2 * Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
4 * This file is part of FFmpeg.
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 #define _SVID_SOURCE //needed for MAP_ANONYMOUS
22 #define _DARWIN_C_SOURCE // needed for MAP_ANON
31 #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
32 #define MAP_ANONYMOUS MAP_ANON
36 #define WIN32_LEAN_AND_MEAN
40 #include "swscale_internal.h"
42 #include "libavutil/intreadwrite.h"
43 #include "libavutil/x86_cpu.h"
44 #include "libavutil/cpu.h"
45 #include "libavutil/avutil.h"
46 #include "libavutil/bswap.h"
47 #include "libavutil/opt.h"
48 #include "libavutil/pixdesc.h"
49 #include "libavutil/avassert.h"
51 unsigned swscale_version(void)
53 return LIBSWSCALE_VERSION_INT;
56 const char *swscale_configuration(void)
58 return FFMPEG_CONFIGURATION;
61 const char *swscale_license(void)
63 #define LICENSE_PREFIX "libswscale license: "
64 return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
67 #define RET 0xC3 //near return opcode for x86
69 typedef struct FormatEntry {
70 int is_supported_in, is_supported_out;
73 const static FormatEntry format_entries[PIX_FMT_NB] = {
74 [PIX_FMT_YUV420P] = { 1 , 1 },
75 [PIX_FMT_YUYV422] = { 1 , 1 },
76 [PIX_FMT_RGB24] = { 1 , 1 },
77 [PIX_FMT_BGR24] = { 1 , 1 },
78 [PIX_FMT_YUV422P] = { 1 , 1 },
79 [PIX_FMT_YUV444P] = { 1 , 1 },
80 [PIX_FMT_YUV410P] = { 1 , 1 },
81 [PIX_FMT_YUV411P] = { 1 , 1 },
82 [PIX_FMT_GRAY8] = { 1 , 1 },
83 [PIX_FMT_MONOWHITE] = { 1 , 1 },
84 [PIX_FMT_MONOBLACK] = { 1 , 1 },
85 [PIX_FMT_PAL8] = { 1 , 0 },
86 [PIX_FMT_YUVJ420P] = { 1 , 1 },
87 [PIX_FMT_YUVJ422P] = { 1 , 1 },
88 [PIX_FMT_YUVJ444P] = { 1 , 1 },
89 [PIX_FMT_UYVY422] = { 1 , 1 },
90 [PIX_FMT_UYYVYY411] = { 0 , 0 },
91 [PIX_FMT_BGR8] = { 1 , 1 },
92 [PIX_FMT_BGR4] = { 0 , 1 },
93 [PIX_FMT_BGR4_BYTE] = { 1 , 1 },
94 [PIX_FMT_RGB8] = { 1 , 1 },
95 [PIX_FMT_RGB4] = { 0 , 1 },
96 [PIX_FMT_RGB4_BYTE] = { 1 , 1 },
97 [PIX_FMT_NV12] = { 1 , 1 },
98 [PIX_FMT_NV21] = { 1 , 1 },
99 [PIX_FMT_ARGB] = { 1 , 1 },
100 [PIX_FMT_RGBA] = { 1 , 1 },
101 [PIX_FMT_ABGR] = { 1 , 1 },
102 [PIX_FMT_BGRA] = { 1 , 1 },
103 [PIX_FMT_GRAY16BE] = { 1 , 1 },
104 [PIX_FMT_GRAY16LE] = { 1 , 1 },
105 [PIX_FMT_YUV440P] = { 1 , 1 },
106 [PIX_FMT_YUVJ440P] = { 1 , 1 },
107 [PIX_FMT_YUVA420P] = { 1 , 1 },
108 [PIX_FMT_RGB48BE] = { 1 , 1 },
109 [PIX_FMT_RGB48LE] = { 1 , 1 },
110 [PIX_FMT_RGB565BE] = { 1 , 1 },
111 [PIX_FMT_RGB565LE] = { 1 , 1 },
112 [PIX_FMT_RGB555BE] = { 1 , 1 },
113 [PIX_FMT_RGB555LE] = { 1 , 1 },
114 [PIX_FMT_BGR565BE] = { 1 , 1 },
115 [PIX_FMT_BGR565LE] = { 1 , 1 },
116 [PIX_FMT_BGR555BE] = { 1 , 1 },
117 [PIX_FMT_BGR555LE] = { 1 , 1 },
118 [PIX_FMT_YUV420P16LE] = { 1 , 1 },
119 [PIX_FMT_YUV420P16BE] = { 1 , 1 },
120 [PIX_FMT_YUV422P16LE] = { 1 , 1 },
121 [PIX_FMT_YUV422P16BE] = { 1 , 1 },
122 [PIX_FMT_YUV444P16LE] = { 1 , 1 },
123 [PIX_FMT_YUV444P16BE] = { 1 , 1 },
124 [PIX_FMT_RGB444LE] = { 0 , 1 },
125 [PIX_FMT_RGB444BE] = { 0 , 1 },
126 [PIX_FMT_BGR444LE] = { 0 , 1 },
127 [PIX_FMT_BGR444BE] = { 0 , 1 },
128 [PIX_FMT_Y400A] = { 1 , 0 },
129 [PIX_FMT_BGR48BE] = { 1 , 1 },
130 [PIX_FMT_BGR48LE] = { 1 , 1 },
131 [PIX_FMT_YUV420P9BE] = { 1 , 1 },
132 [PIX_FMT_YUV420P9LE] = { 1 , 1 },
133 [PIX_FMT_YUV420P10BE] = { 1 , 1 },
134 [PIX_FMT_YUV420P10LE] = { 1 , 1 },
135 [PIX_FMT_YUV422P10BE] = { 1 , 1 },
136 [PIX_FMT_YUV422P10LE] = { 1 , 1 },
137 [PIX_FMT_YUV444P9BE] = { 1 , 0 },
138 [PIX_FMT_YUV444P9LE] = { 1 , 0 },
139 [PIX_FMT_YUV444P10BE] = { 1 , 0 },
140 [PIX_FMT_YUV444P10LE] = { 1 , 0 },
143 int sws_isSupportedInput(enum PixelFormat pix_fmt)
145 return (unsigned)pix_fmt < PIX_FMT_NB ?
146 format_entries[pix_fmt].is_supported_in : 0;
149 int sws_isSupportedOutput(enum PixelFormat pix_fmt)
151 return (unsigned)pix_fmt < PIX_FMT_NB ?
152 format_entries[pix_fmt].is_supported_out : 0;
155 extern const int32_t ff_yuv2rgb_coeffs[8][4];
157 #if FF_API_SWS_FORMAT_NAME
158 const char *sws_format_name(enum PixelFormat format)
160 return av_get_pix_fmt_name(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 +1 is for the MMX scaler which reads over the end
190 FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW+1)*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);
278 int64_t dd = ( d*d)>>30;
279 int64_t ddd= (dd*d)>>30;
282 coeff = (12*(1<<24)-9*B-6*C)*ddd + (-18*(1<<24)+12*B+6*C)*dd + (6*(1<<24)-2*B)*(1<<30);
283 else if (d < 1LL<<31)
284 coeff = (-B-6*C)*ddd + (6*B+30*C)*dd + (-12*B-48*C)*d + (8*B+24*C)*(1<<30);
287 coeff *= fone>>(30+24);
289 /* else if (flags & SWS_X) {
290 double p= param ? param*0.01 : 0.3;
291 coeff = d ? sin(d*M_PI)/(d*M_PI) : 1.0;
292 coeff*= pow(2.0, - p*d*d);
294 else if (flags & SWS_X) {
295 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
299 c = cos(floatd*M_PI);
302 if (c<0.0) c= -pow(-c, A);
304 coeff= (c*0.5 + 0.5)*fone;
305 } else if (flags & SWS_AREA) {
306 int64_t d2= d - (1<<29);
307 if (d2*xInc < -(1LL<<(29+16))) coeff= 1.0 * (1LL<<(30+16));
308 else if (d2*xInc < (1LL<<(29+16))) coeff= -d2*xInc + (1LL<<(29+16));
310 coeff *= fone>>(30+16);
311 } else if (flags & SWS_GAUSS) {
312 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
313 coeff = (pow(2.0, - p*floatd*floatd))*fone;
314 } else if (flags & SWS_SINC) {
315 coeff = (d ? sin(floatd*M_PI)/(floatd*M_PI) : 1.0)*fone;
316 } else if (flags & SWS_LANCZOS) {
317 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
318 coeff = (d ? sin(floatd*M_PI)*sin(floatd*M_PI/p)/(floatd*floatd*M_PI*M_PI/p) : 1.0)*fone;
319 if (floatd>p) coeff=0;
320 } else if (flags & SWS_BILINEAR) {
322 if (coeff<0) coeff=0;
324 } else if (flags & SWS_SPLINE) {
325 double p=-2.196152422706632;
326 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, floatd) * fone;
328 coeff= 0.0; //GCC warning killer
332 filter[i*filterSize + j]= coeff;
339 /* apply src & dst Filter to filter -> filter2
342 assert(filterSize>0);
343 filter2Size= filterSize;
344 if (srcFilter) filter2Size+= srcFilter->length - 1;
345 if (dstFilter) filter2Size+= dstFilter->length - 1;
346 assert(filter2Size>0);
347 FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size*dstW*sizeof(*filter2), fail);
349 for (i=0; i<dstW; i++) {
353 for (k=0; k<srcFilter->length; k++) {
354 for (j=0; j<filterSize; j++)
355 filter2[i*filter2Size + k + j] += srcFilter->coeff[k]*filter[i*filterSize + j];
358 for (j=0; j<filterSize; j++)
359 filter2[i*filter2Size + j]= filter[i*filterSize + j];
363 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
367 /* try to reduce the filter-size (step1 find size and shift left) */
368 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
370 for (i=dstW-1; i>=0; i--) {
371 int min= filter2Size;
375 /* get rid of near zero elements on the left by shifting left */
376 for (j=0; j<filter2Size; j++) {
378 cutOff += FFABS(filter2[i*filter2Size]);
380 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
382 /* preserve monotonicity because the core can't handle the filter otherwise */
383 if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
385 // move filter coefficients left
386 for (k=1; k<filter2Size; k++)
387 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
388 filter2[i*filter2Size + k - 1]= 0;
393 /* count near zeros on the right */
394 for (j=filter2Size-1; j>0; j--) {
395 cutOff += FFABS(filter2[i*filter2Size + j]);
397 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
401 if (min>minFilterSize) minFilterSize= min;
404 if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) {
405 // we can handle the special case 4,
406 // so we don't want to go to the full 8
407 if (minFilterSize < 5)
410 // We really don't want to waste our time
411 // doing useless computation, so fall back on
412 // the scalar C code for very small filters.
413 // Vectorizing is worth it only if you have a
414 // decent-sized vector.
415 if (minFilterSize < 3)
419 if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
420 // special case for unscaled vertical filtering
421 if (minFilterSize == 1 && filterAlign == 2)
425 assert(minFilterSize > 0);
426 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
427 assert(filterSize > 0);
428 filter= av_malloc(filterSize*dstW*sizeof(*filter));
429 if (filterSize >= MAX_FILTER_SIZE*16/((flags&SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
431 *outFilterSize= filterSize;
433 if (flags&SWS_PRINT_INFO)
434 av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
435 /* try to reduce the filter-size (step2 reduce it) */
436 for (i=0; i<dstW; i++) {
439 for (j=0; j<filterSize; j++) {
440 if (j>=filter2Size) filter[i*filterSize + j]= 0;
441 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
442 if((flags & SWS_BITEXACT) && j>=minFilterSize)
443 filter[i*filterSize + j]= 0;
447 //FIXME try to align filterPos if possible
450 for (i=0; i<dstW; i++) {
452 if ((*filterPos)[i] < 0) {
453 // move filter coefficients left to compensate for filterPos
454 for (j=1; j<filterSize; j++) {
455 int left= FFMAX(j + (*filterPos)[i], 0);
456 filter[i*filterSize + left] += filter[i*filterSize + j];
457 filter[i*filterSize + j]=0;
462 if ((*filterPos)[i] + filterSize > srcW) {
463 int shift= (*filterPos)[i] + filterSize - srcW;
464 // move filter coefficients right to compensate for filterPos
465 for (j=filterSize-2; j>=0; j--) {
466 int right= FFMIN(j + shift, filterSize-1);
467 filter[i*filterSize +right] += filter[i*filterSize +j];
468 filter[i*filterSize +j]=0;
470 (*filterPos)[i]= srcW - filterSize;
474 // Note the +1 is for the MMX scaler which reads over the end
475 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
476 FF_ALLOCZ_OR_GOTO(NULL, *outFilter, *outFilterSize*(dstW+1)*sizeof(int16_t), fail);
478 /* normalize & store in outFilter */
479 for (i=0; i<dstW; i++) {
484 for (j=0; j<filterSize; j++) {
485 sum+= filter[i*filterSize + j];
487 sum= (sum + one/2)/ one;
488 for (j=0; j<*outFilterSize; j++) {
489 int64_t v= filter[i*filterSize + j] + error;
490 int intV= ROUNDED_DIV(v, sum);
491 (*outFilter)[i*(*outFilterSize) + j]= intV;
496 (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
497 for (i=0; i<*outFilterSize; i++) {
498 int j= dstW*(*outFilterSize);
499 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
510 static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode, int16_t *filter, int32_t *filterPos, int numSplits)
513 x86_reg imm8OfPShufW1A;
514 x86_reg imm8OfPShufW2A;
515 x86_reg fragmentLengthA;
517 x86_reg imm8OfPShufW1B;
518 x86_reg imm8OfPShufW2B;
519 x86_reg fragmentLengthB;
524 // create an optimized horizontal scaling routine
525 /* This scaler is made of runtime-generated MMX2 code using specially
526 * tuned pshufw instructions. For every four output pixels, if four
527 * input pixels are enough for the fast bilinear scaling, then a chunk
528 * of fragmentB is used. If five input pixels are needed, then a chunk
529 * of fragmentA is used.
538 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
539 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
540 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
541 "punpcklbw %%mm7, %%mm1 \n\t"
542 "punpcklbw %%mm7, %%mm0 \n\t"
543 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
545 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
547 "psubw %%mm1, %%mm0 \n\t"
548 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
549 "pmullw %%mm3, %%mm0 \n\t"
550 "psllw $7, %%mm1 \n\t"
551 "paddw %%mm1, %%mm0 \n\t"
553 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
555 "add $8, %%"REG_a" \n\t"
559 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
560 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
561 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
566 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
570 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
571 "=r" (fragmentLengthA)
578 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
579 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
580 "punpcklbw %%mm7, %%mm0 \n\t"
581 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
583 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
585 "psubw %%mm1, %%mm0 \n\t"
586 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
587 "pmullw %%mm3, %%mm0 \n\t"
588 "psllw $7, %%mm1 \n\t"
589 "paddw %%mm1, %%mm0 \n\t"
591 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
593 "add $8, %%"REG_a" \n\t"
597 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
598 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
599 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
604 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
608 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
609 "=r" (fragmentLengthB)
612 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
615 for (i=0; i<dstW/numSplits; i++) {
620 int b=((xpos+xInc)>>16) - xx;
621 int c=((xpos+xInc*2)>>16) - xx;
622 int d=((xpos+xInc*3)>>16) - xx;
624 uint8_t *fragment = (d+1<4) ? fragmentB : fragmentA;
625 x86_reg imm8OfPShufW1 = (d+1<4) ? imm8OfPShufW1B : imm8OfPShufW1A;
626 x86_reg imm8OfPShufW2 = (d+1<4) ? imm8OfPShufW2B : imm8OfPShufW2A;
627 x86_reg fragmentLength = (d+1<4) ? fragmentLengthB : fragmentLengthA;
628 int maxShift= 3-(d+inc);
632 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
633 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
634 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
635 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
638 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
640 filterCode[fragmentPos + imm8OfPShufW1]=
641 (a+inc) | ((b+inc)<<2) | ((c+inc)<<4) | ((d+inc)<<6);
642 filterCode[fragmentPos + imm8OfPShufW2]=
643 a | (b<<2) | (c<<4) | (d<<6);
645 if (i+4-inc>=dstW) shift=maxShift; //avoid overread
646 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
648 if (shift && i>=shift) {
649 filterCode[fragmentPos + imm8OfPShufW1]+= 0x55*shift;
650 filterCode[fragmentPos + imm8OfPShufW2]+= 0x55*shift;
651 filterPos[i/2]-=shift;
655 fragmentPos+= fragmentLength;
658 filterCode[fragmentPos]= RET;
663 filterPos[((i/2)+1)&(~1)]= xpos>>16; // needed to jump to the next part
665 return fragmentPos + 1;
667 #endif /* HAVE_MMX2 */
669 static void getSubSampleFactors(int *h, int *v, enum PixelFormat format)
671 *h = av_pix_fmt_descriptors[format].log2_chroma_w;
672 *v = av_pix_fmt_descriptors[format].log2_chroma_h;
675 int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],
676 int srcRange, const int table[4], int dstRange,
677 int brightness, int contrast, int saturation)
679 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
680 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
682 c->brightness= brightness;
683 c->contrast = contrast;
684 c->saturation= saturation;
685 c->srcRange = srcRange;
686 c->dstRange = dstRange;
687 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
689 c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->dstFormat]);
690 c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->srcFormat]);
692 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
695 if (HAVE_ALTIVEC && av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)
696 ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness, contrast, saturation);
700 int sws_getColorspaceDetails(struct SwsContext *c, int **inv_table,
701 int *srcRange, int **table, int *dstRange,
702 int *brightness, int *contrast, int *saturation)
704 if (!c || isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
706 *inv_table = c->srcColorspaceTable;
707 *table = c->dstColorspaceTable;
708 *srcRange = c->srcRange;
709 *dstRange = c->dstRange;
710 *brightness= c->brightness;
711 *contrast = c->contrast;
712 *saturation= c->saturation;
717 static int handle_jpeg(enum PixelFormat *format)
720 case PIX_FMT_YUVJ420P: *format = PIX_FMT_YUV420P; return 1;
721 case PIX_FMT_YUVJ422P: *format = PIX_FMT_YUV422P; return 1;
722 case PIX_FMT_YUVJ444P: *format = PIX_FMT_YUV444P; return 1;
723 case PIX_FMT_YUVJ440P: *format = PIX_FMT_YUV440P; return 1;
728 SwsContext *sws_alloc_context(void)
730 SwsContext *c= av_mallocz(sizeof(SwsContext));
732 c->av_class = &sws_context_class;
733 av_opt_set_defaults(c);
738 int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)
741 int usesVFilter, usesHFilter;
743 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
748 int dst_stride = FFALIGN(dstW * sizeof(int16_t)+66, 16);
749 int flags, cpu_flags;
750 enum PixelFormat srcFormat= c->srcFormat;
751 enum PixelFormat dstFormat= c->dstFormat;
753 cpu_flags = av_get_cpu_flags();
756 if (!rgb15to16) sws_rgb2rgb_init();
758 unscaled = (srcW == dstW && srcH == dstH);
760 if (!sws_isSupportedInput(srcFormat)) {
761 av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n", av_get_pix_fmt_name(srcFormat));
762 return AVERROR(EINVAL);
764 if (!sws_isSupportedOutput(dstFormat)) {
765 av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n", av_get_pix_fmt_name(dstFormat));
766 return AVERROR(EINVAL);
769 i= flags & ( SWS_POINT
780 if(!i || (i & (i-1))) {
781 av_log(c, AV_LOG_ERROR, "Exactly one scaler algorithm must be chosen\n");
782 return AVERROR(EINVAL);
785 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
786 av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",
787 srcW, srcH, dstW, dstH);
788 return AVERROR(EINVAL);
791 if (!dstFilter) dstFilter= &dummyFilter;
792 if (!srcFilter) srcFilter= &dummyFilter;
794 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
795 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
796 c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[dstFormat]);
797 c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[srcFormat]);
798 c->vRounder= 4* 0x0001000100010001ULL;
800 usesVFilter = (srcFilter->lumV && srcFilter->lumV->length>1) ||
801 (srcFilter->chrV && srcFilter->chrV->length>1) ||
802 (dstFilter->lumV && dstFilter->lumV->length>1) ||
803 (dstFilter->chrV && dstFilter->chrV->length>1);
804 usesHFilter = (srcFilter->lumH && srcFilter->lumH->length>1) ||
805 (srcFilter->chrH && srcFilter->chrH->length>1) ||
806 (dstFilter->lumH && dstFilter->lumH->length>1) ||
807 (dstFilter->chrH && dstFilter->chrH->length>1);
809 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
810 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
812 // reuse chroma for 2 pixels RGB/BGR unless user wants full chroma interpolation
813 if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
815 // drop some chroma lines if the user wants it
816 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
817 c->chrSrcVSubSample+= c->vChrDrop;
819 // drop every other pixel for chroma calculation unless user wants full chroma
820 if (isAnyRGB(srcFormat) && !(flags&SWS_FULL_CHR_H_INP)
821 && srcFormat!=PIX_FMT_RGB8 && srcFormat!=PIX_FMT_BGR8
822 && srcFormat!=PIX_FMT_RGB4 && srcFormat!=PIX_FMT_BGR4
823 && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE
824 && ((dstW>>c->chrDstHSubSample) <= (srcW>>1) || (flags&SWS_FAST_BILINEAR)))
825 c->chrSrcHSubSample=1;
827 // Note the -((-x)>>y) is so that we always round toward +inf.
828 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
829 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
830 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
831 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
833 /* unscaled special cases */
834 if (unscaled && !usesHFilter && !usesVFilter && (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
835 ff_get_unscaled_swscale(c);
838 if (flags&SWS_PRINT_INFO)
839 av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
840 av_get_pix_fmt_name(srcFormat), av_get_pix_fmt_name(dstFormat));
845 c->scalingBpp = FFMAX(av_pix_fmt_descriptors[srcFormat].comp[0].depth_minus1,
846 av_pix_fmt_descriptors[dstFormat].comp[0].depth_minus1) >= 15 ? 16 : 8;
848 if (c->scalingBpp == 16)
850 av_assert0(c->scalingBpp<=16);
851 FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW*2+78, 16) * 2, fail);
852 if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2 && c->scalingBpp == 8) {
853 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
854 if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR)) {
855 if (flags&SWS_PRINT_INFO)
856 av_log(c, AV_LOG_INFO, "output width is not a multiple of 32 -> no MMX2 scaler\n");
858 if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat)) c->canMMX2BeUsed=0;
863 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
864 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
866 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
867 // but only for the FAST_BILINEAR mode otherwise do correct scaling
868 // n-2 is the last chrominance sample available
869 // this is not perfect, but no one should notice the difference, the more correct variant
870 // would be like the vertical one, but that would require some special code for the
871 // first and last pixel
872 if (flags&SWS_FAST_BILINEAR) {
873 if (c->canMMX2BeUsed) {
877 //we don't use the x86 asm scaler if MMX is available
878 else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX && c->scalingBpp == 8) {
879 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
880 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
884 /* precalculate horizontal scaler filter coefficients */
887 // can't downscale !!!
888 if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {
889 c->lumMmx2FilterCodeSize = initMMX2HScaler( dstW, c->lumXInc, NULL, NULL, NULL, 8);
890 c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc, NULL, NULL, NULL, 4);
893 c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
894 c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
895 #elif HAVE_VIRTUALALLOC
896 c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
897 c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
899 c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);
900 c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);
904 if (c->lumMmx2FilterCode == MAP_FAILED || c->chrMmx2FilterCode == MAP_FAILED)
906 if (!c->lumMmx2FilterCode || !c->chrMmx2FilterCode)
908 return AVERROR(ENOMEM);
909 FF_ALLOCZ_OR_GOTO(c, c->hLumFilter , (dstW /8+8)*sizeof(int16_t), fail);
910 FF_ALLOCZ_OR_GOTO(c, c->hChrFilter , (c->chrDstW /4+8)*sizeof(int16_t), fail);
911 FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW /2/8+8)*sizeof(int32_t), fail);
912 FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW/2/4+8)*sizeof(int32_t), fail);
914 initMMX2HScaler( dstW, c->lumXInc, c->lumMmx2FilterCode, c->hLumFilter, c->hLumFilterPos, 8);
915 initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode, c->hChrFilter, c->hChrFilterPos, 4);
918 mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
919 mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
922 #endif /* HAVE_MMX2 */
924 const int filterAlign=
925 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :
926 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
929 if (initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
930 srcW , dstW, filterAlign, 1<<14,
931 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags, cpu_flags,
932 srcFilter->lumH, dstFilter->lumH, c->param) < 0)
934 if (initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
935 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
936 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, cpu_flags,
937 srcFilter->chrH, dstFilter->chrH, c->param) < 0)
940 } // initialize horizontal stuff
942 /* precalculate vertical scaler filter coefficients */
944 const int filterAlign=
945 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
946 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
949 if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
950 srcH , dstH, filterAlign, (1<<12),
951 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags, cpu_flags,
952 srcFilter->lumV, dstFilter->lumV, c->param) < 0)
954 if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
955 c->chrSrcH, c->chrDstH, filterAlign, (1<<12),
956 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, cpu_flags,
957 srcFilter->chrV, dstFilter->chrV, c->param) < 0)
961 FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof (vector signed short)*c->vLumFilterSize*c->dstH, fail);
962 FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH, fail);
964 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
966 short *p = (short *)&c->vYCoeffsBank[i];
968 p[j] = c->vLumFilter[i];
971 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
973 short *p = (short *)&c->vCCoeffsBank[i];
975 p[j] = c->vChrFilter[i];
980 // calculate buffer sizes so that they won't run out while handling these damn slices
981 c->vLumBufSize= c->vLumFilterSize;
982 c->vChrBufSize= c->vChrFilterSize;
983 for (i=0; i<dstH; i++) {
984 int chrI= (int64_t)i*c->chrDstH / dstH;
985 int nextSlice= FFMAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
986 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
988 nextSlice>>= c->chrSrcVSubSample;
989 nextSlice<<= c->chrSrcVSubSample;
990 if (c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
991 c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
992 if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
993 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
996 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
997 // allocate several megabytes to handle all possible cases)
998 FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
999 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
1000 FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
1001 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
1002 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
1003 //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)
1004 /* align at 16 bytes for AltiVec */
1005 for (i=0; i<c->vLumBufSize; i++) {
1006 FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], dst_stride+16, fail);
1007 c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];
1009 // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)
1010 c->uv_off = (dst_stride>>1) + 64 / c->scalingBpp;
1011 c->uv_offx2 = dst_stride + 16;
1012 for (i=0; i<c->vChrBufSize; i++) {
1013 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i+c->vChrBufSize], dst_stride*2+32, fail);
1014 c->chrUPixBuf[i] = c->chrUPixBuf[i+c->vChrBufSize];
1015 c->chrVPixBuf[i] = c->chrVPixBuf[i+c->vChrBufSize] = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;
1017 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
1018 for (i=0; i<c->vLumBufSize; i++) {
1019 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], dst_stride+16, fail);
1020 c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];
1023 //try to avoid drawing green stuff between the right end and the stride end
1024 for (i=0; i<c->vChrBufSize; i++)
1025 if(av_pix_fmt_descriptors[c->dstFormat].comp[0].depth_minus1 == 15){
1026 av_assert0(c->scalingBpp == 16);
1027 for(j=0; j<dst_stride/2+1; j++)
1028 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;
1030 for(j=0; j<dst_stride+1; j++)
1031 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;
1033 assert(c->chrDstH <= dstH);
1035 if (flags&SWS_PRINT_INFO) {
1036 if (flags&SWS_FAST_BILINEAR) av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
1037 else if (flags&SWS_BILINEAR) av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
1038 else if (flags&SWS_BICUBIC) av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
1039 else if (flags&SWS_X) av_log(c, AV_LOG_INFO, "Experimental scaler, ");
1040 else if (flags&SWS_POINT) av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
1041 else if (flags&SWS_AREA) av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
1042 else if (flags&SWS_BICUBLIN) av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
1043 else if (flags&SWS_GAUSS) av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
1044 else if (flags&SWS_SINC) av_log(c, AV_LOG_INFO, "Sinc scaler, ");
1045 else if (flags&SWS_LANCZOS) av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
1046 else if (flags&SWS_SPLINE) av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
1047 else av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
1049 av_log(c, AV_LOG_INFO, "from %s to %s%s ",
1050 av_get_pix_fmt_name(srcFormat),
1052 dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ||
1053 dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||
1054 dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ? "dithered " : "",
1058 av_get_pix_fmt_name(dstFormat));
1060 if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2) av_log(c, AV_LOG_INFO, "using MMX2\n");
1061 else if (HAVE_AMD3DNOW && cpu_flags & AV_CPU_FLAG_3DNOW) av_log(c, AV_LOG_INFO, "using 3DNOW\n");
1062 else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) av_log(c, AV_LOG_INFO, "using MMX\n");
1063 else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) av_log(c, AV_LOG_INFO, "using AltiVec\n");
1064 else av_log(c, AV_LOG_INFO, "using C\n");
1066 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
1067 av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1068 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
1069 av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1070 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
1073 c->swScale= ff_getSwsFunc(c);
1075 fail: //FIXME replace things by appropriate error codes
1079 #if FF_API_SWS_GETCONTEXT
1080 SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,
1081 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
1082 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
1086 if(!(c=sws_alloc_context()))
1094 c->srcRange = handle_jpeg(&srcFormat);
1095 c->dstRange = handle_jpeg(&dstFormat);
1096 c->srcFormat= srcFormat;
1097 c->dstFormat= dstFormat;
1100 c->param[0] = param[0];
1101 c->param[1] = param[1];
1103 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);
1105 if(sws_init_context(c, srcFilter, dstFilter) < 0){
1114 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
1115 float lumaSharpen, float chromaSharpen,
1116 float chromaHShift, float chromaVShift,
1119 SwsFilter *filter= av_malloc(sizeof(SwsFilter));
1123 if (lumaGBlur!=0.0) {
1124 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
1125 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
1127 filter->lumH= sws_getIdentityVec();
1128 filter->lumV= sws_getIdentityVec();
1131 if (chromaGBlur!=0.0) {
1132 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
1133 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
1135 filter->chrH= sws_getIdentityVec();
1136 filter->chrV= sws_getIdentityVec();
1139 if (chromaSharpen!=0.0) {
1140 SwsVector *id= sws_getIdentityVec();
1141 sws_scaleVec(filter->chrH, -chromaSharpen);
1142 sws_scaleVec(filter->chrV, -chromaSharpen);
1143 sws_addVec(filter->chrH, id);
1144 sws_addVec(filter->chrV, id);
1148 if (lumaSharpen!=0.0) {
1149 SwsVector *id= sws_getIdentityVec();
1150 sws_scaleVec(filter->lumH, -lumaSharpen);
1151 sws_scaleVec(filter->lumV, -lumaSharpen);
1152 sws_addVec(filter->lumH, id);
1153 sws_addVec(filter->lumV, id);
1157 if (chromaHShift != 0.0)
1158 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
1160 if (chromaVShift != 0.0)
1161 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
1163 sws_normalizeVec(filter->chrH, 1.0);
1164 sws_normalizeVec(filter->chrV, 1.0);
1165 sws_normalizeVec(filter->lumH, 1.0);
1166 sws_normalizeVec(filter->lumV, 1.0);
1168 if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
1169 if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
1174 SwsVector *sws_allocVec(int length)
1176 SwsVector *vec = av_malloc(sizeof(SwsVector));
1179 vec->length = length;
1180 vec->coeff = av_malloc(sizeof(double) * length);
1186 SwsVector *sws_getGaussianVec(double variance, double quality)
1188 const int length= (int)(variance*quality + 0.5) | 1;
1190 double middle= (length-1)*0.5;
1191 SwsVector *vec= sws_allocVec(length);
1196 for (i=0; i<length; i++) {
1197 double dist= i-middle;
1198 vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);
1201 sws_normalizeVec(vec, 1.0);
1206 SwsVector *sws_getConstVec(double c, int length)
1209 SwsVector *vec= sws_allocVec(length);
1214 for (i=0; i<length; i++)
1220 SwsVector *sws_getIdentityVec(void)
1222 return sws_getConstVec(1.0, 1);
1225 static double sws_dcVec(SwsVector *a)
1230 for (i=0; i<a->length; i++)
1236 void sws_scaleVec(SwsVector *a, double scalar)
1240 for (i=0; i<a->length; i++)
1241 a->coeff[i]*= scalar;
1244 void sws_normalizeVec(SwsVector *a, double height)
1246 sws_scaleVec(a, height/sws_dcVec(a));
1249 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
1251 int length= a->length + b->length - 1;
1253 SwsVector *vec= sws_getConstVec(0.0, length);
1258 for (i=0; i<a->length; i++) {
1259 for (j=0; j<b->length; j++) {
1260 vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];
1267 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
1269 int length= FFMAX(a->length, b->length);
1271 SwsVector *vec= sws_getConstVec(0.0, length);
1276 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1277 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
1282 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
1284 int length= FFMAX(a->length, b->length);
1286 SwsVector *vec= sws_getConstVec(0.0, length);
1291 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1292 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
1297 /* shift left / or right if "shift" is negative */
1298 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
1300 int length= a->length + FFABS(shift)*2;
1302 SwsVector *vec= sws_getConstVec(0.0, length);
1307 for (i=0; i<a->length; i++) {
1308 vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
1314 void sws_shiftVec(SwsVector *a, int shift)
1316 SwsVector *shifted= sws_getShiftedVec(a, shift);
1318 a->coeff= shifted->coeff;
1319 a->length= shifted->length;
1323 void sws_addVec(SwsVector *a, SwsVector *b)
1325 SwsVector *sum= sws_sumVec(a, b);
1327 a->coeff= sum->coeff;
1328 a->length= sum->length;
1332 void sws_subVec(SwsVector *a, SwsVector *b)
1334 SwsVector *diff= sws_diffVec(a, b);
1336 a->coeff= diff->coeff;
1337 a->length= diff->length;
1341 void sws_convVec(SwsVector *a, SwsVector *b)
1343 SwsVector *conv= sws_getConvVec(a, b);
1345 a->coeff= conv->coeff;
1346 a->length= conv->length;
1350 SwsVector *sws_cloneVec(SwsVector *a)
1353 SwsVector *vec= sws_allocVec(a->length);
1358 for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];
1363 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
1370 for (i=0; i<a->length; i++)
1371 if (a->coeff[i]>max) max= a->coeff[i];
1373 for (i=0; i<a->length; i++)
1374 if (a->coeff[i]<min) min= a->coeff[i];
1378 for (i=0; i<a->length; i++) {
1379 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
1380 av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
1381 for (;x>0; x--) av_log(log_ctx, log_level, " ");
1382 av_log(log_ctx, log_level, "|\n");
1386 void sws_freeVec(SwsVector *a)
1389 av_freep(&a->coeff);
1394 void sws_freeFilter(SwsFilter *filter)
1396 if (!filter) return;
1398 if (filter->lumH) sws_freeVec(filter->lumH);
1399 if (filter->lumV) sws_freeVec(filter->lumV);
1400 if (filter->chrH) sws_freeVec(filter->chrH);
1401 if (filter->chrV) sws_freeVec(filter->chrV);
1405 void sws_freeContext(SwsContext *c)
1411 for (i=0; i<c->vLumBufSize; i++)
1412 av_freep(&c->lumPixBuf[i]);
1413 av_freep(&c->lumPixBuf);
1416 if (c->chrUPixBuf) {
1417 for (i=0; i<c->vChrBufSize; i++)
1418 av_freep(&c->chrUPixBuf[i]);
1419 av_freep(&c->chrUPixBuf);
1420 av_freep(&c->chrVPixBuf);
1423 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1424 for (i=0; i<c->vLumBufSize; i++)
1425 av_freep(&c->alpPixBuf[i]);
1426 av_freep(&c->alpPixBuf);
1429 av_freep(&c->vLumFilter);
1430 av_freep(&c->vChrFilter);
1431 av_freep(&c->hLumFilter);
1432 av_freep(&c->hChrFilter);
1434 av_freep(&c->vYCoeffsBank);
1435 av_freep(&c->vCCoeffsBank);
1438 av_freep(&c->vLumFilterPos);
1439 av_freep(&c->vChrFilterPos);
1440 av_freep(&c->hLumFilterPos);
1441 av_freep(&c->hChrFilterPos);
1444 #ifdef MAP_ANONYMOUS
1445 if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);
1446 if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);
1447 #elif HAVE_VIRTUALALLOC
1448 if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);
1449 if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);
1451 av_free(c->lumMmx2FilterCode);
1452 av_free(c->chrMmx2FilterCode);
1454 c->lumMmx2FilterCode=NULL;
1455 c->chrMmx2FilterCode=NULL;
1456 #endif /* HAVE_MMX */
1458 av_freep(&c->yuvTable);
1459 av_freep(&c->formatConvBuffer);
1464 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
1465 int srcW, int srcH, enum PixelFormat srcFormat,
1466 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
1467 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
1469 static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};
1472 param = default_param;
1475 (context->srcW != srcW ||
1476 context->srcH != srcH ||
1477 context->srcFormat != srcFormat ||
1478 context->dstW != dstW ||
1479 context->dstH != dstH ||
1480 context->dstFormat != dstFormat ||
1481 context->flags != flags ||
1482 context->param[0] != param[0] ||
1483 context->param[1] != param[1])) {
1484 sws_freeContext(context);
1489 if (!(context = sws_alloc_context()))
1491 context->srcW = srcW;
1492 context->srcH = srcH;
1493 context->srcRange = handle_jpeg(&srcFormat);
1494 context->srcFormat = srcFormat;
1495 context->dstW = dstW;
1496 context->dstH = dstH;
1497 context->dstRange = handle_jpeg(&dstFormat);
1498 context->dstFormat = dstFormat;
1499 context->flags = flags;
1500 context->param[0] = param[0];
1501 context->param[1] = param[1];
1502 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);
1503 if (sws_init_context(context, srcFilter, dstFilter) < 0) {
1504 sws_freeContext(context);