SwsVector *a)
1963 av_freep(&a->coeff);
1968 void sws_freeFilter(SwsFilter *filter)
1973 sws_freeVec(filter->lumH);
1974 sws_freeVec(filter->lumV);
1975 sws_freeVec(filter->chrH);
1976 sws_freeVec(filter->chrV);
1980 void sws_freeContext(SwsContext *c)
1987 for (i = 0; i < c->vLumBufSize; i++)
1988 av_freep(&c->lumPixBuf[i]);
1989 av_freep(&c->lumPixBuf);
1992 if (c->chrUPixBuf) {
1993 for (i = 0; i < c->vChrBufSize; i++)
1994 av_freep(&c->chrUPixBuf[i]);
1995 av_freep(&c->chrUPixBuf);
1996 av_freep(&c->chrVPixBuf);
1999 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
2000 for (i = 0; i < c->vLumBufSize; i++)
2001 av_freep(&c->alpPixBuf[i]);
2002 av_freep(&c->alpPixBuf);
2005 for (i = 0; i < 4; i++)
2006 av_freep(&c->dither_error[i]);
2008 av_freep(&c->vLumFilter);
2009 av_freep(&c->vChrFilter);
2010 av_freep(&c->hLumFilter);
2011 av_freep(&c->hChrFilter);
2013 av_freep(&c->vYCoeffsBank);
2014 av_freep(&c->vCCoeffsBank);
2017 av_freep(&c->vLumFilterPos);
2018 av_freep(&c->vChrFilterPos);
2019 av_freep(&c->hLumFilterPos);
2020 av_freep(&c->hChrFilterPos);
2024 if (c->lumMmxextFilterCode)
2025 munmap(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize);
2026 if (c->chrMmxextFilterCode)
2027 munmap(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize);
2028 #elif HAVE_VIRTUALALLOC
2029 if (c->lumMmxextFilterCode)
2030 VirtualFree(c->lumMmxextFilterCode, 0, MEM_RELEASE);
2031 if (c->chrMmxextFilterCode)
2032 VirtualFree(c->chrMmxextFilterCode, 0, MEM_RELEASE);
2034 av_free(c->lumMmxextFilterCode);
2035 av_free(c->chrMmxextFilterCode);
2037 c->lumMmxextFilterCode = NULL;
2038 c->chrMmxextFilterCode = NULL;
2039 #endif /* HAVE_MMX_INLINE */
2041 av_freep(&c->yuvTable);
2042 av_freep(&c->formatConvBuffer);
2047 struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,
2048 int srcH, enum AVPixelFormat srcFormat,
2050 enum AVPixelFormat dstFormat, int flags,
2051 SwsFilter *srcFilter,
2052 SwsFilter *dstFilter,
2053 const double *param)
2055 static const double default_param[2] = { SWS_PARAM_DEFAULT,
2056 SWS_PARAM_DEFAULT };
2059 param = default_param;
2062 (context->srcW != srcW ||
2063 context->srcH != srcH ||
2064 context->srcFormat != srcFormat ||
2065 context->dstW != dstW ||
2066 context->dstH != dstH ||
2067 context->dstFormat != dstFormat ||
2068 context->flags != flags ||
2069 context->param[0] != param[0] ||
2070 context->param[1] != param[1])) {
2071 sws_freeContext(context);
2076 if (!(context = sws_alloc_context()))
2078 context->srcW = srcW;
2079 context->srcH = srcH;
2080 context->srcFormat = srcFormat;
2081 context->dstW = dstW;
2082 context->dstH = dstH;
2083 context->dstFormat = dstFormat;
2084 context->flags = flags;
2085 context->param[0] = param[0];
2086 context->param[1] = param[1];
2087 if (sws_init_context(context, srcFilter, dstFilter) < 0) {
2088 sws_freeContext(context);