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