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
32 #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
33 #define MAP_ANONYMOUS MAP_ANON
37 #define WIN32_LEAN_AND_MEAN
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_cpu.h"
52 #include "swscale_internal.h"
54 unsigned swscale_version(void)
56 av_assert0(LIBSWSCALE_VERSION_MICRO >= 100);
57 return LIBSWSCALE_VERSION_INT;
60 const char *swscale_configuration(void)
62 return FFMPEG_CONFIGURATION;
65 const char *swscale_license(void)
67 #define LICENSE_PREFIX "libswscale license: "
68 return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
71 #define RET 0xC3 // near return opcode for x86
73 typedef struct FormatEntry {
74 int is_supported_in, is_supported_out;
77 static const FormatEntry format_entries[PIX_FMT_NB] = {
78 [PIX_FMT_YUV420P] = { 1, 1 },
79 [PIX_FMT_YUYV422] = { 1, 1 },
80 [PIX_FMT_RGB24] = { 1, 1 },
81 [PIX_FMT_BGR24] = { 1, 1 },
82 [PIX_FMT_YUV422P] = { 1, 1 },
83 [PIX_FMT_YUV444P] = { 1, 1 },
84 [PIX_FMT_YUV410P] = { 1, 1 },
85 [PIX_FMT_YUV411P] = { 1, 1 },
86 [PIX_FMT_GRAY8] = { 1, 1 },
87 [PIX_FMT_MONOWHITE] = { 1, 1 },
88 [PIX_FMT_MONOBLACK] = { 1, 1 },
89 [PIX_FMT_PAL8] = { 1, 0 },
90 [PIX_FMT_YUVJ420P] = { 1, 1 },
91 [PIX_FMT_YUVJ422P] = { 1, 1 },
92 [PIX_FMT_YUVJ444P] = { 1, 1 },
93 [PIX_FMT_UYVY422] = { 1, 1 },
94 [PIX_FMT_UYYVYY411] = { 0, 0 },
95 [PIX_FMT_BGR8] = { 1, 1 },
96 [PIX_FMT_BGR4] = { 0, 1 },
97 [PIX_FMT_BGR4_BYTE] = { 1, 1 },
98 [PIX_FMT_RGB8] = { 1, 1 },
99 [PIX_FMT_RGB4] = { 0, 1 },
100 [PIX_FMT_RGB4_BYTE] = { 1, 1 },
101 [PIX_FMT_NV12] = { 1, 1 },
102 [PIX_FMT_NV21] = { 1, 1 },
103 [PIX_FMT_ARGB] = { 1, 1 },
104 [PIX_FMT_RGBA] = { 1, 1 },
105 [PIX_FMT_ABGR] = { 1, 1 },
106 [PIX_FMT_BGRA] = { 1, 1 },
107 [PIX_FMT_0RGB] = { 1, 1 },
108 [PIX_FMT_RGB0] = { 1, 1 },
109 [PIX_FMT_0BGR] = { 1, 1 },
110 [PIX_FMT_BGR0] = { 1, 1 },
111 [PIX_FMT_GRAY16BE] = { 1, 1 },
112 [PIX_FMT_GRAY16LE] = { 1, 1 },
113 [PIX_FMT_YUV440P] = { 1, 1 },
114 [PIX_FMT_YUVJ440P] = { 1, 1 },
115 [PIX_FMT_YUVA420P] = { 1, 1 },
116 [PIX_FMT_YUVA422P] = { 1, 1 },
117 [PIX_FMT_YUVA444P] = { 1, 1 },
118 [PIX_FMT_RGB48BE] = { 1, 1 },
119 [PIX_FMT_RGB48LE] = { 1, 1 },
120 [PIX_FMT_RGBA64BE] = { 1, 0 },
121 [PIX_FMT_RGBA64LE] = { 1, 0 },
122 [PIX_FMT_RGB565BE] = { 1, 1 },
123 [PIX_FMT_RGB565LE] = { 1, 1 },
124 [PIX_FMT_RGB555BE] = { 1, 1 },
125 [PIX_FMT_RGB555LE] = { 1, 1 },
126 [PIX_FMT_BGR565BE] = { 1, 1 },
127 [PIX_FMT_BGR565LE] = { 1, 1 },
128 [PIX_FMT_BGR555BE] = { 1, 1 },
129 [PIX_FMT_BGR555LE] = { 1, 1 },
130 [PIX_FMT_YUV420P16LE] = { 1, 1 },
131 [PIX_FMT_YUV420P16BE] = { 1, 1 },
132 [PIX_FMT_YUV422P16LE] = { 1, 1 },
133 [PIX_FMT_YUV422P16BE] = { 1, 1 },
134 [PIX_FMT_YUV444P16LE] = { 1, 1 },
135 [PIX_FMT_YUV444P16BE] = { 1, 1 },
136 [PIX_FMT_RGB444LE] = { 1, 1 },
137 [PIX_FMT_RGB444BE] = { 1, 1 },
138 [PIX_FMT_BGR444LE] = { 1, 1 },
139 [PIX_FMT_BGR444BE] = { 1, 1 },
140 [PIX_FMT_Y400A] = { 1, 0 },
141 [PIX_FMT_BGR48BE] = { 1, 1 },
142 [PIX_FMT_BGR48LE] = { 1, 1 },
143 [PIX_FMT_BGRA64BE] = { 0, 0 },
144 [PIX_FMT_BGRA64LE] = { 0, 0 },
145 [PIX_FMT_YUV420P9BE] = { 1, 1 },
146 [PIX_FMT_YUV420P9LE] = { 1, 1 },
147 [PIX_FMT_YUV420P10BE] = { 1, 1 },
148 [PIX_FMT_YUV420P10LE] = { 1, 1 },
149 [PIX_FMT_YUV422P9BE] = { 1, 1 },
150 [PIX_FMT_YUV422P9LE] = { 1, 1 },
151 [PIX_FMT_YUV422P10BE] = { 1, 1 },
152 [PIX_FMT_YUV422P10LE] = { 1, 1 },
153 [PIX_FMT_YUV444P9BE] = { 1, 1 },
154 [PIX_FMT_YUV444P9LE] = { 1, 1 },
155 [PIX_FMT_YUV444P10BE] = { 1, 1 },
156 [PIX_FMT_YUV444P10LE] = { 1, 1 },
157 [PIX_FMT_GBRP] = { 1, 0 },
158 [PIX_FMT_GBRP9LE] = { 1, 0 },
159 [PIX_FMT_GBRP9BE] = { 1, 0 },
160 [PIX_FMT_GBRP10LE] = { 1, 0 },
161 [PIX_FMT_GBRP10BE] = { 1, 0 },
162 [PIX_FMT_GBRP16LE] = { 1, 0 },
163 [PIX_FMT_GBRP16BE] = { 1, 0 },
166 int sws_isSupportedInput(enum PixelFormat pix_fmt)
168 return (unsigned)pix_fmt < PIX_FMT_NB ?
169 format_entries[pix_fmt].is_supported_in : 0;
172 int sws_isSupportedOutput(enum PixelFormat pix_fmt)
174 return (unsigned)pix_fmt < PIX_FMT_NB ?
175 format_entries[pix_fmt].is_supported_out : 0;
178 extern const int32_t ff_yuv2rgb_coeffs[8][4];
180 #if FF_API_SWS_FORMAT_NAME
181 const char *sws_format_name(enum PixelFormat format)
183 return av_get_pix_fmt_name(format);
187 static double getSplineCoeff(double a, double b, double c, double d,
191 return ((d * dist + c) * dist + b) * dist + a;
193 return getSplineCoeff(0.0,
194 b + 2.0 * c + 3.0 * d,
196 -b - 3.0 * c - 6.0 * d,
200 static int initFilter(int16_t **outFilter, int32_t **filterPos,
201 int *outFilterSize, int xInc, int srcW, int dstW,
202 int filterAlign, int one, int flags, int cpu_flags,
203 SwsVector *srcFilter, SwsVector *dstFilter,
210 int64_t *filter = NULL;
211 int64_t *filter2 = NULL;
212 const int64_t fone = 1LL << 54;
215 emms_c(); // FIXME should not be required but IS (even for non-MMX versions)
217 // NOTE: the +3 is for the MMX(+1) / SSE(+3) scaler which reads over the end
218 FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW + 3) * sizeof(**filterPos), fail);
220 if (FFABS(xInc - 0x10000) < 10) { // unscaled
223 FF_ALLOCZ_OR_GOTO(NULL, filter,
224 dstW * sizeof(*filter) * filterSize, fail);
226 for (i = 0; i < dstW; i++) {
227 filter[i * filterSize] = fone;
230 } else if (flags & SWS_POINT) { // lame looking point sampling mode
234 FF_ALLOC_OR_GOTO(NULL, filter,
235 dstW * sizeof(*filter) * filterSize, fail);
237 xDstInSrc = xInc / 2 - 0x8000;
238 for (i = 0; i < dstW; i++) {
239 int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;
241 (*filterPos)[i] = xx;
245 } else if ((xInc <= (1 << 16) && (flags & SWS_AREA)) ||
246 (flags & SWS_FAST_BILINEAR)) { // bilinear upscale
250 FF_ALLOC_OR_GOTO(NULL, filter,
251 dstW * sizeof(*filter) * filterSize, fail);
253 xDstInSrc = xInc / 2 - 0x8000;
254 for (i = 0; i < dstW; i++) {
255 int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;
258 (*filterPos)[i] = xx;
259 // bilinear upscale / linear interpolate / area averaging
260 for (j = 0; j < filterSize; j++) {
261 int64_t coeff= fone - FFABS(((int64_t)xx<<16) - xDstInSrc)*(fone>>16);
264 filter[i * filterSize + j] = coeff;
273 if (flags & SWS_BICUBIC)
275 else if (flags & SWS_X)
277 else if (flags & SWS_AREA)
278 sizeFactor = 1; // downscale only, for upscale it is bilinear
279 else if (flags & SWS_GAUSS)
280 sizeFactor = 8; // infinite ;)
281 else if (flags & SWS_LANCZOS)
282 sizeFactor = param[0] != SWS_PARAM_DEFAULT ? ceil(2 * param[0]) : 6;
283 else if (flags & SWS_SINC)
284 sizeFactor = 20; // infinite ;)
285 else if (flags & SWS_SPLINE)
286 sizeFactor = 20; // infinite ;)
287 else if (flags & SWS_BILINEAR)
290 sizeFactor = 0; // GCC warning killer
295 filterSize = 1 + sizeFactor; // upscale
297 filterSize = 1 + (sizeFactor * srcW + dstW - 1) / dstW;
299 filterSize = FFMIN(filterSize, srcW - 2);
300 filterSize = FFMAX(filterSize, 1);
302 FF_ALLOC_OR_GOTO(NULL, filter,
303 dstW * sizeof(*filter) * filterSize, fail);
305 xDstInSrc = xInc - 0x10000;
306 for (i = 0; i < dstW; i++) {
307 int xx = (xDstInSrc - ((filterSize - 2) << 16)) / (1 << 17);
309 (*filterPos)[i] = xx;
310 for (j = 0; j < filterSize; j++) {
311 int64_t d = (FFABS(((int64_t)xx << 17) - xDstInSrc)) << 13;
317 floatd = d * (1.0 / (1 << 30));
319 if (flags & SWS_BICUBIC) {
320 int64_t B = (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1 << 24);
321 int64_t C = (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1 << 24);
323 if (d >= 1LL << 31) {
326 int64_t dd = (d * d) >> 30;
327 int64_t ddd = (dd * d) >> 30;
330 coeff = (12 * (1 << 24) - 9 * B - 6 * C) * ddd +
331 (-18 * (1 << 24) + 12 * B + 6 * C) * dd +
332 (6 * (1 << 24) - 2 * B) * (1 << 30);
334 coeff = (-B - 6 * C) * ddd +
335 (6 * B + 30 * C) * dd +
336 (-12 * B - 48 * C) * d +
337 (8 * B + 24 * C) * (1 << 30);
339 coeff *= fone >> (30 + 24);
342 else if (flags & SWS_X) {
343 double p = param ? param * 0.01 : 0.3;
344 coeff = d ? sin(d * M_PI) / (d * M_PI) : 1.0;
345 coeff *= pow(2.0, -p * d * d);
348 else if (flags & SWS_X) {
349 double A = param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
353 c = cos(floatd * M_PI);
360 coeff = (c * 0.5 + 0.5) * fone;
361 } else if (flags & SWS_AREA) {
362 int64_t d2 = d - (1 << 29);
363 if (d2 * xInc < -(1LL << (29 + 16)))
364 coeff = 1.0 * (1LL << (30 + 16));
365 else if (d2 * xInc < (1LL << (29 + 16)))
366 coeff = -d2 * xInc + (1LL << (29 + 16));
369 coeff *= fone >> (30 + 16);
370 } else if (flags & SWS_GAUSS) {
371 double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
372 coeff = (pow(2.0, -p * floatd * floatd)) * fone;
373 } else if (flags & SWS_SINC) {
374 coeff = (d ? sin(floatd * M_PI) / (floatd * M_PI) : 1.0) * fone;
375 } else if (flags & SWS_LANCZOS) {
376 double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
377 coeff = (d ? sin(floatd * M_PI) * sin(floatd * M_PI / p) /
378 (floatd * floatd * M_PI * M_PI / p) : 1.0) * fone;
381 } else if (flags & SWS_BILINEAR) {
382 coeff = (1 << 30) - d;
386 } else if (flags & SWS_SPLINE) {
387 double p = -2.196152422706632;
388 coeff = getSplineCoeff(1.0, 0.0, p, -p - 1.0, floatd) * fone;
390 coeff = 0.0; // GCC warning killer
394 filter[i * filterSize + j] = coeff;
397 xDstInSrc += 2 * xInc;
401 /* apply src & dst Filter to filter -> filter2
404 assert(filterSize > 0);
405 filter2Size = filterSize;
407 filter2Size += srcFilter->length - 1;
409 filter2Size += dstFilter->length - 1;
410 assert(filter2Size > 0);
411 FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size * dstW * sizeof(*filter2), fail);
413 for (i = 0; i < dstW; i++) {
417 for (k = 0; k < srcFilter->length; k++) {
418 for (j = 0; j < filterSize; j++)
419 filter2[i * filter2Size + k + j] +=
420 srcFilter->coeff[k] * filter[i * filterSize + j];
423 for (j = 0; j < filterSize; j++)
424 filter2[i * filter2Size + j] = filter[i * filterSize + j];
428 (*filterPos)[i] += (filterSize - 1) / 2 - (filter2Size - 1) / 2;
432 /* try to reduce the filter-size (step1 find size and shift left) */
433 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
435 for (i = dstW - 1; i >= 0; i--) {
436 int min = filter2Size;
438 int64_t cutOff = 0.0;
440 /* get rid of near zero elements on the left by shifting left */
441 for (j = 0; j < filter2Size; j++) {
443 cutOff += FFABS(filter2[i * filter2Size]);
445 if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)
448 /* preserve monotonicity because the core can't handle the
449 * filter otherwise */
450 if (i < dstW - 1 && (*filterPos)[i] >= (*filterPos)[i + 1])
453 // move filter coefficients left
454 for (k = 1; k < filter2Size; k++)
455 filter2[i * filter2Size + k - 1] = filter2[i * filter2Size + k];
456 filter2[i * filter2Size + k - 1] = 0;
461 /* count near zeros on the right */
462 for (j = filter2Size - 1; j > 0; j--) {
463 cutOff += FFABS(filter2[i * filter2Size + j]);
465 if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)
470 if (min > minFilterSize)
474 if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) {
475 // we can handle the special case 4, so we don't want to go the full 8
476 if (minFilterSize < 5)
479 /* We really don't want to waste our time doing useless computation, so
480 * fall back on the scalar C code for very small filters.
481 * Vectorizing is worth it only if you have a decent-sized vector. */
482 if (minFilterSize < 3)
486 if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
487 // special case for unscaled vertical filtering
488 if (minFilterSize == 1 && filterAlign == 2)
492 assert(minFilterSize > 0);
493 filterSize = (minFilterSize + (filterAlign - 1)) & (~(filterAlign - 1));
494 assert(filterSize > 0);
495 filter = av_malloc(filterSize * dstW * sizeof(*filter));
496 if (filterSize >= MAX_FILTER_SIZE * 16 /
497 ((flags & SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
499 *outFilterSize = filterSize;
501 if (flags & SWS_PRINT_INFO)
502 av_log(NULL, AV_LOG_VERBOSE,
503 "SwScaler: reducing / aligning filtersize %d -> %d\n",
504 filter2Size, filterSize);
505 /* try to reduce the filter-size (step2 reduce it) */
506 for (i = 0; i < dstW; i++) {
509 for (j = 0; j < filterSize; j++) {
510 if (j >= filter2Size)
511 filter[i * filterSize + j] = 0;
513 filter[i * filterSize + j] = filter2[i * filter2Size + j];
514 if ((flags & SWS_BITEXACT) && j >= minFilterSize)
515 filter[i * filterSize + j] = 0;
519 // FIXME try to align filterPos if possible
522 for (i = 0; i < dstW; i++) {
524 if ((*filterPos)[i] < 0) {
525 // move filter coefficients left to compensate for filterPos
526 for (j = 1; j < filterSize; j++) {
527 int left = FFMAX(j + (*filterPos)[i], 0);
528 filter[i * filterSize + left] += filter[i * filterSize + j];
529 filter[i * filterSize + j] = 0;
534 if ((*filterPos)[i] + filterSize > srcW) {
535 int shift = (*filterPos)[i] + filterSize - srcW;
536 // move filter coefficients right to compensate for filterPos
537 for (j = filterSize - 2; j >= 0; j--) {
538 int right = FFMIN(j + shift, filterSize - 1);
539 filter[i * filterSize + right] += filter[i * filterSize + j];
540 filter[i * filterSize + j] = 0;
542 (*filterPos)[i]= srcW - filterSize;
546 // Note the +1 is for the MMX scaler which reads over the end
547 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
548 FF_ALLOCZ_OR_GOTO(NULL, *outFilter,
549 *outFilterSize * (dstW + 3) * sizeof(int16_t), fail);
551 /* normalize & store in outFilter */
552 for (i = 0; i < dstW; i++) {
557 for (j = 0; j < filterSize; j++) {
558 sum += filter[i * filterSize + j];
560 sum = (sum + one / 2) / one;
561 for (j = 0; j < *outFilterSize; j++) {
562 int64_t v = filter[i * filterSize + j] + error;
563 int intV = ROUNDED_DIV(v, sum);
564 (*outFilter)[i * (*outFilterSize) + j] = intV;
565 error = v - intV * sum;
569 (*filterPos)[dstW + 0] =
570 (*filterPos)[dstW + 1] =
571 (*filterPos)[dstW + 2] = (*filterPos)[dstW - 1]; /* the MMX/SSE scaler will
572 * read over the end */
573 for (i = 0; i < *outFilterSize; i++) {
574 int k = (dstW - 1) * (*outFilterSize) + i;
575 (*outFilter)[k + 1 * (*outFilterSize)] =
576 (*outFilter)[k + 2 * (*outFilterSize)] =
577 (*outFilter)[k + 3 * (*outFilterSize)] = (*outFilter)[k];
589 static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode,
590 int16_t *filter, int32_t *filterPos, int numSplits)
593 x86_reg imm8OfPShufW1A;
594 x86_reg imm8OfPShufW2A;
595 x86_reg fragmentLengthA;
597 x86_reg imm8OfPShufW1B;
598 x86_reg imm8OfPShufW2B;
599 x86_reg fragmentLengthB;
604 // create an optimized horizontal scaling routine
605 /* This scaler is made of runtime-generated MMX2 code using specially tuned
606 * pshufw instructions. For every four output pixels, if four input pixels
607 * are enough for the fast bilinear scaling, then a chunk of fragmentB is
608 * used. If five input pixels are needed, then a chunk of fragmentA is used.
617 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
618 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
619 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
620 "punpcklbw %%mm7, %%mm1 \n\t"
621 "punpcklbw %%mm7, %%mm0 \n\t"
622 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
624 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
626 "psubw %%mm1, %%mm0 \n\t"
627 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
628 "pmullw %%mm3, %%mm0 \n\t"
629 "psllw $7, %%mm1 \n\t"
630 "paddw %%mm1, %%mm0 \n\t"
632 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
634 "add $8, %%"REG_a" \n\t"
638 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
639 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
640 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
645 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
649 : "=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
650 "=r" (fragmentLengthA)
657 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
658 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
659 "punpcklbw %%mm7, %%mm0 \n\t"
660 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
662 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
664 "psubw %%mm1, %%mm0 \n\t"
665 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
666 "pmullw %%mm3, %%mm0 \n\t"
667 "psllw $7, %%mm1 \n\t"
668 "paddw %%mm1, %%mm0 \n\t"
670 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
672 "add $8, %%"REG_a" \n\t"
676 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
677 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
678 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
683 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
687 : "=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
688 "=r" (fragmentLengthB)
691 xpos = 0; // lumXInc/2 - 0x8000; // difference between pixel centers
694 for (i = 0; i < dstW / numSplits; i++) {
699 int b = ((xpos + xInc) >> 16) - xx;
700 int c = ((xpos + xInc * 2) >> 16) - xx;
701 int d = ((xpos + xInc * 3) >> 16) - xx;
702 int inc = (d + 1 < 4);
703 uint8_t *fragment = (d + 1 < 4) ? fragmentB : fragmentA;
704 x86_reg imm8OfPShufW1 = (d + 1 < 4) ? imm8OfPShufW1B : imm8OfPShufW1A;
705 x86_reg imm8OfPShufW2 = (d + 1 < 4) ? imm8OfPShufW2B : imm8OfPShufW2A;
706 x86_reg fragmentLength = (d + 1 < 4) ? fragmentLengthB : fragmentLengthA;
707 int maxShift = 3 - (d + inc);
711 filter[i] = ((xpos & 0xFFFF) ^ 0xFFFF) >> 9;
712 filter[i + 1] = (((xpos + xInc) & 0xFFFF) ^ 0xFFFF) >> 9;
713 filter[i + 2] = (((xpos + xInc * 2) & 0xFFFF) ^ 0xFFFF) >> 9;
714 filter[i + 3] = (((xpos + xInc * 3) & 0xFFFF) ^ 0xFFFF) >> 9;
715 filterPos[i / 2] = xx;
717 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
719 filterCode[fragmentPos + imm8OfPShufW1] = (a + inc) |
723 filterCode[fragmentPos + imm8OfPShufW2] = a | (b << 2) |
727 if (i + 4 - inc >= dstW)
728 shift = maxShift; // avoid overread
729 else if ((filterPos[i / 2] & 3) <= maxShift)
730 shift = filterPos[i / 2] & 3; // align
732 if (shift && i >= shift) {
733 filterCode[fragmentPos + imm8OfPShufW1] += 0x55 * shift;
734 filterCode[fragmentPos + imm8OfPShufW2] += 0x55 * shift;
735 filterPos[i / 2] -= shift;
739 fragmentPos += fragmentLength;
742 filterCode[fragmentPos] = RET;
747 filterPos[((i / 2) + 1) & (~1)] = xpos >> 16; // needed to jump to the next part
749 return fragmentPos + 1;
751 #endif /* HAVE_MMX2 */
753 static void getSubSampleFactors(int *h, int *v, enum PixelFormat format)
755 *h = av_pix_fmt_descriptors[format].log2_chroma_w;
756 *v = av_pix_fmt_descriptors[format].log2_chroma_h;
759 int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],
760 int srcRange, const int table[4], int dstRange,
761 int brightness, int contrast, int saturation)
763 memcpy(c->srcColorspaceTable, inv_table, sizeof(int) * 4);
764 memcpy(c->dstColorspaceTable, table, sizeof(int) * 4);
766 c->brightness = brightness;
767 c->contrast = contrast;
768 c->saturation = saturation;
769 c->srcRange = srcRange;
770 c->dstRange = dstRange;
771 if (isYUV(c->dstFormat) || isGray(c->dstFormat))
774 c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->dstFormat]);
775 c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->srcFormat]);
777 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness,
778 contrast, saturation);
781 if (HAVE_ALTIVEC && av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)
782 ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness,
783 contrast, saturation);
787 int sws_getColorspaceDetails(struct SwsContext *c, int **inv_table,
788 int *srcRange, int **table, int *dstRange,
789 int *brightness, int *contrast, int *saturation)
791 if (!c || isYUV(c->dstFormat) || isGray(c->dstFormat))
794 *inv_table = c->srcColorspaceTable;
795 *table = c->dstColorspaceTable;
796 *srcRange = c->srcRange;
797 *dstRange = c->dstRange;
798 *brightness = c->brightness;
799 *contrast = c->contrast;
800 *saturation = c->saturation;
805 static int handle_jpeg(enum PixelFormat *format)
808 case PIX_FMT_YUVJ420P:
809 *format = PIX_FMT_YUV420P;
811 case PIX_FMT_YUVJ422P:
812 *format = PIX_FMT_YUV422P;
814 case PIX_FMT_YUVJ444P:
815 *format = PIX_FMT_YUV444P;
817 case PIX_FMT_YUVJ440P:
818 *format = PIX_FMT_YUV440P;
825 static int handle_0alpha(enum PixelFormat *format)
828 case PIX_FMT_0BGR : *format = PIX_FMT_ABGR ; return 1;
829 case PIX_FMT_BGR0 : *format = PIX_FMT_BGRA ; return 4;
830 case PIX_FMT_0RGB : *format = PIX_FMT_ARGB ; return 1;
831 case PIX_FMT_RGB0 : *format = PIX_FMT_RGBA ; return 4;
836 SwsContext *sws_alloc_context(void)
838 SwsContext *c = av_mallocz(sizeof(SwsContext));
840 c->av_class = &sws_context_class;
841 av_opt_set_defaults(c);
846 int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)
849 int usesVFilter, usesHFilter;
851 SwsFilter dummyFilter = { NULL, NULL, NULL, NULL };
856 int dst_stride = FFALIGN(dstW * sizeof(int16_t) + 66, 16);
857 int flags, cpu_flags;
858 enum PixelFormat srcFormat = c->srcFormat;
859 enum PixelFormat dstFormat = c->dstFormat;
861 cpu_flags = av_get_cpu_flags();
867 unscaled = (srcW == dstW && srcH == dstH);
869 handle_jpeg(&srcFormat);
870 handle_jpeg(&dstFormat);
871 handle_0alpha(&srcFormat);
872 handle_0alpha(&dstFormat);
874 if(srcFormat!=c->srcFormat || dstFormat!=c->dstFormat){
875 av_log(c, AV_LOG_WARNING, "deprecated pixel format used, make sure you did set range correctly\n");
876 c->srcFormat= srcFormat;
877 c->dstFormat= dstFormat;
880 if (!sws_isSupportedInput(srcFormat)) {
881 av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n",
882 av_get_pix_fmt_name(srcFormat));
883 return AVERROR(EINVAL);
885 if (!sws_isSupportedOutput(dstFormat)) {
886 av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n",
887 av_get_pix_fmt_name(dstFormat));
888 return AVERROR(EINVAL);
891 i = flags & (SWS_POINT |
902 if (!i || (i & (i - 1))) {
903 av_log(c, AV_LOG_ERROR, "Exactly one scaler algorithm must be chosen, got %X\n", i);
904 return AVERROR(EINVAL);
907 if (srcW < 4 || srcH < 1 || dstW < 8 || dstH < 1) {
908 /* FIXME check if these are enough and try to lower them after
909 * fixing the relevant parts of the code */
910 av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",
911 srcW, srcH, dstW, dstH);
912 return AVERROR(EINVAL);
916 dstFilter = &dummyFilter;
918 srcFilter = &dummyFilter;
920 c->lumXInc = (((int64_t)srcW << 16) + (dstW >> 1)) / dstW;
921 c->lumYInc = (((int64_t)srcH << 16) + (dstH >> 1)) / dstH;
922 c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[dstFormat]);
923 c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[srcFormat]);
924 c->vRounder = 4 * 0x0001000100010001ULL;
926 usesVFilter = (srcFilter->lumV && srcFilter->lumV->length > 1) ||
927 (srcFilter->chrV && srcFilter->chrV->length > 1) ||
928 (dstFilter->lumV && dstFilter->lumV->length > 1) ||
929 (dstFilter->chrV && dstFilter->chrV->length > 1);
930 usesHFilter = (srcFilter->lumH && srcFilter->lumH->length > 1) ||
931 (srcFilter->chrH && srcFilter->chrH->length > 1) ||
932 (dstFilter->lumH && dstFilter->lumH->length > 1) ||
933 (dstFilter->chrH && dstFilter->chrH->length > 1);
935 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
936 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
939 if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) {
941 av_log(c, AV_LOG_DEBUG, "Forcing full internal H chroma due to odd output size\n");
942 flags |= SWS_FULL_CHR_H_INT;
946 /* reuse chroma for 2 pixels RGB/BGR unless user wants full
947 * chroma interpolation */
948 if (flags & SWS_FULL_CHR_H_INT &&
949 isAnyRGB(dstFormat) &&
950 dstFormat != PIX_FMT_RGBA &&
951 dstFormat != PIX_FMT_ARGB &&
952 dstFormat != PIX_FMT_BGRA &&
953 dstFormat != PIX_FMT_ABGR &&
954 dstFormat != PIX_FMT_RGB24 &&
955 dstFormat != PIX_FMT_BGR24) {
956 av_log(c, AV_LOG_WARNING,
957 "full chroma interpolation for destination format '%s' not yet implemented\n",
958 av_get_pix_fmt_name(dstFormat));
959 flags &= ~SWS_FULL_CHR_H_INT;
962 if (isAnyRGB(dstFormat) && !(flags & SWS_FULL_CHR_H_INT))
963 c->chrDstHSubSample = 1;
965 // drop some chroma lines if the user wants it
966 c->vChrDrop = (flags & SWS_SRC_V_CHR_DROP_MASK) >>
967 SWS_SRC_V_CHR_DROP_SHIFT;
968 c->chrSrcVSubSample += c->vChrDrop;
970 /* drop every other pixel for chroma calculation unless user
971 * wants full chroma */
972 if (isAnyRGB(srcFormat) && !(flags & SWS_FULL_CHR_H_INP) &&
973 srcFormat != PIX_FMT_RGB8 && srcFormat != PIX_FMT_BGR8 &&
974 srcFormat != PIX_FMT_RGB4 && srcFormat != PIX_FMT_BGR4 &&
975 srcFormat != PIX_FMT_RGB4_BYTE && srcFormat != PIX_FMT_BGR4_BYTE &&
976 ((dstW >> c->chrDstHSubSample) <= (srcW >> 1) ||
977 (flags & SWS_FAST_BILINEAR)))
978 c->chrSrcHSubSample = 1;
980 // Note the -((-x)>>y) is so that we always round toward +inf.
981 c->chrSrcW = -((-srcW) >> c->chrSrcHSubSample);
982 c->chrSrcH = -((-srcH) >> c->chrSrcVSubSample);
983 c->chrDstW = -((-dstW) >> c->chrDstHSubSample);
984 c->chrDstH = -((-dstH) >> c->chrDstVSubSample);
986 /* unscaled special cases */
987 if (unscaled && !usesHFilter && !usesVFilter &&
988 (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
989 ff_get_unscaled_swscale(c);
992 if (flags & SWS_PRINT_INFO)
993 av_log(c, AV_LOG_INFO,
994 "using unscaled %s -> %s special converter\n",
995 av_get_pix_fmt_name(srcFormat), av_get_pix_fmt_name(dstFormat));
1000 c->srcBpc = 1 + av_pix_fmt_descriptors[srcFormat].comp[0].depth_minus1;
1003 c->dstBpc = 1 + av_pix_fmt_descriptors[dstFormat].comp[0].depth_minus1;
1006 if (isAnyRGB(srcFormat) || srcFormat == PIX_FMT_PAL8)
1008 if (c->dstBpc == 16)
1010 FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW*2+78, 16) * 2, fail);
1011 if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2 &&
1012 c->srcBpc == 8 && c->dstBpc <= 10) {
1013 c->canMMX2BeUsed = (dstW >= srcW && (dstW & 31) == 0 &&
1014 (srcW & 15) == 0) ? 1 : 0;
1015 if (!c->canMMX2BeUsed && dstW >= srcW && (srcW & 15) == 0
1016 && (flags & SWS_FAST_BILINEAR)) {
1017 if (flags & SWS_PRINT_INFO)
1018 av_log(c, AV_LOG_INFO,
1019 "output width is not a multiple of 32 -> no MMX2 scaler\n");
1021 if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat))
1024 c->canMMX2BeUsed = 0;
1026 c->chrXInc = (((int64_t)c->chrSrcW << 16) + (c->chrDstW >> 1)) / c->chrDstW;
1027 c->chrYInc = (((int64_t)c->chrSrcH << 16) + (c->chrDstH >> 1)) / c->chrDstH;
1029 /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src
1030 * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do
1032 * n-2 is the last chrominance sample available.
1033 * This is not perfect, but no one should notice the difference, the more
1034 * correct variant would be like the vertical one, but that would require
1035 * some special code for the first and last pixel */
1036 if (flags & SWS_FAST_BILINEAR) {
1037 if (c->canMMX2BeUsed) {
1041 // we don't use the x86 asm scaler if MMX is available
1042 else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX && c->dstBpc <= 10) {
1043 c->lumXInc = ((int64_t)(srcW - 2) << 16) / (dstW - 2) - 20;
1044 c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;
1048 /* precalculate horizontal scaler filter coefficients */
1051 // can't downscale !!!
1052 if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {
1053 c->lumMmx2FilterCodeSize = initMMX2HScaler(dstW, c->lumXInc, NULL,
1055 c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc,
1056 NULL, NULL, NULL, 4);
1058 #ifdef MAP_ANONYMOUS
1059 c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
1060 c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
1061 #elif HAVE_VIRTUALALLOC
1062 c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
1063 c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
1065 c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);
1066 c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);
1069 #ifdef MAP_ANONYMOUS
1070 if (c->lumMmx2FilterCode == MAP_FAILED || c->chrMmx2FilterCode == MAP_FAILED)
1072 if (!c->lumMmx2FilterCode || !c->chrMmx2FilterCode)
1075 av_log(c, AV_LOG_ERROR, "Failed to allocate MMX2FilterCode\n");
1076 return AVERROR(ENOMEM);
1079 FF_ALLOCZ_OR_GOTO(c, c->hLumFilter, (dstW / 8 + 8) * sizeof(int16_t), fail);
1080 FF_ALLOCZ_OR_GOTO(c, c->hChrFilter, (c->chrDstW / 4 + 8) * sizeof(int16_t), fail);
1081 FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW / 2 / 8 + 8) * sizeof(int32_t), fail);
1082 FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);
1084 initMMX2HScaler( dstW, c->lumXInc, c->lumMmx2FilterCode,
1085 c->hLumFilter, (uint32_t*)c->hLumFilterPos, 8);
1086 initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode,
1087 c->hChrFilter, (uint32_t*)c->hChrFilterPos, 4);
1089 #ifdef MAP_ANONYMOUS
1090 mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
1091 mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
1094 #endif /* HAVE_MMX2 */
1096 const int filterAlign =
1097 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :
1098 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
1101 if (initFilter(&c->hLumFilter, &c->hLumFilterPos,
1102 &c->hLumFilterSize, c->lumXInc,
1103 srcW, dstW, filterAlign, 1 << 14,
1104 (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
1105 cpu_flags, srcFilter->lumH, dstFilter->lumH,
1108 if (initFilter(&c->hChrFilter, &c->hChrFilterPos,
1109 &c->hChrFilterSize, c->chrXInc,
1110 c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,
1111 (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
1112 cpu_flags, srcFilter->chrH, dstFilter->chrH,
1116 } // initialize horizontal stuff
1118 /* precalculate vertical scaler filter coefficients */
1120 const int filterAlign =
1121 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 2 :
1122 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
1125 if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,
1126 c->lumYInc, srcH, dstH, filterAlign, (1 << 12),
1127 (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
1128 cpu_flags, srcFilter->lumV, dstFilter->lumV,
1131 if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,
1132 c->chrYInc, c->chrSrcH, c->chrDstH,
1133 filterAlign, (1 << 12),
1134 (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
1135 cpu_flags, srcFilter->chrV, dstFilter->chrV,
1140 FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH, fail);
1141 FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);
1143 for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {
1145 short *p = (short *)&c->vYCoeffsBank[i];
1146 for (j = 0; j < 8; j++)
1147 p[j] = c->vLumFilter[i];
1150 for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {
1152 short *p = (short *)&c->vCCoeffsBank[i];
1153 for (j = 0; j < 8; j++)
1154 p[j] = c->vChrFilter[i];
1159 // calculate buffer sizes so that they won't run out while handling these damn slices
1160 c->vLumBufSize = c->vLumFilterSize;
1161 c->vChrBufSize = c->vChrFilterSize;
1162 for (i = 0; i < dstH; i++) {
1163 int chrI = (int64_t)i * c->chrDstH / dstH;
1164 int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,
1165 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)
1166 << c->chrSrcVSubSample));
1168 nextSlice >>= c->chrSrcVSubSample;
1169 nextSlice <<= c->chrSrcVSubSample;
1170 if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)
1171 c->vLumBufSize = nextSlice - c->vLumFilterPos[i];
1172 if (c->vChrFilterPos[chrI] + c->vChrBufSize <
1173 (nextSlice >> c->chrSrcVSubSample))
1174 c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -
1175 c->vChrFilterPos[chrI];
1178 /* Allocate pixbufs (we use dynamic allocation because otherwise we would
1179 * need to allocate several megabytes to handle all possible cases) */
1180 FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);
1181 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
1182 FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
1183 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
1184 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);
1185 /* Note we need at least one pixel more at the end because of the MMX code
1186 * (just in case someone wants to replace the 4000/8000). */
1187 /* align at 16 bytes for AltiVec */
1188 for (i = 0; i < c->vLumBufSize; i++) {
1189 FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],
1190 dst_stride + 16, fail);
1191 c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];
1193 // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)
1194 c->uv_off = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);
1195 c->uv_offx2 = dst_stride + 16;
1196 for (i = 0; i < c->vChrBufSize; i++) {
1197 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],
1198 dst_stride * 2 + 32, fail);
1199 c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];
1200 c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]
1201 = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;
1203 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
1204 for (i = 0; i < c->vLumBufSize; i++) {
1205 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i + c->vLumBufSize],
1206 dst_stride + 16, fail);
1207 c->alpPixBuf[i] = c->alpPixBuf[i + c->vLumBufSize];
1210 // try to avoid drawing green stuff between the right end and the stride end
1211 for (i = 0; i < c->vChrBufSize; i++)
1212 if(av_pix_fmt_descriptors[c->dstFormat].comp[0].depth_minus1 == 15){
1213 av_assert0(c->dstBpc > 10);
1214 for(j=0; j<dst_stride/2+1; j++)
1215 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;
1217 for(j=0; j<dst_stride+1; j++)
1218 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;
1220 assert(c->chrDstH <= dstH);
1222 if (flags & SWS_PRINT_INFO) {
1223 if (flags & SWS_FAST_BILINEAR)
1224 av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
1225 else if (flags & SWS_BILINEAR)
1226 av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
1227 else if (flags & SWS_BICUBIC)
1228 av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
1229 else if (flags & SWS_X)
1230 av_log(c, AV_LOG_INFO, "Experimental scaler, ");
1231 else if (flags & SWS_POINT)
1232 av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
1233 else if (flags & SWS_AREA)
1234 av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
1235 else if (flags & SWS_BICUBLIN)
1236 av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
1237 else if (flags & SWS_GAUSS)
1238 av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
1239 else if (flags & SWS_SINC)
1240 av_log(c, AV_LOG_INFO, "Sinc scaler, ");
1241 else if (flags & SWS_LANCZOS)
1242 av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
1243 else if (flags & SWS_SPLINE)
1244 av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
1246 av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
1248 av_log(c, AV_LOG_INFO, "from %s to %s%s ",
1249 av_get_pix_fmt_name(srcFormat),
1251 dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ||
1252 dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||
1253 dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ?
1258 av_get_pix_fmt_name(dstFormat));
1260 if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2)
1261 av_log(c, AV_LOG_INFO, "using MMX2\n");
1262 else if (HAVE_AMD3DNOW && cpu_flags & AV_CPU_FLAG_3DNOW)
1263 av_log(c, AV_LOG_INFO, "using 3DNOW\n");
1264 else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX)
1265 av_log(c, AV_LOG_INFO, "using MMX\n");
1266 else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC)
1267 av_log(c, AV_LOG_INFO, "using AltiVec\n");
1269 av_log(c, AV_LOG_INFO, "using C\n");
1271 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
1272 av_log(c, AV_LOG_DEBUG,
1273 "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1274 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
1275 av_log(c, AV_LOG_DEBUG,
1276 "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1277 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,
1278 c->chrXInc, c->chrYInc);
1281 c->swScale = ff_getSwsFunc(c);
1283 fail: // FIXME replace things by appropriate error codes
1287 #if FF_API_SWS_GETCONTEXT
1288 SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,
1289 int dstW, int dstH, enum PixelFormat dstFormat,
1290 int flags, SwsFilter *srcFilter,
1291 SwsFilter *dstFilter, const double *param)
1295 if (!(c = sws_alloc_context()))
1303 c->srcRange = handle_jpeg(&srcFormat);
1304 c->dstRange = handle_jpeg(&dstFormat);
1305 c->src0Alpha = handle_0alpha(&srcFormat);
1306 c->dst0Alpha = handle_0alpha(&dstFormat);
1307 c->srcFormat = srcFormat;
1308 c->dstFormat = dstFormat;
1311 c->param[0] = param[0];
1312 c->param[1] = param[1];
1314 sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,
1315 ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,
1316 c->dstRange, 0, 1 << 16, 1 << 16);
1318 if (sws_init_context(c, srcFilter, dstFilter) < 0) {
1327 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
1328 float lumaSharpen, float chromaSharpen,
1329 float chromaHShift, float chromaVShift,
1332 SwsFilter *filter = av_malloc(sizeof(SwsFilter));
1336 if (lumaGBlur != 0.0) {
1337 filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);
1338 filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);
1340 filter->lumH = sws_getIdentityVec();
1341 filter->lumV = sws_getIdentityVec();
1344 if (chromaGBlur != 0.0) {
1345 filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);
1346 filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);
1348 filter->chrH = sws_getIdentityVec();
1349 filter->chrV = sws_getIdentityVec();
1352 if (chromaSharpen != 0.0) {
1353 SwsVector *id = sws_getIdentityVec();
1354 sws_scaleVec(filter->chrH, -chromaSharpen);
1355 sws_scaleVec(filter->chrV, -chromaSharpen);
1356 sws_addVec(filter->chrH, id);
1357 sws_addVec(filter->chrV, id);
1361 if (lumaSharpen != 0.0) {
1362 SwsVector *id = sws_getIdentityVec();
1363 sws_scaleVec(filter->lumH, -lumaSharpen);
1364 sws_scaleVec(filter->lumV, -lumaSharpen);
1365 sws_addVec(filter->lumH, id);
1366 sws_addVec(filter->lumV, id);
1370 if (chromaHShift != 0.0)
1371 sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));
1373 if (chromaVShift != 0.0)
1374 sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));
1376 sws_normalizeVec(filter->chrH, 1.0);
1377 sws_normalizeVec(filter->chrV, 1.0);
1378 sws_normalizeVec(filter->lumH, 1.0);
1379 sws_normalizeVec(filter->lumV, 1.0);
1382 sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
1384 sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
1389 SwsVector *sws_allocVec(int length)
1391 SwsVector *vec = av_malloc(sizeof(SwsVector));
1394 vec->length = length;
1395 vec->coeff = av_malloc(sizeof(double) * length);
1401 SwsVector *sws_getGaussianVec(double variance, double quality)
1403 const int length = (int)(variance * quality + 0.5) | 1;
1405 double middle = (length - 1) * 0.5;
1406 SwsVector *vec = sws_allocVec(length);
1411 for (i = 0; i < length; i++) {
1412 double dist = i - middle;
1413 vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /
1414 sqrt(2 * variance * M_PI);
1417 sws_normalizeVec(vec, 1.0);
1422 SwsVector *sws_getConstVec(double c, int length)
1425 SwsVector *vec = sws_allocVec(length);
1430 for (i = 0; i < length; i++)
1436 SwsVector *sws_getIdentityVec(void)
1438 return sws_getConstVec(1.0, 1);
1441 static double sws_dcVec(SwsVector *a)
1446 for (i = 0; i < a->length; i++)
1452 void sws_scaleVec(SwsVector *a, double scalar)
1456 for (i = 0; i < a->length; i++)
1457 a->coeff[i] *= scalar;
1460 void sws_normalizeVec(SwsVector *a, double height)
1462 sws_scaleVec(a, height / sws_dcVec(a));
1465 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
1467 int length = a->length + b->length - 1;
1469 SwsVector *vec = sws_getConstVec(0.0, length);
1474 for (i = 0; i < a->length; i++) {
1475 for (j = 0; j < b->length; j++) {
1476 vec->coeff[i + j] += a->coeff[i] * b->coeff[j];
1483 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
1485 int length = FFMAX(a->length, b->length);
1487 SwsVector *vec = sws_getConstVec(0.0, length);
1492 for (i = 0; i < a->length; i++)
1493 vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
1494 for (i = 0; i < b->length; i++)
1495 vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] += b->coeff[i];
1500 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
1502 int length = FFMAX(a->length, b->length);
1504 SwsVector *vec = sws_getConstVec(0.0, length);
1509 for (i = 0; i < a->length; i++)
1510 vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
1511 for (i = 0; i < b->length; i++)
1512 vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] -= b->coeff[i];
1517 /* shift left / or right if "shift" is negative */
1518 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
1520 int length = a->length + FFABS(shift) * 2;
1522 SwsVector *vec = sws_getConstVec(0.0, length);
1527 for (i = 0; i < a->length; i++) {
1528 vec->coeff[i + (length - 1) / 2 -
1529 (a->length - 1) / 2 - shift] = a->coeff[i];
1535 void sws_shiftVec(SwsVector *a, int shift)
1537 SwsVector *shifted = sws_getShiftedVec(a, shift);
1539 a->coeff = shifted->coeff;
1540 a->length = shifted->length;
1544 void sws_addVec(SwsVector *a, SwsVector *b)
1546 SwsVector *sum = sws_sumVec(a, b);
1548 a->coeff = sum->coeff;
1549 a->length = sum->length;
1553 void sws_subVec(SwsVector *a, SwsVector *b)
1555 SwsVector *diff = sws_diffVec(a, b);
1557 a->coeff = diff->coeff;
1558 a->length = diff->length;
1562 void sws_convVec(SwsVector *a, SwsVector *b)
1564 SwsVector *conv = sws_getConvVec(a, b);
1566 a->coeff = conv->coeff;
1567 a->length = conv->length;
1571 SwsVector *sws_cloneVec(SwsVector *a)
1574 SwsVector *vec = sws_allocVec(a->length);
1579 for (i = 0; i < a->length; i++)
1580 vec->coeff[i] = a->coeff[i];
1585 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
1592 for (i = 0; i < a->length; i++)
1593 if (a->coeff[i] > max)
1596 for (i = 0; i < a->length; i++)
1597 if (a->coeff[i] < min)
1602 for (i = 0; i < a->length; i++) {
1603 int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);
1604 av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
1606 av_log(log_ctx, log_level, " ");
1607 av_log(log_ctx, log_level, "|\n");
1611 void sws_freeVec(SwsVector *a)
1615 av_freep(&a->coeff);
1620 void sws_freeFilter(SwsFilter *filter)
1626 sws_freeVec(filter->lumH);
1628 sws_freeVec(filter->lumV);
1630 sws_freeVec(filter->chrH);
1632 sws_freeVec(filter->chrV);
1636 void sws_freeContext(SwsContext *c)
1643 for (i = 0; i < c->vLumBufSize; i++)
1644 av_freep(&c->lumPixBuf[i]);
1645 av_freep(&c->lumPixBuf);
1648 if (c->chrUPixBuf) {
1649 for (i = 0; i < c->vChrBufSize; i++)
1650 av_freep(&c->chrUPixBuf[i]);
1651 av_freep(&c->chrUPixBuf);
1652 av_freep(&c->chrVPixBuf);
1655 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1656 for (i = 0; i < c->vLumBufSize; i++)
1657 av_freep(&c->alpPixBuf[i]);
1658 av_freep(&c->alpPixBuf);
1661 av_freep(&c->vLumFilter);
1662 av_freep(&c->vChrFilter);
1663 av_freep(&c->hLumFilter);
1664 av_freep(&c->hChrFilter);
1666 av_freep(&c->vYCoeffsBank);
1667 av_freep(&c->vCCoeffsBank);
1670 av_freep(&c->vLumFilterPos);
1671 av_freep(&c->vChrFilterPos);
1672 av_freep(&c->hLumFilterPos);
1673 av_freep(&c->hChrFilterPos);
1676 #ifdef MAP_ANONYMOUS
1677 if (c->lumMmx2FilterCode)
1678 munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);
1679 if (c->chrMmx2FilterCode)
1680 munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);
1681 #elif HAVE_VIRTUALALLOC
1682 if (c->lumMmx2FilterCode)
1683 VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);
1684 if (c->chrMmx2FilterCode)
1685 VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);
1687 av_free(c->lumMmx2FilterCode);
1688 av_free(c->chrMmx2FilterCode);
1690 c->lumMmx2FilterCode = NULL;
1691 c->chrMmx2FilterCode = NULL;
1692 #endif /* HAVE_MMX */
1694 av_freep(&c->yuvTable);
1695 av_freep(&c->formatConvBuffer);
1700 struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,
1701 int srcH, enum PixelFormat srcFormat,
1703 enum PixelFormat dstFormat, int flags,
1704 SwsFilter *srcFilter,
1705 SwsFilter *dstFilter,
1706 const double *param)
1708 static const double default_param[2] = { SWS_PARAM_DEFAULT,
1709 SWS_PARAM_DEFAULT };
1712 param = default_param;
1715 (context->srcW != srcW ||
1716 context->srcH != srcH ||
1717 context->srcFormat != srcFormat ||
1718 context->dstW != dstW ||
1719 context->dstH != dstH ||
1720 context->dstFormat != dstFormat ||
1721 context->flags != flags ||
1722 context->param[0] != param[0] ||
1723 context->param[1] != param[1])) {
1724 sws_freeContext(context);
1729 if (!(context = sws_alloc_context()))
1731 context->srcW = srcW;
1732 context->srcH = srcH;
1733 context->srcRange = handle_jpeg(&srcFormat);
1734 context->src0Alpha = handle_0alpha(&srcFormat);
1735 context->srcFormat = srcFormat;
1736 context->dstW = dstW;
1737 context->dstH = dstH;
1738 context->dstRange = handle_jpeg(&dstFormat);
1739 context->dst0Alpha = handle_0alpha(&dstFormat);
1740 context->dstFormat = dstFormat;
1741 context->flags = flags;
1742 context->param[0] = param[0];
1743 context->param[1] = param[1];
1744 sws_setColorspaceDetails(context, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],
1746 ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,
1747 context->dstRange, 0, 1 << 16, 1 << 16);
1748 if (sws_init_context(context, srcFilter, dstFilter) < 0) {
1749 sws_freeContext(context);