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_YUV422P10BE] = { 1 , 1 },
133 [PIX_FMT_YUV422P10LE] = { 1 , 1 },
134 [PIX_FMT_YUV444P9BE] = { 1 , 0 },
135 [PIX_FMT_YUV444P9LE] = { 1 , 0 },
136 [PIX_FMT_YUV444P10BE] = { 1 , 0 },
137 [PIX_FMT_YUV444P10LE] = { 1 , 0 },
140 int sws_isSupportedInput(enum PixelFormat pix_fmt)
142 return (unsigned)pix_fmt < PIX_FMT_NB ?
143 format_entries[pix_fmt].is_supported_in : 0;
146 int sws_isSupportedOutput(enum PixelFormat pix_fmt)
148 return (unsigned)pix_fmt < PIX_FMT_NB ?
149 format_entries[pix_fmt].is_supported_out : 0;
152 extern const int32_t ff_yuv2rgb_coeffs[8][4];
154 const char *sws_format_name(enum PixelFormat format)
156 if ((unsigned)format < PIX_FMT_NB && av_pix_fmt_descriptors[format].name)
157 return av_pix_fmt_descriptors[format].name;
159 return "Unknown format";
162 static double getSplineCoeff(double a, double b, double c, double d, double dist)
164 if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
165 else return getSplineCoeff( 0.0,
172 static int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
173 int srcW, int dstW, int filterAlign, int one, int flags, int cpu_flags,
174 SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
180 int64_t *filter=NULL;
181 int64_t *filter2=NULL;
182 const int64_t fone= 1LL<<54;
185 emms_c(); //FIXME this should not be required but it IS (even for non-MMX versions)
187 // NOTE: the +3 is for the MMX(+1)/SSE(+3) scaler which reads over the end
188 FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW+3)*sizeof(int16_t), fail);
190 if (FFABS(xInc - 0x10000) <10) { // unscaled
193 FF_ALLOCZ_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
195 for (i=0; i<dstW; i++) {
196 filter[i*filterSize]= fone;
200 } else if (flags&SWS_POINT) { // lame looking point sampling mode
204 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
206 xDstInSrc= xInc/2 - 0x8000;
207 for (i=0; i<dstW; i++) {
208 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
214 } else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) { // bilinear upscale
218 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
220 xDstInSrc= xInc/2 - 0x8000;
221 for (i=0; i<dstW; i++) {
222 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
226 //bilinear upscale / linear interpolate / area averaging
227 for (j=0; j<filterSize; j++) {
228 int64_t coeff= fone - FFABS((xx<<16) - xDstInSrc)*(fone>>16);
229 if (coeff<0) coeff=0;
230 filter[i*filterSize + j]= coeff;
239 if (flags&SWS_BICUBIC) sizeFactor= 4;
240 else if (flags&SWS_X) sizeFactor= 8;
241 else if (flags&SWS_AREA) sizeFactor= 1; //downscale only, for upscale it is bilinear
242 else if (flags&SWS_GAUSS) sizeFactor= 8; // infinite ;)
243 else if (flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? ceil(2*param[0]) : 6;
244 else if (flags&SWS_SINC) sizeFactor= 20; // infinite ;)
245 else if (flags&SWS_SPLINE) sizeFactor= 20; // infinite ;)
246 else if (flags&SWS_BILINEAR) sizeFactor= 2;
248 sizeFactor= 0; //GCC warning killer
252 if (xInc <= 1<<16) filterSize= 1 + sizeFactor; // upscale
253 else filterSize= 1 + (sizeFactor*srcW + dstW - 1)/ dstW;
255 if (filterSize > srcW-2) filterSize=srcW-2;
257 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
259 xDstInSrc= xInc - 0x10000;
260 for (i=0; i<dstW; i++) {
261 int xx= (xDstInSrc - ((filterSize-2)<<16)) / (1<<17);
264 for (j=0; j<filterSize; j++) {
265 int64_t d= ((int64_t)FFABS((xx<<17) - xDstInSrc))<<13;
271 floatd= d * (1.0/(1<<30));
273 if (flags & SWS_BICUBIC) {
274 int64_t B= (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1<<24);
275 int64_t C= (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1<<24);
276 int64_t dd = ( d*d)>>30;
277 int64_t ddd= (dd*d)>>30;
280 coeff = (12*(1<<24)-9*B-6*C)*ddd + (-18*(1<<24)+12*B+6*C)*dd + (6*(1<<24)-2*B)*(1<<30);
281 else if (d < 1LL<<31)
282 coeff = (-B-6*C)*ddd + (6*B+30*C)*dd + (-12*B-48*C)*d + (8*B+24*C)*(1<<30);
285 coeff *= fone>>(30+24);
287 /* else if (flags & SWS_X) {
288 double p= param ? param*0.01 : 0.3;
289 coeff = d ? sin(d*M_PI)/(d*M_PI) : 1.0;
290 coeff*= pow(2.0, - p*d*d);
292 else if (flags & SWS_X) {
293 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
297 c = cos(floatd*M_PI);
300 if (c<0.0) c= -pow(-c, A);
302 coeff= (c*0.5 + 0.5)*fone;
303 } else if (flags & SWS_AREA) {
304 int64_t d2= d - (1<<29);
305 if (d2*xInc < -(1LL<<(29+16))) coeff= 1.0 * (1LL<<(30+16));
306 else if (d2*xInc < (1LL<<(29+16))) coeff= -d2*xInc + (1LL<<(29+16));
308 coeff *= fone>>(30+16);
309 } else if (flags & SWS_GAUSS) {
310 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
311 coeff = (pow(2.0, - p*floatd*floatd))*fone;
312 } else if (flags & SWS_SINC) {
313 coeff = (d ? sin(floatd*M_PI)/(floatd*M_PI) : 1.0)*fone;
314 } else if (flags & SWS_LANCZOS) {
315 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
316 coeff = (d ? sin(floatd*M_PI)*sin(floatd*M_PI/p)/(floatd*floatd*M_PI*M_PI/p) : 1.0)*fone;
317 if (floatd>p) coeff=0;
318 } else if (flags & SWS_BILINEAR) {
320 if (coeff<0) coeff=0;
322 } else if (flags & SWS_SPLINE) {
323 double p=-2.196152422706632;
324 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, floatd) * fone;
326 coeff= 0.0; //GCC warning killer
330 filter[i*filterSize + j]= coeff;
337 /* apply src & dst Filter to filter -> filter2
340 assert(filterSize>0);
341 filter2Size= filterSize;
342 if (srcFilter) filter2Size+= srcFilter->length - 1;
343 if (dstFilter) filter2Size+= dstFilter->length - 1;
344 assert(filter2Size>0);
345 FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size*dstW*sizeof(*filter2), fail);
347 for (i=0; i<dstW; i++) {
351 for (k=0; k<srcFilter->length; k++) {
352 for (j=0; j<filterSize; j++)
353 filter2[i*filter2Size + k + j] += srcFilter->coeff[k]*filter[i*filterSize + j];
356 for (j=0; j<filterSize; j++)
357 filter2[i*filter2Size + j]= filter[i*filterSize + j];
361 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
365 /* try to reduce the filter-size (step1 find size and shift left) */
366 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
368 for (i=dstW-1; i>=0; i--) {
369 int min= filter2Size;
373 /* get rid of near zero elements on the left by shifting left */
374 for (j=0; j<filter2Size; j++) {
376 cutOff += FFABS(filter2[i*filter2Size]);
378 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
380 /* preserve monotonicity because the core can't handle the filter otherwise */
381 if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
383 // move filter coefficients left
384 for (k=1; k<filter2Size; k++)
385 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
386 filter2[i*filter2Size + k - 1]= 0;
391 /* count near zeros on the right */
392 for (j=filter2Size-1; j>0; j--) {
393 cutOff += FFABS(filter2[i*filter2Size + j]);
395 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
399 if (min>minFilterSize) minFilterSize= min;
402 if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) {
403 // we can handle the special case 4,
404 // so we don't want to go to the full 8
405 if (minFilterSize < 5)
408 // We really don't want to waste our time
409 // doing useless computation, so fall back on
410 // the scalar C code for very small filters.
411 // Vectorizing is worth it only if you have a
412 // decent-sized vector.
413 if (minFilterSize < 3)
417 if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
418 // special case for unscaled vertical filtering
419 if (minFilterSize == 1 && filterAlign == 2)
423 assert(minFilterSize > 0);
424 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
425 assert(filterSize > 0);
426 filter= av_malloc(filterSize*dstW*sizeof(*filter));
427 if (filterSize >= MAX_FILTER_SIZE*16/((flags&SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
429 *outFilterSize= filterSize;
431 if (flags&SWS_PRINT_INFO)
432 av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
433 /* try to reduce the filter-size (step2 reduce it) */
434 for (i=0; i<dstW; i++) {
437 for (j=0; j<filterSize; j++) {
438 if (j>=filter2Size) filter[i*filterSize + j]= 0;
439 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
440 if((flags & SWS_BITEXACT) && j>=minFilterSize)
441 filter[i*filterSize + j]= 0;
445 //FIXME try to align filterPos if possible
448 for (i=0; i<dstW; i++) {
450 if ((*filterPos)[i] < 0) {
451 // move filter coefficients left to compensate for filterPos
452 for (j=1; j<filterSize; j++) {
453 int left= FFMAX(j + (*filterPos)[i], 0);
454 filter[i*filterSize + left] += filter[i*filterSize + j];
455 filter[i*filterSize + j]=0;
460 if ((*filterPos)[i] + filterSize > srcW) {
461 int shift= (*filterPos)[i] + filterSize - srcW;
462 // move filter coefficients right to compensate for filterPos
463 for (j=filterSize-2; j>=0; j--) {
464 int right= FFMIN(j + shift, filterSize-1);
465 filter[i*filterSize +right] += filter[i*filterSize +j];
466 filter[i*filterSize +j]=0;
468 (*filterPos)[i]= srcW - filterSize;
472 // Note the +1 is for the MMX scaler which reads over the end
473 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
474 FF_ALLOCZ_OR_GOTO(NULL, *outFilter, *outFilterSize*(dstW+3)*sizeof(int16_t), fail);
476 /* normalize & store in outFilter */
477 for (i=0; i<dstW; i++) {
482 for (j=0; j<filterSize; j++) {
483 sum+= filter[i*filterSize + j];
485 sum= (sum + one/2)/ one;
486 for (j=0; j<*outFilterSize; j++) {
487 int64_t v= filter[i*filterSize + j] + error;
488 int intV= ROUNDED_DIV(v, sum);
489 (*outFilter)[i*(*outFilterSize) + j]= intV;
494 (*filterPos)[dstW+0] =
495 (*filterPos)[dstW+1] =
496 (*filterPos)[dstW+2] = (*filterPos)[dstW-1]; // the MMX/SSE scaler will read over the end
497 for (i=0; i<*outFilterSize; i++) {
498 int k= (dstW - 1) * (*outFilterSize) + i;
499 (*outFilter)[k + 1 * (*outFilterSize)] =
500 (*outFilter)[k + 2 * (*outFilterSize)] =
501 (*outFilter)[k + 3 * (*outFilterSize)] = (*outFilter)[k];
512 static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode, int16_t *filter, int32_t *filterPos, int numSplits)
515 x86_reg imm8OfPShufW1A;
516 x86_reg imm8OfPShufW2A;
517 x86_reg fragmentLengthA;
519 x86_reg imm8OfPShufW1B;
520 x86_reg imm8OfPShufW2B;
521 x86_reg fragmentLengthB;
526 // create an optimized horizontal scaling routine
527 /* This scaler is made of runtime-generated MMX2 code using specially
528 * tuned pshufw instructions. For every four output pixels, if four
529 * input pixels are enough for the fast bilinear scaling, then a chunk
530 * of fragmentB is used. If five input pixels are needed, then a chunk
531 * of fragmentA is used.
540 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
541 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
542 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
543 "punpcklbw %%mm7, %%mm1 \n\t"
544 "punpcklbw %%mm7, %%mm0 \n\t"
545 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
547 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
549 "psubw %%mm1, %%mm0 \n\t"
550 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
551 "pmullw %%mm3, %%mm0 \n\t"
552 "psllw $7, %%mm1 \n\t"
553 "paddw %%mm1, %%mm0 \n\t"
555 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
557 "add $8, %%"REG_a" \n\t"
561 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
562 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
563 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
568 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
572 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
573 "=r" (fragmentLengthA)
580 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
581 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
582 "punpcklbw %%mm7, %%mm0 \n\t"
583 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
585 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
587 "psubw %%mm1, %%mm0 \n\t"
588 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
589 "pmullw %%mm3, %%mm0 \n\t"
590 "psllw $7, %%mm1 \n\t"
591 "paddw %%mm1, %%mm0 \n\t"
593 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
595 "add $8, %%"REG_a" \n\t"
599 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
600 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
601 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
606 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
610 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
611 "=r" (fragmentLengthB)
614 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
617 for (i=0; i<dstW/numSplits; i++) {
622 int b=((xpos+xInc)>>16) - xx;
623 int c=((xpos+xInc*2)>>16) - xx;
624 int d=((xpos+xInc*3)>>16) - xx;
626 uint8_t *fragment = (d+1<4) ? fragmentB : fragmentA;
627 x86_reg imm8OfPShufW1 = (d+1<4) ? imm8OfPShufW1B : imm8OfPShufW1A;
628 x86_reg imm8OfPShufW2 = (d+1<4) ? imm8OfPShufW2B : imm8OfPShufW2A;
629 x86_reg fragmentLength = (d+1<4) ? fragmentLengthB : fragmentLengthA;
630 int maxShift= 3-(d+inc);
634 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
635 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
636 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
637 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
640 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
642 filterCode[fragmentPos + imm8OfPShufW1]=
643 (a+inc) | ((b+inc)<<2) | ((c+inc)<<4) | ((d+inc)<<6);
644 filterCode[fragmentPos + imm8OfPShufW2]=
645 a | (b<<2) | (c<<4) | (d<<6);
647 if (i+4-inc>=dstW) shift=maxShift; //avoid overread
648 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
650 if (shift && i>=shift) {
651 filterCode[fragmentPos + imm8OfPShufW1]+= 0x55*shift;
652 filterCode[fragmentPos + imm8OfPShufW2]+= 0x55*shift;
653 filterPos[i/2]-=shift;
657 fragmentPos+= fragmentLength;
660 filterCode[fragmentPos]= RET;
665 filterPos[((i/2)+1)&(~1)]= xpos>>16; // needed to jump to the next part
667 return fragmentPos + 1;
669 #endif /* HAVE_MMX2 */
671 static void getSubSampleFactors(int *h, int *v, enum PixelFormat format)
673 *h = av_pix_fmt_descriptors[format].log2_chroma_w;
674 *v = av_pix_fmt_descriptors[format].log2_chroma_h;
677 int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],
678 int srcRange, const int table[4], int dstRange,
679 int brightness, int contrast, int saturation)
681 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
682 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
684 c->brightness= brightness;
685 c->contrast = contrast;
686 c->saturation= saturation;
687 c->srcRange = srcRange;
688 c->dstRange = dstRange;
689 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
691 c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->dstFormat]);
692 c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->srcFormat]);
694 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
697 if (HAVE_ALTIVEC && av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)
698 ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness, contrast, saturation);
702 int sws_getColorspaceDetails(struct SwsContext *c, int **inv_table,
703 int *srcRange, int **table, int *dstRange,
704 int *brightness, int *contrast, int *saturation)
706 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
708 *inv_table = c->srcColorspaceTable;
709 *table = c->dstColorspaceTable;
710 *srcRange = c->srcRange;
711 *dstRange = c->dstRange;
712 *brightness= c->brightness;
713 *contrast = c->contrast;
714 *saturation= c->saturation;
719 static int handle_jpeg(enum PixelFormat *format)
722 case PIX_FMT_YUVJ420P: *format = PIX_FMT_YUV420P; return 1;
723 case PIX_FMT_YUVJ422P: *format = PIX_FMT_YUV422P; return 1;
724 case PIX_FMT_YUVJ444P: *format = PIX_FMT_YUV444P; return 1;
725 case PIX_FMT_YUVJ440P: *format = PIX_FMT_YUV440P; return 1;
730 SwsContext *sws_alloc_context(void)
732 SwsContext *c= av_mallocz(sizeof(SwsContext));
734 c->av_class = &sws_context_class;
735 av_opt_set_defaults(c);
740 int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)
743 int usesVFilter, usesHFilter;
745 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
750 int dst_stride = FFALIGN(dstW * sizeof(int16_t) + 16, 16), dst_stride_px = dst_stride >> 1;
751 int flags, cpu_flags;
752 enum PixelFormat srcFormat= c->srcFormat;
753 enum PixelFormat dstFormat= c->dstFormat;
755 cpu_flags = av_get_cpu_flags();
758 if (!rgb15to16) sws_rgb2rgb_init();
760 unscaled = (srcW == dstW && srcH == dstH);
762 if (!sws_isSupportedInput(srcFormat)) {
763 av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n", sws_format_name(srcFormat));
764 return AVERROR(EINVAL);
766 if (!sws_isSupportedOutput(dstFormat)) {
767 av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n", sws_format_name(dstFormat));
768 return AVERROR(EINVAL);
771 i= flags & ( SWS_POINT
782 if(!i || (i & (i-1))) {
783 av_log(c, AV_LOG_ERROR, "Exactly one scaler algorithm must be chosen\n");
784 return AVERROR(EINVAL);
787 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
788 av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",
789 srcW, srcH, dstW, dstH);
790 return AVERROR(EINVAL);
793 if (!dstFilter) dstFilter= &dummyFilter;
794 if (!srcFilter) srcFilter= &dummyFilter;
796 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
797 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
798 c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[dstFormat]);
799 c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[srcFormat]);
800 c->vRounder= 4* 0x0001000100010001ULL;
802 usesVFilter = (srcFilter->lumV && srcFilter->lumV->length>1) ||
803 (srcFilter->chrV && srcFilter->chrV->length>1) ||
804 (dstFilter->lumV && dstFilter->lumV->length>1) ||
805 (dstFilter->chrV && dstFilter->chrV->length>1);
806 usesHFilter = (srcFilter->lumH && srcFilter->lumH->length>1) ||
807 (srcFilter->chrH && srcFilter->chrH->length>1) ||
808 (dstFilter->lumH && dstFilter->lumH->length>1) ||
809 (dstFilter->chrH && dstFilter->chrH->length>1);
811 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
812 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
814 // reuse chroma for 2 pixels RGB/BGR unless user wants full chroma interpolation
815 if (flags & SWS_FULL_CHR_H_INT &&
816 dstFormat != PIX_FMT_RGBA &&
817 dstFormat != PIX_FMT_ARGB &&
818 dstFormat != PIX_FMT_BGRA &&
819 dstFormat != PIX_FMT_ABGR &&
820 dstFormat != PIX_FMT_RGB24 &&
821 dstFormat != PIX_FMT_BGR24) {
822 av_log(c, AV_LOG_ERROR,
823 "full chroma interpolation for destination format '%s' not yet implemented\n",
824 sws_format_name(dstFormat));
825 flags &= ~SWS_FULL_CHR_H_INT;
828 if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
830 // drop some chroma lines if the user wants it
831 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
832 c->chrSrcVSubSample+= c->vChrDrop;
834 // drop every other pixel for chroma calculation unless user wants full chroma
835 if (isAnyRGB(srcFormat) && !(flags&SWS_FULL_CHR_H_INP)
836 && srcFormat!=PIX_FMT_RGB8 && srcFormat!=PIX_FMT_BGR8
837 && srcFormat!=PIX_FMT_RGB4 && srcFormat!=PIX_FMT_BGR4
838 && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE
839 && ((dstW>>c->chrDstHSubSample) <= (srcW>>1) || (flags&SWS_FAST_BILINEAR)))
840 c->chrSrcHSubSample=1;
842 // Note the -((-x)>>y) is so that we always round toward +inf.
843 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
844 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
845 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
846 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
848 /* unscaled special cases */
849 if (unscaled && !usesHFilter && !usesVFilter && (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
850 ff_get_unscaled_swscale(c);
853 if (flags&SWS_PRINT_INFO)
854 av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
855 sws_format_name(srcFormat), sws_format_name(dstFormat));
860 c->srcBpc = 1 + av_pix_fmt_descriptors[srcFormat].comp[0].depth_minus1;
863 c->dstBpc = 1 + av_pix_fmt_descriptors[dstFormat].comp[0].depth_minus1;
868 FF_ALLOC_OR_GOTO(c, c->formatConvBuffer,
869 FFALIGN(srcW, 16) * 2 * FFALIGN(c->srcBpc, 8) >> 3,
871 if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2 && c->srcBpc == 8 && c->dstBpc <= 10) {
872 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
873 if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR)) {
874 if (flags&SWS_PRINT_INFO)
875 av_log(c, AV_LOG_INFO, "output width is not a multiple of 32 -> no MMX2 scaler\n");
877 if (usesHFilter) c->canMMX2BeUsed=0;
882 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
883 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
885 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
886 // but only for the FAST_BILINEAR mode otherwise do correct scaling
887 // n-2 is the last chrominance sample available
888 // this is not perfect, but no one should notice the difference, the more correct variant
889 // would be like the vertical one, but that would require some special code for the
890 // first and last pixel
891 if (flags&SWS_FAST_BILINEAR) {
892 if (c->canMMX2BeUsed) {
896 //we don't use the x86 asm scaler if MMX is available
897 else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
898 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
899 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
903 /* precalculate horizontal scaler filter coefficients */
906 // can't downscale !!!
907 if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {
908 c->lumMmx2FilterCodeSize = initMMX2HScaler( dstW, c->lumXInc, NULL, NULL, NULL, 8);
909 c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc, NULL, NULL, NULL, 4);
912 c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
913 c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
914 #elif HAVE_VIRTUALALLOC
915 c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
916 c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
918 c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);
919 c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);
922 if (!c->lumMmx2FilterCode || !c->chrMmx2FilterCode)
923 return AVERROR(ENOMEM);
924 FF_ALLOCZ_OR_GOTO(c, c->hLumFilter , (dstW /8+8)*sizeof(int16_t), fail);
925 FF_ALLOCZ_OR_GOTO(c, c->hChrFilter , (c->chrDstW /4+8)*sizeof(int16_t), fail);
926 FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW /2/8+8)*sizeof(int32_t), fail);
927 FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW/2/4+8)*sizeof(int32_t), fail);
929 initMMX2HScaler( dstW, c->lumXInc, c->lumMmx2FilterCode, c->hLumFilter, c->hLumFilterPos, 8);
930 initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode, c->hChrFilter, c->hChrFilterPos, 4);
933 mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
934 mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
937 #endif /* HAVE_MMX2 */
939 const int filterAlign=
940 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :
941 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
944 if (initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
945 srcW , dstW, filterAlign, 1<<14,
946 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags, cpu_flags,
947 srcFilter->lumH, dstFilter->lumH, c->param) < 0)
949 if (initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
950 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
951 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, cpu_flags,
952 srcFilter->chrH, dstFilter->chrH, c->param) < 0)
955 } // initialize horizontal stuff
957 /* precalculate vertical scaler filter coefficients */
959 const int filterAlign=
960 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
961 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
964 if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
965 srcH , dstH, filterAlign, (1<<12),
966 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags, cpu_flags,
967 srcFilter->lumV, dstFilter->lumV, c->param) < 0)
969 if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
970 c->chrSrcH, c->chrDstH, filterAlign, (1<<12),
971 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, cpu_flags,
972 srcFilter->chrV, dstFilter->chrV, c->param) < 0)
976 FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof (vector signed short)*c->vLumFilterSize*c->dstH, fail);
977 FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH, fail);
979 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
981 short *p = (short *)&c->vYCoeffsBank[i];
983 p[j] = c->vLumFilter[i];
986 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
988 short *p = (short *)&c->vCCoeffsBank[i];
990 p[j] = c->vChrFilter[i];
995 // calculate buffer sizes so that they won't run out while handling these damn slices
996 c->vLumBufSize= c->vLumFilterSize;
997 c->vChrBufSize= c->vChrFilterSize;
998 for (i=0; i<dstH; i++) {
999 int chrI= i*c->chrDstH / dstH;
1000 int nextSlice= FFMAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
1001 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
1003 nextSlice>>= c->chrSrcVSubSample;
1004 nextSlice<<= c->chrSrcVSubSample;
1005 if (c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
1006 c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
1007 if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
1008 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
1011 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
1012 // allocate several megabytes to handle all possible cases)
1013 FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
1014 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
1015 FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
1016 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
1017 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
1018 //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)
1019 /* align at 16 bytes for AltiVec */
1020 for (i=0; i<c->vLumBufSize; i++) {
1021 FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], dst_stride+16, fail);
1022 c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];
1024 // 64 / (c->dstBpc & ~7) is the same as 16 / sizeof(scaling_intermediate)
1025 c->uv_off_px = dst_stride_px + 64 / (c->dstBpc &~ 7);
1026 c->uv_off_byte = dst_stride + 16;
1027 for (i=0; i<c->vChrBufSize; i++) {
1028 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i+c->vChrBufSize], dst_stride*2+32, fail);
1029 c->chrUPixBuf[i] = c->chrUPixBuf[i+c->vChrBufSize];
1030 c->chrVPixBuf[i] = c->chrVPixBuf[i+c->vChrBufSize] = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;
1032 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
1033 for (i=0; i<c->vLumBufSize; i++) {
1034 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], dst_stride+16, fail);
1035 c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];
1038 //try to avoid drawing green stuff between the right end and the stride end
1039 for (i=0; i<c->vChrBufSize; i++)
1040 memset(c->chrUPixBuf[i], 64, dst_stride*2+1);
1042 assert(c->chrDstH <= dstH);
1044 if (flags&SWS_PRINT_INFO) {
1045 if (flags&SWS_FAST_BILINEAR) av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
1046 else if (flags&SWS_BILINEAR) av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
1047 else if (flags&SWS_BICUBIC) av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
1048 else if (flags&SWS_X) av_log(c, AV_LOG_INFO, "Experimental scaler, ");
1049 else if (flags&SWS_POINT) av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
1050 else if (flags&SWS_AREA) av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
1051 else if (flags&SWS_BICUBLIN) av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
1052 else if (flags&SWS_GAUSS) av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
1053 else if (flags&SWS_SINC) av_log(c, AV_LOG_INFO, "Sinc scaler, ");
1054 else if (flags&SWS_LANCZOS) av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
1055 else if (flags&SWS_SPLINE) av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
1056 else av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
1058 av_log(c, AV_LOG_INFO, "from %s to %s%s ",
1059 sws_format_name(srcFormat),
1061 dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ||
1062 dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||
1063 dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ? "dithered " : "",
1067 sws_format_name(dstFormat));
1069 if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2) av_log(c, AV_LOG_INFO, "using MMX2\n");
1070 else if (HAVE_AMD3DNOW && cpu_flags & AV_CPU_FLAG_3DNOW) av_log(c, AV_LOG_INFO, "using 3DNOW\n");
1071 else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) av_log(c, AV_LOG_INFO, "using MMX\n");
1072 else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) av_log(c, AV_LOG_INFO, "using AltiVec\n");
1073 else av_log(c, AV_LOG_INFO, "using C\n");
1075 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
1076 av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1077 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
1078 av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1079 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
1082 c->swScale= ff_getSwsFunc(c);
1084 fail: //FIXME replace things by appropriate error codes
1088 #if FF_API_SWS_GETCONTEXT
1089 SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,
1090 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
1091 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
1095 if(!(c=sws_alloc_context()))
1103 c->srcRange = handle_jpeg(&srcFormat);
1104 c->dstRange = handle_jpeg(&dstFormat);
1105 c->srcFormat= srcFormat;
1106 c->dstFormat= dstFormat;
1109 c->param[0] = param[0];
1110 c->param[1] = param[1];
1112 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);
1114 if(sws_init_context(c, srcFilter, dstFilter) < 0){
1123 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
1124 float lumaSharpen, float chromaSharpen,
1125 float chromaHShift, float chromaVShift,
1128 SwsFilter *filter= av_malloc(sizeof(SwsFilter));
1132 if (lumaGBlur!=0.0) {
1133 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
1134 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
1136 filter->lumH= sws_getIdentityVec();
1137 filter->lumV= sws_getIdentityVec();
1140 if (chromaGBlur!=0.0) {
1141 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
1142 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
1144 filter->chrH= sws_getIdentityVec();
1145 filter->chrV= sws_getIdentityVec();
1148 if (chromaSharpen!=0.0) {
1149 SwsVector *id= sws_getIdentityVec();
1150 sws_scaleVec(filter->chrH, -chromaSharpen);
1151 sws_scaleVec(filter->chrV, -chromaSharpen);
1152 sws_addVec(filter->chrH, id);
1153 sws_addVec(filter->chrV, id);
1157 if (lumaSharpen!=0.0) {
1158 SwsVector *id= sws_getIdentityVec();
1159 sws_scaleVec(filter->lumH, -lumaSharpen);
1160 sws_scaleVec(filter->lumV, -lumaSharpen);
1161 sws_addVec(filter->lumH, id);
1162 sws_addVec(filter->lumV, id);
1166 if (chromaHShift != 0.0)
1167 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
1169 if (chromaVShift != 0.0)
1170 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
1172 sws_normalizeVec(filter->chrH, 1.0);
1173 sws_normalizeVec(filter->chrV, 1.0);
1174 sws_normalizeVec(filter->lumH, 1.0);
1175 sws_normalizeVec(filter->lumV, 1.0);
1177 if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
1178 if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
1183 SwsVector *sws_allocVec(int length)
1185 SwsVector *vec = av_malloc(sizeof(SwsVector));
1188 vec->length = length;
1189 vec->coeff = av_malloc(sizeof(double) * length);
1195 SwsVector *sws_getGaussianVec(double variance, double quality)
1197 const int length= (int)(variance*quality + 0.5) | 1;
1199 double middle= (length-1)*0.5;
1200 SwsVector *vec= sws_allocVec(length);
1205 for (i=0; i<length; i++) {
1206 double dist= i-middle;
1207 vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);
1210 sws_normalizeVec(vec, 1.0);
1215 SwsVector *sws_getConstVec(double c, int length)
1218 SwsVector *vec= sws_allocVec(length);
1223 for (i=0; i<length; i++)
1229 SwsVector *sws_getIdentityVec(void)
1231 return sws_getConstVec(1.0, 1);
1234 static double sws_dcVec(SwsVector *a)
1239 for (i=0; i<a->length; i++)
1245 void sws_scaleVec(SwsVector *a, double scalar)
1249 for (i=0; i<a->length; i++)
1250 a->coeff[i]*= scalar;
1253 void sws_normalizeVec(SwsVector *a, double height)
1255 sws_scaleVec(a, height/sws_dcVec(a));
1258 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
1260 int length= a->length + b->length - 1;
1262 SwsVector *vec= sws_getConstVec(0.0, length);
1267 for (i=0; i<a->length; i++) {
1268 for (j=0; j<b->length; j++) {
1269 vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];
1276 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
1278 int length= FFMAX(a->length, b->length);
1280 SwsVector *vec= sws_getConstVec(0.0, length);
1285 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1286 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
1291 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
1293 int length= FFMAX(a->length, b->length);
1295 SwsVector *vec= sws_getConstVec(0.0, length);
1300 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1301 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
1306 /* shift left / or right if "shift" is negative */
1307 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
1309 int length= a->length + FFABS(shift)*2;
1311 SwsVector *vec= sws_getConstVec(0.0, length);
1316 for (i=0; i<a->length; i++) {
1317 vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
1323 void sws_shiftVec(SwsVector *a, int shift)
1325 SwsVector *shifted= sws_getShiftedVec(a, shift);
1327 a->coeff= shifted->coeff;
1328 a->length= shifted->length;
1332 void sws_addVec(SwsVector *a, SwsVector *b)
1334 SwsVector *sum= sws_sumVec(a, b);
1336 a->coeff= sum->coeff;
1337 a->length= sum->length;
1341 void sws_subVec(SwsVector *a, SwsVector *b)
1343 SwsVector *diff= sws_diffVec(a, b);
1345 a->coeff= diff->coeff;
1346 a->length= diff->length;
1350 void sws_convVec(SwsVector *a, SwsVector *b)
1352 SwsVector *conv= sws_getConvVec(a, b);
1354 a->coeff= conv->coeff;
1355 a->length= conv->length;
1359 SwsVector *sws_cloneVec(SwsVector *a)
1362 SwsVector *vec= sws_allocVec(a->length);
1367 for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];
1372 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
1379 for (i=0; i<a->length; i++)
1380 if (a->coeff[i]>max) max= a->coeff[i];
1382 for (i=0; i<a->length; i++)
1383 if (a->coeff[i]<min) min= a->coeff[i];
1387 for (i=0; i<a->length; i++) {
1388 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
1389 av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
1390 for (;x>0; x--) av_log(log_ctx, log_level, " ");
1391 av_log(log_ctx, log_level, "|\n");
1395 void sws_freeVec(SwsVector *a)
1398 av_freep(&a->coeff);
1403 void sws_freeFilter(SwsFilter *filter)
1405 if (!filter) return;
1407 if (filter->lumH) sws_freeVec(filter->lumH);
1408 if (filter->lumV) sws_freeVec(filter->lumV);
1409 if (filter->chrH) sws_freeVec(filter->chrH);
1410 if (filter->chrV) sws_freeVec(filter->chrV);
1414 void sws_freeContext(SwsContext *c)
1420 for (i=0; i<c->vLumBufSize; i++)
1421 av_freep(&c->lumPixBuf[i]);
1422 av_freep(&c->lumPixBuf);
1425 if (c->chrUPixBuf) {
1426 for (i=0; i<c->vChrBufSize; i++)
1427 av_freep(&c->chrUPixBuf[i]);
1428 av_freep(&c->chrUPixBuf);
1429 av_freep(&c->chrVPixBuf);
1432 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1433 for (i=0; i<c->vLumBufSize; i++)
1434 av_freep(&c->alpPixBuf[i]);
1435 av_freep(&c->alpPixBuf);
1438 av_freep(&c->vLumFilter);
1439 av_freep(&c->vChrFilter);
1440 av_freep(&c->hLumFilter);
1441 av_freep(&c->hChrFilter);
1443 av_freep(&c->vYCoeffsBank);
1444 av_freep(&c->vCCoeffsBank);
1447 av_freep(&c->vLumFilterPos);
1448 av_freep(&c->vChrFilterPos);
1449 av_freep(&c->hLumFilterPos);
1450 av_freep(&c->hChrFilterPos);
1453 #ifdef MAP_ANONYMOUS
1454 if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);
1455 if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);
1456 #elif HAVE_VIRTUALALLOC
1457 if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);
1458 if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);
1460 av_free(c->lumMmx2FilterCode);
1461 av_free(c->chrMmx2FilterCode);
1463 c->lumMmx2FilterCode=NULL;
1464 c->chrMmx2FilterCode=NULL;
1465 #endif /* HAVE_MMX */
1467 av_freep(&c->yuvTable);
1468 av_free(c->formatConvBuffer);
1473 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
1474 int srcW, int srcH, enum PixelFormat srcFormat,
1475 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
1476 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
1478 static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};
1481 param = default_param;
1484 (context->srcW != srcW ||
1485 context->srcH != srcH ||
1486 context->srcFormat != srcFormat ||
1487 context->dstW != dstW ||
1488 context->dstH != dstH ||
1489 context->dstFormat != dstFormat ||
1490 context->flags != flags ||
1491 context->param[0] != param[0] ||
1492 context->param[1] != param[1])) {
1493 sws_freeContext(context);
1498 if (!(context = sws_alloc_context()))
1500 context->srcW = srcW;
1501 context->srcH = srcH;
1502 context->srcRange = handle_jpeg(&srcFormat);
1503 context->srcFormat = srcFormat;
1504 context->dstW = dstW;
1505 context->dstH = dstH;
1506 context->dstRange = handle_jpeg(&dstFormat);
1507 context->dstFormat = dstFormat;
1508 context->flags = flags;
1509 context->param[0] = param[0];
1510 context->param[1] = param[1];
1511 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);
1512 if (sws_init_context(context, srcFilter, dstFilter) < 0) {
1513 sws_freeContext(context);