FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
dirac_dwt.c
Go to the documentation of this file.
1 /*
2  * Copyright (C) 2004-2010 Michael Niedermayer <michaelni@gmx.at>
3  * Copyright (C) 2008 David Conrad
4  *
5  * This file is part of FFmpeg.
6  *
7  * FFmpeg is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * FFmpeg is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with FFmpeg; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
22 #include "libavutil/attributes.h"
23 #include "libavutil/avassert.h"
24 #include "libavutil/common.h"
25 #include "dsputil.h"
26 #include "dirac_dwt.h"
28 
29 
30 static inline int mirror(int v, int m)
31 {
32  while ((unsigned)v > (unsigned)m) {
33  v = -v;
34  if (v < 0)
35  v += 2 * m;
36  }
37  return v;
38 }
39 
40 static void vertical_compose53iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
41  int width)
42 {
43  int i;
44 
45  for (i = 0; i < width; i++)
46  b1[i] -= (b0[i] + b2[i] + 2) >> 2;
47 }
48 
49 
50 static av_always_inline
51 void interleave(IDWTELEM *dst, IDWTELEM *src0, IDWTELEM *src1, int w2, int add, int shift)
52 {
53  int i;
54  for (i = 0; i < w2; i++) {
55  dst[2*i ] = (src0[i] + add) >> shift;
56  dst[2*i+1] = (src1[i] + add) >> shift;
57  }
58 }
59 
61 {
62  const int w2 = w >> 1;
63  int x;
64 
65  temp[0] = COMPOSE_53iL0(b[w2], b[0], b[w2]);
66  for (x = 1; x < w2; x++) {
67  temp[x ] = COMPOSE_53iL0 (b[x+w2-1], b[x ], b[x+w2]);
68  temp[x+w2-1] = COMPOSE_DIRAC53iH0(temp[x-1], b[x+w2-1], temp[x]);
69  }
70  temp[w-1] = COMPOSE_DIRAC53iH0(temp[w2-1], b[w-1], temp[w2-1]);
71 
72  interleave(b, temp, temp+w2, w2, 1, 1);
73 }
74 
75 static void horizontal_compose_dd97i(IDWTELEM *b, IDWTELEM *tmp, int w)
76 {
77  const int w2 = w >> 1;
78  int x;
79 
80  tmp[0] = COMPOSE_53iL0(b[w2], b[0], b[w2]);
81  for (x = 1; x < w2; x++)
82  tmp[x] = COMPOSE_53iL0(b[x+w2-1], b[x], b[x+w2]);
83 
84  // extend the edges
85  tmp[-1] = tmp[0];
86  tmp[w2+1] = tmp[w2] = tmp[w2-1];
87 
88  for (x = 0; x < w2; x++) {
89  b[2*x ] = (tmp[x] + 1)>>1;
90  b[2*x+1] = (COMPOSE_DD97iH0(tmp[x-1], tmp[x], b[x+w2], tmp[x+1], tmp[x+2]) + 1)>>1;
91  }
92 }
93 
94 static void horizontal_compose_dd137i(IDWTELEM *b, IDWTELEM *tmp, int w)
95 {
96  const int w2 = w >> 1;
97  int x;
98 
99  tmp[0] = COMPOSE_DD137iL0(b[w2], b[w2], b[0], b[w2 ], b[w2+1]);
100  tmp[1] = COMPOSE_DD137iL0(b[w2], b[w2], b[1], b[w2+1], b[w2+2]);
101  for (x = 2; x < w2-1; x++)
102  tmp[x] = COMPOSE_DD137iL0(b[x+w2-2], b[x+w2-1], b[x], b[x+w2], b[x+w2+1]);
103  tmp[w2-1] = COMPOSE_DD137iL0(b[w-3], b[w-2], b[w2-1], b[w-1], b[w-1]);
104 
105  // extend the edges
106  tmp[-1] = tmp[0];
107  tmp[w2+1] = tmp[w2] = tmp[w2-1];
108 
109  for (x = 0; x < w2; x++) {
110  b[2*x ] = (tmp[x] + 1)>>1;
111  b[2*x+1] = (COMPOSE_DD97iH0(tmp[x-1], tmp[x], b[x+w2], tmp[x+1], tmp[x+2]) + 1)>>1;
112  }
113 }
114 
115 static av_always_inline
117 {
118  const int w2 = w >> 1;
119  int x;
120 
121  for (x = 0; x < w2; x++) {
122  temp[x ] = COMPOSE_HAARiL0(b[x ], b[x+w2]);
123  temp[x+w2] = COMPOSE_HAARiH0(b[x+w2], temp[x]);
124  }
125 
126  interleave(b, temp, temp+w2, w2, shift, shift);
127 }
128 
130 {
131  horizontal_compose_haari(b, temp, w, 0);
132 }
133 
135 {
136  horizontal_compose_haari(b, temp, w, 1);
137 }
138 
140 {
141  const int w2 = w >> 1;
142  int i, x;
143  IDWTELEM v[8];
144 
145  for (x = 0; x < w2; x++) {
146  for (i = 0; i < 8; i++)
147  v[i] = b[av_clip(x-3+i, 0, w2-1)];
148  tmp[x] = COMPOSE_FIDELITYiH0(v[0], v[1], v[2], v[3], b[x+w2], v[4], v[5], v[6], v[7]);
149  }
150 
151  for (x = 0; x < w2; x++) {
152  for (i = 0; i < 8; i++)
153  v[i] = tmp[av_clip(x-4+i, 0, w2-1)];
154  tmp[x+w2] = COMPOSE_FIDELITYiL0(v[0], v[1], v[2], v[3], b[x], v[4], v[5], v[6], v[7]);
155  }
156 
157  interleave(b, tmp+w2, tmp, w2, 0, 0);
158 }
159 
161 {
162  const int w2 = w >> 1;
163  int x, b0, b1, b2;
164 
165  temp[0] = COMPOSE_DAUB97iL1(b[w2], b[0], b[w2]);
166  for (x = 1; x < w2; x++) {
167  temp[x ] = COMPOSE_DAUB97iL1(b[x+w2-1], b[x ], b[x+w2]);
168  temp[x+w2-1] = COMPOSE_DAUB97iH1(temp[x-1], b[x+w2-1], temp[x]);
169  }
170  temp[w-1] = COMPOSE_DAUB97iH1(temp[w2-1], b[w-1], temp[w2-1]);
171 
172  // second stage combined with interleave and shift
173  b0 = b2 = COMPOSE_DAUB97iL0(temp[w2], temp[0], temp[w2]);
174  b[0] = (b0 + 1) >> 1;
175  for (x = 1; x < w2; x++) {
176  b2 = COMPOSE_DAUB97iL0(temp[x+w2-1], temp[x ], temp[x+w2]);
177  b1 = COMPOSE_DAUB97iH0( b0, temp[x+w2-1], b2 );
178  b[2*x-1] = (b1 + 1) >> 1;
179  b[2*x ] = (b2 + 1) >> 1;
180  b0 = b2;
181  }
182  b[w-1] = (COMPOSE_DAUB97iH0(b2, temp[w-1], b2) + 1) >> 1;
183 }
184 
186 {
187  int i;
188 
189  for(i=0; i<width; i++){
190  b1[i] = COMPOSE_DIRAC53iH0(b0[i], b1[i], b2[i]);
191  }
192 }
193 
195  IDWTELEM *b3, IDWTELEM *b4, int width)
196 {
197  int i;
198 
199  for(i=0; i<width; i++){
200  b2[i] = COMPOSE_DD97iH0(b0[i], b1[i], b2[i], b3[i], b4[i]);
201  }
202 }
203 
205  IDWTELEM *b3, IDWTELEM *b4, int width)
206 {
207  int i;
208 
209  for(i=0; i<width; i++){
210  b2[i] = COMPOSE_DD137iL0(b0[i], b1[i], b2[i], b3[i], b4[i]);
211  }
212 }
213 
214 static void vertical_compose_haar(IDWTELEM *b0, IDWTELEM *b1, int width)
215 {
216  int i;
217 
218  for (i = 0; i < width; i++) {
219  b0[i] = COMPOSE_HAARiL0(b0[i], b1[i]);
220  b1[i] = COMPOSE_HAARiH0(b1[i], b0[i]);
221  }
222 }
223 
225 {
226  int i;
227 
228  for(i=0; i<width; i++){
229  dst[i] = COMPOSE_FIDELITYiH0(b[0][i], b[1][i], b[2][i], b[3][i], dst[i], b[4][i], b[5][i], b[6][i], b[7][i]);
230  }
231 }
232 
234 {
235  int i;
236 
237  for(i=0; i<width; i++){
238  dst[i] = COMPOSE_FIDELITYiL0(b[0][i], b[1][i], b[2][i], b[3][i], dst[i], b[4][i], b[5][i], b[6][i], b[7][i]);
239  }
240 }
241 
243 {
244  int i;
245 
246  for(i=0; i<width; i++){
247  b1[i] = COMPOSE_DAUB97iH0(b0[i], b1[i], b2[i]);
248  }
249 }
250 
252 {
253  int i;
254 
255  for(i=0; i<width; i++){
256  b1[i] = COMPOSE_DAUB97iH1(b0[i], b1[i], b2[i]);
257  }
258 }
259 
261 {
262  int i;
263 
264  for(i=0; i<width; i++){
265  b1[i] = COMPOSE_DAUB97iL0(b0[i], b1[i], b2[i]);
266  }
267 }
268 
270 {
271  int i;
272 
273  for(i=0; i<width; i++){
274  b1[i] = COMPOSE_DAUB97iL1(b0[i], b1[i], b2[i]);
275  }
276 }
277 
278 
279 static void spatial_compose_dd97i_dy(DWTContext *d, int level, int width, int height, int stride)
280 {
281  vertical_compose_3tap vertical_compose_l0 = (void*)d->vertical_compose_l0;
282  vertical_compose_5tap vertical_compose_h0 = (void*)d->vertical_compose_h0;
283  DWTCompose *cs = d->cs + level;
284 
285  int i, y = cs->y;
286  IDWTELEM *b[8];
287  for (i = 0; i < 6; i++)
288  b[i] = cs->b[i];
289  b[6] = d->buffer + av_clip(y+5, 0, height-2)*stride;
290  b[7] = d->buffer + av_clip(y+6, 1, height-1)*stride;
291 
292  if(y+5<(unsigned)height) vertical_compose_l0( b[5], b[6], b[7], width);
293  if(y+1<(unsigned)height) vertical_compose_h0(b[0], b[2], b[3], b[4], b[6], width);
294 
295  if(y-1<(unsigned)height) d->horizontal_compose(b[0], d->temp, width);
296  if(y+0<(unsigned)height) d->horizontal_compose(b[1], d->temp, width);
297 
298  for (i = 0; i < 6; i++)
299  cs->b[i] = b[i+2];
300  cs->y += 2;
301 }
302 
303 static void spatial_compose_dirac53i_dy(DWTContext *d, int level, int width, int height, int stride)
304 {
305  vertical_compose_3tap vertical_compose_l0 = (void*)d->vertical_compose_l0;
306  vertical_compose_3tap vertical_compose_h0 = (void*)d->vertical_compose_h0;
307  DWTCompose *cs = d->cs + level;
308 
309  int y= cs->y;
310  IDWTELEM *b[4] = { cs->b[0], cs->b[1] };
311  b[2] = d->buffer + mirror(y+1, height-1)*stride;
312  b[3] = d->buffer + mirror(y+2, height-1)*stride;
313 
314  if(y+1<(unsigned)height) vertical_compose_l0(b[1], b[2], b[3], width);
315  if(y+0<(unsigned)height) vertical_compose_h0(b[0], b[1], b[2], width);
316 
317  if(y-1<(unsigned)height) d->horizontal_compose(b[0], d->temp, width);
318  if(y+0<(unsigned)height) d->horizontal_compose(b[1], d->temp, width);
319 
320  cs->b[0] = b[2];
321  cs->b[1] = b[3];
322  cs->y += 2;
323 }
324 
325 
326 static void spatial_compose_dd137i_dy(DWTContext *d, int level, int width, int height, int stride)
327 {
328  vertical_compose_5tap vertical_compose_l0 = (void*)d->vertical_compose_l0;
329  vertical_compose_5tap vertical_compose_h0 = (void*)d->vertical_compose_h0;
330  DWTCompose *cs = d->cs + level;
331 
332  int i, y = cs->y;
333  IDWTELEM *b[10];
334  for (i = 0; i < 8; i++)
335  b[i] = cs->b[i];
336  b[8] = d->buffer + av_clip(y+7, 0, height-2)*stride;
337  b[9] = d->buffer + av_clip(y+8, 1, height-1)*stride;
338 
339  if(y+5<(unsigned)height) vertical_compose_l0(b[3], b[5], b[6], b[7], b[9], width);
340  if(y+1<(unsigned)height) vertical_compose_h0(b[0], b[2], b[3], b[4], b[6], width);
341 
342  if(y-1<(unsigned)height) d->horizontal_compose(b[0], d->temp, width);
343  if(y+0<(unsigned)height) d->horizontal_compose(b[1], d->temp, width);
344 
345  for (i = 0; i < 8; i++)
346  cs->b[i] = b[i+2];
347  cs->y += 2;
348 }
349 
350 // haar makes the assumption that height is even (always true for dirac)
351 static void spatial_compose_haari_dy(DWTContext *d, int level, int width, int height, int stride)
352 {
353  vertical_compose_2tap vertical_compose = (void*)d->vertical_compose;
354  int y = d->cs[level].y;
355  IDWTELEM *b0 = d->buffer + (y-1)*stride;
356  IDWTELEM *b1 = d->buffer + (y )*stride;
357 
358  vertical_compose(b0, b1, width);
359  d->horizontal_compose(b0, d->temp, width);
360  d->horizontal_compose(b1, d->temp, width);
361 
362  d->cs[level].y += 2;
363 }
364 
365 // Don't do sliced idwt for fidelity; the 9 tap filter makes it a bit annoying
366 // Fortunately, this filter isn't used in practice.
367 static void spatial_compose_fidelity(DWTContext *d, int level, int width, int height, int stride)
368 {
369  vertical_compose_9tap vertical_compose_l0 = (void*)d->vertical_compose_l0;
370  vertical_compose_9tap vertical_compose_h0 = (void*)d->vertical_compose_h0;
371  int i, y;
372  IDWTELEM *b[8];
373 
374  for (y = 1; y < height; y += 2) {
375  for (i = 0; i < 8; i++)
376  b[i] = d->buffer + av_clip((y-7 + 2*i), 0, height-2)*stride;
377  vertical_compose_h0(d->buffer + y*stride, b, width);
378  }
379 
380  for (y = 0; y < height; y += 2) {
381  for (i = 0; i < 8; i++)
382  b[i] = d->buffer + av_clip((y-7 + 2*i), 1, height-1)*stride;
383  vertical_compose_l0(d->buffer + y*stride, b, width);
384  }
385 
386  for (y = 0; y < height; y++)
387  d->horizontal_compose(d->buffer + y*stride, d->temp, width);
388 
389  d->cs[level].y = height+1;
390 }
391 
392 static void spatial_compose_daub97i_dy(DWTContext *d, int level, int width, int height, int stride)
393 {
394  vertical_compose_3tap vertical_compose_l0 = (void*)d->vertical_compose_l0;
395  vertical_compose_3tap vertical_compose_h0 = (void*)d->vertical_compose_h0;
396  vertical_compose_3tap vertical_compose_l1 = (void*)d->vertical_compose_l1;
397  vertical_compose_3tap vertical_compose_h1 = (void*)d->vertical_compose_h1;
398  DWTCompose *cs = d->cs + level;
399 
400  int i, y = cs->y;
401  IDWTELEM *b[6];
402  for (i = 0; i < 4; i++)
403  b[i] = cs->b[i];
404  b[4] = d->buffer + mirror(y+3, height-1)*stride;
405  b[5] = d->buffer + mirror(y+4, height-1)*stride;
406 
407  if(y+3<(unsigned)height) vertical_compose_l1(b[3], b[4], b[5], width);
408  if(y+2<(unsigned)height) vertical_compose_h1(b[2], b[3], b[4], width);
409  if(y+1<(unsigned)height) vertical_compose_l0(b[1], b[2], b[3], width);
410  if(y+0<(unsigned)height) vertical_compose_h0(b[0], b[1], b[2], width);
411 
412  if(y-1<(unsigned)height) d->horizontal_compose(b[0], d->temp, width);
413  if(y+0<(unsigned)height) d->horizontal_compose(b[1], d->temp, width);
414 
415  for (i = 0; i < 4; i++)
416  cs->b[i] = b[i+2];
417  cs->y += 2;
418 }
419 
420 
422 {
423  cs->b[0] = buffer + mirror(-3-1, height-1)*stride;
424  cs->b[1] = buffer + mirror(-3 , height-1)*stride;
425  cs->b[2] = buffer + mirror(-3+1, height-1)*stride;
426  cs->b[3] = buffer + mirror(-3+2, height-1)*stride;
427  cs->y = -3;
428 }
429 
431 {
432  cs->b[0] = buffer + mirror(-1-1, height-1)*stride;
433  cs->b[1] = buffer + mirror(-1 , height-1)*stride;
434  cs->y = -1;
435 }
436 
438 {
439  cs->b[0] = buffer + av_clip(-5-1, 0, height-2)*stride;
440  cs->b[1] = buffer + av_clip(-5 , 1, height-1)*stride;
441  cs->b[2] = buffer + av_clip(-5+1, 0, height-2)*stride;
442  cs->b[3] = buffer + av_clip(-5+2, 1, height-1)*stride;
443  cs->b[4] = buffer + av_clip(-5+3, 0, height-2)*stride;
444  cs->b[5] = buffer + av_clip(-5+4, 1, height-1)*stride;
445  cs->y = -5;
446 }
447 
449 {
450  cs->b[0] = buffer + av_clip(-5-1, 0, height-2)*stride;
451  cs->b[1] = buffer + av_clip(-5 , 1, height-1)*stride;
452  cs->b[2] = buffer + av_clip(-5+1, 0, height-2)*stride;
453  cs->b[3] = buffer + av_clip(-5+2, 1, height-1)*stride;
454  cs->b[4] = buffer + av_clip(-5+3, 0, height-2)*stride;
455  cs->b[5] = buffer + av_clip(-5+4, 1, height-1)*stride;
456  cs->b[6] = buffer + av_clip(-5+5, 0, height-2)*stride;
457  cs->b[7] = buffer + av_clip(-5+6, 1, height-1)*stride;
458  cs->y = -5;
459 }
460 
462  int stride, enum dwt_type type, int decomposition_count,
463  IDWTELEM *temp)
464 {
465  int level;
466 
467  d->buffer = buffer;
468  d->width = width;
469  d->height = height;
470  d->stride = stride;
471  d->decomposition_count = decomposition_count;
472  d->temp = temp + 8;
473 
474  for(level=decomposition_count-1; level>=0; level--){
475  int hl = height >> level;
476  int stride_l = stride << level;
477 
478  switch(type){
479  case DWT_DIRAC_DD9_7:
480  spatial_compose_dd97i_init(d->cs+level, buffer, hl, stride_l);
481  break;
482  case DWT_DIRAC_LEGALL5_3:
483  spatial_compose53i_init2(d->cs+level, buffer, hl, stride_l);
484  break;
485  case DWT_DIRAC_DD13_7:
486  spatial_compose_dd137i_init(d->cs+level, buffer, hl, stride_l);
487  break;
488  case DWT_DIRAC_HAAR0:
489  case DWT_DIRAC_HAAR1:
490  d->cs[level].y = 1;
491  break;
492  case DWT_DIRAC_DAUB9_7:
493  spatial_compose97i_init2(d->cs+level, buffer, hl, stride_l);
494  break;
495  default:
496  d->cs[level].y = 0;
497  break;
498  }
499  }
500 
501  switch (type) {
502  case DWT_DIRAC_DD9_7:
507  d->support = 7;
508  break;
509  case DWT_DIRAC_LEGALL5_3:
514  d->support = 3;
515  break;
516  case DWT_DIRAC_DD13_7:
521  d->support = 7;
522  break;
523  case DWT_DIRAC_HAAR0:
524  case DWT_DIRAC_HAAR1:
527  if (type == DWT_DIRAC_HAAR0)
529  else
531  d->support = 1;
532  break;
533  case DWT_DIRAC_FIDELITY:
538  d->support = 0; // not really used
539  break;
540  case DWT_DIRAC_DAUB9_7:
547  d->support = 5;
548  break;
549  default:
550  av_log(NULL, AV_LOG_ERROR, "Unknown wavelet type %d\n", type);
551  return -1;
552  }
553 
554  if (HAVE_MMX) ff_spatial_idwt_init_mmx(d, type);
555 
556  return 0;
557 }
558 
560 {
561  int level, support = d->support;
562 
563  for (level = d->decomposition_count-1; level >= 0; level--) {
564  int wl = d->width >> level;
565  int hl = d->height >> level;
566  int stride_l = d->stride << level;
567 
568  while (d->cs[level].y <= FFMIN((y>>level)+support, hl))
569  d->spatial_compose(d, level, wl, hl, stride_l);
570  }
571 }
572