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
30 #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
31 #define MAP_ANONYMOUS MAP_ANON
35 #define WIN32_LEAN_AND_MEAN
39 #include "swscale_internal.h"
41 #include "libavutil/intreadwrite.h"
42 #include "libavutil/x86_cpu.h"
43 #include "libavutil/avutil.h"
44 #include "libavutil/bswap.h"
45 #include "libavutil/pixdesc.h"
47 unsigned swscale_version(void)
49 return LIBSWSCALE_VERSION_INT;
52 const char *swscale_configuration(void)
54 return FFMPEG_CONFIGURATION;
57 const char *swscale_license(void)
59 #define LICENSE_PREFIX "libswscale license: "
60 return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
63 #define RET 0xC3 //near return opcode for x86
65 #define isSupportedIn(x) ( \
66 (x)==PIX_FMT_YUV420P \
67 || (x)==PIX_FMT_YUVA420P \
68 || (x)==PIX_FMT_YUYV422 \
69 || (x)==PIX_FMT_UYVY422 \
70 || (x)==PIX_FMT_RGB48BE \
71 || (x)==PIX_FMT_RGB48LE \
72 || (x)==PIX_FMT_RGB32 \
73 || (x)==PIX_FMT_RGB32_1 \
74 || (x)==PIX_FMT_BGR24 \
75 || (x)==PIX_FMT_BGR565 \
76 || (x)==PIX_FMT_BGR555 \
77 || (x)==PIX_FMT_BGR32 \
78 || (x)==PIX_FMT_BGR32_1 \
79 || (x)==PIX_FMT_RGB24 \
80 || (x)==PIX_FMT_RGB565 \
81 || (x)==PIX_FMT_RGB555 \
82 || (x)==PIX_FMT_GRAY8 \
83 || (x)==PIX_FMT_YUV410P \
84 || (x)==PIX_FMT_YUV440P \
85 || (x)==PIX_FMT_NV12 \
86 || (x)==PIX_FMT_NV21 \
87 || (x)==PIX_FMT_GRAY16BE \
88 || (x)==PIX_FMT_GRAY16LE \
89 || (x)==PIX_FMT_YUV444P \
90 || (x)==PIX_FMT_YUV422P \
91 || (x)==PIX_FMT_YUV411P \
92 || (x)==PIX_FMT_YUVJ420P \
93 || (x)==PIX_FMT_YUVJ422P \
94 || (x)==PIX_FMT_YUVJ440P \
95 || (x)==PIX_FMT_YUVJ444P \
96 || (x)==PIX_FMT_PAL8 \
97 || (x)==PIX_FMT_BGR8 \
98 || (x)==PIX_FMT_RGB8 \
99 || (x)==PIX_FMT_BGR4_BYTE \
100 || (x)==PIX_FMT_RGB4_BYTE \
101 || (x)==PIX_FMT_YUV440P \
102 || (x)==PIX_FMT_MONOWHITE \
103 || (x)==PIX_FMT_MONOBLACK \
104 || (x)==PIX_FMT_YUV420P16LE \
105 || (x)==PIX_FMT_YUV422P16LE \
106 || (x)==PIX_FMT_YUV444P16LE \
107 || (x)==PIX_FMT_YUV420P16BE \
108 || (x)==PIX_FMT_YUV422P16BE \
109 || (x)==PIX_FMT_YUV444P16BE \
112 int sws_isSupportedInput(enum PixelFormat pix_fmt)
114 return isSupportedIn(pix_fmt);
117 #define isSupportedOut(x) ( \
118 (x)==PIX_FMT_YUV420P \
119 || (x)==PIX_FMT_YUVA420P \
120 || (x)==PIX_FMT_YUYV422 \
121 || (x)==PIX_FMT_UYVY422 \
122 || (x)==PIX_FMT_YUV444P \
123 || (x)==PIX_FMT_YUV422P \
124 || (x)==PIX_FMT_YUV411P \
125 || (x)==PIX_FMT_YUVJ420P \
126 || (x)==PIX_FMT_YUVJ422P \
127 || (x)==PIX_FMT_YUVJ440P \
128 || (x)==PIX_FMT_YUVJ444P \
130 || (x)==PIX_FMT_NV12 \
131 || (x)==PIX_FMT_NV21 \
132 || (x)==PIX_FMT_GRAY16BE \
133 || (x)==PIX_FMT_GRAY16LE \
134 || (x)==PIX_FMT_GRAY8 \
135 || (x)==PIX_FMT_YUV410P \
136 || (x)==PIX_FMT_YUV440P \
137 || (x)==PIX_FMT_YUV420P16LE \
138 || (x)==PIX_FMT_YUV422P16LE \
139 || (x)==PIX_FMT_YUV444P16LE \
140 || (x)==PIX_FMT_YUV420P16BE \
141 || (x)==PIX_FMT_YUV422P16BE \
142 || (x)==PIX_FMT_YUV444P16BE \
145 int sws_isSupportedOutput(enum PixelFormat pix_fmt)
147 return isSupportedOut(pix_fmt);
150 extern const int32_t ff_yuv2rgb_coeffs[8][4];
152 const char *sws_format_name(enum PixelFormat format)
154 if ((unsigned)format < PIX_FMT_NB && av_pix_fmt_descriptors[format].name)
155 return av_pix_fmt_descriptors[format].name;
157 return "Unknown format";
160 static double getSplineCoeff(double a, double b, double c, double d, double dist)
162 // printf("%f %f %f %f %f\n", a,b,c,d,dist);
163 if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
164 else return getSplineCoeff( 0.0,
171 static int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
172 int srcW, int dstW, int filterAlign, int one, int flags,
173 SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
179 int64_t *filter=NULL;
180 int64_t *filter2=NULL;
181 const int64_t fone= 1LL<<54;
184 if (flags & SWS_CPU_CAPS_MMX)
185 __asm__ volatile("emms\n\t"::: "memory"); //FIXME this should not be required but it IS (even for non-MMX versions)
188 // NOTE: the +1 is for the MMX scaler which reads over the end
189 FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW+1)*sizeof(int16_t), fail);
191 if (FFABS(xInc - 0x10000) <10) { // unscaled
194 FF_ALLOCZ_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
196 for (i=0; i<dstW; i++) {
197 filter[i*filterSize]= fone;
201 } else if (flags&SWS_POINT) { // lame looking point sampling mode
205 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
207 xDstInSrc= xInc/2 - 0x8000;
208 for (i=0; i<dstW; i++) {
209 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
215 } else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) { // bilinear upscale
219 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
221 xDstInSrc= xInc/2 - 0x8000;
222 for (i=0; i<dstW; i++) {
223 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
227 //bilinear upscale / linear interpolate / area averaging
228 for (j=0; j<filterSize; j++) {
229 int64_t coeff= fone - FFABS((xx<<16) - xDstInSrc)*(fone>>16);
230 if (coeff<0) coeff=0;
231 filter[i*filterSize + j]= coeff;
240 if (flags&SWS_BICUBIC) sizeFactor= 4;
241 else if (flags&SWS_X) sizeFactor= 8;
242 else if (flags&SWS_AREA) sizeFactor= 1; //downscale only, for upscale it is bilinear
243 else if (flags&SWS_GAUSS) sizeFactor= 8; // infinite ;)
244 else if (flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? ceil(2*param[0]) : 6;
245 else if (flags&SWS_SINC) sizeFactor= 20; // infinite ;)
246 else if (flags&SWS_SPLINE) sizeFactor= 20; // infinite ;)
247 else if (flags&SWS_BILINEAR) sizeFactor= 2;
249 sizeFactor= 0; //GCC warning killer
253 if (xInc <= 1<<16) filterSize= 1 + sizeFactor; // upscale
254 else filterSize= 1 + (sizeFactor*srcW + dstW - 1)/ dstW;
256 if (filterSize > srcW-2) filterSize=srcW-2;
258 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
260 xDstInSrc= xInc - 0x10000;
261 for (i=0; i<dstW; i++) {
262 int xx= (xDstInSrc - ((filterSize-2)<<16)) / (1<<17);
265 for (j=0; j<filterSize; j++) {
266 int64_t d= ((int64_t)FFABS((xx<<17) - xDstInSrc))<<13;
272 floatd= d * (1.0/(1<<30));
274 if (flags & SWS_BICUBIC) {
275 int64_t B= (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1<<24);
276 int64_t C= (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1<<24);
277 int64_t dd = ( d*d)>>30;
278 int64_t ddd= (dd*d)>>30;
281 coeff = (12*(1<<24)-9*B-6*C)*ddd + (-18*(1<<24)+12*B+6*C)*dd + (6*(1<<24)-2*B)*(1<<30);
282 else if (d < 1LL<<31)
283 coeff = (-B-6*C)*ddd + (6*B+30*C)*dd + (-12*B-48*C)*d + (8*B+24*C)*(1<<30);
286 coeff *= fone>>(30+24);
288 /* else if (flags & SWS_X) {
289 double p= param ? param*0.01 : 0.3;
290 coeff = d ? sin(d*M_PI)/(d*M_PI) : 1.0;
291 coeff*= pow(2.0, - p*d*d);
293 else if (flags & SWS_X) {
294 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
298 c = cos(floatd*M_PI);
301 if (c<0.0) c= -pow(-c, A);
303 coeff= (c*0.5 + 0.5)*fone;
304 } else if (flags & SWS_AREA) {
305 int64_t d2= d - (1<<29);
306 if (d2*xInc < -(1LL<<(29+16))) coeff= 1.0 * (1LL<<(30+16));
307 else if (d2*xInc < (1LL<<(29+16))) coeff= -d2*xInc + (1LL<<(29+16));
309 coeff *= fone>>(30+16);
310 } else if (flags & SWS_GAUSS) {
311 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
312 coeff = (pow(2.0, - p*floatd*floatd))*fone;
313 } else if (flags & SWS_SINC) {
314 coeff = (d ? sin(floatd*M_PI)/(floatd*M_PI) : 1.0)*fone;
315 } else if (flags & SWS_LANCZOS) {
316 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
317 coeff = (d ? sin(floatd*M_PI)*sin(floatd*M_PI/p)/(floatd*floatd*M_PI*M_PI/p) : 1.0)*fone;
318 if (floatd>p) coeff=0;
319 } else if (flags & SWS_BILINEAR) {
321 if (coeff<0) coeff=0;
323 } else if (flags & SWS_SPLINE) {
324 double p=-2.196152422706632;
325 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, floatd) * fone;
327 coeff= 0.0; //GCC warning killer
331 filter[i*filterSize + j]= coeff;
338 /* apply src & dst Filter to filter -> filter2
341 assert(filterSize>0);
342 filter2Size= filterSize;
343 if (srcFilter) filter2Size+= srcFilter->length - 1;
344 if (dstFilter) filter2Size+= dstFilter->length - 1;
345 assert(filter2Size>0);
346 FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size*dstW*sizeof(*filter2), fail);
348 for (i=0; i<dstW; i++) {
352 for (k=0; k<srcFilter->length; k++) {
353 for (j=0; j<filterSize; j++)
354 filter2[i*filter2Size + k + j] += srcFilter->coeff[k]*filter[i*filterSize + j];
357 for (j=0; j<filterSize; j++)
358 filter2[i*filter2Size + j]= filter[i*filterSize + j];
362 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
366 /* try to reduce the filter-size (step1 find size and shift left) */
367 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
369 for (i=dstW-1; i>=0; i--) {
370 int min= filter2Size;
374 /* get rid of near zero elements on the left by shifting left */
375 for (j=0; j<filter2Size; j++) {
377 cutOff += FFABS(filter2[i*filter2Size]);
379 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
381 /* preserve monotonicity because the core can't handle the filter otherwise */
382 if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
384 // move filter coefficients left
385 for (k=1; k<filter2Size; k++)
386 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
387 filter2[i*filter2Size + k - 1]= 0;
392 /* count near zeros on the right */
393 for (j=filter2Size-1; j>0; j--) {
394 cutOff += FFABS(filter2[i*filter2Size + j]);
396 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
400 if (min>minFilterSize) minFilterSize= min;
403 if (flags & SWS_CPU_CAPS_ALTIVEC) {
404 // we can handle the special case 4,
405 // so we don't want to go to the full 8
406 if (minFilterSize < 5)
409 // We really don't want to waste our time
410 // doing useless computation, so fall back on
411 // the scalar C code for very small filters.
412 // Vectorizing is worth it only if you have a
413 // decent-sized vector.
414 if (minFilterSize < 3)
418 if (flags & SWS_CPU_CAPS_MMX) {
419 // special case for unscaled vertical filtering
420 if (minFilterSize == 1 && filterAlign == 2)
424 assert(minFilterSize > 0);
425 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
426 assert(filterSize > 0);
427 filter= av_malloc(filterSize*dstW*sizeof(*filter));
428 if (filterSize >= MAX_FILTER_SIZE*16/((flags&SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
430 *outFilterSize= filterSize;
432 if (flags&SWS_PRINT_INFO)
433 av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
434 /* try to reduce the filter-size (step2 reduce it) */
435 for (i=0; i<dstW; i++) {
438 for (j=0; j<filterSize; j++) {
439 if (j>=filter2Size) filter[i*filterSize + j]= 0;
440 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
441 if((flags & SWS_BITEXACT) && j>=minFilterSize)
442 filter[i*filterSize + j]= 0;
446 //FIXME try to align filterPos if possible
449 for (i=0; i<dstW; i++) {
451 if ((*filterPos)[i] < 0) {
452 // move filter coefficients left to compensate for filterPos
453 for (j=1; j<filterSize; j++) {
454 int left= FFMAX(j + (*filterPos)[i], 0);
455 filter[i*filterSize + left] += filter[i*filterSize + j];
456 filter[i*filterSize + j]=0;
461 if ((*filterPos)[i] + filterSize > srcW) {
462 int shift= (*filterPos)[i] + filterSize - srcW;
463 // move filter coefficients right to compensate for filterPos
464 for (j=filterSize-2; j>=0; j--) {
465 int right= FFMIN(j + shift, filterSize-1);
466 filter[i*filterSize +right] += filter[i*filterSize +j];
467 filter[i*filterSize +j]=0;
469 (*filterPos)[i]= srcW - filterSize;
473 // Note the +1 is for the MMX scaler which reads over the end
474 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
475 FF_ALLOCZ_OR_GOTO(NULL, *outFilter, *outFilterSize*(dstW+1)*sizeof(int16_t), fail);
477 /* normalize & store in outFilter */
478 for (i=0; i<dstW; i++) {
483 for (j=0; j<filterSize; j++) {
484 sum+= filter[i*filterSize + j];
486 sum= (sum + one/2)/ one;
487 for (j=0; j<*outFilterSize; j++) {
488 int64_t v= filter[i*filterSize + j] + error;
489 int intV= ROUNDED_DIV(v, sum);
490 (*outFilter)[i*(*outFilterSize) + j]= intV;
495 (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
496 for (i=0; i<*outFilterSize; i++) {
497 int j= dstW*(*outFilterSize);
498 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
508 #if ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT)
509 static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode, int16_t *filter, int32_t *filterPos, int numSplits)
512 x86_reg imm8OfPShufW1A;
513 x86_reg imm8OfPShufW2A;
514 x86_reg fragmentLengthA;
516 x86_reg imm8OfPShufW1B;
517 x86_reg imm8OfPShufW2B;
518 x86_reg fragmentLengthB;
523 // create an optimized horizontal scaling routine
524 /* This scaler is made of runtime-generated MMX2 code using specially
525 * tuned pshufw instructions. For every four output pixels, if four
526 * input pixels are enough for the fast bilinear scaling, then a chunk
527 * of fragmentB is used. If five input pixels are needed, then a chunk
528 * of fragmentA is used.
537 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
538 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
539 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
540 "punpcklbw %%mm7, %%mm1 \n\t"
541 "punpcklbw %%mm7, %%mm0 \n\t"
542 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
544 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
546 "psubw %%mm1, %%mm0 \n\t"
547 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
548 "pmullw %%mm3, %%mm0 \n\t"
549 "psllw $7, %%mm1 \n\t"
550 "paddw %%mm1, %%mm0 \n\t"
552 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
554 "add $8, %%"REG_a" \n\t"
558 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
559 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
560 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
565 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
569 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
570 "=r" (fragmentLengthA)
577 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
578 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
579 "punpcklbw %%mm7, %%mm0 \n\t"
580 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
582 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
584 "psubw %%mm1, %%mm0 \n\t"
585 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
586 "pmullw %%mm3, %%mm0 \n\t"
587 "psllw $7, %%mm1 \n\t"
588 "paddw %%mm1, %%mm0 \n\t"
590 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
592 "add $8, %%"REG_a" \n\t"
596 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
597 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
598 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
603 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
607 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
608 "=r" (fragmentLengthB)
611 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
614 for (i=0; i<dstW/numSplits; i++) {
619 int b=((xpos+xInc)>>16) - xx;
620 int c=((xpos+xInc*2)>>16) - xx;
621 int d=((xpos+xInc*3)>>16) - xx;
623 uint8_t *fragment = (d+1<4) ? fragmentB : fragmentA;
624 x86_reg imm8OfPShufW1 = (d+1<4) ? imm8OfPShufW1B : imm8OfPShufW1A;
625 x86_reg imm8OfPShufW2 = (d+1<4) ? imm8OfPShufW2B : imm8OfPShufW2A;
626 x86_reg fragmentLength = (d+1<4) ? fragmentLengthB : fragmentLengthA;
627 int maxShift= 3-(d+inc);
631 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
632 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
633 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
634 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
637 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
639 filterCode[fragmentPos + imm8OfPShufW1]=
640 (a+inc) | ((b+inc)<<2) | ((c+inc)<<4) | ((d+inc)<<6);
641 filterCode[fragmentPos + imm8OfPShufW2]=
642 a | (b<<2) | (c<<4) | (d<<6);
644 if (i+4-inc>=dstW) shift=maxShift; //avoid overread
645 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
647 if (shift && i>=shift) {
648 filterCode[fragmentPos + imm8OfPShufW1]+= 0x55*shift;
649 filterCode[fragmentPos + imm8OfPShufW2]+= 0x55*shift;
650 filterPos[i/2]-=shift;
654 fragmentPos+= fragmentLength;
657 filterCode[fragmentPos]= RET;
662 filterPos[((i/2)+1)&(~1)]= xpos>>16; // needed to jump to the next part
664 return fragmentPos + 1;
666 #endif /* ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT) */
668 static void getSubSampleFactors(int *h, int *v, enum PixelFormat format)
670 *h = av_pix_fmt_descriptors[format].log2_chroma_w;
671 *v = av_pix_fmt_descriptors[format].log2_chroma_h;
674 static uint16_t roundToInt16(int64_t f)
676 int r= (f + (1<<15))>>16;
677 if (r<-0x7FFF) return 0x8000;
678 else if (r> 0x7FFF) return 0x7FFF;
682 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation)
684 int64_t crv = inv_table[0];
685 int64_t cbu = inv_table[1];
686 int64_t cgu = -inv_table[2];
687 int64_t cgv = -inv_table[3];
691 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
692 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
694 c->brightness= brightness;
695 c->contrast = contrast;
696 c->saturation= saturation;
697 c->srcRange = srcRange;
698 c->dstRange = dstRange;
699 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
701 c->uOffset= 0x0400040004000400LL;
702 c->vOffset= 0x0400040004000400LL;
708 crv= (crv*224) / 255;
709 cbu= (cbu*224) / 255;
710 cgu= (cgu*224) / 255;
711 cgv= (cgv*224) / 255;
714 cy = (cy *contrast )>>16;
715 crv= (crv*contrast * saturation)>>32;
716 cbu= (cbu*contrast * saturation)>>32;
717 cgu= (cgu*contrast * saturation)>>32;
718 cgv= (cgv*contrast * saturation)>>32;
720 oy -= 256*brightness;
722 c->yCoeff= roundToInt16(cy *8192) * 0x0001000100010001ULL;
723 c->vrCoeff= roundToInt16(crv*8192) * 0x0001000100010001ULL;
724 c->ubCoeff= roundToInt16(cbu*8192) * 0x0001000100010001ULL;
725 c->vgCoeff= roundToInt16(cgv*8192) * 0x0001000100010001ULL;
726 c->ugCoeff= roundToInt16(cgu*8192) * 0x0001000100010001ULL;
727 c->yOffset= roundToInt16(oy * 8) * 0x0001000100010001ULL;
729 c->yuv2rgb_y_coeff = (int16_t)roundToInt16(cy <<13);
730 c->yuv2rgb_y_offset = (int16_t)roundToInt16(oy << 9);
731 c->yuv2rgb_v2r_coeff= (int16_t)roundToInt16(crv<<13);
732 c->yuv2rgb_v2g_coeff= (int16_t)roundToInt16(cgv<<13);
733 c->yuv2rgb_u2g_coeff= (int16_t)roundToInt16(cgu<<13);
734 c->yuv2rgb_u2b_coeff= (int16_t)roundToInt16(cbu<<13);
736 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
740 if (c->flags & SWS_CPU_CAPS_ALTIVEC)
741 ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness, contrast, saturation);
746 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation)
748 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
750 *inv_table = c->srcColorspaceTable;
751 *table = c->dstColorspaceTable;
752 *srcRange = c->srcRange;
753 *dstRange = c->dstRange;
754 *brightness= c->brightness;
755 *contrast = c->contrast;
756 *saturation= c->saturation;
761 static int handle_jpeg(enum PixelFormat *format)
764 case PIX_FMT_YUVJ420P:
765 *format = PIX_FMT_YUV420P;
767 case PIX_FMT_YUVJ422P:
768 *format = PIX_FMT_YUV422P;
770 case PIX_FMT_YUVJ444P:
771 *format = PIX_FMT_YUV444P;
773 case PIX_FMT_YUVJ440P:
774 *format = PIX_FMT_YUV440P;
781 SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,
782 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
783 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
787 int usesVFilter, usesHFilter;
789 int srcRange, dstRange;
790 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
792 if (flags & SWS_CPU_CAPS_MMX)
793 __asm__ volatile("emms\n\t"::: "memory");
796 #if !CONFIG_RUNTIME_CPUDETECT //ensure that the flags match the compiled variant if cpudetect is off
797 flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC|SWS_CPU_CAPS_BFIN);
798 flags |= ff_hardcodedcpuflags();
799 #endif /* CONFIG_RUNTIME_CPUDETECT */
800 if (!rgb15to16) sws_rgb2rgb_init(flags);
802 unscaled = (srcW == dstW && srcH == dstH);
804 srcRange = handle_jpeg(&srcFormat);
805 dstRange = handle_jpeg(&dstFormat);
807 if (!isSupportedIn(srcFormat)) {
808 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as input pixel format\n", sws_format_name(srcFormat));
811 if (!isSupportedOut(dstFormat)) {
812 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as output pixel format\n", sws_format_name(dstFormat));
816 i= flags & ( SWS_POINT
827 if(!i || (i & (i-1))) {
828 av_log(NULL, AV_LOG_ERROR, "swScaler: Exactly one scaler algorithm must be chosen\n");
833 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
834 av_log(NULL, AV_LOG_ERROR, "swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
835 srcW, srcH, dstW, dstH);
838 if(srcW > VOFW || dstW > VOFW) {
839 av_log(NULL, AV_LOG_ERROR, "swScaler: Compile-time maximum width is "AV_STRINGIFY(VOFW)" change VOF/VOFW and recompile\n");
843 if (!dstFilter) dstFilter= &dummyFilter;
844 if (!srcFilter) srcFilter= &dummyFilter;
846 FF_ALLOCZ_OR_GOTO(NULL, c, sizeof(SwsContext), fail);
848 c->av_class = &sws_context_class;
853 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
854 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
856 c->dstFormat= dstFormat;
857 c->srcFormat= srcFormat;
858 c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[dstFormat]);
859 c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[srcFormat]);
860 c->vRounder= 4* 0x0001000100010001ULL;
862 usesVFilter = (srcFilter->lumV && srcFilter->lumV->length>1) ||
863 (srcFilter->chrV && srcFilter->chrV->length>1) ||
864 (dstFilter->lumV && dstFilter->lumV->length>1) ||
865 (dstFilter->chrV && dstFilter->chrV->length>1);
866 usesHFilter = (srcFilter->lumH && srcFilter->lumH->length>1) ||
867 (srcFilter->chrH && srcFilter->chrH->length>1) ||
868 (dstFilter->lumH && dstFilter->lumH->length>1) ||
869 (dstFilter->chrH && dstFilter->chrH->length>1);
871 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
872 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
874 // reuse chroma for 2 pixels RGB/BGR unless user wants full chroma interpolation
875 if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
877 // drop some chroma lines if the user wants it
878 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
879 c->chrSrcVSubSample+= c->vChrDrop;
881 // drop every other pixel for chroma calculation unless user wants full chroma
882 if (isAnyRGB(srcFormat) && !(flags&SWS_FULL_CHR_H_INP)
883 && srcFormat!=PIX_FMT_RGB8 && srcFormat!=PIX_FMT_BGR8
884 && srcFormat!=PIX_FMT_RGB4 && srcFormat!=PIX_FMT_BGR4
885 && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE
886 && ((dstW>>c->chrDstHSubSample) <= (srcW>>1) || (flags&(SWS_FAST_BILINEAR|SWS_POINT))))
887 c->chrSrcHSubSample=1;
890 c->param[0] = param[0];
891 c->param[1] = param[1];
894 c->param[1] = SWS_PARAM_DEFAULT;
897 // Note the -((-x)>>y) is so that we always round toward +inf.
898 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
899 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
900 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
901 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
903 sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], srcRange, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/, dstRange, 0, 1<<16, 1<<16);
905 /* unscaled special cases */
906 if (unscaled && !usesHFilter && !usesVFilter && (srcRange == dstRange || isAnyRGB(dstFormat))) {
907 ff_get_unscaled_swscale(c);
910 if (flags&SWS_PRINT_INFO)
911 av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
912 sws_format_name(srcFormat), sws_format_name(dstFormat));
917 if (flags & SWS_CPU_CAPS_MMX2) {
918 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
919 if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR)) {
920 if (flags&SWS_PRINT_INFO)
921 av_log(c, AV_LOG_INFO, "output width is not a multiple of 32 -> no MMX2 scaler\n");
923 if (usesHFilter) c->canMMX2BeUsed=0;
928 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
929 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
931 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
932 // but only for the FAST_BILINEAR mode otherwise do correct scaling
933 // n-2 is the last chrominance sample available
934 // this is not perfect, but no one should notice the difference, the more correct variant
935 // would be like the vertical one, but that would require some special code for the
936 // first and last pixel
937 if (flags&SWS_FAST_BILINEAR) {
938 if (c->canMMX2BeUsed) {
942 //we don't use the x86 asm scaler if MMX is available
943 else if (flags & SWS_CPU_CAPS_MMX) {
944 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
945 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
949 /* precalculate horizontal scaler filter coefficients */
951 #if ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT)
952 // can't downscale !!!
953 if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {
954 c->lumMmx2FilterCodeSize = initMMX2HScaler( dstW, c->lumXInc, NULL, NULL, NULL, 8);
955 c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc, NULL, NULL, NULL, 4);
958 c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
959 c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
960 #elif HAVE_VIRTUALALLOC
961 c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
962 c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
964 c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);
965 c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);
968 if (!c->lumMmx2FilterCode || !c->chrMmx2FilterCode)
970 FF_ALLOCZ_OR_GOTO(c, c->hLumFilter , (dstW /8+8)*sizeof(int16_t), fail);
971 FF_ALLOCZ_OR_GOTO(c, c->hChrFilter , (c->chrDstW /4+8)*sizeof(int16_t), fail);
972 FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW /2/8+8)*sizeof(int32_t), fail);
973 FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW/2/4+8)*sizeof(int32_t), fail);
975 initMMX2HScaler( dstW, c->lumXInc, c->lumMmx2FilterCode, c->hLumFilter, c->hLumFilterPos, 8);
976 initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode, c->hChrFilter, c->hChrFilterPos, 4);
979 mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
980 mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
983 #endif /* ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT) */
985 const int filterAlign=
986 (flags & SWS_CPU_CAPS_MMX) ? 4 :
987 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
990 if (initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
991 srcW , dstW, filterAlign, 1<<14,
992 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
993 srcFilter->lumH, dstFilter->lumH, c->param) < 0)
995 if (initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
996 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
997 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
998 srcFilter->chrH, dstFilter->chrH, c->param) < 0)
1001 } // initialize horizontal stuff
1003 /* precalculate vertical scaler filter coefficients */
1005 const int filterAlign=
1006 (flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
1007 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
1010 if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
1011 srcH , dstH, filterAlign, (1<<12),
1012 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
1013 srcFilter->lumV, dstFilter->lumV, c->param) < 0)
1015 if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
1016 c->chrSrcH, c->chrDstH, filterAlign, (1<<12),
1017 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
1018 srcFilter->chrV, dstFilter->chrV, c->param) < 0)
1022 FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof (vector signed short)*c->vLumFilterSize*c->dstH, fail);
1023 FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH, fail);
1025 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
1027 short *p = (short *)&c->vYCoeffsBank[i];
1029 p[j] = c->vLumFilter[i];
1032 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
1034 short *p = (short *)&c->vCCoeffsBank[i];
1036 p[j] = c->vChrFilter[i];
1041 // calculate buffer sizes so that they won't run out while handling these damn slices
1042 c->vLumBufSize= c->vLumFilterSize;
1043 c->vChrBufSize= c->vChrFilterSize;
1044 for (i=0; i<dstH; i++) {
1045 int chrI= i*c->chrDstH / dstH;
1046 int nextSlice= FFMAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
1047 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
1049 nextSlice>>= c->chrSrcVSubSample;
1050 nextSlice<<= c->chrSrcVSubSample;
1051 if (c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
1052 c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
1053 if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
1054 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
1057 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
1058 // allocate several megabytes to handle all possible cases)
1059 FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
1060 FF_ALLOC_OR_GOTO(c, c->chrPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
1061 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
1062 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
1063 //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)
1064 /* align at 16 bytes for AltiVec */
1065 for (i=0; i<c->vLumBufSize; i++) {
1066 FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], VOF+1, fail);
1067 c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];
1069 for (i=0; i<c->vChrBufSize; i++) {
1070 FF_ALLOC_OR_GOTO(c, c->chrPixBuf[i+c->vChrBufSize], (VOF+1)*2, fail);
1071 c->chrPixBuf[i] = c->chrPixBuf[i+c->vChrBufSize];
1073 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
1074 for (i=0; i<c->vLumBufSize; i++) {
1075 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], VOF+1, fail);
1076 c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];
1079 //try to avoid drawing green stuff between the right end and the stride end
1080 for (i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, (VOF+1)*2);
1082 assert(2*VOFW == VOF);
1084 assert(c->chrDstH <= dstH);
1086 if (flags&SWS_PRINT_INFO) {
1087 if (flags&SWS_FAST_BILINEAR)
1088 av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
1089 else if (flags&SWS_BILINEAR)
1090 av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
1091 else if (flags&SWS_BICUBIC)
1092 av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
1093 else if (flags&SWS_X)
1094 av_log(c, AV_LOG_INFO, "Experimental scaler, ");
1095 else if (flags&SWS_POINT)
1096 av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
1097 else if (flags&SWS_AREA)
1098 av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
1099 else if (flags&SWS_BICUBLIN)
1100 av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
1101 else if (flags&SWS_GAUSS)
1102 av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
1103 else if (flags&SWS_SINC)
1104 av_log(c, AV_LOG_INFO, "Sinc scaler, ");
1105 else if (flags&SWS_LANCZOS)
1106 av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
1107 else if (flags&SWS_SPLINE)
1108 av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
1110 av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
1112 av_log(c, AV_LOG_INFO, "from %s to %s%s ",
1113 sws_format_name(srcFormat),
1115 dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ||
1116 dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||
1117 dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ? "dithered " : "",
1121 sws_format_name(dstFormat));
1123 if (flags & SWS_CPU_CAPS_MMX2)
1124 av_log(c, AV_LOG_INFO, "using MMX2\n");
1125 else if (flags & SWS_CPU_CAPS_3DNOW)
1126 av_log(c, AV_LOG_INFO, "using 3DNOW\n");
1127 else if (flags & SWS_CPU_CAPS_MMX)
1128 av_log(c, AV_LOG_INFO, "using MMX\n");
1129 else if (flags & SWS_CPU_CAPS_ALTIVEC)
1130 av_log(c, AV_LOG_INFO, "using AltiVec\n");
1132 av_log(c, AV_LOG_INFO, "using C\n");
1134 if (flags & SWS_CPU_CAPS_MMX) {
1135 if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
1136 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
1138 if (c->hLumFilterSize==4)
1139 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal luminance scaling\n");
1140 else if (c->hLumFilterSize==8)
1141 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal luminance scaling\n");
1143 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal luminance scaling\n");
1145 if (c->hChrFilterSize==4)
1146 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal chrominance scaling\n");
1147 else if (c->hChrFilterSize==8)
1148 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal chrominance scaling\n");
1150 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal chrominance scaling\n");
1154 av_log(c, AV_LOG_VERBOSE, "using x86 asm scaler for horizontal scaling\n");
1156 if (flags & SWS_FAST_BILINEAR)
1157 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR C scaler for horizontal scaling\n");
1159 av_log(c, AV_LOG_VERBOSE, "using C scaler for horizontal scaling\n");
1162 if (isPlanarYUV(dstFormat)) {
1163 if (c->vLumFilterSize==1)
1164 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1166 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1168 if (c->vLumFilterSize==1 && c->vChrFilterSize==2)
1169 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
1170 " 2-tap scaler for vertical chrominance scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1171 else if (c->vLumFilterSize==2 && c->vChrFilterSize==2)
1172 av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1174 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1177 if (dstFormat==PIX_FMT_BGR24)
1178 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 converter\n",
1179 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
1180 else if (dstFormat==PIX_FMT_RGB32)
1181 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1182 else if (dstFormat==PIX_FMT_BGR565)
1183 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1184 else if (dstFormat==PIX_FMT_BGR555)
1185 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1186 else if (dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||
1187 dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE)
1188 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR12 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1190 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
1191 av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1192 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
1193 av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1194 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
1197 c->swScale= ff_getSwsFunc(c);
1205 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
1206 float lumaSharpen, float chromaSharpen,
1207 float chromaHShift, float chromaVShift,
1210 SwsFilter *filter= av_malloc(sizeof(SwsFilter));
1214 if (lumaGBlur!=0.0) {
1215 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
1216 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
1218 filter->lumH= sws_getIdentityVec();
1219 filter->lumV= sws_getIdentityVec();
1222 if (chromaGBlur!=0.0) {
1223 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
1224 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
1226 filter->chrH= sws_getIdentityVec();
1227 filter->chrV= sws_getIdentityVec();
1230 if (chromaSharpen!=0.0) {
1231 SwsVector *id= sws_getIdentityVec();
1232 sws_scaleVec(filter->chrH, -chromaSharpen);
1233 sws_scaleVec(filter->chrV, -chromaSharpen);
1234 sws_addVec(filter->chrH, id);
1235 sws_addVec(filter->chrV, id);
1239 if (lumaSharpen!=0.0) {
1240 SwsVector *id= sws_getIdentityVec();
1241 sws_scaleVec(filter->lumH, -lumaSharpen);
1242 sws_scaleVec(filter->lumV, -lumaSharpen);
1243 sws_addVec(filter->lumH, id);
1244 sws_addVec(filter->lumV, id);
1248 if (chromaHShift != 0.0)
1249 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
1251 if (chromaVShift != 0.0)
1252 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
1254 sws_normalizeVec(filter->chrH, 1.0);
1255 sws_normalizeVec(filter->chrV, 1.0);
1256 sws_normalizeVec(filter->lumH, 1.0);
1257 sws_normalizeVec(filter->lumV, 1.0);
1259 if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
1260 if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
1265 SwsVector *sws_allocVec(int length)
1267 SwsVector *vec = av_malloc(sizeof(SwsVector));
1270 vec->length = length;
1271 vec->coeff = av_malloc(sizeof(double) * length);
1277 SwsVector *sws_getGaussianVec(double variance, double quality)
1279 const int length= (int)(variance*quality + 0.5) | 1;
1281 double middle= (length-1)*0.5;
1282 SwsVector *vec= sws_allocVec(length);
1287 for (i=0; i<length; i++) {
1288 double dist= i-middle;
1289 vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);
1292 sws_normalizeVec(vec, 1.0);
1297 SwsVector *sws_getConstVec(double c, int length)
1300 SwsVector *vec= sws_allocVec(length);
1305 for (i=0; i<length; i++)
1311 SwsVector *sws_getIdentityVec(void)
1313 return sws_getConstVec(1.0, 1);
1316 static double sws_dcVec(SwsVector *a)
1321 for (i=0; i<a->length; i++)
1327 void sws_scaleVec(SwsVector *a, double scalar)
1331 for (i=0; i<a->length; i++)
1332 a->coeff[i]*= scalar;
1335 void sws_normalizeVec(SwsVector *a, double height)
1337 sws_scaleVec(a, height/sws_dcVec(a));
1340 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
1342 int length= a->length + b->length - 1;
1344 SwsVector *vec= sws_getConstVec(0.0, length);
1349 for (i=0; i<a->length; i++) {
1350 for (j=0; j<b->length; j++) {
1351 vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];
1358 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
1360 int length= FFMAX(a->length, b->length);
1362 SwsVector *vec= sws_getConstVec(0.0, length);
1367 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1368 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
1373 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
1375 int length= FFMAX(a->length, b->length);
1377 SwsVector *vec= sws_getConstVec(0.0, length);
1382 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1383 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
1388 /* shift left / or right if "shift" is negative */
1389 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
1391 int length= a->length + FFABS(shift)*2;
1393 SwsVector *vec= sws_getConstVec(0.0, length);
1398 for (i=0; i<a->length; i++) {
1399 vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
1405 void sws_shiftVec(SwsVector *a, int shift)
1407 SwsVector *shifted= sws_getShiftedVec(a, shift);
1409 a->coeff= shifted->coeff;
1410 a->length= shifted->length;
1414 void sws_addVec(SwsVector *a, SwsVector *b)
1416 SwsVector *sum= sws_sumVec(a, b);
1418 a->coeff= sum->coeff;
1419 a->length= sum->length;
1423 void sws_subVec(SwsVector *a, SwsVector *b)
1425 SwsVector *diff= sws_diffVec(a, b);
1427 a->coeff= diff->coeff;
1428 a->length= diff->length;
1432 void sws_convVec(SwsVector *a, SwsVector *b)
1434 SwsVector *conv= sws_getConvVec(a, b);
1436 a->coeff= conv->coeff;
1437 a->length= conv->length;
1441 SwsVector *sws_cloneVec(SwsVector *a)
1444 SwsVector *vec= sws_allocVec(a->length);
1449 for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];
1454 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
1461 for (i=0; i<a->length; i++)
1462 if (a->coeff[i]>max) max= a->coeff[i];
1464 for (i=0; i<a->length; i++)
1465 if (a->coeff[i]<min) min= a->coeff[i];
1469 for (i=0; i<a->length; i++) {
1470 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
1471 av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
1472 for (;x>0; x--) av_log(log_ctx, log_level, " ");
1473 av_log(log_ctx, log_level, "|\n");
1477 #if LIBSWSCALE_VERSION_MAJOR < 1
1478 void sws_printVec(SwsVector *a)
1480 sws_printVec2(a, NULL, AV_LOG_DEBUG);
1484 void sws_freeVec(SwsVector *a)
1487 av_freep(&a->coeff);
1492 void sws_freeFilter(SwsFilter *filter)
1494 if (!filter) return;
1496 if (filter->lumH) sws_freeVec(filter->lumH);
1497 if (filter->lumV) sws_freeVec(filter->lumV);
1498 if (filter->chrH) sws_freeVec(filter->chrH);
1499 if (filter->chrV) sws_freeVec(filter->chrV);
1503 void sws_freeContext(SwsContext *c)
1509 for (i=0; i<c->vLumBufSize; i++)
1510 av_freep(&c->lumPixBuf[i]);
1511 av_freep(&c->lumPixBuf);
1515 for (i=0; i<c->vChrBufSize; i++)
1516 av_freep(&c->chrPixBuf[i]);
1517 av_freep(&c->chrPixBuf);
1520 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1521 for (i=0; i<c->vLumBufSize; i++)
1522 av_freep(&c->alpPixBuf[i]);
1523 av_freep(&c->alpPixBuf);
1526 av_freep(&c->vLumFilter);
1527 av_freep(&c->vChrFilter);
1528 av_freep(&c->hLumFilter);
1529 av_freep(&c->hChrFilter);
1531 av_freep(&c->vYCoeffsBank);
1532 av_freep(&c->vCCoeffsBank);
1535 av_freep(&c->vLumFilterPos);
1536 av_freep(&c->vChrFilterPos);
1537 av_freep(&c->hLumFilterPos);
1538 av_freep(&c->hChrFilterPos);
1541 #ifdef MAP_ANONYMOUS
1542 if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);
1543 if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);
1544 #elif HAVE_VIRTUALALLOC
1545 if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);
1546 if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);
1548 av_free(c->lumMmx2FilterCode);
1549 av_free(c->chrMmx2FilterCode);
1551 c->lumMmx2FilterCode=NULL;
1552 c->chrMmx2FilterCode=NULL;
1553 #endif /* ARCH_X86 */
1555 av_freep(&c->yuvTable);
1560 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
1561 int srcW, int srcH, enum PixelFormat srcFormat,
1562 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
1563 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
1565 static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};
1568 param = default_param;
1571 (context->srcW != srcW ||
1572 context->srcH != srcH ||
1573 context->srcFormat != srcFormat ||
1574 context->dstW != dstW ||
1575 context->dstH != dstH ||
1576 context->dstFormat != dstFormat ||
1577 context->flags != flags ||
1578 context->param[0] != param[0] ||
1579 context->param[1] != param[1])) {
1580 sws_freeContext(context);
1585 return sws_getContext(srcW, srcH, srcFormat,
1586 dstW, dstH, dstFormat, flags,
1587 srcFilter, dstFilter, param);