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/mathematics.h"
48 #include "libavutil/opt.h"
49 #include "libavutil/pixdesc.h"
50 #include "libavutil/avassert.h"
52 unsigned swscale_version(void)
54 return LIBSWSCALE_VERSION_INT;
57 const char *swscale_configuration(void)
59 return FFMPEG_CONFIGURATION;
62 const char *swscale_license(void)
64 #define LICENSE_PREFIX "libswscale license: "
65 return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
68 #define RET 0xC3 //near return opcode for x86
70 typedef struct FormatEntry {
71 int is_supported_in, is_supported_out;
74 const static FormatEntry format_entries[PIX_FMT_NB] = {
75 [PIX_FMT_YUV420P] = { 1 , 1 },
76 [PIX_FMT_YUYV422] = { 1 , 1 },
77 [PIX_FMT_RGB24] = { 1 , 1 },
78 [PIX_FMT_BGR24] = { 1 , 1 },
79 [PIX_FMT_YUV422P] = { 1 , 1 },
80 [PIX_FMT_YUV444P] = { 1 , 1 },
81 [PIX_FMT_YUV410P] = { 1 , 1 },
82 [PIX_FMT_YUV411P] = { 1 , 1 },
83 [PIX_FMT_GRAY8] = { 1 , 1 },
84 [PIX_FMT_MONOWHITE] = { 1 , 1 },
85 [PIX_FMT_MONOBLACK] = { 1 , 1 },
86 [PIX_FMT_PAL8] = { 1 , 0 },
87 [PIX_FMT_YUVJ420P] = { 1 , 1 },
88 [PIX_FMT_YUVJ422P] = { 1 , 1 },
89 [PIX_FMT_YUVJ444P] = { 1 , 1 },
90 [PIX_FMT_UYVY422] = { 1 , 1 },
91 [PIX_FMT_UYYVYY411] = { 0 , 0 },
92 [PIX_FMT_BGR8] = { 1 , 1 },
93 [PIX_FMT_BGR4] = { 0 , 1 },
94 [PIX_FMT_BGR4_BYTE] = { 1 , 1 },
95 [PIX_FMT_RGB8] = { 1 , 1 },
96 [PIX_FMT_RGB4] = { 0 , 1 },
97 [PIX_FMT_RGB4_BYTE] = { 1 , 1 },
98 [PIX_FMT_NV12] = { 1 , 1 },
99 [PIX_FMT_NV21] = { 1 , 1 },
100 [PIX_FMT_ARGB] = { 1 , 1 },
101 [PIX_FMT_RGBA] = { 1 , 1 },
102 [PIX_FMT_ABGR] = { 1 , 1 },
103 [PIX_FMT_BGRA] = { 1 , 1 },
104 [PIX_FMT_0RGB] = { 1 , 1 },
105 [PIX_FMT_RGB0] = { 1 , 1 },
106 [PIX_FMT_0BGR] = { 1 , 1 },
107 [PIX_FMT_BGR0] = { 1 , 1 },
108 [PIX_FMT_GRAY16BE] = { 1 , 1 },
109 [PIX_FMT_GRAY16LE] = { 1 , 1 },
110 [PIX_FMT_YUV440P] = { 1 , 1 },
111 [PIX_FMT_YUVJ440P] = { 1 , 1 },
112 [PIX_FMT_YUVA420P] = { 1 , 1 },
113 [PIX_FMT_RGB48BE] = { 1 , 1 },
114 [PIX_FMT_RGB48LE] = { 1 , 1 },
115 [PIX_FMT_RGBA64BE] = { 0 , 0 },
116 [PIX_FMT_RGBA64LE] = { 0 , 0 },
117 [PIX_FMT_RGB565BE] = { 1 , 1 },
118 [PIX_FMT_RGB565LE] = { 1 , 1 },
119 [PIX_FMT_RGB555BE] = { 1 , 1 },
120 [PIX_FMT_RGB555LE] = { 1 , 1 },
121 [PIX_FMT_BGR565BE] = { 1 , 1 },
122 [PIX_FMT_BGR565LE] = { 1 , 1 },
123 [PIX_FMT_BGR555BE] = { 1 , 1 },
124 [PIX_FMT_BGR555LE] = { 1 , 1 },
125 [PIX_FMT_YUV420P16LE] = { 1 , 1 },
126 [PIX_FMT_YUV420P16BE] = { 1 , 1 },
127 [PIX_FMT_YUV422P16LE] = { 1 , 1 },
128 [PIX_FMT_YUV422P16BE] = { 1 , 1 },
129 [PIX_FMT_YUV444P16LE] = { 1 , 1 },
130 [PIX_FMT_YUV444P16BE] = { 1 , 1 },
131 [PIX_FMT_RGB444LE] = { 0 , 1 },
132 [PIX_FMT_RGB444BE] = { 0 , 1 },
133 [PIX_FMT_BGR444LE] = { 0 , 1 },
134 [PIX_FMT_BGR444BE] = { 0 , 1 },
135 [PIX_FMT_Y400A] = { 1 , 0 },
136 [PIX_FMT_BGR48BE] = { 1 , 1 },
137 [PIX_FMT_BGR48LE] = { 1 , 1 },
138 [PIX_FMT_BGRA64BE] = { 0 , 0 },
139 [PIX_FMT_BGRA64LE] = { 0 , 0 },
140 [PIX_FMT_YUV420P9BE] = { 1 , 1 },
141 [PIX_FMT_YUV420P9LE] = { 1 , 1 },
142 [PIX_FMT_YUV420P10BE] = { 1 , 1 },
143 [PIX_FMT_YUV420P10LE] = { 1 , 1 },
144 [PIX_FMT_YUV422P9BE] = { 1 , 1 },
145 [PIX_FMT_YUV422P9LE] = { 1 , 1 },
146 [PIX_FMT_YUV422P10BE] = { 1 , 1 },
147 [PIX_FMT_YUV422P10LE] = { 1 , 1 },
148 [PIX_FMT_YUV444P9BE] = { 1 , 1 },
149 [PIX_FMT_YUV444P9LE] = { 1 , 1 },
150 [PIX_FMT_YUV444P10BE] = { 1 , 1 },
151 [PIX_FMT_YUV444P10LE] = { 1 , 1 },
152 [PIX_FMT_GBR24P] = { 1 , 0 },
153 [PIX_FMT_GBRP] = { 1 , 0 },
154 [PIX_FMT_GBRP9LE] = { 1 , 0 },
155 [PIX_FMT_GBRP9BE] = { 1 , 0 },
156 [PIX_FMT_GBRP10LE] = { 1 , 0 },
157 [PIX_FMT_GBRP10BE] = { 1 , 0 },
158 [PIX_FMT_GBRP16LE] = { 1 , 0 },
159 [PIX_FMT_GBRP16BE] = { 1 , 0 },
162 int sws_isSupportedInput(enum PixelFormat pix_fmt)
164 return (unsigned)pix_fmt < PIX_FMT_NB ?
165 format_entries[pix_fmt].is_supported_in : 0;
168 int sws_isSupportedOutput(enum PixelFormat pix_fmt)
170 return (unsigned)pix_fmt < PIX_FMT_NB ?
171 format_entries[pix_fmt].is_supported_out : 0;
174 extern const int32_t ff_yuv2rgb_coeffs[8][4];
176 #if FF_API_SWS_FORMAT_NAME
177 const char *sws_format_name(enum PixelFormat format)
179 return av_get_pix_fmt_name(format);
183 static double getSplineCoeff(double a, double b, double c, double d, double dist)
185 if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
186 else return getSplineCoeff( 0.0,
193 static int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
194 int srcW, int dstW, int filterAlign, int one, int flags, int cpu_flags,
195 SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
201 int64_t *filter=NULL;
202 int64_t *filter2=NULL;
203 const int64_t fone= 1LL<<54;
206 emms_c(); //FIXME this should not be required but it IS (even for non-MMX versions)
208 // NOTE: the +3 is for the MMX(+1)/SSE(+3) scaler which reads over the end
209 FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW+3)*sizeof(int16_t), fail);
211 if (FFABS(xInc - 0x10000) <10) { // unscaled
214 FF_ALLOCZ_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
216 for (i=0; i<dstW; i++) {
217 filter[i*filterSize]= fone;
221 } else if (flags&SWS_POINT) { // lame looking point sampling mode
225 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
227 xDstInSrc= xInc/2 - 0x8000;
228 for (i=0; i<dstW; i++) {
229 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
235 } else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) { // bilinear upscale
239 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
241 xDstInSrc= xInc/2 - 0x8000;
242 for (i=0; i<dstW; i++) {
243 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
247 //bilinear upscale / linear interpolate / area averaging
248 for (j=0; j<filterSize; j++) {
249 int64_t coeff= fone - FFABS((xx<<16) - xDstInSrc)*(fone>>16);
250 if (coeff<0) coeff=0;
251 filter[i*filterSize + j]= coeff;
260 if (flags&SWS_BICUBIC) sizeFactor= 4;
261 else if (flags&SWS_X) sizeFactor= 8;
262 else if (flags&SWS_AREA) sizeFactor= 1; //downscale only, for upscale it is bilinear
263 else if (flags&SWS_GAUSS) sizeFactor= 8; // infinite ;)
264 else if (flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? ceil(2*param[0]) : 6;
265 else if (flags&SWS_SINC) sizeFactor= 20; // infinite ;)
266 else if (flags&SWS_SPLINE) sizeFactor= 20; // infinite ;)
267 else if (flags&SWS_BILINEAR) sizeFactor= 2;
269 sizeFactor= 0; //GCC warning killer
273 if (xInc <= 1<<16) filterSize= 1 + sizeFactor; // upscale
274 else filterSize= 1 + (sizeFactor*srcW + dstW - 1)/ dstW;
276 if (filterSize > srcW-2) filterSize=srcW-2;
278 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
280 xDstInSrc= xInc - 0x10000;
281 for (i=0; i<dstW; i++) {
282 int xx= (xDstInSrc - ((filterSize-2)<<16)) / (1<<17);
285 for (j=0; j<filterSize; j++) {
286 int64_t d= ((int64_t)FFABS((xx<<17) - xDstInSrc))<<13;
292 floatd= d * (1.0/(1<<30));
294 if (flags & SWS_BICUBIC) {
295 int64_t B= (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1<<24);
296 int64_t C= (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1<<24);
301 int64_t dd = (d * d) >> 30;
302 int64_t ddd = (dd * d) >> 30;
305 coeff = (12*(1<<24)-9*B-6*C)*ddd + (-18*(1<<24)+12*B+6*C)*dd + (6*(1<<24)-2*B)*(1<<30);
307 coeff = (-B-6*C)*ddd + (6*B+30*C)*dd + (-12*B-48*C)*d + (8*B+24*C)*(1<<30);
309 coeff *= fone>>(30+24);
311 /* else if (flags & SWS_X) {
312 double p= param ? param*0.01 : 0.3;
313 coeff = d ? sin(d*M_PI)/(d*M_PI) : 1.0;
314 coeff*= pow(2.0, - p*d*d);
316 else if (flags & SWS_X) {
317 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
321 c = cos(floatd*M_PI);
324 if (c<0.0) c= -pow(-c, A);
326 coeff= (c*0.5 + 0.5)*fone;
327 } else if (flags & SWS_AREA) {
328 int64_t d2= d - (1<<29);
329 if (d2*xInc < -(1LL<<(29+16))) coeff= 1.0 * (1LL<<(30+16));
330 else if (d2*xInc < (1LL<<(29+16))) coeff= -d2*xInc + (1LL<<(29+16));
332 coeff *= fone>>(30+16);
333 } else if (flags & SWS_GAUSS) {
334 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
335 coeff = (pow(2.0, - p*floatd*floatd))*fone;
336 } else if (flags & SWS_SINC) {
337 coeff = (d ? sin(floatd*M_PI)/(floatd*M_PI) : 1.0)*fone;
338 } else if (flags & SWS_LANCZOS) {
339 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
340 coeff = (d ? sin(floatd*M_PI)*sin(floatd*M_PI/p)/(floatd*floatd*M_PI*M_PI/p) : 1.0)*fone;
341 if (floatd>p) coeff=0;
342 } else if (flags & SWS_BILINEAR) {
344 if (coeff<0) coeff=0;
346 } else if (flags & SWS_SPLINE) {
347 double p=-2.196152422706632;
348 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, floatd) * fone;
350 coeff= 0.0; //GCC warning killer
354 filter[i*filterSize + j]= coeff;
361 /* apply src & dst Filter to filter -> filter2
364 assert(filterSize>0);
365 filter2Size= filterSize;
366 if (srcFilter) filter2Size+= srcFilter->length - 1;
367 if (dstFilter) filter2Size+= dstFilter->length - 1;
368 assert(filter2Size>0);
369 FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size*dstW*sizeof(*filter2), fail);
371 for (i=0; i<dstW; i++) {
375 for (k=0; k<srcFilter->length; k++) {
376 for (j=0; j<filterSize; j++)
377 filter2[i*filter2Size + k + j] += srcFilter->coeff[k]*filter[i*filterSize + j];
380 for (j=0; j<filterSize; j++)
381 filter2[i*filter2Size + j]= filter[i*filterSize + j];
385 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
389 /* try to reduce the filter-size (step1 find size and shift left) */
390 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
392 for (i=dstW-1; i>=0; i--) {
393 int min= filter2Size;
397 /* get rid of near zero elements on the left by shifting left */
398 for (j=0; j<filter2Size; j++) {
400 cutOff += FFABS(filter2[i*filter2Size]);
402 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
404 /* preserve monotonicity because the core can't handle the filter otherwise */
405 if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
407 // move filter coefficients left
408 for (k=1; k<filter2Size; k++)
409 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
410 filter2[i*filter2Size + k - 1]= 0;
415 /* count near zeros on the right */
416 for (j=filter2Size-1; j>0; j--) {
417 cutOff += FFABS(filter2[i*filter2Size + j]);
419 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
423 if (min>minFilterSize) minFilterSize= min;
426 if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) {
427 // we can handle the special case 4,
428 // so we don't want to go to the full 8
429 if (minFilterSize < 5)
432 // We really don't want to waste our time
433 // doing useless computation, so fall back on
434 // the scalar C code for very small filters.
435 // Vectorizing is worth it only if you have a
436 // decent-sized vector.
437 if (minFilterSize < 3)
441 if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
442 // special case for unscaled vertical filtering
443 if (minFilterSize == 1 && filterAlign == 2)
447 assert(minFilterSize > 0);
448 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
449 assert(filterSize > 0);
450 filter= av_malloc(filterSize*dstW*sizeof(*filter));
451 if (filterSize >= MAX_FILTER_SIZE*16/((flags&SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
453 *outFilterSize= filterSize;
455 if (flags&SWS_PRINT_INFO)
456 av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
457 /* try to reduce the filter-size (step2 reduce it) */
458 for (i=0; i<dstW; i++) {
461 for (j=0; j<filterSize; j++) {
462 if (j>=filter2Size) filter[i*filterSize + j]= 0;
463 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
464 if((flags & SWS_BITEXACT) && j>=minFilterSize)
465 filter[i*filterSize + j]= 0;
469 //FIXME try to align filterPos if possible
472 for (i=0; i<dstW; i++) {
474 if ((*filterPos)[i] < 0) {
475 // move filter coefficients left to compensate for filterPos
476 for (j=1; j<filterSize; j++) {
477 int left= FFMAX(j + (*filterPos)[i], 0);
478 filter[i*filterSize + left] += filter[i*filterSize + j];
479 filter[i*filterSize + j]=0;
484 if ((*filterPos)[i] + filterSize > srcW) {
485 int shift= (*filterPos)[i] + filterSize - srcW;
486 // move filter coefficients right to compensate for filterPos
487 for (j=filterSize-2; j>=0; j--) {
488 int right= FFMIN(j + shift, filterSize-1);
489 filter[i*filterSize +right] += filter[i*filterSize +j];
490 filter[i*filterSize +j]=0;
492 (*filterPos)[i]= srcW - filterSize;
496 // Note the +1 is for the MMX scaler which reads over the end
497 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
498 FF_ALLOCZ_OR_GOTO(NULL, *outFilter, *outFilterSize*(dstW+3)*sizeof(int16_t), fail);
500 /* normalize & store in outFilter */
501 for (i=0; i<dstW; i++) {
506 for (j=0; j<filterSize; j++) {
507 sum+= filter[i*filterSize + j];
509 sum= (sum + one/2)/ one;
510 for (j=0; j<*outFilterSize; j++) {
511 int64_t v= filter[i*filterSize + j] + error;
512 int intV= ROUNDED_DIV(v, sum);
513 (*outFilter)[i*(*outFilterSize) + j]= intV;
518 (*filterPos)[dstW+0] =
519 (*filterPos)[dstW+1] =
520 (*filterPos)[dstW+2] = (*filterPos)[dstW-1]; // the MMX/SSE scaler will read over the end
521 for (i=0; i<*outFilterSize; i++) {
522 int k= (dstW - 1) * (*outFilterSize) + i;
523 (*outFilter)[k + 1 * (*outFilterSize)] =
524 (*outFilter)[k + 2 * (*outFilterSize)] =
525 (*outFilter)[k + 3 * (*outFilterSize)] = (*outFilter)[k];
536 static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode, int16_t *filter, int32_t *filterPos, int numSplits)
539 x86_reg imm8OfPShufW1A;
540 x86_reg imm8OfPShufW2A;
541 x86_reg fragmentLengthA;
543 x86_reg imm8OfPShufW1B;
544 x86_reg imm8OfPShufW2B;
545 x86_reg fragmentLengthB;
550 // create an optimized horizontal scaling routine
551 /* This scaler is made of runtime-generated MMX2 code using specially
552 * tuned pshufw instructions. For every four output pixels, if four
553 * input pixels are enough for the fast bilinear scaling, then a chunk
554 * of fragmentB is used. If five input pixels are needed, then a chunk
555 * of fragmentA is used.
564 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
565 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
566 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
567 "punpcklbw %%mm7, %%mm1 \n\t"
568 "punpcklbw %%mm7, %%mm0 \n\t"
569 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
571 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
573 "psubw %%mm1, %%mm0 \n\t"
574 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
575 "pmullw %%mm3, %%mm0 \n\t"
576 "psllw $7, %%mm1 \n\t"
577 "paddw %%mm1, %%mm0 \n\t"
579 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
581 "add $8, %%"REG_a" \n\t"
585 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
586 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
587 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
592 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
596 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
597 "=r" (fragmentLengthA)
604 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
605 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
606 "punpcklbw %%mm7, %%mm0 \n\t"
607 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
609 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
611 "psubw %%mm1, %%mm0 \n\t"
612 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
613 "pmullw %%mm3, %%mm0 \n\t"
614 "psllw $7, %%mm1 \n\t"
615 "paddw %%mm1, %%mm0 \n\t"
617 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
619 "add $8, %%"REG_a" \n\t"
623 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
624 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
625 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
630 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
634 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
635 "=r" (fragmentLengthB)
638 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
641 for (i=0; i<dstW/numSplits; i++) {
646 int b=((xpos+xInc)>>16) - xx;
647 int c=((xpos+xInc*2)>>16) - xx;
648 int d=((xpos+xInc*3)>>16) - xx;
650 uint8_t *fragment = (d+1<4) ? fragmentB : fragmentA;
651 x86_reg imm8OfPShufW1 = (d+1<4) ? imm8OfPShufW1B : imm8OfPShufW1A;
652 x86_reg imm8OfPShufW2 = (d+1<4) ? imm8OfPShufW2B : imm8OfPShufW2A;
653 x86_reg fragmentLength = (d+1<4) ? fragmentLengthB : fragmentLengthA;
654 int maxShift= 3-(d+inc);
658 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
659 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
660 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
661 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
664 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
666 filterCode[fragmentPos + imm8OfPShufW1]=
667 (a+inc) | ((b+inc)<<2) | ((c+inc)<<4) | ((d+inc)<<6);
668 filterCode[fragmentPos + imm8OfPShufW2]=
669 a | (b<<2) | (c<<4) | (d<<6);
671 if (i+4-inc>=dstW) shift=maxShift; //avoid overread
672 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
674 if (shift && i>=shift) {
675 filterCode[fragmentPos + imm8OfPShufW1]+= 0x55*shift;
676 filterCode[fragmentPos + imm8OfPShufW2]+= 0x55*shift;
677 filterPos[i/2]-=shift;
681 fragmentPos+= fragmentLength;
684 filterCode[fragmentPos]= RET;
689 filterPos[((i/2)+1)&(~1)]= xpos>>16; // needed to jump to the next part
691 return fragmentPos + 1;
693 #endif /* HAVE_MMX2 */
695 static void getSubSampleFactors(int *h, int *v, enum PixelFormat format)
697 *h = av_pix_fmt_descriptors[format].log2_chroma_w;
698 *v = av_pix_fmt_descriptors[format].log2_chroma_h;
701 int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],
702 int srcRange, const int table[4], int dstRange,
703 int brightness, int contrast, int saturation)
705 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
706 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
708 c->brightness= brightness;
709 c->contrast = contrast;
710 c->saturation= saturation;
711 c->srcRange = srcRange;
712 c->dstRange = dstRange;
713 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
715 c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->dstFormat]);
716 c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->srcFormat]);
718 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
721 if (HAVE_ALTIVEC && av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)
722 ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness, contrast, saturation);
726 int sws_getColorspaceDetails(struct SwsContext *c, int **inv_table,
727 int *srcRange, int **table, int *dstRange,
728 int *brightness, int *contrast, int *saturation)
730 if (!c || isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
732 *inv_table = c->srcColorspaceTable;
733 *table = c->dstColorspaceTable;
734 *srcRange = c->srcRange;
735 *dstRange = c->dstRange;
736 *brightness= c->brightness;
737 *contrast = c->contrast;
738 *saturation= c->saturation;
743 static int handle_jpeg(enum PixelFormat *format)
746 case PIX_FMT_YUVJ420P: *format = PIX_FMT_YUV420P; return 1;
747 case PIX_FMT_YUVJ422P: *format = PIX_FMT_YUV422P; return 1;
748 case PIX_FMT_YUVJ444P: *format = PIX_FMT_YUV444P; return 1;
749 case PIX_FMT_YUVJ440P: *format = PIX_FMT_YUV440P; return 1;
750 case PIX_FMT_0BGR : *format = PIX_FMT_ABGR ; return 0;
751 case PIX_FMT_BGR0 : *format = PIX_FMT_BGRA ; return 0;
752 case PIX_FMT_0RGB : *format = PIX_FMT_ARGB ; return 0;
753 case PIX_FMT_RGB0 : *format = PIX_FMT_RGBA ; return 0;
758 SwsContext *sws_alloc_context(void)
760 SwsContext *c= av_mallocz(sizeof(SwsContext));
762 c->av_class = &sws_context_class;
763 av_opt_set_defaults(c);
768 int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)
771 int usesVFilter, usesHFilter;
773 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
778 int dst_stride = FFALIGN(dstW * sizeof(int16_t)+66, 16);
779 int flags, cpu_flags;
780 enum PixelFormat srcFormat= c->srcFormat;
781 enum PixelFormat dstFormat= c->dstFormat;
783 cpu_flags = av_get_cpu_flags();
786 if (!rgb15to16) sws_rgb2rgb_init();
788 unscaled = (srcW == dstW && srcH == dstH);
790 handle_jpeg(&srcFormat);
791 handle_jpeg(&dstFormat);
793 if(srcFormat!=c->srcFormat || dstFormat!=c->dstFormat){
794 av_log(c, AV_LOG_WARNING, "deprecated pixel format used, make sure you did set range correctly\n");
795 c->srcFormat= srcFormat;
796 c->dstFormat= dstFormat;
799 if (!sws_isSupportedInput(srcFormat)) {
800 av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n", av_get_pix_fmt_name(srcFormat));
801 return AVERROR(EINVAL);
803 if (!sws_isSupportedOutput(dstFormat)) {
804 av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n", av_get_pix_fmt_name(dstFormat));
805 return AVERROR(EINVAL);
808 i= flags & ( SWS_POINT
819 if(!i || (i & (i-1))) {
820 av_log(c, AV_LOG_ERROR, "Exactly one scaler algorithm must be chosen\n");
821 return AVERROR(EINVAL);
824 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
825 av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",
826 srcW, srcH, dstW, dstH);
827 return AVERROR(EINVAL);
830 if (!dstFilter) dstFilter= &dummyFilter;
831 if (!srcFilter) srcFilter= &dummyFilter;
833 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
834 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
835 c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[dstFormat]);
836 c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[srcFormat]);
837 c->vRounder= 4* 0x0001000100010001ULL;
839 usesVFilter = (srcFilter->lumV && srcFilter->lumV->length>1) ||
840 (srcFilter->chrV && srcFilter->chrV->length>1) ||
841 (dstFilter->lumV && dstFilter->lumV->length>1) ||
842 (dstFilter->chrV && dstFilter->chrV->length>1);
843 usesHFilter = (srcFilter->lumH && srcFilter->lumH->length>1) ||
844 (srcFilter->chrH && srcFilter->chrH->length>1) ||
845 (dstFilter->lumH && dstFilter->lumH->length>1) ||
846 (dstFilter->chrH && dstFilter->chrH->length>1);
848 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
849 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
851 // reuse chroma for 2 pixels RGB/BGR unless user wants full chroma interpolation
852 if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
854 // drop some chroma lines if the user wants it
855 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
856 c->chrSrcVSubSample+= c->vChrDrop;
858 // drop every other pixel for chroma calculation unless user wants full chroma
859 if (isAnyRGB(srcFormat) && !(flags&SWS_FULL_CHR_H_INP)
860 && srcFormat!=PIX_FMT_RGB8 && srcFormat!=PIX_FMT_BGR8
861 && srcFormat!=PIX_FMT_RGB4 && srcFormat!=PIX_FMT_BGR4
862 && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE
863 && ((dstW>>c->chrDstHSubSample) <= (srcW>>1) || (flags&SWS_FAST_BILINEAR)))
864 c->chrSrcHSubSample=1;
866 // Note the -((-x)>>y) is so that we always round toward +inf.
867 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
868 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
869 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
870 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
872 /* unscaled special cases */
873 if (unscaled && !usesHFilter && !usesVFilter && (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
874 ff_get_unscaled_swscale(c);
877 if (flags&SWS_PRINT_INFO)
878 av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
879 av_get_pix_fmt_name(srcFormat), av_get_pix_fmt_name(dstFormat));
884 c->srcBpc = 1 + av_pix_fmt_descriptors[srcFormat].comp[0].depth_minus1;
887 c->dstBpc = 1 + av_pix_fmt_descriptors[dstFormat].comp[0].depth_minus1;
890 if (isAnyRGB(srcFormat) || srcFormat == PIX_FMT_PAL8)
894 FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW*2+78, 16) * 2, fail);
895 if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2 && c->srcBpc == 8 && c->dstBpc <= 10) {
896 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
897 if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR)) {
898 if (flags&SWS_PRINT_INFO)
899 av_log(c, AV_LOG_INFO, "output width is not a multiple of 32 -> no MMX2 scaler\n");
901 if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat)) c->canMMX2BeUsed=0;
906 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
907 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
909 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
910 // but only for the FAST_BILINEAR mode otherwise do correct scaling
911 // n-2 is the last chrominance sample available
912 // this is not perfect, but no one should notice the difference, the more correct variant
913 // would be like the vertical one, but that would require some special code for the
914 // first and last pixel
915 if (flags&SWS_FAST_BILINEAR) {
916 if (c->canMMX2BeUsed) {
920 //we don't use the x86 asm scaler if MMX is available
921 else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX && c->dstBpc <= 10) {
922 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
923 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
927 /* precalculate horizontal scaler filter coefficients */
930 // can't downscale !!!
931 if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {
932 c->lumMmx2FilterCodeSize = initMMX2HScaler( dstW, c->lumXInc, NULL, NULL, NULL, 8);
933 c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc, NULL, NULL, NULL, 4);
936 c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
937 c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
938 #elif HAVE_VIRTUALALLOC
939 c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
940 c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
942 c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);
943 c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);
947 if (c->lumMmx2FilterCode == MAP_FAILED || c->chrMmx2FilterCode == MAP_FAILED)
949 if (!c->lumMmx2FilterCode || !c->chrMmx2FilterCode)
951 return AVERROR(ENOMEM);
952 FF_ALLOCZ_OR_GOTO(c, c->hLumFilter , (dstW /8+8)*sizeof(int16_t), fail);
953 FF_ALLOCZ_OR_GOTO(c, c->hChrFilter , (c->chrDstW /4+8)*sizeof(int16_t), fail);
954 FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW /2/8+8)*sizeof(int32_t), fail);
955 FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW/2/4+8)*sizeof(int32_t), fail);
957 initMMX2HScaler( dstW, c->lumXInc, c->lumMmx2FilterCode, c->hLumFilter, (uint32_t*)c->hLumFilterPos, 8);
958 initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode, c->hChrFilter, (uint32_t*)c->hChrFilterPos, 4);
961 mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
962 mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
965 #endif /* HAVE_MMX2 */
967 const int filterAlign=
968 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :
969 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
972 if (initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
973 srcW , dstW, filterAlign, 1<<14,
974 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags, cpu_flags,
975 srcFilter->lumH, dstFilter->lumH, c->param) < 0)
977 if (initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
978 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
979 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, cpu_flags,
980 srcFilter->chrH, dstFilter->chrH, c->param) < 0)
983 } // initialize horizontal stuff
985 /* precalculate vertical scaler filter coefficients */
987 const int filterAlign=
988 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 2 :
989 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
992 if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
993 srcH , dstH, filterAlign, (1<<12),
994 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags, cpu_flags,
995 srcFilter->lumV, dstFilter->lumV, c->param) < 0)
997 if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
998 c->chrSrcH, c->chrDstH, filterAlign, (1<<12),
999 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, cpu_flags,
1000 srcFilter->chrV, dstFilter->chrV, c->param) < 0)
1004 FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof (vector signed short)*c->vLumFilterSize*c->dstH, fail);
1005 FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH, fail);
1007 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
1009 short *p = (short *)&c->vYCoeffsBank[i];
1011 p[j] = c->vLumFilter[i];
1014 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
1016 short *p = (short *)&c->vCCoeffsBank[i];
1018 p[j] = c->vChrFilter[i];
1023 // calculate buffer sizes so that they won't run out while handling these damn slices
1024 c->vLumBufSize= c->vLumFilterSize;
1025 c->vChrBufSize= c->vChrFilterSize;
1026 for (i=0; i<dstH; i++) {
1027 int chrI= (int64_t)i*c->chrDstH / dstH;
1028 int nextSlice= FFMAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
1029 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
1031 nextSlice>>= c->chrSrcVSubSample;
1032 nextSlice<<= c->chrSrcVSubSample;
1033 if (c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
1034 c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
1035 if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
1036 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
1039 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
1040 // allocate several megabytes to handle all possible cases)
1041 FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
1042 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
1043 FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
1044 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
1045 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
1046 //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)
1047 /* align at 16 bytes for AltiVec */
1048 for (i=0; i<c->vLumBufSize; i++) {
1049 FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], dst_stride+16, fail);
1050 c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];
1052 // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)
1053 c->uv_off = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);
1054 c->uv_offx2 = dst_stride + 16;
1055 for (i=0; i<c->vChrBufSize; i++) {
1056 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i+c->vChrBufSize], dst_stride*2+32, fail);
1057 c->chrUPixBuf[i] = c->chrUPixBuf[i+c->vChrBufSize];
1058 c->chrVPixBuf[i] = c->chrVPixBuf[i+c->vChrBufSize] = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;
1060 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
1061 for (i=0; i<c->vLumBufSize; i++) {
1062 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], dst_stride+16, fail);
1063 c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];
1066 //try to avoid drawing green stuff between the right end and the stride end
1067 for (i=0; i<c->vChrBufSize; i++)
1068 if(av_pix_fmt_descriptors[c->dstFormat].comp[0].depth_minus1 == 15){
1069 av_assert0(c->dstBpc > 10);
1070 for(j=0; j<dst_stride/2+1; j++)
1071 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;
1073 for(j=0; j<dst_stride+1; j++)
1074 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;
1076 assert(c->chrDstH <= dstH);
1078 if (flags&SWS_PRINT_INFO) {
1079 if (flags&SWS_FAST_BILINEAR) av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
1080 else if (flags&SWS_BILINEAR) av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
1081 else if (flags&SWS_BICUBIC) av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
1082 else if (flags&SWS_X) av_log(c, AV_LOG_INFO, "Experimental scaler, ");
1083 else if (flags&SWS_POINT) av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
1084 else if (flags&SWS_AREA) av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
1085 else if (flags&SWS_BICUBLIN) av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
1086 else if (flags&SWS_GAUSS) av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
1087 else if (flags&SWS_SINC) av_log(c, AV_LOG_INFO, "Sinc scaler, ");
1088 else if (flags&SWS_LANCZOS) av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
1089 else if (flags&SWS_SPLINE) av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
1090 else av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
1092 av_log(c, AV_LOG_INFO, "from %s to %s%s ",
1093 av_get_pix_fmt_name(srcFormat),
1095 dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ||
1096 dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||
1097 dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ? "dithered " : "",
1101 av_get_pix_fmt_name(dstFormat));
1103 if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2) av_log(c, AV_LOG_INFO, "using MMX2\n");
1104 else if (HAVE_AMD3DNOW && cpu_flags & AV_CPU_FLAG_3DNOW) av_log(c, AV_LOG_INFO, "using 3DNOW\n");
1105 else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) av_log(c, AV_LOG_INFO, "using MMX\n");
1106 else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) av_log(c, AV_LOG_INFO, "using AltiVec\n");
1107 else av_log(c, AV_LOG_INFO, "using C\n");
1109 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
1110 av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1111 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
1112 av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1113 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
1116 c->swScale= ff_getSwsFunc(c);
1118 fail: //FIXME replace things by appropriate error codes
1122 #if FF_API_SWS_GETCONTEXT
1123 SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,
1124 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
1125 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
1129 if(!(c=sws_alloc_context()))
1137 c->srcRange = handle_jpeg(&srcFormat);
1138 c->dstRange = handle_jpeg(&dstFormat);
1139 c->srcFormat= srcFormat;
1140 c->dstFormat= dstFormat;
1143 c->param[0] = param[0];
1144 c->param[1] = param[1];
1146 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);
1148 if(sws_init_context(c, srcFilter, dstFilter) < 0){
1157 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
1158 float lumaSharpen, float chromaSharpen,
1159 float chromaHShift, float chromaVShift,
1162 SwsFilter *filter= av_malloc(sizeof(SwsFilter));
1166 if (lumaGBlur!=0.0) {
1167 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
1168 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
1170 filter->lumH= sws_getIdentityVec();
1171 filter->lumV= sws_getIdentityVec();
1174 if (chromaGBlur!=0.0) {
1175 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
1176 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
1178 filter->chrH= sws_getIdentityVec();
1179 filter->chrV= sws_getIdentityVec();
1182 if (chromaSharpen!=0.0) {
1183 SwsVector *id= sws_getIdentityVec();
1184 sws_scaleVec(filter->chrH, -chromaSharpen);
1185 sws_scaleVec(filter->chrV, -chromaSharpen);
1186 sws_addVec(filter->chrH, id);
1187 sws_addVec(filter->chrV, id);
1191 if (lumaSharpen!=0.0) {
1192 SwsVector *id= sws_getIdentityVec();
1193 sws_scaleVec(filter->lumH, -lumaSharpen);
1194 sws_scaleVec(filter->lumV, -lumaSharpen);
1195 sws_addVec(filter->lumH, id);
1196 sws_addVec(filter->lumV, id);
1200 if (chromaHShift != 0.0)
1201 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
1203 if (chromaVShift != 0.0)
1204 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
1206 sws_normalizeVec(filter->chrH, 1.0);
1207 sws_normalizeVec(filter->chrV, 1.0);
1208 sws_normalizeVec(filter->lumH, 1.0);
1209 sws_normalizeVec(filter->lumV, 1.0);
1211 if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
1212 if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
1217 SwsVector *sws_allocVec(int length)
1219 SwsVector *vec = av_malloc(sizeof(SwsVector));
1222 vec->length = length;
1223 vec->coeff = av_malloc(sizeof(double) * length);
1229 SwsVector *sws_getGaussianVec(double variance, double quality)
1231 const int length= (int)(variance*quality + 0.5) | 1;
1233 double middle= (length-1)*0.5;
1234 SwsVector *vec= sws_allocVec(length);
1239 for (i=0; i<length; i++) {
1240 double dist= i-middle;
1241 vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);
1244 sws_normalizeVec(vec, 1.0);
1249 SwsVector *sws_getConstVec(double c, int length)
1252 SwsVector *vec= sws_allocVec(length);
1257 for (i=0; i<length; i++)
1263 SwsVector *sws_getIdentityVec(void)
1265 return sws_getConstVec(1.0, 1);
1268 static double sws_dcVec(SwsVector *a)
1273 for (i=0; i<a->length; i++)
1279 void sws_scaleVec(SwsVector *a, double scalar)
1283 for (i=0; i<a->length; i++)
1284 a->coeff[i]*= scalar;
1287 void sws_normalizeVec(SwsVector *a, double height)
1289 sws_scaleVec(a, height/sws_dcVec(a));
1292 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
1294 int length= a->length + b->length - 1;
1296 SwsVector *vec= sws_getConstVec(0.0, length);
1301 for (i=0; i<a->length; i++) {
1302 for (j=0; j<b->length; j++) {
1303 vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];
1310 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
1312 int length= FFMAX(a->length, b->length);
1314 SwsVector *vec= sws_getConstVec(0.0, length);
1319 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1320 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
1325 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
1327 int length= FFMAX(a->length, b->length);
1329 SwsVector *vec= sws_getConstVec(0.0, length);
1334 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1335 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
1340 /* shift left / or right if "shift" is negative */
1341 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
1343 int length= a->length + FFABS(shift)*2;
1345 SwsVector *vec= sws_getConstVec(0.0, length);
1350 for (i=0; i<a->length; i++) {
1351 vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
1357 void sws_shiftVec(SwsVector *a, int shift)
1359 SwsVector *shifted= sws_getShiftedVec(a, shift);
1361 a->coeff= shifted->coeff;
1362 a->length= shifted->length;
1366 void sws_addVec(SwsVector *a, SwsVector *b)
1368 SwsVector *sum= sws_sumVec(a, b);
1370 a->coeff= sum->coeff;
1371 a->length= sum->length;
1375 void sws_subVec(SwsVector *a, SwsVector *b)
1377 SwsVector *diff= sws_diffVec(a, b);
1379 a->coeff= diff->coeff;
1380 a->length= diff->length;
1384 void sws_convVec(SwsVector *a, SwsVector *b)
1386 SwsVector *conv= sws_getConvVec(a, b);
1388 a->coeff= conv->coeff;
1389 a->length= conv->length;
1393 SwsVector *sws_cloneVec(SwsVector *a)
1396 SwsVector *vec= sws_allocVec(a->length);
1401 for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];
1406 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
1413 for (i=0; i<a->length; i++)
1414 if (a->coeff[i]>max) max= a->coeff[i];
1416 for (i=0; i<a->length; i++)
1417 if (a->coeff[i]<min) min= a->coeff[i];
1421 for (i=0; i<a->length; i++) {
1422 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
1423 av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
1424 for (;x>0; x--) av_log(log_ctx, log_level, " ");
1425 av_log(log_ctx, log_level, "|\n");
1429 void sws_freeVec(SwsVector *a)
1432 av_freep(&a->coeff);
1437 void sws_freeFilter(SwsFilter *filter)
1439 if (!filter) return;
1441 if (filter->lumH) sws_freeVec(filter->lumH);
1442 if (filter->lumV) sws_freeVec(filter->lumV);
1443 if (filter->chrH) sws_freeVec(filter->chrH);
1444 if (filter->chrV) sws_freeVec(filter->chrV);
1448 void sws_freeContext(SwsContext *c)
1454 for (i=0; i<c->vLumBufSize; i++)
1455 av_freep(&c->lumPixBuf[i]);
1456 av_freep(&c->lumPixBuf);
1459 if (c->chrUPixBuf) {
1460 for (i=0; i<c->vChrBufSize; i++)
1461 av_freep(&c->chrUPixBuf[i]);
1462 av_freep(&c->chrUPixBuf);
1463 av_freep(&c->chrVPixBuf);
1466 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1467 for (i=0; i<c->vLumBufSize; i++)
1468 av_freep(&c->alpPixBuf[i]);
1469 av_freep(&c->alpPixBuf);
1472 av_freep(&c->vLumFilter);
1473 av_freep(&c->vChrFilter);
1474 av_freep(&c->hLumFilter);
1475 av_freep(&c->hChrFilter);
1477 av_freep(&c->vYCoeffsBank);
1478 av_freep(&c->vCCoeffsBank);
1481 av_freep(&c->vLumFilterPos);
1482 av_freep(&c->vChrFilterPos);
1483 av_freep(&c->hLumFilterPos);
1484 av_freep(&c->hChrFilterPos);
1487 #ifdef MAP_ANONYMOUS
1488 if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);
1489 if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);
1490 #elif HAVE_VIRTUALALLOC
1491 if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);
1492 if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);
1494 av_free(c->lumMmx2FilterCode);
1495 av_free(c->chrMmx2FilterCode);
1497 c->lumMmx2FilterCode=NULL;
1498 c->chrMmx2FilterCode=NULL;
1499 #endif /* HAVE_MMX */
1501 av_freep(&c->yuvTable);
1502 av_freep(&c->formatConvBuffer);
1507 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
1508 int srcW, int srcH, enum PixelFormat srcFormat,
1509 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
1510 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
1512 static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};
1515 param = default_param;
1518 (context->srcW != srcW ||
1519 context->srcH != srcH ||
1520 context->srcFormat != srcFormat ||
1521 context->dstW != dstW ||
1522 context->dstH != dstH ||
1523 context->dstFormat != dstFormat ||
1524 context->flags != flags ||
1525 context->param[0] != param[0] ||
1526 context->param[1] != param[1])) {
1527 sws_freeContext(context);
1532 if (!(context = sws_alloc_context()))
1534 context->srcW = srcW;
1535 context->srcH = srcH;
1536 context->srcRange = handle_jpeg(&srcFormat);
1537 context->srcFormat = srcFormat;
1538 context->dstW = dstW;
1539 context->dstH = dstH;
1540 context->dstRange = handle_jpeg(&dstFormat);
1541 context->dstFormat = dstFormat;
1542 context->flags = flags;
1543 context->param[0] = param[0];
1544 context->param[1] = param[1];
1545 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);
1546 if (sws_init_context(context, srcFilter, dstFilter) < 0) {
1547 sws_freeContext(context);