swscale: RGB4444 and BGR444 input
[libav.git] / libswscale / utils.c
1 /*
2 * Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
3 *
4 * This file is part of Libav.
5 *
6 * Libav is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
10 *
11 * Libav is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with Libav; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
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/cpu.h"
44 #include "libavutil/avutil.h"
45 #include "libavutil/bswap.h"
46 #include "libavutil/mathematics.h"
47 #include "libavutil/opt.h"
48 #include "libavutil/pixdesc.h"
49
50 unsigned swscale_version(void)
51 {
52 return LIBSWSCALE_VERSION_INT;
53 }
54
55 const char *swscale_configuration(void)
56 {
57 return LIBAV_CONFIGURATION;
58 }
59
60 const char *swscale_license(void)
61 {
62 #define LICENSE_PREFIX "libswscale license: "
63 return LICENSE_PREFIX LIBAV_LICENSE + sizeof(LICENSE_PREFIX) - 1;
64 }
65
66 #define RET 0xC3 //near return opcode for x86
67
68 typedef struct FormatEntry {
69 int is_supported_in, is_supported_out;
70 } FormatEntry;
71
72 const static FormatEntry format_entries[PIX_FMT_NB] = {
73 [PIX_FMT_YUV420P] = { 1 , 1 },
74 [PIX_FMT_YUYV422] = { 1 , 1 },
75 [PIX_FMT_RGB24] = { 1 , 1 },
76 [PIX_FMT_BGR24] = { 1 , 1 },
77 [PIX_FMT_YUV422P] = { 1 , 1 },
78 [PIX_FMT_YUV444P] = { 1 , 1 },
79 [PIX_FMT_YUV410P] = { 1 , 1 },
80 [PIX_FMT_YUV411P] = { 1 , 1 },
81 [PIX_FMT_GRAY8] = { 1 , 1 },
82 [PIX_FMT_MONOWHITE] = { 1 , 1 },
83 [PIX_FMT_MONOBLACK] = { 1 , 1 },
84 [PIX_FMT_PAL8] = { 1 , 0 },
85 [PIX_FMT_YUVJ420P] = { 1 , 1 },
86 [PIX_FMT_YUVJ422P] = { 1 , 1 },
87 [PIX_FMT_YUVJ444P] = { 1 , 1 },
88 [PIX_FMT_UYVY422] = { 1 , 1 },
89 [PIX_FMT_UYYVYY411] = { 0 , 0 },
90 [PIX_FMT_BGR8] = { 1 , 1 },
91 [PIX_FMT_BGR4] = { 0 , 1 },
92 [PIX_FMT_BGR4_BYTE] = { 1 , 1 },
93 [PIX_FMT_RGB8] = { 1 , 1 },
94 [PIX_FMT_RGB4] = { 0 , 1 },
95 [PIX_FMT_RGB4_BYTE] = { 1 , 1 },
96 [PIX_FMT_NV12] = { 1 , 1 },
97 [PIX_FMT_NV21] = { 1 , 1 },
98 [PIX_FMT_ARGB] = { 1 , 1 },
99 [PIX_FMT_RGBA] = { 1 , 1 },
100 [PIX_FMT_ABGR] = { 1 , 1 },
101 [PIX_FMT_BGRA] = { 1 , 1 },
102 [PIX_FMT_GRAY16BE] = { 1 , 1 },
103 [PIX_FMT_GRAY16LE] = { 1 , 1 },
104 [PIX_FMT_YUV440P] = { 1 , 1 },
105 [PIX_FMT_YUVJ440P] = { 1 , 1 },
106 [PIX_FMT_YUVA420P] = { 1 , 1 },
107 [PIX_FMT_RGB48BE] = { 1 , 1 },
108 [PIX_FMT_RGB48LE] = { 1 , 1 },
109 [PIX_FMT_RGB565BE] = { 1 , 1 },
110 [PIX_FMT_RGB565LE] = { 1 , 1 },
111 [PIX_FMT_RGB555BE] = { 1 , 1 },
112 [PIX_FMT_RGB555LE] = { 1 , 1 },
113 [PIX_FMT_BGR565BE] = { 1 , 1 },
114 [PIX_FMT_BGR565LE] = { 1 , 1 },
115 [PIX_FMT_BGR555BE] = { 1 , 1 },
116 [PIX_FMT_BGR555LE] = { 1 , 1 },
117 [PIX_FMT_YUV420P16LE] = { 1 , 1 },
118 [PIX_FMT_YUV420P16BE] = { 1 , 1 },
119 [PIX_FMT_YUV422P16LE] = { 1 , 1 },
120 [PIX_FMT_YUV422P16BE] = { 1 , 1 },
121 [PIX_FMT_YUV444P16LE] = { 1 , 1 },
122 [PIX_FMT_YUV444P16BE] = { 1 , 1 },
123 [PIX_FMT_RGB444LE] = { 1 , 1 },
124 [PIX_FMT_RGB444BE] = { 1 , 1 },
125 [PIX_FMT_BGR444LE] = { 1 , 1 },
126 [PIX_FMT_BGR444BE] = { 1 , 1 },
127 [PIX_FMT_Y400A] = { 1 , 0 },
128 [PIX_FMT_BGR48BE] = { 1 , 1 },
129 [PIX_FMT_BGR48LE] = { 1 , 1 },
130 [PIX_FMT_YUV420P9BE] = { 1 , 1 },
131 [PIX_FMT_YUV420P9LE] = { 1 , 1 },
132 [PIX_FMT_YUV420P10BE] = { 1 , 1 },
133 [PIX_FMT_YUV420P10LE] = { 1 , 1 },
134 [PIX_FMT_YUV422P9BE] = { 1 , 1 },
135 [PIX_FMT_YUV422P9LE] = { 1 , 1 },
136 [PIX_FMT_YUV422P10BE] = { 1 , 1 },
137 [PIX_FMT_YUV422P10LE] = { 1 , 1 },
138 [PIX_FMT_YUV444P9BE] = { 1 , 1 },
139 [PIX_FMT_YUV444P9LE] = { 1 , 1 },
140 [PIX_FMT_YUV444P10BE] = { 1 , 1 },
141 [PIX_FMT_YUV444P10LE] = { 1 , 1 },
142 [PIX_FMT_GBRP] = { 1 , 0 },
143 [PIX_FMT_GBRP9LE] = { 1 , 0 },
144 [PIX_FMT_GBRP9BE] = { 1 , 0 },
145 [PIX_FMT_GBRP10LE] = { 1 , 0 },
146 [PIX_FMT_GBRP10BE] = { 1 , 0 },
147 [PIX_FMT_GBRP16LE] = { 1 , 0 },
148 [PIX_FMT_GBRP16BE] = { 1 , 0 },
149 };
150
151 int sws_isSupportedInput(enum PixelFormat pix_fmt)
152 {
153 return (unsigned)pix_fmt < PIX_FMT_NB ?
154 format_entries[pix_fmt].is_supported_in : 0;
155 }
156
157 int sws_isSupportedOutput(enum PixelFormat pix_fmt)
158 {
159 return (unsigned)pix_fmt < PIX_FMT_NB ?
160 format_entries[pix_fmt].is_supported_out : 0;
161 }
162
163 extern const int32_t ff_yuv2rgb_coeffs[8][4];
164
165 const char *sws_format_name(enum PixelFormat format)
166 {
167 if ((unsigned)format < PIX_FMT_NB && av_pix_fmt_descriptors[format].name)
168 return av_pix_fmt_descriptors[format].name;
169 else
170 return "Unknown format";
171 }
172
173 static double getSplineCoeff(double a, double b, double c, double d, double dist)
174 {
175 if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
176 else return getSplineCoeff( 0.0,
177 b+ 2.0*c + 3.0*d,
178 c + 3.0*d,
179 -b- 3.0*c - 6.0*d,
180 dist-1.0);
181 }
182
183 static int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
184 int srcW, int dstW, int filterAlign, int one, int flags, int cpu_flags,
185 SwsVector *srcFilter, SwsVector *dstFilter, double param[2], int is_horizontal)
186 {
187 int i;
188 int filterSize;
189 int filter2Size;
190 int minFilterSize;
191 int64_t *filter=NULL;
192 int64_t *filter2=NULL;
193 const int64_t fone= 1LL<<54;
194 int ret= -1;
195
196 emms_c(); //FIXME this should not be required but it IS (even for non-MMX versions)
197
198 // NOTE: the +3 is for the MMX(+1)/SSE(+3) scaler which reads over the end
199 FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW+3)*sizeof(int16_t), fail);
200
201 if (FFABS(xInc - 0x10000) <10) { // unscaled
202 int i;
203 filterSize= 1;
204 FF_ALLOCZ_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
205
206 for (i=0; i<dstW; i++) {
207 filter[i*filterSize]= fone;
208 (*filterPos)[i]=i;
209 }
210
211 } else if (flags&SWS_POINT) { // lame looking point sampling mode
212 int i;
213 int xDstInSrc;
214 filterSize= 1;
215 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
216
217 xDstInSrc= xInc/2 - 0x8000;
218 for (i=0; i<dstW; i++) {
219 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
220
221 (*filterPos)[i]= xx;
222 filter[i]= fone;
223 xDstInSrc+= xInc;
224 }
225 } else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) { // bilinear upscale
226 int i;
227 int xDstInSrc;
228 filterSize= 2;
229 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
230
231 xDstInSrc= xInc/2 - 0x8000;
232 for (i=0; i<dstW; i++) {
233 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
234 int j;
235
236 (*filterPos)[i]= xx;
237 //bilinear upscale / linear interpolate / area averaging
238 for (j=0; j<filterSize; j++) {
239 int64_t coeff= fone - FFABS((xx<<16) - xDstInSrc)*(fone>>16);
240 if (coeff<0) coeff=0;
241 filter[i*filterSize + j]= coeff;
242 xx++;
243 }
244 xDstInSrc+= xInc;
245 }
246 } else {
247 int xDstInSrc;
248 int sizeFactor;
249
250 if (flags&SWS_BICUBIC) sizeFactor= 4;
251 else if (flags&SWS_X) sizeFactor= 8;
252 else if (flags&SWS_AREA) sizeFactor= 1; //downscale only, for upscale it is bilinear
253 else if (flags&SWS_GAUSS) sizeFactor= 8; // infinite ;)
254 else if (flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? ceil(2*param[0]) : 6;
255 else if (flags&SWS_SINC) sizeFactor= 20; // infinite ;)
256 else if (flags&SWS_SPLINE) sizeFactor= 20; // infinite ;)
257 else if (flags&SWS_BILINEAR) sizeFactor= 2;
258 else {
259 sizeFactor= 0; //GCC warning killer
260 assert(0);
261 }
262
263 if (xInc <= 1<<16) filterSize= 1 + sizeFactor; // upscale
264 else filterSize= 1 + (sizeFactor*srcW + dstW - 1)/ dstW;
265
266 if (filterSize > srcW-2) filterSize=srcW-2;
267
268 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
269
270 xDstInSrc= xInc - 0x10000;
271 for (i=0; i<dstW; i++) {
272 int xx= (xDstInSrc - ((filterSize-2)<<16)) / (1<<17);
273 int j;
274 (*filterPos)[i]= xx;
275 for (j=0; j<filterSize; j++) {
276 int64_t d= ((int64_t)FFABS((xx<<17) - xDstInSrc))<<13;
277 double floatd;
278 int64_t coeff;
279
280 if (xInc > 1<<16)
281 d= d*dstW/srcW;
282 floatd= d * (1.0/(1<<30));
283
284 if (flags & SWS_BICUBIC) {
285 int64_t B= (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1<<24);
286 int64_t C= (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1<<24);
287
288 if (d >= 1LL<<31) {
289 coeff = 0.0;
290 } else {
291 int64_t dd = (d * d) >> 30;
292 int64_t ddd = (dd * d) >> 30;
293
294 if (d < 1LL<<30)
295 coeff = (12*(1<<24)-9*B-6*C)*ddd + (-18*(1<<24)+12*B+6*C)*dd + (6*(1<<24)-2*B)*(1<<30);
296 else
297 coeff = (-B-6*C)*ddd + (6*B+30*C)*dd + (-12*B-48*C)*d + (8*B+24*C)*(1<<30);
298 }
299 coeff *= fone>>(30+24);
300 }
301 /* else if (flags & SWS_X) {
302 double p= param ? param*0.01 : 0.3;
303 coeff = d ? sin(d*M_PI)/(d*M_PI) : 1.0;
304 coeff*= pow(2.0, - p*d*d);
305 }*/
306 else if (flags & SWS_X) {
307 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
308 double c;
309
310 if (floatd<1.0)
311 c = cos(floatd*M_PI);
312 else
313 c=-1.0;
314 if (c<0.0) c= -pow(-c, A);
315 else c= pow( c, A);
316 coeff= (c*0.5 + 0.5)*fone;
317 } else if (flags & SWS_AREA) {
318 int64_t d2= d - (1<<29);
319 if (d2*xInc < -(1LL<<(29+16))) coeff= 1.0 * (1LL<<(30+16));
320 else if (d2*xInc < (1LL<<(29+16))) coeff= -d2*xInc + (1LL<<(29+16));
321 else coeff=0.0;
322 coeff *= fone>>(30+16);
323 } else if (flags & SWS_GAUSS) {
324 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
325 coeff = (pow(2.0, - p*floatd*floatd))*fone;
326 } else if (flags & SWS_SINC) {
327 coeff = (d ? sin(floatd*M_PI)/(floatd*M_PI) : 1.0)*fone;
328 } else if (flags & SWS_LANCZOS) {
329 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
330 coeff = (d ? sin(floatd*M_PI)*sin(floatd*M_PI/p)/(floatd*floatd*M_PI*M_PI/p) : 1.0)*fone;
331 if (floatd>p) coeff=0;
332 } else if (flags & SWS_BILINEAR) {
333 coeff= (1<<30) - d;
334 if (coeff<0) coeff=0;
335 coeff *= fone >> 30;
336 } else if (flags & SWS_SPLINE) {
337 double p=-2.196152422706632;
338 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, floatd) * fone;
339 } else {
340 coeff= 0.0; //GCC warning killer
341 assert(0);
342 }
343
344 filter[i*filterSize + j]= coeff;
345 xx++;
346 }
347 xDstInSrc+= 2*xInc;
348 }
349 }
350
351 /* apply src & dst Filter to filter -> filter2
352 av_free(filter);
353 */
354 assert(filterSize>0);
355 filter2Size= filterSize;
356 if (srcFilter) filter2Size+= srcFilter->length - 1;
357 if (dstFilter) filter2Size+= dstFilter->length - 1;
358 assert(filter2Size>0);
359 FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size*dstW*sizeof(*filter2), fail);
360
361 for (i=0; i<dstW; i++) {
362 int j, k;
363
364 if(srcFilter) {
365 for (k=0; k<srcFilter->length; k++) {
366 for (j=0; j<filterSize; j++)
367 filter2[i*filter2Size + k + j] += srcFilter->coeff[k]*filter[i*filterSize + j];
368 }
369 } else {
370 for (j=0; j<filterSize; j++)
371 filter2[i*filter2Size + j]= filter[i*filterSize + j];
372 }
373 //FIXME dstFilter
374
375 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
376 }
377 av_freep(&filter);
378
379 /* try to reduce the filter-size (step1 find size and shift left) */
380 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
381 minFilterSize= 0;
382 for (i=dstW-1; i>=0; i--) {
383 int min= filter2Size;
384 int j;
385 int64_t cutOff=0.0;
386
387 /* get rid of near zero elements on the left by shifting left */
388 for (j=0; j<filter2Size; j++) {
389 int k;
390 cutOff += FFABS(filter2[i*filter2Size]);
391
392 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
393
394 /* preserve monotonicity because the core can't handle the filter otherwise */
395 if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
396
397 // move filter coefficients left
398 for (k=1; k<filter2Size; k++)
399 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
400 filter2[i*filter2Size + k - 1]= 0;
401 (*filterPos)[i]++;
402 }
403
404 cutOff=0;
405 /* count near zeros on the right */
406 for (j=filter2Size-1; j>0; j--) {
407 cutOff += FFABS(filter2[i*filter2Size + j]);
408
409 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
410 min--;
411 }
412
413 if (min>minFilterSize) minFilterSize= min;
414 }
415
416 if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) {
417 // we can handle the special case 4,
418 // so we don't want to go to the full 8
419 if (minFilterSize < 5)
420 filterAlign = 4;
421
422 // We really don't want to waste our time
423 // doing useless computation, so fall back on
424 // the scalar C code for very small filters.
425 // Vectorizing is worth it only if you have a
426 // decent-sized vector.
427 if (minFilterSize < 3)
428 filterAlign = 1;
429 }
430
431 if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
432 // special case for unscaled vertical filtering
433 if (minFilterSize == 1 && filterAlign == 2)
434 filterAlign= 1;
435 }
436
437 assert(minFilterSize > 0);
438 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
439 assert(filterSize > 0);
440 filter= av_malloc(filterSize*dstW*sizeof(*filter));
441 if (filterSize >= MAX_FILTER_SIZE*16/((flags&SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
442 goto fail;
443 *outFilterSize= filterSize;
444
445 if (flags&SWS_PRINT_INFO)
446 av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
447 /* try to reduce the filter-size (step2 reduce it) */
448 for (i=0; i<dstW; i++) {
449 int j;
450
451 for (j=0; j<filterSize; j++) {
452 if (j>=filter2Size) filter[i*filterSize + j]= 0;
453 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
454 if((flags & SWS_BITEXACT) && j>=minFilterSize)
455 filter[i*filterSize + j]= 0;
456 }
457 }
458
459 //FIXME try to align filterPos if possible
460
461 //fix borders
462 if (is_horizontal) {
463 for (i = 0; i < dstW; i++) {
464 int j;
465 if ((*filterPos)[i] < 0) {
466 // move filter coefficients left to compensate for filterPos
467 for (j = 1; j < filterSize; j++) {
468 int left = FFMAX(j + (*filterPos)[i], 0);
469 filter[i * filterSize + left] += filter[i * filterSize + j];
470 filter[i * filterSize + j ] = 0;
471 }
472 (*filterPos)[i] = 0;
473 }
474
475 if ((*filterPos)[i] + filterSize > srcW) {
476 int shift = (*filterPos)[i] + filterSize - srcW;
477 // move filter coefficients right to compensate for filterPos
478 for (j = filterSize - 2; j >= 0; j--) {
479 int right = FFMIN(j + shift, filterSize - 1);
480 filter[i * filterSize + right] += filter[i * filterSize + j];
481 filter[i * filterSize + j ] = 0;
482 }
483 (*filterPos)[i] = srcW - filterSize;
484 }
485 }
486 }
487
488 // Note the +1 is for the MMX scaler which reads over the end
489 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
490 FF_ALLOCZ_OR_GOTO(NULL, *outFilter, *outFilterSize*(dstW+3)*sizeof(int16_t), fail);
491
492 /* normalize & store in outFilter */
493 for (i=0; i<dstW; i++) {
494 int j;
495 int64_t error=0;
496 int64_t sum=0;
497
498 for (j=0; j<filterSize; j++) {
499 sum+= filter[i*filterSize + j];
500 }
501 sum= (sum + one/2)/ one;
502 for (j=0; j<*outFilterSize; j++) {
503 int64_t v= filter[i*filterSize + j] + error;
504 int intV= ROUNDED_DIV(v, sum);
505 (*outFilter)[i*(*outFilterSize) + j]= intV;
506 error= v - intV*sum;
507 }
508 }
509
510 (*filterPos)[dstW+0] =
511 (*filterPos)[dstW+1] =
512 (*filterPos)[dstW+2] = (*filterPos)[dstW-1]; // the MMX/SSE scaler will read over the end
513 for (i=0; i<*outFilterSize; i++) {
514 int k= (dstW - 1) * (*outFilterSize) + i;
515 (*outFilter)[k + 1 * (*outFilterSize)] =
516 (*outFilter)[k + 2 * (*outFilterSize)] =
517 (*outFilter)[k + 3 * (*outFilterSize)] = (*outFilter)[k];
518 }
519
520 ret=0;
521 fail:
522 av_free(filter);
523 av_free(filter2);
524 return ret;
525 }
526
527 #if HAVE_MMX2
528 static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode, int16_t *filter, int32_t *filterPos, int numSplits)
529 {
530 uint8_t *fragmentA;
531 x86_reg imm8OfPShufW1A;
532 x86_reg imm8OfPShufW2A;
533 x86_reg fragmentLengthA;
534 uint8_t *fragmentB;
535 x86_reg imm8OfPShufW1B;
536 x86_reg imm8OfPShufW2B;
537 x86_reg fragmentLengthB;
538 int fragmentPos;
539
540 int xpos, i;
541
542 // create an optimized horizontal scaling routine
543 /* This scaler is made of runtime-generated MMX2 code using specially
544 * tuned pshufw instructions. For every four output pixels, if four
545 * input pixels are enough for the fast bilinear scaling, then a chunk
546 * of fragmentB is used. If five input pixels are needed, then a chunk
547 * of fragmentA is used.
548 */
549
550 //code fragment
551
552 __asm__ volatile(
553 "jmp 9f \n\t"
554 // Begin
555 "0: \n\t"
556 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
557 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
558 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
559 "punpcklbw %%mm7, %%mm1 \n\t"
560 "punpcklbw %%mm7, %%mm0 \n\t"
561 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
562 "1: \n\t"
563 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
564 "2: \n\t"
565 "psubw %%mm1, %%mm0 \n\t"
566 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
567 "pmullw %%mm3, %%mm0 \n\t"
568 "psllw $7, %%mm1 \n\t"
569 "paddw %%mm1, %%mm0 \n\t"
570
571 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
572
573 "add $8, %%"REG_a" \n\t"
574 // End
575 "9: \n\t"
576 // "int $3 \n\t"
577 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
578 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
579 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
580 "dec %1 \n\t"
581 "dec %2 \n\t"
582 "sub %0, %1 \n\t"
583 "sub %0, %2 \n\t"
584 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
585 "sub %0, %3 \n\t"
586
587
588 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
589 "=r" (fragmentLengthA)
590 );
591
592 __asm__ volatile(
593 "jmp 9f \n\t"
594 // Begin
595 "0: \n\t"
596 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
597 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
598 "punpcklbw %%mm7, %%mm0 \n\t"
599 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
600 "1: \n\t"
601 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
602 "2: \n\t"
603 "psubw %%mm1, %%mm0 \n\t"
604 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
605 "pmullw %%mm3, %%mm0 \n\t"
606 "psllw $7, %%mm1 \n\t"
607 "paddw %%mm1, %%mm0 \n\t"
608
609 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
610
611 "add $8, %%"REG_a" \n\t"
612 // End
613 "9: \n\t"
614 // "int $3 \n\t"
615 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
616 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
617 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
618 "dec %1 \n\t"
619 "dec %2 \n\t"
620 "sub %0, %1 \n\t"
621 "sub %0, %2 \n\t"
622 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
623 "sub %0, %3 \n\t"
624
625
626 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
627 "=r" (fragmentLengthB)
628 );
629
630 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
631 fragmentPos=0;
632
633 for (i=0; i<dstW/numSplits; i++) {
634 int xx=xpos>>16;
635
636 if ((i&3) == 0) {
637 int a=0;
638 int b=((xpos+xInc)>>16) - xx;
639 int c=((xpos+xInc*2)>>16) - xx;
640 int d=((xpos+xInc*3)>>16) - xx;
641 int inc = (d+1<4);
642 uint8_t *fragment = (d+1<4) ? fragmentB : fragmentA;
643 x86_reg imm8OfPShufW1 = (d+1<4) ? imm8OfPShufW1B : imm8OfPShufW1A;
644 x86_reg imm8OfPShufW2 = (d+1<4) ? imm8OfPShufW2B : imm8OfPShufW2A;
645 x86_reg fragmentLength = (d+1<4) ? fragmentLengthB : fragmentLengthA;
646 int maxShift= 3-(d+inc);
647 int shift=0;
648
649 if (filterCode) {
650 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
651 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
652 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
653 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
654 filterPos[i/2]= xx;
655
656 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
657
658 filterCode[fragmentPos + imm8OfPShufW1]=
659 (a+inc) | ((b+inc)<<2) | ((c+inc)<<4) | ((d+inc)<<6);
660 filterCode[fragmentPos + imm8OfPShufW2]=
661 a | (b<<2) | (c<<4) | (d<<6);
662
663 if (i+4-inc>=dstW) shift=maxShift; //avoid overread
664 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
665
666 if (shift && i>=shift) {
667 filterCode[fragmentPos + imm8OfPShufW1]+= 0x55*shift;
668 filterCode[fragmentPos + imm8OfPShufW2]+= 0x55*shift;
669 filterPos[i/2]-=shift;
670 }
671 }
672
673 fragmentPos+= fragmentLength;
674
675 if (filterCode)
676 filterCode[fragmentPos]= RET;
677 }
678 xpos+=xInc;
679 }
680 if (filterCode)
681 filterPos[((i/2)+1)&(~1)]= xpos>>16; // needed to jump to the next part
682
683 return fragmentPos + 1;
684 }
685 #endif /* HAVE_MMX2 */
686
687 static void getSubSampleFactors(int *h, int *v, enum PixelFormat format)
688 {
689 *h = av_pix_fmt_descriptors[format].log2_chroma_w;
690 *v = av_pix_fmt_descriptors[format].log2_chroma_h;
691 }
692
693 int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],
694 int srcRange, const int table[4], int dstRange,
695 int brightness, int contrast, int saturation)
696 {
697 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
698 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
699
700 c->brightness= brightness;
701 c->contrast = contrast;
702 c->saturation= saturation;
703 c->srcRange = srcRange;
704 c->dstRange = dstRange;
705 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
706
707 c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->dstFormat]);
708 c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->srcFormat]);
709
710 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
711 //FIXME factorize
712
713 if (HAVE_ALTIVEC && av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)
714 ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness, contrast, saturation);
715 return 0;
716 }
717
718 int sws_getColorspaceDetails(struct SwsContext *c, int **inv_table,
719 int *srcRange, int **table, int *dstRange,
720 int *brightness, int *contrast, int *saturation)
721 {
722 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
723
724 *inv_table = c->srcColorspaceTable;
725 *table = c->dstColorspaceTable;
726 *srcRange = c->srcRange;
727 *dstRange = c->dstRange;
728 *brightness= c->brightness;
729 *contrast = c->contrast;
730 *saturation= c->saturation;
731
732 return 0;
733 }
734
735 static int handle_jpeg(enum PixelFormat *format)
736 {
737 switch (*format) {
738 case PIX_FMT_YUVJ420P: *format = PIX_FMT_YUV420P; return 1;
739 case PIX_FMT_YUVJ422P: *format = PIX_FMT_YUV422P; return 1;
740 case PIX_FMT_YUVJ444P: *format = PIX_FMT_YUV444P; return 1;
741 case PIX_FMT_YUVJ440P: *format = PIX_FMT_YUV440P; return 1;
742 default: return 0;
743 }
744 }
745
746 SwsContext *sws_alloc_context(void)
747 {
748 SwsContext *c= av_mallocz(sizeof(SwsContext));
749
750 c->av_class = &sws_context_class;
751 av_opt_set_defaults(c);
752
753 return c;
754 }
755
756 int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)
757 {
758 int i;
759 int usesVFilter, usesHFilter;
760 int unscaled;
761 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
762 int srcW= c->srcW;
763 int srcH= c->srcH;
764 int dstW= c->dstW;
765 int dstH= c->dstH;
766 int dst_stride = FFALIGN(dstW * sizeof(int16_t) + 16, 16), dst_stride_px = dst_stride >> 1;
767 int flags, cpu_flags;
768 enum PixelFormat srcFormat= c->srcFormat;
769 enum PixelFormat dstFormat= c->dstFormat;
770
771 cpu_flags = av_get_cpu_flags();
772 flags = c->flags;
773 emms_c();
774 if (!rgb15to16) sws_rgb2rgb_init();
775
776 unscaled = (srcW == dstW && srcH == dstH);
777
778 if (!sws_isSupportedInput(srcFormat)) {
779 av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n", sws_format_name(srcFormat));
780 return AVERROR(EINVAL);
781 }
782 if (!sws_isSupportedOutput(dstFormat)) {
783 av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n", sws_format_name(dstFormat));
784 return AVERROR(EINVAL);
785 }
786
787 i= flags & ( SWS_POINT
788 |SWS_AREA
789 |SWS_BILINEAR
790 |SWS_FAST_BILINEAR
791 |SWS_BICUBIC
792 |SWS_X
793 |SWS_GAUSS
794 |SWS_LANCZOS
795 |SWS_SINC
796 |SWS_SPLINE
797 |SWS_BICUBLIN);
798 if(!i || (i & (i-1))) {
799 av_log(c, AV_LOG_ERROR, "Exactly one scaler algorithm must be chosen\n");
800 return AVERROR(EINVAL);
801 }
802 /* sanity check */
803 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
804 av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",
805 srcW, srcH, dstW, dstH);
806 return AVERROR(EINVAL);
807 }
808
809 if (!dstFilter) dstFilter= &dummyFilter;
810 if (!srcFilter) srcFilter= &dummyFilter;
811
812 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
813 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
814 c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[dstFormat]);
815 c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[srcFormat]);
816 c->vRounder= 4* 0x0001000100010001ULL;
817
818 usesVFilter = (srcFilter->lumV && srcFilter->lumV->length>1) ||
819 (srcFilter->chrV && srcFilter->chrV->length>1) ||
820 (dstFilter->lumV && dstFilter->lumV->length>1) ||
821 (dstFilter->chrV && dstFilter->chrV->length>1);
822 usesHFilter = (srcFilter->lumH && srcFilter->lumH->length>1) ||
823 (srcFilter->chrH && srcFilter->chrH->length>1) ||
824 (dstFilter->lumH && dstFilter->lumH->length>1) ||
825 (dstFilter->chrH && dstFilter->chrH->length>1);
826
827 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
828 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
829
830 // reuse chroma for 2 pixels RGB/BGR unless user wants full chroma interpolation
831 if (flags & SWS_FULL_CHR_H_INT &&
832 isAnyRGB(dstFormat) &&
833 dstFormat != PIX_FMT_RGBA &&
834 dstFormat != PIX_FMT_ARGB &&
835 dstFormat != PIX_FMT_BGRA &&
836 dstFormat != PIX_FMT_ABGR &&
837 dstFormat != PIX_FMT_RGB24 &&
838 dstFormat != PIX_FMT_BGR24) {
839 av_log(c, AV_LOG_ERROR,
840 "full chroma interpolation for destination format '%s' not yet implemented\n",
841 sws_format_name(dstFormat));
842 flags &= ~SWS_FULL_CHR_H_INT;
843 c->flags = flags;
844 }
845 if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
846
847 // drop some chroma lines if the user wants it
848 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
849 c->chrSrcVSubSample+= c->vChrDrop;
850
851 // drop every other pixel for chroma calculation unless user wants full chroma
852 if (isAnyRGB(srcFormat) && !(flags&SWS_FULL_CHR_H_INP)
853 && srcFormat!=PIX_FMT_RGB8 && srcFormat!=PIX_FMT_BGR8
854 && srcFormat!=PIX_FMT_RGB4 && srcFormat!=PIX_FMT_BGR4
855 && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE
856 && ((dstW>>c->chrDstHSubSample) <= (srcW>>1) || (flags&SWS_FAST_BILINEAR)))
857 c->chrSrcHSubSample=1;
858
859 // Note the -((-x)>>y) is so that we always round toward +inf.
860 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
861 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
862 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
863 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
864
865 /* unscaled special cases */
866 if (unscaled && !usesHFilter && !usesVFilter && (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
867 ff_get_unscaled_swscale(c);
868
869 if (c->swScale) {
870 if (flags&SWS_PRINT_INFO)
871 av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
872 sws_format_name(srcFormat), sws_format_name(dstFormat));
873 return 0;
874 }
875 }
876
877 c->srcBpc = 1 + av_pix_fmt_descriptors[srcFormat].comp[0].depth_minus1;
878 if (c->srcBpc < 8)
879 c->srcBpc = 8;
880 c->dstBpc = 1 + av_pix_fmt_descriptors[dstFormat].comp[0].depth_minus1;
881 if (c->dstBpc < 8)
882 c->dstBpc = 8;
883 if (c->dstBpc == 16)
884 dst_stride <<= 1;
885 FF_ALLOC_OR_GOTO(c, c->formatConvBuffer,
886 (FFALIGN(srcW, 16) * 2 * FFALIGN(c->srcBpc, 8) >> 3) + 16,
887 fail);
888 if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2 && c->srcBpc == 8 && c->dstBpc <= 10) {
889 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
890 if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR)) {
891 if (flags&SWS_PRINT_INFO)
892 av_log(c, AV_LOG_INFO, "output width is not a multiple of 32 -> no MMX2 scaler\n");
893 }
894 if (usesHFilter) c->canMMX2BeUsed=0;
895 }
896 else
897 c->canMMX2BeUsed=0;
898
899 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
900 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
901
902 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
903 // but only for the FAST_BILINEAR mode otherwise do correct scaling
904 // n-2 is the last chrominance sample available
905 // this is not perfect, but no one should notice the difference, the more correct variant
906 // would be like the vertical one, but that would require some special code for the
907 // first and last pixel
908 if (flags&SWS_FAST_BILINEAR) {
909 if (c->canMMX2BeUsed) {
910 c->lumXInc+= 20;
911 c->chrXInc+= 20;
912 }
913 //we don't use the x86 asm scaler if MMX is available
914 else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
915 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
916 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
917 }
918 }
919
920 /* precalculate horizontal scaler filter coefficients */
921 {
922 #if HAVE_MMX2
923 // can't downscale !!!
924 if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {
925 c->lumMmx2FilterCodeSize = initMMX2HScaler( dstW, c->lumXInc, NULL, NULL, NULL, 8);
926 c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc, NULL, NULL, NULL, 4);
927
928 #ifdef MAP_ANONYMOUS
929 c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
930 c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
931 #elif HAVE_VIRTUALALLOC
932 c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
933 c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
934 #else
935 c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);
936 c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);
937 #endif
938
939 if (!c->lumMmx2FilterCode || !c->chrMmx2FilterCode)
940 return AVERROR(ENOMEM);
941 FF_ALLOCZ_OR_GOTO(c, c->hLumFilter , (dstW /8+8)*sizeof(int16_t), fail);
942 FF_ALLOCZ_OR_GOTO(c, c->hChrFilter , (c->chrDstW /4+8)*sizeof(int16_t), fail);
943 FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW /2/8+8)*sizeof(int32_t), fail);
944 FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW/2/4+8)*sizeof(int32_t), fail);
945
946 initMMX2HScaler( dstW, c->lumXInc, c->lumMmx2FilterCode, c->hLumFilter, c->hLumFilterPos, 8);
947 initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode, c->hChrFilter, c->hChrFilterPos, 4);
948
949 #ifdef MAP_ANONYMOUS
950 mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
951 mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
952 #endif
953 } else
954 #endif /* HAVE_MMX2 */
955 {
956 const int filterAlign=
957 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :
958 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
959 1;
960
961 if (initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
962 srcW , dstW, filterAlign, 1<<14,
963 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags, cpu_flags,
964 srcFilter->lumH, dstFilter->lumH, c->param, 1) < 0)
965 goto fail;
966 if (initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
967 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
968 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, cpu_flags,
969 srcFilter->chrH, dstFilter->chrH, c->param, 1) < 0)
970 goto fail;
971 }
972 } // initialize horizontal stuff
973
974 /* precalculate vertical scaler filter coefficients */
975 {
976 const int filterAlign=
977 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 2 :
978 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
979 1;
980
981 if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
982 srcH , dstH, filterAlign, (1<<12),
983 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags, cpu_flags,
984 srcFilter->lumV, dstFilter->lumV, c->param, 0) < 0)
985 goto fail;
986 if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
987 c->chrSrcH, c->chrDstH, filterAlign, (1<<12),
988 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, cpu_flags,
989 srcFilter->chrV, dstFilter->chrV, c->param, 0) < 0)
990 goto fail;
991
992 #if HAVE_ALTIVEC
993 FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof (vector signed short)*c->vLumFilterSize*c->dstH, fail);
994 FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH, fail);
995
996 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
997 int j;
998 short *p = (short *)&c->vYCoeffsBank[i];
999 for (j=0;j<8;j++)
1000 p[j] = c->vLumFilter[i];
1001 }
1002
1003 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
1004 int j;
1005 short *p = (short *)&c->vCCoeffsBank[i];
1006 for (j=0;j<8;j++)
1007 p[j] = c->vChrFilter[i];
1008 }
1009 #endif
1010 }
1011
1012 // calculate buffer sizes so that they won't run out while handling these damn slices
1013 c->vLumBufSize= c->vLumFilterSize;
1014 c->vChrBufSize= c->vChrFilterSize;
1015 for (i=0; i<dstH; i++) {
1016 int chrI= i*c->chrDstH / dstH;
1017 int nextSlice= FFMAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
1018 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
1019
1020 nextSlice>>= c->chrSrcVSubSample;
1021 nextSlice<<= c->chrSrcVSubSample;
1022 if (c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
1023 c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
1024 if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
1025 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
1026 }
1027
1028 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
1029 // allocate several megabytes to handle all possible cases)
1030 FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*3*sizeof(int16_t*), fail);
1031 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize*3*sizeof(int16_t*), fail);
1032 FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize*3*sizeof(int16_t*), fail);
1033 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
1034 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*3*sizeof(int16_t*), fail);
1035 //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)
1036 /* align at 16 bytes for AltiVec */
1037 for (i=0; i<c->vLumBufSize; i++) {
1038 FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], dst_stride+16, fail);
1039 c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];
1040 }
1041 // 64 / (c->dstBpc & ~7) is the same as 16 / sizeof(scaling_intermediate)
1042 c->uv_off_px = dst_stride_px + 64 / (c->dstBpc &~ 7);
1043 c->uv_off_byte = dst_stride + 16;
1044 for (i=0; i<c->vChrBufSize; i++) {
1045 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i+c->vChrBufSize], dst_stride*2+32, fail);
1046 c->chrUPixBuf[i] = c->chrUPixBuf[i+c->vChrBufSize];
1047 c->chrVPixBuf[i] = c->chrVPixBuf[i+c->vChrBufSize] = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;
1048 }
1049 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
1050 for (i=0; i<c->vLumBufSize; i++) {
1051 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], dst_stride+16, fail);
1052 c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];
1053 }
1054
1055 //try to avoid drawing green stuff between the right end and the stride end
1056 for (i=0; i<c->vChrBufSize; i++)
1057 memset(c->chrUPixBuf[i], 64, dst_stride*2+1);
1058
1059 assert(c->chrDstH <= dstH);
1060
1061 if (flags&SWS_PRINT_INFO) {
1062 if (flags&SWS_FAST_BILINEAR) av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
1063 else if (flags&SWS_BILINEAR) av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
1064 else if (flags&SWS_BICUBIC) av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
1065 else if (flags&SWS_X) av_log(c, AV_LOG_INFO, "Experimental scaler, ");
1066 else if (flags&SWS_POINT) av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
1067 else if (flags&SWS_AREA) av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
1068 else if (flags&SWS_BICUBLIN) av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
1069 else if (flags&SWS_GAUSS) av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
1070 else if (flags&SWS_SINC) av_log(c, AV_LOG_INFO, "Sinc scaler, ");
1071 else if (flags&SWS_LANCZOS) av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
1072 else if (flags&SWS_SPLINE) av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
1073 else av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
1074
1075 av_log(c, AV_LOG_INFO, "from %s to %s%s ",
1076 sws_format_name(srcFormat),
1077 #ifdef DITHER1XBPP
1078 dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ||
1079 dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||
1080 dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ? "dithered " : "",
1081 #else
1082 "",
1083 #endif
1084 sws_format_name(dstFormat));
1085
1086 if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2) av_log(c, AV_LOG_INFO, "using MMX2\n");
1087 else if (HAVE_AMD3DNOW && cpu_flags & AV_CPU_FLAG_3DNOW) av_log(c, AV_LOG_INFO, "using 3DNOW\n");
1088 else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) av_log(c, AV_LOG_INFO, "using MMX\n");
1089 else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) av_log(c, AV_LOG_INFO, "using AltiVec\n");
1090 else av_log(c, AV_LOG_INFO, "using C\n");
1091
1092 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
1093 av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1094 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
1095 av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1096 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
1097 }
1098
1099 c->swScale= ff_getSwsFunc(c);
1100 return 0;
1101 fail: //FIXME replace things by appropriate error codes
1102 return -1;
1103 }
1104
1105 #if FF_API_SWS_GETCONTEXT
1106 SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,
1107 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
1108 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
1109 {
1110 SwsContext *c;
1111
1112 if(!(c=sws_alloc_context()))
1113 return NULL;
1114
1115 c->flags= flags;
1116 c->srcW= srcW;
1117 c->srcH= srcH;
1118 c->dstW= dstW;
1119 c->dstH= dstH;
1120 c->srcRange = handle_jpeg(&srcFormat);
1121 c->dstRange = handle_jpeg(&dstFormat);
1122 c->srcFormat= srcFormat;
1123 c->dstFormat= dstFormat;
1124
1125 if (param) {
1126 c->param[0] = param[0];
1127 c->param[1] = param[1];
1128 }
1129 sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/, c->dstRange, 0, 1<<16, 1<<16);
1130
1131 if(sws_init_context(c, srcFilter, dstFilter) < 0){
1132 sws_freeContext(c);
1133 return NULL;
1134 }
1135
1136 return c;
1137 }
1138 #endif
1139
1140 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
1141 float lumaSharpen, float chromaSharpen,
1142 float chromaHShift, float chromaVShift,
1143 int verbose)
1144 {
1145 SwsFilter *filter= av_malloc(sizeof(SwsFilter));
1146 if (!filter)
1147 return NULL;
1148
1149 if (lumaGBlur!=0.0) {
1150 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
1151 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
1152 } else {
1153 filter->lumH= sws_getIdentityVec();
1154 filter->lumV= sws_getIdentityVec();
1155 }
1156
1157 if (chromaGBlur!=0.0) {
1158 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
1159 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
1160 } else {
1161 filter->chrH= sws_getIdentityVec();
1162 filter->chrV= sws_getIdentityVec();
1163 }
1164
1165 if (chromaSharpen!=0.0) {
1166 SwsVector *id= sws_getIdentityVec();
1167 sws_scaleVec(filter->chrH, -chromaSharpen);
1168 sws_scaleVec(filter->chrV, -chromaSharpen);
1169 sws_addVec(filter->chrH, id);
1170 sws_addVec(filter->chrV, id);
1171 sws_freeVec(id);
1172 }
1173
1174 if (lumaSharpen!=0.0) {
1175 SwsVector *id= sws_getIdentityVec();
1176 sws_scaleVec(filter->lumH, -lumaSharpen);
1177 sws_scaleVec(filter->lumV, -lumaSharpen);
1178 sws_addVec(filter->lumH, id);
1179 sws_addVec(filter->lumV, id);
1180 sws_freeVec(id);
1181 }
1182
1183 if (chromaHShift != 0.0)
1184 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
1185
1186 if (chromaVShift != 0.0)
1187 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
1188
1189 sws_normalizeVec(filter->chrH, 1.0);
1190 sws_normalizeVec(filter->chrV, 1.0);
1191 sws_normalizeVec(filter->lumH, 1.0);
1192 sws_normalizeVec(filter->lumV, 1.0);
1193
1194 if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
1195 if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
1196
1197 return filter;
1198 }
1199
1200 SwsVector *sws_allocVec(int length)
1201 {
1202 SwsVector *vec = av_malloc(sizeof(SwsVector));
1203 if (!vec)
1204 return NULL;
1205 vec->length = length;
1206 vec->coeff = av_malloc(sizeof(double) * length);
1207 if (!vec->coeff)
1208 av_freep(&vec);
1209 return vec;
1210 }
1211
1212 SwsVector *sws_getGaussianVec(double variance, double quality)
1213 {
1214 const int length= (int)(variance*quality + 0.5) | 1;
1215 int i;
1216 double middle= (length-1)*0.5;
1217 SwsVector *vec= sws_allocVec(length);
1218
1219 if (!vec)
1220 return NULL;
1221
1222 for (i=0; i<length; i++) {
1223 double dist= i-middle;
1224 vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);
1225 }
1226
1227 sws_normalizeVec(vec, 1.0);
1228
1229 return vec;
1230 }
1231
1232 SwsVector *sws_getConstVec(double c, int length)
1233 {
1234 int i;
1235 SwsVector *vec= sws_allocVec(length);
1236
1237 if (!vec)
1238 return NULL;
1239
1240 for (i=0; i<length; i++)
1241 vec->coeff[i]= c;
1242
1243 return vec;
1244 }
1245
1246 SwsVector *sws_getIdentityVec(void)
1247 {
1248 return sws_getConstVec(1.0, 1);
1249 }
1250
1251 static double sws_dcVec(SwsVector *a)
1252 {
1253 int i;
1254 double sum=0;
1255
1256 for (i=0; i<a->length; i++)
1257 sum+= a->coeff[i];
1258
1259 return sum;
1260 }
1261
1262 void sws_scaleVec(SwsVector *a, double scalar)
1263 {
1264 int i;
1265
1266 for (i=0; i<a->length; i++)
1267 a->coeff[i]*= scalar;
1268 }
1269
1270 void sws_normalizeVec(SwsVector *a, double height)
1271 {
1272 sws_scaleVec(a, height/sws_dcVec(a));
1273 }
1274
1275 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
1276 {
1277 int length= a->length + b->length - 1;
1278 int i, j;
1279 SwsVector *vec= sws_getConstVec(0.0, length);
1280
1281 if (!vec)
1282 return NULL;
1283
1284 for (i=0; i<a->length; i++) {
1285 for (j=0; j<b->length; j++) {
1286 vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];
1287 }
1288 }
1289
1290 return vec;
1291 }
1292
1293 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
1294 {
1295 int length= FFMAX(a->length, b->length);
1296 int i;
1297 SwsVector *vec= sws_getConstVec(0.0, length);
1298
1299 if (!vec)
1300 return NULL;
1301
1302 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1303 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
1304
1305 return vec;
1306 }
1307
1308 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
1309 {
1310 int length= FFMAX(a->length, b->length);
1311 int i;
1312 SwsVector *vec= sws_getConstVec(0.0, length);
1313
1314 if (!vec)
1315 return NULL;
1316
1317 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1318 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
1319
1320 return vec;
1321 }
1322
1323 /* shift left / or right if "shift" is negative */
1324 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
1325 {
1326 int length= a->length + FFABS(shift)*2;
1327 int i;
1328 SwsVector *vec= sws_getConstVec(0.0, length);
1329
1330 if (!vec)
1331 return NULL;
1332
1333 for (i=0; i<a->length; i++) {
1334 vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
1335 }
1336
1337 return vec;
1338 }
1339
1340 void sws_shiftVec(SwsVector *a, int shift)
1341 {
1342 SwsVector *shifted= sws_getShiftedVec(a, shift);
1343 av_free(a->coeff);
1344 a->coeff= shifted->coeff;
1345 a->length= shifted->length;
1346 av_free(shifted);
1347 }
1348
1349 void sws_addVec(SwsVector *a, SwsVector *b)
1350 {
1351 SwsVector *sum= sws_sumVec(a, b);
1352 av_free(a->coeff);
1353 a->coeff= sum->coeff;
1354 a->length= sum->length;
1355 av_free(sum);
1356 }
1357
1358 void sws_subVec(SwsVector *a, SwsVector *b)
1359 {
1360 SwsVector *diff= sws_diffVec(a, b);
1361 av_free(a->coeff);
1362 a->coeff= diff->coeff;
1363 a->length= diff->length;
1364 av_free(diff);
1365 }
1366
1367 void sws_convVec(SwsVector *a, SwsVector *b)
1368 {
1369 SwsVector *conv= sws_getConvVec(a, b);
1370 av_free(a->coeff);
1371 a->coeff= conv->coeff;
1372 a->length= conv->length;
1373 av_free(conv);
1374 }
1375
1376 SwsVector *sws_cloneVec(SwsVector *a)
1377 {
1378 int i;
1379 SwsVector *vec= sws_allocVec(a->length);
1380
1381 if (!vec)
1382 return NULL;
1383
1384 for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];
1385
1386 return vec;
1387 }
1388
1389 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
1390 {
1391 int i;
1392 double max=0;
1393 double min=0;
1394 double range;
1395
1396 for (i=0; i<a->length; i++)
1397 if (a->coeff[i]>max) max= a->coeff[i];
1398
1399 for (i=0; i<a->length; i++)
1400 if (a->coeff[i]<min) min= a->coeff[i];
1401
1402 range= max - min;
1403
1404 for (i=0; i<a->length; i++) {
1405 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
1406 av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
1407 for (;x>0; x--) av_log(log_ctx, log_level, " ");
1408 av_log(log_ctx, log_level, "|\n");
1409 }
1410 }
1411
1412 void sws_freeVec(SwsVector *a)
1413 {
1414 if (!a) return;
1415 av_freep(&a->coeff);
1416 a->length=0;
1417 av_free(a);
1418 }
1419
1420 void sws_freeFilter(SwsFilter *filter)
1421 {
1422 if (!filter) return;
1423
1424 if (filter->lumH) sws_freeVec(filter->lumH);
1425 if (filter->lumV) sws_freeVec(filter->lumV);
1426 if (filter->chrH) sws_freeVec(filter->chrH);
1427 if (filter->chrV) sws_freeVec(filter->chrV);
1428 av_free(filter);
1429 }
1430
1431 void sws_freeContext(SwsContext *c)
1432 {
1433 int i;
1434 if (!c) return;
1435
1436 if (c->lumPixBuf) {
1437 for (i=0; i<c->vLumBufSize; i++)
1438 av_freep(&c->lumPixBuf[i]);
1439 av_freep(&c->lumPixBuf);
1440 }
1441
1442 if (c->chrUPixBuf) {
1443 for (i=0; i<c->vChrBufSize; i++)
1444 av_freep(&c->chrUPixBuf[i]);
1445 av_freep(&c->chrUPixBuf);
1446 av_freep(&c->chrVPixBuf);
1447 }
1448
1449 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1450 for (i=0; i<c->vLumBufSize; i++)
1451 av_freep(&c->alpPixBuf[i]);
1452 av_freep(&c->alpPixBuf);
1453 }
1454
1455 av_freep(&c->vLumFilter);
1456 av_freep(&c->vChrFilter);
1457 av_freep(&c->hLumFilter);
1458 av_freep(&c->hChrFilter);
1459 #if HAVE_ALTIVEC
1460 av_freep(&c->vYCoeffsBank);
1461 av_freep(&c->vCCoeffsBank);
1462 #endif
1463
1464 av_freep(&c->vLumFilterPos);
1465 av_freep(&c->vChrFilterPos);
1466 av_freep(&c->hLumFilterPos);
1467 av_freep(&c->hChrFilterPos);
1468
1469 #if HAVE_MMX
1470 #ifdef MAP_ANONYMOUS
1471 if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);
1472 if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);
1473 #elif HAVE_VIRTUALALLOC
1474 if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);
1475 if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);
1476 #else
1477 av_free(c->lumMmx2FilterCode);
1478 av_free(c->chrMmx2FilterCode);
1479 #endif
1480 c->lumMmx2FilterCode=NULL;
1481 c->chrMmx2FilterCode=NULL;
1482 #endif /* HAVE_MMX */
1483
1484 av_freep(&c->yuvTable);
1485 av_free(c->formatConvBuffer);
1486
1487 av_free(c);
1488 }
1489
1490 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
1491 int srcW, int srcH, enum PixelFormat srcFormat,
1492 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
1493 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
1494 {
1495 static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};
1496
1497 if (!param)
1498 param = default_param;
1499
1500 if (context &&
1501 (context->srcW != srcW ||
1502 context->srcH != srcH ||
1503 context->srcFormat != srcFormat ||
1504 context->dstW != dstW ||
1505 context->dstH != dstH ||
1506 context->dstFormat != dstFormat ||
1507 context->flags != flags ||
1508 context->param[0] != param[0] ||
1509 context->param[1] != param[1])) {
1510 sws_freeContext(context);
1511 context = NULL;
1512 }
1513
1514 if (!context) {
1515 if (!(context = sws_alloc_context()))
1516 return NULL;
1517 context->srcW = srcW;
1518 context->srcH = srcH;
1519 context->srcRange = handle_jpeg(&srcFormat);
1520 context->srcFormat = srcFormat;
1521 context->dstW = dstW;
1522 context->dstH = dstH;
1523 context->dstRange = handle_jpeg(&dstFormat);
1524 context->dstFormat = dstFormat;
1525 context->flags = flags;
1526 context->param[0] = param[0];
1527 context->param[1] = param[1];
1528 sws_setColorspaceDetails(context, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], context->srcRange, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/, context->dstRange, 0, 1<<16, 1<<16);
1529 if (sws_init_context(context, srcFilter, dstFilter) < 0) {
1530 sws_freeContext(context);
1531 return NULL;
1532 }
1533 }
1534 return context;
1535 }
1536