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
21 #define _SVID_SOURCE //needed for MAP_ANONYMOUS
30 #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
31 #define MAP_ANONYMOUS MAP_ANON
35 #define WIN32_LEAN_AND_MEAN
39 #include "swscale_internal.h"
41 #include "libavutil/intreadwrite.h"
42 #include "libavutil/x86_cpu.h"
43 #include "libavutil/cpu.h"
44 #include "libavutil/avutil.h"
45 #include "libavutil/bswap.h"
46 #include "libavutil/opt.h"
47 #include "libavutil/pixdesc.h"
49 unsigned swscale_version(void)
51 return LIBSWSCALE_VERSION_INT;
54 const char *swscale_configuration(void)
56 return LIBAV_CONFIGURATION;
59 const char *swscale_license(void)
61 #define LICENSE_PREFIX "libswscale license: "
62 return LICENSE_PREFIX LIBAV_LICENSE + sizeof(LICENSE_PREFIX) - 1;
65 #define RET 0xC3 //near return opcode for x86
67 typedef struct FormatEntry {
68 int is_supported_in, is_supported_out;
71 const static FormatEntry format_entries[PIX_FMT_NB] = {
72 [PIX_FMT_YUV420P] = { 1 , 1 },
73 [PIX_FMT_YUYV422] = { 1 , 1 },
74 [PIX_FMT_RGB24] = { 1 , 1 },
75 [PIX_FMT_BGR24] = { 1 , 1 },
76 [PIX_FMT_YUV422P] = { 1 , 1 },
77 [PIX_FMT_YUV444P] = { 1 , 1 },
78 [PIX_FMT_YUV410P] = { 1 , 1 },
79 [PIX_FMT_YUV411P] = { 1 , 1 },
80 [PIX_FMT_GRAY8] = { 1 , 1 },
81 [PIX_FMT_MONOWHITE] = { 1 , 1 },
82 [PIX_FMT_MONOBLACK] = { 1 , 1 },
83 [PIX_FMT_PAL8] = { 1 , 0 },
84 [PIX_FMT_YUVJ420P] = { 1 , 1 },
85 [PIX_FMT_YUVJ422P] = { 1 , 1 },
86 [PIX_FMT_YUVJ444P] = { 1 , 1 },
87 [PIX_FMT_UYVY422] = { 1 , 1 },
88 [PIX_FMT_UYYVYY411] = { 0 , 0 },
89 [PIX_FMT_BGR8] = { 1 , 1 },
90 [PIX_FMT_BGR4] = { 0 , 1 },
91 [PIX_FMT_BGR4_BYTE] = { 1 , 1 },
92 [PIX_FMT_RGB8] = { 1 , 1 },
93 [PIX_FMT_RGB4] = { 0 , 1 },
94 [PIX_FMT_RGB4_BYTE] = { 1 , 1 },
95 [PIX_FMT_NV12] = { 1 , 1 },
96 [PIX_FMT_NV21] = { 1 , 1 },
97 [PIX_FMT_ARGB] = { 1 , 1 },
98 [PIX_FMT_RGBA] = { 1 , 1 },
99 [PIX_FMT_ABGR] = { 1 , 1 },
100 [PIX_FMT_BGRA] = { 1 , 1 },
101 [PIX_FMT_GRAY16BE] = { 1 , 1 },
102 [PIX_FMT_GRAY16LE] = { 1 , 1 },
103 [PIX_FMT_YUV440P] = { 1 , 1 },
104 [PIX_FMT_YUVJ440P] = { 1 , 1 },
105 [PIX_FMT_YUVA420P] = { 1 , 1 },
106 [PIX_FMT_RGB48BE] = { 1 , 1 },
107 [PIX_FMT_RGB48LE] = { 1 , 1 },
108 [PIX_FMT_RGB565BE] = { 1 , 1 },
109 [PIX_FMT_RGB565LE] = { 1 , 1 },
110 [PIX_FMT_RGB555BE] = { 1 , 1 },
111 [PIX_FMT_RGB555LE] = { 1 , 1 },
112 [PIX_FMT_BGR565BE] = { 1 , 1 },
113 [PIX_FMT_BGR565LE] = { 1 , 1 },
114 [PIX_FMT_BGR555BE] = { 1 , 1 },
115 [PIX_FMT_BGR555LE] = { 1 , 1 },
116 [PIX_FMT_YUV420P16LE] = { 1 , 1 },
117 [PIX_FMT_YUV420P16BE] = { 1 , 1 },
118 [PIX_FMT_YUV422P16LE] = { 1 , 1 },
119 [PIX_FMT_YUV422P16BE] = { 1 , 1 },
120 [PIX_FMT_YUV444P16LE] = { 1 , 1 },
121 [PIX_FMT_YUV444P16BE] = { 1 , 1 },
122 [PIX_FMT_RGB444LE] = { 0 , 1 },
123 [PIX_FMT_RGB444BE] = { 0 , 1 },
124 [PIX_FMT_BGR444LE] = { 0 , 1 },
125 [PIX_FMT_BGR444BE] = { 0 , 1 },
126 [PIX_FMT_Y400A] = { 1 , 0 },
127 [PIX_FMT_BGR48BE] = { 1 , 1 },
128 [PIX_FMT_BGR48LE] = { 1 , 1 },
129 [PIX_FMT_YUV420P9BE] = { 1 , 1 },
130 [PIX_FMT_YUV420P9LE] = { 1 , 1 },
131 [PIX_FMT_YUV420P10BE] = { 1 , 1 },
132 [PIX_FMT_YUV420P10LE] = { 1 , 1 },
133 [PIX_FMT_YUV422P9BE] = { 1 , 1 },
134 [PIX_FMT_YUV422P9LE] = { 1 , 1 },
135 [PIX_FMT_YUV422P10BE] = { 1 , 1 },
136 [PIX_FMT_YUV422P10LE] = { 1 , 1 },
137 [PIX_FMT_YUV444P9BE] = { 1 , 1 },
138 [PIX_FMT_YUV444P9LE] = { 1 , 1 },
139 [PIX_FMT_YUV444P10BE] = { 1 , 1 },
140 [PIX_FMT_YUV444P10LE] = { 1 , 1 },
141 [PIX_FMT_GBRP] = { 1 , 0 },
142 [PIX_FMT_GBRP9LE] = { 1 , 0 },
143 [PIX_FMT_GBRP9BE] = { 1 , 0 },
144 [PIX_FMT_GBRP10LE] = { 1 , 0 },
145 [PIX_FMT_GBRP10BE] = { 1 , 0 },
146 [PIX_FMT_GBRP16LE] = { 1 , 0 },
147 [PIX_FMT_GBRP16BE] = { 1 , 0 },
150 int sws_isSupportedInput(enum PixelFormat pix_fmt)
152 return (unsigned)pix_fmt < PIX_FMT_NB ?
153 format_entries[pix_fmt].is_supported_in : 0;
156 int sws_isSupportedOutput(enum PixelFormat pix_fmt)
158 return (unsigned)pix_fmt < PIX_FMT_NB ?
159 format_entries[pix_fmt].is_supported_out : 0;
162 extern const int32_t ff_yuv2rgb_coeffs[8][4];
164 const char *sws_format_name(enum PixelFormat format)
166 if ((unsigned)format < PIX_FMT_NB && av_pix_fmt_descriptors[format].name)
167 return av_pix_fmt_descriptors[format].name;
169 return "Unknown format";
172 static double getSplineCoeff(double a, double b, double c, double d, double 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, int cpu_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 emms_c(); //FIXME this should not be required but it IS (even for non-MMX versions)
197 // NOTE: the +3 is for the MMX(+1)/SSE(+3) scaler which reads over the end
198 FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW+3)*sizeof(int16_t), fail);
200 if (FFABS(xInc - 0x10000) <10) { // unscaled
203 FF_ALLOCZ_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
205 for (i=0; i<dstW; i++) {
206 filter[i*filterSize]= fone;
210 } else if (flags&SWS_POINT) { // lame looking point sampling mode
214 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
216 xDstInSrc= xInc/2 - 0x8000;
217 for (i=0; i<dstW; i++) {
218 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
224 } else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) { // bilinear upscale
228 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
230 xDstInSrc= xInc/2 - 0x8000;
231 for (i=0; i<dstW; i++) {
232 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
236 //bilinear upscale / linear interpolate / area averaging
237 for (j=0; j<filterSize; j++) {
238 int64_t coeff= fone - FFABS((xx<<16) - xDstInSrc)*(fone>>16);
239 if (coeff<0) coeff=0;
240 filter[i*filterSize + j]= coeff;
249 if (flags&SWS_BICUBIC) sizeFactor= 4;
250 else if (flags&SWS_X) sizeFactor= 8;
251 else if (flags&SWS_AREA) sizeFactor= 1; //downscale only, for upscale it is bilinear
252 else if (flags&SWS_GAUSS) sizeFactor= 8; // infinite ;)
253 else if (flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? ceil(2*param[0]) : 6;
254 else if (flags&SWS_SINC) sizeFactor= 20; // infinite ;)
255 else if (flags&SWS_SPLINE) sizeFactor= 20; // infinite ;)
256 else if (flags&SWS_BILINEAR) sizeFactor= 2;
258 sizeFactor= 0; //GCC warning killer
262 if (xInc <= 1<<16) filterSize= 1 + sizeFactor; // upscale
263 else filterSize= 1 + (sizeFactor*srcW + dstW - 1)/ dstW;
265 if (filterSize > srcW-2) filterSize=srcW-2;
267 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
269 xDstInSrc= xInc - 0x10000;
270 for (i=0; i<dstW; i++) {
271 int xx= (xDstInSrc - ((filterSize-2)<<16)) / (1<<17);
274 for (j=0; j<filterSize; j++) {
275 int64_t d= ((int64_t)FFABS((xx<<17) - xDstInSrc))<<13;
281 floatd= d * (1.0/(1<<30));
283 if (flags & SWS_BICUBIC) {
284 int64_t B= (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1<<24);
285 int64_t C= (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1<<24);
290 int64_t dd = (d * d) >> 30;
291 int64_t ddd = (dd * d) >> 30;
294 coeff = (12*(1<<24)-9*B-6*C)*ddd + (-18*(1<<24)+12*B+6*C)*dd + (6*(1<<24)-2*B)*(1<<30);
296 coeff = (-B-6*C)*ddd + (6*B+30*C)*dd + (-12*B-48*C)*d + (8*B+24*C)*(1<<30);
298 coeff *= fone>>(30+24);
300 /* else if (flags & SWS_X) {
301 double p= param ? param*0.01 : 0.3;
302 coeff = d ? sin(d*M_PI)/(d*M_PI) : 1.0;
303 coeff*= pow(2.0, - p*d*d);
305 else if (flags & SWS_X) {
306 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
310 c = cos(floatd*M_PI);
313 if (c<0.0) c= -pow(-c, A);
315 coeff= (c*0.5 + 0.5)*fone;
316 } else if (flags & SWS_AREA) {
317 int64_t d2= d - (1<<29);
318 if (d2*xInc < -(1LL<<(29+16))) coeff= 1.0 * (1LL<<(30+16));
319 else if (d2*xInc < (1LL<<(29+16))) coeff= -d2*xInc + (1LL<<(29+16));
321 coeff *= fone>>(30+16);
322 } else if (flags & SWS_GAUSS) {
323 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
324 coeff = (pow(2.0, - p*floatd*floatd))*fone;
325 } else if (flags & SWS_SINC) {
326 coeff = (d ? sin(floatd*M_PI)/(floatd*M_PI) : 1.0)*fone;
327 } else if (flags & SWS_LANCZOS) {
328 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
329 coeff = (d ? sin(floatd*M_PI)*sin(floatd*M_PI/p)/(floatd*floatd*M_PI*M_PI/p) : 1.0)*fone;
330 if (floatd>p) coeff=0;
331 } else if (flags & SWS_BILINEAR) {
333 if (coeff<0) coeff=0;
335 } else if (flags & SWS_SPLINE) {
336 double p=-2.196152422706632;
337 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, floatd) * fone;
339 coeff= 0.0; //GCC warning killer
343 filter[i*filterSize + j]= coeff;
350 /* apply src & dst Filter to filter -> filter2
353 assert(filterSize>0);
354 filter2Size= filterSize;
355 if (srcFilter) filter2Size+= srcFilter->length - 1;
356 if (dstFilter) filter2Size+= dstFilter->length - 1;
357 assert(filter2Size>0);
358 FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size*dstW*sizeof(*filter2), fail);
360 for (i=0; i<dstW; i++) {
364 for (k=0; k<srcFilter->length; k++) {
365 for (j=0; j<filterSize; j++)
366 filter2[i*filter2Size + k + j] += srcFilter->coeff[k]*filter[i*filterSize + j];
369 for (j=0; j<filterSize; j++)
370 filter2[i*filter2Size + j]= filter[i*filterSize + j];
374 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
378 /* try to reduce the filter-size (step1 find size and shift left) */
379 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
381 for (i=dstW-1; i>=0; i--) {
382 int min= filter2Size;
386 /* get rid of near zero elements on the left by shifting left */
387 for (j=0; j<filter2Size; j++) {
389 cutOff += FFABS(filter2[i*filter2Size]);
391 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
393 /* preserve monotonicity because the core can't handle the filter otherwise */
394 if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
396 // move filter coefficients left
397 for (k=1; k<filter2Size; k++)
398 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
399 filter2[i*filter2Size + k - 1]= 0;
404 /* count near zeros on the right */
405 for (j=filter2Size-1; j>0; j--) {
406 cutOff += FFABS(filter2[i*filter2Size + j]);
408 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
412 if (min>minFilterSize) minFilterSize= min;
415 if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) {
416 // we can handle the special case 4,
417 // so we don't want to go to the full 8
418 if (minFilterSize < 5)
421 // We really don't want to waste our time
422 // doing useless computation, so fall back on
423 // the scalar C code for very small filters.
424 // Vectorizing is worth it only if you have a
425 // decent-sized vector.
426 if (minFilterSize < 3)
430 if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
431 // special case for unscaled vertical filtering
432 if (minFilterSize == 1 && filterAlign == 2)
436 assert(minFilterSize > 0);
437 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
438 assert(filterSize > 0);
439 filter= av_malloc(filterSize*dstW*sizeof(*filter));
440 if (filterSize >= MAX_FILTER_SIZE*16/((flags&SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
442 *outFilterSize= filterSize;
444 if (flags&SWS_PRINT_INFO)
445 av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
446 /* try to reduce the filter-size (step2 reduce it) */
447 for (i=0; i<dstW; i++) {
450 for (j=0; j<filterSize; j++) {
451 if (j>=filter2Size) filter[i*filterSize + j]= 0;
452 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
453 if((flags & SWS_BITEXACT) && j>=minFilterSize)
454 filter[i*filterSize + j]= 0;
458 //FIXME try to align filterPos if possible
461 for (i=0; i<dstW; i++) {
463 if ((*filterPos)[i] < 0) {
464 // move filter coefficients left to compensate for filterPos
465 for (j=1; j<filterSize; j++) {
466 int left= FFMAX(j + (*filterPos)[i], 0);
467 filter[i*filterSize + left] += filter[i*filterSize + j];
468 filter[i*filterSize + j]=0;
473 if ((*filterPos)[i] + filterSize > srcW) {
474 int shift= (*filterPos)[i] + filterSize - srcW;
475 // move filter coefficients right to compensate for filterPos
476 for (j=filterSize-2; j>=0; j--) {
477 int right= FFMIN(j + shift, filterSize-1);
478 filter[i*filterSize +right] += filter[i*filterSize +j];
479 filter[i*filterSize +j]=0;
481 (*filterPos)[i]= srcW - filterSize;
485 // Note the +1 is for the MMX scaler which reads over the end
486 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
487 FF_ALLOCZ_OR_GOTO(NULL, *outFilter, *outFilterSize*(dstW+3)*sizeof(int16_t), fail);
489 /* normalize & store in outFilter */
490 for (i=0; i<dstW; i++) {
495 for (j=0; j<filterSize; j++) {
496 sum+= filter[i*filterSize + j];
498 sum= (sum + one/2)/ one;
499 for (j=0; j<*outFilterSize; j++) {
500 int64_t v= filter[i*filterSize + j] + error;
501 int intV= ROUNDED_DIV(v, sum);
502 (*outFilter)[i*(*outFilterSize) + j]= intV;
507 (*filterPos)[dstW+0] =
508 (*filterPos)[dstW+1] =
509 (*filterPos)[dstW+2] = (*filterPos)[dstW-1]; // the MMX/SSE scaler will read over the end
510 for (i=0; i<*outFilterSize; i++) {
511 int k= (dstW - 1) * (*outFilterSize) + i;
512 (*outFilter)[k + 1 * (*outFilterSize)] =
513 (*outFilter)[k + 2 * (*outFilterSize)] =
514 (*outFilter)[k + 3 * (*outFilterSize)] = (*outFilter)[k];
525 static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode, int16_t *filter, int32_t *filterPos, int numSplits)
528 x86_reg imm8OfPShufW1A;
529 x86_reg imm8OfPShufW2A;
530 x86_reg fragmentLengthA;
532 x86_reg imm8OfPShufW1B;
533 x86_reg imm8OfPShufW2B;
534 x86_reg fragmentLengthB;
539 // create an optimized horizontal scaling routine
540 /* This scaler is made of runtime-generated MMX2 code using specially
541 * tuned pshufw instructions. For every four output pixels, if four
542 * input pixels are enough for the fast bilinear scaling, then a chunk
543 * of fragmentB is used. If five input pixels are needed, then a chunk
544 * of fragmentA is used.
553 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
554 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
555 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
556 "punpcklbw %%mm7, %%mm1 \n\t"
557 "punpcklbw %%mm7, %%mm0 \n\t"
558 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
560 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
562 "psubw %%mm1, %%mm0 \n\t"
563 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
564 "pmullw %%mm3, %%mm0 \n\t"
565 "psllw $7, %%mm1 \n\t"
566 "paddw %%mm1, %%mm0 \n\t"
568 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
570 "add $8, %%"REG_a" \n\t"
574 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
575 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
576 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
581 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
585 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
586 "=r" (fragmentLengthA)
593 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
594 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
595 "punpcklbw %%mm7, %%mm0 \n\t"
596 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
598 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
600 "psubw %%mm1, %%mm0 \n\t"
601 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
602 "pmullw %%mm3, %%mm0 \n\t"
603 "psllw $7, %%mm1 \n\t"
604 "paddw %%mm1, %%mm0 \n\t"
606 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
608 "add $8, %%"REG_a" \n\t"
612 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
613 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
614 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
619 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
623 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
624 "=r" (fragmentLengthB)
627 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
630 for (i=0; i<dstW/numSplits; i++) {
635 int b=((xpos+xInc)>>16) - xx;
636 int c=((xpos+xInc*2)>>16) - xx;
637 int d=((xpos+xInc*3)>>16) - xx;
639 uint8_t *fragment = (d+1<4) ? fragmentB : fragmentA;
640 x86_reg imm8OfPShufW1 = (d+1<4) ? imm8OfPShufW1B : imm8OfPShufW1A;
641 x86_reg imm8OfPShufW2 = (d+1<4) ? imm8OfPShufW2B : imm8OfPShufW2A;
642 x86_reg fragmentLength = (d+1<4) ? fragmentLengthB : fragmentLengthA;
643 int maxShift= 3-(d+inc);
647 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
648 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
649 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
650 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
653 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
655 filterCode[fragmentPos + imm8OfPShufW1]=
656 (a+inc) | ((b+inc)<<2) | ((c+inc)<<4) | ((d+inc)<<6);
657 filterCode[fragmentPos + imm8OfPShufW2]=
658 a | (b<<2) | (c<<4) | (d<<6);
660 if (i+4-inc>=dstW) shift=maxShift; //avoid overread
661 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
663 if (shift && i>=shift) {
664 filterCode[fragmentPos + imm8OfPShufW1]+= 0x55*shift;
665 filterCode[fragmentPos + imm8OfPShufW2]+= 0x55*shift;
666 filterPos[i/2]-=shift;
670 fragmentPos+= fragmentLength;
673 filterCode[fragmentPos]= RET;
678 filterPos[((i/2)+1)&(~1)]= xpos>>16; // needed to jump to the next part
680 return fragmentPos + 1;
682 #endif /* HAVE_MMX2 */
684 static void getSubSampleFactors(int *h, int *v, enum PixelFormat format)
686 *h = av_pix_fmt_descriptors[format].log2_chroma_w;
687 *v = av_pix_fmt_descriptors[format].log2_chroma_h;
690 int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],
691 int srcRange, const int table[4], int dstRange,
692 int brightness, int contrast, int saturation)
694 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
695 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
697 c->brightness= brightness;
698 c->contrast = contrast;
699 c->saturation= saturation;
700 c->srcRange = srcRange;
701 c->dstRange = dstRange;
702 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
704 c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->dstFormat]);
705 c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->srcFormat]);
707 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
710 if (HAVE_ALTIVEC && av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)
711 ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness, contrast, saturation);
715 int sws_getColorspaceDetails(struct SwsContext *c, int **inv_table,
716 int *srcRange, int **table, int *dstRange,
717 int *brightness, int *contrast, int *saturation)
719 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
721 *inv_table = c->srcColorspaceTable;
722 *table = c->dstColorspaceTable;
723 *srcRange = c->srcRange;
724 *dstRange = c->dstRange;
725 *brightness= c->brightness;
726 *contrast = c->contrast;
727 *saturation= c->saturation;
732 static int handle_jpeg(enum PixelFormat *format)
735 case PIX_FMT_YUVJ420P: *format = PIX_FMT_YUV420P; return 1;
736 case PIX_FMT_YUVJ422P: *format = PIX_FMT_YUV422P; return 1;
737 case PIX_FMT_YUVJ444P: *format = PIX_FMT_YUV444P; return 1;
738 case PIX_FMT_YUVJ440P: *format = PIX_FMT_YUV440P; return 1;
743 SwsContext *sws_alloc_context(void)
745 SwsContext *c= av_mallocz(sizeof(SwsContext));
747 c->av_class = &sws_context_class;
748 av_opt_set_defaults(c);
753 int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)
756 int usesVFilter, usesHFilter;
758 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
763 int dst_stride = FFALIGN(dstW * sizeof(int16_t) + 16, 16), dst_stride_px = dst_stride >> 1;
764 int flags, cpu_flags;
765 enum PixelFormat srcFormat= c->srcFormat;
766 enum PixelFormat dstFormat= c->dstFormat;
768 cpu_flags = av_get_cpu_flags();
771 if (!rgb15to16) sws_rgb2rgb_init();
773 unscaled = (srcW == dstW && srcH == dstH);
775 if (!sws_isSupportedInput(srcFormat)) {
776 av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n", sws_format_name(srcFormat));
777 return AVERROR(EINVAL);
779 if (!sws_isSupportedOutput(dstFormat)) {
780 av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n", sws_format_name(dstFormat));
781 return AVERROR(EINVAL);
784 i= flags & ( SWS_POINT
795 if(!i || (i & (i-1))) {
796 av_log(c, AV_LOG_ERROR, "Exactly one scaler algorithm must be chosen\n");
797 return AVERROR(EINVAL);
800 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
801 av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",
802 srcW, srcH, dstW, dstH);
803 return AVERROR(EINVAL);
806 if (!dstFilter) dstFilter= &dummyFilter;
807 if (!srcFilter) srcFilter= &dummyFilter;
809 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
810 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
811 c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[dstFormat]);
812 c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[srcFormat]);
813 c->vRounder= 4* 0x0001000100010001ULL;
815 usesVFilter = (srcFilter->lumV && srcFilter->lumV->length>1) ||
816 (srcFilter->chrV && srcFilter->chrV->length>1) ||
817 (dstFilter->lumV && dstFilter->lumV->length>1) ||
818 (dstFilter->chrV && dstFilter->chrV->length>1);
819 usesHFilter = (srcFilter->lumH && srcFilter->lumH->length>1) ||
820 (srcFilter->chrH && srcFilter->chrH->length>1) ||
821 (dstFilter->lumH && dstFilter->lumH->length>1) ||
822 (dstFilter->chrH && dstFilter->chrH->length>1);
824 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
825 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
827 // reuse chroma for 2 pixels RGB/BGR unless user wants full chroma interpolation
828 if (flags & SWS_FULL_CHR_H_INT &&
829 dstFormat != PIX_FMT_RGBA &&
830 dstFormat != PIX_FMT_ARGB &&
831 dstFormat != PIX_FMT_BGRA &&
832 dstFormat != PIX_FMT_ABGR &&
833 dstFormat != PIX_FMT_RGB24 &&
834 dstFormat != PIX_FMT_BGR24) {
835 av_log(c, AV_LOG_ERROR,
836 "full chroma interpolation for destination format '%s' not yet implemented\n",
837 sws_format_name(dstFormat));
838 flags &= ~SWS_FULL_CHR_H_INT;
841 if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
843 // drop some chroma lines if the user wants it
844 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
845 c->chrSrcVSubSample+= c->vChrDrop;
847 // drop every other pixel for chroma calculation unless user wants full chroma
848 if (isAnyRGB(srcFormat) && !(flags&SWS_FULL_CHR_H_INP)
849 && srcFormat!=PIX_FMT_RGB8 && srcFormat!=PIX_FMT_BGR8
850 && srcFormat!=PIX_FMT_RGB4 && srcFormat!=PIX_FMT_BGR4
851 && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE
852 && ((dstW>>c->chrDstHSubSample) <= (srcW>>1) || (flags&SWS_FAST_BILINEAR)))
853 c->chrSrcHSubSample=1;
855 // Note the -((-x)>>y) is so that we always round toward +inf.
856 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
857 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
858 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
859 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
861 /* unscaled special cases */
862 if (unscaled && !usesHFilter && !usesVFilter && (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
863 ff_get_unscaled_swscale(c);
866 if (flags&SWS_PRINT_INFO)
867 av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
868 sws_format_name(srcFormat), sws_format_name(dstFormat));
873 c->srcBpc = 1 + av_pix_fmt_descriptors[srcFormat].comp[0].depth_minus1;
876 c->dstBpc = 1 + av_pix_fmt_descriptors[dstFormat].comp[0].depth_minus1;
881 FF_ALLOC_OR_GOTO(c, c->formatConvBuffer,
882 (FFALIGN(srcW, 16) * 2 * FFALIGN(c->srcBpc, 8) >> 3) + 16,
884 if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2 && c->srcBpc == 8 && c->dstBpc <= 10) {
885 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
886 if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR)) {
887 if (flags&SWS_PRINT_INFO)
888 av_log(c, AV_LOG_INFO, "output width is not a multiple of 32 -> no MMX2 scaler\n");
890 if (usesHFilter) c->canMMX2BeUsed=0;
895 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
896 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
898 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
899 // but only for the FAST_BILINEAR mode otherwise do correct scaling
900 // n-2 is the last chrominance sample available
901 // this is not perfect, but no one should notice the difference, the more correct variant
902 // would be like the vertical one, but that would require some special code for the
903 // first and last pixel
904 if (flags&SWS_FAST_BILINEAR) {
905 if (c->canMMX2BeUsed) {
909 //we don't use the x86 asm scaler if MMX is available
910 else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
911 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
912 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
916 /* precalculate horizontal scaler filter coefficients */
919 // can't downscale !!!
920 if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {
921 c->lumMmx2FilterCodeSize = initMMX2HScaler( dstW, c->lumXInc, NULL, NULL, NULL, 8);
922 c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc, NULL, NULL, NULL, 4);
925 c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
926 c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
927 #elif HAVE_VIRTUALALLOC
928 c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
929 c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
931 c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);
932 c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);
935 if (!c->lumMmx2FilterCode || !c->chrMmx2FilterCode)
936 return AVERROR(ENOMEM);
937 FF_ALLOCZ_OR_GOTO(c, c->hLumFilter , (dstW /8+8)*sizeof(int16_t), fail);
938 FF_ALLOCZ_OR_GOTO(c, c->hChrFilter , (c->chrDstW /4+8)*sizeof(int16_t), fail);
939 FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW /2/8+8)*sizeof(int32_t), fail);
940 FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW/2/4+8)*sizeof(int32_t), fail);
942 initMMX2HScaler( dstW, c->lumXInc, c->lumMmx2FilterCode, c->hLumFilter, c->hLumFilterPos, 8);
943 initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode, c->hChrFilter, c->hChrFilterPos, 4);
946 mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
947 mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
950 #endif /* HAVE_MMX2 */
952 const int filterAlign=
953 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :
954 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
957 if (initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
958 srcW , dstW, filterAlign, 1<<14,
959 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags, cpu_flags,
960 srcFilter->lumH, dstFilter->lumH, c->param) < 0)
962 if (initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
963 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
964 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, cpu_flags,
965 srcFilter->chrH, dstFilter->chrH, c->param) < 0)
968 } // initialize horizontal stuff
970 /* precalculate vertical scaler filter coefficients */
972 const int filterAlign=
973 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 2 :
974 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
977 if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
978 srcH , dstH, filterAlign, (1<<12),
979 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags, cpu_flags,
980 srcFilter->lumV, dstFilter->lumV, c->param) < 0)
982 if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
983 c->chrSrcH, c->chrDstH, filterAlign, (1<<12),
984 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, cpu_flags,
985 srcFilter->chrV, dstFilter->chrV, c->param) < 0)
989 FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof (vector signed short)*c->vLumFilterSize*c->dstH, fail);
990 FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH, fail);
992 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
994 short *p = (short *)&c->vYCoeffsBank[i];
996 p[j] = c->vLumFilter[i];
999 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
1001 short *p = (short *)&c->vCCoeffsBank[i];
1003 p[j] = c->vChrFilter[i];
1008 // calculate buffer sizes so that they won't run out while handling these damn slices
1009 c->vLumBufSize= c->vLumFilterSize;
1010 c->vChrBufSize= c->vChrFilterSize;
1011 for (i=0; i<dstH; i++) {
1012 int chrI= i*c->chrDstH / dstH;
1013 int nextSlice= FFMAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
1014 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
1016 nextSlice>>= c->chrSrcVSubSample;
1017 nextSlice<<= c->chrSrcVSubSample;
1018 if (c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
1019 c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
1020 if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
1021 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
1024 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
1025 // allocate several megabytes to handle all possible cases)
1026 FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
1027 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
1028 FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
1029 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
1030 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
1031 //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)
1032 /* align at 16 bytes for AltiVec */
1033 for (i=0; i<c->vLumBufSize; i++) {
1034 FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], dst_stride+16, fail);
1035 c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];
1037 // 64 / (c->dstBpc & ~7) is the same as 16 / sizeof(scaling_intermediate)
1038 c->uv_off_px = dst_stride_px + 64 / (c->dstBpc &~ 7);
1039 c->uv_off_byte = dst_stride + 16;
1040 for (i=0; i<c->vChrBufSize; i++) {
1041 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i+c->vChrBufSize], dst_stride*2+32, fail);
1042 c->chrUPixBuf[i] = c->chrUPixBuf[i+c->vChrBufSize];
1043 c->chrVPixBuf[i] = c->chrVPixBuf[i+c->vChrBufSize] = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;
1045 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
1046 for (i=0; i<c->vLumBufSize; i++) {
1047 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], dst_stride+16, fail);
1048 c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];
1051 //try to avoid drawing green stuff between the right end and the stride end
1052 for (i=0; i<c->vChrBufSize; i++)
1053 memset(c->chrUPixBuf[i], 64, dst_stride*2+1);
1055 assert(c->chrDstH <= dstH);
1057 if (flags&SWS_PRINT_INFO) {
1058 if (flags&SWS_FAST_BILINEAR) av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
1059 else if (flags&SWS_BILINEAR) av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
1060 else if (flags&SWS_BICUBIC) av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
1061 else if (flags&SWS_X) av_log(c, AV_LOG_INFO, "Experimental scaler, ");
1062 else if (flags&SWS_POINT) av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
1063 else if (flags&SWS_AREA) av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
1064 else if (flags&SWS_BICUBLIN) av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
1065 else if (flags&SWS_GAUSS) av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
1066 else if (flags&SWS_SINC) av_log(c, AV_LOG_INFO, "Sinc scaler, ");
1067 else if (flags&SWS_LANCZOS) av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
1068 else if (flags&SWS_SPLINE) av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
1069 else av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
1071 av_log(c, AV_LOG_INFO, "from %s to %s%s ",
1072 sws_format_name(srcFormat),
1074 dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ||
1075 dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||
1076 dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ? "dithered " : "",
1080 sws_format_name(dstFormat));
1082 if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2) av_log(c, AV_LOG_INFO, "using MMX2\n");
1083 else if (HAVE_AMD3DNOW && cpu_flags & AV_CPU_FLAG_3DNOW) av_log(c, AV_LOG_INFO, "using 3DNOW\n");
1084 else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) av_log(c, AV_LOG_INFO, "using MMX\n");
1085 else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) av_log(c, AV_LOG_INFO, "using AltiVec\n");
1086 else av_log(c, AV_LOG_INFO, "using C\n");
1088 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
1089 av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1090 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
1091 av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1092 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
1095 c->swScale= ff_getSwsFunc(c);
1097 fail: //FIXME replace things by appropriate error codes
1101 #if FF_API_SWS_GETCONTEXT
1102 SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,
1103 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
1104 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
1108 if(!(c=sws_alloc_context()))
1116 c->srcRange = handle_jpeg(&srcFormat);
1117 c->dstRange = handle_jpeg(&dstFormat);
1118 c->srcFormat= srcFormat;
1119 c->dstFormat= dstFormat;
1122 c->param[0] = param[0];
1123 c->param[1] = param[1];
1125 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);
1127 if(sws_init_context(c, srcFilter, dstFilter) < 0){
1136 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
1137 float lumaSharpen, float chromaSharpen,
1138 float chromaHShift, float chromaVShift,
1141 SwsFilter *filter= av_malloc(sizeof(SwsFilter));
1145 if (lumaGBlur!=0.0) {
1146 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
1147 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
1149 filter->lumH= sws_getIdentityVec();
1150 filter->lumV= sws_getIdentityVec();
1153 if (chromaGBlur!=0.0) {
1154 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
1155 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
1157 filter->chrH= sws_getIdentityVec();
1158 filter->chrV= sws_getIdentityVec();
1161 if (chromaSharpen!=0.0) {
1162 SwsVector *id= sws_getIdentityVec();
1163 sws_scaleVec(filter->chrH, -chromaSharpen);
1164 sws_scaleVec(filter->chrV, -chromaSharpen);
1165 sws_addVec(filter->chrH, id);
1166 sws_addVec(filter->chrV, id);
1170 if (lumaSharpen!=0.0) {
1171 SwsVector *id= sws_getIdentityVec();
1172 sws_scaleVec(filter->lumH, -lumaSharpen);
1173 sws_scaleVec(filter->lumV, -lumaSharpen);
1174 sws_addVec(filter->lumH, id);
1175 sws_addVec(filter->lumV, id);
1179 if (chromaHShift != 0.0)
1180 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
1182 if (chromaVShift != 0.0)
1183 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
1185 sws_normalizeVec(filter->chrH, 1.0);
1186 sws_normalizeVec(filter->chrV, 1.0);
1187 sws_normalizeVec(filter->lumH, 1.0);
1188 sws_normalizeVec(filter->lumV, 1.0);
1190 if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
1191 if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
1196 SwsVector *sws_allocVec(int length)
1198 SwsVector *vec = av_malloc(sizeof(SwsVector));
1201 vec->length = length;
1202 vec->coeff = av_malloc(sizeof(double) * length);
1208 SwsVector *sws_getGaussianVec(double variance, double quality)
1210 const int length= (int)(variance*quality + 0.5) | 1;
1212 double middle= (length-1)*0.5;
1213 SwsVector *vec= sws_allocVec(length);
1218 for (i=0; i<length; i++) {
1219 double dist= i-middle;
1220 vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);
1223 sws_normalizeVec(vec, 1.0);
1228 SwsVector *sws_getConstVec(double c, int length)
1231 SwsVector *vec= sws_allocVec(length);
1236 for (i=0; i<length; i++)
1242 SwsVector *sws_getIdentityVec(void)
1244 return sws_getConstVec(1.0, 1);
1247 static double sws_dcVec(SwsVector *a)
1252 for (i=0; i<a->length; i++)
1258 void sws_scaleVec(SwsVector *a, double scalar)
1262 for (i=0; i<a->length; i++)
1263 a->coeff[i]*= scalar;
1266 void sws_normalizeVec(SwsVector *a, double height)
1268 sws_scaleVec(a, height/sws_dcVec(a));
1271 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
1273 int length= a->length + b->length - 1;
1275 SwsVector *vec= sws_getConstVec(0.0, length);
1280 for (i=0; i<a->length; i++) {
1281 for (j=0; j<b->length; j++) {
1282 vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];
1289 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
1291 int length= FFMAX(a->length, b->length);
1293 SwsVector *vec= sws_getConstVec(0.0, length);
1298 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1299 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
1304 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
1306 int length= FFMAX(a->length, b->length);
1308 SwsVector *vec= sws_getConstVec(0.0, length);
1313 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1314 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
1319 /* shift left / or right if "shift" is negative */
1320 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
1322 int length= a->length + FFABS(shift)*2;
1324 SwsVector *vec= sws_getConstVec(0.0, length);
1329 for (i=0; i<a->length; i++) {
1330 vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
1336 void sws_shiftVec(SwsVector *a, int shift)
1338 SwsVector *shifted= sws_getShiftedVec(a, shift);
1340 a->coeff= shifted->coeff;
1341 a->length= shifted->length;
1345 void sws_addVec(SwsVector *a, SwsVector *b)
1347 SwsVector *sum= sws_sumVec(a, b);
1349 a->coeff= sum->coeff;
1350 a->length= sum->length;
1354 void sws_subVec(SwsVector *a, SwsVector *b)
1356 SwsVector *diff= sws_diffVec(a, b);
1358 a->coeff= diff->coeff;
1359 a->length= diff->length;
1363 void sws_convVec(SwsVector *a, SwsVector *b)
1365 SwsVector *conv= sws_getConvVec(a, b);
1367 a->coeff= conv->coeff;
1368 a->length= conv->length;
1372 SwsVector *sws_cloneVec(SwsVector *a)
1375 SwsVector *vec= sws_allocVec(a->length);
1380 for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];
1385 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
1392 for (i=0; i<a->length; i++)
1393 if (a->coeff[i]>max) max= a->coeff[i];
1395 for (i=0; i<a->length; i++)
1396 if (a->coeff[i]<min) min= a->coeff[i];
1400 for (i=0; i<a->length; i++) {
1401 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
1402 av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
1403 for (;x>0; x--) av_log(log_ctx, log_level, " ");
1404 av_log(log_ctx, log_level, "|\n");
1408 void sws_freeVec(SwsVector *a)
1411 av_freep(&a->coeff);
1416 void sws_freeFilter(SwsFilter *filter)
1418 if (!filter) return;
1420 if (filter->lumH) sws_freeVec(filter->lumH);
1421 if (filter->lumV) sws_freeVec(filter->lumV);
1422 if (filter->chrH) sws_freeVec(filter->chrH);
1423 if (filter->chrV) sws_freeVec(filter->chrV);
1427 void sws_freeContext(SwsContext *c)
1433 for (i=0; i<c->vLumBufSize; i++)
1434 av_freep(&c->lumPixBuf[i]);
1435 av_freep(&c->lumPixBuf);
1438 if (c->chrUPixBuf) {
1439 for (i=0; i<c->vChrBufSize; i++)
1440 av_freep(&c->chrUPixBuf[i]);
1441 av_freep(&c->chrUPixBuf);
1442 av_freep(&c->chrVPixBuf);
1445 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1446 for (i=0; i<c->vLumBufSize; i++)
1447 av_freep(&c->alpPixBuf[i]);
1448 av_freep(&c->alpPixBuf);
1451 av_freep(&c->vLumFilter);
1452 av_freep(&c->vChrFilter);
1453 av_freep(&c->hLumFilter);
1454 av_freep(&c->hChrFilter);
1456 av_freep(&c->vYCoeffsBank);
1457 av_freep(&c->vCCoeffsBank);
1460 av_freep(&c->vLumFilterPos);
1461 av_freep(&c->vChrFilterPos);
1462 av_freep(&c->hLumFilterPos);
1463 av_freep(&c->hChrFilterPos);
1466 #ifdef MAP_ANONYMOUS
1467 if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);
1468 if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);
1469 #elif HAVE_VIRTUALALLOC
1470 if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);
1471 if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);
1473 av_free(c->lumMmx2FilterCode);
1474 av_free(c->chrMmx2FilterCode);
1476 c->lumMmx2FilterCode=NULL;
1477 c->chrMmx2FilterCode=NULL;
1478 #endif /* HAVE_MMX */
1480 av_freep(&c->yuvTable);
1481 av_free(c->formatConvBuffer);
1486 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
1487 int srcW, int srcH, enum PixelFormat srcFormat,
1488 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
1489 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
1491 static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};
1494 param = default_param;
1497 (context->srcW != srcW ||
1498 context->srcH != srcH ||
1499 context->srcFormat != srcFormat ||
1500 context->dstW != dstW ||
1501 context->dstH != dstH ||
1502 context->dstFormat != dstFormat ||
1503 context->flags != flags ||
1504 context->param[0] != param[0] ||
1505 context->param[1] != param[1])) {
1506 sws_freeContext(context);
1511 if (!(context = sws_alloc_context()))
1513 context->srcW = srcW;
1514 context->srcH = srcH;
1515 context->srcRange = handle_jpeg(&srcFormat);
1516 context->srcFormat = srcFormat;
1517 context->dstW = dstW;
1518 context->dstH = dstH;
1519 context->dstRange = handle_jpeg(&dstFormat);
1520 context->dstFormat = dstFormat;
1521 context->flags = flags;
1522 context->param[0] = param[0];
1523 context->param[1] = param[1];
1524 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);
1525 if (sws_init_context(context, srcFilter, dstFilter) < 0) {
1526 sws_freeContext(context);