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
21 #define _SVID_SOURCE //needed for MAP_ANONYMOUS
22 #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 "swscale_internal.h"
42 #include "libavutil/intreadwrite.h"
43 #include "libavutil/x86_cpu.h"
44 #include "libavutil/cpu.h"
45 #include "libavutil/avutil.h"
46 #include "libavutil/bswap.h"
47 #include "libavutil/opt.h"
48 #include "libavutil/pixdesc.h"
50 unsigned swscale_version(void)
52 return LIBSWSCALE_VERSION_INT;
55 const char *swscale_configuration(void)
57 return FFMPEG_CONFIGURATION;
60 const char *swscale_license(void)
62 #define LICENSE_PREFIX "libswscale license: "
63 return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
66 #define RET 0xC3 //near return opcode for x86
68 #define isSupportedIn(x) ( \
69 (x)==PIX_FMT_YUV420P \
70 || (x)==PIX_FMT_YUVA420P \
71 || (x)==PIX_FMT_YUYV422 \
72 || (x)==PIX_FMT_UYVY422 \
73 || (x)==PIX_FMT_RGB48BE \
74 || (x)==PIX_FMT_RGB48LE \
75 || (x)==PIX_FMT_RGB32 \
76 || (x)==PIX_FMT_RGB32_1 \
77 || (x)==PIX_FMT_BGR48BE \
78 || (x)==PIX_FMT_BGR48LE \
79 || (x)==PIX_FMT_BGR24 \
80 || (x)==PIX_FMT_BGR565 \
81 || (x)==PIX_FMT_BGR555 \
82 || (x)==PIX_FMT_BGR32 \
83 || (x)==PIX_FMT_BGR32_1 \
84 || (x)==PIX_FMT_RGB24 \
85 || (x)==PIX_FMT_RGB565 \
86 || (x)==PIX_FMT_RGB555 \
87 || (x)==PIX_FMT_GRAY8 \
88 || (x)==PIX_FMT_GRAY8A \
89 || (x)==PIX_FMT_YUV410P \
90 || (x)==PIX_FMT_YUV440P \
91 || (x)==PIX_FMT_NV12 \
92 || (x)==PIX_FMT_NV21 \
93 || (x)==PIX_FMT_GRAY16BE \
94 || (x)==PIX_FMT_GRAY16LE \
95 || (x)==PIX_FMT_YUV444P \
96 || (x)==PIX_FMT_YUV422P \
97 || (x)==PIX_FMT_YUV411P \
98 || (x)==PIX_FMT_YUVJ420P \
99 || (x)==PIX_FMT_YUVJ422P \
100 || (x)==PIX_FMT_YUVJ440P \
101 || (x)==PIX_FMT_YUVJ444P \
102 || (x)==PIX_FMT_PAL8 \
103 || (x)==PIX_FMT_BGR8 \
104 || (x)==PIX_FMT_RGB8 \
105 || (x)==PIX_FMT_BGR4_BYTE \
106 || (x)==PIX_FMT_RGB4_BYTE \
107 || (x)==PIX_FMT_YUV440P \
108 || (x)==PIX_FMT_MONOWHITE \
109 || (x)==PIX_FMT_MONOBLACK \
110 || (x)==PIX_FMT_YUV420P9LE \
111 || (x)==PIX_FMT_YUV420P10LE \
112 || (x)==PIX_FMT_YUV420P16LE \
113 || (x)==PIX_FMT_YUV422P16LE \
114 || (x)==PIX_FMT_YUV444P16LE \
115 || (x)==PIX_FMT_YUV420P9BE \
116 || (x)==PIX_FMT_YUV420P10BE \
117 || (x)==PIX_FMT_YUV420P16BE \
118 || (x)==PIX_FMT_YUV422P16BE \
119 || (x)==PIX_FMT_YUV444P16BE \
120 || (x)==PIX_FMT_YUV422P10 \
123 int sws_isSupportedInput(enum PixelFormat pix_fmt)
125 return isSupportedIn(pix_fmt);
128 #define isSupportedOut(x) ( \
129 (x)==PIX_FMT_YUV420P \
130 || (x)==PIX_FMT_YUVA420P \
131 || (x)==PIX_FMT_YUYV422 \
132 || (x)==PIX_FMT_UYVY422 \
133 || (x)==PIX_FMT_YUV444P \
134 || (x)==PIX_FMT_YUV422P \
135 || (x)==PIX_FMT_YUV411P \
136 || (x)==PIX_FMT_YUVJ420P \
137 || (x)==PIX_FMT_YUVJ422P \
138 || (x)==PIX_FMT_YUVJ440P \
139 || (x)==PIX_FMT_YUVJ444P \
141 || (x)==PIX_FMT_NV12 \
142 || (x)==PIX_FMT_NV21 \
143 || (x)==PIX_FMT_GRAY16BE \
144 || (x)==PIX_FMT_GRAY16LE \
145 || (x)==PIX_FMT_GRAY8 \
146 || (x)==PIX_FMT_YUV410P \
147 || (x)==PIX_FMT_YUV440P \
148 || (x)==PIX_FMT_YUV422P10 \
149 || (x)==PIX_FMT_YUV420P9LE \
150 || (x)==PIX_FMT_YUV420P10LE \
151 || (x)==PIX_FMT_YUV420P16LE \
152 || (x)==PIX_FMT_YUV422P16LE \
153 || (x)==PIX_FMT_YUV444P16LE \
154 || (x)==PIX_FMT_YUV420P9BE \
155 || (x)==PIX_FMT_YUV420P10BE \
156 || (x)==PIX_FMT_YUV420P16BE \
157 || (x)==PIX_FMT_YUV422P16BE \
158 || (x)==PIX_FMT_YUV444P16BE \
161 int sws_isSupportedOutput(enum PixelFormat pix_fmt)
163 return isSupportedOut(pix_fmt);
166 extern const int32_t ff_yuv2rgb_coeffs[8][4];
168 const char *sws_format_name(enum PixelFormat format)
170 if ((unsigned)format < PIX_FMT_NB && av_pix_fmt_descriptors[format].name)
171 return av_pix_fmt_descriptors[format].name;
173 return "Unknown format";
176 static double getSplineCoeff(double a, double b, double c, double d, double dist)
178 if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
179 else return getSplineCoeff( 0.0,
186 static int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
187 int srcW, int dstW, int filterAlign, int one, int flags, int cpu_flags,
188 SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
194 int64_t *filter=NULL;
195 int64_t *filter2=NULL;
196 const int64_t fone= 1LL<<54;
199 emms_c(); //FIXME this should not be required but it IS (even for non-MMX versions)
201 // NOTE: the +1 is for the MMX scaler which reads over the end
202 FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW+1)*sizeof(int16_t), fail);
204 if (FFABS(xInc - 0x10000) <10) { // unscaled
207 FF_ALLOCZ_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
209 for (i=0; i<dstW; i++) {
210 filter[i*filterSize]= fone;
214 } else if (flags&SWS_POINT) { // lame looking point sampling mode
218 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
220 xDstInSrc= xInc/2 - 0x8000;
221 for (i=0; i<dstW; i++) {
222 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
228 } else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) { // bilinear upscale
232 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
234 xDstInSrc= xInc/2 - 0x8000;
235 for (i=0; i<dstW; i++) {
236 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
240 //bilinear upscale / linear interpolate / area averaging
241 for (j=0; j<filterSize; j++) {
242 int64_t coeff= fone - FFABS((xx<<16) - xDstInSrc)*(fone>>16);
243 if (coeff<0) coeff=0;
244 filter[i*filterSize + j]= coeff;
253 if (flags&SWS_BICUBIC) sizeFactor= 4;
254 else if (flags&SWS_X) sizeFactor= 8;
255 else if (flags&SWS_AREA) sizeFactor= 1; //downscale only, for upscale it is bilinear
256 else if (flags&SWS_GAUSS) sizeFactor= 8; // infinite ;)
257 else if (flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? ceil(2*param[0]) : 6;
258 else if (flags&SWS_SINC) sizeFactor= 20; // infinite ;)
259 else if (flags&SWS_SPLINE) sizeFactor= 20; // infinite ;)
260 else if (flags&SWS_BILINEAR) sizeFactor= 2;
262 sizeFactor= 0; //GCC warning killer
266 if (xInc <= 1<<16) filterSize= 1 + sizeFactor; // upscale
267 else filterSize= 1 + (sizeFactor*srcW + dstW - 1)/ dstW;
269 if (filterSize > srcW-2) filterSize=srcW-2;
271 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
273 xDstInSrc= xInc - 0x10000;
274 for (i=0; i<dstW; i++) {
275 int xx= (xDstInSrc - ((filterSize-2)<<16)) / (1<<17);
278 for (j=0; j<filterSize; j++) {
279 int64_t d= ((int64_t)FFABS((xx<<17) - xDstInSrc))<<13;
285 floatd= d * (1.0/(1<<30));
287 if (flags & SWS_BICUBIC) {
288 int64_t B= (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1<<24);
289 int64_t C= (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1<<24);
290 int64_t dd = ( d*d)>>30;
291 int64_t ddd= (dd*d)>>30;
294 coeff = (12*(1<<24)-9*B-6*C)*ddd + (-18*(1<<24)+12*B+6*C)*dd + (6*(1<<24)-2*B)*(1<<30);
295 else if (d < 1LL<<31)
296 coeff = (-B-6*C)*ddd + (6*B+30*C)*dd + (-12*B-48*C)*d + (8*B+24*C)*(1<<30);
299 coeff *= fone>>(30+24);
301 /* else if (flags & SWS_X) {
302 double p= param ? param*0.01 : 0.3;
303 coeff = d ? sin(d*M_PI)/(d*M_PI) : 1.0;
304 coeff*= pow(2.0, - p*d*d);
306 else if (flags & SWS_X) {
307 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
311 c = cos(floatd*M_PI);
314 if (c<0.0) c= -pow(-c, A);
316 coeff= (c*0.5 + 0.5)*fone;
317 } else if (flags & SWS_AREA) {
318 int64_t d2= d - (1<<29);
319 if (d2*xInc < -(1LL<<(29+16))) coeff= 1.0 * (1LL<<(30+16));
320 else if (d2*xInc < (1LL<<(29+16))) coeff= -d2*xInc + (1LL<<(29+16));
322 coeff *= fone>>(30+16);
323 } else if (flags & SWS_GAUSS) {
324 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
325 coeff = (pow(2.0, - p*floatd*floatd))*fone;
326 } else if (flags & SWS_SINC) {
327 coeff = (d ? sin(floatd*M_PI)/(floatd*M_PI) : 1.0)*fone;
328 } else if (flags & SWS_LANCZOS) {
329 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
330 coeff = (d ? sin(floatd*M_PI)*sin(floatd*M_PI/p)/(floatd*floatd*M_PI*M_PI/p) : 1.0)*fone;
331 if (floatd>p) coeff=0;
332 } else if (flags & SWS_BILINEAR) {
334 if (coeff<0) coeff=0;
336 } else if (flags & SWS_SPLINE) {
337 double p=-2.196152422706632;
338 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, floatd) * fone;
340 coeff= 0.0; //GCC warning killer
344 filter[i*filterSize + j]= coeff;
351 /* apply src & dst Filter to filter -> filter2
354 assert(filterSize>0);
355 filter2Size= filterSize;
356 if (srcFilter) filter2Size+= srcFilter->length - 1;
357 if (dstFilter) filter2Size+= dstFilter->length - 1;
358 assert(filter2Size>0);
359 FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size*dstW*sizeof(*filter2), fail);
361 for (i=0; i<dstW; i++) {
365 for (k=0; k<srcFilter->length; k++) {
366 for (j=0; j<filterSize; j++)
367 filter2[i*filter2Size + k + j] += srcFilter->coeff[k]*filter[i*filterSize + j];
370 for (j=0; j<filterSize; j++)
371 filter2[i*filter2Size + j]= filter[i*filterSize + j];
375 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
379 /* try to reduce the filter-size (step1 find size and shift left) */
380 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
382 for (i=dstW-1; i>=0; i--) {
383 int min= filter2Size;
387 /* get rid of near zero elements on the left by shifting left */
388 for (j=0; j<filter2Size; j++) {
390 cutOff += FFABS(filter2[i*filter2Size]);
392 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
394 /* preserve monotonicity because the core can't handle the filter otherwise */
395 if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
397 // move filter coefficients left
398 for (k=1; k<filter2Size; k++)
399 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
400 filter2[i*filter2Size + k - 1]= 0;
405 /* count near zeros on the right */
406 for (j=filter2Size-1; j>0; j--) {
407 cutOff += FFABS(filter2[i*filter2Size + j]);
409 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
413 if (min>minFilterSize) minFilterSize= min;
416 if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) {
417 // we can handle the special case 4,
418 // so we don't want to go to the full 8
419 if (minFilterSize < 5)
422 // We really don't want to waste our time
423 // doing useless computation, so fall back on
424 // the scalar C code for very small filters.
425 // Vectorizing is worth it only if you have a
426 // decent-sized vector.
427 if (minFilterSize < 3)
431 if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
432 // special case for unscaled vertical filtering
433 if (minFilterSize == 1 && filterAlign == 2)
437 assert(minFilterSize > 0);
438 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
439 assert(filterSize > 0);
440 filter= av_malloc(filterSize*dstW*sizeof(*filter));
441 if (filterSize >= MAX_FILTER_SIZE*16/((flags&SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
443 *outFilterSize= filterSize;
445 if (flags&SWS_PRINT_INFO)
446 av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
447 /* try to reduce the filter-size (step2 reduce it) */
448 for (i=0; i<dstW; i++) {
451 for (j=0; j<filterSize; j++) {
452 if (j>=filter2Size) filter[i*filterSize + j]= 0;
453 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
454 if((flags & SWS_BITEXACT) && j>=minFilterSize)
455 filter[i*filterSize + j]= 0;
459 //FIXME try to align filterPos if possible
462 for (i=0; i<dstW; i++) {
464 if ((*filterPos)[i] < 0) {
465 // move filter coefficients left to compensate for filterPos
466 for (j=1; j<filterSize; j++) {
467 int left= FFMAX(j + (*filterPos)[i], 0);
468 filter[i*filterSize + left] += filter[i*filterSize + j];
469 filter[i*filterSize + j]=0;
474 if ((*filterPos)[i] + filterSize > srcW) {
475 int shift= (*filterPos)[i] + filterSize - srcW;
476 // move filter coefficients right to compensate for filterPos
477 for (j=filterSize-2; j>=0; j--) {
478 int right= FFMIN(j + shift, filterSize-1);
479 filter[i*filterSize +right] += filter[i*filterSize +j];
480 filter[i*filterSize +j]=0;
482 (*filterPos)[i]= srcW - filterSize;
486 // Note the +1 is for the MMX scaler which reads over the end
487 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
488 FF_ALLOCZ_OR_GOTO(NULL, *outFilter, *outFilterSize*(dstW+1)*sizeof(int16_t), fail);
490 /* normalize & store in outFilter */
491 for (i=0; i<dstW; i++) {
496 for (j=0; j<filterSize; j++) {
497 sum+= filter[i*filterSize + j];
499 sum= (sum + one/2)/ one;
500 for (j=0; j<*outFilterSize; j++) {
501 int64_t v= filter[i*filterSize + j] + error;
502 int intV= ROUNDED_DIV(v, sum);
503 (*outFilter)[i*(*outFilterSize) + j]= intV;
508 (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
509 for (i=0; i<*outFilterSize; i++) {
510 int j= dstW*(*outFilterSize);
511 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
522 static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode, int16_t *filter, int32_t *filterPos, int numSplits)
525 x86_reg imm8OfPShufW1A;
526 x86_reg imm8OfPShufW2A;
527 x86_reg fragmentLengthA;
529 x86_reg imm8OfPShufW1B;
530 x86_reg imm8OfPShufW2B;
531 x86_reg fragmentLengthB;
536 // create an optimized horizontal scaling routine
537 /* This scaler is made of runtime-generated MMX2 code using specially
538 * tuned pshufw instructions. For every four output pixels, if four
539 * input pixels are enough for the fast bilinear scaling, then a chunk
540 * of fragmentB is used. If five input pixels are needed, then a chunk
541 * of fragmentA is used.
550 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
551 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
552 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
553 "punpcklbw %%mm7, %%mm1 \n\t"
554 "punpcklbw %%mm7, %%mm0 \n\t"
555 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
557 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
559 "psubw %%mm1, %%mm0 \n\t"
560 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
561 "pmullw %%mm3, %%mm0 \n\t"
562 "psllw $7, %%mm1 \n\t"
563 "paddw %%mm1, %%mm0 \n\t"
565 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
567 "add $8, %%"REG_a" \n\t"
571 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
572 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
573 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
578 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
582 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
583 "=r" (fragmentLengthA)
590 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
591 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
592 "punpcklbw %%mm7, %%mm0 \n\t"
593 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
595 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
597 "psubw %%mm1, %%mm0 \n\t"
598 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
599 "pmullw %%mm3, %%mm0 \n\t"
600 "psllw $7, %%mm1 \n\t"
601 "paddw %%mm1, %%mm0 \n\t"
603 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
605 "add $8, %%"REG_a" \n\t"
609 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
610 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
611 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
616 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
620 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
621 "=r" (fragmentLengthB)
624 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
627 for (i=0; i<dstW/numSplits; i++) {
632 int b=((xpos+xInc)>>16) - xx;
633 int c=((xpos+xInc*2)>>16) - xx;
634 int d=((xpos+xInc*3)>>16) - xx;
636 uint8_t *fragment = (d+1<4) ? fragmentB : fragmentA;
637 x86_reg imm8OfPShufW1 = (d+1<4) ? imm8OfPShufW1B : imm8OfPShufW1A;
638 x86_reg imm8OfPShufW2 = (d+1<4) ? imm8OfPShufW2B : imm8OfPShufW2A;
639 x86_reg fragmentLength = (d+1<4) ? fragmentLengthB : fragmentLengthA;
640 int maxShift= 3-(d+inc);
644 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
645 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
646 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
647 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
650 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
652 filterCode[fragmentPos + imm8OfPShufW1]=
653 (a+inc) | ((b+inc)<<2) | ((c+inc)<<4) | ((d+inc)<<6);
654 filterCode[fragmentPos + imm8OfPShufW2]=
655 a | (b<<2) | (c<<4) | (d<<6);
657 if (i+4-inc>=dstW) shift=maxShift; //avoid overread
658 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
660 if (shift && i>=shift) {
661 filterCode[fragmentPos + imm8OfPShufW1]+= 0x55*shift;
662 filterCode[fragmentPos + imm8OfPShufW2]+= 0x55*shift;
663 filterPos[i/2]-=shift;
667 fragmentPos+= fragmentLength;
670 filterCode[fragmentPos]= RET;
675 filterPos[((i/2)+1)&(~1)]= xpos>>16; // needed to jump to the next part
677 return fragmentPos + 1;
679 #endif /* HAVE_MMX2 */
681 static void getSubSampleFactors(int *h, int *v, enum PixelFormat format)
683 *h = av_pix_fmt_descriptors[format].log2_chroma_w;
684 *v = av_pix_fmt_descriptors[format].log2_chroma_h;
687 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation)
689 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
690 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
692 c->brightness= brightness;
693 c->contrast = contrast;
694 c->saturation= saturation;
695 c->srcRange = srcRange;
696 c->dstRange = dstRange;
697 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
699 c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->dstFormat]);
700 c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->srcFormat]);
702 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
705 if (HAVE_ALTIVEC && av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)
706 ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness, contrast, saturation);
710 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation)
712 if (!c || isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
714 *inv_table = c->srcColorspaceTable;
715 *table = c->dstColorspaceTable;
716 *srcRange = c->srcRange;
717 *dstRange = c->dstRange;
718 *brightness= c->brightness;
719 *contrast = c->contrast;
720 *saturation= c->saturation;
725 static int handle_jpeg(enum PixelFormat *format)
728 case PIX_FMT_YUVJ420P: *format = PIX_FMT_YUV420P; return 1;
729 case PIX_FMT_YUVJ422P: *format = PIX_FMT_YUV422P; return 1;
730 case PIX_FMT_YUVJ444P: *format = PIX_FMT_YUV444P; return 1;
731 case PIX_FMT_YUVJ440P: *format = PIX_FMT_YUV440P; return 1;
736 SwsContext *sws_alloc_context(void)
738 SwsContext *c= av_mallocz(sizeof(SwsContext));
740 c->av_class = &sws_context_class;
741 av_opt_set_defaults(c);
746 int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)
749 int usesVFilter, usesHFilter;
751 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
756 int dst_stride = FFALIGN(dstW * sizeof(int16_t)+66, 16), dst_stride_px = dst_stride >> 1;
757 int flags, cpu_flags;
758 enum PixelFormat srcFormat= c->srcFormat;
759 enum PixelFormat dstFormat= c->dstFormat;
761 cpu_flags = av_get_cpu_flags();
764 if (!rgb15to16) sws_rgb2rgb_init();
766 unscaled = (srcW == dstW && srcH == dstH);
768 if (!isSupportedIn(srcFormat)) {
769 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as input pixel format\n", sws_format_name(srcFormat));
770 return AVERROR(EINVAL);
772 if (!isSupportedOut(dstFormat)) {
773 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as output pixel format\n", sws_format_name(dstFormat));
774 return AVERROR(EINVAL);
777 i= flags & ( SWS_POINT
788 if(!i || (i & (i-1))) {
789 av_log(NULL, AV_LOG_ERROR, "swScaler: Exactly one scaler algorithm must be chosen\n");
790 return AVERROR(EINVAL);
793 if (srcW<4 || srcH<1 || dstW<8 || dstH<1) { //FIXME check if these are enough and try to lowwer them after fixing the relevant parts of the code
794 av_log(NULL, AV_LOG_ERROR, "swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
795 srcW, srcH, dstW, dstH);
796 return AVERROR(EINVAL);
799 if (!dstFilter) dstFilter= &dummyFilter;
800 if (!srcFilter) srcFilter= &dummyFilter;
802 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
803 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
804 c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[dstFormat]);
805 c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[srcFormat]);
806 c->vRounder= 4* 0x0001000100010001ULL;
808 usesVFilter = (srcFilter->lumV && srcFilter->lumV->length>1) ||
809 (srcFilter->chrV && srcFilter->chrV->length>1) ||
810 (dstFilter->lumV && dstFilter->lumV->length>1) ||
811 (dstFilter->chrV && dstFilter->chrV->length>1);
812 usesHFilter = (srcFilter->lumH && srcFilter->lumH->length>1) ||
813 (srcFilter->chrH && srcFilter->chrH->length>1) ||
814 (dstFilter->lumH && dstFilter->lumH->length>1) ||
815 (dstFilter->chrH && dstFilter->chrH->length>1);
817 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
818 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
820 // reuse chroma for 2 pixels RGB/BGR unless user wants full chroma interpolation
821 if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
823 // drop some chroma lines if the user wants it
824 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
825 c->chrSrcVSubSample+= c->vChrDrop;
827 // drop every other pixel for chroma calculation unless user wants full chroma
828 if (isAnyRGB(srcFormat) && !(flags&SWS_FULL_CHR_H_INP)
829 && srcFormat!=PIX_FMT_RGB8 && srcFormat!=PIX_FMT_BGR8
830 && srcFormat!=PIX_FMT_RGB4 && srcFormat!=PIX_FMT_BGR4
831 && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE
832 && ((dstW>>c->chrDstHSubSample) <= (srcW>>1) || (flags&SWS_FAST_BILINEAR)))
833 c->chrSrcHSubSample=1;
835 // Note the -((-x)>>y) is so that we always round toward +inf.
836 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
837 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
838 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
839 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
841 /* unscaled special cases */
842 if (unscaled && !usesHFilter && !usesVFilter && (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
843 ff_get_unscaled_swscale(c);
846 if (flags&SWS_PRINT_INFO)
847 av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
848 sws_format_name(srcFormat), sws_format_name(dstFormat));
853 FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW*2+78, 16) * 2, fail);
854 if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2) {
855 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
856 if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR)) {
857 if (flags&SWS_PRINT_INFO)
858 av_log(c, AV_LOG_INFO, "output width is not a multiple of 32 -> no MMX2 scaler\n");
860 if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat)) c->canMMX2BeUsed=0;
865 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
866 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
868 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
869 // but only for the FAST_BILINEAR mode otherwise do correct scaling
870 // n-2 is the last chrominance sample available
871 // this is not perfect, but no one should notice the difference, the more correct variant
872 // would be like the vertical one, but that would require some special code for the
873 // first and last pixel
874 if (flags&SWS_FAST_BILINEAR) {
875 if (c->canMMX2BeUsed) {
879 //we don't use the x86 asm scaler if MMX is available
880 else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
881 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
882 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
886 /* precalculate horizontal scaler filter coefficients */
889 // can't downscale !!!
890 if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {
891 c->lumMmx2FilterCodeSize = initMMX2HScaler( dstW, c->lumXInc, NULL, NULL, NULL, 8);
892 c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc, NULL, NULL, NULL, 4);
895 c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
896 c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
897 #elif HAVE_VIRTUALALLOC
898 c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
899 c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
901 c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);
902 c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);
906 if (c->lumMmx2FilterCode == MAP_FAILED || c->chrMmx2FilterCode == MAP_FAILED)
908 if (!c->lumMmx2FilterCode || !c->chrMmx2FilterCode)
910 return AVERROR(ENOMEM);
911 FF_ALLOCZ_OR_GOTO(c, c->hLumFilter , (dstW /8+8)*sizeof(int16_t), fail);
912 FF_ALLOCZ_OR_GOTO(c, c->hChrFilter , (c->chrDstW /4+8)*sizeof(int16_t), fail);
913 FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW /2/8+8)*sizeof(int32_t), fail);
914 FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW/2/4+8)*sizeof(int32_t), fail);
916 initMMX2HScaler( dstW, c->lumXInc, c->lumMmx2FilterCode, c->hLumFilter, c->hLumFilterPos, 8);
917 initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode, c->hChrFilter, c->hChrFilterPos, 4);
920 mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
921 mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
924 #endif /* HAVE_MMX2 */
926 const int filterAlign=
927 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :
928 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
931 if (initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
932 srcW , dstW, filterAlign, 1<<14,
933 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags, cpu_flags,
934 srcFilter->lumH, dstFilter->lumH, c->param) < 0)
936 if (initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
937 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
938 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, cpu_flags,
939 srcFilter->chrH, dstFilter->chrH, c->param) < 0)
942 } // initialize horizontal stuff
944 /* precalculate vertical scaler filter coefficients */
946 const int filterAlign=
947 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
948 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
951 if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
952 srcH , dstH, filterAlign, (1<<12),
953 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags, cpu_flags,
954 srcFilter->lumV, dstFilter->lumV, c->param) < 0)
956 if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
957 c->chrSrcH, c->chrDstH, filterAlign, (1<<12),
958 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, cpu_flags,
959 srcFilter->chrV, dstFilter->chrV, c->param) < 0)
963 FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof (vector signed short)*c->vLumFilterSize*c->dstH, fail);
964 FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH, fail);
966 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
968 short *p = (short *)&c->vYCoeffsBank[i];
970 p[j] = c->vLumFilter[i];
973 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
975 short *p = (short *)&c->vCCoeffsBank[i];
977 p[j] = c->vChrFilter[i];
982 // calculate buffer sizes so that they won't run out while handling these damn slices
983 c->vLumBufSize= c->vLumFilterSize;
984 c->vChrBufSize= c->vChrFilterSize;
985 for (i=0; i<dstH; i++) {
986 int chrI= (int64_t)i*c->chrDstH / dstH;
987 int nextSlice= FFMAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
988 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
990 nextSlice>>= c->chrSrcVSubSample;
991 nextSlice<<= c->chrSrcVSubSample;
992 if (c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
993 c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
994 if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
995 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
998 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
999 // allocate several megabytes to handle all possible cases)
1000 FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
1001 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
1002 FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
1003 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
1004 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
1005 //Note we need at least one pixel more at the end because of the MMX code (just in case someone wanna replace the 4000/8000)
1006 /* align at 16 bytes for AltiVec */
1007 for (i=0; i<c->vLumBufSize; i++) {
1008 FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], dst_stride+1, fail);
1009 c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];
1011 c->uv_off = dst_stride_px;
1012 c->uv_offx2 = dst_stride;
1013 for (i=0; i<c->vChrBufSize; i++) {
1014 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i+c->vChrBufSize], dst_stride*2+1, fail);
1015 c->chrUPixBuf[i] = c->chrUPixBuf[i+c->vChrBufSize];
1016 c->chrVPixBuf[i] = c->chrVPixBuf[i+c->vChrBufSize] = c->chrUPixBuf[i] + dst_stride_px;
1018 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
1019 for (i=0; i<c->vLumBufSize; i++) {
1020 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], dst_stride+1, fail);
1021 c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];
1024 //try to avoid drawing green stuff between the right end and the stride end
1025 for (i=0; i<c->vChrBufSize; i++)
1026 memset(c->chrUPixBuf[i], 64, dst_stride*2+1);
1028 assert(c->chrDstH <= dstH);
1030 if (flags&SWS_PRINT_INFO) {
1031 if (flags&SWS_FAST_BILINEAR) av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
1032 else if (flags&SWS_BILINEAR) av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
1033 else if (flags&SWS_BICUBIC) av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
1034 else if (flags&SWS_X) av_log(c, AV_LOG_INFO, "Experimental scaler, ");
1035 else if (flags&SWS_POINT) av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
1036 else if (flags&SWS_AREA) av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
1037 else if (flags&SWS_BICUBLIN) av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
1038 else if (flags&SWS_GAUSS) av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
1039 else if (flags&SWS_SINC) av_log(c, AV_LOG_INFO, "Sinc scaler, ");
1040 else if (flags&SWS_LANCZOS) av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
1041 else if (flags&SWS_SPLINE) av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
1042 else av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
1044 av_log(c, AV_LOG_INFO, "from %s to %s%s ",
1045 sws_format_name(srcFormat),
1047 dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ||
1048 dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||
1049 dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ? "dithered " : "",
1053 sws_format_name(dstFormat));
1055 if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2) av_log(c, AV_LOG_INFO, "using MMX2\n");
1056 else if (HAVE_AMD3DNOW && cpu_flags & AV_CPU_FLAG_3DNOW) av_log(c, AV_LOG_INFO, "using 3DNOW\n");
1057 else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) av_log(c, AV_LOG_INFO, "using MMX\n");
1058 else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) av_log(c, AV_LOG_INFO, "using AltiVec\n");
1059 else av_log(c, AV_LOG_INFO, "using C\n");
1061 if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
1062 if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
1063 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
1065 if (c->hLumFilterSize==4)
1066 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal luminance scaling\n");
1067 else if (c->hLumFilterSize==8)
1068 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal luminance scaling\n");
1070 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal luminance scaling\n");
1072 if (c->hChrFilterSize==4)
1073 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal chrominance scaling\n");
1074 else if (c->hChrFilterSize==8)
1075 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal chrominance scaling\n");
1077 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal chrominance scaling\n");
1081 av_log(c, AV_LOG_VERBOSE, "using x86 asm scaler for horizontal scaling\n");
1083 if (flags & SWS_FAST_BILINEAR)
1084 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR C scaler for horizontal scaling\n");
1086 av_log(c, AV_LOG_VERBOSE, "using C scaler for horizontal scaling\n");
1089 if (isPlanarYUV(dstFormat)) {
1090 if (c->vLumFilterSize==1)
1091 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n",
1092 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
1094 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n",
1095 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
1097 if (c->vLumFilterSize==1 && c->vChrFilterSize==2)
1098 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
1099 " 2-tap scaler for vertical chrominance scaling (BGR)\n",
1100 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
1101 else if (c->vLumFilterSize==2 && c->vChrFilterSize==2)
1102 av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n",
1103 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
1105 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n",
1106 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
1109 if (dstFormat==PIX_FMT_BGR24)
1110 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 converter\n",
1111 (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2) ? "MMX2" :
1112 ((HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C"));
1113 else if (dstFormat==PIX_FMT_RGB32)
1114 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 converter\n",
1115 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
1116 else if (dstFormat==PIX_FMT_BGR565)
1117 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 converter\n",
1118 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
1119 else if (dstFormat==PIX_FMT_BGR555)
1120 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 converter\n",
1121 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
1122 else if (dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||
1123 dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE)
1124 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR12 converter\n",
1125 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
1127 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
1128 av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1129 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
1130 av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1131 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
1134 c->swScale= ff_getSwsFunc(c);
1136 fail: //FIXME replace things by appropriate error codes
1140 #if FF_API_SWS_GETCONTEXT
1141 SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,
1142 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
1143 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
1147 if(!(c=sws_alloc_context()))
1155 c->srcRange = handle_jpeg(&srcFormat);
1156 c->dstRange = handle_jpeg(&dstFormat);
1157 c->srcFormat= srcFormat;
1158 c->dstFormat= dstFormat;
1161 c->param[0] = param[0];
1162 c->param[1] = param[1];
1164 sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/, c->dstRange, 0, 1<<16, 1<<16);
1166 if(sws_init_context(c, srcFilter, dstFilter) < 0){
1175 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
1176 float lumaSharpen, float chromaSharpen,
1177 float chromaHShift, float chromaVShift,
1180 SwsFilter *filter= av_malloc(sizeof(SwsFilter));
1184 if (lumaGBlur!=0.0) {
1185 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
1186 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
1188 filter->lumH= sws_getIdentityVec();
1189 filter->lumV= sws_getIdentityVec();
1192 if (chromaGBlur!=0.0) {
1193 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
1194 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
1196 filter->chrH= sws_getIdentityVec();
1197 filter->chrV= sws_getIdentityVec();
1200 if (chromaSharpen!=0.0) {
1201 SwsVector *id= sws_getIdentityVec();
1202 sws_scaleVec(filter->chrH, -chromaSharpen);
1203 sws_scaleVec(filter->chrV, -chromaSharpen);
1204 sws_addVec(filter->chrH, id);
1205 sws_addVec(filter->chrV, id);
1209 if (lumaSharpen!=0.0) {
1210 SwsVector *id= sws_getIdentityVec();
1211 sws_scaleVec(filter->lumH, -lumaSharpen);
1212 sws_scaleVec(filter->lumV, -lumaSharpen);
1213 sws_addVec(filter->lumH, id);
1214 sws_addVec(filter->lumV, id);
1218 if (chromaHShift != 0.0)
1219 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
1221 if (chromaVShift != 0.0)
1222 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
1224 sws_normalizeVec(filter->chrH, 1.0);
1225 sws_normalizeVec(filter->chrV, 1.0);
1226 sws_normalizeVec(filter->lumH, 1.0);
1227 sws_normalizeVec(filter->lumV, 1.0);
1229 if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
1230 if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
1235 SwsVector *sws_allocVec(int length)
1237 SwsVector *vec = av_malloc(sizeof(SwsVector));
1240 vec->length = length;
1241 vec->coeff = av_malloc(sizeof(double) * length);
1247 SwsVector *sws_getGaussianVec(double variance, double quality)
1249 const int length= (int)(variance*quality + 0.5) | 1;
1251 double middle= (length-1)*0.5;
1252 SwsVector *vec= sws_allocVec(length);
1257 for (i=0; i<length; i++) {
1258 double dist= i-middle;
1259 vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);
1262 sws_normalizeVec(vec, 1.0);
1267 SwsVector *sws_getConstVec(double c, int length)
1270 SwsVector *vec= sws_allocVec(length);
1275 for (i=0; i<length; i++)
1281 SwsVector *sws_getIdentityVec(void)
1283 return sws_getConstVec(1.0, 1);
1286 static double sws_dcVec(SwsVector *a)
1291 for (i=0; i<a->length; i++)
1297 void sws_scaleVec(SwsVector *a, double scalar)
1301 for (i=0; i<a->length; i++)
1302 a->coeff[i]*= scalar;
1305 void sws_normalizeVec(SwsVector *a, double height)
1307 sws_scaleVec(a, height/sws_dcVec(a));
1310 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
1312 int length= a->length + b->length - 1;
1314 SwsVector *vec= sws_getConstVec(0.0, length);
1319 for (i=0; i<a->length; i++) {
1320 for (j=0; j<b->length; j++) {
1321 vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];
1328 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
1330 int length= FFMAX(a->length, b->length);
1332 SwsVector *vec= sws_getConstVec(0.0, length);
1337 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1338 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
1343 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
1345 int length= FFMAX(a->length, b->length);
1347 SwsVector *vec= sws_getConstVec(0.0, length);
1352 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1353 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
1358 /* shift left / or right if "shift" is negative */
1359 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
1361 int length= a->length + FFABS(shift)*2;
1363 SwsVector *vec= sws_getConstVec(0.0, length);
1368 for (i=0; i<a->length; i++) {
1369 vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
1375 void sws_shiftVec(SwsVector *a, int shift)
1377 SwsVector *shifted= sws_getShiftedVec(a, shift);
1379 a->coeff= shifted->coeff;
1380 a->length= shifted->length;
1384 void sws_addVec(SwsVector *a, SwsVector *b)
1386 SwsVector *sum= sws_sumVec(a, b);
1388 a->coeff= sum->coeff;
1389 a->length= sum->length;
1393 void sws_subVec(SwsVector *a, SwsVector *b)
1395 SwsVector *diff= sws_diffVec(a, b);
1397 a->coeff= diff->coeff;
1398 a->length= diff->length;
1402 void sws_convVec(SwsVector *a, SwsVector *b)
1404 SwsVector *conv= sws_getConvVec(a, b);
1406 a->coeff= conv->coeff;
1407 a->length= conv->length;
1411 SwsVector *sws_cloneVec(SwsVector *a)
1414 SwsVector *vec= sws_allocVec(a->length);
1419 for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];
1424 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
1431 for (i=0; i<a->length; i++)
1432 if (a->coeff[i]>max) max= a->coeff[i];
1434 for (i=0; i<a->length; i++)
1435 if (a->coeff[i]<min) min= a->coeff[i];
1439 for (i=0; i<a->length; i++) {
1440 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
1441 av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
1442 for (;x>0; x--) av_log(log_ctx, log_level, " ");
1443 av_log(log_ctx, log_level, "|\n");
1447 #if LIBSWSCALE_VERSION_MAJOR < 1
1448 void sws_printVec(SwsVector *a)
1450 sws_printVec2(a, NULL, AV_LOG_DEBUG);
1454 void sws_freeVec(SwsVector *a)
1457 av_freep(&a->coeff);
1462 void sws_freeFilter(SwsFilter *filter)
1464 if (!filter) return;
1466 if (filter->lumH) sws_freeVec(filter->lumH);
1467 if (filter->lumV) sws_freeVec(filter->lumV);
1468 if (filter->chrH) sws_freeVec(filter->chrH);
1469 if (filter->chrV) sws_freeVec(filter->chrV);
1473 void sws_freeContext(SwsContext *c)
1479 for (i=0; i<c->vLumBufSize; i++)
1480 av_freep(&c->lumPixBuf[i]);
1481 av_freep(&c->lumPixBuf);
1484 if (c->chrUPixBuf) {
1485 for (i=0; i<c->vChrBufSize; i++)
1486 av_freep(&c->chrUPixBuf[i]);
1487 av_freep(&c->chrUPixBuf);
1488 av_freep(&c->chrVPixBuf);
1491 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1492 for (i=0; i<c->vLumBufSize; i++)
1493 av_freep(&c->alpPixBuf[i]);
1494 av_freep(&c->alpPixBuf);
1497 av_freep(&c->vLumFilter);
1498 av_freep(&c->vChrFilter);
1499 av_freep(&c->hLumFilter);
1500 av_freep(&c->hChrFilter);
1502 av_freep(&c->vYCoeffsBank);
1503 av_freep(&c->vCCoeffsBank);
1506 av_freep(&c->vLumFilterPos);
1507 av_freep(&c->vChrFilterPos);
1508 av_freep(&c->hLumFilterPos);
1509 av_freep(&c->hChrFilterPos);
1512 #ifdef MAP_ANONYMOUS
1513 if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);
1514 if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);
1515 #elif HAVE_VIRTUALALLOC
1516 if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);
1517 if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);
1519 av_free(c->lumMmx2FilterCode);
1520 av_free(c->chrMmx2FilterCode);
1522 c->lumMmx2FilterCode=NULL;
1523 c->chrMmx2FilterCode=NULL;
1524 #endif /* HAVE_MMX */
1526 av_freep(&c->yuvTable);
1527 av_freep(&c->formatConvBuffer);
1532 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
1533 int srcW, int srcH, enum PixelFormat srcFormat,
1534 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
1535 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
1537 static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};
1540 param = default_param;
1543 (context->srcW != srcW ||
1544 context->srcH != srcH ||
1545 context->srcFormat != srcFormat ||
1546 context->dstW != dstW ||
1547 context->dstH != dstH ||
1548 context->dstFormat != dstFormat ||
1549 context->flags != flags ||
1550 context->param[0] != param[0] ||
1551 context->param[1] != param[1])) {
1552 sws_freeContext(context);
1557 if (!(context = sws_alloc_context()))
1559 context->srcW = srcW;
1560 context->srcH = srcH;
1561 context->srcRange = handle_jpeg(&srcFormat);
1562 context->srcFormat = srcFormat;
1563 context->dstW = dstW;
1564 context->dstH = dstH;
1565 context->dstRange = handle_jpeg(&dstFormat);
1566 context->dstFormat = dstFormat;
1567 context->flags = flags;
1568 context->param[0] = param[0];
1569 context->param[1] = param[1];
1570 sws_setColorspaceDetails(context, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], context->srcRange, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/, context->dstRange, 0, 1<<16, 1<<16);
1571 if (sws_init_context(context, srcFilter, dstFilter) < 0) {
1572 sws_freeContext(context);