snowdec.c
Go to the documentation of this file.
1 /*
2  * Copyright (C) 2004 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 #include "libavutil/intmath.h"
22 #include "libavutil/log.h"
23 #include "libavutil/opt.h"
24 #include "avcodec.h"
25 #include "dsputil.h"
26 #include "dwt.h"
27 #include "internal.h"
28 #include "snow.h"
29 
30 #include "rangecoder.h"
31 #include "mathops.h"
32 
33 #include "mpegvideo.h"
34 #include "h263.h"
35 
36 #undef NDEBUG
37 #include <assert.h>
38 
39 static av_always_inline void predict_slice_buffered(SnowContext *s, slice_buffer * sb, IDWTELEM * old_buffer, int plane_index, int add, int mb_y){
40  Plane *p= &s->plane[plane_index];
41  const int mb_w= s->b_width << s->block_max_depth;
42  const int mb_h= s->b_height << s->block_max_depth;
43  int x, y, mb_x;
44  int block_size = MB_SIZE >> s->block_max_depth;
45  int block_w = plane_index ? block_size/2 : block_size;
46  const uint8_t *obmc = plane_index ? ff_obmc_tab[s->block_max_depth+1] : ff_obmc_tab[s->block_max_depth];
47  int obmc_stride= plane_index ? block_size : 2*block_size;
48  int ref_stride= s->current_picture.linesize[plane_index];
49  uint8_t *dst8= s->current_picture.data[plane_index];
50  int w= p->width;
51  int h= p->height;
52 
53  if(s->keyframe || (s->avctx->debug&512)){
54  if(mb_y==mb_h)
55  return;
56 
57  if(add){
58  for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++){
59 // DWTELEM * line = slice_buffer_get_line(sb, y);
60  IDWTELEM * line = sb->line[y];
61  for(x=0; x<w; x++){
62 // int v= buf[x + y*w] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1));
63  int v= line[x] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1));
64  v >>= FRAC_BITS;
65  if(v&(~255)) v= ~(v>>31);
66  dst8[x + y*ref_stride]= v;
67  }
68  }
69  }else{
70  for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++){
71 // DWTELEM * line = slice_buffer_get_line(sb, y);
72  IDWTELEM * line = sb->line[y];
73  for(x=0; x<w; x++){
74  line[x] -= 128 << FRAC_BITS;
75 // buf[x + y*w]-= 128<<FRAC_BITS;
76  }
77  }
78  }
79 
80  return;
81  }
82 
83  for(mb_x=0; mb_x<=mb_w; mb_x++){
84  add_yblock(s, 1, sb, old_buffer, dst8, obmc,
85  block_w*mb_x - block_w/2,
86  block_w*mb_y - block_w/2,
87  block_w, block_w,
88  w, h,
89  w, ref_stride, obmc_stride,
90  mb_x - 1, mb_y - 1,
91  add, 0, plane_index);
92  }
93 }
94 
95 static inline void decode_subband_slice_buffered(SnowContext *s, SubBand *b, slice_buffer * sb, int start_y, int h, int save_state[1]){
96  const int w= b->width;
97  int y;
98  const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16);
99  int qmul= ff_qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
100  int qadd= (s->qbias*qmul)>>QBIAS_SHIFT;
101  int new_index = 0;
102 
103  if(b->ibuf == s->spatial_idwt_buffer || s->qlog == LOSSLESS_QLOG){
104  qadd= 0;
105  qmul= 1<<QEXPSHIFT;
106  }
107 
108  /* If we are on the second or later slice, restore our index. */
109  if (start_y != 0)
110  new_index = save_state[0];
111 
112 
113  for(y=start_y; y<h; y++){
114  int x = 0;
115  int v;
117  memset(line, 0, b->width*sizeof(IDWTELEM));
118  v = b->x_coeff[new_index].coeff;
119  x = b->x_coeff[new_index++].x;
120  while(x < w){
121  register int t= ( (v>>1)*qmul + qadd)>>QEXPSHIFT;
122  register int u= -(v&1);
123  line[x] = (t^u) - u;
124 
125  v = b->x_coeff[new_index].coeff;
126  x = b->x_coeff[new_index++].x;
127  }
128  }
129 
130  /* Save our variables for the next slice. */
131  save_state[0] = new_index;
132 
133  return;
134 }
135 
136 static int decode_q_branch(SnowContext *s, int level, int x, int y){
137  const int w= s->b_width << s->block_max_depth;
138  const int rem_depth= s->block_max_depth - level;
139  const int index= (x + y*w) << rem_depth;
140  int trx= (x+1)<<rem_depth;
141  const BlockNode *left = x ? &s->block[index-1] : &null_block;
142  const BlockNode *top = y ? &s->block[index-w] : &null_block;
143  const BlockNode *tl = y && x ? &s->block[index-w-1] : left;
144  const BlockNode *tr = y && trx<w && ((x&1)==0 || level==0) ? &s->block[index-w+(1<<rem_depth)] : tl; //FIXME use lt
145  int s_context= 2*left->level + 2*top->level + tl->level + tr->level;
146  int res;
147 
148  if(s->keyframe){
150  return 0;
151  }
152 
153  if(level==s->block_max_depth || get_rac(&s->c, &s->block_state[4 + s_context])){
154  int type, mx, my;
155  int l = left->color[0];
156  int cb= left->color[1];
157  int cr= left->color[2];
158  int ref = 0;
159  int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref);
160  int mx_context= av_log2(2*FFABS(left->mx - top->mx)) + 0*av_log2(2*FFABS(tr->mx - top->mx));
161  int my_context= av_log2(2*FFABS(left->my - top->my)) + 0*av_log2(2*FFABS(tr->my - top->my));
162 
163  type= get_rac(&s->c, &s->block_state[1 + left->type + top->type]) ? BLOCK_INTRA : 0;
164 
165  if(type){
166  pred_mv(s, &mx, &my, 0, left, top, tr);
167  l += get_symbol(&s->c, &s->block_state[32], 1);
168  cb+= get_symbol(&s->c, &s->block_state[64], 1);
169  cr+= get_symbol(&s->c, &s->block_state[96], 1);
170  }else{
171  if(s->ref_frames > 1)
172  ref= get_symbol(&s->c, &s->block_state[128 + 1024 + 32*ref_context], 0);
173  if (ref >= s->ref_frames) {
174  av_log(s->avctx, AV_LOG_ERROR, "Invalid ref\n");
175  return AVERROR_INVALIDDATA;
176  }
177  pred_mv(s, &mx, &my, ref, left, top, tr);
178  mx+= get_symbol(&s->c, &s->block_state[128 + 32*(mx_context + 16*!!ref)], 1);
179  my+= get_symbol(&s->c, &s->block_state[128 + 32*(my_context + 16*!!ref)], 1);
180  }
181  set_blocks(s, level, x, y, l, cb, cr, mx, my, ref, type);
182  }else{
183  if ((res = decode_q_branch(s, level+1, 2*x+0, 2*y+0)) < 0 ||
184  (res = decode_q_branch(s, level+1, 2*x+1, 2*y+0)) < 0 ||
185  (res = decode_q_branch(s, level+1, 2*x+0, 2*y+1)) < 0 ||
186  (res = decode_q_branch(s, level+1, 2*x+1, 2*y+1)) < 0)
187  return res;
188  }
189  return 0;
190 }
191 
192 static void dequantize_slice_buffered(SnowContext *s, slice_buffer * sb, SubBand *b, IDWTELEM *src, int stride, int start_y, int end_y){
193  const int w= b->width;
194  const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16);
195  const int qmul= ff_qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
196  const int qadd= (s->qbias*qmul)>>QBIAS_SHIFT;
197  int x,y;
198 
199  if(s->qlog == LOSSLESS_QLOG) return;
200 
201  for(y=start_y; y<end_y; y++){
202 // DWTELEM * line = slice_buffer_get_line_from_address(sb, src + (y * stride));
204  for(x=0; x<w; x++){
205  int i= line[x];
206  if(i<0){
207  line[x]= -((-i*qmul + qadd)>>(QEXPSHIFT)); //FIXME try different bias
208  }else if(i>0){
209  line[x]= (( i*qmul + qadd)>>(QEXPSHIFT));
210  }
211  }
212  }
213 }
214 
215 static void correlate_slice_buffered(SnowContext *s, slice_buffer * sb, SubBand *b, IDWTELEM *src, int stride, int inverse, int use_median, int start_y, int end_y){
216  const int w= b->width;
217  int x,y;
218 
219  IDWTELEM * line=0; // silence silly "could be used without having been initialized" warning
220  IDWTELEM * prev;
221 
222  if (start_y != 0)
223  line = slice_buffer_get_line(sb, ((start_y - 1) * b->stride_line) + b->buf_y_offset) + b->buf_x_offset;
224 
225  for(y=start_y; y<end_y; y++){
226  prev = line;
227 // line = slice_buffer_get_line_from_address(sb, src + (y * stride));
228  line = slice_buffer_get_line(sb, (y * b->stride_line) + b->buf_y_offset) + b->buf_x_offset;
229  for(x=0; x<w; x++){
230  if(x){
231  if(use_median){
232  if(y && x+1<w) line[x] += mid_pred(line[x - 1], prev[x], prev[x + 1]);
233  else line[x] += line[x - 1];
234  }else{
235  if(y) line[x] += mid_pred(line[x - 1], prev[x], line[x - 1] + prev[x] - prev[x - 1]);
236  else line[x] += line[x - 1];
237  }
238  }else{
239  if(y) line[x] += prev[x];
240  }
241  }
242  }
243 }
244 
245 static void decode_qlogs(SnowContext *s){
246  int plane_index, level, orientation;
247 
248  for(plane_index=0; plane_index<3; plane_index++){
249  for(level=0; level<s->spatial_decomposition_count; level++){
250  for(orientation=level ? 1:0; orientation<4; orientation++){
251  int q;
252  if (plane_index==2) q= s->plane[1].band[level][orientation].qlog;
253  else if(orientation==2) q= s->plane[plane_index].band[level][1].qlog;
254  else q= get_symbol(&s->c, s->header_state, 1);
255  s->plane[plane_index].band[level][orientation].qlog= q;
256  }
257  }
258  }
259 }
260 
261 #define GET_S(dst, check) \
262  tmp= get_symbol(&s->c, s->header_state, 0);\
263  if(!(check)){\
264  av_log(s->avctx, AV_LOG_ERROR, "Error " #dst " is %d\n", tmp);\
265  return -1;\
266  }\
267  dst= tmp;
268 
269 static int decode_header(SnowContext *s){
270  int plane_index, tmp;
271  uint8_t kstate[32];
272 
273  memset(kstate, MID_STATE, sizeof(kstate));
274 
275  s->keyframe= get_rac(&s->c, kstate);
276  if(s->keyframe || s->always_reset){
279  s->qlog=
280  s->qbias=
281  s->mv_scale=
282  s->block_max_depth= 0;
283  }
284  if(s->keyframe){
285  GET_S(s->version, tmp <= 0U)
286  s->always_reset= get_rac(&s->c, s->header_state);
290  s->colorspace_type= get_symbol(&s->c, s->header_state, 0);
291  s->chroma_h_shift= get_symbol(&s->c, s->header_state, 0);
292  s->chroma_v_shift= get_symbol(&s->c, s->header_state, 0);
294 // s->rate_scalability= get_rac(&s->c, s->header_state);
295  GET_S(s->max_ref_frames, tmp < (unsigned)MAX_REF_FRAMES)
296  s->max_ref_frames++;
297 
298  decode_qlogs(s);
299  }
300 
301  if(!s->keyframe){
302  if(get_rac(&s->c, s->header_state)){
303  for(plane_index=0; plane_index<2; plane_index++){
304  int htaps, i, sum=0;
305  Plane *p= &s->plane[plane_index];
306  p->diag_mc= get_rac(&s->c, s->header_state);
307  htaps= get_symbol(&s->c, s->header_state, 0)*2 + 2;
308  if((unsigned)htaps > HTAPS_MAX || htaps==0)
309  return -1;
310  p->htaps= htaps;
311  for(i= htaps/2; i; i--){
312  p->hcoeff[i]= get_symbol(&s->c, s->header_state, 0) * (1-2*(i&1));
313  sum += p->hcoeff[i];
314  }
315  p->hcoeff[0]= 32-sum;
316  }
317  s->plane[2].diag_mc= s->plane[1].diag_mc;
318  s->plane[2].htaps = s->plane[1].htaps;
319  memcpy(s->plane[2].hcoeff, s->plane[1].hcoeff, sizeof(s->plane[1].hcoeff));
320  }
321  if(get_rac(&s->c, s->header_state)){
323  decode_qlogs(s);
324  }
325  }
326 
328  if(s->spatial_decomposition_type > 1U){
329  av_log(s->avctx, AV_LOG_ERROR, "spatial_decomposition_type %d not supported", s->spatial_decomposition_type);
330  return -1;
331  }
332  if(FFMIN(s->avctx-> width>>s->chroma_h_shift,
333  s->avctx->height>>s->chroma_v_shift) >> (s->spatial_decomposition_count-1) <= 0){
334  av_log(s->avctx, AV_LOG_ERROR, "spatial_decomposition_count %d too large for size", s->spatial_decomposition_count);
335  return -1;
336  }
337 
338  if (s->chroma_h_shift != 1 || s->chroma_v_shift != 1) {
339  av_log(s->avctx, AV_LOG_ERROR, "Invalid chroma shift\n");
340  return AVERROR_PATCHWELCOME;
341  }
342 
343  s->qlog += get_symbol(&s->c, s->header_state, 1);
344  s->mv_scale += get_symbol(&s->c, s->header_state, 1);
345  s->qbias += get_symbol(&s->c, s->header_state, 1);
346  s->block_max_depth+= get_symbol(&s->c, s->header_state, 1);
347  if(s->block_max_depth > 1 || s->block_max_depth < 0){
348  av_log(s->avctx, AV_LOG_ERROR, "block_max_depth= %d is too large", s->block_max_depth);
349  s->block_max_depth= 0;
350  return -1;
351  }
352 
353  return 0;
354 }
355 
357 {
358  int ret;
359 
360  avctx->pix_fmt= AV_PIX_FMT_YUV420P;
361 
362  if ((ret = ff_snow_common_init(avctx)) < 0) {
364  return ret;
365  }
366 
367  return 0;
368 }
369 
370 static int decode_blocks(SnowContext *s){
371  int x, y;
372  int w= s->b_width;
373  int h= s->b_height;
374  int res;
375 
376  for(y=0; y<h; y++){
377  for(x=0; x<w; x++){
378  if ((res = decode_q_branch(s, 0, x, y)) < 0)
379  return res;
380  }
381  }
382  return 0;
383 }
384 
385 static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame,
386  AVPacket *avpkt)
387 {
388  const uint8_t *buf = avpkt->data;
389  int buf_size = avpkt->size;
390  SnowContext *s = avctx->priv_data;
391  RangeCoder * const c= &s->c;
392  int bytes_read;
393  AVFrame *picture = data;
394  int level, orientation, plane_index;
395  int res;
396 
397  ff_init_range_decoder(c, buf, buf_size);
398  ff_build_rac_states(c, 0.05*(1LL<<32), 256-8);
399 
400  s->current_picture.pict_type= AV_PICTURE_TYPE_I; //FIXME I vs. P
401  if(decode_header(s)<0)
402  return -1;
404 
405  // realloc slice buffer for the case that spatial_decomposition_count changed
407  if ((res = ff_slice_buffer_init(&s->sb, s->plane[0].height,
408  (MB_SIZE >> s->block_max_depth) +
409  s->spatial_decomposition_count * 8 + 1,
410  s->plane[0].width,
411  s->spatial_idwt_buffer)) < 0)
412  return res;
413 
414  for(plane_index=0; plane_index<3; plane_index++){
415  Plane *p= &s->plane[plane_index];
416  p->fast_mc= p->diag_mc && p->htaps==6 && p->hcoeff[0]==40
417  && p->hcoeff[1]==-10
418  && p->hcoeff[2]==2;
419  }
420 
422 
423  if(ff_snow_frame_start(s) < 0)
424  return -1;
425  //keyframe flag duplication mess FIXME
426  if(avctx->debug&FF_DEBUG_PICT_INFO)
427  av_log(avctx, AV_LOG_ERROR, "keyframe:%d qlog:%d\n", s->keyframe, s->qlog);
428 
429  if ((res = decode_blocks(s)) < 0)
430  return res;
431 
432  for(plane_index=0; plane_index<3; plane_index++){
433  Plane *p= &s->plane[plane_index];
434  int w= p->width;
435  int h= p->height;
436  int x, y;
437  int decode_state[MAX_DECOMPOSITIONS][4][1]; /* Stored state info for unpack_coeffs. 1 variable per instance. */
438 
439  if(s->avctx->debug&2048){
440  memset(s->spatial_dwt_buffer, 0, sizeof(DWTELEM)*w*h);
441  predict_plane(s, s->spatial_idwt_buffer, plane_index, 1);
442 
443  for(y=0; y<h; y++){
444  for(x=0; x<w; x++){
445  int v= s->current_picture.data[plane_index][y*s->current_picture.linesize[plane_index] + x];
446  s->mconly_picture.data[plane_index][y*s->mconly_picture.linesize[plane_index] + x]= v;
447  }
448  }
449  }
450 
451  {
452  for(level=0; level<s->spatial_decomposition_count; level++){
453  for(orientation=level ? 1 : 0; orientation<4; orientation++){
454  SubBand *b= &p->band[level][orientation];
455  unpack_coeffs(s, b, b->parent, orientation);
456  }
457  }
458  }
459 
460  {
461  const int mb_h= s->b_height << s->block_max_depth;
462  const int block_size = MB_SIZE >> s->block_max_depth;
463  const int block_w = plane_index ? block_size/2 : block_size;
464  int mb_y;
466  int yd=0, yq=0;
467  int y;
468  int end_y;
469 
471  for(mb_y=0; mb_y<=mb_h; mb_y++){
472 
473  int slice_starty = block_w*mb_y;
474  int slice_h = block_w*(mb_y+1);
475  if (!(s->keyframe || s->avctx->debug&512)){
476  slice_starty = FFMAX(0, slice_starty - (block_w >> 1));
477  slice_h -= (block_w >> 1);
478  }
479 
480  for(level=0; level<s->spatial_decomposition_count; level++){
481  for(orientation=level ? 1 : 0; orientation<4; orientation++){
482  SubBand *b= &p->band[level][orientation];
483  int start_y;
484  int end_y;
485  int our_mb_start = mb_y;
486  int our_mb_end = (mb_y + 1);
487  const int extra= 3;
488  start_y = (mb_y ? ((block_w * our_mb_start) >> (s->spatial_decomposition_count - level)) + s->spatial_decomposition_count - level + extra: 0);
489  end_y = (((block_w * our_mb_end) >> (s->spatial_decomposition_count - level)) + s->spatial_decomposition_count - level + extra);
490  if (!(s->keyframe || s->avctx->debug&512)){
491  start_y = FFMAX(0, start_y - (block_w >> (1+s->spatial_decomposition_count - level)));
492  end_y = FFMAX(0, end_y - (block_w >> (1+s->spatial_decomposition_count - level)));
493  }
494  start_y = FFMIN(b->height, start_y);
495  end_y = FFMIN(b->height, end_y);
496 
497  if (start_y != end_y){
498  if (orientation == 0){
499  SubBand * correlate_band = &p->band[0][0];
500  int correlate_end_y = FFMIN(b->height, end_y + 1);
501  int correlate_start_y = FFMIN(b->height, (start_y ? start_y + 1 : 0));
502  decode_subband_slice_buffered(s, correlate_band, &s->sb, correlate_start_y, correlate_end_y, decode_state[0][0]);
503  correlate_slice_buffered(s, &s->sb, correlate_band, correlate_band->ibuf, correlate_band->stride, 1, 0, correlate_start_y, correlate_end_y);
504  dequantize_slice_buffered(s, &s->sb, correlate_band, correlate_band->ibuf, correlate_band->stride, start_y, end_y);
505  }
506  else
507  decode_subband_slice_buffered(s, b, &s->sb, start_y, end_y, decode_state[level][orientation]);
508  }
509  }
510  }
511 
512  for(; yd<slice_h; yd+=4){
514  }
515 
516  if(s->qlog == LOSSLESS_QLOG){
517  for(; yq<slice_h && yq<h; yq++){
518  IDWTELEM * line = slice_buffer_get_line(&s->sb, yq);
519  for(x=0; x<w; x++){
520  line[x] <<= FRAC_BITS;
521  }
522  }
523  }
524 
525  predict_slice_buffered(s, &s->sb, s->spatial_idwt_buffer, plane_index, 1, mb_y);
526 
527  y = FFMIN(p->height, slice_starty);
528  end_y = FFMIN(p->height, slice_h);
529  while(y < end_y)
530  ff_slice_buffer_release(&s->sb, y++);
531  }
532 
534  }
535 
536  }
537 
538  emms_c();
539 
540  ff_snow_release_buffer(avctx);
541 
542  if(!(s->avctx->debug&2048))
543  *picture= s->current_picture;
544  else
545  *picture= s->mconly_picture;
546 
547  *got_frame = 1;
548 
549  bytes_read= c->bytestream - c->bytestream_start;
550  if(bytes_read ==0) av_log(s->avctx, AV_LOG_ERROR, "error at end of frame\n"); //FIXME
551 
552  return bytes_read;
553 }
554 
556 {
557  SnowContext *s = avctx->priv_data;
558 
560 
562 
563  return 0;
564 }
565 
567  .name = "snow",
568  .type = AVMEDIA_TYPE_VIDEO,
569  .id = AV_CODEC_ID_SNOW,
570  .priv_data_size = sizeof(SnowContext),
571  .init = decode_init,
572  .close = decode_end,
573  .decode = decode_frame,
574  .capabilities = CODEC_CAP_DR1 /*| CODEC_CAP_DRAW_HORIZ_BAND*/,
575  .long_name = NULL_IF_CONFIG_SMALL("Snow"),
576 };