2 * Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
4 * This file is part of FFmpeg.
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 #define _SVID_SOURCE // needed for MAP_ANONYMOUS
24 #define _DARWIN_C_SOURCE // needed for MAP_ANON
31 #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
32 #define MAP_ANONYMOUS MAP_ANON
36 #define WIN32_LEAN_AND_MEAN
40 #include "libavutil/attributes.h"
41 #include "libavutil/avassert.h"
42 #include "libavutil/avutil.h"
43 #include "libavutil/bswap.h"
44 #include "libavutil/cpu.h"
45 #include "libavutil/intreadwrite.h"
46 #include "libavutil/mathematics.h"
47 #include "libavutil/opt.h"
48 #include "libavutil/pixdesc.h"
49 #include "libavutil/x86/asm.h"
50 #include "libavutil/x86/cpu.h"
53 #include "swscale_internal.h"
55 unsigned swscale_version(void)
57 av_assert0(LIBSWSCALE_VERSION_MICRO >= 100);
58 return LIBSWSCALE_VERSION_INT;
61 const char *swscale_configuration(void)
63 return FFMPEG_CONFIGURATION;
66 const char *swscale_license(void)
68 #define LICENSE_PREFIX "libswscale license: "
69 return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
72 #define RET 0xC3 // near return opcode for x86
74 typedef struct FormatEntry {
75 int is_supported_in, is_supported_out;
78 static const FormatEntry format_entries[AV_PIX_FMT_NB] = {
79 [AV_PIX_FMT_YUV420P] = { 1, 1 },
80 [AV_PIX_FMT_YUYV422] = { 1, 1 },
81 [AV_PIX_FMT_RGB24] = { 1, 1 },
82 [AV_PIX_FMT_BGR24] = { 1, 1 },
83 [AV_PIX_FMT_YUV422P] = { 1, 1 },
84 [AV_PIX_FMT_YUV444P] = { 1, 1 },
85 [AV_PIX_FMT_YUV410P] = { 1, 1 },
86 [AV_PIX_FMT_YUV411P] = { 1, 1 },
87 [AV_PIX_FMT_GRAY8] = { 1, 1 },
88 [AV_PIX_FMT_MONOWHITE] = { 1, 1 },
89 [AV_PIX_FMT_MONOBLACK] = { 1, 1 },
90 [AV_PIX_FMT_PAL8] = { 1, 0 },
91 [AV_PIX_FMT_YUVJ420P] = { 1, 1 },
92 [AV_PIX_FMT_YUVJ422P] = { 1, 1 },
93 [AV_PIX_FMT_YUVJ444P] = { 1, 1 },
94 [AV_PIX_FMT_UYVY422] = { 1, 1 },
95 [AV_PIX_FMT_UYYVYY411] = { 0, 0 },
96 [AV_PIX_FMT_BGR8] = { 1, 1 },
97 [AV_PIX_FMT_BGR4] = { 0, 1 },
98 [AV_PIX_FMT_BGR4_BYTE] = { 1, 1 },
99 [AV_PIX_FMT_RGB8] = { 1, 1 },
100 [AV_PIX_FMT_RGB4] = { 0, 1 },
101 [AV_PIX_FMT_RGB4_BYTE] = { 1, 1 },
102 [AV_PIX_FMT_NV12] = { 1, 1 },
103 [AV_PIX_FMT_NV21] = { 1, 1 },
104 [AV_PIX_FMT_ARGB] = { 1, 1 },
105 [AV_PIX_FMT_RGBA] = { 1, 1 },
106 [AV_PIX_FMT_ABGR] = { 1, 1 },
107 [AV_PIX_FMT_BGRA] = { 1, 1 },
108 [AV_PIX_FMT_0RGB] = { 1, 1 },
109 [AV_PIX_FMT_RGB0] = { 1, 1 },
110 [AV_PIX_FMT_0BGR] = { 1, 1 },
111 [AV_PIX_FMT_BGR0] = { 1, 1 },
112 [AV_PIX_FMT_GRAY16BE] = { 1, 1 },
113 [AV_PIX_FMT_GRAY16LE] = { 1, 1 },
114 [AV_PIX_FMT_YUV440P] = { 1, 1 },
115 [AV_PIX_FMT_YUVJ440P] = { 1, 1 },
116 [AV_PIX_FMT_YUVA420P] = { 1, 1 },
117 [AV_PIX_FMT_YUVA422P] = { 1, 1 },
118 [AV_PIX_FMT_YUVA444P] = { 1, 1 },
119 [AV_PIX_FMT_YUVA420P9BE] = { 1, 1 },
120 [AV_PIX_FMT_YUVA420P9LE] = { 1, 1 },
121 [AV_PIX_FMT_YUVA422P9BE] = { 1, 1 },
122 [AV_PIX_FMT_YUVA422P9LE] = { 1, 1 },
123 [AV_PIX_FMT_YUVA444P9BE] = { 1, 1 },
124 [AV_PIX_FMT_YUVA444P9LE] = { 1, 1 },
125 [AV_PIX_FMT_YUVA420P10BE]= { 1, 1 },
126 [AV_PIX_FMT_YUVA420P10LE]= { 1, 1 },
127 [AV_PIX_FMT_YUVA422P10BE]= { 1, 1 },
128 [AV_PIX_FMT_YUVA422P10LE]= { 1, 1 },
129 [AV_PIX_FMT_YUVA444P10BE]= { 1, 1 },
130 [AV_PIX_FMT_YUVA444P10LE]= { 1, 1 },
131 [AV_PIX_FMT_YUVA420P16BE]= { 1, 1 },
132 [AV_PIX_FMT_YUVA420P16LE]= { 1, 1 },
133 [AV_PIX_FMT_YUVA422P16BE]= { 1, 1 },
134 [AV_PIX_FMT_YUVA422P16LE]= { 1, 1 },
135 [AV_PIX_FMT_YUVA444P16BE]= { 1, 1 },
136 [AV_PIX_FMT_YUVA444P16LE]= { 1, 1 },
137 [AV_PIX_FMT_RGB48BE] = { 1, 1 },
138 [AV_PIX_FMT_RGB48LE] = { 1, 1 },
139 [AV_PIX_FMT_RGBA64BE] = { 1, 0 },
140 [AV_PIX_FMT_RGBA64LE] = { 1, 0 },
141 [AV_PIX_FMT_RGB565BE] = { 1, 1 },
142 [AV_PIX_FMT_RGB565LE] = { 1, 1 },
143 [AV_PIX_FMT_RGB555BE] = { 1, 1 },
144 [AV_PIX_FMT_RGB555LE] = { 1, 1 },
145 [AV_PIX_FMT_BGR565BE] = { 1, 1 },
146 [AV_PIX_FMT_BGR565LE] = { 1, 1 },
147 [AV_PIX_FMT_BGR555BE] = { 1, 1 },
148 [AV_PIX_FMT_BGR555LE] = { 1, 1 },
149 [AV_PIX_FMT_YUV420P16LE] = { 1, 1 },
150 [AV_PIX_FMT_YUV420P16BE] = { 1, 1 },
151 [AV_PIX_FMT_YUV422P16LE] = { 1, 1 },
152 [AV_PIX_FMT_YUV422P16BE] = { 1, 1 },
153 [AV_PIX_FMT_YUV444P16LE] = { 1, 1 },
154 [AV_PIX_FMT_YUV444P16BE] = { 1, 1 },
155 [AV_PIX_FMT_RGB444LE] = { 1, 1 },
156 [AV_PIX_FMT_RGB444BE] = { 1, 1 },
157 [AV_PIX_FMT_BGR444LE] = { 1, 1 },
158 [AV_PIX_FMT_BGR444BE] = { 1, 1 },
159 [AV_PIX_FMT_Y400A] = { 1, 0 },
160 [AV_PIX_FMT_BGR48BE] = { 1, 1 },
161 [AV_PIX_FMT_BGR48LE] = { 1, 1 },
162 [AV_PIX_FMT_BGRA64BE] = { 0, 0 },
163 [AV_PIX_FMT_BGRA64LE] = { 0, 0 },
164 [AV_PIX_FMT_YUV420P9BE] = { 1, 1 },
165 [AV_PIX_FMT_YUV420P9LE] = { 1, 1 },
166 [AV_PIX_FMT_YUV420P10BE] = { 1, 1 },
167 [AV_PIX_FMT_YUV420P10LE] = { 1, 1 },
168 [AV_PIX_FMT_YUV420P12BE] = { 1, 1 },
169 [AV_PIX_FMT_YUV420P12LE] = { 1, 1 },
170 [AV_PIX_FMT_YUV420P14BE] = { 1, 1 },
171 [AV_PIX_FMT_YUV420P14LE] = { 1, 1 },
172 [AV_PIX_FMT_YUV422P9BE] = { 1, 1 },
173 [AV_PIX_FMT_YUV422P9LE] = { 1, 1 },
174 [AV_PIX_FMT_YUV422P10BE] = { 1, 1 },
175 [AV_PIX_FMT_YUV422P10LE] = { 1, 1 },
176 [AV_PIX_FMT_YUV422P12BE] = { 1, 1 },
177 [AV_PIX_FMT_YUV422P12LE] = { 1, 1 },
178 [AV_PIX_FMT_YUV422P14BE] = { 1, 1 },
179 [AV_PIX_FMT_YUV422P14LE] = { 1, 1 },
180 [AV_PIX_FMT_YUV444P9BE] = { 1, 1 },
181 [AV_PIX_FMT_YUV444P9LE] = { 1, 1 },
182 [AV_PIX_FMT_YUV444P10BE] = { 1, 1 },
183 [AV_PIX_FMT_YUV444P10LE] = { 1, 1 },
184 [AV_PIX_FMT_YUV444P12BE] = { 1, 1 },
185 [AV_PIX_FMT_YUV444P12LE] = { 1, 1 },
186 [AV_PIX_FMT_YUV444P14BE] = { 1, 1 },
187 [AV_PIX_FMT_YUV444P14LE] = { 1, 1 },
188 [AV_PIX_FMT_GBRP] = { 1, 0 },
189 [AV_PIX_FMT_GBRP9LE] = { 1, 0 },
190 [AV_PIX_FMT_GBRP9BE] = { 1, 0 },
191 [AV_PIX_FMT_GBRP10LE] = { 1, 0 },
192 [AV_PIX_FMT_GBRP10BE] = { 1, 0 },
193 [AV_PIX_FMT_GBRP12LE] = { 1, 0 },
194 [AV_PIX_FMT_GBRP12BE] = { 1, 0 },
195 [AV_PIX_FMT_GBRP14LE] = { 1, 0 },
196 [AV_PIX_FMT_GBRP14BE] = { 1, 0 },
197 [AV_PIX_FMT_GBRP16LE] = { 1, 0 },
198 [AV_PIX_FMT_GBRP16BE] = { 1, 0 },
201 int sws_isSupportedInput(enum AVPixelFormat pix_fmt)
203 return (unsigned)pix_fmt < AV_PIX_FMT_NB ?
204 format_entries[pix_fmt].is_supported_in : 0;
207 int sws_isSupportedOutput(enum AVPixelFormat pix_fmt)
209 return (unsigned)pix_fmt < AV_PIX_FMT_NB ?
210 format_entries[pix_fmt].is_supported_out : 0;
213 extern const int32_t ff_yuv2rgb_coeffs[8][4];
215 #if FF_API_SWS_FORMAT_NAME
216 const char *sws_format_name(enum AVPixelFormat format)
218 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(format);
222 return "Unknown format";
226 static double getSplineCoeff(double a, double b, double c, double d,
230 return ((d * dist + c) * dist + b) * dist + a;
232 return getSplineCoeff(0.0,
233 b + 2.0 * c + 3.0 * d,
235 -b - 3.0 * c - 6.0 * d,
239 static int initFilter(int16_t **outFilter, int32_t **filterPos,
240 int *outFilterSize, int xInc, int srcW, int dstW,
241 int filterAlign, int one, int flags, int cpu_flags,
242 SwsVector *srcFilter, SwsVector *dstFilter,
249 int64_t *filter = NULL;
250 int64_t *filter2 = NULL;
251 const int64_t fone = 1LL << (54 - FFMIN(av_log2(srcW/dstW), 8));
254 emms_c(); // FIXME should not be required but IS (even for non-MMX versions)
256 // NOTE: the +3 is for the MMX(+1) / SSE(+3) scaler which reads over the end
257 FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW + 3) * sizeof(**filterPos), fail);
259 if (FFABS(xInc - 0x10000) < 10) { // unscaled
262 FF_ALLOCZ_OR_GOTO(NULL, filter,
263 dstW * sizeof(*filter) * filterSize, fail);
265 for (i = 0; i < dstW; i++) {
266 filter[i * filterSize] = fone;
269 } else if (flags & SWS_POINT) { // lame looking point sampling mode
273 FF_ALLOC_OR_GOTO(NULL, filter,
274 dstW * sizeof(*filter) * filterSize, fail);
276 xDstInSrc = xInc / 2 - 0x8000;
277 for (i = 0; i < dstW; i++) {
278 int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;
280 (*filterPos)[i] = xx;
284 } else if ((xInc <= (1 << 16) && (flags & SWS_AREA)) ||
285 (flags & SWS_FAST_BILINEAR)) { // bilinear upscale
289 FF_ALLOC_OR_GOTO(NULL, filter,
290 dstW * sizeof(*filter) * filterSize, fail);
292 xDstInSrc = xInc / 2 - 0x8000;
293 for (i = 0; i < dstW; i++) {
294 int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;
297 (*filterPos)[i] = xx;
298 // bilinear upscale / linear interpolate / area averaging
299 for (j = 0; j < filterSize; j++) {
300 int64_t coeff= fone - FFABS(((int64_t)xx<<16) - xDstInSrc)*(fone>>16);
303 filter[i * filterSize + j] = coeff;
312 if (flags & SWS_BICUBIC)
314 else if (flags & SWS_X)
316 else if (flags & SWS_AREA)
317 sizeFactor = 1; // downscale only, for upscale it is bilinear
318 else if (flags & SWS_GAUSS)
319 sizeFactor = 8; // infinite ;)
320 else if (flags & SWS_LANCZOS)
321 sizeFactor = param[0] != SWS_PARAM_DEFAULT ? ceil(2 * param[0]) : 6;
322 else if (flags & SWS_SINC)
323 sizeFactor = 20; // infinite ;)
324 else if (flags & SWS_SPLINE)
325 sizeFactor = 20; // infinite ;)
326 else if (flags & SWS_BILINEAR)
333 filterSize = 1 + sizeFactor; // upscale
335 filterSize = 1 + (sizeFactor * srcW + dstW - 1) / dstW;
337 filterSize = FFMIN(filterSize, srcW - 2);
338 filterSize = FFMAX(filterSize, 1);
340 FF_ALLOC_OR_GOTO(NULL, filter,
341 dstW * sizeof(*filter) * filterSize, fail);
343 xDstInSrc = xInc - 0x10000;
344 for (i = 0; i < dstW; i++) {
345 int xx = (xDstInSrc - ((filterSize - 2) << 16)) / (1 << 17);
347 (*filterPos)[i] = xx;
348 for (j = 0; j < filterSize; j++) {
349 int64_t d = (FFABS(((int64_t)xx << 17) - xDstInSrc)) << 13;
355 floatd = d * (1.0 / (1 << 30));
357 if (flags & SWS_BICUBIC) {
358 int64_t B = (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1 << 24);
359 int64_t C = (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1 << 24);
361 if (d >= 1LL << 31) {
364 int64_t dd = (d * d) >> 30;
365 int64_t ddd = (dd * d) >> 30;
368 coeff = (12 * (1 << 24) - 9 * B - 6 * C) * ddd +
369 (-18 * (1 << 24) + 12 * B + 6 * C) * dd +
370 (6 * (1 << 24) - 2 * B) * (1 << 30);
372 coeff = (-B - 6 * C) * ddd +
373 (6 * B + 30 * C) * dd +
374 (-12 * B - 48 * C) * d +
375 (8 * B + 24 * C) * (1 << 30);
377 coeff /= (1LL<<54)/fone;
380 else if (flags & SWS_X) {
381 double p = param ? param * 0.01 : 0.3;
382 coeff = d ? sin(d * M_PI) / (d * M_PI) : 1.0;
383 coeff *= pow(2.0, -p * d * d);
386 else if (flags & SWS_X) {
387 double A = param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
391 c = cos(floatd * M_PI);
398 coeff = (c * 0.5 + 0.5) * fone;
399 } else if (flags & SWS_AREA) {
400 int64_t d2 = d - (1 << 29);
401 if (d2 * xInc < -(1LL << (29 + 16)))
402 coeff = 1.0 * (1LL << (30 + 16));
403 else if (d2 * xInc < (1LL << (29 + 16)))
404 coeff = -d2 * xInc + (1LL << (29 + 16));
407 coeff *= fone >> (30 + 16);
408 } else if (flags & SWS_GAUSS) {
409 double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
410 coeff = (pow(2.0, -p * floatd * floatd)) * fone;
411 } else if (flags & SWS_SINC) {
412 coeff = (d ? sin(floatd * M_PI) / (floatd * M_PI) : 1.0) * fone;
413 } else if (flags & SWS_LANCZOS) {
414 double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
415 coeff = (d ? sin(floatd * M_PI) * sin(floatd * M_PI / p) /
416 (floatd * floatd * M_PI * M_PI / p) : 1.0) * fone;
419 } else if (flags & SWS_BILINEAR) {
420 coeff = (1 << 30) - d;
424 } else if (flags & SWS_SPLINE) {
425 double p = -2.196152422706632;
426 coeff = getSplineCoeff(1.0, 0.0, p, -p - 1.0, floatd) * fone;
431 filter[i * filterSize + j] = coeff;
434 xDstInSrc += 2 * xInc;
438 /* apply src & dst Filter to filter -> filter2
441 av_assert0(filterSize > 0);
442 filter2Size = filterSize;
444 filter2Size += srcFilter->length - 1;
446 filter2Size += dstFilter->length - 1;
447 av_assert0(filter2Size > 0);
448 FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size * dstW * sizeof(*filter2), fail);
450 for (i = 0; i < dstW; i++) {
454 for (k = 0; k < srcFilter->length; k++) {
455 for (j = 0; j < filterSize; j++)
456 filter2[i * filter2Size + k + j] +=
457 srcFilter->coeff[k] * filter[i * filterSize + j];
460 for (j = 0; j < filterSize; j++)
461 filter2[i * filter2Size + j] = filter[i * filterSize + j];
465 (*filterPos)[i] += (filterSize - 1) / 2 - (filter2Size - 1) / 2;
469 /* try to reduce the filter-size (step1 find size and shift left) */
470 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
472 for (i = dstW - 1; i >= 0; i--) {
473 int min = filter2Size;
475 int64_t cutOff = 0.0;
477 /* get rid of near zero elements on the left by shifting left */
478 for (j = 0; j < filter2Size; j++) {
480 cutOff += FFABS(filter2[i * filter2Size]);
482 if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)
485 /* preserve monotonicity because the core can't handle the
486 * filter otherwise */
487 if (i < dstW - 1 && (*filterPos)[i] >= (*filterPos)[i + 1])
490 // move filter coefficients left
491 for (k = 1; k < filter2Size; k++)
492 filter2[i * filter2Size + k - 1] = filter2[i * filter2Size + k];
493 filter2[i * filter2Size + k - 1] = 0;
498 /* count near zeros on the right */
499 for (j = filter2Size - 1; j > 0; j--) {
500 cutOff += FFABS(filter2[i * filter2Size + j]);
502 if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)
507 if (min > minFilterSize)
511 if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) {
512 // we can handle the special case 4, so we don't want to go the full 8
513 if (minFilterSize < 5)
516 /* We really don't want to waste our time doing useless computation, so
517 * fall back on the scalar C code for very small filters.
518 * Vectorizing is worth it only if you have a decent-sized vector. */
519 if (minFilterSize < 3)
523 if (INLINE_MMX(cpu_flags)) {
524 // special case for unscaled vertical filtering
525 if (minFilterSize == 1 && filterAlign == 2)
529 av_assert0(minFilterSize > 0);
530 filterSize = (minFilterSize + (filterAlign - 1)) & (~(filterAlign - 1));
531 av_assert0(filterSize > 0);
532 filter = av_malloc(filterSize * dstW * sizeof(*filter));
533 if (filterSize >= MAX_FILTER_SIZE * 16 /
534 ((flags & SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter) {
535 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);
538 *outFilterSize = filterSize;
540 if (flags & SWS_PRINT_INFO)
541 av_log(NULL, AV_LOG_VERBOSE,
542 "SwScaler: reducing / aligning filtersize %d -> %d\n",
543 filter2Size, filterSize);
544 /* try to reduce the filter-size (step2 reduce it) */
545 for (i = 0; i < dstW; i++) {
548 for (j = 0; j < filterSize; j++) {
549 if (j >= filter2Size)
550 filter[i * filterSize + j] = 0;
552 filter[i * filterSize + j] = filter2[i * filter2Size + j];
553 if ((flags & SWS_BITEXACT) && j >= minFilterSize)
554 filter[i * filterSize + j] = 0;
558 // FIXME try to align filterPos if possible
561 for (i = 0; i < dstW; i++) {
563 if ((*filterPos)[i] < 0) {
564 // move filter coefficients left to compensate for filterPos
565 for (j = 1; j < filterSize; j++) {
566 int left = FFMAX(j + (*filterPos)[i], 0);
567 filter[i * filterSize + left] += filter[i * filterSize + j];
568 filter[i * filterSize + j] = 0;
573 if ((*filterPos)[i] + filterSize > srcW) {
574 int shift = (*filterPos)[i] + filterSize - srcW;
575 // move filter coefficients right to compensate for filterPos
576 for (j = filterSize - 2; j >= 0; j--) {
577 int right = FFMIN(j + shift, filterSize - 1);
578 filter[i * filterSize + right] += filter[i * filterSize + j];
579 filter[i * filterSize + j] = 0;
581 (*filterPos)[i]= srcW - filterSize;
585 // Note the +1 is for the MMX scaler which reads over the end
586 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
587 FF_ALLOCZ_OR_GOTO(NULL, *outFilter,
588 *outFilterSize * (dstW + 3) * sizeof(int16_t), fail);
590 /* normalize & store in outFilter */
591 for (i = 0; i < dstW; i++) {
596 for (j = 0; j < filterSize; j++) {
597 sum += filter[i * filterSize + j];
599 sum = (sum + one / 2) / one;
600 for (j = 0; j < *outFilterSize; j++) {
601 int64_t v = filter[i * filterSize + j] + error;
602 int intV = ROUNDED_DIV(v, sum);
603 (*outFilter)[i * (*outFilterSize) + j] = intV;
604 error = v - intV * sum;
608 (*filterPos)[dstW + 0] =
609 (*filterPos)[dstW + 1] =
610 (*filterPos)[dstW + 2] = (*filterPos)[dstW - 1]; /* the MMX/SSE scaler will
611 * read over the end */
612 for (i = 0; i < *outFilterSize; i++) {
613 int k = (dstW - 1) * (*outFilterSize) + i;
614 (*outFilter)[k + 1 * (*outFilterSize)] =
615 (*outFilter)[k + 2 * (*outFilterSize)] =
616 (*outFilter)[k + 3 * (*outFilterSize)] = (*outFilter)[k];
623 av_log(NULL, AV_LOG_ERROR, "sws: initFilter failed\n");
629 #if HAVE_MMXEXT_INLINE
630 static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode,
631 int16_t *filter, int32_t *filterPos, int numSplits)
634 x86_reg imm8OfPShufW1A;
635 x86_reg imm8OfPShufW2A;
636 x86_reg fragmentLengthA;
638 x86_reg imm8OfPShufW1B;
639 x86_reg imm8OfPShufW2B;
640 x86_reg fragmentLengthB;
645 // create an optimized horizontal scaling routine
646 /* This scaler is made of runtime-generated MMX2 code using specially tuned
647 * pshufw instructions. For every four output pixels, if four input pixels
648 * are enough for the fast bilinear scaling, then a chunk of fragmentB is
649 * used. If five input pixels are needed, then a chunk of fragmentA is used.
658 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
659 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
660 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
661 "punpcklbw %%mm7, %%mm1 \n\t"
662 "punpcklbw %%mm7, %%mm0 \n\t"
663 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
665 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
667 "psubw %%mm1, %%mm0 \n\t"
668 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
669 "pmullw %%mm3, %%mm0 \n\t"
670 "psllw $7, %%mm1 \n\t"
671 "paddw %%mm1, %%mm0 \n\t"
673 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
675 "add $8, %%"REG_a" \n\t"
679 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
680 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
681 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
686 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
690 : "=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
691 "=r" (fragmentLengthA)
698 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
699 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
700 "punpcklbw %%mm7, %%mm0 \n\t"
701 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
703 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
705 "psubw %%mm1, %%mm0 \n\t"
706 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
707 "pmullw %%mm3, %%mm0 \n\t"
708 "psllw $7, %%mm1 \n\t"
709 "paddw %%mm1, %%mm0 \n\t"
711 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
713 "add $8, %%"REG_a" \n\t"
717 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
718 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
719 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
724 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
728 : "=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
729 "=r" (fragmentLengthB)
732 xpos = 0; // lumXInc/2 - 0x8000; // difference between pixel centers
735 for (i = 0; i < dstW / numSplits; i++) {
740 int b = ((xpos + xInc) >> 16) - xx;
741 int c = ((xpos + xInc * 2) >> 16) - xx;
742 int d = ((xpos + xInc * 3) >> 16) - xx;
743 int inc = (d + 1 < 4);
744 uint8_t *fragment = (d + 1 < 4) ? fragmentB : fragmentA;
745 x86_reg imm8OfPShufW1 = (d + 1 < 4) ? imm8OfPShufW1B : imm8OfPShufW1A;
746 x86_reg imm8OfPShufW2 = (d + 1 < 4) ? imm8OfPShufW2B : imm8OfPShufW2A;
747 x86_reg fragmentLength = (d + 1 < 4) ? fragmentLengthB : fragmentLengthA;
748 int maxShift = 3 - (d + inc);
752 filter[i] = ((xpos & 0xFFFF) ^ 0xFFFF) >> 9;
753 filter[i + 1] = (((xpos + xInc) & 0xFFFF) ^ 0xFFFF) >> 9;
754 filter[i + 2] = (((xpos + xInc * 2) & 0xFFFF) ^ 0xFFFF) >> 9;
755 filter[i + 3] = (((xpos + xInc * 3) & 0xFFFF) ^ 0xFFFF) >> 9;
756 filterPos[i / 2] = xx;
758 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
760 filterCode[fragmentPos + imm8OfPShufW1] = (a + inc) |
764 filterCode[fragmentPos + imm8OfPShufW2] = a | (b << 2) |
768 if (i + 4 - inc >= dstW)
769 shift = maxShift; // avoid overread
770 else if ((filterPos[i / 2] & 3) <= maxShift)
771 shift = filterPos[i / 2] & 3; // align
773 if (shift && i >= shift) {
774 filterCode[fragmentPos + imm8OfPShufW1] += 0x55 * shift;
775 filterCode[fragmentPos + imm8OfPShufW2] += 0x55 * shift;
776 filterPos[i / 2] -= shift;
780 fragmentPos += fragmentLength;
783 filterCode[fragmentPos] = RET;
788 filterPos[((i / 2) + 1) & (~1)] = xpos >> 16; // needed to jump to the next part
790 return fragmentPos + 1;
792 #endif /* HAVE_MMXEXT_INLINE */
794 static void getSubSampleFactors(int *h, int *v, enum AVPixelFormat format)
796 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(format);
797 *h = desc->log2_chroma_w;
798 *v = desc->log2_chroma_h;
801 int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],
802 int srcRange, const int table[4], int dstRange,
803 int brightness, int contrast, int saturation)
805 const AVPixFmtDescriptor *desc_dst = av_pix_fmt_desc_get(c->dstFormat);
806 const AVPixFmtDescriptor *desc_src = av_pix_fmt_desc_get(c->srcFormat);
807 memcpy(c->srcColorspaceTable, inv_table, sizeof(int) * 4);
808 memcpy(c->dstColorspaceTable, table, sizeof(int) * 4);
810 if(!isYUV(c->dstFormat) && !isGray(c->dstFormat))
812 if(!isYUV(c->srcFormat) && !isGray(c->srcFormat))
815 c->brightness = brightness;
816 c->contrast = contrast;
817 c->saturation = saturation;
818 c->srcRange = srcRange;
819 c->dstRange = dstRange;
821 if (isYUV(c->dstFormat) || isGray(c->dstFormat))
824 c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);
825 c->srcFormatBpp = av_get_bits_per_pixel(desc_src);
827 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness,
828 contrast, saturation);
831 if (HAVE_ALTIVEC && av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)
832 ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness,
833 contrast, saturation);
837 int sws_getColorspaceDetails(struct SwsContext *c, int **inv_table,
838 int *srcRange, int **table, int *dstRange,
839 int *brightness, int *contrast, int *saturation)
841 if (!c || isYUV(c->dstFormat) || isGray(c->dstFormat))
844 *inv_table = c->srcColorspaceTable;
845 *table = c->dstColorspaceTable;
846 *srcRange = c->srcRange;
847 *dstRange = c->dstRange;
848 *brightness = c->brightness;
849 *contrast = c->contrast;
850 *saturation = c->saturation;
855 static int handle_jpeg(enum AVPixelFormat *format)
858 case AV_PIX_FMT_YUVJ420P:
859 *format = AV_PIX_FMT_YUV420P;
861 case AV_PIX_FMT_YUVJ422P:
862 *format = AV_PIX_FMT_YUV422P;
864 case AV_PIX_FMT_YUVJ444P:
865 *format = AV_PIX_FMT_YUV444P;
867 case AV_PIX_FMT_YUVJ440P:
868 *format = AV_PIX_FMT_YUV440P;
875 static int handle_0alpha(enum AVPixelFormat *format)
878 case AV_PIX_FMT_0BGR : *format = AV_PIX_FMT_ABGR ; return 1;
879 case AV_PIX_FMT_BGR0 : *format = AV_PIX_FMT_BGRA ; return 4;
880 case AV_PIX_FMT_0RGB : *format = AV_PIX_FMT_ARGB ; return 1;
881 case AV_PIX_FMT_RGB0 : *format = AV_PIX_FMT_RGBA ; return 4;
886 SwsContext *sws_alloc_context(void)
888 SwsContext *c = av_mallocz(sizeof(SwsContext));
890 c->av_class = &sws_context_class;
891 av_opt_set_defaults(c);
896 av_cold int sws_init_context(SwsContext *c, SwsFilter *srcFilter,
897 SwsFilter *dstFilter)
900 int usesVFilter, usesHFilter;
902 SwsFilter dummyFilter = { NULL, NULL, NULL, NULL };
907 int dst_stride = FFALIGN(dstW * sizeof(int16_t) + 66, 16);
908 int flags, cpu_flags;
909 enum AVPixelFormat srcFormat = c->srcFormat;
910 enum AVPixelFormat dstFormat = c->dstFormat;
911 const AVPixFmtDescriptor *desc_src = av_pix_fmt_desc_get(srcFormat);
912 const AVPixFmtDescriptor *desc_dst = av_pix_fmt_desc_get(dstFormat);
914 cpu_flags = av_get_cpu_flags();
920 unscaled = (srcW == dstW && srcH == dstH);
922 handle_jpeg(&srcFormat);
923 handle_jpeg(&dstFormat);
924 handle_0alpha(&srcFormat);
925 handle_0alpha(&dstFormat);
927 if(srcFormat!=c->srcFormat || dstFormat!=c->dstFormat){
928 av_log(c, AV_LOG_WARNING, "deprecated pixel format used, make sure you did set range correctly\n");
929 c->srcFormat= srcFormat;
930 c->dstFormat= dstFormat;
933 if (!sws_isSupportedInput(srcFormat)) {
934 av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n",
935 av_get_pix_fmt_name(srcFormat));
936 return AVERROR(EINVAL);
938 if (!sws_isSupportedOutput(dstFormat)) {
939 av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n",
940 av_get_pix_fmt_name(dstFormat));
941 return AVERROR(EINVAL);
944 i = flags & (SWS_POINT |
955 if (!i || (i & (i - 1))) {
956 av_log(c, AV_LOG_ERROR, "Exactly one scaler algorithm must be chosen, got %X\n", i);
957 return AVERROR(EINVAL);
960 if (srcW < 1 || srcH < 1 || dstW < 1 || dstH < 1) {
961 /* FIXME check if these are enough and try to lower them after
962 * fixing the relevant parts of the code */
963 av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",
964 srcW, srcH, dstW, dstH);
965 return AVERROR(EINVAL);
969 dstFilter = &dummyFilter;
971 srcFilter = &dummyFilter;
973 c->lumXInc = (((int64_t)srcW << 16) + (dstW >> 1)) / dstW;
974 c->lumYInc = (((int64_t)srcH << 16) + (dstH >> 1)) / dstH;
975 c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);
976 c->srcFormatBpp = av_get_bits_per_pixel(desc_src);
977 c->vRounder = 4 * 0x0001000100010001ULL;
979 usesVFilter = (srcFilter->lumV && srcFilter->lumV->length > 1) ||
980 (srcFilter->chrV && srcFilter->chrV->length > 1) ||
981 (dstFilter->lumV && dstFilter->lumV->length > 1) ||
982 (dstFilter->chrV && dstFilter->chrV->length > 1);
983 usesHFilter = (srcFilter->lumH && srcFilter->lumH->length > 1) ||
984 (srcFilter->chrH && srcFilter->chrH->length > 1) ||
985 (dstFilter->lumH && dstFilter->lumH->length > 1) ||
986 (dstFilter->chrH && dstFilter->chrH->length > 1);
988 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
989 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
992 if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) {
994 av_log(c, AV_LOG_DEBUG, "Forcing full internal H chroma due to odd output size\n");
995 flags |= SWS_FULL_CHR_H_INT;
999 /* reuse chroma for 2 pixels RGB/BGR unless user wants full
1000 * chroma interpolation */
1001 if (flags & SWS_FULL_CHR_H_INT &&
1002 isAnyRGB(dstFormat) &&
1003 dstFormat != AV_PIX_FMT_RGBA &&
1004 dstFormat != AV_PIX_FMT_ARGB &&
1005 dstFormat != AV_PIX_FMT_BGRA &&
1006 dstFormat != AV_PIX_FMT_ABGR &&
1007 dstFormat != AV_PIX_FMT_RGB24 &&
1008 dstFormat != AV_PIX_FMT_BGR24) {
1009 av_log(c, AV_LOG_WARNING,
1010 "full chroma interpolation for destination format '%s' not yet implemented\n",
1011 av_get_pix_fmt_name(dstFormat));
1012 flags &= ~SWS_FULL_CHR_H_INT;
1015 if (isAnyRGB(dstFormat) && !(flags & SWS_FULL_CHR_H_INT))
1016 c->chrDstHSubSample = 1;
1018 // drop some chroma lines if the user wants it
1019 c->vChrDrop = (flags & SWS_SRC_V_CHR_DROP_MASK) >>
1020 SWS_SRC_V_CHR_DROP_SHIFT;
1021 c->chrSrcVSubSample += c->vChrDrop;
1023 /* drop every other pixel for chroma calculation unless user
1024 * wants full chroma */
1025 if (isAnyRGB(srcFormat) && !(flags & SWS_FULL_CHR_H_INP) &&
1026 srcFormat != AV_PIX_FMT_RGB8 && srcFormat != AV_PIX_FMT_BGR8 &&
1027 srcFormat != AV_PIX_FMT_RGB4 && srcFormat != AV_PIX_FMT_BGR4 &&
1028 srcFormat != AV_PIX_FMT_RGB4_BYTE && srcFormat != AV_PIX_FMT_BGR4_BYTE &&
1029 ((dstW >> c->chrDstHSubSample) <= (srcW >> 1) ||
1030 (flags & SWS_FAST_BILINEAR)))
1031 c->chrSrcHSubSample = 1;
1033 // Note the -((-x)>>y) is so that we always round toward +inf.
1034 c->chrSrcW = -((-srcW) >> c->chrSrcHSubSample);
1035 c->chrSrcH = -((-srcH) >> c->chrSrcVSubSample);
1036 c->chrDstW = -((-dstW) >> c->chrDstHSubSample);
1037 c->chrDstH = -((-dstH) >> c->chrDstVSubSample);
1039 FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW*2+78, 16) * 2, fail);
1041 /* unscaled special cases */
1042 if (unscaled && !usesHFilter && !usesVFilter &&
1043 (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
1044 ff_get_unscaled_swscale(c);
1047 if (flags & SWS_PRINT_INFO)
1048 av_log(c, AV_LOG_INFO,
1049 "using unscaled %s -> %s special converter\n",
1050 av_get_pix_fmt_name(srcFormat), av_get_pix_fmt_name(dstFormat));
1055 c->srcBpc = 1 + desc_src->comp[0].depth_minus1;
1058 c->dstBpc = 1 + desc_dst->comp[0].depth_minus1;
1061 if (isAnyRGB(srcFormat) || srcFormat == AV_PIX_FMT_PAL8)
1063 if (c->dstBpc == 16)
1065 if (INLINE_MMXEXT(cpu_flags) && c->srcBpc == 8 && c->dstBpc <= 14) {
1066 c->canMMX2BeUsed = (dstW >= srcW && (dstW & 31) == 0 &&
1067 (srcW & 15) == 0) ? 1 : 0;
1068 if (!c->canMMX2BeUsed && dstW >= srcW && (srcW & 15) == 0
1069 && (flags & SWS_FAST_BILINEAR)) {
1070 if (flags & SWS_PRINT_INFO)
1071 av_log(c, AV_LOG_INFO,
1072 "output width is not a multiple of 32 -> no MMX2 scaler\n");
1074 if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat))
1077 c->canMMX2BeUsed = 0;
1079 c->chrXInc = (((int64_t)c->chrSrcW << 16) + (c->chrDstW >> 1)) / c->chrDstW;
1080 c->chrYInc = (((int64_t)c->chrSrcH << 16) + (c->chrDstH >> 1)) / c->chrDstH;
1082 /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src
1083 * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do
1085 * n-2 is the last chrominance sample available.
1086 * This is not perfect, but no one should notice the difference, the more
1087 * correct variant would be like the vertical one, but that would require
1088 * some special code for the first and last pixel */
1089 if (flags & SWS_FAST_BILINEAR) {
1090 if (c->canMMX2BeUsed) {
1094 // we don't use the x86 asm scaler if MMX is available
1095 else if (INLINE_MMX(cpu_flags) && c->dstBpc <= 14) {
1096 c->lumXInc = ((int64_t)(srcW - 2) << 16) / (dstW - 2) - 20;
1097 c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;
1101 #define USE_MMAP (HAVE_MMAP && HAVE_MPROTECT && defined MAP_ANONYMOUS)
1103 /* precalculate horizontal scaler filter coefficients */
1105 #if HAVE_MMXEXT_INLINE
1106 // can't downscale !!!
1107 if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {
1108 c->lumMmx2FilterCodeSize = initMMX2HScaler(dstW, c->lumXInc, NULL,
1110 c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc,
1111 NULL, NULL, NULL, 4);
1114 c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
1115 c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
1116 #elif HAVE_VIRTUALALLOC
1117 c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
1118 c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
1120 c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);
1121 c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);
1124 #ifdef MAP_ANONYMOUS
1125 if (c->lumMmx2FilterCode == MAP_FAILED || c->chrMmx2FilterCode == MAP_FAILED)
1127 if (!c->lumMmx2FilterCode || !c->chrMmx2FilterCode)
1130 av_log(c, AV_LOG_ERROR, "Failed to allocate MMX2FilterCode\n");
1131 return AVERROR(ENOMEM);
1134 FF_ALLOCZ_OR_GOTO(c, c->hLumFilter, (dstW / 8 + 8) * sizeof(int16_t), fail);
1135 FF_ALLOCZ_OR_GOTO(c, c->hChrFilter, (c->chrDstW / 4 + 8) * sizeof(int16_t), fail);
1136 FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW / 2 / 8 + 8) * sizeof(int32_t), fail);
1137 FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);
1139 initMMX2HScaler( dstW, c->lumXInc, c->lumMmx2FilterCode,
1140 c->hLumFilter, (uint32_t*)c->hLumFilterPos, 8);
1141 initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode,
1142 c->hChrFilter, (uint32_t*)c->hChrFilterPos, 4);
1145 mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
1146 mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
1149 #endif /* HAVE_MMXEXT_INLINE */
1151 const int filterAlign =
1152 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :
1153 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
1156 if (initFilter(&c->hLumFilter, &c->hLumFilterPos,
1157 &c->hLumFilterSize, c->lumXInc,
1158 srcW, dstW, filterAlign, 1 << 14,
1159 (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
1160 cpu_flags, srcFilter->lumH, dstFilter->lumH,
1163 if (initFilter(&c->hChrFilter, &c->hChrFilterPos,
1164 &c->hChrFilterSize, c->chrXInc,
1165 c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,
1166 (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
1167 cpu_flags, srcFilter->chrH, dstFilter->chrH,
1171 } // initialize horizontal stuff
1173 /* precalculate vertical scaler filter coefficients */
1175 const int filterAlign =
1176 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 2 :
1177 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
1180 if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,
1181 c->lumYInc, srcH, dstH, filterAlign, (1 << 12),
1182 (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
1183 cpu_flags, srcFilter->lumV, dstFilter->lumV,
1186 if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,
1187 c->chrYInc, c->chrSrcH, c->chrDstH,
1188 filterAlign, (1 << 12),
1189 (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
1190 cpu_flags, srcFilter->chrV, dstFilter->chrV,
1195 FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH, fail);
1196 FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);
1198 for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {
1200 short *p = (short *)&c->vYCoeffsBank[i];
1201 for (j = 0; j < 8; j++)
1202 p[j] = c->vLumFilter[i];
1205 for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {
1207 short *p = (short *)&c->vCCoeffsBank[i];
1208 for (j = 0; j < 8; j++)
1209 p[j] = c->vChrFilter[i];
1214 // calculate buffer sizes so that they won't run out while handling these damn slices
1215 c->vLumBufSize = c->vLumFilterSize;
1216 c->vChrBufSize = c->vChrFilterSize;
1217 for (i = 0; i < dstH; i++) {
1218 int chrI = (int64_t)i * c->chrDstH / dstH;
1219 int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,
1220 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)
1221 << c->chrSrcVSubSample));
1223 nextSlice >>= c->chrSrcVSubSample;
1224 nextSlice <<= c->chrSrcVSubSample;
1225 if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)
1226 c->vLumBufSize = nextSlice - c->vLumFilterPos[i];
1227 if (c->vChrFilterPos[chrI] + c->vChrBufSize <
1228 (nextSlice >> c->chrSrcVSubSample))
1229 c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -
1230 c->vChrFilterPos[chrI];
1233 /* Allocate pixbufs (we use dynamic allocation because otherwise we would
1234 * need to allocate several megabytes to handle all possible cases) */
1235 FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);
1236 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
1237 FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
1238 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
1239 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);
1240 /* Note we need at least one pixel more at the end because of the MMX code
1241 * (just in case someone wants to replace the 4000/8000). */
1242 /* align at 16 bytes for AltiVec */
1243 for (i = 0; i < c->vLumBufSize; i++) {
1244 FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],
1245 dst_stride + 16, fail);
1246 c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];
1248 // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)
1249 c->uv_off = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);
1250 c->uv_offx2 = dst_stride + 16;
1251 for (i = 0; i < c->vChrBufSize; i++) {
1252 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],
1253 dst_stride * 2 + 32, fail);
1254 c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];
1255 c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]
1256 = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;
1258 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
1259 for (i = 0; i < c->vLumBufSize; i++) {
1260 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i + c->vLumBufSize],
1261 dst_stride + 16, fail);
1262 c->alpPixBuf[i] = c->alpPixBuf[i + c->vLumBufSize];
1265 // try to avoid drawing green stuff between the right end and the stride end
1266 for (i = 0; i < c->vChrBufSize; i++)
1267 if(desc_dst->comp[0].depth_minus1 == 15){
1268 av_assert0(c->dstBpc > 14);
1269 for(j=0; j<dst_stride/2+1; j++)
1270 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;
1272 for(j=0; j<dst_stride+1; j++)
1273 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;
1275 av_assert0(c->chrDstH <= dstH);
1277 if (flags & SWS_PRINT_INFO) {
1278 if (flags & SWS_FAST_BILINEAR)
1279 av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
1280 else if (flags & SWS_BILINEAR)
1281 av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
1282 else if (flags & SWS_BICUBIC)
1283 av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
1284 else if (flags & SWS_X)
1285 av_log(c, AV_LOG_INFO, "Experimental scaler, ");
1286 else if (flags & SWS_POINT)
1287 av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
1288 else if (flags & SWS_AREA)
1289 av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
1290 else if (flags & SWS_BICUBLIN)
1291 av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
1292 else if (flags & SWS_GAUSS)
1293 av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
1294 else if (flags & SWS_SINC)
1295 av_log(c, AV_LOG_INFO, "Sinc scaler, ");
1296 else if (flags & SWS_LANCZOS)
1297 av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
1298 else if (flags & SWS_SPLINE)
1299 av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
1301 av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
1303 av_log(c, AV_LOG_INFO, "from %s to %s%s ",
1304 av_get_pix_fmt_name(srcFormat),
1306 dstFormat == AV_PIX_FMT_BGR555 || dstFormat == AV_PIX_FMT_BGR565 ||
1307 dstFormat == AV_PIX_FMT_RGB444BE || dstFormat == AV_PIX_FMT_RGB444LE ||
1308 dstFormat == AV_PIX_FMT_BGR444BE || dstFormat == AV_PIX_FMT_BGR444LE ?
1313 av_get_pix_fmt_name(dstFormat));
1315 if (INLINE_MMXEXT(cpu_flags))
1316 av_log(c, AV_LOG_INFO, "using MMX2\n");
1317 else if (INLINE_AMD3DNOW(cpu_flags))
1318 av_log(c, AV_LOG_INFO, "using 3DNOW\n");
1319 else if (INLINE_MMX(cpu_flags))
1320 av_log(c, AV_LOG_INFO, "using MMX\n");
1321 else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC)
1322 av_log(c, AV_LOG_INFO, "using AltiVec\n");
1324 av_log(c, AV_LOG_INFO, "using C\n");
1326 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
1327 av_log(c, AV_LOG_DEBUG,
1328 "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1329 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
1330 av_log(c, AV_LOG_DEBUG,
1331 "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1332 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,
1333 c->chrXInc, c->chrYInc);
1336 c->swScale = ff_getSwsFunc(c);
1338 fail: // FIXME replace things by appropriate error codes
1342 #if FF_API_SWS_GETCONTEXT
1343 SwsContext *sws_getContext(int srcW, int srcH, enum AVPixelFormat srcFormat,
1344 int dstW, int dstH, enum AVPixelFormat dstFormat,
1345 int flags, SwsFilter *srcFilter,
1346 SwsFilter *dstFilter, const double *param)
1350 if (!(c = sws_alloc_context()))
1358 c->srcRange = handle_jpeg(&srcFormat);
1359 c->dstRange = handle_jpeg(&dstFormat);
1360 c->src0Alpha = handle_0alpha(&srcFormat);
1361 c->dst0Alpha = handle_0alpha(&dstFormat);
1362 c->srcFormat = srcFormat;
1363 c->dstFormat = dstFormat;
1366 c->param[0] = param[0];
1367 c->param[1] = param[1];
1369 sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,
1370 ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,
1371 c->dstRange, 0, 1 << 16, 1 << 16);
1373 if (sws_init_context(c, srcFilter, dstFilter) < 0) {
1382 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
1383 float lumaSharpen, float chromaSharpen,
1384 float chromaHShift, float chromaVShift,
1387 SwsFilter *filter = av_malloc(sizeof(SwsFilter));
1391 if (lumaGBlur != 0.0) {
1392 filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);
1393 filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);
1395 filter->lumH = sws_getIdentityVec();
1396 filter->lumV = sws_getIdentityVec();
1399 if (chromaGBlur != 0.0) {
1400 filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);
1401 filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);
1403 filter->chrH = sws_getIdentityVec();
1404 filter->chrV = sws_getIdentityVec();
1407 if (chromaSharpen != 0.0) {
1408 SwsVector *id = sws_getIdentityVec();
1409 sws_scaleVec(filter->chrH, -chromaSharpen);
1410 sws_scaleVec(filter->chrV, -chromaSharpen);
1411 sws_addVec(filter->chrH, id);
1412 sws_addVec(filter->chrV, id);
1416 if (lumaSharpen != 0.0) {
1417 SwsVector *id = sws_getIdentityVec();
1418 sws_scaleVec(filter->lumH, -lumaSharpen);
1419 sws_scaleVec(filter->lumV, -lumaSharpen);
1420 sws_addVec(filter->lumH, id);
1421 sws_addVec(filter->lumV, id);
1425 if (chromaHShift != 0.0)
1426 sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));
1428 if (chromaVShift != 0.0)
1429 sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));
1431 sws_normalizeVec(filter->chrH, 1.0);
1432 sws_normalizeVec(filter->chrV, 1.0);
1433 sws_normalizeVec(filter->lumH, 1.0);
1434 sws_normalizeVec(filter->lumV, 1.0);
1437 sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
1439 sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
1444 SwsVector *sws_allocVec(int length)
1448 if(length <= 0 || length > INT_MAX/ sizeof(double))
1451 vec = av_malloc(sizeof(SwsVector));
1454 vec->length = length;
1455 vec->coeff = av_malloc(sizeof(double) * length);
1461 SwsVector *sws_getGaussianVec(double variance, double quality)
1463 const int length = (int)(variance * quality + 0.5) | 1;
1465 double middle = (length - 1) * 0.5;
1468 if(variance < 0 || quality < 0)
1471 vec = sws_allocVec(length);
1476 for (i = 0; i < length; i++) {
1477 double dist = i - middle;
1478 vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /
1479 sqrt(2 * variance * M_PI);
1482 sws_normalizeVec(vec, 1.0);
1487 SwsVector *sws_getConstVec(double c, int length)
1490 SwsVector *vec = sws_allocVec(length);
1495 for (i = 0; i < length; i++)
1501 SwsVector *sws_getIdentityVec(void)
1503 return sws_getConstVec(1.0, 1);
1506 static double sws_dcVec(SwsVector *a)
1511 for (i = 0; i < a->length; i++)
1517 void sws_scaleVec(SwsVector *a, double scalar)
1521 for (i = 0; i < a->length; i++)
1522 a->coeff[i] *= scalar;
1525 void sws_normalizeVec(SwsVector *a, double height)
1527 sws_scaleVec(a, height / sws_dcVec(a));
1530 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
1532 int length = a->length + b->length - 1;
1534 SwsVector *vec = sws_getConstVec(0.0, length);
1539 for (i = 0; i < a->length; i++) {
1540 for (j = 0; j < b->length; j++) {
1541 vec->coeff[i + j] += a->coeff[i] * b->coeff[j];
1548 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
1550 int length = FFMAX(a->length, b->length);
1552 SwsVector *vec = sws_getConstVec(0.0, length);
1557 for (i = 0; i < a->length; i++)
1558 vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
1559 for (i = 0; i < b->length; i++)
1560 vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] += b->coeff[i];
1565 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
1567 int length = FFMAX(a->length, b->length);
1569 SwsVector *vec = sws_getConstVec(0.0, length);
1574 for (i = 0; i < a->length; i++)
1575 vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
1576 for (i = 0; i < b->length; i++)
1577 vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] -= b->coeff[i];
1582 /* shift left / or right if "shift" is negative */
1583 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
1585 int length = a->length + FFABS(shift) * 2;
1587 SwsVector *vec = sws_getConstVec(0.0, length);
1592 for (i = 0; i < a->length; i++) {
1593 vec->coeff[i + (length - 1) / 2 -
1594 (a->length - 1) / 2 - shift] = a->coeff[i];
1600 void sws_shiftVec(SwsVector *a, int shift)
1602 SwsVector *shifted = sws_getShiftedVec(a, shift);
1604 a->coeff = shifted->coeff;
1605 a->length = shifted->length;
1609 void sws_addVec(SwsVector *a, SwsVector *b)
1611 SwsVector *sum = sws_sumVec(a, b);
1613 a->coeff = sum->coeff;
1614 a->length = sum->length;
1618 void sws_subVec(SwsVector *a, SwsVector *b)
1620 SwsVector *diff = sws_diffVec(a, b);
1622 a->coeff = diff->coeff;
1623 a->length = diff->length;
1627 void sws_convVec(SwsVector *a, SwsVector *b)
1629 SwsVector *conv = sws_getConvVec(a, b);
1631 a->coeff = conv->coeff;
1632 a->length = conv->length;
1636 SwsVector *sws_cloneVec(SwsVector *a)
1639 SwsVector *vec = sws_allocVec(a->length);
1644 for (i = 0; i < a->length; i++)
1645 vec->coeff[i] = a->coeff[i];
1650 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
1657 for (i = 0; i < a->length; i++)
1658 if (a->coeff[i] > max)
1661 for (i = 0; i < a->length; i++)
1662 if (a->coeff[i] < min)
1667 for (i = 0; i < a->length; i++) {
1668 int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);
1669 av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
1671 av_log(log_ctx, log_level, " ");
1672 av_log(log_ctx, log_level, "|\n");
1676 void sws_freeVec(SwsVector *a)
1680 av_freep(&a->coeff);
1685 void sws_freeFilter(SwsFilter *filter)
1691 sws_freeVec(filter->lumH);
1693 sws_freeVec(filter->lumV);
1695 sws_freeVec(filter->chrH);
1697 sws_freeVec(filter->chrV);
1701 void sws_freeContext(SwsContext *c)
1708 for (i = 0; i < c->vLumBufSize; i++)
1709 av_freep(&c->lumPixBuf[i]);
1710 av_freep(&c->lumPixBuf);
1713 if (c->chrUPixBuf) {
1714 for (i = 0; i < c->vChrBufSize; i++)
1715 av_freep(&c->chrUPixBuf[i]);
1716 av_freep(&c->chrUPixBuf);
1717 av_freep(&c->chrVPixBuf);
1720 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1721 for (i = 0; i < c->vLumBufSize; i++)
1722 av_freep(&c->alpPixBuf[i]);
1723 av_freep(&c->alpPixBuf);
1726 av_freep(&c->vLumFilter);
1727 av_freep(&c->vChrFilter);
1728 av_freep(&c->hLumFilter);
1729 av_freep(&c->hChrFilter);
1731 av_freep(&c->vYCoeffsBank);
1732 av_freep(&c->vCCoeffsBank);
1735 av_freep(&c->vLumFilterPos);
1736 av_freep(&c->vChrFilterPos);
1737 av_freep(&c->hLumFilterPos);
1738 av_freep(&c->hChrFilterPos);
1742 if (c->lumMmx2FilterCode)
1743 munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);
1744 if (c->chrMmx2FilterCode)
1745 munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);
1746 #elif HAVE_VIRTUALALLOC
1747 if (c->lumMmx2FilterCode)
1748 VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);
1749 if (c->chrMmx2FilterCode)
1750 VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);
1752 av_free(c->lumMmx2FilterCode);
1753 av_free(c->chrMmx2FilterCode);
1755 c->lumMmx2FilterCode = NULL;
1756 c->chrMmx2FilterCode = NULL;
1757 #endif /* HAVE_MMX_INLINE */
1759 av_freep(&c->yuvTable);
1760 av_freep(&c->formatConvBuffer);
1765 struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,
1766 int srcH, enum AVPixelFormat srcFormat,
1768 enum AVPixelFormat dstFormat, int flags,
1769 SwsFilter *srcFilter,
1770 SwsFilter *dstFilter,
1771 const double *param)
1773 static const double default_param[2] = { SWS_PARAM_DEFAULT,
1774 SWS_PARAM_DEFAULT };
1777 param = default_param;
1780 (context->srcW != srcW ||
1781 context->srcH != srcH ||
1782 context->srcFormat != srcFormat ||
1783 context->dstW != dstW ||
1784 context->dstH != dstH ||
1785 context->dstFormat != dstFormat ||
1786 context->flags != flags ||
1787 context->param[0] != param[0] ||
1788 context->param[1] != param[1])) {
1789 sws_freeContext(context);
1794 if (!(context = sws_alloc_context()))
1796 context->srcW = srcW;
1797 context->srcH = srcH;
1798 context->srcRange = handle_jpeg(&srcFormat);
1799 context->src0Alpha = handle_0alpha(&srcFormat);
1800 context->srcFormat = srcFormat;
1801 context->dstW = dstW;
1802 context->dstH = dstH;
1803 context->dstRange = handle_jpeg(&dstFormat);
1804 context->dst0Alpha = handle_0alpha(&dstFormat);
1805 context->dstFormat = dstFormat;
1806 context->flags = flags;
1807 context->param[0] = param[0];
1808 context->param[1] = param[1];
1809 sws_setColorspaceDetails(context, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],
1811 ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,
1812 context->dstRange, 0, 1 << 16, 1 << 16);
1813 if (sws_init_context(context, srcFilter, dstFilter) < 0) {
1814 sws_freeContext(context);