2 * Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
4 * This file is part of FFmpeg.
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 #define _SVID_SOURCE // needed for MAP_ANONYMOUS
24 #define _DARWIN_C_SOURCE // needed for MAP_ANON
31 #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
32 #define MAP_ANONYMOUS MAP_ANON
36 #define WIN32_LEAN_AND_MEAN
40 #include "libavutil/attributes.h"
41 #include "libavutil/avassert.h"
42 #include "libavutil/avutil.h"
43 #include "libavutil/bswap.h"
44 #include "libavutil/cpu.h"
45 #include "libavutil/intreadwrite.h"
46 #include "libavutil/mathematics.h"
47 #include "libavutil/opt.h"
48 #include "libavutil/pixdesc.h"
49 #include "libavutil/x86/asm.h"
50 #include "libavutil/x86/cpu.h"
53 #include "swscale_internal.h"
55 unsigned swscale_version(void)
57 av_assert0(LIBSWSCALE_VERSION_MICRO >= 100);
58 return LIBSWSCALE_VERSION_INT;
61 const char *swscale_configuration(void)
63 return FFMPEG_CONFIGURATION;
66 const char *swscale_license(void)
68 #define LICENSE_PREFIX "libswscale license: "
69 return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
72 #define RET 0xC3 // near return opcode for x86
74 typedef struct FormatEntry {
75 int is_supported_in, is_supported_out;
78 static const FormatEntry format_entries[AV_PIX_FMT_NB] = {
79 [AV_PIX_FMT_YUV420P] = { 1, 1 },
80 [AV_PIX_FMT_YUYV422] = { 1, 1 },
81 [AV_PIX_FMT_RGB24] = { 1, 1 },
82 [AV_PIX_FMT_BGR24] = { 1, 1 },
83 [AV_PIX_FMT_YUV422P] = { 1, 1 },
84 [AV_PIX_FMT_YUV444P] = { 1, 1 },
85 [AV_PIX_FMT_YUV410P] = { 1, 1 },
86 [AV_PIX_FMT_YUV411P] = { 1, 1 },
87 [AV_PIX_FMT_GRAY8] = { 1, 1 },
88 [AV_PIX_FMT_MONOWHITE] = { 1, 1 },
89 [AV_PIX_FMT_MONOBLACK] = { 1, 1 },
90 [AV_PIX_FMT_PAL8] = { 1, 0 },
91 [AV_PIX_FMT_YUVJ420P] = { 1, 1 },
92 [AV_PIX_FMT_YUVJ422P] = { 1, 1 },
93 [AV_PIX_FMT_YUVJ444P] = { 1, 1 },
94 [AV_PIX_FMT_UYVY422] = { 1, 1 },
95 [AV_PIX_FMT_UYYVYY411] = { 0, 0 },
96 [AV_PIX_FMT_BGR8] = { 1, 1 },
97 [AV_PIX_FMT_BGR4] = { 0, 1 },
98 [AV_PIX_FMT_BGR4_BYTE] = { 1, 1 },
99 [AV_PIX_FMT_RGB8] = { 1, 1 },
100 [AV_PIX_FMT_RGB4] = { 0, 1 },
101 [AV_PIX_FMT_RGB4_BYTE] = { 1, 1 },
102 [AV_PIX_FMT_NV12] = { 1, 1 },
103 [AV_PIX_FMT_NV21] = { 1, 1 },
104 [AV_PIX_FMT_ARGB] = { 1, 1 },
105 [AV_PIX_FMT_RGBA] = { 1, 1 },
106 [AV_PIX_FMT_ABGR] = { 1, 1 },
107 [AV_PIX_FMT_BGRA] = { 1, 1 },
108 [AV_PIX_FMT_0RGB] = { 1, 1 },
109 [AV_PIX_FMT_RGB0] = { 1, 1 },
110 [AV_PIX_FMT_0BGR] = { 1, 1 },
111 [AV_PIX_FMT_BGR0] = { 1, 1 },
112 [AV_PIX_FMT_GRAY16BE] = { 1, 1 },
113 [AV_PIX_FMT_GRAY16LE] = { 1, 1 },
114 [AV_PIX_FMT_YUV440P] = { 1, 1 },
115 [AV_PIX_FMT_YUVJ440P] = { 1, 1 },
116 [AV_PIX_FMT_YUVA420P] = { 1, 1 },
117 [AV_PIX_FMT_YUVA422P] = { 1, 1 },
118 [AV_PIX_FMT_YUVA444P] = { 1, 1 },
119 [AV_PIX_FMT_YUVA420P9BE] = { 1, 1 },
120 [AV_PIX_FMT_YUVA420P9LE] = { 1, 1 },
121 [AV_PIX_FMT_YUVA422P9BE] = { 1, 1 },
122 [AV_PIX_FMT_YUVA422P9LE] = { 1, 1 },
123 [AV_PIX_FMT_YUVA444P9BE] = { 1, 1 },
124 [AV_PIX_FMT_YUVA444P9LE] = { 1, 1 },
125 [AV_PIX_FMT_YUVA420P10BE]= { 1, 1 },
126 [AV_PIX_FMT_YUVA420P10LE]= { 1, 1 },
127 [AV_PIX_FMT_YUVA422P10BE]= { 1, 1 },
128 [AV_PIX_FMT_YUVA422P10LE]= { 1, 1 },
129 [AV_PIX_FMT_YUVA444P10BE]= { 1, 1 },
130 [AV_PIX_FMT_YUVA444P10LE]= { 1, 1 },
131 [AV_PIX_FMT_YUVA420P16BE]= { 1, 1 },
132 [AV_PIX_FMT_YUVA420P16LE]= { 1, 1 },
133 [AV_PIX_FMT_YUVA422P16BE]= { 1, 1 },
134 [AV_PIX_FMT_YUVA422P16LE]= { 1, 1 },
135 [AV_PIX_FMT_YUVA444P16BE]= { 1, 1 },
136 [AV_PIX_FMT_YUVA444P16LE]= { 1, 1 },
137 [AV_PIX_FMT_RGB48BE] = { 1, 1 },
138 [AV_PIX_FMT_RGB48LE] = { 1, 1 },
139 [AV_PIX_FMT_RGBA64BE] = { 1, 0 },
140 [AV_PIX_FMT_RGBA64LE] = { 1, 0 },
141 [AV_PIX_FMT_RGB565BE] = { 1, 1 },
142 [AV_PIX_FMT_RGB565LE] = { 1, 1 },
143 [AV_PIX_FMT_RGB555BE] = { 1, 1 },
144 [AV_PIX_FMT_RGB555LE] = { 1, 1 },
145 [AV_PIX_FMT_BGR565BE] = { 1, 1 },
146 [AV_PIX_FMT_BGR565LE] = { 1, 1 },
147 [AV_PIX_FMT_BGR555BE] = { 1, 1 },
148 [AV_PIX_FMT_BGR555LE] = { 1, 1 },
149 [AV_PIX_FMT_YUV420P16LE] = { 1, 1 },
150 [AV_PIX_FMT_YUV420P16BE] = { 1, 1 },
151 [AV_PIX_FMT_YUV422P16LE] = { 1, 1 },
152 [AV_PIX_FMT_YUV422P16BE] = { 1, 1 },
153 [AV_PIX_FMT_YUV444P16LE] = { 1, 1 },
154 [AV_PIX_FMT_YUV444P16BE] = { 1, 1 },
155 [AV_PIX_FMT_RGB444LE] = { 1, 1 },
156 [AV_PIX_FMT_RGB444BE] = { 1, 1 },
157 [AV_PIX_FMT_BGR444LE] = { 1, 1 },
158 [AV_PIX_FMT_BGR444BE] = { 1, 1 },
159 [AV_PIX_FMT_Y400A] = { 1, 0 },
160 [AV_PIX_FMT_BGR48BE] = { 1, 1 },
161 [AV_PIX_FMT_BGR48LE] = { 1, 1 },
162 [AV_PIX_FMT_BGRA64BE] = { 0, 0 },
163 [AV_PIX_FMT_BGRA64LE] = { 0, 0 },
164 [AV_PIX_FMT_YUV420P9BE] = { 1, 1 },
165 [AV_PIX_FMT_YUV420P9LE] = { 1, 1 },
166 [AV_PIX_FMT_YUV420P10BE] = { 1, 1 },
167 [AV_PIX_FMT_YUV420P10LE] = { 1, 1 },
168 [AV_PIX_FMT_YUV420P12BE] = { 1, 1 },
169 [AV_PIX_FMT_YUV420P12LE] = { 1, 1 },
170 [AV_PIX_FMT_YUV420P14BE] = { 1, 1 },
171 [AV_PIX_FMT_YUV420P14LE] = { 1, 1 },
172 [AV_PIX_FMT_YUV422P9BE] = { 1, 1 },
173 [AV_PIX_FMT_YUV422P9LE] = { 1, 1 },
174 [AV_PIX_FMT_YUV422P10BE] = { 1, 1 },
175 [AV_PIX_FMT_YUV422P10LE] = { 1, 1 },
176 [AV_PIX_FMT_YUV422P12BE] = { 1, 1 },
177 [AV_PIX_FMT_YUV422P12LE] = { 1, 1 },
178 [AV_PIX_FMT_YUV422P14BE] = { 1, 1 },
179 [AV_PIX_FMT_YUV422P14LE] = { 1, 1 },
180 [AV_PIX_FMT_YUV444P9BE] = { 1, 1 },
181 [AV_PIX_FMT_YUV444P9LE] = { 1, 1 },
182 [AV_PIX_FMT_YUV444P10BE] = { 1, 1 },
183 [AV_PIX_FMT_YUV444P10LE] = { 1, 1 },
184 [AV_PIX_FMT_YUV444P12BE] = { 1, 1 },
185 [AV_PIX_FMT_YUV444P12LE] = { 1, 1 },
186 [AV_PIX_FMT_YUV444P14BE] = { 1, 1 },
187 [AV_PIX_FMT_YUV444P14LE] = { 1, 1 },
188 [AV_PIX_FMT_GBRP] = { 1, 1 },
189 [AV_PIX_FMT_GBRP9LE] = { 1, 1 },
190 [AV_PIX_FMT_GBRP9BE] = { 1, 1 },
191 [AV_PIX_FMT_GBRP10LE] = { 1, 1 },
192 [AV_PIX_FMT_GBRP10BE] = { 1, 1 },
193 [AV_PIX_FMT_GBRP12LE] = { 1, 1 },
194 [AV_PIX_FMT_GBRP12BE] = { 1, 1 },
195 [AV_PIX_FMT_GBRP14LE] = { 1, 1 },
196 [AV_PIX_FMT_GBRP14BE] = { 1, 1 },
197 [AV_PIX_FMT_GBRP16LE] = { 1, 0 },
198 [AV_PIX_FMT_GBRP16BE] = { 1, 0 },
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 init_hscaler_mmxext(int dstW, int xInc, uint8_t *filterCode,
631 int16_t *filter, int32_t *filterPos,
635 x86_reg imm8OfPShufW1A;
636 x86_reg imm8OfPShufW2A;
637 x86_reg fragmentLengthA;
639 x86_reg imm8OfPShufW1B;
640 x86_reg imm8OfPShufW2B;
641 x86_reg fragmentLengthB;
646 // create an optimized horizontal scaling routine
647 /* This scaler is made of runtime-generated MMXEXT code using specially tuned
648 * pshufw instructions. For every four output pixels, if four input pixels
649 * are enough for the fast bilinear scaling, then a chunk of fragmentB is
650 * used. If five input pixels are needed, then a chunk of fragmentA is used.
659 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
660 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
661 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
662 "punpcklbw %%mm7, %%mm1 \n\t"
663 "punpcklbw %%mm7, %%mm0 \n\t"
664 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
666 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
668 "psubw %%mm1, %%mm0 \n\t"
669 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
670 "pmullw %%mm3, %%mm0 \n\t"
671 "psllw $7, %%mm1 \n\t"
672 "paddw %%mm1, %%mm0 \n\t"
674 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
676 "add $8, %%"REG_a" \n\t"
680 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
681 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
682 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
687 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
691 : "=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
692 "=r" (fragmentLengthA)
699 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
700 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
701 "punpcklbw %%mm7, %%mm0 \n\t"
702 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
704 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
706 "psubw %%mm1, %%mm0 \n\t"
707 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
708 "pmullw %%mm3, %%mm0 \n\t"
709 "psllw $7, %%mm1 \n\t"
710 "paddw %%mm1, %%mm0 \n\t"
712 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
714 "add $8, %%"REG_a" \n\t"
718 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
719 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
720 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
725 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
729 : "=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
730 "=r" (fragmentLengthB)
733 xpos = 0; // lumXInc/2 - 0x8000; // difference between pixel centers
736 for (i = 0; i < dstW / numSplits; i++) {
741 int b = ((xpos + xInc) >> 16) - xx;
742 int c = ((xpos + xInc * 2) >> 16) - xx;
743 int d = ((xpos + xInc * 3) >> 16) - xx;
744 int inc = (d + 1 < 4);
745 uint8_t *fragment = (d + 1 < 4) ? fragmentB : fragmentA;
746 x86_reg imm8OfPShufW1 = (d + 1 < 4) ? imm8OfPShufW1B : imm8OfPShufW1A;
747 x86_reg imm8OfPShufW2 = (d + 1 < 4) ? imm8OfPShufW2B : imm8OfPShufW2A;
748 x86_reg fragmentLength = (d + 1 < 4) ? fragmentLengthB : fragmentLengthA;
749 int maxShift = 3 - (d + inc);
753 filter[i] = ((xpos & 0xFFFF) ^ 0xFFFF) >> 9;
754 filter[i + 1] = (((xpos + xInc) & 0xFFFF) ^ 0xFFFF) >> 9;
755 filter[i + 2] = (((xpos + xInc * 2) & 0xFFFF) ^ 0xFFFF) >> 9;
756 filter[i + 3] = (((xpos + xInc * 3) & 0xFFFF) ^ 0xFFFF) >> 9;
757 filterPos[i / 2] = xx;
759 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
761 filterCode[fragmentPos + imm8OfPShufW1] = (a + inc) |
765 filterCode[fragmentPos + imm8OfPShufW2] = a | (b << 2) |
769 if (i + 4 - inc >= dstW)
770 shift = maxShift; // avoid overread
771 else if ((filterPos[i / 2] & 3) <= maxShift)
772 shift = filterPos[i / 2] & 3; // align
774 if (shift && i >= shift) {
775 filterCode[fragmentPos + imm8OfPShufW1] += 0x55 * shift;
776 filterCode[fragmentPos + imm8OfPShufW2] += 0x55 * shift;
777 filterPos[i / 2] -= shift;
781 fragmentPos += fragmentLength;
784 filterCode[fragmentPos] = RET;
789 filterPos[((i / 2) + 1) & (~1)] = xpos >> 16; // needed to jump to the next part
791 return fragmentPos + 1;
793 #endif /* HAVE_MMXEXT_INLINE */
795 static void getSubSampleFactors(int *h, int *v, enum AVPixelFormat format)
797 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(format);
798 *h = desc->log2_chroma_w;
799 *v = desc->log2_chroma_h;
802 int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],
803 int srcRange, const int table[4], int dstRange,
804 int brightness, int contrast, int saturation)
806 const AVPixFmtDescriptor *desc_dst = av_pix_fmt_desc_get(c->dstFormat);
807 const AVPixFmtDescriptor *desc_src = av_pix_fmt_desc_get(c->srcFormat);
808 memcpy(c->srcColorspaceTable, inv_table, sizeof(int) * 4);
809 memcpy(c->dstColorspaceTable, table, sizeof(int) * 4);
811 if(!isYUV(c->dstFormat) && !isGray(c->dstFormat))
813 if(!isYUV(c->srcFormat) && !isGray(c->srcFormat))
816 c->brightness = brightness;
817 c->contrast = contrast;
818 c->saturation = saturation;
819 c->srcRange = srcRange;
820 c->dstRange = dstRange;
822 if (isYUV(c->dstFormat) || isGray(c->dstFormat))
825 c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);
826 c->srcFormatBpp = av_get_bits_per_pixel(desc_src);
828 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness,
829 contrast, saturation);
832 if (HAVE_ALTIVEC && av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)
833 ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness,
834 contrast, saturation);
838 int sws_getColorspaceDetails(struct SwsContext *c, int **inv_table,
839 int *srcRange, int **table, int *dstRange,
840 int *brightness, int *contrast, int *saturation)
842 if (!c || isYUV(c->dstFormat) || isGray(c->dstFormat))
845 *inv_table = c->srcColorspaceTable;
846 *table = c->dstColorspaceTable;
847 *srcRange = c->srcRange;
848 *dstRange = c->dstRange;
849 *brightness = c->brightness;
850 *contrast = c->contrast;
851 *saturation = c->saturation;
856 static int handle_jpeg(enum AVPixelFormat *format)
859 case AV_PIX_FMT_YUVJ420P:
860 *format = AV_PIX_FMT_YUV420P;
862 case AV_PIX_FMT_YUVJ422P:
863 *format = AV_PIX_FMT_YUV422P;
865 case AV_PIX_FMT_YUVJ444P:
866 *format = AV_PIX_FMT_YUV444P;
868 case AV_PIX_FMT_YUVJ440P:
869 *format = AV_PIX_FMT_YUV440P;
876 static int handle_0alpha(enum AVPixelFormat *format)
879 case AV_PIX_FMT_0BGR : *format = AV_PIX_FMT_ABGR ; return 1;
880 case AV_PIX_FMT_BGR0 : *format = AV_PIX_FMT_BGRA ; return 4;
881 case AV_PIX_FMT_0RGB : *format = AV_PIX_FMT_ARGB ; return 1;
882 case AV_PIX_FMT_RGB0 : *format = AV_PIX_FMT_RGBA ; return 4;
887 SwsContext *sws_alloc_context(void)
889 SwsContext *c = av_mallocz(sizeof(SwsContext));
892 c->av_class = &sws_context_class;
893 av_opt_set_defaults(c);
899 av_cold int sws_init_context(SwsContext *c, SwsFilter *srcFilter,
900 SwsFilter *dstFilter)
903 int usesVFilter, usesHFilter;
905 SwsFilter dummyFilter = { NULL, NULL, NULL, NULL };
910 int dst_stride = FFALIGN(dstW * sizeof(int16_t) + 66, 16);
911 int flags, cpu_flags;
912 enum AVPixelFormat srcFormat = c->srcFormat;
913 enum AVPixelFormat dstFormat = c->dstFormat;
914 const AVPixFmtDescriptor *desc_src = av_pix_fmt_desc_get(srcFormat);
915 const AVPixFmtDescriptor *desc_dst = av_pix_fmt_desc_get(dstFormat);
917 cpu_flags = av_get_cpu_flags();
923 unscaled = (srcW == dstW && srcH == dstH);
925 handle_jpeg(&srcFormat);
926 handle_jpeg(&dstFormat);
927 handle_0alpha(&srcFormat);
928 handle_0alpha(&dstFormat);
930 if(srcFormat!=c->srcFormat || dstFormat!=c->dstFormat){
931 av_log(c, AV_LOG_WARNING, "deprecated pixel format used, make sure you did set range correctly\n");
932 c->srcFormat= srcFormat;
933 c->dstFormat= dstFormat;
936 if (!sws_isSupportedInput(srcFormat)) {
937 av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n",
938 av_get_pix_fmt_name(srcFormat));
939 return AVERROR(EINVAL);
941 if (!sws_isSupportedOutput(dstFormat)) {
942 av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n",
943 av_get_pix_fmt_name(dstFormat));
944 return AVERROR(EINVAL);
947 i = flags & (SWS_POINT |
958 if (!i || (i & (i - 1))) {
959 av_log(c, AV_LOG_ERROR, "Exactly one scaler algorithm must be chosen, got %X\n", i);
960 return AVERROR(EINVAL);
963 if (srcW < 1 || srcH < 1 || dstW < 1 || dstH < 1) {
964 /* FIXME check if these are enough and try to lower them after
965 * fixing the relevant parts of the code */
966 av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",
967 srcW, srcH, dstW, dstH);
968 return AVERROR(EINVAL);
972 dstFilter = &dummyFilter;
974 srcFilter = &dummyFilter;
976 c->lumXInc = (((int64_t)srcW << 16) + (dstW >> 1)) / dstW;
977 c->lumYInc = (((int64_t)srcH << 16) + (dstH >> 1)) / dstH;
978 c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);
979 c->srcFormatBpp = av_get_bits_per_pixel(desc_src);
980 c->vRounder = 4 * 0x0001000100010001ULL;
982 usesVFilter = (srcFilter->lumV && srcFilter->lumV->length > 1) ||
983 (srcFilter->chrV && srcFilter->chrV->length > 1) ||
984 (dstFilter->lumV && dstFilter->lumV->length > 1) ||
985 (dstFilter->chrV && dstFilter->chrV->length > 1);
986 usesHFilter = (srcFilter->lumH && srcFilter->lumH->length > 1) ||
987 (srcFilter->chrH && srcFilter->chrH->length > 1) ||
988 (dstFilter->lumH && dstFilter->lumH->length > 1) ||
989 (dstFilter->chrH && dstFilter->chrH->length > 1);
991 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
992 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
994 if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) {
996 av_log(c, AV_LOG_DEBUG, "Forcing full internal H chroma due to odd output size\n");
997 flags |= SWS_FULL_CHR_H_INT;
1002 if(dstFormat == AV_PIX_FMT_BGR4_BYTE ||
1003 dstFormat == AV_PIX_FMT_RGB4_BYTE ||
1004 dstFormat == AV_PIX_FMT_BGR8 ||
1005 dstFormat == AV_PIX_FMT_RGB8) {
1006 if (flags & SWS_ERROR_DIFFUSION && !(flags & SWS_FULL_CHR_H_INT)) {
1007 av_log(c, AV_LOG_DEBUG,
1008 "Error diffusion dither is only supported in full chroma interpolation for destination format '%s'\n",
1009 av_get_pix_fmt_name(dstFormat));
1010 flags |= SWS_FULL_CHR_H_INT;
1013 if (!(flags & SWS_ERROR_DIFFUSION) && (flags & SWS_FULL_CHR_H_INT)) {
1014 av_log(c, AV_LOG_DEBUG,
1015 "Ordered dither is not supported in full chroma interpolation for destination format '%s'\n",
1016 av_get_pix_fmt_name(dstFormat));
1017 flags |= SWS_ERROR_DIFFUSION;
1021 if (isPlanarRGB(dstFormat)) {
1022 if (!(flags & SWS_FULL_CHR_H_INT)) {
1023 av_log(c, AV_LOG_DEBUG,
1024 "%s output is not supported with half chroma resolution, switching to full\n",
1025 av_get_pix_fmt_name(dstFormat));
1026 flags |= SWS_FULL_CHR_H_INT;
1031 /* reuse chroma for 2 pixels RGB/BGR unless user wants full
1032 * chroma interpolation */
1033 if (flags & SWS_FULL_CHR_H_INT &&
1034 isAnyRGB(dstFormat) &&
1035 !isPlanarRGB(dstFormat) &&
1036 dstFormat != AV_PIX_FMT_RGBA &&
1037 dstFormat != AV_PIX_FMT_ARGB &&
1038 dstFormat != AV_PIX_FMT_BGRA &&
1039 dstFormat != AV_PIX_FMT_ABGR &&
1040 dstFormat != AV_PIX_FMT_RGB24 &&
1041 dstFormat != AV_PIX_FMT_BGR24 &&
1042 dstFormat != AV_PIX_FMT_BGR4_BYTE &&
1043 dstFormat != AV_PIX_FMT_RGB4_BYTE &&
1044 dstFormat != AV_PIX_FMT_BGR8 &&
1045 dstFormat != AV_PIX_FMT_RGB8
1047 av_log(c, AV_LOG_WARNING,
1048 "full chroma interpolation for destination format '%s' not yet implemented\n",
1049 av_get_pix_fmt_name(dstFormat));
1050 flags &= ~SWS_FULL_CHR_H_INT;
1053 if (isAnyRGB(dstFormat) && !(flags & SWS_FULL_CHR_H_INT))
1054 c->chrDstHSubSample = 1;
1056 // drop some chroma lines if the user wants it
1057 c->vChrDrop = (flags & SWS_SRC_V_CHR_DROP_MASK) >>
1058 SWS_SRC_V_CHR_DROP_SHIFT;
1059 c->chrSrcVSubSample += c->vChrDrop;
1061 /* drop every other pixel for chroma calculation unless user
1062 * wants full chroma */
1063 if (isAnyRGB(srcFormat) && !(flags & SWS_FULL_CHR_H_INP) &&
1064 srcFormat != AV_PIX_FMT_RGB8 && srcFormat != AV_PIX_FMT_BGR8 &&
1065 srcFormat != AV_PIX_FMT_RGB4 && srcFormat != AV_PIX_FMT_BGR4 &&
1066 srcFormat != AV_PIX_FMT_RGB4_BYTE && srcFormat != AV_PIX_FMT_BGR4_BYTE &&
1067 srcFormat != AV_PIX_FMT_GBRP9BE && srcFormat != AV_PIX_FMT_GBRP9LE &&
1068 srcFormat != AV_PIX_FMT_GBRP10BE && srcFormat != AV_PIX_FMT_GBRP10LE &&
1069 srcFormat != AV_PIX_FMT_GBRP12BE && srcFormat != AV_PIX_FMT_GBRP12LE &&
1070 srcFormat != AV_PIX_FMT_GBRP14BE && srcFormat != AV_PIX_FMT_GBRP14LE &&
1071 srcFormat != AV_PIX_FMT_GBRP16BE && srcFormat != AV_PIX_FMT_GBRP16LE &&
1072 ((dstW >> c->chrDstHSubSample) <= (srcW >> 1) ||
1073 (flags & SWS_FAST_BILINEAR)))
1074 c->chrSrcHSubSample = 1;
1076 // Note the -((-x)>>y) is so that we always round toward +inf.
1077 c->chrSrcW = -((-srcW) >> c->chrSrcHSubSample);
1078 c->chrSrcH = -((-srcH) >> c->chrSrcVSubSample);
1079 c->chrDstW = -((-dstW) >> c->chrDstHSubSample);
1080 c->chrDstH = -((-dstH) >> c->chrDstVSubSample);
1082 FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW*2+78, 16) * 2, fail);
1084 /* unscaled special cases */
1085 if (unscaled && !usesHFilter && !usesVFilter &&
1086 (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
1087 ff_get_unscaled_swscale(c);
1090 if (flags & SWS_PRINT_INFO)
1091 av_log(c, AV_LOG_INFO,
1092 "using unscaled %s -> %s special converter\n",
1093 av_get_pix_fmt_name(srcFormat), av_get_pix_fmt_name(dstFormat));
1098 c->srcBpc = 1 + desc_src->comp[0].depth_minus1;
1101 c->dstBpc = 1 + desc_dst->comp[0].depth_minus1;
1104 if (isAnyRGB(srcFormat) || srcFormat == AV_PIX_FMT_PAL8)
1106 if (c->dstBpc == 16)
1109 if (INLINE_MMXEXT(cpu_flags) && c->srcBpc == 8 && c->dstBpc <= 14) {
1110 c->canMMXEXTBeUsed = (dstW >= srcW && (dstW & 31) == 0 &&
1111 (srcW & 15) == 0) ? 1 : 0;
1112 if (!c->canMMXEXTBeUsed && dstW >= srcW && (srcW & 15) == 0
1114 && (flags & SWS_FAST_BILINEAR)) {
1115 if (flags & SWS_PRINT_INFO)
1116 av_log(c, AV_LOG_INFO,
1117 "output width is not a multiple of 32 -> no MMXEXT scaler\n");
1119 if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat))
1120 c->canMMXEXTBeUsed = 0;
1122 c->canMMXEXTBeUsed = 0;
1124 c->chrXInc = (((int64_t)c->chrSrcW << 16) + (c->chrDstW >> 1)) / c->chrDstW;
1125 c->chrYInc = (((int64_t)c->chrSrcH << 16) + (c->chrDstH >> 1)) / c->chrDstH;
1127 /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src
1128 * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do
1130 * n-2 is the last chrominance sample available.
1131 * This is not perfect, but no one should notice the difference, the more
1132 * correct variant would be like the vertical one, but that would require
1133 * some special code for the first and last pixel */
1134 if (flags & SWS_FAST_BILINEAR) {
1135 if (c->canMMXEXTBeUsed) {
1139 // we don't use the x86 asm scaler if MMX is available
1140 else if (INLINE_MMX(cpu_flags) && c->dstBpc <= 14) {
1141 c->lumXInc = ((int64_t)(srcW - 2) << 16) / (dstW - 2) - 20;
1142 c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;
1146 #define USE_MMAP (HAVE_MMAP && HAVE_MPROTECT && defined MAP_ANONYMOUS)
1148 /* precalculate horizontal scaler filter coefficients */
1150 #if HAVE_MMXEXT_INLINE
1151 // can't downscale !!!
1152 if (c->canMMXEXTBeUsed && (flags & SWS_FAST_BILINEAR)) {
1153 c->lumMmxextFilterCodeSize = init_hscaler_mmxext(dstW, c->lumXInc, NULL,
1155 c->chrMmxextFilterCodeSize = init_hscaler_mmxext(c->chrDstW, c->chrXInc,
1156 NULL, NULL, NULL, 4);
1159 c->lumMmxextFilterCode = mmap(NULL, c->lumMmxextFilterCodeSize,
1160 PROT_READ | PROT_WRITE,
1161 MAP_PRIVATE | MAP_ANONYMOUS,
1163 c->chrMmxextFilterCode = mmap(NULL, c->chrMmxextFilterCodeSize,
1164 PROT_READ | PROT_WRITE,
1165 MAP_PRIVATE | MAP_ANONYMOUS,
1167 #elif HAVE_VIRTUALALLOC
1168 c->lumMmxextFilterCode = VirtualAlloc(NULL,
1169 c->lumMmxextFilterCodeSize,
1171 PAGE_EXECUTE_READWRITE);
1172 c->chrMmxextFilterCode = VirtualAlloc(NULL,
1173 c->chrMmxextFilterCodeSize,
1175 PAGE_EXECUTE_READWRITE);
1177 c->lumMmxextFilterCode = av_malloc(c->lumMmxextFilterCodeSize);
1178 c->chrMmxextFilterCode = av_malloc(c->chrMmxextFilterCodeSize);
1181 #ifdef MAP_ANONYMOUS
1182 if (c->lumMmxextFilterCode == MAP_FAILED || c->chrMmxextFilterCode == MAP_FAILED)
1184 if (!c->lumMmxextFilterCode || !c->chrMmxextFilterCode)
1187 av_log(c, AV_LOG_ERROR, "Failed to allocate MMX2FilterCode\n");
1188 return AVERROR(ENOMEM);
1191 FF_ALLOCZ_OR_GOTO(c, c->hLumFilter, (dstW / 8 + 8) * sizeof(int16_t), fail);
1192 FF_ALLOCZ_OR_GOTO(c, c->hChrFilter, (c->chrDstW / 4 + 8) * sizeof(int16_t), fail);
1193 FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW / 2 / 8 + 8) * sizeof(int32_t), fail);
1194 FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);
1196 init_hscaler_mmxext( dstW, c->lumXInc, c->lumMmxextFilterCode,
1197 c->hLumFilter, (uint32_t*)c->hLumFilterPos, 8);
1198 init_hscaler_mmxext(c->chrDstW, c->chrXInc, c->chrMmxextFilterCode,
1199 c->hChrFilter, (uint32_t*)c->hChrFilterPos, 4);
1202 mprotect(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize, PROT_EXEC | PROT_READ);
1203 mprotect(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize, PROT_EXEC | PROT_READ);
1206 #endif /* HAVE_MMXEXT_INLINE */
1208 const int filterAlign =
1209 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :
1210 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
1213 if (initFilter(&c->hLumFilter, &c->hLumFilterPos,
1214 &c->hLumFilterSize, c->lumXInc,
1215 srcW, dstW, filterAlign, 1 << 14,
1216 (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
1217 cpu_flags, srcFilter->lumH, dstFilter->lumH,
1220 if (initFilter(&c->hChrFilter, &c->hChrFilterPos,
1221 &c->hChrFilterSize, c->chrXInc,
1222 c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,
1223 (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
1224 cpu_flags, srcFilter->chrH, dstFilter->chrH,
1228 } // initialize horizontal stuff
1230 /* precalculate vertical scaler filter coefficients */
1232 const int filterAlign =
1233 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 2 :
1234 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
1237 if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,
1238 c->lumYInc, srcH, dstH, filterAlign, (1 << 12),
1239 (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
1240 cpu_flags, srcFilter->lumV, dstFilter->lumV,
1243 if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,
1244 c->chrYInc, c->chrSrcH, c->chrDstH,
1245 filterAlign, (1 << 12),
1246 (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
1247 cpu_flags, srcFilter->chrV, dstFilter->chrV,
1252 FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH, fail);
1253 FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);
1255 for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {
1257 short *p = (short *)&c->vYCoeffsBank[i];
1258 for (j = 0; j < 8; j++)
1259 p[j] = c->vLumFilter[i];
1262 for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {
1264 short *p = (short *)&c->vCCoeffsBank[i];
1265 for (j = 0; j < 8; j++)
1266 p[j] = c->vChrFilter[i];
1271 // calculate buffer sizes so that they won't run out while handling these damn slices
1272 c->vLumBufSize = c->vLumFilterSize;
1273 c->vChrBufSize = c->vChrFilterSize;
1274 for (i = 0; i < dstH; i++) {
1275 int chrI = (int64_t)i * c->chrDstH / dstH;
1276 int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,
1277 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)
1278 << c->chrSrcVSubSample));
1280 nextSlice >>= c->chrSrcVSubSample;
1281 nextSlice <<= c->chrSrcVSubSample;
1282 if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)
1283 c->vLumBufSize = nextSlice - c->vLumFilterPos[i];
1284 if (c->vChrFilterPos[chrI] + c->vChrBufSize <
1285 (nextSlice >> c->chrSrcVSubSample))
1286 c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -
1287 c->vChrFilterPos[chrI];
1290 for (i = 0; i < 4; i++)
1291 FF_ALLOCZ_OR_GOTO(c, c->dither_error[i], (c->dstW+2) * sizeof(int), fail);
1293 /* Allocate pixbufs (we use dynamic allocation because otherwise we would
1294 * need to allocate several megabytes to handle all possible cases) */
1295 FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);
1296 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
1297 FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
1298 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
1299 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);
1300 /* Note we need at least one pixel more at the end because of the MMX code
1301 * (just in case someone wants to replace the 4000/8000). */
1302 /* align at 16 bytes for AltiVec */
1303 for (i = 0; i < c->vLumBufSize; i++) {
1304 FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],
1305 dst_stride + 16, fail);
1306 c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];
1308 // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)
1309 c->uv_off = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);
1310 c->uv_offx2 = dst_stride + 16;
1311 for (i = 0; i < c->vChrBufSize; i++) {
1312 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],
1313 dst_stride * 2 + 32, fail);
1314 c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];
1315 c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]
1316 = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;
1318 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
1319 for (i = 0; i < c->vLumBufSize; i++) {
1320 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i + c->vLumBufSize],
1321 dst_stride + 16, fail);
1322 c->alpPixBuf[i] = c->alpPixBuf[i + c->vLumBufSize];
1325 // try to avoid drawing green stuff between the right end and the stride end
1326 for (i = 0; i < c->vChrBufSize; i++)
1327 if(desc_dst->comp[0].depth_minus1 == 15){
1328 av_assert0(c->dstBpc > 14);
1329 for(j=0; j<dst_stride/2+1; j++)
1330 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;
1332 for(j=0; j<dst_stride+1; j++)
1333 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;
1335 av_assert0(c->chrDstH <= dstH);
1337 if (flags & SWS_PRINT_INFO) {
1338 if (flags & SWS_FAST_BILINEAR)
1339 av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
1340 else if (flags & SWS_BILINEAR)
1341 av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
1342 else if (flags & SWS_BICUBIC)
1343 av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
1344 else if (flags & SWS_X)
1345 av_log(c, AV_LOG_INFO, "Experimental scaler, ");
1346 else if (flags & SWS_POINT)
1347 av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
1348 else if (flags & SWS_AREA)
1349 av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
1350 else if (flags & SWS_BICUBLIN)
1351 av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
1352 else if (flags & SWS_GAUSS)
1353 av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
1354 else if (flags & SWS_SINC)
1355 av_log(c, AV_LOG_INFO, "Sinc scaler, ");
1356 else if (flags & SWS_LANCZOS)
1357 av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
1358 else if (flags & SWS_SPLINE)
1359 av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
1361 av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
1363 av_log(c, AV_LOG_INFO, "from %s to %s%s ",
1364 av_get_pix_fmt_name(srcFormat),
1366 dstFormat == AV_PIX_FMT_BGR555 || dstFormat == AV_PIX_FMT_BGR565 ||
1367 dstFormat == AV_PIX_FMT_RGB444BE || dstFormat == AV_PIX_FMT_RGB444LE ||
1368 dstFormat == AV_PIX_FMT_BGR444BE || dstFormat == AV_PIX_FMT_BGR444LE ?
1373 av_get_pix_fmt_name(dstFormat));
1375 if (INLINE_MMXEXT(cpu_flags))
1376 av_log(c, AV_LOG_INFO, "using MMXEXT\n");
1377 else if (INLINE_AMD3DNOW(cpu_flags))
1378 av_log(c, AV_LOG_INFO, "using 3DNOW\n");
1379 else if (INLINE_MMX(cpu_flags))
1380 av_log(c, AV_LOG_INFO, "using MMX\n");
1381 else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC)
1382 av_log(c, AV_LOG_INFO, "using AltiVec\n");
1384 av_log(c, AV_LOG_INFO, "using C\n");
1386 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
1387 av_log(c, AV_LOG_DEBUG,
1388 "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1389 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
1390 av_log(c, AV_LOG_DEBUG,
1391 "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1392 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,
1393 c->chrXInc, c->chrYInc);
1396 c->swScale = ff_getSwsFunc(c);
1398 fail: // FIXME replace things by appropriate error codes
1402 #if FF_API_SWS_GETCONTEXT
1403 SwsContext *sws_getContext(int srcW, int srcH, enum AVPixelFormat srcFormat,
1404 int dstW, int dstH, enum AVPixelFormat dstFormat,
1405 int flags, SwsFilter *srcFilter,
1406 SwsFilter *dstFilter, const double *param)
1410 if (!(c = sws_alloc_context()))
1418 c->srcRange = handle_jpeg(&srcFormat);
1419 c->dstRange = handle_jpeg(&dstFormat);
1420 c->src0Alpha = handle_0alpha(&srcFormat);
1421 c->dst0Alpha = handle_0alpha(&dstFormat);
1422 c->srcFormat = srcFormat;
1423 c->dstFormat = dstFormat;
1426 c->param[0] = param[0];
1427 c->param[1] = param[1];
1429 sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,
1430 ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,
1431 c->dstRange, 0, 1 << 16, 1 << 16);
1433 if (sws_init_context(c, srcFilter, dstFilter) < 0) {
1442 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
1443 float lumaSharpen, float chromaSharpen,
1444 float chromaHShift, float chromaVShift,
1447 SwsFilter *filter = av_malloc(sizeof(SwsFilter));
1451 if (lumaGBlur != 0.0) {
1452 filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);
1453 filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);
1455 filter->lumH = sws_getIdentityVec();
1456 filter->lumV = sws_getIdentityVec();
1459 if (chromaGBlur != 0.0) {
1460 filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);
1461 filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);
1463 filter->chrH = sws_getIdentityVec();
1464 filter->chrV = sws_getIdentityVec();
1467 if (chromaSharpen != 0.0) {
1468 SwsVector *id = sws_getIdentityVec();
1469 sws_scaleVec(filter->chrH, -chromaSharpen);
1470 sws_scaleVec(filter->chrV, -chromaSharpen);
1471 sws_addVec(filter->chrH, id);
1472 sws_addVec(filter->chrV, id);
1476 if (lumaSharpen != 0.0) {
1477 SwsVector *id = sws_getIdentityVec();
1478 sws_scaleVec(filter->lumH, -lumaSharpen);
1479 sws_scaleVec(filter->lumV, -lumaSharpen);
1480 sws_addVec(filter->lumH, id);
1481 sws_addVec(filter->lumV, id);
1485 if (chromaHShift != 0.0)
1486 sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));
1488 if (chromaVShift != 0.0)
1489 sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));
1491 sws_normalizeVec(filter->chrH, 1.0);
1492 sws_normalizeVec(filter->chrV, 1.0);
1493 sws_normalizeVec(filter->lumH, 1.0);
1494 sws_normalizeVec(filter->lumV, 1.0);
1497 sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
1499 sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
1504 SwsVector *sws_allocVec(int length)
1508 if(length <= 0 || length > INT_MAX/ sizeof(double))
1511 vec = av_malloc(sizeof(SwsVector));
1514 vec->length = length;
1515 vec->coeff = av_malloc(sizeof(double) * length);
1521 SwsVector *sws_getGaussianVec(double variance, double quality)
1523 const int length = (int)(variance * quality + 0.5) | 1;
1525 double middle = (length - 1) * 0.5;
1528 if(variance < 0 || quality < 0)
1531 vec = sws_allocVec(length);
1536 for (i = 0; i < length; i++) {
1537 double dist = i - middle;
1538 vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /
1539 sqrt(2 * variance * M_PI);
1542 sws_normalizeVec(vec, 1.0);
1547 SwsVector *sws_getConstVec(double c, int length)
1550 SwsVector *vec = sws_allocVec(length);
1555 for (i = 0; i < length; i++)
1561 SwsVector *sws_getIdentityVec(void)
1563 return sws_getConstVec(1.0, 1);
1566 static double sws_dcVec(SwsVector *a)
1571 for (i = 0; i < a->length; i++)
1577 void sws_scaleVec(SwsVector *a, double scalar)
1581 for (i = 0; i < a->length; i++)
1582 a->coeff[i] *= scalar;
1585 void sws_normalizeVec(SwsVector *a, double height)
1587 sws_scaleVec(a, height / sws_dcVec(a));
1590 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
1592 int length = a->length + b->length - 1;
1594 SwsVector *vec = sws_getConstVec(0.0, length);
1599 for (i = 0; i < a->length; i++) {
1600 for (j = 0; j < b->length; j++) {
1601 vec->coeff[i + j] += a->coeff[i] * b->coeff[j];
1608 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
1610 int length = FFMAX(a->length, b->length);
1612 SwsVector *vec = sws_getConstVec(0.0, length);
1617 for (i = 0; i < a->length; i++)
1618 vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
1619 for (i = 0; i < b->length; i++)
1620 vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] += b->coeff[i];
1625 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
1627 int length = FFMAX(a->length, b->length);
1629 SwsVector *vec = sws_getConstVec(0.0, length);
1634 for (i = 0; i < a->length; i++)
1635 vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
1636 for (i = 0; i < b->length; i++)
1637 vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] -= b->coeff[i];
1642 /* shift left / or right if "shift" is negative */
1643 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
1645 int length = a->length + FFABS(shift) * 2;
1647 SwsVector *vec = sws_getConstVec(0.0, length);
1652 for (i = 0; i < a->length; i++) {
1653 vec->coeff[i + (length - 1) / 2 -
1654 (a->length - 1) / 2 - shift] = a->coeff[i];
1660 void sws_shiftVec(SwsVector *a, int shift)
1662 SwsVector *shifted = sws_getShiftedVec(a, shift);
1664 a->coeff = shifted->coeff;
1665 a->length = shifted->length;
1669 void sws_addVec(SwsVector *a, SwsVector *b)
1671 SwsVector *sum = sws_sumVec(a, b);
1673 a->coeff = sum->coeff;
1674 a->length = sum->length;
1678 void sws_subVec(SwsVector *a, SwsVector *b)
1680 SwsVector *diff = sws_diffVec(a, b);
1682 a->coeff = diff->coeff;
1683 a->length = diff->length;
1687 void sws_convVec(SwsVector *a, SwsVector *b)
1689 SwsVector *conv = sws_getConvVec(a, b);
1691 a->coeff = conv->coeff;
1692 a->length = conv->length;
1696 SwsVector *sws_cloneVec(SwsVector *a)
1699 SwsVector *vec = sws_allocVec(a->length);
1704 for (i = 0; i < a->length; i++)
1705 vec->coeff[i] = a->coeff[i];
1710 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
1717 for (i = 0; i < a->length; i++)
1718 if (a->coeff[i] > max)
1721 for (i = 0; i < a->length; i++)
1722 if (a->coeff[i] < min)
1727 for (i = 0; i < a->length; i++) {
1728 int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);
1729 av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
1731 av_log(log_ctx, log_level, " ");
1732 av_log(log_ctx, log_level, "|\n");
1736 void sws_freeVec(SwsVector *a)
1740 av_freep(&a->coeff);
1745 void sws_freeFilter(SwsFilter *filter)
1751 sws_freeVec(filter->lumH);
1753 sws_freeVec(filter->lumV);
1755 sws_freeVec(filter->chrH);
1757 sws_freeVec(filter->chrV);
1761 void sws_freeContext(SwsContext *c)
1768 for (i = 0; i < c->vLumBufSize; i++)
1769 av_freep(&c->lumPixBuf[i]);
1770 av_freep(&c->lumPixBuf);
1773 if (c->chrUPixBuf) {
1774 for (i = 0; i < c->vChrBufSize; i++)
1775 av_freep(&c->chrUPixBuf[i]);
1776 av_freep(&c->chrUPixBuf);
1777 av_freep(&c->chrVPixBuf);
1780 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1781 for (i = 0; i < c->vLumBufSize; i++)
1782 av_freep(&c->alpPixBuf[i]);
1783 av_freep(&c->alpPixBuf);
1786 for (i = 0; i < 4; i++)
1787 av_freep(&c->dither_error[i]);
1789 av_freep(&c->vLumFilter);
1790 av_freep(&c->vChrFilter);
1791 av_freep(&c->hLumFilter);
1792 av_freep(&c->hChrFilter);
1794 av_freep(&c->vYCoeffsBank);
1795 av_freep(&c->vCCoeffsBank);
1798 av_freep(&c->vLumFilterPos);
1799 av_freep(&c->vChrFilterPos);
1800 av_freep(&c->hLumFilterPos);
1801 av_freep(&c->hChrFilterPos);
1805 if (c->lumMmxextFilterCode)
1806 munmap(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize);
1807 if (c->chrMmxextFilterCode)
1808 munmap(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize);
1809 #elif HAVE_VIRTUALALLOC
1810 if (c->lumMmxextFilterCode)
1811 VirtualFree(c->lumMmxextFilterCode, 0, MEM_RELEASE);
1812 if (c->chrMmxextFilterCode)
1813 VirtualFree(c->chrMmxextFilterCode, 0, MEM_RELEASE);
1815 av_free(c->lumMmxextFilterCode);
1816 av_free(c->chrMmxextFilterCode);
1818 c->lumMmxextFilterCode = NULL;
1819 c->chrMmxextFilterCode = NULL;
1820 #endif /* HAVE_MMX_INLINE */
1822 av_freep(&c->yuvTable);
1823 av_freep(&c->formatConvBuffer);
1828 struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,
1829 int srcH, enum AVPixelFormat srcFormat,
1831 enum AVPixelFormat dstFormat, int flags,
1832 SwsFilter *srcFilter,
1833 SwsFilter *dstFilter,
1834 const double *param)
1836 static const double default_param[2] = { SWS_PARAM_DEFAULT,
1837 SWS_PARAM_DEFAULT };
1840 param = default_param;
1843 (context->srcW != srcW ||
1844 context->srcH != srcH ||
1845 context->srcFormat != srcFormat ||
1846 context->dstW != dstW ||
1847 context->dstH != dstH ||
1848 context->dstFormat != dstFormat ||
1849 context->flags != flags ||
1850 context->param[0] != param[0] ||
1851 context->param[1] != param[1])) {
1852 sws_freeContext(context);
1857 if (!(context = sws_alloc_context()))
1859 context->srcW = srcW;
1860 context->srcH = srcH;
1861 context->srcRange = handle_jpeg(&srcFormat);
1862 context->src0Alpha = handle_0alpha(&srcFormat);
1863 context->srcFormat = srcFormat;
1864 context->dstW = dstW;
1865 context->dstH = dstH;
1866 context->dstRange = handle_jpeg(&dstFormat);
1867 context->dst0Alpha = handle_0alpha(&dstFormat);
1868 context->dstFormat = dstFormat;
1869 context->flags = flags;
1870 context->param[0] = param[0];
1871 context->param[1] = param[1];
1872 sws_setColorspaceDetails(context, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],
1874 ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,
1875 context->dstRange, 0, 1 << 16, 1 << 16);
1876 if (sws_init_context(context, srcFilter, dstFilter) < 0) {
1877 sws_freeContext(context);