Add documentation for the returned value of sws_init_context().
[libav.git] / libswscale / utils.c
CommitLineData
a4388ebd
RP
1/*
2 * Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
3 *
4 * This file is part of FFmpeg.
5 *
819ee683
DB
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.
a4388ebd
RP
10 *
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
819ee683
DB
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
a4388ebd 15 *
819ee683
DB
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
a4388ebd 18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
a4388ebd
RP
19 */
20
21#define _SVID_SOURCE //needed for MAP_ANONYMOUS
262b6dba 22#define _DARWIN_C_SOURCE // needed for MAP_ANON
a4388ebd
RP
23#include <inttypes.h>
24#include <string.h>
25#include <math.h>
26#include <stdio.h>
27#include "config.h"
28#include <assert.h>
29#if HAVE_SYS_MMAN_H
30#include <sys/mman.h>
31#if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
32#define MAP_ANONYMOUS MAP_ANON
33#endif
34#endif
35#if HAVE_VIRTUALALLOC
36#define WIN32_LEAN_AND_MEAN
37#include <windows.h>
38#endif
39#include "swscale.h"
40#include "swscale_internal.h"
41#include "rgb2rgb.h"
42#include "libavutil/intreadwrite.h"
43#include "libavutil/x86_cpu.h"
44#include "libavutil/avutil.h"
45#include "libavutil/bswap.h"
46#include "libavutil/pixdesc.h"
47
48unsigned swscale_version(void)
49{
50 return LIBSWSCALE_VERSION_INT;
51}
52
53const char *swscale_configuration(void)
54{
55 return FFMPEG_CONFIGURATION;
56}
57
58const char *swscale_license(void)
59{
60#define LICENSE_PREFIX "libswscale license: "
61 return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
62}
63
64#define RET 0xC3 //near return opcode for x86
65
a4388ebd
RP
66#define isSupportedIn(x) ( \
67 (x)==PIX_FMT_YUV420P \
68 || (x)==PIX_FMT_YUVA420P \
69 || (x)==PIX_FMT_YUYV422 \
70 || (x)==PIX_FMT_UYVY422 \
71 || (x)==PIX_FMT_RGB48BE \
72 || (x)==PIX_FMT_RGB48LE \
73 || (x)==PIX_FMT_RGB32 \
74 || (x)==PIX_FMT_RGB32_1 \
75 || (x)==PIX_FMT_BGR24 \
76 || (x)==PIX_FMT_BGR565 \
77 || (x)==PIX_FMT_BGR555 \
78 || (x)==PIX_FMT_BGR32 \
79 || (x)==PIX_FMT_BGR32_1 \
80 || (x)==PIX_FMT_RGB24 \
81 || (x)==PIX_FMT_RGB565 \
82 || (x)==PIX_FMT_RGB555 \
83 || (x)==PIX_FMT_GRAY8 \
df0743e0 84 || (x)==PIX_FMT_Y400A \
a4388ebd
RP
85 || (x)==PIX_FMT_YUV410P \
86 || (x)==PIX_FMT_YUV440P \
87 || (x)==PIX_FMT_NV12 \
88 || (x)==PIX_FMT_NV21 \
89 || (x)==PIX_FMT_GRAY16BE \
90 || (x)==PIX_FMT_GRAY16LE \
91 || (x)==PIX_FMT_YUV444P \
92 || (x)==PIX_FMT_YUV422P \
93 || (x)==PIX_FMT_YUV411P \
af0e6621
SS
94 || (x)==PIX_FMT_YUVJ420P \
95 || (x)==PIX_FMT_YUVJ422P \
96 || (x)==PIX_FMT_YUVJ440P \
97 || (x)==PIX_FMT_YUVJ444P \
a4388ebd
RP
98 || (x)==PIX_FMT_PAL8 \
99 || (x)==PIX_FMT_BGR8 \
100 || (x)==PIX_FMT_RGB8 \
101 || (x)==PIX_FMT_BGR4_BYTE \
102 || (x)==PIX_FMT_RGB4_BYTE \
103 || (x)==PIX_FMT_YUV440P \
104 || (x)==PIX_FMT_MONOWHITE \
105 || (x)==PIX_FMT_MONOBLACK \
106 || (x)==PIX_FMT_YUV420P16LE \
107 || (x)==PIX_FMT_YUV422P16LE \
108 || (x)==PIX_FMT_YUV444P16LE \
109 || (x)==PIX_FMT_YUV420P16BE \
110 || (x)==PIX_FMT_YUV422P16BE \
111 || (x)==PIX_FMT_YUV444P16BE \
112 )
113
114int sws_isSupportedInput(enum PixelFormat pix_fmt)
115{
116 return isSupportedIn(pix_fmt);
117}
118
119#define isSupportedOut(x) ( \
120 (x)==PIX_FMT_YUV420P \
121 || (x)==PIX_FMT_YUVA420P \
122 || (x)==PIX_FMT_YUYV422 \
123 || (x)==PIX_FMT_UYVY422 \
124 || (x)==PIX_FMT_YUV444P \
125 || (x)==PIX_FMT_YUV422P \
126 || (x)==PIX_FMT_YUV411P \
af0e6621
SS
127 || (x)==PIX_FMT_YUVJ420P \
128 || (x)==PIX_FMT_YUVJ422P \
129 || (x)==PIX_FMT_YUVJ440P \
130 || (x)==PIX_FMT_YUVJ444P \
60222557 131 || isAnyRGB(x) \
a4388ebd
RP
132 || (x)==PIX_FMT_NV12 \
133 || (x)==PIX_FMT_NV21 \
134 || (x)==PIX_FMT_GRAY16BE \
135 || (x)==PIX_FMT_GRAY16LE \
136 || (x)==PIX_FMT_GRAY8 \
137 || (x)==PIX_FMT_YUV410P \
138 || (x)==PIX_FMT_YUV440P \
139 || (x)==PIX_FMT_YUV420P16LE \
140 || (x)==PIX_FMT_YUV422P16LE \
141 || (x)==PIX_FMT_YUV444P16LE \
142 || (x)==PIX_FMT_YUV420P16BE \
143 || (x)==PIX_FMT_YUV422P16BE \
144 || (x)==PIX_FMT_YUV444P16BE \
145 )
146
147int sws_isSupportedOutput(enum PixelFormat pix_fmt)
148{
149 return isSupportedOut(pix_fmt);
150}
151
a4388ebd
RP
152extern const int32_t ff_yuv2rgb_coeffs[8][4];
153
154const char *sws_format_name(enum PixelFormat format)
155{
156 if ((unsigned)format < PIX_FMT_NB && av_pix_fmt_descriptors[format].name)
157 return av_pix_fmt_descriptors[format].name;
158 else
159 return "Unknown format";
160}
161
162static double getSplineCoeff(double a, double b, double c, double d, double dist)
163{
164// printf("%f %f %f %f %f\n", a,b,c,d,dist);
165 if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
166 else return getSplineCoeff( 0.0,
167 b+ 2.0*c + 3.0*d,
168 c + 3.0*d,
169 -b- 3.0*c - 6.0*d,
170 dist-1.0);
171}
172
173static int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
174 int srcW, int dstW, int filterAlign, int one, int flags,
175 SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
176{
177 int i;
178 int filterSize;
179 int filter2Size;
180 int minFilterSize;
181 int64_t *filter=NULL;
182 int64_t *filter2=NULL;
183 const int64_t fone= 1LL<<54;
184 int ret= -1;
185#if ARCH_X86
186 if (flags & SWS_CPU_CAPS_MMX)
187 __asm__ volatile("emms\n\t"::: "memory"); //FIXME this should not be required but it IS (even for non-MMX versions)
188#endif
189
190 // NOTE: the +1 is for the MMX scaler which reads over the end
191 FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW+1)*sizeof(int16_t), fail);
192
193 if (FFABS(xInc - 0x10000) <10) { // unscaled
194 int i;
195 filterSize= 1;
196 FF_ALLOCZ_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
197
198 for (i=0; i<dstW; i++) {
199 filter[i*filterSize]= fone;
200 (*filterPos)[i]=i;
201 }
202
203 } else if (flags&SWS_POINT) { // lame looking point sampling mode
204 int i;
205 int xDstInSrc;
206 filterSize= 1;
207 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
208
209 xDstInSrc= xInc/2 - 0x8000;
210 for (i=0; i<dstW; i++) {
211 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
212
213 (*filterPos)[i]= xx;
214 filter[i]= fone;
215 xDstInSrc+= xInc;
216 }
217 } else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) { // bilinear upscale
218 int i;
219 int xDstInSrc;
220 filterSize= 2;
221 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
222
223 xDstInSrc= xInc/2 - 0x8000;
224 for (i=0; i<dstW; i++) {
225 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
226 int j;
227
228 (*filterPos)[i]= xx;
229 //bilinear upscale / linear interpolate / area averaging
230 for (j=0; j<filterSize; j++) {
231 int64_t coeff= fone - FFABS((xx<<16) - xDstInSrc)*(fone>>16);
232 if (coeff<0) coeff=0;
233 filter[i*filterSize + j]= coeff;
234 xx++;
235 }
236 xDstInSrc+= xInc;
237 }
238 } else {
239 int xDstInSrc;
240 int sizeFactor;
241
242 if (flags&SWS_BICUBIC) sizeFactor= 4;
243 else if (flags&SWS_X) sizeFactor= 8;
244 else if (flags&SWS_AREA) sizeFactor= 1; //downscale only, for upscale it is bilinear
245 else if (flags&SWS_GAUSS) sizeFactor= 8; // infinite ;)
246 else if (flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? ceil(2*param[0]) : 6;
247 else if (flags&SWS_SINC) sizeFactor= 20; // infinite ;)
248 else if (flags&SWS_SPLINE) sizeFactor= 20; // infinite ;)
249 else if (flags&SWS_BILINEAR) sizeFactor= 2;
250 else {
251 sizeFactor= 0; //GCC warning killer
252 assert(0);
253 }
254
255 if (xInc <= 1<<16) filterSize= 1 + sizeFactor; // upscale
256 else filterSize= 1 + (sizeFactor*srcW + dstW - 1)/ dstW;
257
258 if (filterSize > srcW-2) filterSize=srcW-2;
259
260 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
261
262 xDstInSrc= xInc - 0x10000;
263 for (i=0; i<dstW; i++) {
264 int xx= (xDstInSrc - ((filterSize-2)<<16)) / (1<<17);
265 int j;
266 (*filterPos)[i]= xx;
267 for (j=0; j<filterSize; j++) {
268 int64_t d= ((int64_t)FFABS((xx<<17) - xDstInSrc))<<13;
269 double floatd;
270 int64_t coeff;
271
272 if (xInc > 1<<16)
273 d= d*dstW/srcW;
274 floatd= d * (1.0/(1<<30));
275
276 if (flags & SWS_BICUBIC) {
277 int64_t B= (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1<<24);
278 int64_t C= (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1<<24);
279 int64_t dd = ( d*d)>>30;
280 int64_t ddd= (dd*d)>>30;
281
282 if (d < 1LL<<30)
283 coeff = (12*(1<<24)-9*B-6*C)*ddd + (-18*(1<<24)+12*B+6*C)*dd + (6*(1<<24)-2*B)*(1<<30);
284 else if (d < 1LL<<31)
285 coeff = (-B-6*C)*ddd + (6*B+30*C)*dd + (-12*B-48*C)*d + (8*B+24*C)*(1<<30);
286 else
287 coeff=0.0;
288 coeff *= fone>>(30+24);
289 }
290/* else if (flags & SWS_X) {
291 double p= param ? param*0.01 : 0.3;
4e74187d 292 coeff = d ? sin(d*M_PI)/(d*M_PI) : 1.0;
a4388ebd
RP
293 coeff*= pow(2.0, - p*d*d);
294 }*/
295 else if (flags & SWS_X) {
296 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
297 double c;
298
299 if (floatd<1.0)
2658e7a6 300 c = cos(floatd*M_PI);
a4388ebd
RP
301 else
302 c=-1.0;
303 if (c<0.0) c= -pow(-c, A);
304 else c= pow( c, A);
305 coeff= (c*0.5 + 0.5)*fone;
306 } else if (flags & SWS_AREA) {
307 int64_t d2= d - (1<<29);
308 if (d2*xInc < -(1LL<<(29+16))) coeff= 1.0 * (1LL<<(30+16));
309 else if (d2*xInc < (1LL<<(29+16))) coeff= -d2*xInc + (1LL<<(29+16));
310 else coeff=0.0;
311 coeff *= fone>>(30+16);
312 } else if (flags & SWS_GAUSS) {
313 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
314 coeff = (pow(2.0, - p*floatd*floatd))*fone;
315 } else if (flags & SWS_SINC) {
2658e7a6 316 coeff = (d ? sin(floatd*M_PI)/(floatd*M_PI) : 1.0)*fone;
a4388ebd
RP
317 } else if (flags & SWS_LANCZOS) {
318 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
2658e7a6 319 coeff = (d ? sin(floatd*M_PI)*sin(floatd*M_PI/p)/(floatd*floatd*M_PI*M_PI/p) : 1.0)*fone;
a4388ebd
RP
320 if (floatd>p) coeff=0;
321 } else if (flags & SWS_BILINEAR) {
322 coeff= (1<<30) - d;
323 if (coeff<0) coeff=0;
324 coeff *= fone >> 30;
325 } else if (flags & SWS_SPLINE) {
326 double p=-2.196152422706632;
327 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, floatd) * fone;
328 } else {
329 coeff= 0.0; //GCC warning killer
330 assert(0);
331 }
332
333 filter[i*filterSize + j]= coeff;
334 xx++;
335 }
336 xDstInSrc+= 2*xInc;
337 }
338 }
339
340 /* apply src & dst Filter to filter -> filter2
341 av_free(filter);
342 */
343 assert(filterSize>0);
344 filter2Size= filterSize;
345 if (srcFilter) filter2Size+= srcFilter->length - 1;
346 if (dstFilter) filter2Size+= dstFilter->length - 1;
347 assert(filter2Size>0);
348 FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size*dstW*sizeof(*filter2), fail);
349
350 for (i=0; i<dstW; i++) {
351 int j, k;
352
353 if(srcFilter) {
354 for (k=0; k<srcFilter->length; k++) {
355 for (j=0; j<filterSize; j++)
356 filter2[i*filter2Size + k + j] += srcFilter->coeff[k]*filter[i*filterSize + j];
357 }
358 } else {
359 for (j=0; j<filterSize; j++)
360 filter2[i*filter2Size + j]= filter[i*filterSize + j];
361 }
362 //FIXME dstFilter
363
364 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
365 }
366 av_freep(&filter);
367
368 /* try to reduce the filter-size (step1 find size and shift left) */
369 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
370 minFilterSize= 0;
371 for (i=dstW-1; i>=0; i--) {
372 int min= filter2Size;
373 int j;
374 int64_t cutOff=0.0;
375
376 /* get rid of near zero elements on the left by shifting left */
377 for (j=0; j<filter2Size; j++) {
378 int k;
379 cutOff += FFABS(filter2[i*filter2Size]);
380
381 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
382
383 /* preserve monotonicity because the core can't handle the filter otherwise */
384 if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
385
386 // move filter coefficients left
387 for (k=1; k<filter2Size; k++)
388 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
389 filter2[i*filter2Size + k - 1]= 0;
390 (*filterPos)[i]++;
391 }
392
393 cutOff=0;
394 /* count near zeros on the right */
395 for (j=filter2Size-1; j>0; j--) {
396 cutOff += FFABS(filter2[i*filter2Size + j]);
397
398 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
399 min--;
400 }
401
402 if (min>minFilterSize) minFilterSize= min;
403 }
404
405 if (flags & SWS_CPU_CAPS_ALTIVEC) {
406 // we can handle the special case 4,
407 // so we don't want to go to the full 8
408 if (minFilterSize < 5)
409 filterAlign = 4;
410
411 // We really don't want to waste our time
412 // doing useless computation, so fall back on
413 // the scalar C code for very small filters.
414 // Vectorizing is worth it only if you have a
415 // decent-sized vector.
416 if (minFilterSize < 3)
417 filterAlign = 1;
418 }
419
420 if (flags & SWS_CPU_CAPS_MMX) {
421 // special case for unscaled vertical filtering
422 if (minFilterSize == 1 && filterAlign == 2)
423 filterAlign= 1;
424 }
425
426 assert(minFilterSize > 0);
427 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
428 assert(filterSize > 0);
429 filter= av_malloc(filterSize*dstW*sizeof(*filter));
430 if (filterSize >= MAX_FILTER_SIZE*16/((flags&SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
431 goto fail;
432 *outFilterSize= filterSize;
433
434 if (flags&SWS_PRINT_INFO)
435 av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
436 /* try to reduce the filter-size (step2 reduce it) */
437 for (i=0; i<dstW; i++) {
438 int j;
439
440 for (j=0; j<filterSize; j++) {
441 if (j>=filter2Size) filter[i*filterSize + j]= 0;
442 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
443 if((flags & SWS_BITEXACT) && j>=minFilterSize)
444 filter[i*filterSize + j]= 0;
445 }
446 }
447
448 //FIXME try to align filterPos if possible
449
450 //fix borders
451 for (i=0; i<dstW; i++) {
452 int j;
453 if ((*filterPos)[i] < 0) {
454 // move filter coefficients left to compensate for filterPos
455 for (j=1; j<filterSize; j++) {
456 int left= FFMAX(j + (*filterPos)[i], 0);
457 filter[i*filterSize + left] += filter[i*filterSize + j];
458 filter[i*filterSize + j]=0;
459 }
460 (*filterPos)[i]= 0;
461 }
462
463 if ((*filterPos)[i] + filterSize > srcW) {
464 int shift= (*filterPos)[i] + filterSize - srcW;
465 // move filter coefficients right to compensate for filterPos
466 for (j=filterSize-2; j>=0; j--) {
467 int right= FFMIN(j + shift, filterSize-1);
468 filter[i*filterSize +right] += filter[i*filterSize +j];
469 filter[i*filterSize +j]=0;
470 }
471 (*filterPos)[i]= srcW - filterSize;
472 }
473 }
474
475 // Note the +1 is for the MMX scaler which reads over the end
476 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
477 FF_ALLOCZ_OR_GOTO(NULL, *outFilter, *outFilterSize*(dstW+1)*sizeof(int16_t), fail);
478
479 /* normalize & store in outFilter */
480 for (i=0; i<dstW; i++) {
481 int j;
482 int64_t error=0;
483 int64_t sum=0;
484
485 for (j=0; j<filterSize; j++) {
486 sum+= filter[i*filterSize + j];
487 }
488 sum= (sum + one/2)/ one;
489 for (j=0; j<*outFilterSize; j++) {
490 int64_t v= filter[i*filterSize + j] + error;
491 int intV= ROUNDED_DIV(v, sum);
492 (*outFilter)[i*(*outFilterSize) + j]= intV;
493 error= v - intV*sum;
494 }
495 }
496
497 (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
498 for (i=0; i<*outFilterSize; i++) {
499 int j= dstW*(*outFilterSize);
500 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
501 }
502
503 ret=0;
504fail:
505 av_free(filter);
506 av_free(filter2);
507 return ret;
508}
509
819ee683 510#if ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT)
a4388ebd
RP
511static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode, int16_t *filter, int32_t *filterPos, int numSplits)
512{
513 uint8_t *fragmentA;
514 x86_reg imm8OfPShufW1A;
515 x86_reg imm8OfPShufW2A;
516 x86_reg fragmentLengthA;
517 uint8_t *fragmentB;
518 x86_reg imm8OfPShufW1B;
519 x86_reg imm8OfPShufW2B;
520 x86_reg fragmentLengthB;
521 int fragmentPos;
522
523 int xpos, i;
524
525 // create an optimized horizontal scaling routine
526 /* This scaler is made of runtime-generated MMX2 code using specially
527 * tuned pshufw instructions. For every four output pixels, if four
528 * input pixels are enough for the fast bilinear scaling, then a chunk
529 * of fragmentB is used. If five input pixels are needed, then a chunk
530 * of fragmentA is used.
531 */
532
533 //code fragment
534
535 __asm__ volatile(
536 "jmp 9f \n\t"
537 // Begin
538 "0: \n\t"
539 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
540 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
541 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
542 "punpcklbw %%mm7, %%mm1 \n\t"
543 "punpcklbw %%mm7, %%mm0 \n\t"
544 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
545 "1: \n\t"
546 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
547 "2: \n\t"
548 "psubw %%mm1, %%mm0 \n\t"
549 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
550 "pmullw %%mm3, %%mm0 \n\t"
551 "psllw $7, %%mm1 \n\t"
552 "paddw %%mm1, %%mm0 \n\t"
553
554 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
555
556 "add $8, %%"REG_a" \n\t"
557 // End
558 "9: \n\t"
559// "int $3 \n\t"
560 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
561 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
562 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
563 "dec %1 \n\t"
564 "dec %2 \n\t"
565 "sub %0, %1 \n\t"
566 "sub %0, %2 \n\t"
567 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
568 "sub %0, %3 \n\t"
569
570
571 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
572 "=r" (fragmentLengthA)
573 );
574
575 __asm__ volatile(
576 "jmp 9f \n\t"
577 // Begin
578 "0: \n\t"
579 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
580 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
581 "punpcklbw %%mm7, %%mm0 \n\t"
582 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
583 "1: \n\t"
584 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
585 "2: \n\t"
586 "psubw %%mm1, %%mm0 \n\t"
587 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
588 "pmullw %%mm3, %%mm0 \n\t"
589 "psllw $7, %%mm1 \n\t"
590 "paddw %%mm1, %%mm0 \n\t"
591
592 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
593
594 "add $8, %%"REG_a" \n\t"
595 // End
596 "9: \n\t"
597// "int $3 \n\t"
598 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
599 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
600 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
601 "dec %1 \n\t"
602 "dec %2 \n\t"
603 "sub %0, %1 \n\t"
604 "sub %0, %2 \n\t"
605 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
606 "sub %0, %3 \n\t"
607
608
609 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
610 "=r" (fragmentLengthB)
611 );
612
613 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
614 fragmentPos=0;
615
616 for (i=0; i<dstW/numSplits; i++) {
617 int xx=xpos>>16;
618
619 if ((i&3) == 0) {
620 int a=0;
621 int b=((xpos+xInc)>>16) - xx;
622 int c=((xpos+xInc*2)>>16) - xx;
623 int d=((xpos+xInc*3)>>16) - xx;
624 int inc = (d+1<4);
625 uint8_t *fragment = (d+1<4) ? fragmentB : fragmentA;
626 x86_reg imm8OfPShufW1 = (d+1<4) ? imm8OfPShufW1B : imm8OfPShufW1A;
627 x86_reg imm8OfPShufW2 = (d+1<4) ? imm8OfPShufW2B : imm8OfPShufW2A;
628 x86_reg fragmentLength = (d+1<4) ? fragmentLengthB : fragmentLengthA;
629 int maxShift= 3-(d+inc);
630 int shift=0;
631
632 if (filterCode) {
633 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
634 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
635 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
636 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
637 filterPos[i/2]= xx;
638
639 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
640
641 filterCode[fragmentPos + imm8OfPShufW1]=
642 (a+inc) | ((b+inc)<<2) | ((c+inc)<<4) | ((d+inc)<<6);
643 filterCode[fragmentPos + imm8OfPShufW2]=
644 a | (b<<2) | (c<<4) | (d<<6);
645
646 if (i+4-inc>=dstW) shift=maxShift; //avoid overread
647 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
648
649 if (shift && i>=shift) {
650 filterCode[fragmentPos + imm8OfPShufW1]+= 0x55*shift;
651 filterCode[fragmentPos + imm8OfPShufW2]+= 0x55*shift;
652 filterPos[i/2]-=shift;
653 }
654 }
655
656 fragmentPos+= fragmentLength;
657
658 if (filterCode)
659 filterCode[fragmentPos]= RET;
660 }
661 xpos+=xInc;
662 }
663 if (filterCode)
664 filterPos[((i/2)+1)&(~1)]= xpos>>16; // needed to jump to the next part
665
666 return fragmentPos + 1;
667}
819ee683 668#endif /* ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT) */
a4388ebd
RP
669
670static void getSubSampleFactors(int *h, int *v, enum PixelFormat format)
671{
672 *h = av_pix_fmt_descriptors[format].log2_chroma_w;
673 *v = av_pix_fmt_descriptors[format].log2_chroma_h;
674}
675
635d4aed
MN
676static int update_flags_cpu(int flags);
677
a4388ebd
RP
678int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation)
679{
a4388ebd
RP
680 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
681 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
682
683 c->brightness= brightness;
684 c->contrast = contrast;
685 c->saturation= saturation;
686 c->srcRange = srcRange;
687 c->dstRange = dstRange;
688 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
689
635d4aed
MN
690 c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->dstFormat]);
691 c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->srcFormat]);
692 c->flags = update_flags_cpu(c->flags);
693
a4388ebd
RP
694 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
695 //FIXME factorize
696
f684f3c5 697#if HAVE_ALTIVEC
a4388ebd
RP
698 if (c->flags & SWS_CPU_CAPS_ALTIVEC)
699 ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness, contrast, saturation);
700#endif
701 return 0;
702}
703
704int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation)
705{
706 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
707
708 *inv_table = c->srcColorspaceTable;
709 *table = c->dstColorspaceTable;
710 *srcRange = c->srcRange;
711 *dstRange = c->dstRange;
712 *brightness= c->brightness;
713 *contrast = c->contrast;
714 *saturation= c->saturation;
715
716 return 0;
717}
718
719static int handle_jpeg(enum PixelFormat *format)
720{
721 switch (*format) {
b8dbff33
SS
722 case PIX_FMT_YUVJ420P: *format = PIX_FMT_YUV420P; return 1;
723 case PIX_FMT_YUVJ422P: *format = PIX_FMT_YUV422P; return 1;
724 case PIX_FMT_YUVJ444P: *format = PIX_FMT_YUV444P; return 1;
725 case PIX_FMT_YUVJ440P: *format = PIX_FMT_YUV440P; return 1;
726 default: return 0;
a4388ebd
RP
727 }
728}
729
c17f5a4e
RP
730static int update_flags_cpu(int flags)
731{
732#if !CONFIG_RUNTIME_CPUDETECT //ensure that the flags match the compiled variant if cpudetect is off
829e5daf
JG
733 flags &= ~( SWS_CPU_CAPS_MMX
734 |SWS_CPU_CAPS_MMX2
735 |SWS_CPU_CAPS_3DNOW
9f694f0b 736 |SWS_CPU_CAPS_SSE2
829e5daf
JG
737 |SWS_CPU_CAPS_ALTIVEC
738 |SWS_CPU_CAPS_BFIN);
c17f5a4e
RP
739 flags |= ff_hardcodedcpuflags();
740#endif /* CONFIG_RUNTIME_CPUDETECT */
741 return flags;
742}
743
635d4aed
MN
744SwsContext *sws_alloc_context(void){
745 SwsContext *c= av_mallocz(sizeof(SwsContext));
746
747 c->av_class = &sws_context_class;
748
749 return c;
750}
751
752int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter){
a4388ebd
RP
753 int i;
754 int usesVFilter, usesHFilter;
755 int unscaled;
a4388ebd 756 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
635d4aed
MN
757 int srcW= c->srcW;
758 int srcH= c->srcH;
759 int dstW= c->dstW;
760 int dstH= c->dstH;
761 int flags;
762 enum PixelFormat srcFormat= c->srcFormat;
763 enum PixelFormat dstFormat= c->dstFormat;
764
765 flags= c->flags = update_flags_cpu(c->flags);
a4388ebd
RP
766#if ARCH_X86
767 if (flags & SWS_CPU_CAPS_MMX)
768 __asm__ volatile("emms\n\t"::: "memory");
769#endif
a4388ebd
RP
770 if (!rgb15to16) sws_rgb2rgb_init(flags);
771
772 unscaled = (srcW == dstW && srcH == dstH);
773
a4388ebd
RP
774 if (!isSupportedIn(srcFormat)) {
775 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as input pixel format\n", sws_format_name(srcFormat));
635d4aed 776 return AVERROR(EINVAL);
a4388ebd
RP
777 }
778 if (!isSupportedOut(dstFormat)) {
779 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as output pixel format\n", sws_format_name(dstFormat));
635d4aed 780 return AVERROR(EINVAL);
a4388ebd
RP
781 }
782
783 i= flags & ( SWS_POINT
784 |SWS_AREA
785 |SWS_BILINEAR
786 |SWS_FAST_BILINEAR
787 |SWS_BICUBIC
788 |SWS_X
789 |SWS_GAUSS
790 |SWS_LANCZOS
791 |SWS_SINC
792 |SWS_SPLINE
793 |SWS_BICUBLIN);
794 if(!i || (i & (i-1))) {
795 av_log(NULL, AV_LOG_ERROR, "swScaler: Exactly one scaler algorithm must be chosen\n");
635d4aed 796 return AVERROR(EINVAL);
a4388ebd 797 }
a4388ebd
RP
798 /* sanity check */
799 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
800 av_log(NULL, AV_LOG_ERROR, "swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
801 srcW, srcH, dstW, dstH);
635d4aed 802 return AVERROR(EINVAL);
a4388ebd
RP
803 }
804 if(srcW > VOFW || dstW > VOFW) {
805 av_log(NULL, AV_LOG_ERROR, "swScaler: Compile-time maximum width is "AV_STRINGIFY(VOFW)" change VOF/VOFW and recompile\n");
635d4aed 806 return AVERROR(EINVAL);
a4388ebd
RP
807 }
808
809 if (!dstFilter) dstFilter= &dummyFilter;
810 if (!srcFilter) srcFilter= &dummyFilter;
811
a4388ebd
RP
812 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
813 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
30b61475
SS
814 c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[dstFormat]);
815 c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[srcFormat]);
a4388ebd
RP
816 c->vRounder= 4* 0x0001000100010001ULL;
817
62bb4ca1
SS
818 usesVFilter = (srcFilter->lumV && srcFilter->lumV->length>1) ||
819 (srcFilter->chrV && srcFilter->chrV->length>1) ||
820 (dstFilter->lumV && dstFilter->lumV->length>1) ||
821 (dstFilter->chrV && dstFilter->chrV->length>1);
822 usesHFilter = (srcFilter->lumH && srcFilter->lumH->length>1) ||
823 (srcFilter->chrH && srcFilter->chrH->length>1) ||
824 (dstFilter->lumH && dstFilter->lumH->length>1) ||
825 (dstFilter->chrH && dstFilter->chrH->length>1);
a4388ebd
RP
826
827 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
828 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
829
830 // reuse chroma for 2 pixels RGB/BGR unless user wants full chroma interpolation
60222557 831 if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
a4388ebd
RP
832
833 // drop some chroma lines if the user wants it
834 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
835 c->chrSrcVSubSample+= c->vChrDrop;
836
837 // drop every other pixel for chroma calculation unless user wants full chroma
60222557 838 if (isAnyRGB(srcFormat) && !(flags&SWS_FULL_CHR_H_INP)
a4388ebd
RP
839 && srcFormat!=PIX_FMT_RGB8 && srcFormat!=PIX_FMT_BGR8
840 && srcFormat!=PIX_FMT_RGB4 && srcFormat!=PIX_FMT_BGR4
841 && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE
f4203ff3 842 && ((dstW>>c->chrDstHSubSample) <= (srcW>>1) || (flags&SWS_FAST_BILINEAR)))
a4388ebd
RP
843 c->chrSrcHSubSample=1;
844
a4388ebd
RP
845 // Note the -((-x)>>y) is so that we always round toward +inf.
846 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
847 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
848 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
849 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
850
a4388ebd 851 /* unscaled special cases */
635d4aed 852 if (unscaled && !usesHFilter && !usesVFilter && (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
a4388ebd
RP
853 ff_get_unscaled_swscale(c);
854
855 if (c->swScale) {
856 if (flags&SWS_PRINT_INFO)
857 av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
858 sws_format_name(srcFormat), sws_format_name(dstFormat));
635d4aed 859 return 0;
a4388ebd
RP
860 }
861 }
862
863 if (flags & SWS_CPU_CAPS_MMX2) {
864 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
865 if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR)) {
866 if (flags&SWS_PRINT_INFO)
867 av_log(c, AV_LOG_INFO, "output width is not a multiple of 32 -> no MMX2 scaler\n");
868 }
869 if (usesHFilter) c->canMMX2BeUsed=0;
870 }
871 else
872 c->canMMX2BeUsed=0;
873
874 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
875 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
876
877 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
878 // but only for the FAST_BILINEAR mode otherwise do correct scaling
879 // n-2 is the last chrominance sample available
880 // this is not perfect, but no one should notice the difference, the more correct variant
881 // would be like the vertical one, but that would require some special code for the
882 // first and last pixel
883 if (flags&SWS_FAST_BILINEAR) {
884 if (c->canMMX2BeUsed) {
885 c->lumXInc+= 20;
886 c->chrXInc+= 20;
887 }
888 //we don't use the x86 asm scaler if MMX is available
889 else if (flags & SWS_CPU_CAPS_MMX) {
890 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
891 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
892 }
893 }
894
895 /* precalculate horizontal scaler filter coefficients */
896 {
819ee683 897#if ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT)
a4388ebd
RP
898// can't downscale !!!
899 if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {
900 c->lumMmx2FilterCodeSize = initMMX2HScaler( dstW, c->lumXInc, NULL, NULL, NULL, 8);
901 c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc, NULL, NULL, NULL, 4);
902
903#ifdef MAP_ANONYMOUS
53bc0dc2
GC
904 c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
905 c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
a4388ebd
RP
906#elif HAVE_VIRTUALALLOC
907 c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
908 c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
909#else
910 c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);
911 c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);
912#endif
913
97cda76a 914 if (!c->lumMmx2FilterCode || !c->chrMmx2FilterCode)
635d4aed 915 return AVERROR(ENOMEM);
a4388ebd
RP
916 FF_ALLOCZ_OR_GOTO(c, c->hLumFilter , (dstW /8+8)*sizeof(int16_t), fail);
917 FF_ALLOCZ_OR_GOTO(c, c->hChrFilter , (c->chrDstW /4+8)*sizeof(int16_t), fail);
918 FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW /2/8+8)*sizeof(int32_t), fail);
919 FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW/2/4+8)*sizeof(int32_t), fail);
920
921 initMMX2HScaler( dstW, c->lumXInc, c->lumMmx2FilterCode, c->hLumFilter, c->hLumFilterPos, 8);
922 initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode, c->hChrFilter, c->hChrFilterPos, 4);
923
924#ifdef MAP_ANONYMOUS
925 mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
926 mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
927#endif
928 } else
819ee683 929#endif /* ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT) */
a4388ebd
RP
930 {
931 const int filterAlign=
932 (flags & SWS_CPU_CAPS_MMX) ? 4 :
933 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
934 1;
935
936 if (initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
937 srcW , dstW, filterAlign, 1<<14,
938 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
939 srcFilter->lumH, dstFilter->lumH, c->param) < 0)
940 goto fail;
941 if (initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
942 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
943 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
944 srcFilter->chrH, dstFilter->chrH, c->param) < 0)
945 goto fail;
946 }
947 } // initialize horizontal stuff
948
949 /* precalculate vertical scaler filter coefficients */
950 {
951 const int filterAlign=
952 (flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
953 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
954 1;
955
956 if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
957 srcH , dstH, filterAlign, (1<<12),
958 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
959 srcFilter->lumV, dstFilter->lumV, c->param) < 0)
960 goto fail;
961 if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
962 c->chrSrcH, c->chrDstH, filterAlign, (1<<12),
963 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
964 srcFilter->chrV, dstFilter->chrV, c->param) < 0)
965 goto fail;
966
f684f3c5 967#if HAVE_ALTIVEC
a4388ebd
RP
968 FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof (vector signed short)*c->vLumFilterSize*c->dstH, fail);
969 FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH, fail);
970
971 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
972 int j;
973 short *p = (short *)&c->vYCoeffsBank[i];
974 for (j=0;j<8;j++)
975 p[j] = c->vLumFilter[i];
976 }
977
978 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
979 int j;
980 short *p = (short *)&c->vCCoeffsBank[i];
981 for (j=0;j<8;j++)
982 p[j] = c->vChrFilter[i];
983 }
984#endif
985 }
986
987 // calculate buffer sizes so that they won't run out while handling these damn slices
988 c->vLumBufSize= c->vLumFilterSize;
989 c->vChrBufSize= c->vChrFilterSize;
990 for (i=0; i<dstH; i++) {
991 int chrI= i*c->chrDstH / dstH;
992 int nextSlice= FFMAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
993 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
994
995 nextSlice>>= c->chrSrcVSubSample;
996 nextSlice<<= c->chrSrcVSubSample;
997 if (c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
998 c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
999 if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
1000 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
1001 }
1002
1003 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
1004 // allocate several megabytes to handle all possible cases)
1005 FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
1006 FF_ALLOC_OR_GOTO(c, c->chrPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
1007 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
1008 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
1009 //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)
1010 /* align at 16 bytes for AltiVec */
1011 for (i=0; i<c->vLumBufSize; i++) {
1012 FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], VOF+1, fail);
1013 c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];
1014 }
1015 for (i=0; i<c->vChrBufSize; i++) {
1016 FF_ALLOC_OR_GOTO(c, c->chrPixBuf[i+c->vChrBufSize], (VOF+1)*2, fail);
1017 c->chrPixBuf[i] = c->chrPixBuf[i+c->vChrBufSize];
1018 }
1019 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
1020 for (i=0; i<c->vLumBufSize; i++) {
1021 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], VOF+1, fail);
1022 c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];
1023 }
1024
1025 //try to avoid drawing green stuff between the right end and the stride end
1026 for (i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, (VOF+1)*2);
1027
1028 assert(2*VOFW == VOF);
1029
1030 assert(c->chrDstH <= dstH);
1031
1032 if (flags&SWS_PRINT_INFO) {
1033 if (flags&SWS_FAST_BILINEAR)
1034 av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
1035 else if (flags&SWS_BILINEAR)
1036 av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
1037 else if (flags&SWS_BICUBIC)
1038 av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
1039 else if (flags&SWS_X)
1040 av_log(c, AV_LOG_INFO, "Experimental scaler, ");
1041 else if (flags&SWS_POINT)
1042 av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
1043 else if (flags&SWS_AREA)
1044 av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
1045 else if (flags&SWS_BICUBLIN)
1046 av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
1047 else if (flags&SWS_GAUSS)
1048 av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
1049 else if (flags&SWS_SINC)
1050 av_log(c, AV_LOG_INFO, "Sinc scaler, ");
1051 else if (flags&SWS_LANCZOS)
1052 av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
1053 else if (flags&SWS_SPLINE)
1054 av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
1055 else
1056 av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
1057
1058 av_log(c, AV_LOG_INFO, "from %s to %s%s ",
1059 sws_format_name(srcFormat),
1060#ifdef DITHER1XBPP
22e8222f
JK
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 " : "",
a4388ebd
RP
1064#else
1065 "",
1066#endif
1067 sws_format_name(dstFormat));
1068
1069 if (flags & SWS_CPU_CAPS_MMX2)
1070 av_log(c, AV_LOG_INFO, "using MMX2\n");
1071 else if (flags & SWS_CPU_CAPS_3DNOW)
1072 av_log(c, AV_LOG_INFO, "using 3DNOW\n");
1073 else if (flags & SWS_CPU_CAPS_MMX)
1074 av_log(c, AV_LOG_INFO, "using MMX\n");
1075 else if (flags & SWS_CPU_CAPS_ALTIVEC)
1076 av_log(c, AV_LOG_INFO, "using AltiVec\n");
1077 else
1078 av_log(c, AV_LOG_INFO, "using C\n");
a4388ebd 1079
a4388ebd
RP
1080 if (flags & SWS_CPU_CAPS_MMX) {
1081 if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
1082 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
1083 else {
1084 if (c->hLumFilterSize==4)
1085 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal luminance scaling\n");
1086 else if (c->hLumFilterSize==8)
1087 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal luminance scaling\n");
1088 else
1089 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal luminance scaling\n");
1090
1091 if (c->hChrFilterSize==4)
1092 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal chrominance scaling\n");
1093 else if (c->hChrFilterSize==8)
1094 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal chrominance scaling\n");
1095 else
1096 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal chrominance scaling\n");
1097 }
1098 } else {
1099#if ARCH_X86
1100 av_log(c, AV_LOG_VERBOSE, "using x86 asm scaler for horizontal scaling\n");
1101#else
1102 if (flags & SWS_FAST_BILINEAR)
1103 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR C scaler for horizontal scaling\n");
1104 else
1105 av_log(c, AV_LOG_VERBOSE, "using C scaler for horizontal scaling\n");
1106#endif
1107 }
1108 if (isPlanarYUV(dstFormat)) {
1109 if (c->vLumFilterSize==1)
1110 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1111 else
1112 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1113 } else {
1114 if (c->vLumFilterSize==1 && c->vChrFilterSize==2)
1115 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
1116 " 2-tap scaler for vertical chrominance scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1117 else if (c->vLumFilterSize==2 && c->vChrFilterSize==2)
1118 av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1119 else
1120 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1121 }
1122
1123 if (dstFormat==PIX_FMT_BGR24)
1124 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 converter\n",
1125 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
1126 else if (dstFormat==PIX_FMT_RGB32)
1127 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1128 else if (dstFormat==PIX_FMT_BGR565)
1129 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1130 else if (dstFormat==PIX_FMT_BGR555)
1131 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
22e8222f
JK
1132 else if (dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||
1133 dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE)
1134 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR12 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
a4388ebd
RP
1135
1136 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
a4388ebd
RP
1137 av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1138 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
1139 av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1140 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
1141 }
1142
1143 c->swScale= ff_getSwsFunc(c);
635d4aed
MN
1144 return 0;
1145fail: //FIXME replace things by appropriate error codes
1146 return -1;
1147}
a4388ebd 1148
635d4aed
MN
1149SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,
1150 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
1151 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
1152{
1153 SwsContext *c;
1154
1155 if(!(c=sws_alloc_context()))
1156 return NULL;
1157
1158 c->flags= flags;
1159 c->srcW= srcW;
1160 c->srcH= srcH;
1161 c->dstW= dstW;
1162 c->dstH= dstH;
1163 c->srcRange = handle_jpeg(&srcFormat);
1164 c->dstRange = handle_jpeg(&dstFormat);
1165 c->srcFormat= srcFormat;
1166 c->dstFormat= dstFormat;
1167
1168 if (param) {
1169 c->param[0] = param[0];
1170 c->param[1] = param[1];
1171 } else {
1172 c->param[0] =
1173 c->param[1] = SWS_PARAM_DEFAULT;
1174 }
1175 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);
1176
1177 if(sws_init_context(c, srcFilter, dstFilter) < 0){
1178 sws_freeContext(c);
1179 return NULL;
1180 }
1181
1182 return c;
a4388ebd
RP
1183}
1184
1185SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
1186 float lumaSharpen, float chromaSharpen,
1187 float chromaHShift, float chromaVShift,
1188 int verbose)
1189{
1190 SwsFilter *filter= av_malloc(sizeof(SwsFilter));
1191 if (!filter)
1192 return NULL;
1193
1194 if (lumaGBlur!=0.0) {
1195 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
1196 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
1197 } else {
1198 filter->lumH= sws_getIdentityVec();
1199 filter->lumV= sws_getIdentityVec();
1200 }
1201
1202 if (chromaGBlur!=0.0) {
1203 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
1204 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
1205 } else {
1206 filter->chrH= sws_getIdentityVec();
1207 filter->chrV= sws_getIdentityVec();
1208 }
1209
1210 if (chromaSharpen!=0.0) {
1211 SwsVector *id= sws_getIdentityVec();
1212 sws_scaleVec(filter->chrH, -chromaSharpen);
1213 sws_scaleVec(filter->chrV, -chromaSharpen);
1214 sws_addVec(filter->chrH, id);
1215 sws_addVec(filter->chrV, id);
1216 sws_freeVec(id);
1217 }
1218
1219 if (lumaSharpen!=0.0) {
1220 SwsVector *id= sws_getIdentityVec();
1221 sws_scaleVec(filter->lumH, -lumaSharpen);
1222 sws_scaleVec(filter->lumV, -lumaSharpen);
1223 sws_addVec(filter->lumH, id);
1224 sws_addVec(filter->lumV, id);
1225 sws_freeVec(id);
1226 }
1227
1228 if (chromaHShift != 0.0)
1229 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
1230
1231 if (chromaVShift != 0.0)
1232 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
1233
1234 sws_normalizeVec(filter->chrH, 1.0);
1235 sws_normalizeVec(filter->chrV, 1.0);
1236 sws_normalizeVec(filter->lumH, 1.0);
1237 sws_normalizeVec(filter->lumV, 1.0);
1238
1239 if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
1240 if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
1241
1242 return filter;
1243}
1244
1245SwsVector *sws_allocVec(int length)
1246{
1247 SwsVector *vec = av_malloc(sizeof(SwsVector));
1248 if (!vec)
1249 return NULL;
1250 vec->length = length;
1251 vec->coeff = av_malloc(sizeof(double) * length);
1252 if (!vec->coeff)
1253 av_freep(&vec);
1254 return vec;
1255}
1256
1257SwsVector *sws_getGaussianVec(double variance, double quality)
1258{
1259 const int length= (int)(variance*quality + 0.5) | 1;
1260 int i;
1261 double middle= (length-1)*0.5;
1262 SwsVector *vec= sws_allocVec(length);
1263
1264 if (!vec)
1265 return NULL;
1266
1267 for (i=0; i<length; i++) {
1268 double dist= i-middle;
2658e7a6 1269 vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);
a4388ebd
RP
1270 }
1271
1272 sws_normalizeVec(vec, 1.0);
1273
1274 return vec;
1275}
1276
1277SwsVector *sws_getConstVec(double c, int length)
1278{
1279 int i;
1280 SwsVector *vec= sws_allocVec(length);
1281
1282 if (!vec)
1283 return NULL;
1284
1285 for (i=0; i<length; i++)
1286 vec->coeff[i]= c;
1287
1288 return vec;
1289}
1290
1291SwsVector *sws_getIdentityVec(void)
1292{
1293 return sws_getConstVec(1.0, 1);
1294}
1295
627686e6 1296static double sws_dcVec(SwsVector *a)
a4388ebd
RP
1297{
1298 int i;
1299 double sum=0;
1300
1301 for (i=0; i<a->length; i++)
1302 sum+= a->coeff[i];
1303
1304 return sum;
1305}
1306
1307void sws_scaleVec(SwsVector *a, double scalar)
1308{
1309 int i;
1310
1311 for (i=0; i<a->length; i++)
1312 a->coeff[i]*= scalar;
1313}
1314
1315void sws_normalizeVec(SwsVector *a, double height)
1316{
1317 sws_scaleVec(a, height/sws_dcVec(a));
1318}
1319
1320static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
1321{
1322 int length= a->length + b->length - 1;
1323 int i, j;
1324 SwsVector *vec= sws_getConstVec(0.0, length);
1325
1326 if (!vec)
1327 return NULL;
1328
1329 for (i=0; i<a->length; i++) {
1330 for (j=0; j<b->length; j++) {
1331 vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];
1332 }
1333 }
1334
1335 return vec;
1336}
1337
1338static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
1339{
1340 int length= FFMAX(a->length, b->length);
1341 int i;
1342 SwsVector *vec= sws_getConstVec(0.0, length);
1343
1344 if (!vec)
1345 return NULL;
1346
1347 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1348 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
1349
1350 return vec;
1351}
1352
1353static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
1354{
1355 int length= FFMAX(a->length, b->length);
1356 int i;
1357 SwsVector *vec= sws_getConstVec(0.0, length);
1358
1359 if (!vec)
1360 return NULL;
1361
1362 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1363 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
1364
1365 return vec;
1366}
1367
1368/* shift left / or right if "shift" is negative */
1369static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
1370{
1371 int length= a->length + FFABS(shift)*2;
1372 int i;
1373 SwsVector *vec= sws_getConstVec(0.0, length);
1374
1375 if (!vec)
1376 return NULL;
1377
1378 for (i=0; i<a->length; i++) {
1379 vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
1380 }
1381
1382 return vec;
1383}
1384
1385void sws_shiftVec(SwsVector *a, int shift)
1386{
1387 SwsVector *shifted= sws_getShiftedVec(a, shift);
1388 av_free(a->coeff);
1389 a->coeff= shifted->coeff;
1390 a->length= shifted->length;
1391 av_free(shifted);
1392}
1393
1394void sws_addVec(SwsVector *a, SwsVector *b)
1395{
1396 SwsVector *sum= sws_sumVec(a, b);
1397 av_free(a->coeff);
1398 a->coeff= sum->coeff;
1399 a->length= sum->length;
1400 av_free(sum);
1401}
1402
1403void sws_subVec(SwsVector *a, SwsVector *b)
1404{
1405 SwsVector *diff= sws_diffVec(a, b);
1406 av_free(a->coeff);
1407 a->coeff= diff->coeff;
1408 a->length= diff->length;
1409 av_free(diff);
1410}
1411
1412void sws_convVec(SwsVector *a, SwsVector *b)
1413{
1414 SwsVector *conv= sws_getConvVec(a, b);
1415 av_free(a->coeff);
1416 a->coeff= conv->coeff;
1417 a->length= conv->length;
1418 av_free(conv);
1419}
1420
1421SwsVector *sws_cloneVec(SwsVector *a)
1422{
1423 int i;
1424 SwsVector *vec= sws_allocVec(a->length);
1425
1426 if (!vec)
1427 return NULL;
1428
1429 for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];
1430
1431 return vec;
1432}
1433
1434void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
1435{
1436 int i;
1437 double max=0;
1438 double min=0;
1439 double range;
1440
1441 for (i=0; i<a->length; i++)
1442 if (a->coeff[i]>max) max= a->coeff[i];
1443
1444 for (i=0; i<a->length; i++)
1445 if (a->coeff[i]<min) min= a->coeff[i];
1446
1447 range= max - min;
1448
1449 for (i=0; i<a->length; i++) {
1450 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
1451 av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
1452 for (;x>0; x--) av_log(log_ctx, log_level, " ");
1453 av_log(log_ctx, log_level, "|\n");
1454 }
1455}
1456
1457#if LIBSWSCALE_VERSION_MAJOR < 1
1458void sws_printVec(SwsVector *a)
1459{
1460 sws_printVec2(a, NULL, AV_LOG_DEBUG);
1461}
1462#endif
1463
1464void sws_freeVec(SwsVector *a)
1465{
1466 if (!a) return;
1467 av_freep(&a->coeff);
1468 a->length=0;
1469 av_free(a);
1470}
1471
1472void sws_freeFilter(SwsFilter *filter)
1473{
1474 if (!filter) return;
1475
1476 if (filter->lumH) sws_freeVec(filter->lumH);
1477 if (filter->lumV) sws_freeVec(filter->lumV);
1478 if (filter->chrH) sws_freeVec(filter->chrH);
1479 if (filter->chrV) sws_freeVec(filter->chrV);
1480 av_free(filter);
1481}
1482
1483void sws_freeContext(SwsContext *c)
1484{
1485 int i;
1486 if (!c) return;
1487
1488 if (c->lumPixBuf) {
1489 for (i=0; i<c->vLumBufSize; i++)
1490 av_freep(&c->lumPixBuf[i]);
1491 av_freep(&c->lumPixBuf);
1492 }
1493
1494 if (c->chrPixBuf) {
1495 for (i=0; i<c->vChrBufSize; i++)
1496 av_freep(&c->chrPixBuf[i]);
1497 av_freep(&c->chrPixBuf);
1498 }
1499
1500 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1501 for (i=0; i<c->vLumBufSize; i++)
1502 av_freep(&c->alpPixBuf[i]);
1503 av_freep(&c->alpPixBuf);
1504 }
1505
1506 av_freep(&c->vLumFilter);
1507 av_freep(&c->vChrFilter);
1508 av_freep(&c->hLumFilter);
1509 av_freep(&c->hChrFilter);
f684f3c5 1510#if HAVE_ALTIVEC
a4388ebd
RP
1511 av_freep(&c->vYCoeffsBank);
1512 av_freep(&c->vCCoeffsBank);
1513#endif
1514
1515 av_freep(&c->vLumFilterPos);
1516 av_freep(&c->vChrFilterPos);
1517 av_freep(&c->hLumFilterPos);
1518 av_freep(&c->hChrFilterPos);
1519
819ee683 1520#if ARCH_X86
a4388ebd
RP
1521#ifdef MAP_ANONYMOUS
1522 if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);
1523 if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);
1524#elif HAVE_VIRTUALALLOC
12423f17
RD
1525 if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);
1526 if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);
a4388ebd
RP
1527#else
1528 av_free(c->lumMmx2FilterCode);
1529 av_free(c->chrMmx2FilterCode);
1530#endif
1531 c->lumMmx2FilterCode=NULL;
1532 c->chrMmx2FilterCode=NULL;
819ee683 1533#endif /* ARCH_X86 */
a4388ebd
RP
1534
1535 av_freep(&c->yuvTable);
1536
1537 av_free(c);
1538}
1539
1540struct SwsContext *sws_getCachedContext(struct SwsContext *context,
1541 int srcW, int srcH, enum PixelFormat srcFormat,
1542 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
1543 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
1544{
1545 static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};
1546
1547 if (!param)
1548 param = default_param;
1549
c17f5a4e
RP
1550 flags = update_flags_cpu(flags);
1551
5a011d8b 1552 if (context &&
22b6a24c
SS
1553 (context->srcW != srcW ||
1554 context->srcH != srcH ||
1555 context->srcFormat != srcFormat ||
1556 context->dstW != dstW ||
1557 context->dstH != dstH ||
1558 context->dstFormat != dstFormat ||
1559 context->flags != flags ||
1560 context->param[0] != param[0] ||
b623d0cb
SS
1561 context->param[1] != param[1])) {
1562 sws_freeContext(context);
1563 context = NULL;
1564 }
5a011d8b 1565
a4388ebd
RP
1566 if (!context) {
1567 return sws_getContext(srcW, srcH, srcFormat,
1568 dstW, dstH, dstFormat, flags,
1569 srcFilter, dstFilter, param);
1570 }
1571 return context;
1572}
1573