snow.h
Go to the documentation of this file.
1 /*
2  * Copyright (C) 2004 Michael Niedermayer <michaelni@gmx.at>
3  * Copyright (C) 2006 Robert Edele <yartrebo@earthlink.net>
4  *
5  * This file is part of Libav.
6  *
7  * Libav 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  * Libav 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 Libav; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
22 #ifndef AVCODEC_SNOW_H
23 #define AVCODEC_SNOW_H
24 
25 #include "dsputil.h"
26 #include "dwt.h"
27 
28 #include "rangecoder.h"
29 #include "mathops.h"
30 #include "mpegvideo.h"
31 
32 #define MID_STATE 128
33 
34 #define MAX_PLANES 4
35 #define QSHIFT 5
36 #define QROOT (1<<QSHIFT)
37 #define LOSSLESS_QLOG -128
38 #define FRAC_BITS 4
39 #define MAX_REF_FRAMES 8
40 
41 #define LOG2_OBMC_MAX 8
42 #define OBMC_MAX (1<<(LOG2_OBMC_MAX))
43 typedef struct BlockNode{
44  int16_t mx;
45  int16_t my;
46  uint8_t ref;
47  uint8_t color[3];
48  uint8_t type;
49 //#define TYPE_SPLIT 1
50 #define BLOCK_INTRA 1
51 #define BLOCK_OPT 2
52 //#define TYPE_NOCOLOR 4
53  uint8_t level; //FIXME merge into type?
54 }BlockNode;
55 
56 static const BlockNode null_block= { //FIXME add border maybe
57  .color= {128,128,128},
58  .mx= 0,
59  .my= 0,
60  .ref= 0,
61  .type= 0,
62  .level= 0,
63 };
64 
65 #define LOG2_MB_SIZE 4
66 #define MB_SIZE (1<<LOG2_MB_SIZE)
67 #define ENCODER_EXTRA_BITS 4
68 #define HTAPS_MAX 8
69 
70 typedef struct x_and_coeff{
71  int16_t x;
72  uint16_t coeff;
73 } x_and_coeff;
74 
75 typedef struct SubBand{
76  int level;
77  int stride;
78  int width;
79  int height;
80  int qlog;
87  struct SubBand *parent;
88  uint8_t state[/*7*2*/ 7 + 512][32];
89 }SubBand;
90 
91 typedef struct Plane{
92  int width;
93  int height;
95 
96  int htaps;
97  int8_t hcoeff[HTAPS_MAX/2];
98  int diag_mc;
99  int fast_mc;
100 
104 }Plane;
105 
106 typedef struct SnowContext{
107  AVClass *class;
116  uint8_t *halfpel_plane[MAX_REF_FRAMES][4][4];
118 // uint8_t q_context[16];
119  uint8_t header_state[32];
120  uint8_t block_state[128 + 32*128];
121  int keyframe;
123  int version;
132  int16_t (*ref_mvs[MAX_REF_FRAMES])[2];
140  int qlog;
142  int lambda;
143  int lambda2;
144  int pass1_rc;
145  int mv_scale;
147  int qbias;
149 #define QBIAS_SHIFT 3
150  int b_width;
151  int b_height;
156 #define ME_CACHE_SIZE 1024
161 
162  MpegEncContext m; // needed for motion estimation, should not be used for anything else, the idea is to eventually make the motion estimation independent of MpegEncContext, so this will be removed then (FIXME/XXX)
163 
164  uint8_t *scratchbuf;
165 }SnowContext;
166 
167 /* Tables */
168 extern const uint8_t * const obmc_tab[4];
169 #ifdef __sgi
170 // Avoid a name clash on SGI IRIX
171 #undef qexp
172 #endif
173 extern uint8_t qexp[QROOT];
175 
176 /* C bits used by mmx/sse2/altivec */
177 
178 static av_always_inline void snow_interleave_line_header(int * i, int width, IDWTELEM * low, IDWTELEM * high){
179  (*i) = (width) - 2;
180 
181  if (width & 1){
182  low[(*i)+1] = low[((*i)+1)>>1];
183  (*i)--;
184  }
185 }
186 
187 static av_always_inline void snow_interleave_line_footer(int * i, IDWTELEM * low, IDWTELEM * high){
188  for (; (*i)>=0; (*i)-=2){
189  low[(*i)+1] = high[(*i)>>1];
190  low[*i] = low[(*i)>>1];
191  }
192 }
193 
194 static av_always_inline void snow_horizontal_compose_lift_lead_out(int i, IDWTELEM * dst, IDWTELEM * src, IDWTELEM * ref, int width, int w, int lift_high, int mul, int add, int shift){
195  for(; i<w; i++){
196  dst[i] = src[i] - ((mul * (ref[i] + ref[i + 1]) + add) >> shift);
197  }
198 
199  if((width^lift_high)&1){
200  dst[w] = src[w] - ((mul * 2 * ref[w] + add) >> shift);
201  }
202 }
203 
204 static av_always_inline void snow_horizontal_compose_liftS_lead_out(int i, IDWTELEM * dst, IDWTELEM * src, IDWTELEM * ref, int width, int w){
205  for(; i<w; i++){
206  dst[i] = src[i] + ((ref[i] + ref[(i+1)]+W_BO + 4 * src[i]) >> W_BS);
207  }
208 
209  if(width&1){
210  dst[w] = src[w] + ((2 * ref[w] + W_BO + 4 * src[w]) >> W_BS);
211  }
212 }
213 
214 /* common code */
215 
223 void ff_snow_pred_block(SnowContext *s, uint8_t *dst, uint8_t *tmp, int stride,
224  int sx, int sy, int b_w, int b_h, BlockNode *block,
225  int plane_index, int w, int h);
226 /* common inline functions */
227 //XXX doublecheck all of them should stay inlined
228 
229 static inline void snow_set_blocks(SnowContext *s, int level, int x, int y, int l, int cb, int cr, int mx, int my, int ref, int type){
230  const int w= s->b_width << s->block_max_depth;
231  const int rem_depth= s->block_max_depth - level;
232  const int index= (x + y*w) << rem_depth;
233  const int block_w= 1<<rem_depth;
235  int i,j;
236 
237  block.color[0]= l;
238  block.color[1]= cb;
239  block.color[2]= cr;
240  block.mx= mx;
241  block.my= my;
242  block.ref= ref;
243  block.type= type;
244  block.level= level;
245 
246  for(j=0; j<block_w; j++){
247  for(i=0; i<block_w; i++){
248  s->block[index + i + j*w]= block;
249  }
250  }
251 }
252 
253 static inline void pred_mv(SnowContext *s, int *mx, int *my, int ref,
254  const BlockNode *left, const BlockNode *top, const BlockNode *tr){
255  if(s->ref_frames == 1){
256  *mx = mid_pred(left->mx, top->mx, tr->mx);
257  *my = mid_pred(left->my, top->my, tr->my);
258  }else{
259  const int *scale = scale_mv_ref[ref];
260  *mx = mid_pred((left->mx * scale[left->ref] + 128) >>8,
261  (top ->mx * scale[top ->ref] + 128) >>8,
262  (tr ->mx * scale[tr ->ref] + 128) >>8);
263  *my = mid_pred((left->my * scale[left->ref] + 128) >>8,
264  (top ->my * scale[top ->ref] + 128) >>8,
265  (tr ->my * scale[tr ->ref] + 128) >>8);
266  }
267 }
268 
270  if((a->type&BLOCK_INTRA) && (b->type&BLOCK_INTRA)){
271  return !((a->color[0] - b->color[0]) | (a->color[1] - b->color[1]) | (a->color[2] - b->color[2]));
272  }else{
273  return !((a->mx - b->mx) | (a->my - b->my) | (a->ref - b->ref) | ((a->type ^ b->type)&BLOCK_INTRA));
274  }
275 }
276 
277 //FIXME name cleanup (b_w, block_w, b_width stuff)
278 //XXX should we really inline it?
279 static av_always_inline void add_yblock(SnowContext *s, int sliced, slice_buffer *sb, IDWTELEM *dst, uint8_t *dst8, const uint8_t *obmc, int src_x, int src_y, int b_w, int b_h, int w, int h, int dst_stride, int src_stride, int obmc_stride, int b_x, int b_y, int add, int offset_dst, int plane_index){
280  const int b_width = s->b_width << s->block_max_depth;
281  const int b_height= s->b_height << s->block_max_depth;
282  const int b_stride= b_width;
283  BlockNode *lt= &s->block[b_x + b_y*b_stride];
284  BlockNode *rt= lt+1;
285  BlockNode *lb= lt+b_stride;
286  BlockNode *rb= lb+1;
287  uint8_t *block[4];
288  int tmp_step= src_stride >= 7*MB_SIZE ? MB_SIZE : MB_SIZE*src_stride;
289  uint8_t *tmp = s->scratchbuf;
290  uint8_t *ptmp;
291  int x,y;
292 
293  if(b_x<0){
294  lt= rt;
295  lb= rb;
296  }else if(b_x + 1 >= b_width){
297  rt= lt;
298  rb= lb;
299  }
300  if(b_y<0){
301  lt= lb;
302  rt= rb;
303  }else if(b_y + 1 >= b_height){
304  lb= lt;
305  rb= rt;
306  }
307 
308  if(src_x<0){ //FIXME merge with prev & always round internal width up to *16
309  obmc -= src_x;
310  b_w += src_x;
311  if(!sliced && !offset_dst)
312  dst -= src_x;
313  src_x=0;
314  }else if(src_x + b_w > w){
315  b_w = w - src_x;
316  }
317  if(src_y<0){
318  obmc -= src_y*obmc_stride;
319  b_h += src_y;
320  if(!sliced && !offset_dst)
321  dst -= src_y*dst_stride;
322  src_y=0;
323  }else if(src_y + b_h> h){
324  b_h = h - src_y;
325  }
326 
327  if(b_w<=0 || b_h<=0) return;
328 
329  assert(src_stride > 2*MB_SIZE + 5);
330 
331  if(!sliced && offset_dst)
332  dst += src_x + src_y*dst_stride;
333  dst8+= src_x + src_y*src_stride;
334 // src += src_x + src_y*src_stride;
335 
336  ptmp= tmp + 3*tmp_step;
337  block[0]= ptmp;
338  ptmp+=tmp_step;
339  ff_snow_pred_block(s, block[0], tmp, src_stride, src_x, src_y, b_w, b_h, lt, plane_index, w, h);
340 
341  if(same_block(lt, rt)){
342  block[1]= block[0];
343  }else{
344  block[1]= ptmp;
345  ptmp+=tmp_step;
346  ff_snow_pred_block(s, block[1], tmp, src_stride, src_x, src_y, b_w, b_h, rt, plane_index, w, h);
347  }
348 
349  if(same_block(lt, lb)){
350  block[2]= block[0];
351  }else if(same_block(rt, lb)){
352  block[2]= block[1];
353  }else{
354  block[2]= ptmp;
355  ptmp+=tmp_step;
356  ff_snow_pred_block(s, block[2], tmp, src_stride, src_x, src_y, b_w, b_h, lb, plane_index, w, h);
357  }
358 
359  if(same_block(lt, rb) ){
360  block[3]= block[0];
361  }else if(same_block(rt, rb)){
362  block[3]= block[1];
363  }else if(same_block(lb, rb)){
364  block[3]= block[2];
365  }else{
366  block[3]= ptmp;
367  ff_snow_pred_block(s, block[3], tmp, src_stride, src_x, src_y, b_w, b_h, rb, plane_index, w, h);
368  }
369  if(sliced){
370  s->dwt.inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
371  }else{
372  for(y=0; y<b_h; y++){
373  //FIXME ugly misuse of obmc_stride
374  const uint8_t *obmc1= obmc + y*obmc_stride;
375  const uint8_t *obmc2= obmc1+ (obmc_stride>>1);
376  const uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1);
377  const uint8_t *obmc4= obmc3+ (obmc_stride>>1);
378  for(x=0; x<b_w; x++){
379  int v= obmc1[x] * block[3][x + y*src_stride]
380  +obmc2[x] * block[2][x + y*src_stride]
381  +obmc3[x] * block[1][x + y*src_stride]
382  +obmc4[x] * block[0][x + y*src_stride];
383 
384  v <<= 8 - LOG2_OBMC_MAX;
385  if(FRAC_BITS != 8){
386  v >>= 8 - FRAC_BITS;
387  }
388  if(add){
389  v += dst[x + y*dst_stride];
390  v = (v + (1<<(FRAC_BITS-1))) >> FRAC_BITS;
391  if(v&(~255)) v= ~(v>>31);
392  dst8[x + y*src_stride] = v;
393  }else{
394  dst[x + y*dst_stride] -= v;
395  }
396  }
397  }
398  }
399 }
400 
401 static av_always_inline void predict_slice(SnowContext *s, IDWTELEM *buf, int plane_index, int add, int mb_y){
402  Plane *p= &s->plane[plane_index];
403  const int mb_w= s->b_width << s->block_max_depth;
404  const int mb_h= s->b_height << s->block_max_depth;
405  int x, y, mb_x;
406  int block_size = MB_SIZE >> s->block_max_depth;
407  int block_w = plane_index ? block_size/2 : block_size;
408  const uint8_t *obmc = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth];
409  const int obmc_stride= plane_index ? block_size : 2*block_size;
410  int ref_stride= s->current_picture.linesize[plane_index];
411  uint8_t *dst8= s->current_picture.data[plane_index];
412  int w= p->width;
413  int h= p->height;
414 
415  if(s->keyframe || (s->avctx->debug&512)){
416  if(mb_y==mb_h)
417  return;
418 
419  if(add){
420  for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++){
421  for(x=0; x<w; x++){
422  int v= buf[x + y*w] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1));
423  v >>= FRAC_BITS;
424  if(v&(~255)) v= ~(v>>31);
425  dst8[x + y*ref_stride]= v;
426  }
427  }
428  }else{
429  for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++){
430  for(x=0; x<w; x++){
431  buf[x + y*w]-= 128<<FRAC_BITS;
432  }
433  }
434  }
435 
436  return;
437  }
438 
439  for(mb_x=0; mb_x<=mb_w; mb_x++){
440  add_yblock(s, 0, NULL, buf, dst8, obmc,
441  block_w*mb_x - block_w/2,
442  block_w*mb_y - block_w/2,
443  block_w, block_w,
444  w, h,
445  w, ref_stride, obmc_stride,
446  mb_x - 1, mb_y - 1,
447  add, 1, plane_index);
448  }
449 }
450 
451 static av_always_inline void predict_plane(SnowContext *s, IDWTELEM *buf, int plane_index, int add){
452  const int mb_h= s->b_height << s->block_max_depth;
453  int mb_y;
454  for(mb_y=0; mb_y<=mb_h; mb_y++)
455  predict_slice(s, buf, plane_index, add, mb_y);
456 }
457 
458 static inline void set_blocks(SnowContext *s, int level, int x, int y, int l, int cb, int cr, int mx, int my, int ref, int type){
459  const int w= s->b_width << s->block_max_depth;
460  const int rem_depth= s->block_max_depth - level;
461  const int index= (x + y*w) << rem_depth;
462  const int block_w= 1<<rem_depth;
464  int i,j;
465 
466  block.color[0]= l;
467  block.color[1]= cb;
468  block.color[2]= cr;
469  block.mx= mx;
470  block.my= my;
471  block.ref= ref;
472  block.type= type;
473  block.level= level;
474 
475  for(j=0; j<block_w; j++){
476  for(i=0; i<block_w; i++){
477  s->block[index + i + j*w]= block;
478  }
479  }
480 }
481 
482 static inline void init_ref(MotionEstContext *c, uint8_t *src[3], uint8_t *ref[3], uint8_t *ref2[3], int x, int y, int ref_index){
483  const int offset[3]= {
484  y*c-> stride + x,
485  ((y*c->uvstride + x)>>1),
486  ((y*c->uvstride + x)>>1),
487  };
488  int i;
489  for(i=0; i<3; i++){
490  c->src[0][i]= src [i];
491  c->ref[0][i]= ref [i] + offset[i];
492  }
493  assert(!ref_index);
494 }
495 
496 
497 /* bitstream functions */
498 
499 extern const int8_t quant3bA[256];
500 
501 #define QEXPSHIFT (7-FRAC_BITS+8) //FIXME try to change this to 0
502 
503 static inline void put_symbol(RangeCoder *c, uint8_t *state, int v, int is_signed){
504  int i;
505 
506  if(v){
507  const int a= FFABS(v);
508  const int e= av_log2(a);
509  const int el= FFMIN(e, 10);
510  put_rac(c, state+0, 0);
511 
512  for(i=0; i<el; i++){
513  put_rac(c, state+1+i, 1); //1..10
514  }
515  for(; i<e; i++){
516  put_rac(c, state+1+9, 1); //1..10
517  }
518  put_rac(c, state+1+FFMIN(i,9), 0);
519 
520  for(i=e-1; i>=el; i--){
521  put_rac(c, state+22+9, (a>>i)&1); //22..31
522  }
523  for(; i>=0; i--){
524  put_rac(c, state+22+i, (a>>i)&1); //22..31
525  }
526 
527  if(is_signed)
528  put_rac(c, state+11 + el, v < 0); //11..21
529  }else{
530  put_rac(c, state+0, 1);
531  }
532 }
533 
534 static inline int get_symbol(RangeCoder *c, uint8_t *state, int is_signed){
535  if(get_rac(c, state+0))
536  return 0;
537  else{
538  int i, e, a;
539  e= 0;
540  while(get_rac(c, state+1 + FFMIN(e,9))){ //1..10
541  e++;
542  }
543 
544  a= 1;
545  for(i=e-1; i>=0; i--){
546  a += a + get_rac(c, state+22 + FFMIN(i,9)); //22..31
547  }
548 
549  e= -(is_signed && get_rac(c, state+11 + FFMIN(e,10))); //11..21
550  return (a^e)-e;
551  }
552 }
553 
554 static inline void put_symbol2(RangeCoder *c, uint8_t *state, int v, int log2){
555  int i;
556  int r= log2>=0 ? 1<<log2 : 1;
557 
558  assert(v>=0);
559  assert(log2>=-4);
560 
561  while(v >= r){
562  put_rac(c, state+4+log2, 1);
563  v -= r;
564  log2++;
565  if(log2>0) r+=r;
566  }
567  put_rac(c, state+4+log2, 0);
568 
569  for(i=log2-1; i>=0; i--){
570  put_rac(c, state+31-i, (v>>i)&1);
571  }
572 }
573 
574 static inline int get_symbol2(RangeCoder *c, uint8_t *state, int log2){
575  int i;
576  int r= log2>=0 ? 1<<log2 : 1;
577  int v=0;
578 
579  assert(log2>=-4);
580 
581  while(get_rac(c, state+4+log2)){
582  v+= r;
583  log2++;
584  if(log2>0) r+=r;
585  }
586 
587  for(i=log2-1; i>=0; i--){
588  v+= get_rac(c, state+31-i)<<i;
589  }
590 
591  return v;
592 }
593 
594 static inline void unpack_coeffs(SnowContext *s, SubBand *b, SubBand * parent, int orientation){
595  const int w= b->width;
596  const int h= b->height;
597  int x,y;
598 
599  int run, runs;
600  x_and_coeff *xc= b->x_coeff;
601  x_and_coeff *prev_xc= NULL;
602  x_and_coeff *prev2_xc= xc;
603  x_and_coeff *parent_xc= parent ? parent->x_coeff : NULL;
604  x_and_coeff *prev_parent_xc= parent_xc;
605 
606  runs= get_symbol2(&s->c, b->state[30], 0);
607  if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3);
608  else run= INT_MAX;
609 
610  for(y=0; y<h; y++){
611  int v=0;
612  int lt=0, t=0, rt=0;
613 
614  if(y && prev_xc->x == 0){
615  rt= prev_xc->coeff;
616  }
617  for(x=0; x<w; x++){
618  int p=0;
619  const int l= v;
620 
621  lt= t; t= rt;
622 
623  if(y){
624  if(prev_xc->x <= x)
625  prev_xc++;
626  if(prev_xc->x == x + 1)
627  rt= prev_xc->coeff;
628  else
629  rt=0;
630  }
631  if(parent_xc){
632  if(x>>1 > parent_xc->x){
633  parent_xc++;
634  }
635  if(x>>1 == parent_xc->x){
636  p= parent_xc->coeff;
637  }
638  }
639  if(/*ll|*/l|lt|t|rt|p){
640  int context= av_log2(/*FFABS(ll) + */3*(l>>1) + (lt>>1) + (t&~1) + (rt>>1) + (p>>1));
641 
642  v=get_rac(&s->c, &b->state[0][context]);
643  if(v){
644  v= 2*(get_symbol2(&s->c, b->state[context + 2], context-4) + 1);
645  v+=get_rac(&s->c, &b->state[0][16 + 1 + 3 + quant3bA[l&0xFF] + 3*quant3bA[t&0xFF]]);
646 
647  xc->x=x;
648  (xc++)->coeff= v;
649  }
650  }else{
651  if(!run){
652  if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3);
653  else run= INT_MAX;
654  v= 2*(get_symbol2(&s->c, b->state[0 + 2], 0-4) + 1);
655  v+=get_rac(&s->c, &b->state[0][16 + 1 + 3]);
656 
657  xc->x=x;
658  (xc++)->coeff= v;
659  }else{
660  int max_run;
661  run--;
662  v=0;
663 
664  if(y) max_run= FFMIN(run, prev_xc->x - x - 2);
665  else max_run= FFMIN(run, w-x-1);
666  if(parent_xc)
667  max_run= FFMIN(max_run, 2*parent_xc->x - x - 1);
668  x+= max_run;
669  run-= max_run;
670  }
671  }
672  }
673  (xc++)->x= w+1; //end marker
674  prev_xc= prev2_xc;
675  prev2_xc= xc;
676 
677  if(parent_xc){
678  if(y&1){
679  while(parent_xc->x != parent->width+1)
680  parent_xc++;
681  parent_xc++;
682  prev_parent_xc= parent_xc;
683  }else{
684  parent_xc= prev_parent_xc;
685  }
686  }
687  }
688 
689  (xc++)->x= w+1; //end marker
690 }
691 
692 #endif /* AVCODEC_SNOW_H */