Cosmetics: make more compact the code for handle_jpeg(), this also allows
[libav.git] / libswscale / utils.c
1 /*
2 * Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
3 *
4 * This file is part of FFmpeg.
5 *
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.
10 *
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 */
20
21 #define _SVID_SOURCE //needed for MAP_ANONYMOUS
22 #include <inttypes.h>
23 #include <string.h>
24 #include <math.h>
25 #include <stdio.h>
26 #include "config.h"
27 #include <assert.h>
28 #if HAVE_SYS_MMAN_H
29 #include <sys/mman.h>
30 #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
31 #define MAP_ANONYMOUS MAP_ANON
32 #endif
33 #endif
34 #if HAVE_VIRTUALALLOC
35 #define WIN32_LEAN_AND_MEAN
36 #include <windows.h>
37 #endif
38 #include "swscale.h"
39 #include "swscale_internal.h"
40 #include "rgb2rgb.h"
41 #include "libavutil/intreadwrite.h"
42 #include "libavutil/x86_cpu.h"
43 #include "libavutil/avutil.h"
44 #include "libavutil/bswap.h"
45 #include "libavutil/pixdesc.h"
46
47 unsigned swscale_version(void)
48 {
49 return LIBSWSCALE_VERSION_INT;
50 }
51
52 const char *swscale_configuration(void)
53 {
54 return FFMPEG_CONFIGURATION;
55 }
56
57 const char *swscale_license(void)
58 {
59 #define LICENSE_PREFIX "libswscale license: "
60 return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
61 }
62
63 #define RET 0xC3 //near return opcode for x86
64
65 #define isSupportedIn(x) ( \
66 (x)==PIX_FMT_YUV420P \
67 || (x)==PIX_FMT_YUVA420P \
68 || (x)==PIX_FMT_YUYV422 \
69 || (x)==PIX_FMT_UYVY422 \
70 || (x)==PIX_FMT_RGB48BE \
71 || (x)==PIX_FMT_RGB48LE \
72 || (x)==PIX_FMT_RGB32 \
73 || (x)==PIX_FMT_RGB32_1 \
74 || (x)==PIX_FMT_BGR24 \
75 || (x)==PIX_FMT_BGR565 \
76 || (x)==PIX_FMT_BGR555 \
77 || (x)==PIX_FMT_BGR32 \
78 || (x)==PIX_FMT_BGR32_1 \
79 || (x)==PIX_FMT_RGB24 \
80 || (x)==PIX_FMT_RGB565 \
81 || (x)==PIX_FMT_RGB555 \
82 || (x)==PIX_FMT_GRAY8 \
83 || (x)==PIX_FMT_YUV410P \
84 || (x)==PIX_FMT_YUV440P \
85 || (x)==PIX_FMT_NV12 \
86 || (x)==PIX_FMT_NV21 \
87 || (x)==PIX_FMT_GRAY16BE \
88 || (x)==PIX_FMT_GRAY16LE \
89 || (x)==PIX_FMT_YUV444P \
90 || (x)==PIX_FMT_YUV422P \
91 || (x)==PIX_FMT_YUV411P \
92 || (x)==PIX_FMT_YUVJ420P \
93 || (x)==PIX_FMT_YUVJ422P \
94 || (x)==PIX_FMT_YUVJ440P \
95 || (x)==PIX_FMT_YUVJ444P \
96 || (x)==PIX_FMT_PAL8 \
97 || (x)==PIX_FMT_BGR8 \
98 || (x)==PIX_FMT_RGB8 \
99 || (x)==PIX_FMT_BGR4_BYTE \
100 || (x)==PIX_FMT_RGB4_BYTE \
101 || (x)==PIX_FMT_YUV440P \
102 || (x)==PIX_FMT_MONOWHITE \
103 || (x)==PIX_FMT_MONOBLACK \
104 || (x)==PIX_FMT_YUV420P16LE \
105 || (x)==PIX_FMT_YUV422P16LE \
106 || (x)==PIX_FMT_YUV444P16LE \
107 || (x)==PIX_FMT_YUV420P16BE \
108 || (x)==PIX_FMT_YUV422P16BE \
109 || (x)==PIX_FMT_YUV444P16BE \
110 )
111
112 int sws_isSupportedInput(enum PixelFormat pix_fmt)
113 {
114 return isSupportedIn(pix_fmt);
115 }
116
117 #define isSupportedOut(x) ( \
118 (x)==PIX_FMT_YUV420P \
119 || (x)==PIX_FMT_YUVA420P \
120 || (x)==PIX_FMT_YUYV422 \
121 || (x)==PIX_FMT_UYVY422 \
122 || (x)==PIX_FMT_YUV444P \
123 || (x)==PIX_FMT_YUV422P \
124 || (x)==PIX_FMT_YUV411P \
125 || (x)==PIX_FMT_YUVJ420P \
126 || (x)==PIX_FMT_YUVJ422P \
127 || (x)==PIX_FMT_YUVJ440P \
128 || (x)==PIX_FMT_YUVJ444P \
129 || isAnyRGB(x) \
130 || (x)==PIX_FMT_NV12 \
131 || (x)==PIX_FMT_NV21 \
132 || (x)==PIX_FMT_GRAY16BE \
133 || (x)==PIX_FMT_GRAY16LE \
134 || (x)==PIX_FMT_GRAY8 \
135 || (x)==PIX_FMT_YUV410P \
136 || (x)==PIX_FMT_YUV440P \
137 || (x)==PIX_FMT_YUV420P16LE \
138 || (x)==PIX_FMT_YUV422P16LE \
139 || (x)==PIX_FMT_YUV444P16LE \
140 || (x)==PIX_FMT_YUV420P16BE \
141 || (x)==PIX_FMT_YUV422P16BE \
142 || (x)==PIX_FMT_YUV444P16BE \
143 )
144
145 int sws_isSupportedOutput(enum PixelFormat pix_fmt)
146 {
147 return isSupportedOut(pix_fmt);
148 }
149
150 extern const int32_t ff_yuv2rgb_coeffs[8][4];
151
152 const char *sws_format_name(enum PixelFormat format)
153 {
154 if ((unsigned)format < PIX_FMT_NB && av_pix_fmt_descriptors[format].name)
155 return av_pix_fmt_descriptors[format].name;
156 else
157 return "Unknown format";
158 }
159
160 static double getSplineCoeff(double a, double b, double c, double d, double dist)
161 {
162 // printf("%f %f %f %f %f\n", a,b,c,d,dist);
163 if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
164 else return getSplineCoeff( 0.0,
165 b+ 2.0*c + 3.0*d,
166 c + 3.0*d,
167 -b- 3.0*c - 6.0*d,
168 dist-1.0);
169 }
170
171 static int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
172 int srcW, int dstW, int filterAlign, int one, int flags,
173 SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
174 {
175 int i;
176 int filterSize;
177 int filter2Size;
178 int minFilterSize;
179 int64_t *filter=NULL;
180 int64_t *filter2=NULL;
181 const int64_t fone= 1LL<<54;
182 int ret= -1;
183 #if ARCH_X86
184 if (flags & SWS_CPU_CAPS_MMX)
185 __asm__ volatile("emms\n\t"::: "memory"); //FIXME this should not be required but it IS (even for non-MMX versions)
186 #endif
187
188 // NOTE: the +1 is for the MMX scaler which reads over the end
189 FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW+1)*sizeof(int16_t), fail);
190
191 if (FFABS(xInc - 0x10000) <10) { // unscaled
192 int i;
193 filterSize= 1;
194 FF_ALLOCZ_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
195
196 for (i=0; i<dstW; i++) {
197 filter[i*filterSize]= fone;
198 (*filterPos)[i]=i;
199 }
200
201 } else if (flags&SWS_POINT) { // lame looking point sampling mode
202 int i;
203 int xDstInSrc;
204 filterSize= 1;
205 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
206
207 xDstInSrc= xInc/2 - 0x8000;
208 for (i=0; i<dstW; i++) {
209 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
210
211 (*filterPos)[i]= xx;
212 filter[i]= fone;
213 xDstInSrc+= xInc;
214 }
215 } else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) { // bilinear upscale
216 int i;
217 int xDstInSrc;
218 filterSize= 2;
219 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
220
221 xDstInSrc= xInc/2 - 0x8000;
222 for (i=0; i<dstW; i++) {
223 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
224 int j;
225
226 (*filterPos)[i]= xx;
227 //bilinear upscale / linear interpolate / area averaging
228 for (j=0; j<filterSize; j++) {
229 int64_t coeff= fone - FFABS((xx<<16) - xDstInSrc)*(fone>>16);
230 if (coeff<0) coeff=0;
231 filter[i*filterSize + j]= coeff;
232 xx++;
233 }
234 xDstInSrc+= xInc;
235 }
236 } else {
237 int xDstInSrc;
238 int sizeFactor;
239
240 if (flags&SWS_BICUBIC) sizeFactor= 4;
241 else if (flags&SWS_X) sizeFactor= 8;
242 else if (flags&SWS_AREA) sizeFactor= 1; //downscale only, for upscale it is bilinear
243 else if (flags&SWS_GAUSS) sizeFactor= 8; // infinite ;)
244 else if (flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? ceil(2*param[0]) : 6;
245 else if (flags&SWS_SINC) sizeFactor= 20; // infinite ;)
246 else if (flags&SWS_SPLINE) sizeFactor= 20; // infinite ;)
247 else if (flags&SWS_BILINEAR) sizeFactor= 2;
248 else {
249 sizeFactor= 0; //GCC warning killer
250 assert(0);
251 }
252
253 if (xInc <= 1<<16) filterSize= 1 + sizeFactor; // upscale
254 else filterSize= 1 + (sizeFactor*srcW + dstW - 1)/ dstW;
255
256 if (filterSize > srcW-2) filterSize=srcW-2;
257
258 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
259
260 xDstInSrc= xInc - 0x10000;
261 for (i=0; i<dstW; i++) {
262 int xx= (xDstInSrc - ((filterSize-2)<<16)) / (1<<17);
263 int j;
264 (*filterPos)[i]= xx;
265 for (j=0; j<filterSize; j++) {
266 int64_t d= ((int64_t)FFABS((xx<<17) - xDstInSrc))<<13;
267 double floatd;
268 int64_t coeff;
269
270 if (xInc > 1<<16)
271 d= d*dstW/srcW;
272 floatd= d * (1.0/(1<<30));
273
274 if (flags & SWS_BICUBIC) {
275 int64_t B= (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1<<24);
276 int64_t C= (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1<<24);
277 int64_t dd = ( d*d)>>30;
278 int64_t ddd= (dd*d)>>30;
279
280 if (d < 1LL<<30)
281 coeff = (12*(1<<24)-9*B-6*C)*ddd + (-18*(1<<24)+12*B+6*C)*dd + (6*(1<<24)-2*B)*(1<<30);
282 else if (d < 1LL<<31)
283 coeff = (-B-6*C)*ddd + (6*B+30*C)*dd + (-12*B-48*C)*d + (8*B+24*C)*(1<<30);
284 else
285 coeff=0.0;
286 coeff *= fone>>(30+24);
287 }
288 /* else if (flags & SWS_X) {
289 double p= param ? param*0.01 : 0.3;
290 coeff = d ? sin(d*M_PI)/(d*M_PI) : 1.0;
291 coeff*= pow(2.0, - p*d*d);
292 }*/
293 else if (flags & SWS_X) {
294 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
295 double c;
296
297 if (floatd<1.0)
298 c = cos(floatd*M_PI);
299 else
300 c=-1.0;
301 if (c<0.0) c= -pow(-c, A);
302 else c= pow( c, A);
303 coeff= (c*0.5 + 0.5)*fone;
304 } else if (flags & SWS_AREA) {
305 int64_t d2= d - (1<<29);
306 if (d2*xInc < -(1LL<<(29+16))) coeff= 1.0 * (1LL<<(30+16));
307 else if (d2*xInc < (1LL<<(29+16))) coeff= -d2*xInc + (1LL<<(29+16));
308 else coeff=0.0;
309 coeff *= fone>>(30+16);
310 } else if (flags & SWS_GAUSS) {
311 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
312 coeff = (pow(2.0, - p*floatd*floatd))*fone;
313 } else if (flags & SWS_SINC) {
314 coeff = (d ? sin(floatd*M_PI)/(floatd*M_PI) : 1.0)*fone;
315 } else if (flags & SWS_LANCZOS) {
316 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
317 coeff = (d ? sin(floatd*M_PI)*sin(floatd*M_PI/p)/(floatd*floatd*M_PI*M_PI/p) : 1.0)*fone;
318 if (floatd>p) coeff=0;
319 } else if (flags & SWS_BILINEAR) {
320 coeff= (1<<30) - d;
321 if (coeff<0) coeff=0;
322 coeff *= fone >> 30;
323 } else if (flags & SWS_SPLINE) {
324 double p=-2.196152422706632;
325 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, floatd) * fone;
326 } else {
327 coeff= 0.0; //GCC warning killer
328 assert(0);
329 }
330
331 filter[i*filterSize + j]= coeff;
332 xx++;
333 }
334 xDstInSrc+= 2*xInc;
335 }
336 }
337
338 /* apply src & dst Filter to filter -> filter2
339 av_free(filter);
340 */
341 assert(filterSize>0);
342 filter2Size= filterSize;
343 if (srcFilter) filter2Size+= srcFilter->length - 1;
344 if (dstFilter) filter2Size+= dstFilter->length - 1;
345 assert(filter2Size>0);
346 FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size*dstW*sizeof(*filter2), fail);
347
348 for (i=0; i<dstW; i++) {
349 int j, k;
350
351 if(srcFilter) {
352 for (k=0; k<srcFilter->length; k++) {
353 for (j=0; j<filterSize; j++)
354 filter2[i*filter2Size + k + j] += srcFilter->coeff[k]*filter[i*filterSize + j];
355 }
356 } else {
357 for (j=0; j<filterSize; j++)
358 filter2[i*filter2Size + j]= filter[i*filterSize + j];
359 }
360 //FIXME dstFilter
361
362 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
363 }
364 av_freep(&filter);
365
366 /* try to reduce the filter-size (step1 find size and shift left) */
367 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
368 minFilterSize= 0;
369 for (i=dstW-1; i>=0; i--) {
370 int min= filter2Size;
371 int j;
372 int64_t cutOff=0.0;
373
374 /* get rid of near zero elements on the left by shifting left */
375 for (j=0; j<filter2Size; j++) {
376 int k;
377 cutOff += FFABS(filter2[i*filter2Size]);
378
379 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
380
381 /* preserve monotonicity because the core can't handle the filter otherwise */
382 if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
383
384 // move filter coefficients left
385 for (k=1; k<filter2Size; k++)
386 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
387 filter2[i*filter2Size + k - 1]= 0;
388 (*filterPos)[i]++;
389 }
390
391 cutOff=0;
392 /* count near zeros on the right */
393 for (j=filter2Size-1; j>0; j--) {
394 cutOff += FFABS(filter2[i*filter2Size + j]);
395
396 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
397 min--;
398 }
399
400 if (min>minFilterSize) minFilterSize= min;
401 }
402
403 if (flags & SWS_CPU_CAPS_ALTIVEC) {
404 // we can handle the special case 4,
405 // so we don't want to go to the full 8
406 if (minFilterSize < 5)
407 filterAlign = 4;
408
409 // We really don't want to waste our time
410 // doing useless computation, so fall back on
411 // the scalar C code for very small filters.
412 // Vectorizing is worth it only if you have a
413 // decent-sized vector.
414 if (minFilterSize < 3)
415 filterAlign = 1;
416 }
417
418 if (flags & SWS_CPU_CAPS_MMX) {
419 // special case for unscaled vertical filtering
420 if (minFilterSize == 1 && filterAlign == 2)
421 filterAlign= 1;
422 }
423
424 assert(minFilterSize > 0);
425 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
426 assert(filterSize > 0);
427 filter= av_malloc(filterSize*dstW*sizeof(*filter));
428 if (filterSize >= MAX_FILTER_SIZE*16/((flags&SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
429 goto fail;
430 *outFilterSize= filterSize;
431
432 if (flags&SWS_PRINT_INFO)
433 av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
434 /* try to reduce the filter-size (step2 reduce it) */
435 for (i=0; i<dstW; i++) {
436 int j;
437
438 for (j=0; j<filterSize; j++) {
439 if (j>=filter2Size) filter[i*filterSize + j]= 0;
440 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
441 if((flags & SWS_BITEXACT) && j>=minFilterSize)
442 filter[i*filterSize + j]= 0;
443 }
444 }
445
446 //FIXME try to align filterPos if possible
447
448 //fix borders
449 for (i=0; i<dstW; i++) {
450 int j;
451 if ((*filterPos)[i] < 0) {
452 // move filter coefficients left to compensate for filterPos
453 for (j=1; j<filterSize; j++) {
454 int left= FFMAX(j + (*filterPos)[i], 0);
455 filter[i*filterSize + left] += filter[i*filterSize + j];
456 filter[i*filterSize + j]=0;
457 }
458 (*filterPos)[i]= 0;
459 }
460
461 if ((*filterPos)[i] + filterSize > srcW) {
462 int shift= (*filterPos)[i] + filterSize - srcW;
463 // move filter coefficients right to compensate for filterPos
464 for (j=filterSize-2; j>=0; j--) {
465 int right= FFMIN(j + shift, filterSize-1);
466 filter[i*filterSize +right] += filter[i*filterSize +j];
467 filter[i*filterSize +j]=0;
468 }
469 (*filterPos)[i]= srcW - filterSize;
470 }
471 }
472
473 // Note the +1 is for the MMX scaler which reads over the end
474 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
475 FF_ALLOCZ_OR_GOTO(NULL, *outFilter, *outFilterSize*(dstW+1)*sizeof(int16_t), fail);
476
477 /* normalize & store in outFilter */
478 for (i=0; i<dstW; i++) {
479 int j;
480 int64_t error=0;
481 int64_t sum=0;
482
483 for (j=0; j<filterSize; j++) {
484 sum+= filter[i*filterSize + j];
485 }
486 sum= (sum + one/2)/ one;
487 for (j=0; j<*outFilterSize; j++) {
488 int64_t v= filter[i*filterSize + j] + error;
489 int intV= ROUNDED_DIV(v, sum);
490 (*outFilter)[i*(*outFilterSize) + j]= intV;
491 error= v - intV*sum;
492 }
493 }
494
495 (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
496 for (i=0; i<*outFilterSize; i++) {
497 int j= dstW*(*outFilterSize);
498 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
499 }
500
501 ret=0;
502 fail:
503 av_free(filter);
504 av_free(filter2);
505 return ret;
506 }
507
508 #if ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT)
509 static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode, int16_t *filter, int32_t *filterPos, int numSplits)
510 {
511 uint8_t *fragmentA;
512 x86_reg imm8OfPShufW1A;
513 x86_reg imm8OfPShufW2A;
514 x86_reg fragmentLengthA;
515 uint8_t *fragmentB;
516 x86_reg imm8OfPShufW1B;
517 x86_reg imm8OfPShufW2B;
518 x86_reg fragmentLengthB;
519 int fragmentPos;
520
521 int xpos, i;
522
523 // create an optimized horizontal scaling routine
524 /* This scaler is made of runtime-generated MMX2 code using specially
525 * tuned pshufw instructions. For every four output pixels, if four
526 * input pixels are enough for the fast bilinear scaling, then a chunk
527 * of fragmentB is used. If five input pixels are needed, then a chunk
528 * of fragmentA is used.
529 */
530
531 //code fragment
532
533 __asm__ volatile(
534 "jmp 9f \n\t"
535 // Begin
536 "0: \n\t"
537 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
538 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
539 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
540 "punpcklbw %%mm7, %%mm1 \n\t"
541 "punpcklbw %%mm7, %%mm0 \n\t"
542 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
543 "1: \n\t"
544 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
545 "2: \n\t"
546 "psubw %%mm1, %%mm0 \n\t"
547 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
548 "pmullw %%mm3, %%mm0 \n\t"
549 "psllw $7, %%mm1 \n\t"
550 "paddw %%mm1, %%mm0 \n\t"
551
552 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
553
554 "add $8, %%"REG_a" \n\t"
555 // End
556 "9: \n\t"
557 // "int $3 \n\t"
558 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
559 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
560 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
561 "dec %1 \n\t"
562 "dec %2 \n\t"
563 "sub %0, %1 \n\t"
564 "sub %0, %2 \n\t"
565 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
566 "sub %0, %3 \n\t"
567
568
569 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
570 "=r" (fragmentLengthA)
571 );
572
573 __asm__ volatile(
574 "jmp 9f \n\t"
575 // Begin
576 "0: \n\t"
577 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
578 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
579 "punpcklbw %%mm7, %%mm0 \n\t"
580 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
581 "1: \n\t"
582 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
583 "2: \n\t"
584 "psubw %%mm1, %%mm0 \n\t"
585 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
586 "pmullw %%mm3, %%mm0 \n\t"
587 "psllw $7, %%mm1 \n\t"
588 "paddw %%mm1, %%mm0 \n\t"
589
590 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
591
592 "add $8, %%"REG_a" \n\t"
593 // End
594 "9: \n\t"
595 // "int $3 \n\t"
596 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
597 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
598 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
599 "dec %1 \n\t"
600 "dec %2 \n\t"
601 "sub %0, %1 \n\t"
602 "sub %0, %2 \n\t"
603 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
604 "sub %0, %3 \n\t"
605
606
607 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
608 "=r" (fragmentLengthB)
609 );
610
611 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
612 fragmentPos=0;
613
614 for (i=0; i<dstW/numSplits; i++) {
615 int xx=xpos>>16;
616
617 if ((i&3) == 0) {
618 int a=0;
619 int b=((xpos+xInc)>>16) - xx;
620 int c=((xpos+xInc*2)>>16) - xx;
621 int d=((xpos+xInc*3)>>16) - xx;
622 int inc = (d+1<4);
623 uint8_t *fragment = (d+1<4) ? fragmentB : fragmentA;
624 x86_reg imm8OfPShufW1 = (d+1<4) ? imm8OfPShufW1B : imm8OfPShufW1A;
625 x86_reg imm8OfPShufW2 = (d+1<4) ? imm8OfPShufW2B : imm8OfPShufW2A;
626 x86_reg fragmentLength = (d+1<4) ? fragmentLengthB : fragmentLengthA;
627 int maxShift= 3-(d+inc);
628 int shift=0;
629
630 if (filterCode) {
631 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
632 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
633 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
634 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
635 filterPos[i/2]= xx;
636
637 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
638
639 filterCode[fragmentPos + imm8OfPShufW1]=
640 (a+inc) | ((b+inc)<<2) | ((c+inc)<<4) | ((d+inc)<<6);
641 filterCode[fragmentPos + imm8OfPShufW2]=
642 a | (b<<2) | (c<<4) | (d<<6);
643
644 if (i+4-inc>=dstW) shift=maxShift; //avoid overread
645 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
646
647 if (shift && i>=shift) {
648 filterCode[fragmentPos + imm8OfPShufW1]+= 0x55*shift;
649 filterCode[fragmentPos + imm8OfPShufW2]+= 0x55*shift;
650 filterPos[i/2]-=shift;
651 }
652 }
653
654 fragmentPos+= fragmentLength;
655
656 if (filterCode)
657 filterCode[fragmentPos]= RET;
658 }
659 xpos+=xInc;
660 }
661 if (filterCode)
662 filterPos[((i/2)+1)&(~1)]= xpos>>16; // needed to jump to the next part
663
664 return fragmentPos + 1;
665 }
666 #endif /* ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT) */
667
668 static void getSubSampleFactors(int *h, int *v, enum PixelFormat format)
669 {
670 *h = av_pix_fmt_descriptors[format].log2_chroma_w;
671 *v = av_pix_fmt_descriptors[format].log2_chroma_h;
672 }
673
674 static uint16_t roundToInt16(int64_t f)
675 {
676 int r= (f + (1<<15))>>16;
677 if (r<-0x7FFF) return 0x8000;
678 else if (r> 0x7FFF) return 0x7FFF;
679 else return r;
680 }
681
682 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation)
683 {
684 int64_t crv = inv_table[0];
685 int64_t cbu = inv_table[1];
686 int64_t cgu = -inv_table[2];
687 int64_t cgv = -inv_table[3];
688 int64_t cy = 1<<16;
689 int64_t oy = 0;
690
691 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
692 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
693
694 c->brightness= brightness;
695 c->contrast = contrast;
696 c->saturation= saturation;
697 c->srcRange = srcRange;
698 c->dstRange = dstRange;
699 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
700
701 c->uOffset= 0x0400040004000400LL;
702 c->vOffset= 0x0400040004000400LL;
703
704 if (!srcRange) {
705 cy= (cy*255) / 219;
706 oy= 16<<16;
707 } else {
708 crv= (crv*224) / 255;
709 cbu= (cbu*224) / 255;
710 cgu= (cgu*224) / 255;
711 cgv= (cgv*224) / 255;
712 }
713
714 cy = (cy *contrast )>>16;
715 crv= (crv*contrast * saturation)>>32;
716 cbu= (cbu*contrast * saturation)>>32;
717 cgu= (cgu*contrast * saturation)>>32;
718 cgv= (cgv*contrast * saturation)>>32;
719
720 oy -= 256*brightness;
721
722 c->yCoeff= roundToInt16(cy *8192) * 0x0001000100010001ULL;
723 c->vrCoeff= roundToInt16(crv*8192) * 0x0001000100010001ULL;
724 c->ubCoeff= roundToInt16(cbu*8192) * 0x0001000100010001ULL;
725 c->vgCoeff= roundToInt16(cgv*8192) * 0x0001000100010001ULL;
726 c->ugCoeff= roundToInt16(cgu*8192) * 0x0001000100010001ULL;
727 c->yOffset= roundToInt16(oy * 8) * 0x0001000100010001ULL;
728
729 c->yuv2rgb_y_coeff = (int16_t)roundToInt16(cy <<13);
730 c->yuv2rgb_y_offset = (int16_t)roundToInt16(oy << 9);
731 c->yuv2rgb_v2r_coeff= (int16_t)roundToInt16(crv<<13);
732 c->yuv2rgb_v2g_coeff= (int16_t)roundToInt16(cgv<<13);
733 c->yuv2rgb_u2g_coeff= (int16_t)roundToInt16(cgu<<13);
734 c->yuv2rgb_u2b_coeff= (int16_t)roundToInt16(cbu<<13);
735
736 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
737 //FIXME factorize
738
739 #if HAVE_ALTIVEC
740 if (c->flags & SWS_CPU_CAPS_ALTIVEC)
741 ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness, contrast, saturation);
742 #endif
743 return 0;
744 }
745
746 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation)
747 {
748 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
749
750 *inv_table = c->srcColorspaceTable;
751 *table = c->dstColorspaceTable;
752 *srcRange = c->srcRange;
753 *dstRange = c->dstRange;
754 *brightness= c->brightness;
755 *contrast = c->contrast;
756 *saturation= c->saturation;
757
758 return 0;
759 }
760
761 static int handle_jpeg(enum PixelFormat *format)
762 {
763 switch (*format) {
764 case PIX_FMT_YUVJ420P: *format = PIX_FMT_YUV420P; return 1;
765 case PIX_FMT_YUVJ422P: *format = PIX_FMT_YUV422P; return 1;
766 case PIX_FMT_YUVJ444P: *format = PIX_FMT_YUV444P; return 1;
767 case PIX_FMT_YUVJ440P: *format = PIX_FMT_YUV440P; return 1;
768 default: return 0;
769 }
770 }
771
772 SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,
773 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
774 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
775 {
776 SwsContext *c;
777 int i;
778 int usesVFilter, usesHFilter;
779 int unscaled;
780 int srcRange, dstRange;
781 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
782 #if ARCH_X86
783 if (flags & SWS_CPU_CAPS_MMX)
784 __asm__ volatile("emms\n\t"::: "memory");
785 #endif
786
787 #if !CONFIG_RUNTIME_CPUDETECT //ensure that the flags match the compiled variant if cpudetect is off
788 flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC|SWS_CPU_CAPS_BFIN);
789 flags |= ff_hardcodedcpuflags();
790 #endif /* CONFIG_RUNTIME_CPUDETECT */
791 if (!rgb15to16) sws_rgb2rgb_init(flags);
792
793 unscaled = (srcW == dstW && srcH == dstH);
794
795 srcRange = handle_jpeg(&srcFormat);
796 dstRange = handle_jpeg(&dstFormat);
797
798 if (!isSupportedIn(srcFormat)) {
799 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as input pixel format\n", sws_format_name(srcFormat));
800 return NULL;
801 }
802 if (!isSupportedOut(dstFormat)) {
803 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as output pixel format\n", sws_format_name(dstFormat));
804 return NULL;
805 }
806
807 i= flags & ( SWS_POINT
808 |SWS_AREA
809 |SWS_BILINEAR
810 |SWS_FAST_BILINEAR
811 |SWS_BICUBIC
812 |SWS_X
813 |SWS_GAUSS
814 |SWS_LANCZOS
815 |SWS_SINC
816 |SWS_SPLINE
817 |SWS_BICUBLIN);
818 if(!i || (i & (i-1))) {
819 av_log(NULL, AV_LOG_ERROR, "swScaler: Exactly one scaler algorithm must be chosen\n");
820 return NULL;
821 }
822
823 /* sanity check */
824 if (srcW<4 || srcH<1 || dstW<8 || dstH<1) { //FIXME check if these are enough and try to lowwer them after fixing the relevant parts of the code
825 av_log(NULL, AV_LOG_ERROR, "swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
826 srcW, srcH, dstW, dstH);
827 return NULL;
828 }
829 if(srcW > VOFW || dstW > VOFW) {
830 av_log(NULL, AV_LOG_ERROR, "swScaler: Compile-time maximum width is "AV_STRINGIFY(VOFW)" change VOF/VOFW and recompile\n");
831 return NULL;
832 }
833
834 if (!dstFilter) dstFilter= &dummyFilter;
835 if (!srcFilter) srcFilter= &dummyFilter;
836
837 FF_ALLOCZ_OR_GOTO(NULL, c, sizeof(SwsContext), fail);
838
839 c->av_class = &sws_context_class;
840 c->srcW= srcW;
841 c->srcH= srcH;
842 c->dstW= dstW;
843 c->dstH= dstH;
844 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
845 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
846 c->flags= flags;
847 c->dstFormat= dstFormat;
848 c->srcFormat= srcFormat;
849 c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[dstFormat]);
850 c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[srcFormat]);
851 c->vRounder= 4* 0x0001000100010001ULL;
852
853 usesVFilter = (srcFilter->lumV && srcFilter->lumV->length>1) ||
854 (srcFilter->chrV && srcFilter->chrV->length>1) ||
855 (dstFilter->lumV && dstFilter->lumV->length>1) ||
856 (dstFilter->chrV && dstFilter->chrV->length>1);
857 usesHFilter = (srcFilter->lumH && srcFilter->lumH->length>1) ||
858 (srcFilter->chrH && srcFilter->chrH->length>1) ||
859 (dstFilter->lumH && dstFilter->lumH->length>1) ||
860 (dstFilter->chrH && dstFilter->chrH->length>1);
861
862 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
863 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
864
865 // reuse chroma for 2 pixels RGB/BGR unless user wants full chroma interpolation
866 if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
867
868 // drop some chroma lines if the user wants it
869 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
870 c->chrSrcVSubSample+= c->vChrDrop;
871
872 // drop every other pixel for chroma calculation unless user wants full chroma
873 if (isAnyRGB(srcFormat) && !(flags&SWS_FULL_CHR_H_INP)
874 && srcFormat!=PIX_FMT_RGB8 && srcFormat!=PIX_FMT_BGR8
875 && srcFormat!=PIX_FMT_RGB4 && srcFormat!=PIX_FMT_BGR4
876 && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE
877 && ((dstW>>c->chrDstHSubSample) <= (srcW>>1) || (flags&(SWS_FAST_BILINEAR|SWS_POINT))))
878 c->chrSrcHSubSample=1;
879
880 if (param) {
881 c->param[0] = param[0];
882 c->param[1] = param[1];
883 } else {
884 c->param[0] =
885 c->param[1] = SWS_PARAM_DEFAULT;
886 }
887
888 // Note the -((-x)>>y) is so that we always round toward +inf.
889 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
890 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
891 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
892 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
893
894 sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], srcRange, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/, dstRange, 0, 1<<16, 1<<16);
895
896 /* unscaled special cases */
897 if (unscaled && !usesHFilter && !usesVFilter && (srcRange == dstRange || isAnyRGB(dstFormat))) {
898 ff_get_unscaled_swscale(c);
899
900 if (c->swScale) {
901 if (flags&SWS_PRINT_INFO)
902 av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
903 sws_format_name(srcFormat), sws_format_name(dstFormat));
904 return c;
905 }
906 }
907
908 if (flags & SWS_CPU_CAPS_MMX2) {
909 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
910 if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR)) {
911 if (flags&SWS_PRINT_INFO)
912 av_log(c, AV_LOG_INFO, "output width is not a multiple of 32 -> no MMX2 scaler\n");
913 }
914 if (usesHFilter) c->canMMX2BeUsed=0;
915 }
916 else
917 c->canMMX2BeUsed=0;
918
919 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
920 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
921
922 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
923 // but only for the FAST_BILINEAR mode otherwise do correct scaling
924 // n-2 is the last chrominance sample available
925 // this is not perfect, but no one should notice the difference, the more correct variant
926 // would be like the vertical one, but that would require some special code for the
927 // first and last pixel
928 if (flags&SWS_FAST_BILINEAR) {
929 if (c->canMMX2BeUsed) {
930 c->lumXInc+= 20;
931 c->chrXInc+= 20;
932 }
933 //we don't use the x86 asm scaler if MMX is available
934 else if (flags & SWS_CPU_CAPS_MMX) {
935 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
936 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
937 }
938 }
939
940 /* precalculate horizontal scaler filter coefficients */
941 {
942 #if ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT)
943 // can't downscale !!!
944 if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {
945 c->lumMmx2FilterCodeSize = initMMX2HScaler( dstW, c->lumXInc, NULL, NULL, NULL, 8);
946 c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc, NULL, NULL, NULL, 4);
947
948 #ifdef MAP_ANONYMOUS
949 c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
950 c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
951 #elif HAVE_VIRTUALALLOC
952 c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
953 c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
954 #else
955 c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);
956 c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);
957 #endif
958
959 if (!c->lumMmx2FilterCode || !c->chrMmx2FilterCode)
960 goto fail;
961 FF_ALLOCZ_OR_GOTO(c, c->hLumFilter , (dstW /8+8)*sizeof(int16_t), fail);
962 FF_ALLOCZ_OR_GOTO(c, c->hChrFilter , (c->chrDstW /4+8)*sizeof(int16_t), fail);
963 FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW /2/8+8)*sizeof(int32_t), fail);
964 FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW/2/4+8)*sizeof(int32_t), fail);
965
966 initMMX2HScaler( dstW, c->lumXInc, c->lumMmx2FilterCode, c->hLumFilter, c->hLumFilterPos, 8);
967 initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode, c->hChrFilter, c->hChrFilterPos, 4);
968
969 #ifdef MAP_ANONYMOUS
970 mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
971 mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
972 #endif
973 } else
974 #endif /* ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT) */
975 {
976 const int filterAlign=
977 (flags & SWS_CPU_CAPS_MMX) ? 4 :
978 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
979 1;
980
981 if (initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
982 srcW , dstW, filterAlign, 1<<14,
983 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
984 srcFilter->lumH, dstFilter->lumH, c->param) < 0)
985 goto fail;
986 if (initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
987 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
988 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
989 srcFilter->chrH, dstFilter->chrH, c->param) < 0)
990 goto fail;
991 }
992 } // initialize horizontal stuff
993
994 /* precalculate vertical scaler filter coefficients */
995 {
996 const int filterAlign=
997 (flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
998 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
999 1;
1000
1001 if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
1002 srcH , dstH, filterAlign, (1<<12),
1003 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
1004 srcFilter->lumV, dstFilter->lumV, c->param) < 0)
1005 goto fail;
1006 if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
1007 c->chrSrcH, c->chrDstH, filterAlign, (1<<12),
1008 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
1009 srcFilter->chrV, dstFilter->chrV, c->param) < 0)
1010 goto fail;
1011
1012 #if HAVE_ALTIVEC
1013 FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof (vector signed short)*c->vLumFilterSize*c->dstH, fail);
1014 FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH, fail);
1015
1016 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
1017 int j;
1018 short *p = (short *)&c->vYCoeffsBank[i];
1019 for (j=0;j<8;j++)
1020 p[j] = c->vLumFilter[i];
1021 }
1022
1023 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
1024 int j;
1025 short *p = (short *)&c->vCCoeffsBank[i];
1026 for (j=0;j<8;j++)
1027 p[j] = c->vChrFilter[i];
1028 }
1029 #endif
1030 }
1031
1032 // calculate buffer sizes so that they won't run out while handling these damn slices
1033 c->vLumBufSize= c->vLumFilterSize;
1034 c->vChrBufSize= c->vChrFilterSize;
1035 for (i=0; i<dstH; i++) {
1036 int chrI= i*c->chrDstH / dstH;
1037 int nextSlice= FFMAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
1038 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
1039
1040 nextSlice>>= c->chrSrcVSubSample;
1041 nextSlice<<= c->chrSrcVSubSample;
1042 if (c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
1043 c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
1044 if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
1045 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
1046 }
1047
1048 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
1049 // allocate several megabytes to handle all possible cases)
1050 FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
1051 FF_ALLOC_OR_GOTO(c, c->chrPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
1052 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
1053 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
1054 //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)
1055 /* align at 16 bytes for AltiVec */
1056 for (i=0; i<c->vLumBufSize; i++) {
1057 FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], VOF+1, fail);
1058 c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];
1059 }
1060 for (i=0; i<c->vChrBufSize; i++) {
1061 FF_ALLOC_OR_GOTO(c, c->chrPixBuf[i+c->vChrBufSize], (VOF+1)*2, fail);
1062 c->chrPixBuf[i] = c->chrPixBuf[i+c->vChrBufSize];
1063 }
1064 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
1065 for (i=0; i<c->vLumBufSize; i++) {
1066 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], VOF+1, fail);
1067 c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];
1068 }
1069
1070 //try to avoid drawing green stuff between the right end and the stride end
1071 for (i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, (VOF+1)*2);
1072
1073 assert(2*VOFW == VOF);
1074
1075 assert(c->chrDstH <= dstH);
1076
1077 if (flags&SWS_PRINT_INFO) {
1078 if (flags&SWS_FAST_BILINEAR)
1079 av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
1080 else if (flags&SWS_BILINEAR)
1081 av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
1082 else if (flags&SWS_BICUBIC)
1083 av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
1084 else if (flags&SWS_X)
1085 av_log(c, AV_LOG_INFO, "Experimental scaler, ");
1086 else if (flags&SWS_POINT)
1087 av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
1088 else if (flags&SWS_AREA)
1089 av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
1090 else if (flags&SWS_BICUBLIN)
1091 av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
1092 else if (flags&SWS_GAUSS)
1093 av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
1094 else if (flags&SWS_SINC)
1095 av_log(c, AV_LOG_INFO, "Sinc scaler, ");
1096 else if (flags&SWS_LANCZOS)
1097 av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
1098 else if (flags&SWS_SPLINE)
1099 av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
1100 else
1101 av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
1102
1103 av_log(c, AV_LOG_INFO, "from %s to %s%s ",
1104 sws_format_name(srcFormat),
1105 #ifdef DITHER1XBPP
1106 dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ||
1107 dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||
1108 dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ? "dithered " : "",
1109 #else
1110 "",
1111 #endif
1112 sws_format_name(dstFormat));
1113
1114 if (flags & SWS_CPU_CAPS_MMX2)
1115 av_log(c, AV_LOG_INFO, "using MMX2\n");
1116 else if (flags & SWS_CPU_CAPS_3DNOW)
1117 av_log(c, AV_LOG_INFO, "using 3DNOW\n");
1118 else if (flags & SWS_CPU_CAPS_MMX)
1119 av_log(c, AV_LOG_INFO, "using MMX\n");
1120 else if (flags & SWS_CPU_CAPS_ALTIVEC)
1121 av_log(c, AV_LOG_INFO, "using AltiVec\n");
1122 else
1123 av_log(c, AV_LOG_INFO, "using C\n");
1124
1125 if (flags & SWS_CPU_CAPS_MMX) {
1126 if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
1127 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
1128 else {
1129 if (c->hLumFilterSize==4)
1130 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal luminance scaling\n");
1131 else if (c->hLumFilterSize==8)
1132 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal luminance scaling\n");
1133 else
1134 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal luminance scaling\n");
1135
1136 if (c->hChrFilterSize==4)
1137 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal chrominance scaling\n");
1138 else if (c->hChrFilterSize==8)
1139 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal chrominance scaling\n");
1140 else
1141 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal chrominance scaling\n");
1142 }
1143 } else {
1144 #if ARCH_X86
1145 av_log(c, AV_LOG_VERBOSE, "using x86 asm scaler for horizontal scaling\n");
1146 #else
1147 if (flags & SWS_FAST_BILINEAR)
1148 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR C scaler for horizontal scaling\n");
1149 else
1150 av_log(c, AV_LOG_VERBOSE, "using C scaler for horizontal scaling\n");
1151 #endif
1152 }
1153 if (isPlanarYUV(dstFormat)) {
1154 if (c->vLumFilterSize==1)
1155 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1156 else
1157 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1158 } else {
1159 if (c->vLumFilterSize==1 && c->vChrFilterSize==2)
1160 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
1161 " 2-tap scaler for vertical chrominance scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1162 else if (c->vLumFilterSize==2 && c->vChrFilterSize==2)
1163 av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1164 else
1165 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1166 }
1167
1168 if (dstFormat==PIX_FMT_BGR24)
1169 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 converter\n",
1170 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
1171 else if (dstFormat==PIX_FMT_RGB32)
1172 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1173 else if (dstFormat==PIX_FMT_BGR565)
1174 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1175 else if (dstFormat==PIX_FMT_BGR555)
1176 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1177 else if (dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||
1178 dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE)
1179 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR12 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1180
1181 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
1182 av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1183 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
1184 av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1185 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
1186 }
1187
1188 c->swScale= ff_getSwsFunc(c);
1189 return c;
1190
1191 fail:
1192 sws_freeContext(c);
1193 return NULL;
1194 }
1195
1196 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
1197 float lumaSharpen, float chromaSharpen,
1198 float chromaHShift, float chromaVShift,
1199 int verbose)
1200 {
1201 SwsFilter *filter= av_malloc(sizeof(SwsFilter));
1202 if (!filter)
1203 return NULL;
1204
1205 if (lumaGBlur!=0.0) {
1206 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
1207 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
1208 } else {
1209 filter->lumH= sws_getIdentityVec();
1210 filter->lumV= sws_getIdentityVec();
1211 }
1212
1213 if (chromaGBlur!=0.0) {
1214 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
1215 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
1216 } else {
1217 filter->chrH= sws_getIdentityVec();
1218 filter->chrV= sws_getIdentityVec();
1219 }
1220
1221 if (chromaSharpen!=0.0) {
1222 SwsVector *id= sws_getIdentityVec();
1223 sws_scaleVec(filter->chrH, -chromaSharpen);
1224 sws_scaleVec(filter->chrV, -chromaSharpen);
1225 sws_addVec(filter->chrH, id);
1226 sws_addVec(filter->chrV, id);
1227 sws_freeVec(id);
1228 }
1229
1230 if (lumaSharpen!=0.0) {
1231 SwsVector *id= sws_getIdentityVec();
1232 sws_scaleVec(filter->lumH, -lumaSharpen);
1233 sws_scaleVec(filter->lumV, -lumaSharpen);
1234 sws_addVec(filter->lumH, id);
1235 sws_addVec(filter->lumV, id);
1236 sws_freeVec(id);
1237 }
1238
1239 if (chromaHShift != 0.0)
1240 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
1241
1242 if (chromaVShift != 0.0)
1243 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
1244
1245 sws_normalizeVec(filter->chrH, 1.0);
1246 sws_normalizeVec(filter->chrV, 1.0);
1247 sws_normalizeVec(filter->lumH, 1.0);
1248 sws_normalizeVec(filter->lumV, 1.0);
1249
1250 if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
1251 if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
1252
1253 return filter;
1254 }
1255
1256 SwsVector *sws_allocVec(int length)
1257 {
1258 SwsVector *vec = av_malloc(sizeof(SwsVector));
1259 if (!vec)
1260 return NULL;
1261 vec->length = length;
1262 vec->coeff = av_malloc(sizeof(double) * length);
1263 if (!vec->coeff)
1264 av_freep(&vec);
1265 return vec;
1266 }
1267
1268 SwsVector *sws_getGaussianVec(double variance, double quality)
1269 {
1270 const int length= (int)(variance*quality + 0.5) | 1;
1271 int i;
1272 double middle= (length-1)*0.5;
1273 SwsVector *vec= sws_allocVec(length);
1274
1275 if (!vec)
1276 return NULL;
1277
1278 for (i=0; i<length; i++) {
1279 double dist= i-middle;
1280 vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);
1281 }
1282
1283 sws_normalizeVec(vec, 1.0);
1284
1285 return vec;
1286 }
1287
1288 SwsVector *sws_getConstVec(double c, int length)
1289 {
1290 int i;
1291 SwsVector *vec= sws_allocVec(length);
1292
1293 if (!vec)
1294 return NULL;
1295
1296 for (i=0; i<length; i++)
1297 vec->coeff[i]= c;
1298
1299 return vec;
1300 }
1301
1302 SwsVector *sws_getIdentityVec(void)
1303 {
1304 return sws_getConstVec(1.0, 1);
1305 }
1306
1307 static double sws_dcVec(SwsVector *a)
1308 {
1309 int i;
1310 double sum=0;
1311
1312 for (i=0; i<a->length; i++)
1313 sum+= a->coeff[i];
1314
1315 return sum;
1316 }
1317
1318 void sws_scaleVec(SwsVector *a, double scalar)
1319 {
1320 int i;
1321
1322 for (i=0; i<a->length; i++)
1323 a->coeff[i]*= scalar;
1324 }
1325
1326 void sws_normalizeVec(SwsVector *a, double height)
1327 {
1328 sws_scaleVec(a, height/sws_dcVec(a));
1329 }
1330
1331 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
1332 {
1333 int length= a->length + b->length - 1;
1334 int i, j;
1335 SwsVector *vec= sws_getConstVec(0.0, length);
1336
1337 if (!vec)
1338 return NULL;
1339
1340 for (i=0; i<a->length; i++) {
1341 for (j=0; j<b->length; j++) {
1342 vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];
1343 }
1344 }
1345
1346 return vec;
1347 }
1348
1349 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
1350 {
1351 int length= FFMAX(a->length, b->length);
1352 int i;
1353 SwsVector *vec= sws_getConstVec(0.0, length);
1354
1355 if (!vec)
1356 return NULL;
1357
1358 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1359 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
1360
1361 return vec;
1362 }
1363
1364 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
1365 {
1366 int length= FFMAX(a->length, b->length);
1367 int i;
1368 SwsVector *vec= sws_getConstVec(0.0, length);
1369
1370 if (!vec)
1371 return NULL;
1372
1373 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1374 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
1375
1376 return vec;
1377 }
1378
1379 /* shift left / or right if "shift" is negative */
1380 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
1381 {
1382 int length= a->length + FFABS(shift)*2;
1383 int i;
1384 SwsVector *vec= sws_getConstVec(0.0, length);
1385
1386 if (!vec)
1387 return NULL;
1388
1389 for (i=0; i<a->length; i++) {
1390 vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
1391 }
1392
1393 return vec;
1394 }
1395
1396 void sws_shiftVec(SwsVector *a, int shift)
1397 {
1398 SwsVector *shifted= sws_getShiftedVec(a, shift);
1399 av_free(a->coeff);
1400 a->coeff= shifted->coeff;
1401 a->length= shifted->length;
1402 av_free(shifted);
1403 }
1404
1405 void sws_addVec(SwsVector *a, SwsVector *b)
1406 {
1407 SwsVector *sum= sws_sumVec(a, b);
1408 av_free(a->coeff);
1409 a->coeff= sum->coeff;
1410 a->length= sum->length;
1411 av_free(sum);
1412 }
1413
1414 void sws_subVec(SwsVector *a, SwsVector *b)
1415 {
1416 SwsVector *diff= sws_diffVec(a, b);
1417 av_free(a->coeff);
1418 a->coeff= diff->coeff;
1419 a->length= diff->length;
1420 av_free(diff);
1421 }
1422
1423 void sws_convVec(SwsVector *a, SwsVector *b)
1424 {
1425 SwsVector *conv= sws_getConvVec(a, b);
1426 av_free(a->coeff);
1427 a->coeff= conv->coeff;
1428 a->length= conv->length;
1429 av_free(conv);
1430 }
1431
1432 SwsVector *sws_cloneVec(SwsVector *a)
1433 {
1434 int i;
1435 SwsVector *vec= sws_allocVec(a->length);
1436
1437 if (!vec)
1438 return NULL;
1439
1440 for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];
1441
1442 return vec;
1443 }
1444
1445 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
1446 {
1447 int i;
1448 double max=0;
1449 double min=0;
1450 double range;
1451
1452 for (i=0; i<a->length; i++)
1453 if (a->coeff[i]>max) max= a->coeff[i];
1454
1455 for (i=0; i<a->length; i++)
1456 if (a->coeff[i]<min) min= a->coeff[i];
1457
1458 range= max - min;
1459
1460 for (i=0; i<a->length; i++) {
1461 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
1462 av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
1463 for (;x>0; x--) av_log(log_ctx, log_level, " ");
1464 av_log(log_ctx, log_level, "|\n");
1465 }
1466 }
1467
1468 #if LIBSWSCALE_VERSION_MAJOR < 1
1469 void sws_printVec(SwsVector *a)
1470 {
1471 sws_printVec2(a, NULL, AV_LOG_DEBUG);
1472 }
1473 #endif
1474
1475 void sws_freeVec(SwsVector *a)
1476 {
1477 if (!a) return;
1478 av_freep(&a->coeff);
1479 a->length=0;
1480 av_free(a);
1481 }
1482
1483 void sws_freeFilter(SwsFilter *filter)
1484 {
1485 if (!filter) return;
1486
1487 if (filter->lumH) sws_freeVec(filter->lumH);
1488 if (filter->lumV) sws_freeVec(filter->lumV);
1489 if (filter->chrH) sws_freeVec(filter->chrH);
1490 if (filter->chrV) sws_freeVec(filter->chrV);
1491 av_free(filter);
1492 }
1493
1494 void sws_freeContext(SwsContext *c)
1495 {
1496 int i;
1497 if (!c) return;
1498
1499 if (c->lumPixBuf) {
1500 for (i=0; i<c->vLumBufSize; i++)
1501 av_freep(&c->lumPixBuf[i]);
1502 av_freep(&c->lumPixBuf);
1503 }
1504
1505 if (c->chrPixBuf) {
1506 for (i=0; i<c->vChrBufSize; i++)
1507 av_freep(&c->chrPixBuf[i]);
1508 av_freep(&c->chrPixBuf);
1509 }
1510
1511 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1512 for (i=0; i<c->vLumBufSize; i++)
1513 av_freep(&c->alpPixBuf[i]);
1514 av_freep(&c->alpPixBuf);
1515 }
1516
1517 av_freep(&c->vLumFilter);
1518 av_freep(&c->vChrFilter);
1519 av_freep(&c->hLumFilter);
1520 av_freep(&c->hChrFilter);
1521 #if HAVE_ALTIVEC
1522 av_freep(&c->vYCoeffsBank);
1523 av_freep(&c->vCCoeffsBank);
1524 #endif
1525
1526 av_freep(&c->vLumFilterPos);
1527 av_freep(&c->vChrFilterPos);
1528 av_freep(&c->hLumFilterPos);
1529 av_freep(&c->hChrFilterPos);
1530
1531 #if ARCH_X86
1532 #ifdef MAP_ANONYMOUS
1533 if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);
1534 if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);
1535 #elif HAVE_VIRTUALALLOC
1536 if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);
1537 if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);
1538 #else
1539 av_free(c->lumMmx2FilterCode);
1540 av_free(c->chrMmx2FilterCode);
1541 #endif
1542 c->lumMmx2FilterCode=NULL;
1543 c->chrMmx2FilterCode=NULL;
1544 #endif /* ARCH_X86 */
1545
1546 av_freep(&c->yuvTable);
1547
1548 av_free(c);
1549 }
1550
1551 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
1552 int srcW, int srcH, enum PixelFormat srcFormat,
1553 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
1554 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
1555 {
1556 static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};
1557
1558 if (!param)
1559 param = default_param;
1560
1561 if (context &&
1562 (context->srcW != srcW ||
1563 context->srcH != srcH ||
1564 context->srcFormat != srcFormat ||
1565 context->dstW != dstW ||
1566 context->dstH != dstH ||
1567 context->dstFormat != dstFormat ||
1568 context->flags != flags ||
1569 context->param[0] != param[0] ||
1570 context->param[1] != param[1])) {
1571 sws_freeContext(context);
1572 context = NULL;
1573 }
1574
1575 if (!context) {
1576 return sws_getContext(srcW, srcH, srcFormat,
1577 dstW, dstH, dstFormat, flags,
1578 srcFilter, dstFilter, param);
1579 }
1580 return context;
1581 }
1582