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/avutil.h"
43 #include "libavutil/bswap.h"
44 #include "libavutil/opt.h"
45 #include "libavutil/pixdesc.h"
47 unsigned swscale_version(void)
49 return LIBSWSCALE_VERSION_INT;
52 const char *swscale_configuration(void)
54 return LIBAV_CONFIGURATION;
57 const char *swscale_license(void)
59 #define LICENSE_PREFIX "libswscale license: "
60 return LICENSE_PREFIX LIBAV_LICENSE + sizeof(LICENSE_PREFIX) - 1;
63 #define RET 0xC3 //near return opcode for x86
65 #define isSupportedIn(x) ( \
66 (x)==PIX_FMT_YUV420P \
67 || (x)==PIX_FMT_YUVA420P \
68 || (x)==PIX_FMT_YUYV422 \
69 || (x)==PIX_FMT_UYVY422 \
70 || (x)==PIX_FMT_RGB48BE \
71 || (x)==PIX_FMT_RGB48LE \
72 || (x)==PIX_FMT_RGB32 \
73 || (x)==PIX_FMT_RGB32_1 \
74 || (x)==PIX_FMT_BGR48BE \
75 || (x)==PIX_FMT_BGR48LE \
76 || (x)==PIX_FMT_BGR24 \
77 || (x)==PIX_FMT_BGR565 \
78 || (x)==PIX_FMT_BGR555 \
79 || (x)==PIX_FMT_BGR32 \
80 || (x)==PIX_FMT_BGR32_1 \
81 || (x)==PIX_FMT_RGB24 \
82 || (x)==PIX_FMT_RGB565 \
83 || (x)==PIX_FMT_RGB555 \
84 || (x)==PIX_FMT_GRAY8 \
85 || (x)==PIX_FMT_Y400A \
86 || (x)==PIX_FMT_YUV410P \
87 || (x)==PIX_FMT_YUV440P \
88 || (x)==PIX_FMT_NV12 \
89 || (x)==PIX_FMT_NV21 \
90 || (x)==PIX_FMT_GRAY16BE \
91 || (x)==PIX_FMT_GRAY16LE \
92 || (x)==PIX_FMT_YUV444P \
93 || (x)==PIX_FMT_YUV422P \
94 || (x)==PIX_FMT_YUV411P \
95 || (x)==PIX_FMT_YUVJ420P \
96 || (x)==PIX_FMT_YUVJ422P \
97 || (x)==PIX_FMT_YUVJ440P \
98 || (x)==PIX_FMT_YUVJ444P \
99 || (x)==PIX_FMT_PAL8 \
100 || (x)==PIX_FMT_BGR8 \
101 || (x)==PIX_FMT_RGB8 \
102 || (x)==PIX_FMT_BGR4_BYTE \
103 || (x)==PIX_FMT_RGB4_BYTE \
104 || (x)==PIX_FMT_YUV440P \
105 || (x)==PIX_FMT_MONOWHITE \
106 || (x)==PIX_FMT_MONOBLACK \
107 || (x)==PIX_FMT_YUV420P9LE \
108 || (x)==PIX_FMT_YUV420P10LE \
109 || (x)==PIX_FMT_YUV420P16LE \
110 || (x)==PIX_FMT_YUV422P16LE \
111 || (x)==PIX_FMT_YUV444P16LE \
112 || (x)==PIX_FMT_YUV420P9BE \
113 || (x)==PIX_FMT_YUV420P10BE \
114 || (x)==PIX_FMT_YUV420P16BE \
115 || (x)==PIX_FMT_YUV422P16BE \
116 || (x)==PIX_FMT_YUV444P16BE \
119 int sws_isSupportedInput(enum PixelFormat pix_fmt)
121 return isSupportedIn(pix_fmt);
124 #define isSupportedOut(x) ( \
125 (x)==PIX_FMT_YUV420P \
126 || (x)==PIX_FMT_YUVA420P \
127 || (x)==PIX_FMT_YUYV422 \
128 || (x)==PIX_FMT_UYVY422 \
129 || (x)==PIX_FMT_YUV444P \
130 || (x)==PIX_FMT_YUV422P \
131 || (x)==PIX_FMT_YUV411P \
132 || (x)==PIX_FMT_YUVJ420P \
133 || (x)==PIX_FMT_YUVJ422P \
134 || (x)==PIX_FMT_YUVJ440P \
135 || (x)==PIX_FMT_YUVJ444P \
137 || (x)==PIX_FMT_NV12 \
138 || (x)==PIX_FMT_NV21 \
139 || (x)==PIX_FMT_GRAY16BE \
140 || (x)==PIX_FMT_GRAY16LE \
141 || (x)==PIX_FMT_GRAY8 \
142 || (x)==PIX_FMT_YUV410P \
143 || (x)==PIX_FMT_YUV440P \
144 || (x)==PIX_FMT_YUV420P9LE \
145 || (x)==PIX_FMT_YUV420P10LE \
146 || (x)==PIX_FMT_YUV420P16LE \
147 || (x)==PIX_FMT_YUV422P16LE \
148 || (x)==PIX_FMT_YUV444P16LE \
149 || (x)==PIX_FMT_YUV420P9BE \
150 || (x)==PIX_FMT_YUV420P10BE \
151 || (x)==PIX_FMT_YUV420P16BE \
152 || (x)==PIX_FMT_YUV422P16BE \
153 || (x)==PIX_FMT_YUV444P16BE \
156 int sws_isSupportedOutput(enum PixelFormat pix_fmt)
158 return isSupportedOut(pix_fmt);
161 extern const int32_t ff_yuv2rgb_coeffs[8][4];
163 const char *sws_format_name(enum PixelFormat format)
165 if ((unsigned)format < PIX_FMT_NB && av_pix_fmt_descriptors[format].name)
166 return av_pix_fmt_descriptors[format].name;
168 return "Unknown format";
171 static double getSplineCoeff(double a, double b, double c, double d, double dist)
173 // printf("%f %f %f %f %f\n", a,b,c,d,dist);
174 if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
175 else return getSplineCoeff( 0.0,
182 static int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
183 int srcW, int dstW, int filterAlign, int one, int flags,
184 SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
190 int64_t *filter=NULL;
191 int64_t *filter2=NULL;
192 const int64_t fone= 1LL<<54;
195 if (flags & SWS_CPU_CAPS_MMX)
196 __asm__ volatile("emms\n\t"::: "memory"); //FIXME this should not be required but it IS (even for non-MMX versions)
199 // NOTE: the +1 is for the MMX scaler which reads over the end
200 FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW+1)*sizeof(int16_t), fail);
202 if (FFABS(xInc - 0x10000) <10) { // unscaled
205 FF_ALLOCZ_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
207 for (i=0; i<dstW; i++) {
208 filter[i*filterSize]= fone;
212 } else if (flags&SWS_POINT) { // lame looking point sampling mode
216 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
218 xDstInSrc= xInc/2 - 0x8000;
219 for (i=0; i<dstW; i++) {
220 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
226 } else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) { // bilinear upscale
230 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
232 xDstInSrc= xInc/2 - 0x8000;
233 for (i=0; i<dstW; i++) {
234 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
238 //bilinear upscale / linear interpolate / area averaging
239 for (j=0; j<filterSize; j++) {
240 int64_t coeff= fone - FFABS((xx<<16) - xDstInSrc)*(fone>>16);
241 if (coeff<0) coeff=0;
242 filter[i*filterSize + j]= coeff;
251 if (flags&SWS_BICUBIC) sizeFactor= 4;
252 else if (flags&SWS_X) sizeFactor= 8;
253 else if (flags&SWS_AREA) sizeFactor= 1; //downscale only, for upscale it is bilinear
254 else if (flags&SWS_GAUSS) sizeFactor= 8; // infinite ;)
255 else if (flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? ceil(2*param[0]) : 6;
256 else if (flags&SWS_SINC) sizeFactor= 20; // infinite ;)
257 else if (flags&SWS_SPLINE) sizeFactor= 20; // infinite ;)
258 else if (flags&SWS_BILINEAR) sizeFactor= 2;
260 sizeFactor= 0; //GCC warning killer
264 if (xInc <= 1<<16) filterSize= 1 + sizeFactor; // upscale
265 else filterSize= 1 + (sizeFactor*srcW + dstW - 1)/ dstW;
267 if (filterSize > srcW-2) filterSize=srcW-2;
269 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
271 xDstInSrc= xInc - 0x10000;
272 for (i=0; i<dstW; i++) {
273 int xx= (xDstInSrc - ((filterSize-2)<<16)) / (1<<17);
276 for (j=0; j<filterSize; j++) {
277 int64_t d= ((int64_t)FFABS((xx<<17) - xDstInSrc))<<13;
283 floatd= d * (1.0/(1<<30));
285 if (flags & SWS_BICUBIC) {
286 int64_t B= (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1<<24);
287 int64_t C= (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1<<24);
288 int64_t dd = ( d*d)>>30;
289 int64_t ddd= (dd*d)>>30;
292 coeff = (12*(1<<24)-9*B-6*C)*ddd + (-18*(1<<24)+12*B+6*C)*dd + (6*(1<<24)-2*B)*(1<<30);
293 else if (d < 1LL<<31)
294 coeff = (-B-6*C)*ddd + (6*B+30*C)*dd + (-12*B-48*C)*d + (8*B+24*C)*(1<<30);
297 coeff *= fone>>(30+24);
299 /* else if (flags & SWS_X) {
300 double p= param ? param*0.01 : 0.3;
301 coeff = d ? sin(d*M_PI)/(d*M_PI) : 1.0;
302 coeff*= pow(2.0, - p*d*d);
304 else if (flags & SWS_X) {
305 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
309 c = cos(floatd*M_PI);
312 if (c<0.0) c= -pow(-c, A);
314 coeff= (c*0.5 + 0.5)*fone;
315 } else if (flags & SWS_AREA) {
316 int64_t d2= d - (1<<29);
317 if (d2*xInc < -(1LL<<(29+16))) coeff= 1.0 * (1LL<<(30+16));
318 else if (d2*xInc < (1LL<<(29+16))) coeff= -d2*xInc + (1LL<<(29+16));
320 coeff *= fone>>(30+16);
321 } else if (flags & SWS_GAUSS) {
322 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
323 coeff = (pow(2.0, - p*floatd*floatd))*fone;
324 } else if (flags & SWS_SINC) {
325 coeff = (d ? sin(floatd*M_PI)/(floatd*M_PI) : 1.0)*fone;
326 } else if (flags & SWS_LANCZOS) {
327 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
328 coeff = (d ? sin(floatd*M_PI)*sin(floatd*M_PI/p)/(floatd*floatd*M_PI*M_PI/p) : 1.0)*fone;
329 if (floatd>p) coeff=0;
330 } else if (flags & SWS_BILINEAR) {
332 if (coeff<0) coeff=0;
334 } else if (flags & SWS_SPLINE) {
335 double p=-2.196152422706632;
336 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, floatd) * fone;
338 coeff= 0.0; //GCC warning killer
342 filter[i*filterSize + j]= coeff;
349 /* apply src & dst Filter to filter -> filter2
352 assert(filterSize>0);
353 filter2Size= filterSize;
354 if (srcFilter) filter2Size+= srcFilter->length - 1;
355 if (dstFilter) filter2Size+= dstFilter->length - 1;
356 assert(filter2Size>0);
357 FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size*dstW*sizeof(*filter2), fail);
359 for (i=0; i<dstW; i++) {
363 for (k=0; k<srcFilter->length; k++) {
364 for (j=0; j<filterSize; j++)
365 filter2[i*filter2Size + k + j] += srcFilter->coeff[k]*filter[i*filterSize + j];
368 for (j=0; j<filterSize; j++)
369 filter2[i*filter2Size + j]= filter[i*filterSize + j];
373 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
377 /* try to reduce the filter-size (step1 find size and shift left) */
378 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
380 for (i=dstW-1; i>=0; i--) {
381 int min= filter2Size;
385 /* get rid of near zero elements on the left by shifting left */
386 for (j=0; j<filter2Size; j++) {
388 cutOff += FFABS(filter2[i*filter2Size]);
390 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
392 /* preserve monotonicity because the core can't handle the filter otherwise */
393 if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
395 // move filter coefficients left
396 for (k=1; k<filter2Size; k++)
397 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
398 filter2[i*filter2Size + k - 1]= 0;
403 /* count near zeros on the right */
404 for (j=filter2Size-1; j>0; j--) {
405 cutOff += FFABS(filter2[i*filter2Size + j]);
407 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
411 if (min>minFilterSize) minFilterSize= min;
414 if (flags & SWS_CPU_CAPS_ALTIVEC) {
415 // we can handle the special case 4,
416 // so we don't want to go to the full 8
417 if (minFilterSize < 5)
420 // We really don't want to waste our time
421 // doing useless computation, so fall back on
422 // the scalar C code for very small filters.
423 // Vectorizing is worth it only if you have a
424 // decent-sized vector.
425 if (minFilterSize < 3)
429 if (flags & SWS_CPU_CAPS_MMX) {
430 // special case for unscaled vertical filtering
431 if (minFilterSize == 1 && filterAlign == 2)
435 assert(minFilterSize > 0);
436 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
437 assert(filterSize > 0);
438 filter= av_malloc(filterSize*dstW*sizeof(*filter));
439 if (filterSize >= MAX_FILTER_SIZE*16/((flags&SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
441 *outFilterSize= filterSize;
443 if (flags&SWS_PRINT_INFO)
444 av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
445 /* try to reduce the filter-size (step2 reduce it) */
446 for (i=0; i<dstW; i++) {
449 for (j=0; j<filterSize; j++) {
450 if (j>=filter2Size) filter[i*filterSize + j]= 0;
451 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
452 if((flags & SWS_BITEXACT) && j>=minFilterSize)
453 filter[i*filterSize + j]= 0;
457 //FIXME try to align filterPos if possible
460 for (i=0; i<dstW; i++) {
462 if ((*filterPos)[i] < 0) {
463 // move filter coefficients left to compensate for filterPos
464 for (j=1; j<filterSize; j++) {
465 int left= FFMAX(j + (*filterPos)[i], 0);
466 filter[i*filterSize + left] += filter[i*filterSize + j];
467 filter[i*filterSize + j]=0;
472 if ((*filterPos)[i] + filterSize > srcW) {
473 int shift= (*filterPos)[i] + filterSize - srcW;
474 // move filter coefficients right to compensate for filterPos
475 for (j=filterSize-2; j>=0; j--) {
476 int right= FFMIN(j + shift, filterSize-1);
477 filter[i*filterSize +right] += filter[i*filterSize +j];
478 filter[i*filterSize +j]=0;
480 (*filterPos)[i]= srcW - filterSize;
484 // Note the +1 is for the MMX scaler which reads over the end
485 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
486 FF_ALLOCZ_OR_GOTO(NULL, *outFilter, *outFilterSize*(dstW+1)*sizeof(int16_t), fail);
488 /* normalize & store in outFilter */
489 for (i=0; i<dstW; i++) {
494 for (j=0; j<filterSize; j++) {
495 sum+= filter[i*filterSize + j];
497 sum= (sum + one/2)/ one;
498 for (j=0; j<*outFilterSize; j++) {
499 int64_t v= filter[i*filterSize + j] + error;
500 int intV= ROUNDED_DIV(v, sum);
501 (*outFilter)[i*(*outFilterSize) + j]= intV;
506 (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
507 for (i=0; i<*outFilterSize; i++) {
508 int j= dstW*(*outFilterSize);
509 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
519 #if ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT)
520 static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode, int16_t *filter, int32_t *filterPos, int numSplits)
523 x86_reg imm8OfPShufW1A;
524 x86_reg imm8OfPShufW2A;
525 x86_reg fragmentLengthA;
527 x86_reg imm8OfPShufW1B;
528 x86_reg imm8OfPShufW2B;
529 x86_reg fragmentLengthB;
534 // create an optimized horizontal scaling routine
535 /* This scaler is made of runtime-generated MMX2 code using specially
536 * tuned pshufw instructions. For every four output pixels, if four
537 * input pixels are enough for the fast bilinear scaling, then a chunk
538 * of fragmentB is used. If five input pixels are needed, then a chunk
539 * of fragmentA is used.
548 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
549 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
550 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
551 "punpcklbw %%mm7, %%mm1 \n\t"
552 "punpcklbw %%mm7, %%mm0 \n\t"
553 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
555 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
557 "psubw %%mm1, %%mm0 \n\t"
558 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
559 "pmullw %%mm3, %%mm0 \n\t"
560 "psllw $7, %%mm1 \n\t"
561 "paddw %%mm1, %%mm0 \n\t"
563 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
565 "add $8, %%"REG_a" \n\t"
569 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
570 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
571 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
576 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
580 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
581 "=r" (fragmentLengthA)
588 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
589 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
590 "punpcklbw %%mm7, %%mm0 \n\t"
591 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
593 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
595 "psubw %%mm1, %%mm0 \n\t"
596 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
597 "pmullw %%mm3, %%mm0 \n\t"
598 "psllw $7, %%mm1 \n\t"
599 "paddw %%mm1, %%mm0 \n\t"
601 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
603 "add $8, %%"REG_a" \n\t"
607 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
608 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
609 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
614 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
618 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
619 "=r" (fragmentLengthB)
622 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
625 for (i=0; i<dstW/numSplits; i++) {
630 int b=((xpos+xInc)>>16) - xx;
631 int c=((xpos+xInc*2)>>16) - xx;
632 int d=((xpos+xInc*3)>>16) - xx;
634 uint8_t *fragment = (d+1<4) ? fragmentB : fragmentA;
635 x86_reg imm8OfPShufW1 = (d+1<4) ? imm8OfPShufW1B : imm8OfPShufW1A;
636 x86_reg imm8OfPShufW2 = (d+1<4) ? imm8OfPShufW2B : imm8OfPShufW2A;
637 x86_reg fragmentLength = (d+1<4) ? fragmentLengthB : fragmentLengthA;
638 int maxShift= 3-(d+inc);
642 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
643 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
644 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
645 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
648 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
650 filterCode[fragmentPos + imm8OfPShufW1]=
651 (a+inc) | ((b+inc)<<2) | ((c+inc)<<4) | ((d+inc)<<6);
652 filterCode[fragmentPos + imm8OfPShufW2]=
653 a | (b<<2) | (c<<4) | (d<<6);
655 if (i+4-inc>=dstW) shift=maxShift; //avoid overread
656 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
658 if (shift && i>=shift) {
659 filterCode[fragmentPos + imm8OfPShufW1]+= 0x55*shift;
660 filterCode[fragmentPos + imm8OfPShufW2]+= 0x55*shift;
661 filterPos[i/2]-=shift;
665 fragmentPos+= fragmentLength;
668 filterCode[fragmentPos]= RET;
673 filterPos[((i/2)+1)&(~1)]= xpos>>16; // needed to jump to the next part
675 return fragmentPos + 1;
677 #endif /* ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT) */
679 static void getSubSampleFactors(int *h, int *v, enum PixelFormat format)
681 *h = av_pix_fmt_descriptors[format].log2_chroma_w;
682 *v = av_pix_fmt_descriptors[format].log2_chroma_h;
685 static int update_flags_cpu(int flags);
687 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation)
689 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
690 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
692 c->brightness= brightness;
693 c->contrast = contrast;
694 c->saturation= saturation;
695 c->srcRange = srcRange;
696 c->dstRange = dstRange;
697 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
699 c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->dstFormat]);
700 c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->srcFormat]);
701 c->flags = update_flags_cpu(c->flags);
703 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
707 if (c->flags & SWS_CPU_CAPS_ALTIVEC)
708 ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness, contrast, saturation);
713 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation)
715 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
717 *inv_table = c->srcColorspaceTable;
718 *table = c->dstColorspaceTable;
719 *srcRange = c->srcRange;
720 *dstRange = c->dstRange;
721 *brightness= c->brightness;
722 *contrast = c->contrast;
723 *saturation= c->saturation;
728 static int handle_jpeg(enum PixelFormat *format)
731 case PIX_FMT_YUVJ420P: *format = PIX_FMT_YUV420P; return 1;
732 case PIX_FMT_YUVJ422P: *format = PIX_FMT_YUV422P; return 1;
733 case PIX_FMT_YUVJ444P: *format = PIX_FMT_YUV444P; return 1;
734 case PIX_FMT_YUVJ440P: *format = PIX_FMT_YUV440P; return 1;
739 static int update_flags_cpu(int flags)
741 #if !CONFIG_RUNTIME_CPUDETECT //ensure that the flags match the compiled variant if cpudetect is off
742 flags &= ~( SWS_CPU_CAPS_MMX
746 |SWS_CPU_CAPS_ALTIVEC
748 flags |= ff_hardcodedcpuflags();
749 #endif /* CONFIG_RUNTIME_CPUDETECT */
753 SwsContext *sws_alloc_context(void)
755 SwsContext *c= av_mallocz(sizeof(SwsContext));
757 c->av_class = &sws_context_class;
758 av_opt_set_defaults(c);
763 int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)
766 int usesVFilter, usesHFilter;
768 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
774 enum PixelFormat srcFormat= c->srcFormat;
775 enum PixelFormat dstFormat= c->dstFormat;
777 flags= c->flags = update_flags_cpu(c->flags);
779 if (flags & SWS_CPU_CAPS_MMX)
780 __asm__ volatile("emms\n\t"::: "memory");
782 if (!rgb15to16) sws_rgb2rgb_init(flags);
784 unscaled = (srcW == dstW && srcH == dstH);
786 if (!isSupportedIn(srcFormat)) {
787 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as input pixel format\n", sws_format_name(srcFormat));
788 return AVERROR(EINVAL);
790 if (!isSupportedOut(dstFormat)) {
791 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as output pixel format\n", sws_format_name(dstFormat));
792 return AVERROR(EINVAL);
795 i= flags & ( SWS_POINT
806 if(!i || (i & (i-1))) {
807 av_log(NULL, AV_LOG_ERROR, "swScaler: Exactly one scaler algorithm must be chosen\n");
808 return AVERROR(EINVAL);
811 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
812 av_log(NULL, AV_LOG_ERROR, "swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
813 srcW, srcH, dstW, dstH);
814 return AVERROR(EINVAL);
816 if(srcW > VOFW || dstW > VOFW) {
817 av_log(NULL, AV_LOG_ERROR, "swScaler: Compile-time maximum width is "AV_STRINGIFY(VOFW)" change VOF/VOFW and recompile\n");
818 return AVERROR(EINVAL);
821 if (!dstFilter) dstFilter= &dummyFilter;
822 if (!srcFilter) srcFilter= &dummyFilter;
824 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
825 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
826 c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[dstFormat]);
827 c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[srcFormat]);
828 c->vRounder= 4* 0x0001000100010001ULL;
830 usesVFilter = (srcFilter->lumV && srcFilter->lumV->length>1) ||
831 (srcFilter->chrV && srcFilter->chrV->length>1) ||
832 (dstFilter->lumV && dstFilter->lumV->length>1) ||
833 (dstFilter->chrV && dstFilter->chrV->length>1);
834 usesHFilter = (srcFilter->lumH && srcFilter->lumH->length>1) ||
835 (srcFilter->chrH && srcFilter->chrH->length>1) ||
836 (dstFilter->lumH && dstFilter->lumH->length>1) ||
837 (dstFilter->chrH && dstFilter->chrH->length>1);
839 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
840 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
842 // reuse chroma for 2 pixels RGB/BGR unless user wants full chroma interpolation
843 if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
845 // drop some chroma lines if the user wants it
846 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
847 c->chrSrcVSubSample+= c->vChrDrop;
849 // drop every other pixel for chroma calculation unless user wants full chroma
850 if (isAnyRGB(srcFormat) && !(flags&SWS_FULL_CHR_H_INP)
851 && srcFormat!=PIX_FMT_RGB8 && srcFormat!=PIX_FMT_BGR8
852 && srcFormat!=PIX_FMT_RGB4 && srcFormat!=PIX_FMT_BGR4
853 && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE
854 && ((dstW>>c->chrDstHSubSample) <= (srcW>>1) || (flags&SWS_FAST_BILINEAR)))
855 c->chrSrcHSubSample=1;
857 // Note the -((-x)>>y) is so that we always round toward +inf.
858 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
859 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
860 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
861 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
863 /* unscaled special cases */
864 if (unscaled && !usesHFilter && !usesVFilter && (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
865 ff_get_unscaled_swscale(c);
868 if (flags&SWS_PRINT_INFO)
869 av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
870 sws_format_name(srcFormat), sws_format_name(dstFormat));
875 if (flags & SWS_CPU_CAPS_MMX2) {
876 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
877 if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR)) {
878 if (flags&SWS_PRINT_INFO)
879 av_log(c, AV_LOG_INFO, "output width is not a multiple of 32 -> no MMX2 scaler\n");
881 if (usesHFilter) c->canMMX2BeUsed=0;
886 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
887 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
889 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
890 // but only for the FAST_BILINEAR mode otherwise do correct scaling
891 // n-2 is the last chrominance sample available
892 // this is not perfect, but no one should notice the difference, the more correct variant
893 // would be like the vertical one, but that would require some special code for the
894 // first and last pixel
895 if (flags&SWS_FAST_BILINEAR) {
896 if (c->canMMX2BeUsed) {
900 //we don't use the x86 asm scaler if MMX is available
901 else if (flags & SWS_CPU_CAPS_MMX) {
902 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
903 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
907 /* precalculate horizontal scaler filter coefficients */
909 #if ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT)
910 // can't downscale !!!
911 if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {
912 c->lumMmx2FilterCodeSize = initMMX2HScaler( dstW, c->lumXInc, NULL, NULL, NULL, 8);
913 c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc, NULL, NULL, NULL, 4);
916 c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
917 c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
918 #elif HAVE_VIRTUALALLOC
919 c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
920 c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
922 c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);
923 c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);
926 if (!c->lumMmx2FilterCode || !c->chrMmx2FilterCode)
927 return AVERROR(ENOMEM);
928 FF_ALLOCZ_OR_GOTO(c, c->hLumFilter , (dstW /8+8)*sizeof(int16_t), fail);
929 FF_ALLOCZ_OR_GOTO(c, c->hChrFilter , (c->chrDstW /4+8)*sizeof(int16_t), fail);
930 FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW /2/8+8)*sizeof(int32_t), fail);
931 FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW/2/4+8)*sizeof(int32_t), fail);
933 initMMX2HScaler( dstW, c->lumXInc, c->lumMmx2FilterCode, c->hLumFilter, c->hLumFilterPos, 8);
934 initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode, c->hChrFilter, c->hChrFilterPos, 4);
937 mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
938 mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
941 #endif /* ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT) */
943 const int filterAlign=
944 (flags & SWS_CPU_CAPS_MMX) ? 4 :
945 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
948 if (initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
949 srcW , dstW, filterAlign, 1<<14,
950 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
951 srcFilter->lumH, dstFilter->lumH, c->param) < 0)
953 if (initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
954 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
955 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
956 srcFilter->chrH, dstFilter->chrH, c->param) < 0)
959 } // initialize horizontal stuff
961 /* precalculate vertical scaler filter coefficients */
963 const int filterAlign=
964 (flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
965 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
968 if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
969 srcH , dstH, filterAlign, (1<<12),
970 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
971 srcFilter->lumV, dstFilter->lumV, c->param) < 0)
973 if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
974 c->chrSrcH, c->chrDstH, filterAlign, (1<<12),
975 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
976 srcFilter->chrV, dstFilter->chrV, c->param) < 0)
980 FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof (vector signed short)*c->vLumFilterSize*c->dstH, fail);
981 FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH, fail);
983 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
985 short *p = (short *)&c->vYCoeffsBank[i];
987 p[j] = c->vLumFilter[i];
990 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
992 short *p = (short *)&c->vCCoeffsBank[i];
994 p[j] = c->vChrFilter[i];
999 // calculate buffer sizes so that they won't run out while handling these damn slices
1000 c->vLumBufSize= c->vLumFilterSize;
1001 c->vChrBufSize= c->vChrFilterSize;
1002 for (i=0; i<dstH; i++) {
1003 int chrI= i*c->chrDstH / dstH;
1004 int nextSlice= FFMAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
1005 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
1007 nextSlice>>= c->chrSrcVSubSample;
1008 nextSlice<<= c->chrSrcVSubSample;
1009 if (c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
1010 c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
1011 if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
1012 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
1015 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
1016 // allocate several megabytes to handle all possible cases)
1017 FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
1018 FF_ALLOC_OR_GOTO(c, c->chrPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
1019 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
1020 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
1021 //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)
1022 /* align at 16 bytes for AltiVec */
1023 for (i=0; i<c->vLumBufSize; i++) {
1024 FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], VOF+1, fail);
1025 c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];
1027 for (i=0; i<c->vChrBufSize; i++) {
1028 FF_ALLOC_OR_GOTO(c, c->chrPixBuf[i+c->vChrBufSize], (VOF+1)*2, fail);
1029 c->chrPixBuf[i] = c->chrPixBuf[i+c->vChrBufSize];
1031 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
1032 for (i=0; i<c->vLumBufSize; i++) {
1033 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], VOF+1, fail);
1034 c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];
1037 //try to avoid drawing green stuff between the right end and the stride end
1038 for (i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, (VOF+1)*2);
1040 assert(2*VOFW == VOF);
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 (flags & SWS_CPU_CAPS_MMX2) av_log(c, AV_LOG_INFO, "using MMX2\n");
1070 else if (flags & SWS_CPU_CAPS_3DNOW) av_log(c, AV_LOG_INFO, "using 3DNOW\n");
1071 else if (flags & SWS_CPU_CAPS_MMX) av_log(c, AV_LOG_INFO, "using MMX\n");
1072 else if (flags & SWS_CPU_CAPS_ALTIVEC) av_log(c, AV_LOG_INFO, "using AltiVec\n");
1073 else av_log(c, AV_LOG_INFO, "using C\n");
1075 if (flags & SWS_CPU_CAPS_MMX) {
1076 if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
1077 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
1079 if (c->hLumFilterSize==4)
1080 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal luminance scaling\n");
1081 else if (c->hLumFilterSize==8)
1082 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal luminance scaling\n");
1084 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal luminance scaling\n");
1086 if (c->hChrFilterSize==4)
1087 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal chrominance scaling\n");
1088 else if (c->hChrFilterSize==8)
1089 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal chrominance scaling\n");
1091 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal chrominance scaling\n");
1095 av_log(c, AV_LOG_VERBOSE, "using x86 asm scaler for horizontal scaling\n");
1097 if (flags & SWS_FAST_BILINEAR)
1098 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR C scaler for horizontal scaling\n");
1100 av_log(c, AV_LOG_VERBOSE, "using C scaler for horizontal scaling\n");
1103 if (isPlanarYUV(dstFormat)) {
1104 if (c->vLumFilterSize==1)
1105 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1107 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1109 if (c->vLumFilterSize==1 && c->vChrFilterSize==2)
1110 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
1111 " 2-tap scaler for vertical chrominance scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1112 else if (c->vLumFilterSize==2 && c->vChrFilterSize==2)
1113 av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1115 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1118 if (dstFormat==PIX_FMT_BGR24)
1119 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 converter\n",
1120 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
1121 else if (dstFormat==PIX_FMT_RGB32)
1122 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1123 else if (dstFormat==PIX_FMT_BGR565)
1124 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1125 else if (dstFormat==PIX_FMT_BGR555)
1126 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1127 else if (dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||
1128 dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE)
1129 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR12 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1131 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
1132 av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1133 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
1134 av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1135 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
1138 c->swScale= ff_getSwsFunc(c);
1140 fail: //FIXME replace things by appropriate error codes
1144 #if FF_API_SWS_GETCONTEXT
1145 SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,
1146 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
1147 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
1151 if(!(c=sws_alloc_context()))
1159 c->srcRange = handle_jpeg(&srcFormat);
1160 c->dstRange = handle_jpeg(&dstFormat);
1161 c->srcFormat= srcFormat;
1162 c->dstFormat= dstFormat;
1165 c->param[0] = param[0];
1166 c->param[1] = param[1];
1168 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);
1170 if(sws_init_context(c, srcFilter, dstFilter) < 0){
1179 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
1180 float lumaSharpen, float chromaSharpen,
1181 float chromaHShift, float chromaVShift,
1184 SwsFilter *filter= av_malloc(sizeof(SwsFilter));
1188 if (lumaGBlur!=0.0) {
1189 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
1190 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
1192 filter->lumH= sws_getIdentityVec();
1193 filter->lumV= sws_getIdentityVec();
1196 if (chromaGBlur!=0.0) {
1197 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
1198 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
1200 filter->chrH= sws_getIdentityVec();
1201 filter->chrV= sws_getIdentityVec();
1204 if (chromaSharpen!=0.0) {
1205 SwsVector *id= sws_getIdentityVec();
1206 sws_scaleVec(filter->chrH, -chromaSharpen);
1207 sws_scaleVec(filter->chrV, -chromaSharpen);
1208 sws_addVec(filter->chrH, id);
1209 sws_addVec(filter->chrV, id);
1213 if (lumaSharpen!=0.0) {
1214 SwsVector *id= sws_getIdentityVec();
1215 sws_scaleVec(filter->lumH, -lumaSharpen);
1216 sws_scaleVec(filter->lumV, -lumaSharpen);
1217 sws_addVec(filter->lumH, id);
1218 sws_addVec(filter->lumV, id);
1222 if (chromaHShift != 0.0)
1223 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
1225 if (chromaVShift != 0.0)
1226 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
1228 sws_normalizeVec(filter->chrH, 1.0);
1229 sws_normalizeVec(filter->chrV, 1.0);
1230 sws_normalizeVec(filter->lumH, 1.0);
1231 sws_normalizeVec(filter->lumV, 1.0);
1233 if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
1234 if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
1239 SwsVector *sws_allocVec(int length)
1241 SwsVector *vec = av_malloc(sizeof(SwsVector));
1244 vec->length = length;
1245 vec->coeff = av_malloc(sizeof(double) * length);
1251 SwsVector *sws_getGaussianVec(double variance, double quality)
1253 const int length= (int)(variance*quality + 0.5) | 1;
1255 double middle= (length-1)*0.5;
1256 SwsVector *vec= sws_allocVec(length);
1261 for (i=0; i<length; i++) {
1262 double dist= i-middle;
1263 vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);
1266 sws_normalizeVec(vec, 1.0);
1271 SwsVector *sws_getConstVec(double c, int length)
1274 SwsVector *vec= sws_allocVec(length);
1279 for (i=0; i<length; i++)
1285 SwsVector *sws_getIdentityVec(void)
1287 return sws_getConstVec(1.0, 1);
1290 static double sws_dcVec(SwsVector *a)
1295 for (i=0; i<a->length; i++)
1301 void sws_scaleVec(SwsVector *a, double scalar)
1305 for (i=0; i<a->length; i++)
1306 a->coeff[i]*= scalar;
1309 void sws_normalizeVec(SwsVector *a, double height)
1311 sws_scaleVec(a, height/sws_dcVec(a));
1314 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
1316 int length= a->length + b->length - 1;
1318 SwsVector *vec= sws_getConstVec(0.0, length);
1323 for (i=0; i<a->length; i++) {
1324 for (j=0; j<b->length; j++) {
1325 vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];
1332 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
1334 int length= FFMAX(a->length, b->length);
1336 SwsVector *vec= sws_getConstVec(0.0, length);
1341 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1342 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
1347 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
1349 int length= FFMAX(a->length, b->length);
1351 SwsVector *vec= sws_getConstVec(0.0, length);
1356 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1357 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
1362 /* shift left / or right if "shift" is negative */
1363 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
1365 int length= a->length + FFABS(shift)*2;
1367 SwsVector *vec= sws_getConstVec(0.0, length);
1372 for (i=0; i<a->length; i++) {
1373 vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
1379 void sws_shiftVec(SwsVector *a, int shift)
1381 SwsVector *shifted= sws_getShiftedVec(a, shift);
1383 a->coeff= shifted->coeff;
1384 a->length= shifted->length;
1388 void sws_addVec(SwsVector *a, SwsVector *b)
1390 SwsVector *sum= sws_sumVec(a, b);
1392 a->coeff= sum->coeff;
1393 a->length= sum->length;
1397 void sws_subVec(SwsVector *a, SwsVector *b)
1399 SwsVector *diff= sws_diffVec(a, b);
1401 a->coeff= diff->coeff;
1402 a->length= diff->length;
1406 void sws_convVec(SwsVector *a, SwsVector *b)
1408 SwsVector *conv= sws_getConvVec(a, b);
1410 a->coeff= conv->coeff;
1411 a->length= conv->length;
1415 SwsVector *sws_cloneVec(SwsVector *a)
1418 SwsVector *vec= sws_allocVec(a->length);
1423 for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];
1428 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
1435 for (i=0; i<a->length; i++)
1436 if (a->coeff[i]>max) max= a->coeff[i];
1438 for (i=0; i<a->length; i++)
1439 if (a->coeff[i]<min) min= a->coeff[i];
1443 for (i=0; i<a->length; i++) {
1444 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
1445 av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
1446 for (;x>0; x--) av_log(log_ctx, log_level, " ");
1447 av_log(log_ctx, log_level, "|\n");
1451 void sws_freeVec(SwsVector *a)
1454 av_freep(&a->coeff);
1459 void sws_freeFilter(SwsFilter *filter)
1461 if (!filter) return;
1463 if (filter->lumH) sws_freeVec(filter->lumH);
1464 if (filter->lumV) sws_freeVec(filter->lumV);
1465 if (filter->chrH) sws_freeVec(filter->chrH);
1466 if (filter->chrV) sws_freeVec(filter->chrV);
1470 void sws_freeContext(SwsContext *c)
1476 for (i=0; i<c->vLumBufSize; i++)
1477 av_freep(&c->lumPixBuf[i]);
1478 av_freep(&c->lumPixBuf);
1482 for (i=0; i<c->vChrBufSize; i++)
1483 av_freep(&c->chrPixBuf[i]);
1484 av_freep(&c->chrPixBuf);
1487 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1488 for (i=0; i<c->vLumBufSize; i++)
1489 av_freep(&c->alpPixBuf[i]);
1490 av_freep(&c->alpPixBuf);
1493 av_freep(&c->vLumFilter);
1494 av_freep(&c->vChrFilter);
1495 av_freep(&c->hLumFilter);
1496 av_freep(&c->hChrFilter);
1498 av_freep(&c->vYCoeffsBank);
1499 av_freep(&c->vCCoeffsBank);
1502 av_freep(&c->vLumFilterPos);
1503 av_freep(&c->vChrFilterPos);
1504 av_freep(&c->hLumFilterPos);
1505 av_freep(&c->hChrFilterPos);
1508 #ifdef MAP_ANONYMOUS
1509 if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);
1510 if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);
1511 #elif HAVE_VIRTUALALLOC
1512 if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);
1513 if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);
1515 av_free(c->lumMmx2FilterCode);
1516 av_free(c->chrMmx2FilterCode);
1518 c->lumMmx2FilterCode=NULL;
1519 c->chrMmx2FilterCode=NULL;
1520 #endif /* ARCH_X86 */
1522 av_freep(&c->yuvTable);
1527 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
1528 int srcW, int srcH, enum PixelFormat srcFormat,
1529 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
1530 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
1532 static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};
1535 param = default_param;
1537 flags = update_flags_cpu(flags);
1540 (context->srcW != srcW ||
1541 context->srcH != srcH ||
1542 context->srcFormat != srcFormat ||
1543 context->dstW != dstW ||
1544 context->dstH != dstH ||
1545 context->dstFormat != dstFormat ||
1546 context->flags != flags ||
1547 context->param[0] != param[0] ||
1548 context->param[1] != param[1])) {
1549 sws_freeContext(context);
1554 if (!(context = sws_alloc_context()))
1556 context->srcW = srcW;
1557 context->srcH = srcH;
1558 context->srcRange = handle_jpeg(&srcFormat);
1559 context->srcFormat = srcFormat;
1560 context->dstW = dstW;
1561 context->dstH = dstH;
1562 context->dstRange = handle_jpeg(&dstFormat);
1563 context->dstFormat = dstFormat;
1564 context->flags = flags;
1565 context->param[0] = param[0];
1566 context->param[1] = param[1];
1567 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);
1568 if (sws_init_context(context, srcFilter, dstFilter) < 0) {
1569 sws_freeContext(context);