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 typedef struct FormatEntry {
67 int is_supported_in, is_supported_out;
70 const static FormatEntry format_entries[PIX_FMT_NB] = {
71 [PIX_FMT_YUV420P] = { 1 , 1 },
72 [PIX_FMT_YUYV422] = { 1 , 1 },
73 [PIX_FMT_RGB24] = { 1 , 1 },
74 [PIX_FMT_BGR24] = { 1 , 1 },
75 [PIX_FMT_YUV422P] = { 1 , 1 },
76 [PIX_FMT_YUV444P] = { 1 , 1 },
77 [PIX_FMT_YUV410P] = { 1 , 1 },
78 [PIX_FMT_YUV411P] = { 1 , 1 },
79 [PIX_FMT_GRAY8] = { 1 , 1 },
80 [PIX_FMT_MONOWHITE] = { 1 , 1 },
81 [PIX_FMT_MONOBLACK] = { 1 , 1 },
82 [PIX_FMT_PAL8] = { 1 , 0 },
83 [PIX_FMT_YUVJ420P] = { 1 , 1 },
84 [PIX_FMT_YUVJ422P] = { 1 , 1 },
85 [PIX_FMT_YUVJ444P] = { 1 , 1 },
86 [PIX_FMT_UYVY422] = { 1 , 1 },
87 [PIX_FMT_UYYVYY411] = { 0 , 0 },
88 [PIX_FMT_BGR8] = { 1 , 1 },
89 [PIX_FMT_BGR4] = { 0 , 1 },
90 [PIX_FMT_BGR4_BYTE] = { 1 , 1 },
91 [PIX_FMT_RGB8] = { 1 , 1 },
92 [PIX_FMT_RGB4] = { 0 , 1 },
93 [PIX_FMT_RGB4_BYTE] = { 1 , 1 },
94 [PIX_FMT_NV12] = { 1 , 1 },
95 [PIX_FMT_NV21] = { 1 , 1 },
96 [PIX_FMT_ARGB] = { 1 , 1 },
97 [PIX_FMT_RGBA] = { 1 , 1 },
98 [PIX_FMT_ABGR] = { 1 , 1 },
99 [PIX_FMT_BGRA] = { 1 , 1 },
100 [PIX_FMT_GRAY16BE] = { 1 , 1 },
101 [PIX_FMT_GRAY16LE] = { 1 , 1 },
102 [PIX_FMT_YUV440P] = { 1 , 1 },
103 [PIX_FMT_YUVJ440P] = { 1 , 1 },
104 [PIX_FMT_YUVA420P] = { 1 , 1 },
105 [PIX_FMT_RGB48BE] = { 1 , 1 },
106 [PIX_FMT_RGB48LE] = { 1 , 1 },
107 [PIX_FMT_RGB565BE] = { 1 , 1 },
108 [PIX_FMT_RGB565LE] = { 1 , 1 },
109 [PIX_FMT_RGB555BE] = { 1 , 1 },
110 [PIX_FMT_RGB555LE] = { 1 , 1 },
111 [PIX_FMT_BGR565BE] = { 1 , 1 },
112 [PIX_FMT_BGR565LE] = { 1 , 1 },
113 [PIX_FMT_BGR555BE] = { 1 , 1 },
114 [PIX_FMT_BGR555LE] = { 1 , 1 },
115 [PIX_FMT_YUV420P16LE] = { 1 , 1 },
116 [PIX_FMT_YUV420P16BE] = { 1 , 1 },
117 [PIX_FMT_YUV422P16LE] = { 1 , 1 },
118 [PIX_FMT_YUV422P16BE] = { 1 , 1 },
119 [PIX_FMT_YUV444P16LE] = { 1 , 1 },
120 [PIX_FMT_YUV444P16BE] = { 1 , 1 },
121 [PIX_FMT_RGB444LE] = { 0 , 1 },
122 [PIX_FMT_RGB444BE] = { 0 , 1 },
123 [PIX_FMT_BGR444LE] = { 0 , 1 },
124 [PIX_FMT_BGR444BE] = { 0 , 1 },
125 [PIX_FMT_Y400A] = { 1 , 0 },
126 [PIX_FMT_BGR48BE] = { 1 , 1 },
127 [PIX_FMT_BGR48LE] = { 1 , 1 },
128 [PIX_FMT_YUV420P9BE] = { 1 , 1 },
129 [PIX_FMT_YUV420P9LE] = { 1 , 1 },
130 [PIX_FMT_YUV420P10BE] = { 1 , 1 },
131 [PIX_FMT_YUV420P10LE] = { 1 , 1 },
132 [PIX_FMT_YUV422P9BE] = { 1 , 1 },
133 [PIX_FMT_YUV422P9LE] = { 1 , 1 },
134 [PIX_FMT_YUV422P10BE] = { 1 , 1 },
135 [PIX_FMT_YUV422P10LE] = { 1 , 1 },
136 [PIX_FMT_YUV444P9BE] = { 1 , 1 },
137 [PIX_FMT_YUV444P9LE] = { 1 , 1 },
138 [PIX_FMT_YUV444P10BE] = { 1 , 1 },
139 [PIX_FMT_YUV444P10LE] = { 1 , 1 },
142 int sws_isSupportedInput(enum PixelFormat pix_fmt)
144 return (unsigned)pix_fmt < PIX_FMT_NB ?
145 format_entries[pix_fmt].is_supported_in : 0;
148 int sws_isSupportedOutput(enum PixelFormat pix_fmt)
150 return (unsigned)pix_fmt < PIX_FMT_NB ?
151 format_entries[pix_fmt].is_supported_out : 0;
154 extern const int32_t ff_yuv2rgb_coeffs[8][4];
156 const char *sws_format_name(enum PixelFormat format)
158 if ((unsigned)format < PIX_FMT_NB && av_pix_fmt_descriptors[format].name)
159 return av_pix_fmt_descriptors[format].name;
161 return "Unknown format";
164 static double getSplineCoeff(double a, double b, double c, double d, double dist)
166 if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
167 else return getSplineCoeff( 0.0,
174 static int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
175 int srcW, int dstW, int filterAlign, int one, int flags, int cpu_flags,
176 SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
182 int64_t *filter=NULL;
183 int64_t *filter2=NULL;
184 const int64_t fone= 1LL<<54;
187 emms_c(); //FIXME this should not be required but it IS (even for non-MMX versions)
189 // NOTE: the +3 is for the MMX(+1)/SSE(+3) scaler which reads over the end
190 FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW+3)*sizeof(int16_t), fail);
192 if (FFABS(xInc - 0x10000) <10) { // unscaled
195 FF_ALLOCZ_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
197 for (i=0; i<dstW; i++) {
198 filter[i*filterSize]= fone;
202 } else if (flags&SWS_POINT) { // lame looking point sampling mode
206 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
208 xDstInSrc= xInc/2 - 0x8000;
209 for (i=0; i<dstW; i++) {
210 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
216 } else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) { // bilinear upscale
220 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
222 xDstInSrc= xInc/2 - 0x8000;
223 for (i=0; i<dstW; i++) {
224 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
228 //bilinear upscale / linear interpolate / area averaging
229 for (j=0; j<filterSize; j++) {
230 int64_t coeff= fone - FFABS((xx<<16) - xDstInSrc)*(fone>>16);
231 if (coeff<0) coeff=0;
232 filter[i*filterSize + j]= coeff;
241 if (flags&SWS_BICUBIC) sizeFactor= 4;
242 else if (flags&SWS_X) sizeFactor= 8;
243 else if (flags&SWS_AREA) sizeFactor= 1; //downscale only, for upscale it is bilinear
244 else if (flags&SWS_GAUSS) sizeFactor= 8; // infinite ;)
245 else if (flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? ceil(2*param[0]) : 6;
246 else if (flags&SWS_SINC) sizeFactor= 20; // infinite ;)
247 else if (flags&SWS_SPLINE) sizeFactor= 20; // infinite ;)
248 else if (flags&SWS_BILINEAR) sizeFactor= 2;
250 sizeFactor= 0; //GCC warning killer
254 if (xInc <= 1<<16) filterSize= 1 + sizeFactor; // upscale
255 else filterSize= 1 + (sizeFactor*srcW + dstW - 1)/ dstW;
257 if (filterSize > srcW-2) filterSize=srcW-2;
259 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
261 xDstInSrc= xInc - 0x10000;
262 for (i=0; i<dstW; i++) {
263 int xx= (xDstInSrc - ((filterSize-2)<<16)) / (1<<17);
266 for (j=0; j<filterSize; j++) {
267 int64_t d= ((int64_t)FFABS((xx<<17) - xDstInSrc))<<13;
273 floatd= d * (1.0/(1<<30));
275 if (flags & SWS_BICUBIC) {
276 int64_t B= (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1<<24);
277 int64_t C= (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1<<24);
282 int64_t dd = (d * d) >> 30;
283 int64_t ddd = (dd * d) >> 30;
286 coeff = (12*(1<<24)-9*B-6*C)*ddd + (-18*(1<<24)+12*B+6*C)*dd + (6*(1<<24)-2*B)*(1<<30);
288 coeff = (-B-6*C)*ddd + (6*B+30*C)*dd + (-12*B-48*C)*d + (8*B+24*C)*(1<<30);
290 coeff *= fone>>(30+24);
292 /* else if (flags & SWS_X) {
293 double p= param ? param*0.01 : 0.3;
294 coeff = d ? sin(d*M_PI)/(d*M_PI) : 1.0;
295 coeff*= pow(2.0, - p*d*d);
297 else if (flags & SWS_X) {
298 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
302 c = cos(floatd*M_PI);
305 if (c<0.0) c= -pow(-c, A);
307 coeff= (c*0.5 + 0.5)*fone;
308 } else if (flags & SWS_AREA) {
309 int64_t d2= d - (1<<29);
310 if (d2*xInc < -(1LL<<(29+16))) coeff= 1.0 * (1LL<<(30+16));
311 else if (d2*xInc < (1LL<<(29+16))) coeff= -d2*xInc + (1LL<<(29+16));
313 coeff *= fone>>(30+16);
314 } else if (flags & SWS_GAUSS) {
315 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
316 coeff = (pow(2.0, - p*floatd*floatd))*fone;
317 } else if (flags & SWS_SINC) {
318 coeff = (d ? sin(floatd*M_PI)/(floatd*M_PI) : 1.0)*fone;
319 } else if (flags & SWS_LANCZOS) {
320 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
321 coeff = (d ? sin(floatd*M_PI)*sin(floatd*M_PI/p)/(floatd*floatd*M_PI*M_PI/p) : 1.0)*fone;
322 if (floatd>p) coeff=0;
323 } else if (flags & SWS_BILINEAR) {
325 if (coeff<0) coeff=0;
327 } else if (flags & SWS_SPLINE) {
328 double p=-2.196152422706632;
329 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, floatd) * fone;
331 coeff= 0.0; //GCC warning killer
335 filter[i*filterSize + j]= coeff;
342 /* apply src & dst Filter to filter -> filter2
345 assert(filterSize>0);
346 filter2Size= filterSize;
347 if (srcFilter) filter2Size+= srcFilter->length - 1;
348 if (dstFilter) filter2Size+= dstFilter->length - 1;
349 assert(filter2Size>0);
350 FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size*dstW*sizeof(*filter2), fail);
352 for (i=0; i<dstW; i++) {
356 for (k=0; k<srcFilter->length; k++) {
357 for (j=0; j<filterSize; j++)
358 filter2[i*filter2Size + k + j] += srcFilter->coeff[k]*filter[i*filterSize + j];
361 for (j=0; j<filterSize; j++)
362 filter2[i*filter2Size + j]= filter[i*filterSize + j];
366 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
370 /* try to reduce the filter-size (step1 find size and shift left) */
371 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
373 for (i=dstW-1; i>=0; i--) {
374 int min= filter2Size;
378 /* get rid of near zero elements on the left by shifting left */
379 for (j=0; j<filter2Size; j++) {
381 cutOff += FFABS(filter2[i*filter2Size]);
383 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
385 /* preserve monotonicity because the core can't handle the filter otherwise */
386 if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
388 // move filter coefficients left
389 for (k=1; k<filter2Size; k++)
390 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
391 filter2[i*filter2Size + k - 1]= 0;
396 /* count near zeros on the right */
397 for (j=filter2Size-1; j>0; j--) {
398 cutOff += FFABS(filter2[i*filter2Size + j]);
400 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
404 if (min>minFilterSize) minFilterSize= min;
407 if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) {
408 // we can handle the special case 4,
409 // so we don't want to go to the full 8
410 if (minFilterSize < 5)
413 // We really don't want to waste our time
414 // doing useless computation, so fall back on
415 // the scalar C code for very small filters.
416 // Vectorizing is worth it only if you have a
417 // decent-sized vector.
418 if (minFilterSize < 3)
422 if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
423 // special case for unscaled vertical filtering
424 if (minFilterSize == 1 && filterAlign == 2)
428 assert(minFilterSize > 0);
429 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
430 assert(filterSize > 0);
431 filter= av_malloc(filterSize*dstW*sizeof(*filter));
432 if (filterSize >= MAX_FILTER_SIZE*16/((flags&SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
434 *outFilterSize= filterSize;
436 if (flags&SWS_PRINT_INFO)
437 av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
438 /* try to reduce the filter-size (step2 reduce it) */
439 for (i=0; i<dstW; i++) {
442 for (j=0; j<filterSize; j++) {
443 if (j>=filter2Size) filter[i*filterSize + j]= 0;
444 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
445 if((flags & SWS_BITEXACT) && j>=minFilterSize)
446 filter[i*filterSize + j]= 0;
450 //FIXME try to align filterPos if possible
453 for (i=0; i<dstW; i++) {
455 if ((*filterPos)[i] < 0) {
456 // move filter coefficients left to compensate for filterPos
457 for (j=1; j<filterSize; j++) {
458 int left= FFMAX(j + (*filterPos)[i], 0);
459 filter[i*filterSize + left] += filter[i*filterSize + j];
460 filter[i*filterSize + j]=0;
465 if ((*filterPos)[i] + filterSize > srcW) {
466 int shift= (*filterPos)[i] + filterSize - srcW;
467 // move filter coefficients right to compensate for filterPos
468 for (j=filterSize-2; j>=0; j--) {
469 int right= FFMIN(j + shift, filterSize-1);
470 filter[i*filterSize +right] += filter[i*filterSize +j];
471 filter[i*filterSize +j]=0;
473 (*filterPos)[i]= srcW - filterSize;
477 // Note the +1 is for the MMX scaler which reads over the end
478 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
479 FF_ALLOCZ_OR_GOTO(NULL, *outFilter, *outFilterSize*(dstW+3)*sizeof(int16_t), fail);
481 /* normalize & store in outFilter */
482 for (i=0; i<dstW; i++) {
487 for (j=0; j<filterSize; j++) {
488 sum+= filter[i*filterSize + j];
490 sum= (sum + one/2)/ one;
491 for (j=0; j<*outFilterSize; j++) {
492 int64_t v= filter[i*filterSize + j] + error;
493 int intV= ROUNDED_DIV(v, sum);
494 (*outFilter)[i*(*outFilterSize) + j]= intV;
499 (*filterPos)[dstW+0] =
500 (*filterPos)[dstW+1] =
501 (*filterPos)[dstW+2] = (*filterPos)[dstW-1]; // the MMX/SSE scaler will read over the end
502 for (i=0; i<*outFilterSize; i++) {
503 int k= (dstW - 1) * (*outFilterSize) + i;
504 (*outFilter)[k + 1 * (*outFilterSize)] =
505 (*outFilter)[k + 2 * (*outFilterSize)] =
506 (*outFilter)[k + 3 * (*outFilterSize)] = (*outFilter)[k];
517 static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode, int16_t *filter, int32_t *filterPos, int numSplits)
520 x86_reg imm8OfPShufW1A;
521 x86_reg imm8OfPShufW2A;
522 x86_reg fragmentLengthA;
524 x86_reg imm8OfPShufW1B;
525 x86_reg imm8OfPShufW2B;
526 x86_reg fragmentLengthB;
531 // create an optimized horizontal scaling routine
532 /* This scaler is made of runtime-generated MMX2 code using specially
533 * tuned pshufw instructions. For every four output pixels, if four
534 * input pixels are enough for the fast bilinear scaling, then a chunk
535 * of fragmentB is used. If five input pixels are needed, then a chunk
536 * of fragmentA is used.
545 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
546 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
547 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
548 "punpcklbw %%mm7, %%mm1 \n\t"
549 "punpcklbw %%mm7, %%mm0 \n\t"
550 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
552 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
554 "psubw %%mm1, %%mm0 \n\t"
555 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
556 "pmullw %%mm3, %%mm0 \n\t"
557 "psllw $7, %%mm1 \n\t"
558 "paddw %%mm1, %%mm0 \n\t"
560 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
562 "add $8, %%"REG_a" \n\t"
566 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
567 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
568 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
573 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
577 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
578 "=r" (fragmentLengthA)
585 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
586 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
587 "punpcklbw %%mm7, %%mm0 \n\t"
588 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
590 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
592 "psubw %%mm1, %%mm0 \n\t"
593 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
594 "pmullw %%mm3, %%mm0 \n\t"
595 "psllw $7, %%mm1 \n\t"
596 "paddw %%mm1, %%mm0 \n\t"
598 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
600 "add $8, %%"REG_a" \n\t"
604 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
605 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
606 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
611 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
615 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
616 "=r" (fragmentLengthB)
619 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
622 for (i=0; i<dstW/numSplits; i++) {
627 int b=((xpos+xInc)>>16) - xx;
628 int c=((xpos+xInc*2)>>16) - xx;
629 int d=((xpos+xInc*3)>>16) - xx;
631 uint8_t *fragment = (d+1<4) ? fragmentB : fragmentA;
632 x86_reg imm8OfPShufW1 = (d+1<4) ? imm8OfPShufW1B : imm8OfPShufW1A;
633 x86_reg imm8OfPShufW2 = (d+1<4) ? imm8OfPShufW2B : imm8OfPShufW2A;
634 x86_reg fragmentLength = (d+1<4) ? fragmentLengthB : fragmentLengthA;
635 int maxShift= 3-(d+inc);
639 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
640 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
641 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
642 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
645 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
647 filterCode[fragmentPos + imm8OfPShufW1]=
648 (a+inc) | ((b+inc)<<2) | ((c+inc)<<4) | ((d+inc)<<6);
649 filterCode[fragmentPos + imm8OfPShufW2]=
650 a | (b<<2) | (c<<4) | (d<<6);
652 if (i+4-inc>=dstW) shift=maxShift; //avoid overread
653 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
655 if (shift && i>=shift) {
656 filterCode[fragmentPos + imm8OfPShufW1]+= 0x55*shift;
657 filterCode[fragmentPos + imm8OfPShufW2]+= 0x55*shift;
658 filterPos[i/2]-=shift;
662 fragmentPos+= fragmentLength;
665 filterCode[fragmentPos]= RET;
670 filterPos[((i/2)+1)&(~1)]= xpos>>16; // needed to jump to the next part
672 return fragmentPos + 1;
674 #endif /* HAVE_MMX2 */
676 static void getSubSampleFactors(int *h, int *v, enum PixelFormat format)
678 *h = av_pix_fmt_descriptors[format].log2_chroma_w;
679 *v = av_pix_fmt_descriptors[format].log2_chroma_h;
682 int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],
683 int srcRange, const int table[4], int dstRange,
684 int brightness, int contrast, int saturation)
686 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
687 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
689 c->brightness= brightness;
690 c->contrast = contrast;
691 c->saturation= saturation;
692 c->srcRange = srcRange;
693 c->dstRange = dstRange;
694 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
696 c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->dstFormat]);
697 c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->srcFormat]);
699 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
702 if (HAVE_ALTIVEC && av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)
703 ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness, contrast, saturation);
707 int sws_getColorspaceDetails(struct SwsContext *c, int **inv_table,
708 int *srcRange, int **table, int *dstRange,
709 int *brightness, int *contrast, int *saturation)
711 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
713 *inv_table = c->srcColorspaceTable;
714 *table = c->dstColorspaceTable;
715 *srcRange = c->srcRange;
716 *dstRange = c->dstRange;
717 *brightness= c->brightness;
718 *contrast = c->contrast;
719 *saturation= c->saturation;
724 static int handle_jpeg(enum PixelFormat *format)
727 case PIX_FMT_YUVJ420P: *format = PIX_FMT_YUV420P; return 1;
728 case PIX_FMT_YUVJ422P: *format = PIX_FMT_YUV422P; return 1;
729 case PIX_FMT_YUVJ444P: *format = PIX_FMT_YUV444P; return 1;
730 case PIX_FMT_YUVJ440P: *format = PIX_FMT_YUV440P; return 1;
735 SwsContext *sws_alloc_context(void)
737 SwsContext *c= av_mallocz(sizeof(SwsContext));
739 c->av_class = &sws_context_class;
740 av_opt_set_defaults(c);
745 int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)
748 int usesVFilter, usesHFilter;
750 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
755 int dst_stride = FFALIGN(dstW * sizeof(int16_t) + 16, 16), dst_stride_px = dst_stride >> 1;
756 int flags, cpu_flags;
757 enum PixelFormat srcFormat= c->srcFormat;
758 enum PixelFormat dstFormat= c->dstFormat;
760 cpu_flags = av_get_cpu_flags();
763 if (!rgb15to16) sws_rgb2rgb_init();
765 unscaled = (srcW == dstW && srcH == dstH);
767 if (!sws_isSupportedInput(srcFormat)) {
768 av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n", sws_format_name(srcFormat));
769 return AVERROR(EINVAL);
771 if (!sws_isSupportedOutput(dstFormat)) {
772 av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n", sws_format_name(dstFormat));
773 return AVERROR(EINVAL);
776 i= flags & ( SWS_POINT
787 if(!i || (i & (i-1))) {
788 av_log(c, AV_LOG_ERROR, "Exactly one scaler algorithm must be chosen\n");
789 return AVERROR(EINVAL);
792 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
793 av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",
794 srcW, srcH, dstW, dstH);
795 return AVERROR(EINVAL);
798 if (!dstFilter) dstFilter= &dummyFilter;
799 if (!srcFilter) srcFilter= &dummyFilter;
801 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
802 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
803 c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[dstFormat]);
804 c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[srcFormat]);
805 c->vRounder= 4* 0x0001000100010001ULL;
807 usesVFilter = (srcFilter->lumV && srcFilter->lumV->length>1) ||
808 (srcFilter->chrV && srcFilter->chrV->length>1) ||
809 (dstFilter->lumV && dstFilter->lumV->length>1) ||
810 (dstFilter->chrV && dstFilter->chrV->length>1);
811 usesHFilter = (srcFilter->lumH && srcFilter->lumH->length>1) ||
812 (srcFilter->chrH && srcFilter->chrH->length>1) ||
813 (dstFilter->lumH && dstFilter->lumH->length>1) ||
814 (dstFilter->chrH && dstFilter->chrH->length>1);
816 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
817 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
819 // reuse chroma for 2 pixels RGB/BGR unless user wants full chroma interpolation
820 if (flags & SWS_FULL_CHR_H_INT &&
821 dstFormat != PIX_FMT_RGBA &&
822 dstFormat != PIX_FMT_ARGB &&
823 dstFormat != PIX_FMT_BGRA &&
824 dstFormat != PIX_FMT_ABGR &&
825 dstFormat != PIX_FMT_RGB24 &&
826 dstFormat != PIX_FMT_BGR24) {
827 av_log(c, AV_LOG_ERROR,
828 "full chroma interpolation for destination format '%s' not yet implemented\n",
829 sws_format_name(dstFormat));
830 flags &= ~SWS_FULL_CHR_H_INT;
833 if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
835 // drop some chroma lines if the user wants it
836 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
837 c->chrSrcVSubSample+= c->vChrDrop;
839 // drop every other pixel for chroma calculation unless user wants full chroma
840 if (isAnyRGB(srcFormat) && !(flags&SWS_FULL_CHR_H_INP)
841 && srcFormat!=PIX_FMT_RGB8 && srcFormat!=PIX_FMT_BGR8
842 && srcFormat!=PIX_FMT_RGB4 && srcFormat!=PIX_FMT_BGR4
843 && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE
844 && ((dstW>>c->chrDstHSubSample) <= (srcW>>1) || (flags&SWS_FAST_BILINEAR)))
845 c->chrSrcHSubSample=1;
847 // Note the -((-x)>>y) is so that we always round toward +inf.
848 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
849 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
850 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
851 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
853 /* unscaled special cases */
854 if (unscaled && !usesHFilter && !usesVFilter && (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
855 ff_get_unscaled_swscale(c);
858 if (flags&SWS_PRINT_INFO)
859 av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
860 sws_format_name(srcFormat), sws_format_name(dstFormat));
865 c->srcBpc = 1 + av_pix_fmt_descriptors[srcFormat].comp[0].depth_minus1;
868 c->dstBpc = 1 + av_pix_fmt_descriptors[dstFormat].comp[0].depth_minus1;
873 FF_ALLOC_OR_GOTO(c, c->formatConvBuffer,
874 FFALIGN(srcW, 16) * 2 * FFALIGN(c->srcBpc, 8) >> 3,
876 if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2 && c->srcBpc == 8 && c->dstBpc <= 10) {
877 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
878 if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR)) {
879 if (flags&SWS_PRINT_INFO)
880 av_log(c, AV_LOG_INFO, "output width is not a multiple of 32 -> no MMX2 scaler\n");
882 if (usesHFilter) c->canMMX2BeUsed=0;
887 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
888 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
890 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
891 // but only for the FAST_BILINEAR mode otherwise do correct scaling
892 // n-2 is the last chrominance sample available
893 // this is not perfect, but no one should notice the difference, the more correct variant
894 // would be like the vertical one, but that would require some special code for the
895 // first and last pixel
896 if (flags&SWS_FAST_BILINEAR) {
897 if (c->canMMX2BeUsed) {
901 //we don't use the x86 asm scaler if MMX is available
902 else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
903 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
904 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
908 /* precalculate horizontal scaler filter coefficients */
911 // can't downscale !!!
912 if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {
913 c->lumMmx2FilterCodeSize = initMMX2HScaler( dstW, c->lumXInc, NULL, NULL, NULL, 8);
914 c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc, NULL, NULL, NULL, 4);
917 c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
918 c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
919 #elif HAVE_VIRTUALALLOC
920 c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
921 c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
923 c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);
924 c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);
927 if (!c->lumMmx2FilterCode || !c->chrMmx2FilterCode)
928 return AVERROR(ENOMEM);
929 FF_ALLOCZ_OR_GOTO(c, c->hLumFilter , (dstW /8+8)*sizeof(int16_t), fail);
930 FF_ALLOCZ_OR_GOTO(c, c->hChrFilter , (c->chrDstW /4+8)*sizeof(int16_t), fail);
931 FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW /2/8+8)*sizeof(int32_t), fail);
932 FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW/2/4+8)*sizeof(int32_t), fail);
934 initMMX2HScaler( dstW, c->lumXInc, c->lumMmx2FilterCode, c->hLumFilter, c->hLumFilterPos, 8);
935 initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode, c->hChrFilter, c->hChrFilterPos, 4);
938 mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
939 mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
942 #endif /* HAVE_MMX2 */
944 const int filterAlign=
945 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :
946 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
949 if (initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
950 srcW , dstW, filterAlign, 1<<14,
951 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags, cpu_flags,
952 srcFilter->lumH, dstFilter->lumH, c->param) < 0)
954 if (initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
955 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
956 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, cpu_flags,
957 srcFilter->chrH, dstFilter->chrH, c->param) < 0)
960 } // initialize horizontal stuff
962 /* precalculate vertical scaler filter coefficients */
964 const int filterAlign=
965 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
966 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
969 if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
970 srcH , dstH, filterAlign, (1<<12),
971 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags, cpu_flags,
972 srcFilter->lumV, dstFilter->lumV, c->param) < 0)
974 if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
975 c->chrSrcH, c->chrDstH, filterAlign, (1<<12),
976 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, cpu_flags,
977 srcFilter->chrV, dstFilter->chrV, c->param) < 0)
981 FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof (vector signed short)*c->vLumFilterSize*c->dstH, fail);
982 FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH, fail);
984 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
986 short *p = (short *)&c->vYCoeffsBank[i];
988 p[j] = c->vLumFilter[i];
991 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
993 short *p = (short *)&c->vCCoeffsBank[i];
995 p[j] = c->vChrFilter[i];
1000 // calculate buffer sizes so that they won't run out while handling these damn slices
1001 c->vLumBufSize= c->vLumFilterSize;
1002 c->vChrBufSize= c->vChrFilterSize;
1003 for (i=0; i<dstH; i++) {
1004 int chrI= i*c->chrDstH / dstH;
1005 int nextSlice= FFMAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
1006 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
1008 nextSlice>>= c->chrSrcVSubSample;
1009 nextSlice<<= c->chrSrcVSubSample;
1010 if (c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
1011 c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
1012 if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
1013 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
1016 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
1017 // allocate several megabytes to handle all possible cases)
1018 FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
1019 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
1020 FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
1021 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
1022 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
1023 //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)
1024 /* align at 16 bytes for AltiVec */
1025 for (i=0; i<c->vLumBufSize; i++) {
1026 FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], dst_stride+16, fail);
1027 c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];
1029 // 64 / (c->dstBpc & ~7) is the same as 16 / sizeof(scaling_intermediate)
1030 c->uv_off_px = dst_stride_px + 64 / (c->dstBpc &~ 7);
1031 c->uv_off_byte = dst_stride + 16;
1032 for (i=0; i<c->vChrBufSize; i++) {
1033 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i+c->vChrBufSize], dst_stride*2+32, fail);
1034 c->chrUPixBuf[i] = c->chrUPixBuf[i+c->vChrBufSize];
1035 c->chrVPixBuf[i] = c->chrVPixBuf[i+c->vChrBufSize] = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;
1037 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
1038 for (i=0; i<c->vLumBufSize; i++) {
1039 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], dst_stride+16, fail);
1040 c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];
1043 //try to avoid drawing green stuff between the right end and the stride end
1044 for (i=0; i<c->vChrBufSize; i++)
1045 memset(c->chrUPixBuf[i], 64, dst_stride*2+1);
1047 assert(c->chrDstH <= dstH);
1049 if (flags&SWS_PRINT_INFO) {
1050 if (flags&SWS_FAST_BILINEAR) av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
1051 else if (flags&SWS_BILINEAR) av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
1052 else if (flags&SWS_BICUBIC) av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
1053 else if (flags&SWS_X) av_log(c, AV_LOG_INFO, "Experimental scaler, ");
1054 else if (flags&SWS_POINT) av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
1055 else if (flags&SWS_AREA) av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
1056 else if (flags&SWS_BICUBLIN) av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
1057 else if (flags&SWS_GAUSS) av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
1058 else if (flags&SWS_SINC) av_log(c, AV_LOG_INFO, "Sinc scaler, ");
1059 else if (flags&SWS_LANCZOS) av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
1060 else if (flags&SWS_SPLINE) av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
1061 else av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
1063 av_log(c, AV_LOG_INFO, "from %s to %s%s ",
1064 sws_format_name(srcFormat),
1066 dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ||
1067 dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||
1068 dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ? "dithered " : "",
1072 sws_format_name(dstFormat));
1074 if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2) av_log(c, AV_LOG_INFO, "using MMX2\n");
1075 else if (HAVE_AMD3DNOW && cpu_flags & AV_CPU_FLAG_3DNOW) av_log(c, AV_LOG_INFO, "using 3DNOW\n");
1076 else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) av_log(c, AV_LOG_INFO, "using MMX\n");
1077 else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) av_log(c, AV_LOG_INFO, "using AltiVec\n");
1078 else av_log(c, AV_LOG_INFO, "using C\n");
1080 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
1081 av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1082 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
1083 av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1084 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
1087 c->swScale= ff_getSwsFunc(c);
1089 fail: //FIXME replace things by appropriate error codes
1093 #if FF_API_SWS_GETCONTEXT
1094 SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,
1095 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
1096 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
1100 if(!(c=sws_alloc_context()))
1108 c->srcRange = handle_jpeg(&srcFormat);
1109 c->dstRange = handle_jpeg(&dstFormat);
1110 c->srcFormat= srcFormat;
1111 c->dstFormat= dstFormat;
1114 c->param[0] = param[0];
1115 c->param[1] = param[1];
1117 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);
1119 if(sws_init_context(c, srcFilter, dstFilter) < 0){
1128 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
1129 float lumaSharpen, float chromaSharpen,
1130 float chromaHShift, float chromaVShift,
1133 SwsFilter *filter= av_malloc(sizeof(SwsFilter));
1137 if (lumaGBlur!=0.0) {
1138 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
1139 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
1141 filter->lumH= sws_getIdentityVec();
1142 filter->lumV= sws_getIdentityVec();
1145 if (chromaGBlur!=0.0) {
1146 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
1147 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
1149 filter->chrH= sws_getIdentityVec();
1150 filter->chrV= sws_getIdentityVec();
1153 if (chromaSharpen!=0.0) {
1154 SwsVector *id= sws_getIdentityVec();
1155 sws_scaleVec(filter->chrH, -chromaSharpen);
1156 sws_scaleVec(filter->chrV, -chromaSharpen);
1157 sws_addVec(filter->chrH, id);
1158 sws_addVec(filter->chrV, id);
1162 if (lumaSharpen!=0.0) {
1163 SwsVector *id= sws_getIdentityVec();
1164 sws_scaleVec(filter->lumH, -lumaSharpen);
1165 sws_scaleVec(filter->lumV, -lumaSharpen);
1166 sws_addVec(filter->lumH, id);
1167 sws_addVec(filter->lumV, id);
1171 if (chromaHShift != 0.0)
1172 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
1174 if (chromaVShift != 0.0)
1175 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
1177 sws_normalizeVec(filter->chrH, 1.0);
1178 sws_normalizeVec(filter->chrV, 1.0);
1179 sws_normalizeVec(filter->lumH, 1.0);
1180 sws_normalizeVec(filter->lumV, 1.0);
1182 if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
1183 if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
1188 SwsVector *sws_allocVec(int length)
1190 SwsVector *vec = av_malloc(sizeof(SwsVector));
1193 vec->length = length;
1194 vec->coeff = av_malloc(sizeof(double) * length);
1200 SwsVector *sws_getGaussianVec(double variance, double quality)
1202 const int length= (int)(variance*quality + 0.5) | 1;
1204 double middle= (length-1)*0.5;
1205 SwsVector *vec= sws_allocVec(length);
1210 for (i=0; i<length; i++) {
1211 double dist= i-middle;
1212 vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);
1215 sws_normalizeVec(vec, 1.0);
1220 SwsVector *sws_getConstVec(double c, int length)
1223 SwsVector *vec= sws_allocVec(length);
1228 for (i=0; i<length; i++)
1234 SwsVector *sws_getIdentityVec(void)
1236 return sws_getConstVec(1.0, 1);
1239 static double sws_dcVec(SwsVector *a)
1244 for (i=0; i<a->length; i++)
1250 void sws_scaleVec(SwsVector *a, double scalar)
1254 for (i=0; i<a->length; i++)
1255 a->coeff[i]*= scalar;
1258 void sws_normalizeVec(SwsVector *a, double height)
1260 sws_scaleVec(a, height/sws_dcVec(a));
1263 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
1265 int length= a->length + b->length - 1;
1267 SwsVector *vec= sws_getConstVec(0.0, length);
1272 for (i=0; i<a->length; i++) {
1273 for (j=0; j<b->length; j++) {
1274 vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];
1281 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
1283 int length= FFMAX(a->length, b->length);
1285 SwsVector *vec= sws_getConstVec(0.0, length);
1290 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1291 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
1296 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
1298 int length= FFMAX(a->length, b->length);
1300 SwsVector *vec= sws_getConstVec(0.0, length);
1305 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1306 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
1311 /* shift left / or right if "shift" is negative */
1312 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
1314 int length= a->length + FFABS(shift)*2;
1316 SwsVector *vec= sws_getConstVec(0.0, length);
1321 for (i=0; i<a->length; i++) {
1322 vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
1328 void sws_shiftVec(SwsVector *a, int shift)
1330 SwsVector *shifted= sws_getShiftedVec(a, shift);
1332 a->coeff= shifted->coeff;
1333 a->length= shifted->length;
1337 void sws_addVec(SwsVector *a, SwsVector *b)
1339 SwsVector *sum= sws_sumVec(a, b);
1341 a->coeff= sum->coeff;
1342 a->length= sum->length;
1346 void sws_subVec(SwsVector *a, SwsVector *b)
1348 SwsVector *diff= sws_diffVec(a, b);
1350 a->coeff= diff->coeff;
1351 a->length= diff->length;
1355 void sws_convVec(SwsVector *a, SwsVector *b)
1357 SwsVector *conv= sws_getConvVec(a, b);
1359 a->coeff= conv->coeff;
1360 a->length= conv->length;
1364 SwsVector *sws_cloneVec(SwsVector *a)
1367 SwsVector *vec= sws_allocVec(a->length);
1372 for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];
1377 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
1384 for (i=0; i<a->length; i++)
1385 if (a->coeff[i]>max) max= a->coeff[i];
1387 for (i=0; i<a->length; i++)
1388 if (a->coeff[i]<min) min= a->coeff[i];
1392 for (i=0; i<a->length; i++) {
1393 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
1394 av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
1395 for (;x>0; x--) av_log(log_ctx, log_level, " ");
1396 av_log(log_ctx, log_level, "|\n");
1400 void sws_freeVec(SwsVector *a)
1403 av_freep(&a->coeff);
1408 void sws_freeFilter(SwsFilter *filter)
1410 if (!filter) return;
1412 if (filter->lumH) sws_freeVec(filter->lumH);
1413 if (filter->lumV) sws_freeVec(filter->lumV);
1414 if (filter->chrH) sws_freeVec(filter->chrH);
1415 if (filter->chrV) sws_freeVec(filter->chrV);
1419 void sws_freeContext(SwsContext *c)
1425 for (i=0; i<c->vLumBufSize; i++)
1426 av_freep(&c->lumPixBuf[i]);
1427 av_freep(&c->lumPixBuf);
1430 if (c->chrUPixBuf) {
1431 for (i=0; i<c->vChrBufSize; i++)
1432 av_freep(&c->chrUPixBuf[i]);
1433 av_freep(&c->chrUPixBuf);
1434 av_freep(&c->chrVPixBuf);
1437 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1438 for (i=0; i<c->vLumBufSize; i++)
1439 av_freep(&c->alpPixBuf[i]);
1440 av_freep(&c->alpPixBuf);
1443 av_freep(&c->vLumFilter);
1444 av_freep(&c->vChrFilter);
1445 av_freep(&c->hLumFilter);
1446 av_freep(&c->hChrFilter);
1448 av_freep(&c->vYCoeffsBank);
1449 av_freep(&c->vCCoeffsBank);
1452 av_freep(&c->vLumFilterPos);
1453 av_freep(&c->vChrFilterPos);
1454 av_freep(&c->hLumFilterPos);
1455 av_freep(&c->hChrFilterPos);
1458 #ifdef MAP_ANONYMOUS
1459 if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);
1460 if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);
1461 #elif HAVE_VIRTUALALLOC
1462 if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);
1463 if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);
1465 av_free(c->lumMmx2FilterCode);
1466 av_free(c->chrMmx2FilterCode);
1468 c->lumMmx2FilterCode=NULL;
1469 c->chrMmx2FilterCode=NULL;
1470 #endif /* HAVE_MMX */
1472 av_freep(&c->yuvTable);
1473 av_free(c->formatConvBuffer);
1478 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
1479 int srcW, int srcH, enum PixelFormat srcFormat,
1480 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
1481 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
1483 static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};
1486 param = default_param;
1489 (context->srcW != srcW ||
1490 context->srcH != srcH ||
1491 context->srcFormat != srcFormat ||
1492 context->dstW != dstW ||
1493 context->dstH != dstH ||
1494 context->dstFormat != dstFormat ||
1495 context->flags != flags ||
1496 context->param[0] != param[0] ||
1497 context->param[1] != param[1])) {
1498 sws_freeContext(context);
1503 if (!(context = sws_alloc_context()))
1505 context->srcW = srcW;
1506 context->srcH = srcH;
1507 context->srcRange = handle_jpeg(&srcFormat);
1508 context->srcFormat = srcFormat;
1509 context->dstW = dstW;
1510 context->dstH = dstH;
1511 context->dstRange = handle_jpeg(&dstFormat);
1512 context->dstFormat = dstFormat;
1513 context->flags = flags;
1514 context->param[0] = param[0];
1515 context->param[1] = param[1];
1516 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);
1517 if (sws_init_context(context, srcFilter, dstFilter) < 0) {
1518 sws_freeContext(context);