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_BGR565LE \
79 || (x)==PIX_FMT_BGR565BE \
80 || (x)==PIX_FMT_BGR555LE \
81 || (x)==PIX_FMT_BGR555BE \
82 || (x)==PIX_FMT_BGR32 \
83 || (x)==PIX_FMT_BGR32_1 \
84 || (x)==PIX_FMT_RGB24 \
85 || (x)==PIX_FMT_RGB565LE \
86 || (x)==PIX_FMT_RGB565BE \
87 || (x)==PIX_FMT_RGB555LE \
88 || (x)==PIX_FMT_RGB555BE \
89 || (x)==PIX_FMT_GRAY8 \
90 || (x)==PIX_FMT_Y400A \
91 || (x)==PIX_FMT_YUV410P \
92 || (x)==PIX_FMT_YUV440P \
93 || (x)==PIX_FMT_NV12 \
94 || (x)==PIX_FMT_NV21 \
95 || (x)==PIX_FMT_GRAY16BE \
96 || (x)==PIX_FMT_GRAY16LE \
97 || (x)==PIX_FMT_YUV444P \
98 || (x)==PIX_FMT_YUV422P \
99 || (x)==PIX_FMT_YUV411P \
100 || (x)==PIX_FMT_YUVJ420P \
101 || (x)==PIX_FMT_YUVJ422P \
102 || (x)==PIX_FMT_YUVJ440P \
103 || (x)==PIX_FMT_YUVJ444P \
104 || (x)==PIX_FMT_PAL8 \
105 || (x)==PIX_FMT_BGR8 \
106 || (x)==PIX_FMT_RGB8 \
107 || (x)==PIX_FMT_BGR4_BYTE \
108 || (x)==PIX_FMT_RGB4_BYTE \
109 || (x)==PIX_FMT_YUV440P \
110 || (x)==PIX_FMT_MONOWHITE \
111 || (x)==PIX_FMT_MONOBLACK \
112 || (x)==PIX_FMT_YUV420P9LE \
113 || (x)==PIX_FMT_YUV444P9LE \
114 || (x)==PIX_FMT_YUV420P10LE \
115 || (x)==PIX_FMT_YUV422P10LE \
116 || (x)==PIX_FMT_YUV444P10LE \
117 || (x)==PIX_FMT_YUV420P16LE \
118 || (x)==PIX_FMT_YUV422P16LE \
119 || (x)==PIX_FMT_YUV444P16LE \
120 || (x)==PIX_FMT_YUV420P9BE \
121 || (x)==PIX_FMT_YUV444P9BE \
122 || (x)==PIX_FMT_YUV420P10BE \
123 || (x)==PIX_FMT_YUV444P10BE \
124 || (x)==PIX_FMT_YUV422P10BE \
125 || (x)==PIX_FMT_YUV420P16BE \
126 || (x)==PIX_FMT_YUV422P16BE \
127 || (x)==PIX_FMT_YUV444P16BE \
130 int sws_isSupportedInput(enum PixelFormat pix_fmt)
132 return isSupportedIn(pix_fmt);
135 #define isSupportedOut(x) ( \
136 (x)==PIX_FMT_YUV420P \
137 || (x)==PIX_FMT_YUVA420P \
138 || (x)==PIX_FMT_YUYV422 \
139 || (x)==PIX_FMT_UYVY422 \
140 || (x)==PIX_FMT_YUV444P \
141 || (x)==PIX_FMT_YUV422P \
142 || (x)==PIX_FMT_YUV411P \
143 || (x)==PIX_FMT_YUVJ420P \
144 || (x)==PIX_FMT_YUVJ422P \
145 || (x)==PIX_FMT_YUVJ440P \
146 || (x)==PIX_FMT_YUVJ444P \
149 || (x)==PIX_FMT_RGB565LE \
150 || (x)==PIX_FMT_RGB565BE \
151 || (x)==PIX_FMT_RGB555LE \
152 || (x)==PIX_FMT_RGB555BE \
153 || (x)==PIX_FMT_RGB444LE \
154 || (x)==PIX_FMT_RGB444BE \
155 || (x)==PIX_FMT_BGR565LE \
156 || (x)==PIX_FMT_BGR565BE \
157 || (x)==PIX_FMT_BGR555LE \
158 || (x)==PIX_FMT_BGR555BE \
159 || (x)==PIX_FMT_BGR444LE \
160 || (x)==PIX_FMT_BGR444BE \
161 || (x)==PIX_FMT_RGB8 \
162 || (x)==PIX_FMT_BGR8 \
163 || (x)==PIX_FMT_RGB4_BYTE \
164 || (x)==PIX_FMT_BGR4_BYTE \
165 || (x)==PIX_FMT_RGB4 \
166 || (x)==PIX_FMT_BGR4 \
167 || (x)==PIX_FMT_MONOBLACK \
168 || (x)==PIX_FMT_MONOWHITE \
169 || (x)==PIX_FMT_NV12 \
170 || (x)==PIX_FMT_NV21 \
171 || (x)==PIX_FMT_GRAY16BE \
172 || (x)==PIX_FMT_GRAY16LE \
173 || (x)==PIX_FMT_GRAY8 \
174 || (x)==PIX_FMT_YUV410P \
175 || (x)==PIX_FMT_YUV440P \
176 || (x)==PIX_FMT_YUV420P9LE \
177 || (x)==PIX_FMT_YUV420P10LE \
178 || (x)==PIX_FMT_YUV420P16LE \
179 || (x)==PIX_FMT_YUV422P16LE \
180 || (x)==PIX_FMT_YUV444P16LE \
181 || (x)==PIX_FMT_YUV420P9BE \
182 || (x)==PIX_FMT_YUV420P10BE \
183 || (x)==PIX_FMT_YUV420P16BE \
184 || (x)==PIX_FMT_YUV422P16BE \
185 || (x)==PIX_FMT_YUV444P16BE \
188 int sws_isSupportedOutput(enum PixelFormat pix_fmt)
190 return isSupportedOut(pix_fmt);
193 extern const int32_t ff_yuv2rgb_coeffs[8][4];
195 const char *sws_format_name(enum PixelFormat format)
197 if ((unsigned)format < PIX_FMT_NB && av_pix_fmt_descriptors[format].name)
198 return av_pix_fmt_descriptors[format].name;
200 return "Unknown format";
203 static double getSplineCoeff(double a, double b, double c, double d, double dist)
205 if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
206 else return getSplineCoeff( 0.0,
213 static int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
214 int srcW, int dstW, int filterAlign, int one, int flags, int cpu_flags,
215 SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
221 int64_t *filter=NULL;
222 int64_t *filter2=NULL;
223 const int64_t fone= 1LL<<54;
226 emms_c(); //FIXME this should not be required but it IS (even for non-MMX versions)
228 // NOTE: the +1 is for the MMX scaler which reads over the end
229 FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW+1)*sizeof(int16_t), fail);
231 if (FFABS(xInc - 0x10000) <10) { // unscaled
234 FF_ALLOCZ_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
236 for (i=0; i<dstW; i++) {
237 filter[i*filterSize]= fone;
241 } else if (flags&SWS_POINT) { // lame looking point sampling mode
245 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
247 xDstInSrc= xInc/2 - 0x8000;
248 for (i=0; i<dstW; i++) {
249 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
255 } else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) { // bilinear upscale
259 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
261 xDstInSrc= xInc/2 - 0x8000;
262 for (i=0; i<dstW; i++) {
263 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
267 //bilinear upscale / linear interpolate / area averaging
268 for (j=0; j<filterSize; j++) {
269 int64_t coeff= fone - FFABS((xx<<16) - xDstInSrc)*(fone>>16);
270 if (coeff<0) coeff=0;
271 filter[i*filterSize + j]= coeff;
280 if (flags&SWS_BICUBIC) sizeFactor= 4;
281 else if (flags&SWS_X) sizeFactor= 8;
282 else if (flags&SWS_AREA) sizeFactor= 1; //downscale only, for upscale it is bilinear
283 else if (flags&SWS_GAUSS) sizeFactor= 8; // infinite ;)
284 else if (flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? ceil(2*param[0]) : 6;
285 else if (flags&SWS_SINC) sizeFactor= 20; // infinite ;)
286 else if (flags&SWS_SPLINE) sizeFactor= 20; // infinite ;)
287 else if (flags&SWS_BILINEAR) sizeFactor= 2;
289 sizeFactor= 0; //GCC warning killer
293 if (xInc <= 1<<16) filterSize= 1 + sizeFactor; // upscale
294 else filterSize= 1 + (sizeFactor*srcW + dstW - 1)/ dstW;
296 if (filterSize > srcW-2) filterSize=srcW-2;
298 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
300 xDstInSrc= xInc - 0x10000;
301 for (i=0; i<dstW; i++) {
302 int xx= (xDstInSrc - ((filterSize-2)<<16)) / (1<<17);
305 for (j=0; j<filterSize; j++) {
306 int64_t d= ((int64_t)FFABS((xx<<17) - xDstInSrc))<<13;
312 floatd= d * (1.0/(1<<30));
314 if (flags & SWS_BICUBIC) {
315 int64_t B= (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1<<24);
316 int64_t C= (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1<<24);
317 int64_t dd = ( d*d)>>30;
318 int64_t ddd= (dd*d)>>30;
321 coeff = (12*(1<<24)-9*B-6*C)*ddd + (-18*(1<<24)+12*B+6*C)*dd + (6*(1<<24)-2*B)*(1<<30);
322 else if (d < 1LL<<31)
323 coeff = (-B-6*C)*ddd + (6*B+30*C)*dd + (-12*B-48*C)*d + (8*B+24*C)*(1<<30);
326 coeff *= fone>>(30+24);
328 /* else if (flags & SWS_X) {
329 double p= param ? param*0.01 : 0.3;
330 coeff = d ? sin(d*M_PI)/(d*M_PI) : 1.0;
331 coeff*= pow(2.0, - p*d*d);
333 else if (flags & SWS_X) {
334 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
338 c = cos(floatd*M_PI);
341 if (c<0.0) c= -pow(-c, A);
343 coeff= (c*0.5 + 0.5)*fone;
344 } else if (flags & SWS_AREA) {
345 int64_t d2= d - (1<<29);
346 if (d2*xInc < -(1LL<<(29+16))) coeff= 1.0 * (1LL<<(30+16));
347 else if (d2*xInc < (1LL<<(29+16))) coeff= -d2*xInc + (1LL<<(29+16));
349 coeff *= fone>>(30+16);
350 } else if (flags & SWS_GAUSS) {
351 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
352 coeff = (pow(2.0, - p*floatd*floatd))*fone;
353 } else if (flags & SWS_SINC) {
354 coeff = (d ? sin(floatd*M_PI)/(floatd*M_PI) : 1.0)*fone;
355 } else if (flags & SWS_LANCZOS) {
356 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
357 coeff = (d ? sin(floatd*M_PI)*sin(floatd*M_PI/p)/(floatd*floatd*M_PI*M_PI/p) : 1.0)*fone;
358 if (floatd>p) coeff=0;
359 } else if (flags & SWS_BILINEAR) {
361 if (coeff<0) coeff=0;
363 } else if (flags & SWS_SPLINE) {
364 double p=-2.196152422706632;
365 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, floatd) * fone;
367 coeff= 0.0; //GCC warning killer
371 filter[i*filterSize + j]= coeff;
378 /* apply src & dst Filter to filter -> filter2
381 assert(filterSize>0);
382 filter2Size= filterSize;
383 if (srcFilter) filter2Size+= srcFilter->length - 1;
384 if (dstFilter) filter2Size+= dstFilter->length - 1;
385 assert(filter2Size>0);
386 FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size*dstW*sizeof(*filter2), fail);
388 for (i=0; i<dstW; i++) {
392 for (k=0; k<srcFilter->length; k++) {
393 for (j=0; j<filterSize; j++)
394 filter2[i*filter2Size + k + j] += srcFilter->coeff[k]*filter[i*filterSize + j];
397 for (j=0; j<filterSize; j++)
398 filter2[i*filter2Size + j]= filter[i*filterSize + j];
402 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
406 /* try to reduce the filter-size (step1 find size and shift left) */
407 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
409 for (i=dstW-1; i>=0; i--) {
410 int min= filter2Size;
414 /* get rid of near zero elements on the left by shifting left */
415 for (j=0; j<filter2Size; j++) {
417 cutOff += FFABS(filter2[i*filter2Size]);
419 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
421 /* preserve monotonicity because the core can't handle the filter otherwise */
422 if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
424 // move filter coefficients left
425 for (k=1; k<filter2Size; k++)
426 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
427 filter2[i*filter2Size + k - 1]= 0;
432 /* count near zeros on the right */
433 for (j=filter2Size-1; j>0; j--) {
434 cutOff += FFABS(filter2[i*filter2Size + j]);
436 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
440 if (min>minFilterSize) minFilterSize= min;
443 if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) {
444 // we can handle the special case 4,
445 // so we don't want to go to the full 8
446 if (minFilterSize < 5)
449 // We really don't want to waste our time
450 // doing useless computation, so fall back on
451 // the scalar C code for very small filters.
452 // Vectorizing is worth it only if you have a
453 // decent-sized vector.
454 if (minFilterSize < 3)
458 if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
459 // special case for unscaled vertical filtering
460 if (minFilterSize == 1 && filterAlign == 2)
464 assert(minFilterSize > 0);
465 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
466 assert(filterSize > 0);
467 filter= av_malloc(filterSize*dstW*sizeof(*filter));
468 if (filterSize >= MAX_FILTER_SIZE*16/((flags&SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
470 *outFilterSize= filterSize;
472 if (flags&SWS_PRINT_INFO)
473 av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
474 /* try to reduce the filter-size (step2 reduce it) */
475 for (i=0; i<dstW; i++) {
478 for (j=0; j<filterSize; j++) {
479 if (j>=filter2Size) filter[i*filterSize + j]= 0;
480 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
481 if((flags & SWS_BITEXACT) && j>=minFilterSize)
482 filter[i*filterSize + j]= 0;
486 //FIXME try to align filterPos if possible
489 for (i=0; i<dstW; i++) {
491 if ((*filterPos)[i] < 0) {
492 // move filter coefficients left to compensate for filterPos
493 for (j=1; j<filterSize; j++) {
494 int left= FFMAX(j + (*filterPos)[i], 0);
495 filter[i*filterSize + left] += filter[i*filterSize + j];
496 filter[i*filterSize + j]=0;
501 if ((*filterPos)[i] + filterSize > srcW) {
502 int shift= (*filterPos)[i] + filterSize - srcW;
503 // move filter coefficients right to compensate for filterPos
504 for (j=filterSize-2; j>=0; j--) {
505 int right= FFMIN(j + shift, filterSize-1);
506 filter[i*filterSize +right] += filter[i*filterSize +j];
507 filter[i*filterSize +j]=0;
509 (*filterPos)[i]= srcW - filterSize;
513 // Note the +1 is for the MMX scaler which reads over the end
514 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
515 FF_ALLOCZ_OR_GOTO(NULL, *outFilter, *outFilterSize*(dstW+1)*sizeof(int16_t), fail);
517 /* normalize & store in outFilter */
518 for (i=0; i<dstW; i++) {
523 for (j=0; j<filterSize; j++) {
524 sum+= filter[i*filterSize + j];
526 sum= (sum + one/2)/ one;
527 for (j=0; j<*outFilterSize; j++) {
528 int64_t v= filter[i*filterSize + j] + error;
529 int intV= ROUNDED_DIV(v, sum);
530 (*outFilter)[i*(*outFilterSize) + j]= intV;
535 (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
536 for (i=0; i<*outFilterSize; i++) {
537 int j= dstW*(*outFilterSize);
538 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
549 static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode, int16_t *filter, int32_t *filterPos, int numSplits)
552 x86_reg imm8OfPShufW1A;
553 x86_reg imm8OfPShufW2A;
554 x86_reg fragmentLengthA;
556 x86_reg imm8OfPShufW1B;
557 x86_reg imm8OfPShufW2B;
558 x86_reg fragmentLengthB;
563 // create an optimized horizontal scaling routine
564 /* This scaler is made of runtime-generated MMX2 code using specially
565 * tuned pshufw instructions. For every four output pixels, if four
566 * input pixels are enough for the fast bilinear scaling, then a chunk
567 * of fragmentB is used. If five input pixels are needed, then a chunk
568 * of fragmentA is used.
577 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
578 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
579 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
580 "punpcklbw %%mm7, %%mm1 \n\t"
581 "punpcklbw %%mm7, %%mm0 \n\t"
582 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
584 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
586 "psubw %%mm1, %%mm0 \n\t"
587 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
588 "pmullw %%mm3, %%mm0 \n\t"
589 "psllw $7, %%mm1 \n\t"
590 "paddw %%mm1, %%mm0 \n\t"
592 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
594 "add $8, %%"REG_a" \n\t"
598 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
599 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
600 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
605 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
609 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
610 "=r" (fragmentLengthA)
617 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
618 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
619 "punpcklbw %%mm7, %%mm0 \n\t"
620 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
622 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
624 "psubw %%mm1, %%mm0 \n\t"
625 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
626 "pmullw %%mm3, %%mm0 \n\t"
627 "psllw $7, %%mm1 \n\t"
628 "paddw %%mm1, %%mm0 \n\t"
630 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
632 "add $8, %%"REG_a" \n\t"
636 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
637 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
638 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
643 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
647 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
648 "=r" (fragmentLengthB)
651 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
654 for (i=0; i<dstW/numSplits; i++) {
659 int b=((xpos+xInc)>>16) - xx;
660 int c=((xpos+xInc*2)>>16) - xx;
661 int d=((xpos+xInc*3)>>16) - xx;
663 uint8_t *fragment = (d+1<4) ? fragmentB : fragmentA;
664 x86_reg imm8OfPShufW1 = (d+1<4) ? imm8OfPShufW1B : imm8OfPShufW1A;
665 x86_reg imm8OfPShufW2 = (d+1<4) ? imm8OfPShufW2B : imm8OfPShufW2A;
666 x86_reg fragmentLength = (d+1<4) ? fragmentLengthB : fragmentLengthA;
667 int maxShift= 3-(d+inc);
671 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
672 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
673 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
674 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
677 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
679 filterCode[fragmentPos + imm8OfPShufW1]=
680 (a+inc) | ((b+inc)<<2) | ((c+inc)<<4) | ((d+inc)<<6);
681 filterCode[fragmentPos + imm8OfPShufW2]=
682 a | (b<<2) | (c<<4) | (d<<6);
684 if (i+4-inc>=dstW) shift=maxShift; //avoid overread
685 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
687 if (shift && i>=shift) {
688 filterCode[fragmentPos + imm8OfPShufW1]+= 0x55*shift;
689 filterCode[fragmentPos + imm8OfPShufW2]+= 0x55*shift;
690 filterPos[i/2]-=shift;
694 fragmentPos+= fragmentLength;
697 filterCode[fragmentPos]= RET;
702 filterPos[((i/2)+1)&(~1)]= xpos>>16; // needed to jump to the next part
704 return fragmentPos + 1;
706 #endif /* HAVE_MMX2 */
708 static void getSubSampleFactors(int *h, int *v, enum PixelFormat format)
710 *h = av_pix_fmt_descriptors[format].log2_chroma_w;
711 *v = av_pix_fmt_descriptors[format].log2_chroma_h;
714 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation)
716 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
717 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
719 c->brightness= brightness;
720 c->contrast = contrast;
721 c->saturation= saturation;
722 c->srcRange = srcRange;
723 c->dstRange = dstRange;
724 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
726 c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->dstFormat]);
727 c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->srcFormat]);
729 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
732 if (HAVE_ALTIVEC && av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)
733 ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness, contrast, saturation);
737 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation)
739 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
741 *inv_table = c->srcColorspaceTable;
742 *table = c->dstColorspaceTable;
743 *srcRange = c->srcRange;
744 *dstRange = c->dstRange;
745 *brightness= c->brightness;
746 *contrast = c->contrast;
747 *saturation= c->saturation;
752 static int handle_jpeg(enum PixelFormat *format)
755 case PIX_FMT_YUVJ420P: *format = PIX_FMT_YUV420P; return 1;
756 case PIX_FMT_YUVJ422P: *format = PIX_FMT_YUV422P; return 1;
757 case PIX_FMT_YUVJ444P: *format = PIX_FMT_YUV444P; return 1;
758 case PIX_FMT_YUVJ440P: *format = PIX_FMT_YUV440P; return 1;
763 SwsContext *sws_alloc_context(void)
765 SwsContext *c= av_mallocz(sizeof(SwsContext));
767 c->av_class = &sws_context_class;
768 av_opt_set_defaults(c);
773 int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)
776 int usesVFilter, usesHFilter;
778 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
783 int dst_stride = FFALIGN(dstW * sizeof(int16_t) + 16, 16), dst_stride_px = dst_stride >> 1;
784 int flags, cpu_flags;
785 enum PixelFormat srcFormat= c->srcFormat;
786 enum PixelFormat dstFormat= c->dstFormat;
788 cpu_flags = av_get_cpu_flags();
791 if (!rgb15to16) sws_rgb2rgb_init();
793 unscaled = (srcW == dstW && srcH == dstH);
795 if (!isSupportedIn(srcFormat)) {
796 av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n", sws_format_name(srcFormat));
797 return AVERROR(EINVAL);
799 if (!isSupportedOut(dstFormat)) {
800 av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n", sws_format_name(dstFormat));
801 return AVERROR(EINVAL);
804 i= flags & ( SWS_POINT
815 if(!i || (i & (i-1))) {
816 av_log(c, AV_LOG_ERROR, "Exactly one scaler algorithm must be chosen\n");
817 return AVERROR(EINVAL);
820 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
821 av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",
822 srcW, srcH, dstW, dstH);
823 return AVERROR(EINVAL);
826 if (!dstFilter) dstFilter= &dummyFilter;
827 if (!srcFilter) srcFilter= &dummyFilter;
829 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
830 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
831 c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[dstFormat]);
832 c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[srcFormat]);
833 c->vRounder= 4* 0x0001000100010001ULL;
835 usesVFilter = (srcFilter->lumV && srcFilter->lumV->length>1) ||
836 (srcFilter->chrV && srcFilter->chrV->length>1) ||
837 (dstFilter->lumV && dstFilter->lumV->length>1) ||
838 (dstFilter->chrV && dstFilter->chrV->length>1);
839 usesHFilter = (srcFilter->lumH && srcFilter->lumH->length>1) ||
840 (srcFilter->chrH && srcFilter->chrH->length>1) ||
841 (dstFilter->lumH && dstFilter->lumH->length>1) ||
842 (dstFilter->chrH && dstFilter->chrH->length>1);
844 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
845 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
847 // reuse chroma for 2 pixels RGB/BGR unless user wants full chroma interpolation
848 if (flags & SWS_FULL_CHR_H_INT &&
849 dstFormat != PIX_FMT_RGBA &&
850 dstFormat != PIX_FMT_ARGB &&
851 dstFormat != PIX_FMT_BGRA &&
852 dstFormat != PIX_FMT_ABGR &&
853 dstFormat != PIX_FMT_RGB24 &&
854 dstFormat != PIX_FMT_BGR24) {
855 av_log(c, AV_LOG_ERROR,
856 "full chroma interpolation for destination format '%s' not yet implemented\n",
857 sws_format_name(dstFormat));
858 flags &= ~SWS_FULL_CHR_H_INT;
861 if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
863 // drop some chroma lines if the user wants it
864 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
865 c->chrSrcVSubSample+= c->vChrDrop;
867 // drop every other pixel for chroma calculation unless user wants full chroma
868 if (isAnyRGB(srcFormat) && !(flags&SWS_FULL_CHR_H_INP)
869 && srcFormat!=PIX_FMT_RGB8 && srcFormat!=PIX_FMT_BGR8
870 && srcFormat!=PIX_FMT_RGB4 && srcFormat!=PIX_FMT_BGR4
871 && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE
872 && ((dstW>>c->chrDstHSubSample) <= (srcW>>1) || (flags&SWS_FAST_BILINEAR)))
873 c->chrSrcHSubSample=1;
875 // Note the -((-x)>>y) is so that we always round toward +inf.
876 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
877 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
878 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
879 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
881 /* unscaled special cases */
882 if (unscaled && !usesHFilter && !usesVFilter && (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
883 ff_get_unscaled_swscale(c);
886 if (flags&SWS_PRINT_INFO)
887 av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
888 sws_format_name(srcFormat), sws_format_name(dstFormat));
893 c->scalingBpp = FFMAX(av_pix_fmt_descriptors[srcFormat].comp[0].depth_minus1,
894 av_pix_fmt_descriptors[dstFormat].comp[0].depth_minus1) >= 8 ? 16 : 8;
895 if (c->scalingBpp == 16)
897 FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW, 16) * 2 * c->scalingBpp >> 3, fail);
898 if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2 && c->scalingBpp == 8) {
899 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
900 if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR)) {
901 if (flags&SWS_PRINT_INFO)
902 av_log(c, AV_LOG_INFO, "output width is not a multiple of 32 -> no MMX2 scaler\n");
904 if (usesHFilter) c->canMMX2BeUsed=0;
909 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
910 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
912 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
913 // but only for the FAST_BILINEAR mode otherwise do correct scaling
914 // n-2 is the last chrominance sample available
915 // this is not perfect, but no one should notice the difference, the more correct variant
916 // would be like the vertical one, but that would require some special code for the
917 // first and last pixel
918 if (flags&SWS_FAST_BILINEAR) {
919 if (c->canMMX2BeUsed) {
923 //we don't use the x86 asm scaler if MMX is available
924 else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
925 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
926 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
930 /* precalculate horizontal scaler filter coefficients */
933 // can't downscale !!!
934 if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {
935 c->lumMmx2FilterCodeSize = initMMX2HScaler( dstW, c->lumXInc, NULL, NULL, NULL, 8);
936 c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc, NULL, NULL, NULL, 4);
939 c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
940 c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
941 #elif HAVE_VIRTUALALLOC
942 c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
943 c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
945 c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);
946 c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);
949 if (!c->lumMmx2FilterCode || !c->chrMmx2FilterCode)
950 return AVERROR(ENOMEM);
951 FF_ALLOCZ_OR_GOTO(c, c->hLumFilter , (dstW /8+8)*sizeof(int16_t), fail);
952 FF_ALLOCZ_OR_GOTO(c, c->hChrFilter , (c->chrDstW /4+8)*sizeof(int16_t), fail);
953 FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW /2/8+8)*sizeof(int32_t), fail);
954 FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW/2/4+8)*sizeof(int32_t), fail);
956 initMMX2HScaler( dstW, c->lumXInc, c->lumMmx2FilterCode, c->hLumFilter, c->hLumFilterPos, 8);
957 initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode, c->hChrFilter, c->hChrFilterPos, 4);
960 mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
961 mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
964 #endif /* HAVE_MMX2 */
966 const int filterAlign=
967 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :
968 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
971 if (initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
972 srcW , dstW, filterAlign, 1<<14,
973 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags, cpu_flags,
974 srcFilter->lumH, dstFilter->lumH, c->param) < 0)
976 if (initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
977 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
978 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, cpu_flags,
979 srcFilter->chrH, dstFilter->chrH, c->param) < 0)
982 } // initialize horizontal stuff
984 /* precalculate vertical scaler filter coefficients */
986 const int filterAlign=
987 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
988 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
991 if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
992 srcH , dstH, filterAlign, (1<<12),
993 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags, cpu_flags,
994 srcFilter->lumV, dstFilter->lumV, c->param) < 0)
996 if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
997 c->chrSrcH, c->chrDstH, filterAlign, (1<<12),
998 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, cpu_flags,
999 srcFilter->chrV, dstFilter->chrV, c->param) < 0)
1003 FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof (vector signed short)*c->vLumFilterSize*c->dstH, fail);
1004 FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH, fail);
1006 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
1008 short *p = (short *)&c->vYCoeffsBank[i];
1010 p[j] = c->vLumFilter[i];
1013 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
1015 short *p = (short *)&c->vCCoeffsBank[i];
1017 p[j] = c->vChrFilter[i];
1022 // calculate buffer sizes so that they won't run out while handling these damn slices
1023 c->vLumBufSize= c->vLumFilterSize;
1024 c->vChrBufSize= c->vChrFilterSize;
1025 for (i=0; i<dstH; i++) {
1026 int chrI= i*c->chrDstH / dstH;
1027 int nextSlice= FFMAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
1028 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
1030 nextSlice>>= c->chrSrcVSubSample;
1031 nextSlice<<= c->chrSrcVSubSample;
1032 if (c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
1033 c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
1034 if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
1035 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
1038 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
1039 // allocate several megabytes to handle all possible cases)
1040 FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
1041 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
1042 FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
1043 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
1044 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
1045 //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)
1046 /* align at 16 bytes for AltiVec */
1047 for (i=0; i<c->vLumBufSize; i++) {
1048 FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], dst_stride+1, fail);
1049 c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];
1051 c->uv_off = dst_stride_px;
1052 c->uv_offx2 = dst_stride;
1053 for (i=0; i<c->vChrBufSize; i++) {
1054 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i+c->vChrBufSize], dst_stride*2+1, fail);
1055 c->chrUPixBuf[i] = c->chrUPixBuf[i+c->vChrBufSize];
1056 c->chrVPixBuf[i] = c->chrVPixBuf[i+c->vChrBufSize] = c->chrUPixBuf[i] + dst_stride_px;
1058 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
1059 for (i=0; i<c->vLumBufSize; i++) {
1060 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], dst_stride+1, fail);
1061 c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];
1064 //try to avoid drawing green stuff between the right end and the stride end
1065 for (i=0; i<c->vChrBufSize; i++)
1066 memset(c->chrUPixBuf[i], 64, dst_stride*2+1);
1068 assert(c->chrDstH <= dstH);
1070 if (flags&SWS_PRINT_INFO) {
1071 if (flags&SWS_FAST_BILINEAR) av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
1072 else if (flags&SWS_BILINEAR) av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
1073 else if (flags&SWS_BICUBIC) av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
1074 else if (flags&SWS_X) av_log(c, AV_LOG_INFO, "Experimental scaler, ");
1075 else if (flags&SWS_POINT) av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
1076 else if (flags&SWS_AREA) av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
1077 else if (flags&SWS_BICUBLIN) av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
1078 else if (flags&SWS_GAUSS) av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
1079 else if (flags&SWS_SINC) av_log(c, AV_LOG_INFO, "Sinc scaler, ");
1080 else if (flags&SWS_LANCZOS) av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
1081 else if (flags&SWS_SPLINE) av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
1082 else av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
1084 av_log(c, AV_LOG_INFO, "from %s to %s%s ",
1085 sws_format_name(srcFormat),
1087 dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ||
1088 dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||
1089 dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ? "dithered " : "",
1093 sws_format_name(dstFormat));
1095 if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2) av_log(c, AV_LOG_INFO, "using MMX2\n");
1096 else if (HAVE_AMD3DNOW && cpu_flags & AV_CPU_FLAG_3DNOW) av_log(c, AV_LOG_INFO, "using 3DNOW\n");
1097 else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) av_log(c, AV_LOG_INFO, "using MMX\n");
1098 else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) av_log(c, AV_LOG_INFO, "using AltiVec\n");
1099 else av_log(c, AV_LOG_INFO, "using C\n");
1101 if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
1102 if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
1103 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
1105 if (c->hLumFilterSize==4)
1106 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal luminance scaling\n");
1107 else if (c->hLumFilterSize==8)
1108 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal luminance scaling\n");
1110 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal luminance scaling\n");
1112 if (c->hChrFilterSize==4)
1113 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal chrominance scaling\n");
1114 else if (c->hChrFilterSize==8)
1115 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal chrominance scaling\n");
1117 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal chrominance scaling\n");
1121 av_log(c, AV_LOG_VERBOSE, "using x86 asm scaler for horizontal scaling\n");
1123 if (flags & SWS_FAST_BILINEAR)
1124 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR C scaler for horizontal scaling\n");
1126 av_log(c, AV_LOG_VERBOSE, "using C scaler for horizontal scaling\n");
1129 if (isPlanarYUV(dstFormat)) {
1130 if (c->vLumFilterSize==1)
1131 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n",
1132 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
1134 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n",
1135 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
1137 if (c->vLumFilterSize==1 && c->vChrFilterSize==2)
1138 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
1139 " 2-tap scaler for vertical chrominance scaling (BGR)\n",
1140 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
1141 else if (c->vLumFilterSize==2 && c->vChrFilterSize==2)
1142 av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n",
1143 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
1145 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n",
1146 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
1149 if (dstFormat==PIX_FMT_BGR24)
1150 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 converter\n",
1151 (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2) ? "MMX2" :
1152 ((HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C"));
1153 else if (dstFormat==PIX_FMT_RGB32)
1154 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 converter\n",
1155 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
1156 else if (dstFormat==PIX_FMT_BGR565)
1157 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 converter\n",
1158 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
1159 else if (dstFormat==PIX_FMT_BGR555)
1160 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 converter\n",
1161 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
1162 else if (dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||
1163 dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE)
1164 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR12 converter\n",
1165 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
1167 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
1168 av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1169 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
1170 av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1171 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
1174 c->swScale= ff_getSwsFunc(c);
1176 fail: //FIXME replace things by appropriate error codes
1180 #if FF_API_SWS_GETCONTEXT
1181 SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,
1182 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
1183 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
1187 if(!(c=sws_alloc_context()))
1195 c->srcRange = handle_jpeg(&srcFormat);
1196 c->dstRange = handle_jpeg(&dstFormat);
1197 c->srcFormat= srcFormat;
1198 c->dstFormat= dstFormat;
1201 c->param[0] = param[0];
1202 c->param[1] = param[1];
1204 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);
1206 if(sws_init_context(c, srcFilter, dstFilter) < 0){
1215 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
1216 float lumaSharpen, float chromaSharpen,
1217 float chromaHShift, float chromaVShift,
1220 SwsFilter *filter= av_malloc(sizeof(SwsFilter));
1224 if (lumaGBlur!=0.0) {
1225 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
1226 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
1228 filter->lumH= sws_getIdentityVec();
1229 filter->lumV= sws_getIdentityVec();
1232 if (chromaGBlur!=0.0) {
1233 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
1234 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
1236 filter->chrH= sws_getIdentityVec();
1237 filter->chrV= sws_getIdentityVec();
1240 if (chromaSharpen!=0.0) {
1241 SwsVector *id= sws_getIdentityVec();
1242 sws_scaleVec(filter->chrH, -chromaSharpen);
1243 sws_scaleVec(filter->chrV, -chromaSharpen);
1244 sws_addVec(filter->chrH, id);
1245 sws_addVec(filter->chrV, id);
1249 if (lumaSharpen!=0.0) {
1250 SwsVector *id= sws_getIdentityVec();
1251 sws_scaleVec(filter->lumH, -lumaSharpen);
1252 sws_scaleVec(filter->lumV, -lumaSharpen);
1253 sws_addVec(filter->lumH, id);
1254 sws_addVec(filter->lumV, id);
1258 if (chromaHShift != 0.0)
1259 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
1261 if (chromaVShift != 0.0)
1262 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
1264 sws_normalizeVec(filter->chrH, 1.0);
1265 sws_normalizeVec(filter->chrV, 1.0);
1266 sws_normalizeVec(filter->lumH, 1.0);
1267 sws_normalizeVec(filter->lumV, 1.0);
1269 if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
1270 if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
1275 SwsVector *sws_allocVec(int length)
1277 SwsVector *vec = av_malloc(sizeof(SwsVector));
1280 vec->length = length;
1281 vec->coeff = av_malloc(sizeof(double) * length);
1287 SwsVector *sws_getGaussianVec(double variance, double quality)
1289 const int length= (int)(variance*quality + 0.5) | 1;
1291 double middle= (length-1)*0.5;
1292 SwsVector *vec= sws_allocVec(length);
1297 for (i=0; i<length; i++) {
1298 double dist= i-middle;
1299 vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);
1302 sws_normalizeVec(vec, 1.0);
1307 SwsVector *sws_getConstVec(double c, int length)
1310 SwsVector *vec= sws_allocVec(length);
1315 for (i=0; i<length; i++)
1321 SwsVector *sws_getIdentityVec(void)
1323 return sws_getConstVec(1.0, 1);
1326 static double sws_dcVec(SwsVector *a)
1331 for (i=0; i<a->length; i++)
1337 void sws_scaleVec(SwsVector *a, double scalar)
1341 for (i=0; i<a->length; i++)
1342 a->coeff[i]*= scalar;
1345 void sws_normalizeVec(SwsVector *a, double height)
1347 sws_scaleVec(a, height/sws_dcVec(a));
1350 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
1352 int length= a->length + b->length - 1;
1354 SwsVector *vec= sws_getConstVec(0.0, length);
1359 for (i=0; i<a->length; i++) {
1360 for (j=0; j<b->length; j++) {
1361 vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];
1368 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
1370 int length= FFMAX(a->length, b->length);
1372 SwsVector *vec= sws_getConstVec(0.0, length);
1377 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1378 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
1383 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
1385 int length= FFMAX(a->length, b->length);
1387 SwsVector *vec= sws_getConstVec(0.0, length);
1392 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1393 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
1398 /* shift left / or right if "shift" is negative */
1399 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
1401 int length= a->length + FFABS(shift)*2;
1403 SwsVector *vec= sws_getConstVec(0.0, length);
1408 for (i=0; i<a->length; i++) {
1409 vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
1415 void sws_shiftVec(SwsVector *a, int shift)
1417 SwsVector *shifted= sws_getShiftedVec(a, shift);
1419 a->coeff= shifted->coeff;
1420 a->length= shifted->length;
1424 void sws_addVec(SwsVector *a, SwsVector *b)
1426 SwsVector *sum= sws_sumVec(a, b);
1428 a->coeff= sum->coeff;
1429 a->length= sum->length;
1433 void sws_subVec(SwsVector *a, SwsVector *b)
1435 SwsVector *diff= sws_diffVec(a, b);
1437 a->coeff= diff->coeff;
1438 a->length= diff->length;
1442 void sws_convVec(SwsVector *a, SwsVector *b)
1444 SwsVector *conv= sws_getConvVec(a, b);
1446 a->coeff= conv->coeff;
1447 a->length= conv->length;
1451 SwsVector *sws_cloneVec(SwsVector *a)
1454 SwsVector *vec= sws_allocVec(a->length);
1459 for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];
1464 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
1471 for (i=0; i<a->length; i++)
1472 if (a->coeff[i]>max) max= a->coeff[i];
1474 for (i=0; i<a->length; i++)
1475 if (a->coeff[i]<min) min= a->coeff[i];
1479 for (i=0; i<a->length; i++) {
1480 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
1481 av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
1482 for (;x>0; x--) av_log(log_ctx, log_level, " ");
1483 av_log(log_ctx, log_level, "|\n");
1487 void sws_freeVec(SwsVector *a)
1490 av_freep(&a->coeff);
1495 void sws_freeFilter(SwsFilter *filter)
1497 if (!filter) return;
1499 if (filter->lumH) sws_freeVec(filter->lumH);
1500 if (filter->lumV) sws_freeVec(filter->lumV);
1501 if (filter->chrH) sws_freeVec(filter->chrH);
1502 if (filter->chrV) sws_freeVec(filter->chrV);
1506 void sws_freeContext(SwsContext *c)
1512 for (i=0; i<c->vLumBufSize; i++)
1513 av_freep(&c->lumPixBuf[i]);
1514 av_freep(&c->lumPixBuf);
1517 if (c->chrUPixBuf) {
1518 for (i=0; i<c->vChrBufSize; i++)
1519 av_freep(&c->chrUPixBuf[i]);
1520 av_freep(&c->chrUPixBuf);
1521 av_freep(&c->chrVPixBuf);
1524 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1525 for (i=0; i<c->vLumBufSize; i++)
1526 av_freep(&c->alpPixBuf[i]);
1527 av_freep(&c->alpPixBuf);
1530 av_freep(&c->vLumFilter);
1531 av_freep(&c->vChrFilter);
1532 av_freep(&c->hLumFilter);
1533 av_freep(&c->hChrFilter);
1535 av_freep(&c->vYCoeffsBank);
1536 av_freep(&c->vCCoeffsBank);
1539 av_freep(&c->vLumFilterPos);
1540 av_freep(&c->vChrFilterPos);
1541 av_freep(&c->hLumFilterPos);
1542 av_freep(&c->hChrFilterPos);
1545 #ifdef MAP_ANONYMOUS
1546 if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);
1547 if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);
1548 #elif HAVE_VIRTUALALLOC
1549 if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);
1550 if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);
1552 av_free(c->lumMmx2FilterCode);
1553 av_free(c->chrMmx2FilterCode);
1555 c->lumMmx2FilterCode=NULL;
1556 c->chrMmx2FilterCode=NULL;
1557 #endif /* HAVE_MMX */
1559 av_freep(&c->yuvTable);
1560 av_free(c->formatConvBuffer);
1565 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
1566 int srcW, int srcH, enum PixelFormat srcFormat,
1567 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
1568 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
1570 static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};
1573 param = default_param;
1576 (context->srcW != srcW ||
1577 context->srcH != srcH ||
1578 context->srcFormat != srcFormat ||
1579 context->dstW != dstW ||
1580 context->dstH != dstH ||
1581 context->dstFormat != dstFormat ||
1582 context->flags != flags ||
1583 context->param[0] != param[0] ||
1584 context->param[1] != param[1])) {
1585 sws_freeContext(context);
1590 if (!(context = sws_alloc_context()))
1592 context->srcW = srcW;
1593 context->srcH = srcH;
1594 context->srcRange = handle_jpeg(&srcFormat);
1595 context->srcFormat = srcFormat;
1596 context->dstW = dstW;
1597 context->dstH = dstH;
1598 context->dstRange = handle_jpeg(&dstFormat);
1599 context->dstFormat = dstFormat;
1600 context->flags = flags;
1601 context->param[0] = param[0];
1602 context->param[1] = param[1];
1603 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);
1604 if (sws_init_context(context, srcFilter, dstFilter) < 0) {
1605 sws_freeContext(context);