utils.c
Go to the documentation of this file.
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 {
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 {
70 } FormatEntry;
71 
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 
152 {
153  return (unsigned)pix_fmt < PIX_FMT_NB ?
154  format_entries[pix_fmt].is_supported_in : 0;
155 }
156 
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, int32_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(**filterPos), 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  int64_t 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  filterSize = av_clip(filterSize, 1, 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 - ((int64_t)(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 {
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 
709 
710  ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
711  //FIXME factorize
712 
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 
747 {
748  SwsContext *c= av_mallocz(sizeof(SwsContext));
749 
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
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= (((int64_t)srcW<<16) + (dstW>>1))/dstW;
813  c->lumYInc= (((int64_t)srcH<<16) + (dstH>>1))/dstH;
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 
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
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))) {
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;
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= (((int64_t)c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
900  c->chrYInc= (((int64_t)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 = ((int64_t)(srcW-2)<<16)/(dstW-2) - 20;
916  c->chrXInc = ((int64_t)(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)) {
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
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 
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 
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;
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 
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;
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 = (int64_t) 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);
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) {
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) {
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 
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 
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 
1271 {
1272  sws_scaleVec(a, height/sws_dcVec(a));
1273 }
1274 
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 
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 
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 
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 
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 
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 
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 
1413 {
1414  if (!a) return;
1415  av_freep(&a->coeff);
1416  a->length=0;
1417  av_free(a);
1418 }
1419 
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 
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
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
1479 #endif
1482 #endif /* HAVE_MMX */
1483 
1484  av_freep(&c->yuvTable);
1486 
1487  av_free(c);
1488 }
1489 
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