2 * Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
4 * This file is part of Libav.
6 * Libav 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 * Libav 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 Libav; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
29 #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
30 #define MAP_ANONYMOUS MAP_ANON
34 #define WIN32_LEAN_AND_MEAN
38 #include "swscale_internal.h"
40 #include "libavutil/intreadwrite.h"
41 #include "libavutil/x86_cpu.h"
42 #include "libavutil/cpu.h"
43 #include "libavutil/avutil.h"
44 #include "libavutil/bswap.h"
45 #include "libavutil/opt.h"
46 #include "libavutil/pixdesc.h"
48 unsigned swscale_version(void)
50 return LIBSWSCALE_VERSION_INT;
53 const char *swscale_configuration(void)
55 return LIBAV_CONFIGURATION;
58 const char *swscale_license(void)
60 #define LICENSE_PREFIX "libswscale license: "
61 return LICENSE_PREFIX LIBAV_LICENSE + sizeof(LICENSE_PREFIX) - 1;
64 #define RET 0xC3 //near return opcode for x86
66 #define isSupportedIn(x) ( \
67 (x)==PIX_FMT_YUV420P \
68 || (x)==PIX_FMT_YUVA420P \
69 || (x)==PIX_FMT_YUYV422 \
70 || (x)==PIX_FMT_UYVY422 \
71 || (x)==PIX_FMT_RGB48BE \
72 || (x)==PIX_FMT_RGB48LE \
73 || (x)==PIX_FMT_RGB32 \
74 || (x)==PIX_FMT_RGB32_1 \
75 || (x)==PIX_FMT_BGR48BE \
76 || (x)==PIX_FMT_BGR48LE \
77 || (x)==PIX_FMT_BGR24 \
78 || (x)==PIX_FMT_BGR565 \
79 || (x)==PIX_FMT_BGR555 \
80 || (x)==PIX_FMT_BGR32 \
81 || (x)==PIX_FMT_BGR32_1 \
82 || (x)==PIX_FMT_RGB24 \
83 || (x)==PIX_FMT_RGB565 \
84 || (x)==PIX_FMT_RGB555 \
85 || (x)==PIX_FMT_GRAY8 \
86 || (x)==PIX_FMT_Y400A \
87 || (x)==PIX_FMT_YUV410P \
88 || (x)==PIX_FMT_YUV440P \
89 || (x)==PIX_FMT_NV12 \
90 || (x)==PIX_FMT_NV21 \
91 || (x)==PIX_FMT_GRAY16BE \
92 || (x)==PIX_FMT_GRAY16LE \
93 || (x)==PIX_FMT_YUV444P \
94 || (x)==PIX_FMT_YUV422P \
95 || (x)==PIX_FMT_YUV411P \
96 || (x)==PIX_FMT_YUVJ420P \
97 || (x)==PIX_FMT_YUVJ422P \
98 || (x)==PIX_FMT_YUVJ440P \
99 || (x)==PIX_FMT_YUVJ444P \
100 || (x)==PIX_FMT_PAL8 \
101 || (x)==PIX_FMT_BGR8 \
102 || (x)==PIX_FMT_RGB8 \
103 || (x)==PIX_FMT_BGR4_BYTE \
104 || (x)==PIX_FMT_RGB4_BYTE \
105 || (x)==PIX_FMT_YUV440P \
106 || (x)==PIX_FMT_MONOWHITE \
107 || (x)==PIX_FMT_MONOBLACK \
108 || (x)==PIX_FMT_YUV420P9LE \
109 || (x)==PIX_FMT_YUV420P10LE \
110 || (x)==PIX_FMT_YUV420P16LE \
111 || (x)==PIX_FMT_YUV422P16LE \
112 || (x)==PIX_FMT_YUV444P16LE \
113 || (x)==PIX_FMT_YUV420P9BE \
114 || (x)==PIX_FMT_YUV420P10BE \
115 || (x)==PIX_FMT_YUV420P16BE \
116 || (x)==PIX_FMT_YUV422P16BE \
117 || (x)==PIX_FMT_YUV444P16BE \
120 int sws_isSupportedInput(enum PixelFormat pix_fmt)
122 return isSupportedIn(pix_fmt);
125 #define isSupportedOut(x) ( \
126 (x)==PIX_FMT_YUV420P \
127 || (x)==PIX_FMT_YUVA420P \
128 || (x)==PIX_FMT_YUYV422 \
129 || (x)==PIX_FMT_UYVY422 \
130 || (x)==PIX_FMT_YUV444P \
131 || (x)==PIX_FMT_YUV422P \
132 || (x)==PIX_FMT_YUV411P \
133 || (x)==PIX_FMT_YUVJ420P \
134 || (x)==PIX_FMT_YUVJ422P \
135 || (x)==PIX_FMT_YUVJ440P \
136 || (x)==PIX_FMT_YUVJ444P \
138 || (x)==PIX_FMT_NV12 \
139 || (x)==PIX_FMT_NV21 \
140 || (x)==PIX_FMT_GRAY16BE \
141 || (x)==PIX_FMT_GRAY16LE \
142 || (x)==PIX_FMT_GRAY8 \
143 || (x)==PIX_FMT_YUV410P \
144 || (x)==PIX_FMT_YUV440P \
145 || (x)==PIX_FMT_YUV420P9LE \
146 || (x)==PIX_FMT_YUV420P10LE \
147 || (x)==PIX_FMT_YUV420P16LE \
148 || (x)==PIX_FMT_YUV422P16LE \
149 || (x)==PIX_FMT_YUV444P16LE \
150 || (x)==PIX_FMT_YUV420P9BE \
151 || (x)==PIX_FMT_YUV420P10BE \
152 || (x)==PIX_FMT_YUV420P16BE \
153 || (x)==PIX_FMT_YUV422P16BE \
154 || (x)==PIX_FMT_YUV444P16BE \
157 int sws_isSupportedOutput(enum PixelFormat pix_fmt)
159 return isSupportedOut(pix_fmt);
162 extern const int32_t ff_yuv2rgb_coeffs[8][4];
164 const char *sws_format_name(enum PixelFormat format)
166 if ((unsigned)format < PIX_FMT_NB && av_pix_fmt_descriptors[format].name)
167 return av_pix_fmt_descriptors[format].name;
169 return "Unknown format";
172 static double getSplineCoeff(double a, double b, double c, double d, double dist)
174 if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
175 else return getSplineCoeff( 0.0,
182 static int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
183 int srcW, int dstW, int filterAlign, int one, int flags, int cpu_flags,
184 SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
190 int64_t *filter=NULL;
191 int64_t *filter2=NULL;
192 const int64_t fone= 1LL<<54;
195 emms_c(); //FIXME this should not be required but it IS (even for non-MMX versions)
197 // NOTE: the +1 is for the MMX scaler which reads over the end
198 FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW+1)*sizeof(int16_t), fail);
200 if (FFABS(xInc - 0x10000) <10) { // unscaled
203 FF_ALLOCZ_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
205 for (i=0; i<dstW; i++) {
206 filter[i*filterSize]= fone;
210 } else if (flags&SWS_POINT) { // lame looking point sampling mode
214 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
216 xDstInSrc= xInc/2 - 0x8000;
217 for (i=0; i<dstW; i++) {
218 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
224 } else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) { // bilinear upscale
228 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
230 xDstInSrc= xInc/2 - 0x8000;
231 for (i=0; i<dstW; i++) {
232 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
236 //bilinear upscale / linear interpolate / area averaging
237 for (j=0; j<filterSize; j++) {
238 int64_t coeff= fone - FFABS((xx<<16) - xDstInSrc)*(fone>>16);
239 if (coeff<0) coeff=0;
240 filter[i*filterSize + j]= coeff;
249 if (flags&SWS_BICUBIC) sizeFactor= 4;
250 else if (flags&SWS_X) sizeFactor= 8;
251 else if (flags&SWS_AREA) sizeFactor= 1; //downscale only, for upscale it is bilinear
252 else if (flags&SWS_GAUSS) sizeFactor= 8; // infinite ;)
253 else if (flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? ceil(2*param[0]) : 6;
254 else if (flags&SWS_SINC) sizeFactor= 20; // infinite ;)
255 else if (flags&SWS_SPLINE) sizeFactor= 20; // infinite ;)
256 else if (flags&SWS_BILINEAR) sizeFactor= 2;
258 sizeFactor= 0; //GCC warning killer
262 if (xInc <= 1<<16) filterSize= 1 + sizeFactor; // upscale
263 else filterSize= 1 + (sizeFactor*srcW + dstW - 1)/ dstW;
265 if (filterSize > srcW-2) filterSize=srcW-2;
267 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
269 xDstInSrc= xInc - 0x10000;
270 for (i=0; i<dstW; i++) {
271 int xx= (xDstInSrc - ((filterSize-2)<<16)) / (1<<17);
274 for (j=0; j<filterSize; j++) {
275 int64_t d= ((int64_t)FFABS((xx<<17) - xDstInSrc))<<13;
281 floatd= d * (1.0/(1<<30));
283 if (flags & SWS_BICUBIC) {
284 int64_t B= (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1<<24);
285 int64_t C= (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1<<24);
286 int64_t dd = ( d*d)>>30;
287 int64_t ddd= (dd*d)>>30;
290 coeff = (12*(1<<24)-9*B-6*C)*ddd + (-18*(1<<24)+12*B+6*C)*dd + (6*(1<<24)-2*B)*(1<<30);
291 else if (d < 1LL<<31)
292 coeff = (-B-6*C)*ddd + (6*B+30*C)*dd + (-12*B-48*C)*d + (8*B+24*C)*(1<<30);
295 coeff *= fone>>(30+24);
297 /* else if (flags & SWS_X) {
298 double p= param ? param*0.01 : 0.3;
299 coeff = d ? sin(d*M_PI)/(d*M_PI) : 1.0;
300 coeff*= pow(2.0, - p*d*d);
302 else if (flags & SWS_X) {
303 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
307 c = cos(floatd*M_PI);
310 if (c<0.0) c= -pow(-c, A);
312 coeff= (c*0.5 + 0.5)*fone;
313 } else if (flags & SWS_AREA) {
314 int64_t d2= d - (1<<29);
315 if (d2*xInc < -(1LL<<(29+16))) coeff= 1.0 * (1LL<<(30+16));
316 else if (d2*xInc < (1LL<<(29+16))) coeff= -d2*xInc + (1LL<<(29+16));
318 coeff *= fone>>(30+16);
319 } else if (flags & SWS_GAUSS) {
320 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
321 coeff = (pow(2.0, - p*floatd*floatd))*fone;
322 } else if (flags & SWS_SINC) {
323 coeff = (d ? sin(floatd*M_PI)/(floatd*M_PI) : 1.0)*fone;
324 } else if (flags & SWS_LANCZOS) {
325 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
326 coeff = (d ? sin(floatd*M_PI)*sin(floatd*M_PI/p)/(floatd*floatd*M_PI*M_PI/p) : 1.0)*fone;
327 if (floatd>p) coeff=0;
328 } else if (flags & SWS_BILINEAR) {
330 if (coeff<0) coeff=0;
332 } else if (flags & SWS_SPLINE) {
333 double p=-2.196152422706632;
334 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, floatd) * fone;
336 coeff= 0.0; //GCC warning killer
340 filter[i*filterSize + j]= coeff;
347 /* apply src & dst Filter to filter -> filter2
350 assert(filterSize>0);
351 filter2Size= filterSize;
352 if (srcFilter) filter2Size+= srcFilter->length - 1;
353 if (dstFilter) filter2Size+= dstFilter->length - 1;
354 assert(filter2Size>0);
355 FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size*dstW*sizeof(*filter2), fail);
357 for (i=0; i<dstW; i++) {
361 for (k=0; k<srcFilter->length; k++) {
362 for (j=0; j<filterSize; j++)
363 filter2[i*filter2Size + k + j] += srcFilter->coeff[k]*filter[i*filterSize + j];
366 for (j=0; j<filterSize; j++)
367 filter2[i*filter2Size + j]= filter[i*filterSize + j];
371 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
375 /* try to reduce the filter-size (step1 find size and shift left) */
376 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
378 for (i=dstW-1; i>=0; i--) {
379 int min= filter2Size;
383 /* get rid of near zero elements on the left by shifting left */
384 for (j=0; j<filter2Size; j++) {
386 cutOff += FFABS(filter2[i*filter2Size]);
388 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
390 /* preserve monotonicity because the core can't handle the filter otherwise */
391 if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
393 // move filter coefficients left
394 for (k=1; k<filter2Size; k++)
395 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
396 filter2[i*filter2Size + k - 1]= 0;
401 /* count near zeros on the right */
402 for (j=filter2Size-1; j>0; j--) {
403 cutOff += FFABS(filter2[i*filter2Size + j]);
405 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
409 if (min>minFilterSize) minFilterSize= min;
412 if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) {
413 // we can handle the special case 4,
414 // so we don't want to go to the full 8
415 if (minFilterSize < 5)
418 // We really don't want to waste our time
419 // doing useless computation, so fall back on
420 // the scalar C code for very small filters.
421 // Vectorizing is worth it only if you have a
422 // decent-sized vector.
423 if (minFilterSize < 3)
427 if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
428 // special case for unscaled vertical filtering
429 if (minFilterSize == 1 && filterAlign == 2)
433 assert(minFilterSize > 0);
434 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
435 assert(filterSize > 0);
436 filter= av_malloc(filterSize*dstW*sizeof(*filter));
437 if (filterSize >= MAX_FILTER_SIZE*16/((flags&SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
439 *outFilterSize= filterSize;
441 if (flags&SWS_PRINT_INFO)
442 av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
443 /* try to reduce the filter-size (step2 reduce it) */
444 for (i=0; i<dstW; i++) {
447 for (j=0; j<filterSize; j++) {
448 if (j>=filter2Size) filter[i*filterSize + j]= 0;
449 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
450 if((flags & SWS_BITEXACT) && j>=minFilterSize)
451 filter[i*filterSize + j]= 0;
455 //FIXME try to align filterPos if possible
458 for (i=0; i<dstW; i++) {
460 if ((*filterPos)[i] < 0) {
461 // move filter coefficients left to compensate for filterPos
462 for (j=1; j<filterSize; j++) {
463 int left= FFMAX(j + (*filterPos)[i], 0);
464 filter[i*filterSize + left] += filter[i*filterSize + j];
465 filter[i*filterSize + j]=0;
470 if ((*filterPos)[i] + filterSize > srcW) {
471 int shift= (*filterPos)[i] + filterSize - srcW;
472 // move filter coefficients right to compensate for filterPos
473 for (j=filterSize-2; j>=0; j--) {
474 int right= FFMIN(j + shift, filterSize-1);
475 filter[i*filterSize +right] += filter[i*filterSize +j];
476 filter[i*filterSize +j]=0;
478 (*filterPos)[i]= srcW - filterSize;
482 // Note the +1 is for the MMX scaler which reads over the end
483 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
484 FF_ALLOCZ_OR_GOTO(NULL, *outFilter, *outFilterSize*(dstW+1)*sizeof(int16_t), fail);
486 /* normalize & store in outFilter */
487 for (i=0; i<dstW; i++) {
492 for (j=0; j<filterSize; j++) {
493 sum+= filter[i*filterSize + j];
495 sum= (sum + one/2)/ one;
496 for (j=0; j<*outFilterSize; j++) {
497 int64_t v= filter[i*filterSize + j] + error;
498 int intV= ROUNDED_DIV(v, sum);
499 (*outFilter)[i*(*outFilterSize) + j]= intV;
504 (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
505 for (i=0; i<*outFilterSize; i++) {
506 int j= dstW*(*outFilterSize);
507 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
518 static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode, int16_t *filter, int32_t *filterPos, int numSplits)
521 x86_reg imm8OfPShufW1A;
522 x86_reg imm8OfPShufW2A;
523 x86_reg fragmentLengthA;
525 x86_reg imm8OfPShufW1B;
526 x86_reg imm8OfPShufW2B;
527 x86_reg fragmentLengthB;
532 // create an optimized horizontal scaling routine
533 /* This scaler is made of runtime-generated MMX2 code using specially
534 * tuned pshufw instructions. For every four output pixels, if four
535 * input pixels are enough for the fast bilinear scaling, then a chunk
536 * of fragmentB is used. If five input pixels are needed, then a chunk
537 * of fragmentA is used.
546 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
547 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
548 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
549 "punpcklbw %%mm7, %%mm1 \n\t"
550 "punpcklbw %%mm7, %%mm0 \n\t"
551 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
553 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
555 "psubw %%mm1, %%mm0 \n\t"
556 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
557 "pmullw %%mm3, %%mm0 \n\t"
558 "psllw $7, %%mm1 \n\t"
559 "paddw %%mm1, %%mm0 \n\t"
561 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
563 "add $8, %%"REG_a" \n\t"
567 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
568 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
569 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
574 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
578 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
579 "=r" (fragmentLengthA)
586 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
587 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
588 "punpcklbw %%mm7, %%mm0 \n\t"
589 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
591 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
593 "psubw %%mm1, %%mm0 \n\t"
594 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
595 "pmullw %%mm3, %%mm0 \n\t"
596 "psllw $7, %%mm1 \n\t"
597 "paddw %%mm1, %%mm0 \n\t"
599 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
601 "add $8, %%"REG_a" \n\t"
605 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
606 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
607 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
612 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
616 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
617 "=r" (fragmentLengthB)
620 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
623 for (i=0; i<dstW/numSplits; i++) {
628 int b=((xpos+xInc)>>16) - xx;
629 int c=((xpos+xInc*2)>>16) - xx;
630 int d=((xpos+xInc*3)>>16) - xx;
632 uint8_t *fragment = (d+1<4) ? fragmentB : fragmentA;
633 x86_reg imm8OfPShufW1 = (d+1<4) ? imm8OfPShufW1B : imm8OfPShufW1A;
634 x86_reg imm8OfPShufW2 = (d+1<4) ? imm8OfPShufW2B : imm8OfPShufW2A;
635 x86_reg fragmentLength = (d+1<4) ? fragmentLengthB : fragmentLengthA;
636 int maxShift= 3-(d+inc);
640 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
641 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
642 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
643 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
646 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
648 filterCode[fragmentPos + imm8OfPShufW1]=
649 (a+inc) | ((b+inc)<<2) | ((c+inc)<<4) | ((d+inc)<<6);
650 filterCode[fragmentPos + imm8OfPShufW2]=
651 a | (b<<2) | (c<<4) | (d<<6);
653 if (i+4-inc>=dstW) shift=maxShift; //avoid overread
654 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
656 if (shift && i>=shift) {
657 filterCode[fragmentPos + imm8OfPShufW1]+= 0x55*shift;
658 filterCode[fragmentPos + imm8OfPShufW2]+= 0x55*shift;
659 filterPos[i/2]-=shift;
663 fragmentPos+= fragmentLength;
666 filterCode[fragmentPos]= RET;
671 filterPos[((i/2)+1)&(~1)]= xpos>>16; // needed to jump to the next part
673 return fragmentPos + 1;
675 #endif /* HAVE_MMX2 */
677 static void getSubSampleFactors(int *h, int *v, enum PixelFormat format)
679 *h = av_pix_fmt_descriptors[format].log2_chroma_w;
680 *v = av_pix_fmt_descriptors[format].log2_chroma_h;
683 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation)
685 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
686 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
688 c->brightness= brightness;
689 c->contrast = contrast;
690 c->saturation= saturation;
691 c->srcRange = srcRange;
692 c->dstRange = dstRange;
693 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
695 c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->dstFormat]);
696 c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->srcFormat]);
698 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
701 if (HAVE_ALTIVEC && av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)
702 ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness, contrast, saturation);
706 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation)
708 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
710 *inv_table = c->srcColorspaceTable;
711 *table = c->dstColorspaceTable;
712 *srcRange = c->srcRange;
713 *dstRange = c->dstRange;
714 *brightness= c->brightness;
715 *contrast = c->contrast;
716 *saturation= c->saturation;
721 static int handle_jpeg(enum PixelFormat *format)
724 case PIX_FMT_YUVJ420P: *format = PIX_FMT_YUV420P; return 1;
725 case PIX_FMT_YUVJ422P: *format = PIX_FMT_YUV422P; return 1;
726 case PIX_FMT_YUVJ444P: *format = PIX_FMT_YUV444P; return 1;
727 case PIX_FMT_YUVJ440P: *format = PIX_FMT_YUV440P; return 1;
732 SwsContext *sws_alloc_context(void)
734 SwsContext *c= av_mallocz(sizeof(SwsContext));
736 c->av_class = &sws_context_class;
737 av_opt_set_defaults(c);
742 int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)
745 int usesVFilter, usesHFilter;
747 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
752 int dst_stride = FFALIGN(dstW * sizeof(int16_t) + 16, 16), dst_stride_px = dst_stride >> 1;
753 int flags, cpu_flags;
754 enum PixelFormat srcFormat= c->srcFormat;
755 enum PixelFormat dstFormat= c->dstFormat;
757 cpu_flags = av_get_cpu_flags();
760 if (!rgb15to16) sws_rgb2rgb_init();
762 unscaled = (srcW == dstW && srcH == dstH);
764 if (!isSupportedIn(srcFormat)) {
765 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as input pixel format\n", sws_format_name(srcFormat));
766 return AVERROR(EINVAL);
768 if (!isSupportedOut(dstFormat)) {
769 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as output pixel format\n", sws_format_name(dstFormat));
770 return AVERROR(EINVAL);
773 i= flags & ( SWS_POINT
784 if(!i || (i & (i-1))) {
785 av_log(NULL, AV_LOG_ERROR, "swScaler: Exactly one scaler algorithm must be chosen\n");
786 return AVERROR(EINVAL);
789 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
790 av_log(NULL, AV_LOG_ERROR, "swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
791 srcW, srcH, dstW, dstH);
792 return AVERROR(EINVAL);
795 if (!dstFilter) dstFilter= &dummyFilter;
796 if (!srcFilter) srcFilter= &dummyFilter;
798 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
799 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
800 c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[dstFormat]);
801 c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[srcFormat]);
802 c->vRounder= 4* 0x0001000100010001ULL;
804 usesVFilter = (srcFilter->lumV && srcFilter->lumV->length>1) ||
805 (srcFilter->chrV && srcFilter->chrV->length>1) ||
806 (dstFilter->lumV && dstFilter->lumV->length>1) ||
807 (dstFilter->chrV && dstFilter->chrV->length>1);
808 usesHFilter = (srcFilter->lumH && srcFilter->lumH->length>1) ||
809 (srcFilter->chrH && srcFilter->chrH->length>1) ||
810 (dstFilter->lumH && dstFilter->lumH->length>1) ||
811 (dstFilter->chrH && dstFilter->chrH->length>1);
813 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
814 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
816 // reuse chroma for 2 pixels RGB/BGR unless user wants full chroma interpolation
817 if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
819 // drop some chroma lines if the user wants it
820 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
821 c->chrSrcVSubSample+= c->vChrDrop;
823 // drop every other pixel for chroma calculation unless user wants full chroma
824 if (isAnyRGB(srcFormat) && !(flags&SWS_FULL_CHR_H_INP)
825 && srcFormat!=PIX_FMT_RGB8 && srcFormat!=PIX_FMT_BGR8
826 && srcFormat!=PIX_FMT_RGB4 && srcFormat!=PIX_FMT_BGR4
827 && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE
828 && ((dstW>>c->chrDstHSubSample) <= (srcW>>1) || (flags&SWS_FAST_BILINEAR)))
829 c->chrSrcHSubSample=1;
831 // Note the -((-x)>>y) is so that we always round toward +inf.
832 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
833 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
834 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
835 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
837 /* unscaled special cases */
838 if (unscaled && !usesHFilter && !usesVFilter && (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
839 ff_get_unscaled_swscale(c);
842 if (flags&SWS_PRINT_INFO)
843 av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
844 sws_format_name(srcFormat), sws_format_name(dstFormat));
849 FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW, 16) * 2, fail);
850 if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2) {
851 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
852 if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR)) {
853 if (flags&SWS_PRINT_INFO)
854 av_log(c, AV_LOG_INFO, "output width is not a multiple of 32 -> no MMX2 scaler\n");
856 if (usesHFilter) c->canMMX2BeUsed=0;
861 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
862 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
864 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
865 // but only for the FAST_BILINEAR mode otherwise do correct scaling
866 // n-2 is the last chrominance sample available
867 // this is not perfect, but no one should notice the difference, the more correct variant
868 // would be like the vertical one, but that would require some special code for the
869 // first and last pixel
870 if (flags&SWS_FAST_BILINEAR) {
871 if (c->canMMX2BeUsed) {
875 //we don't use the x86 asm scaler if MMX is available
876 else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
877 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
878 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
882 /* precalculate horizontal scaler filter coefficients */
885 // can't downscale !!!
886 if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {
887 c->lumMmx2FilterCodeSize = initMMX2HScaler( dstW, c->lumXInc, NULL, NULL, NULL, 8);
888 c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc, NULL, NULL, NULL, 4);
891 c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
892 c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
893 #elif HAVE_VIRTUALALLOC
894 c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
895 c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
897 c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);
898 c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);
901 if (!c->lumMmx2FilterCode || !c->chrMmx2FilterCode)
902 return AVERROR(ENOMEM);
903 FF_ALLOCZ_OR_GOTO(c, c->hLumFilter , (dstW /8+8)*sizeof(int16_t), fail);
904 FF_ALLOCZ_OR_GOTO(c, c->hChrFilter , (c->chrDstW /4+8)*sizeof(int16_t), fail);
905 FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW /2/8+8)*sizeof(int32_t), fail);
906 FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW/2/4+8)*sizeof(int32_t), fail);
908 initMMX2HScaler( dstW, c->lumXInc, c->lumMmx2FilterCode, c->hLumFilter, c->hLumFilterPos, 8);
909 initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode, c->hChrFilter, c->hChrFilterPos, 4);
912 mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
913 mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
916 #endif /* HAVE_MMX2 */
918 const int filterAlign=
919 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :
920 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
923 if (initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
924 srcW , dstW, filterAlign, 1<<14,
925 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags, cpu_flags,
926 srcFilter->lumH, dstFilter->lumH, c->param) < 0)
928 if (initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
929 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
930 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, cpu_flags,
931 srcFilter->chrH, dstFilter->chrH, c->param) < 0)
934 } // initialize horizontal stuff
936 /* precalculate vertical scaler filter coefficients */
938 const int filterAlign=
939 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
940 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
943 if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
944 srcH , dstH, filterAlign, (1<<12),
945 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags, cpu_flags,
946 srcFilter->lumV, dstFilter->lumV, c->param) < 0)
948 if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
949 c->chrSrcH, c->chrDstH, filterAlign, (1<<12),
950 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, cpu_flags,
951 srcFilter->chrV, dstFilter->chrV, c->param) < 0)
955 FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof (vector signed short)*c->vLumFilterSize*c->dstH, fail);
956 FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH, fail);
958 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
960 short *p = (short *)&c->vYCoeffsBank[i];
962 p[j] = c->vLumFilter[i];
965 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
967 short *p = (short *)&c->vCCoeffsBank[i];
969 p[j] = c->vChrFilter[i];
974 // calculate buffer sizes so that they won't run out while handling these damn slices
975 c->vLumBufSize= c->vLumFilterSize;
976 c->vChrBufSize= c->vChrFilterSize;
977 for (i=0; i<dstH; i++) {
978 int chrI= i*c->chrDstH / dstH;
979 int nextSlice= FFMAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
980 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
982 nextSlice>>= c->chrSrcVSubSample;
983 nextSlice<<= c->chrSrcVSubSample;
984 if (c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
985 c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
986 if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
987 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
990 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
991 // allocate several megabytes to handle all possible cases)
992 FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
993 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
994 FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
995 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
996 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
997 //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)
998 /* align at 16 bytes for AltiVec */
999 for (i=0; i<c->vLumBufSize; i++) {
1000 FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], dst_stride+1, fail);
1001 c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];
1003 c->uv_off = dst_stride_px;
1004 c->uv_offx2 = dst_stride;
1005 for (i=0; i<c->vChrBufSize; i++) {
1006 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i+c->vChrBufSize], dst_stride*2+1, fail);
1007 c->chrUPixBuf[i] = c->chrUPixBuf[i+c->vChrBufSize];
1008 c->chrVPixBuf[i] = c->chrVPixBuf[i+c->vChrBufSize] = c->chrUPixBuf[i] + dst_stride_px;
1010 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
1011 for (i=0; i<c->vLumBufSize; i++) {
1012 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], dst_stride+1, fail);
1013 c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];
1016 //try to avoid drawing green stuff between the right end and the stride end
1017 for (i=0; i<c->vChrBufSize; i++)
1018 memset(c->chrUPixBuf[i], 64, dst_stride*2+1);
1020 assert(c->chrDstH <= dstH);
1022 if (flags&SWS_PRINT_INFO) {
1023 if (flags&SWS_FAST_BILINEAR) av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
1024 else if (flags&SWS_BILINEAR) av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
1025 else if (flags&SWS_BICUBIC) av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
1026 else if (flags&SWS_X) av_log(c, AV_LOG_INFO, "Experimental scaler, ");
1027 else if (flags&SWS_POINT) av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
1028 else if (flags&SWS_AREA) av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
1029 else if (flags&SWS_BICUBLIN) av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
1030 else if (flags&SWS_GAUSS) av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
1031 else if (flags&SWS_SINC) av_log(c, AV_LOG_INFO, "Sinc scaler, ");
1032 else if (flags&SWS_LANCZOS) av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
1033 else if (flags&SWS_SPLINE) av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
1034 else av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
1036 av_log(c, AV_LOG_INFO, "from %s to %s%s ",
1037 sws_format_name(srcFormat),
1039 dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ||
1040 dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||
1041 dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ? "dithered " : "",
1045 sws_format_name(dstFormat));
1047 if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2) av_log(c, AV_LOG_INFO, "using MMX2\n");
1048 else if (HAVE_AMD3DNOW && cpu_flags & AV_CPU_FLAG_3DNOW) av_log(c, AV_LOG_INFO, "using 3DNOW\n");
1049 else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) av_log(c, AV_LOG_INFO, "using MMX\n");
1050 else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) av_log(c, AV_LOG_INFO, "using AltiVec\n");
1051 else av_log(c, AV_LOG_INFO, "using C\n");
1053 if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
1054 if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
1055 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
1057 if (c->hLumFilterSize==4)
1058 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal luminance scaling\n");
1059 else if (c->hLumFilterSize==8)
1060 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal luminance scaling\n");
1062 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal luminance scaling\n");
1064 if (c->hChrFilterSize==4)
1065 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal chrominance scaling\n");
1066 else if (c->hChrFilterSize==8)
1067 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal chrominance scaling\n");
1069 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal chrominance scaling\n");
1073 av_log(c, AV_LOG_VERBOSE, "using x86 asm scaler for horizontal scaling\n");
1075 if (flags & SWS_FAST_BILINEAR)
1076 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR C scaler for horizontal scaling\n");
1078 av_log(c, AV_LOG_VERBOSE, "using C scaler for horizontal scaling\n");
1081 if (isPlanarYUV(dstFormat)) {
1082 if (c->vLumFilterSize==1)
1083 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n",
1084 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
1086 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n",
1087 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
1089 if (c->vLumFilterSize==1 && c->vChrFilterSize==2)
1090 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
1091 " 2-tap scaler for vertical chrominance scaling (BGR)\n",
1092 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
1093 else if (c->vLumFilterSize==2 && c->vChrFilterSize==2)
1094 av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n",
1095 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
1097 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n",
1098 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
1101 if (dstFormat==PIX_FMT_BGR24)
1102 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 converter\n",
1103 (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2) ? "MMX2" :
1104 ((HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C"));
1105 else if (dstFormat==PIX_FMT_RGB32)
1106 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 converter\n",
1107 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
1108 else if (dstFormat==PIX_FMT_BGR565)
1109 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 converter\n",
1110 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
1111 else if (dstFormat==PIX_FMT_BGR555)
1112 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 converter\n",
1113 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
1114 else if (dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||
1115 dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE)
1116 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR12 converter\n",
1117 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
1119 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
1120 av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1121 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
1122 av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1123 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
1126 c->swScale= ff_getSwsFunc(c);
1128 fail: //FIXME replace things by appropriate error codes
1132 #if FF_API_SWS_GETCONTEXT
1133 SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,
1134 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
1135 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
1139 if(!(c=sws_alloc_context()))
1147 c->srcRange = handle_jpeg(&srcFormat);
1148 c->dstRange = handle_jpeg(&dstFormat);
1149 c->srcFormat= srcFormat;
1150 c->dstFormat= dstFormat;
1153 c->param[0] = param[0];
1154 c->param[1] = param[1];
1156 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);
1158 if(sws_init_context(c, srcFilter, dstFilter) < 0){
1167 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
1168 float lumaSharpen, float chromaSharpen,
1169 float chromaHShift, float chromaVShift,
1172 SwsFilter *filter= av_malloc(sizeof(SwsFilter));
1176 if (lumaGBlur!=0.0) {
1177 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
1178 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
1180 filter->lumH= sws_getIdentityVec();
1181 filter->lumV= sws_getIdentityVec();
1184 if (chromaGBlur!=0.0) {
1185 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
1186 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
1188 filter->chrH= sws_getIdentityVec();
1189 filter->chrV= sws_getIdentityVec();
1192 if (chromaSharpen!=0.0) {
1193 SwsVector *id= sws_getIdentityVec();
1194 sws_scaleVec(filter->chrH, -chromaSharpen);
1195 sws_scaleVec(filter->chrV, -chromaSharpen);
1196 sws_addVec(filter->chrH, id);
1197 sws_addVec(filter->chrV, id);
1201 if (lumaSharpen!=0.0) {
1202 SwsVector *id= sws_getIdentityVec();
1203 sws_scaleVec(filter->lumH, -lumaSharpen);
1204 sws_scaleVec(filter->lumV, -lumaSharpen);
1205 sws_addVec(filter->lumH, id);
1206 sws_addVec(filter->lumV, id);
1210 if (chromaHShift != 0.0)
1211 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
1213 if (chromaVShift != 0.0)
1214 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
1216 sws_normalizeVec(filter->chrH, 1.0);
1217 sws_normalizeVec(filter->chrV, 1.0);
1218 sws_normalizeVec(filter->lumH, 1.0);
1219 sws_normalizeVec(filter->lumV, 1.0);
1221 if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
1222 if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
1227 SwsVector *sws_allocVec(int length)
1229 SwsVector *vec = av_malloc(sizeof(SwsVector));
1232 vec->length = length;
1233 vec->coeff = av_malloc(sizeof(double) * length);
1239 SwsVector *sws_getGaussianVec(double variance, double quality)
1241 const int length= (int)(variance*quality + 0.5) | 1;
1243 double middle= (length-1)*0.5;
1244 SwsVector *vec= sws_allocVec(length);
1249 for (i=0; i<length; i++) {
1250 double dist= i-middle;
1251 vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);
1254 sws_normalizeVec(vec, 1.0);
1259 SwsVector *sws_getConstVec(double c, int length)
1262 SwsVector *vec= sws_allocVec(length);
1267 for (i=0; i<length; i++)
1273 SwsVector *sws_getIdentityVec(void)
1275 return sws_getConstVec(1.0, 1);
1278 static double sws_dcVec(SwsVector *a)
1283 for (i=0; i<a->length; i++)
1289 void sws_scaleVec(SwsVector *a, double scalar)
1293 for (i=0; i<a->length; i++)
1294 a->coeff[i]*= scalar;
1297 void sws_normalizeVec(SwsVector *a, double height)
1299 sws_scaleVec(a, height/sws_dcVec(a));
1302 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
1304 int length= a->length + b->length - 1;
1306 SwsVector *vec= sws_getConstVec(0.0, length);
1311 for (i=0; i<a->length; i++) {
1312 for (j=0; j<b->length; j++) {
1313 vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];
1320 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
1322 int length= FFMAX(a->length, b->length);
1324 SwsVector *vec= sws_getConstVec(0.0, length);
1329 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1330 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
1335 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
1337 int length= FFMAX(a->length, b->length);
1339 SwsVector *vec= sws_getConstVec(0.0, length);
1344 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1345 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
1350 /* shift left / or right if "shift" is negative */
1351 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
1353 int length= a->length + FFABS(shift)*2;
1355 SwsVector *vec= sws_getConstVec(0.0, length);
1360 for (i=0; i<a->length; i++) {
1361 vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
1367 void sws_shiftVec(SwsVector *a, int shift)
1369 SwsVector *shifted= sws_getShiftedVec(a, shift);
1371 a->coeff= shifted->coeff;
1372 a->length= shifted->length;
1376 void sws_addVec(SwsVector *a, SwsVector *b)
1378 SwsVector *sum= sws_sumVec(a, b);
1380 a->coeff= sum->coeff;
1381 a->length= sum->length;
1385 void sws_subVec(SwsVector *a, SwsVector *b)
1387 SwsVector *diff= sws_diffVec(a, b);
1389 a->coeff= diff->coeff;
1390 a->length= diff->length;
1394 void sws_convVec(SwsVector *a, SwsVector *b)
1396 SwsVector *conv= sws_getConvVec(a, b);
1398 a->coeff= conv->coeff;
1399 a->length= conv->length;
1403 SwsVector *sws_cloneVec(SwsVector *a)
1406 SwsVector *vec= sws_allocVec(a->length);
1411 for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];
1416 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
1423 for (i=0; i<a->length; i++)
1424 if (a->coeff[i]>max) max= a->coeff[i];
1426 for (i=0; i<a->length; i++)
1427 if (a->coeff[i]<min) min= a->coeff[i];
1431 for (i=0; i<a->length; i++) {
1432 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
1433 av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
1434 for (;x>0; x--) av_log(log_ctx, log_level, " ");
1435 av_log(log_ctx, log_level, "|\n");
1439 void sws_freeVec(SwsVector *a)
1442 av_freep(&a->coeff);
1447 void sws_freeFilter(SwsFilter *filter)
1449 if (!filter) return;
1451 if (filter->lumH) sws_freeVec(filter->lumH);
1452 if (filter->lumV) sws_freeVec(filter->lumV);
1453 if (filter->chrH) sws_freeVec(filter->chrH);
1454 if (filter->chrV) sws_freeVec(filter->chrV);
1458 void sws_freeContext(SwsContext *c)
1464 for (i=0; i<c->vLumBufSize; i++)
1465 av_freep(&c->lumPixBuf[i]);
1466 av_freep(&c->lumPixBuf);
1469 if (c->chrUPixBuf) {
1470 for (i=0; i<c->vChrBufSize; i++)
1471 av_freep(&c->chrUPixBuf[i]);
1472 av_freep(&c->chrUPixBuf);
1473 av_freep(&c->chrVPixBuf);
1476 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1477 for (i=0; i<c->vLumBufSize; i++)
1478 av_freep(&c->alpPixBuf[i]);
1479 av_freep(&c->alpPixBuf);
1482 av_freep(&c->vLumFilter);
1483 av_freep(&c->vChrFilter);
1484 av_freep(&c->hLumFilter);
1485 av_freep(&c->hChrFilter);
1487 av_freep(&c->vYCoeffsBank);
1488 av_freep(&c->vCCoeffsBank);
1491 av_freep(&c->vLumFilterPos);
1492 av_freep(&c->vChrFilterPos);
1493 av_freep(&c->hLumFilterPos);
1494 av_freep(&c->hChrFilterPos);
1497 #ifdef MAP_ANONYMOUS
1498 if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);
1499 if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);
1500 #elif HAVE_VIRTUALALLOC
1501 if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);
1502 if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);
1504 av_free(c->lumMmx2FilterCode);
1505 av_free(c->chrMmx2FilterCode);
1507 c->lumMmx2FilterCode=NULL;
1508 c->chrMmx2FilterCode=NULL;
1509 #endif /* HAVE_MMX */
1511 av_freep(&c->yuvTable);
1512 av_free(c->formatConvBuffer);
1517 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
1518 int srcW, int srcH, enum PixelFormat srcFormat,
1519 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
1520 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
1522 static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};
1525 param = default_param;
1528 (context->srcW != srcW ||
1529 context->srcH != srcH ||
1530 context->srcFormat != srcFormat ||
1531 context->dstW != dstW ||
1532 context->dstH != dstH ||
1533 context->dstFormat != dstFormat ||
1534 context->flags != flags ||
1535 context->param[0] != param[0] ||
1536 context->param[1] != param[1])) {
1537 sws_freeContext(context);
1542 if (!(context = sws_alloc_context()))
1544 context->srcW = srcW;
1545 context->srcH = srcH;
1546 context->srcRange = handle_jpeg(&srcFormat);
1547 context->srcFormat = srcFormat;
1548 context->dstW = dstW;
1549 context->dstH = dstH;
1550 context->dstRange = handle_jpeg(&dstFormat);
1551 context->dstFormat = dstFormat;
1552 context->flags = flags;
1553 context->param[0] = param[0];
1554 context->param[1] = param[1];
1555 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);
1556 if (sws_init_context(context, srcFilter, dstFilter) < 0) {
1557 sws_freeContext(context);