snowenc.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 "snow.h"
28 
29 #include "rangecoder.h"
30 #include "mathops.h"
31 
32 #include "mpegvideo.h"
33 #include "h263.h"
34 
35 #undef NDEBUG
36 #include <assert.h>
37 
38 #define QUANTIZE2 0
39 
40 #if QUANTIZE2==1
41 #define Q2_STEP 8
42 
43 static void find_sse(SnowContext *s, Plane *p, int *score, int score_stride, IDWTELEM *r0, IDWTELEM *r1, int level, int orientation){
44  SubBand *b= &p->band[level][orientation];
45  int x, y;
46  int xo=0;
47  int yo=0;
48  int step= 1 << (s->spatial_decomposition_count - level);
49 
50  if(orientation&1)
51  xo= step>>1;
52  if(orientation&2)
53  yo= step>>1;
54 
55  //FIXME bias for nonzero ?
56  //FIXME optimize
57  memset(score, 0, sizeof(*score)*score_stride*((p->height + Q2_STEP-1)/Q2_STEP));
58  for(y=0; y<p->height; y++){
59  for(x=0; x<p->width; x++){
60  int sx= (x-xo + step/2) / step / Q2_STEP;
61  int sy= (y-yo + step/2) / step / Q2_STEP;
62  int v= r0[x + y*p->width] - r1[x + y*p->width];
63  assert(sx>=0 && sy>=0 && sx < score_stride);
64  v= ((v+8)>>4)<<4;
65  score[sx + sy*score_stride] += v*v;
66  assert(score[sx + sy*score_stride] >= 0);
67  }
68  }
69 }
70 
71 static void dequantize_all(SnowContext *s, Plane *p, IDWTELEM *buffer, int width, int height){
72  int level, orientation;
73 
74  for(level=0; level<s->spatial_decomposition_count; level++){
75  for(orientation=level ? 1 : 0; orientation<4; orientation++){
76  SubBand *b= &p->band[level][orientation];
77  IDWTELEM *dst= buffer + (b->ibuf - s->spatial_idwt_buffer);
78 
79  dequantize(s, b, dst, b->stride);
80  }
81  }
82 }
83 
84 static void dwt_quantize(SnowContext *s, Plane *p, DWTELEM *buffer, int width, int height, int stride, int type){
85  int level, orientation, ys, xs, x, y, pass;
86  IDWTELEM best_dequant[height * stride];
87  IDWTELEM idwt2_buffer[height * stride];
88  const int score_stride= (width + 10)/Q2_STEP;
89  int best_score[(width + 10)/Q2_STEP * (height + 10)/Q2_STEP]; //FIXME size
90  int score[(width + 10)/Q2_STEP * (height + 10)/Q2_STEP]; //FIXME size
91  int threshold= (s->m.lambda * s->m.lambda) >> 6;
92 
93  //FIXME pass the copy cleanly ?
94 
95 // memcpy(dwt_buffer, buffer, height * stride * sizeof(DWTELEM));
96  ff_spatial_dwt(buffer, width, height, stride, type, s->spatial_decomposition_count);
97 
98  for(level=0; level<s->spatial_decomposition_count; level++){
99  for(orientation=level ? 1 : 0; orientation<4; orientation++){
100  SubBand *b= &p->band[level][orientation];
101  IDWTELEM *dst= best_dequant + (b->ibuf - s->spatial_idwt_buffer);
102  DWTELEM *src= buffer + (b-> buf - s->spatial_dwt_buffer);
103  assert(src == b->buf); // code does not depend on this but it is true currently
104 
105  quantize(s, b, dst, src, b->stride, s->qbias);
106  }
107  }
108  for(pass=0; pass<1; pass++){
109  if(s->qbias == 0) //keyframe
110  continue;
111  for(level=0; level<s->spatial_decomposition_count; level++){
112  for(orientation=level ? 1 : 0; orientation<4; orientation++){
113  SubBand *b= &p->band[level][orientation];
114  IDWTELEM *dst= idwt2_buffer + (b->ibuf - s->spatial_idwt_buffer);
115  IDWTELEM *best_dst= best_dequant + (b->ibuf - s->spatial_idwt_buffer);
116 
117  for(ys= 0; ys<Q2_STEP; ys++){
118  for(xs= 0; xs<Q2_STEP; xs++){
119  memcpy(idwt2_buffer, best_dequant, height * stride * sizeof(IDWTELEM));
120  dequantize_all(s, p, idwt2_buffer, width, height);
121  ff_spatial_idwt(idwt2_buffer, width, height, stride, type, s->spatial_decomposition_count);
122  find_sse(s, p, best_score, score_stride, idwt2_buffer, s->spatial_idwt_buffer, level, orientation);
123  memcpy(idwt2_buffer, best_dequant, height * stride * sizeof(IDWTELEM));
124  for(y=ys; y<b->height; y+= Q2_STEP){
125  for(x=xs; x<b->width; x+= Q2_STEP){
126  if(dst[x + y*b->stride]<0) dst[x + y*b->stride]++;
127  if(dst[x + y*b->stride]>0) dst[x + y*b->stride]--;
128  //FIXME try more than just --
129  }
130  }
131  dequantize_all(s, p, idwt2_buffer, width, height);
132  ff_spatial_idwt(idwt2_buffer, width, height, stride, type, s->spatial_decomposition_count);
133  find_sse(s, p, score, score_stride, idwt2_buffer, s->spatial_idwt_buffer, level, orientation);
134  for(y=ys; y<b->height; y+= Q2_STEP){
135  for(x=xs; x<b->width; x+= Q2_STEP){
136  int score_idx= x/Q2_STEP + (y/Q2_STEP)*score_stride;
137  if(score[score_idx] <= best_score[score_idx] + threshold){
138  best_score[score_idx]= score[score_idx];
139  if(best_dst[x + y*b->stride]<0) best_dst[x + y*b->stride]++;
140  if(best_dst[x + y*b->stride]>0) best_dst[x + y*b->stride]--;
141  //FIXME copy instead
142  }
143  }
144  }
145  }
146  }
147  }
148  }
149  }
150  memcpy(s->spatial_idwt_buffer, best_dequant, height * stride * sizeof(IDWTELEM)); //FIXME work with that directly instead of copy at the end
151 }
152 
153 #endif /* QUANTIZE2==1 */
154 
155 #if CONFIG_SNOW_ENCODER
156 static av_cold int encode_init(AVCodecContext *avctx)
157 {
158  SnowContext *s = avctx->priv_data;
159  int plane_index, ret;
160 
162  av_log(avctx, AV_LOG_ERROR, "This codec is under development, files encoded with it may not be decodable with future versions!!!\n"
163  "Use vstrict=-2 / -strict -2 to use it anyway.\n");
164  return -1;
165  }
166 
167  if(avctx->prediction_method == DWT_97
168  && (avctx->flags & CODEC_FLAG_QSCALE)
169  && avctx->global_quality == 0){
170  av_log(avctx, AV_LOG_ERROR, "The 9/7 wavelet is incompatible with lossless mode.\n");
171  return -1;
172  }
173 
174  s->spatial_decomposition_type= avctx->prediction_method; //FIXME add decorrelator type r transform_type
175 
176  s->mv_scale = (avctx->flags & CODEC_FLAG_QPEL) ? 2 : 4;
177  s->block_max_depth= (avctx->flags & CODEC_FLAG_4MV ) ? 1 : 0;
178 
179  for(plane_index=0; plane_index<3; plane_index++){
180  s->plane[plane_index].diag_mc= 1;
181  s->plane[plane_index].htaps= 6;
182  s->plane[plane_index].hcoeff[0]= 40;
183  s->plane[plane_index].hcoeff[1]= -10;
184  s->plane[plane_index].hcoeff[2]= 2;
185  s->plane[plane_index].fast_mc= 1;
186  }
187 
188  if ((ret = ff_snow_common_init(avctx)) < 0) {
190  return ret;
191  }
193 
194  s->version=0;
195 
196  s->m.avctx = avctx;
197  s->m.flags = avctx->flags;
198  s->m.bit_rate= avctx->bit_rate;
199 
200  s->m.me.temp =
201  s->m.me.scratchpad= av_mallocz((avctx->width+64)*2*16*2*sizeof(uint8_t));
202  s->m.me.map = av_mallocz(ME_MAP_SIZE*sizeof(uint32_t));
203  s->m.me.score_map = av_mallocz(ME_MAP_SIZE*sizeof(uint32_t));
204  s->m.obmc_scratchpad= av_mallocz(MB_SIZE*MB_SIZE*12*sizeof(uint32_t));
205  ff_h263_encode_init(&s->m); //mv_penalty
206 
207  s->max_ref_frames = FFMAX(FFMIN(avctx->refs, MAX_REF_FRAMES), 1);
208 
209  if(avctx->flags&CODEC_FLAG_PASS1){
210  if(!avctx->stats_out)
211  avctx->stats_out = av_mallocz(256);
212  }
213  if((avctx->flags&CODEC_FLAG_PASS2) || !(avctx->flags&CODEC_FLAG_QSCALE)){
214  if(ff_rate_control_init(&s->m) < 0)
215  return -1;
216  }
218 
219  avctx->coded_frame= &s->current_picture;
220  switch(avctx->pix_fmt){
221 // case PIX_FMT_YUV444P:
222 // case PIX_FMT_YUV422P:
223  case PIX_FMT_YUV420P:
224  case PIX_FMT_GRAY8:
225 // case PIX_FMT_YUV411P:
226 // case PIX_FMT_YUV410P:
227  s->colorspace_type= 0;
228  break;
229 /* case PIX_FMT_RGB32:
230  s->colorspace= 1;
231  break;*/
232  default:
233  av_log(avctx, AV_LOG_ERROR, "pixel format not supported\n");
234  return -1;
235  }
236 // avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_h_shift, &s->chroma_v_shift);
237  s->chroma_h_shift= 1;
238  s->chroma_v_shift= 1;
239 
240  ff_set_cmp(&s->dsp, s->dsp.me_cmp, s->avctx->me_cmp);
242 
243  s->avctx->get_buffer(s->avctx, &s->input_picture);
244 
245  if(s->avctx->me_method == ME_ITER){
246  int i;
247  int size= s->b_width * s->b_height << 2*s->block_max_depth;
248  for(i=0; i<s->max_ref_frames; i++){
249  s->ref_mvs[i]= av_mallocz(size*sizeof(int16_t[2]));
250  s->ref_scores[i]= av_mallocz(size*sizeof(uint32_t));
251  }
252  }
253 
254  return 0;
255 }
256 
257 //near copy & paste from dsputil, FIXME
258 static int pix_sum(uint8_t * pix, int line_size, int w)
259 {
260  int s, i, j;
261 
262  s = 0;
263  for (i = 0; i < w; i++) {
264  for (j = 0; j < w; j++) {
265  s += pix[0];
266  pix ++;
267  }
268  pix += line_size - w;
269  }
270  return s;
271 }
272 
273 //near copy & paste from dsputil, FIXME
274 static int pix_norm1(uint8_t * pix, int line_size, int w)
275 {
276  int s, i, j;
277  uint32_t *sq = ff_squareTbl + 256;
278 
279  s = 0;
280  for (i = 0; i < w; i++) {
281  for (j = 0; j < w; j ++) {
282  s += sq[pix[0]];
283  pix ++;
284  }
285  pix += line_size - w;
286  }
287  return s;
288 }
289 
290 //FIXME copy&paste
291 #define P_LEFT P[1]
292 #define P_TOP P[2]
293 #define P_TOPRIGHT P[3]
294 #define P_MEDIAN P[4]
295 #define P_MV1 P[9]
296 #define FLAG_QPEL 1 //must be 1
297 
298 static int encode_q_branch(SnowContext *s, int level, int x, int y){
299  uint8_t p_buffer[1024];
300  uint8_t i_buffer[1024];
301  uint8_t p_state[sizeof(s->block_state)];
302  uint8_t i_state[sizeof(s->block_state)];
303  RangeCoder pc, ic;
304  uint8_t *pbbak= s->c.bytestream;
305  uint8_t *pbbak_start= s->c.bytestream_start;
306  int score, score2, iscore, i_len, p_len, block_s, sum, base_bits;
307  const int w= s->b_width << s->block_max_depth;
308  const int h= s->b_height << s->block_max_depth;
309  const int rem_depth= s->block_max_depth - level;
310  const int index= (x + y*w) << rem_depth;
311  const int block_w= 1<<(LOG2_MB_SIZE - level);
312  int trx= (x+1)<<rem_depth;
313  int try= (y+1)<<rem_depth;
314  const BlockNode *left = x ? &s->block[index-1] : &null_block;
315  const BlockNode *top = y ? &s->block[index-w] : &null_block;
316  const BlockNode *right = trx<w ? &s->block[index+1] : &null_block;
317  const BlockNode *bottom= try<h ? &s->block[index+w] : &null_block;
318  const BlockNode *tl = y && x ? &s->block[index-w-1] : left;
319  const BlockNode *tr = y && trx<w && ((x&1)==0 || level==0) ? &s->block[index-w+(1<<rem_depth)] : tl; //FIXME use lt
320  int pl = left->color[0];
321  int pcb= left->color[1];
322  int pcr= left->color[2];
323  int pmx, pmy;
324  int mx=0, my=0;
325  int l,cr,cb;
326  const int stride= s->current_picture.linesize[0];
327  const int uvstride= s->current_picture.linesize[1];
328  uint8_t *current_data[3]= { s->input_picture.data[0] + (x + y* stride)*block_w,
329  s->input_picture.data[1] + (x + y*uvstride)*block_w/2,
330  s->input_picture.data[2] + (x + y*uvstride)*block_w/2};
331  int P[10][2];
332  int16_t last_mv[3][2];
333  int qpel= !!(s->avctx->flags & CODEC_FLAG_QPEL); //unused
334  const int shift= 1+qpel;
335  MotionEstContext *c= &s->m.me;
336  int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref);
337  int mx_context= av_log2(2*FFABS(left->mx - top->mx));
338  int my_context= av_log2(2*FFABS(left->my - top->my));
339  int s_context= 2*left->level + 2*top->level + tl->level + tr->level;
340  int ref, best_ref, ref_score, ref_mx, ref_my;
341 
342  assert(sizeof(s->block_state) >= 256);
343  if(s->keyframe){
344  set_blocks(s, level, x, y, pl, pcb, pcr, 0, 0, 0, BLOCK_INTRA);
345  return 0;
346  }
347 
348 // clip predictors / edge ?
349 
350  P_LEFT[0]= left->mx;
351  P_LEFT[1]= left->my;
352  P_TOP [0]= top->mx;
353  P_TOP [1]= top->my;
354  P_TOPRIGHT[0]= tr->mx;
355  P_TOPRIGHT[1]= tr->my;
356 
357  last_mv[0][0]= s->block[index].mx;
358  last_mv[0][1]= s->block[index].my;
359  last_mv[1][0]= right->mx;
360  last_mv[1][1]= right->my;
361  last_mv[2][0]= bottom->mx;
362  last_mv[2][1]= bottom->my;
363 
364  s->m.mb_stride=2;
365  s->m.mb_x=
366  s->m.mb_y= 0;
367  c->skip= 0;
368 
369  assert(c-> stride == stride);
370  assert(c->uvstride == uvstride);
371 
372  c->penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp);
373  c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp);
374  c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp);
375  c->current_mv_penalty= c->mv_penalty[s->m.f_code=1] + MAX_MV;
376 
377  c->xmin = - x*block_w - 16+3;
378  c->ymin = - y*block_w - 16+3;
379  c->xmax = - (x+1)*block_w + (w<<(LOG2_MB_SIZE - s->block_max_depth)) + 16-3;
380  c->ymax = - (y+1)*block_w + (h<<(LOG2_MB_SIZE - s->block_max_depth)) + 16-3;
381 
382  if(P_LEFT[0] > (c->xmax<<shift)) P_LEFT[0] = (c->xmax<<shift);
383  if(P_LEFT[1] > (c->ymax<<shift)) P_LEFT[1] = (c->ymax<<shift);
384  if(P_TOP[0] > (c->xmax<<shift)) P_TOP[0] = (c->xmax<<shift);
385  if(P_TOP[1] > (c->ymax<<shift)) P_TOP[1] = (c->ymax<<shift);
386  if(P_TOPRIGHT[0] < (c->xmin<<shift)) P_TOPRIGHT[0]= (c->xmin<<shift);
387  if(P_TOPRIGHT[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<shift); //due to pmx no clip
388  if(P_TOPRIGHT[1] > (c->ymax<<shift)) P_TOPRIGHT[1]= (c->ymax<<shift);
389 
390  P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
391  P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
392 
393  if (!y) {
394  c->pred_x= P_LEFT[0];
395  c->pred_y= P_LEFT[1];
396  } else {
397  c->pred_x = P_MEDIAN[0];
398  c->pred_y = P_MEDIAN[1];
399  }
400 
401  score= INT_MAX;
402  best_ref= 0;
403  for(ref=0; ref<s->ref_frames; ref++){
404  init_ref(c, current_data, s->last_picture[ref].data, NULL, block_w*x, block_w*y, 0);
405 
406  ref_score= ff_epzs_motion_search(&s->m, &ref_mx, &ref_my, P, 0, /*ref_index*/ 0, last_mv,
407  (1<<16)>>shift, level-LOG2_MB_SIZE+4, block_w);
408 
409  assert(ref_mx >= c->xmin);
410  assert(ref_mx <= c->xmax);
411  assert(ref_my >= c->ymin);
412  assert(ref_my <= c->ymax);
413 
414  ref_score= c->sub_motion_search(&s->m, &ref_mx, &ref_my, ref_score, 0, 0, level-LOG2_MB_SIZE+4, block_w);
415  ref_score= ff_get_mb_score(&s->m, ref_mx, ref_my, 0, 0, level-LOG2_MB_SIZE+4, block_w, 0);
416  ref_score+= 2*av_log2(2*ref)*c->penalty_factor;
417  if(s->ref_mvs[ref]){
418  s->ref_mvs[ref][index][0]= ref_mx;
419  s->ref_mvs[ref][index][1]= ref_my;
420  s->ref_scores[ref][index]= ref_score;
421  }
422  if(score > ref_score){
423  score= ref_score;
424  best_ref= ref;
425  mx= ref_mx;
426  my= ref_my;
427  }
428  }
429  //FIXME if mb_cmp != SSE then intra cannot be compared currently and mb_penalty vs. lambda2
430 
431  // subpel search
432  base_bits= get_rac_count(&s->c) - 8*(s->c.bytestream - s->c.bytestream_start);
433  pc= s->c;
434  pc.bytestream_start=
435  pc.bytestream= p_buffer; //FIXME end/start? and at the other stoo
436  memcpy(p_state, s->block_state, sizeof(s->block_state));
437 
438  if(level!=s->block_max_depth)
439  put_rac(&pc, &p_state[4 + s_context], 1);
440  put_rac(&pc, &p_state[1 + left->type + top->type], 0);
441  if(s->ref_frames > 1)
442  put_symbol(&pc, &p_state[128 + 1024 + 32*ref_context], best_ref, 0);
443  pred_mv(s, &pmx, &pmy, best_ref, left, top, tr);
444  put_symbol(&pc, &p_state[128 + 32*(mx_context + 16*!!best_ref)], mx - pmx, 1);
445  put_symbol(&pc, &p_state[128 + 32*(my_context + 16*!!best_ref)], my - pmy, 1);
446  p_len= pc.bytestream - pc.bytestream_start;
447  score += (s->lambda2*(get_rac_count(&pc)-base_bits))>>FF_LAMBDA_SHIFT;
448 
449  block_s= block_w*block_w;
450  sum = pix_sum(current_data[0], stride, block_w);
451  l= (sum + block_s/2)/block_s;
452  iscore = pix_norm1(current_data[0], stride, block_w) - 2*l*sum + l*l*block_s;
453 
454  block_s= block_w*block_w>>2;
455  sum = pix_sum(current_data[1], uvstride, block_w>>1);
456  cb= (sum + block_s/2)/block_s;
457 // iscore += pix_norm1(&current_mb[1][0], uvstride, block_w>>1) - 2*cb*sum + cb*cb*block_s;
458  sum = pix_sum(current_data[2], uvstride, block_w>>1);
459  cr= (sum + block_s/2)/block_s;
460 // iscore += pix_norm1(&current_mb[2][0], uvstride, block_w>>1) - 2*cr*sum + cr*cr*block_s;
461 
462  ic= s->c;
463  ic.bytestream_start=
464  ic.bytestream= i_buffer; //FIXME end/start? and at the other stoo
465  memcpy(i_state, s->block_state, sizeof(s->block_state));
466  if(level!=s->block_max_depth)
467  put_rac(&ic, &i_state[4 + s_context], 1);
468  put_rac(&ic, &i_state[1 + left->type + top->type], 1);
469  put_symbol(&ic, &i_state[32], l-pl , 1);
470  put_symbol(&ic, &i_state[64], cb-pcb, 1);
471  put_symbol(&ic, &i_state[96], cr-pcr, 1);
472  i_len= ic.bytestream - ic.bytestream_start;
473  iscore += (s->lambda2*(get_rac_count(&ic)-base_bits))>>FF_LAMBDA_SHIFT;
474 
475 // assert(score==256*256*256*64-1);
476  assert(iscore < 255*255*256 + s->lambda2*10);
477  assert(iscore >= 0);
478  assert(l>=0 && l<=255);
479  assert(pl>=0 && pl<=255);
480 
481  if(level==0){
482  int varc= iscore >> 8;
483  int vard= score >> 8;
484  if (vard <= 64 || vard < varc)
485  c->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);
486  else
487  c->scene_change_score+= s->m.qscale;
488  }
489 
490  if(level!=s->block_max_depth){
491  put_rac(&s->c, &s->block_state[4 + s_context], 0);
492  score2 = encode_q_branch(s, level+1, 2*x+0, 2*y+0);
493  score2+= encode_q_branch(s, level+1, 2*x+1, 2*y+0);
494  score2+= encode_q_branch(s, level+1, 2*x+0, 2*y+1);
495  score2+= encode_q_branch(s, level+1, 2*x+1, 2*y+1);
496  score2+= s->lambda2>>FF_LAMBDA_SHIFT; //FIXME exact split overhead
497 
498  if(score2 < score && score2 < iscore)
499  return score2;
500  }
501 
502  if(iscore < score){
503  pred_mv(s, &pmx, &pmy, 0, left, top, tr);
504  memcpy(pbbak, i_buffer, i_len);
505  s->c= ic;
506  s->c.bytestream_start= pbbak_start;
507  s->c.bytestream= pbbak + i_len;
508  set_blocks(s, level, x, y, l, cb, cr, pmx, pmy, 0, BLOCK_INTRA);
509  memcpy(s->block_state, i_state, sizeof(s->block_state));
510  return iscore;
511  }else{
512  memcpy(pbbak, p_buffer, p_len);
513  s->c= pc;
514  s->c.bytestream_start= pbbak_start;
515  s->c.bytestream= pbbak + p_len;
516  set_blocks(s, level, x, y, pl, pcb, pcr, mx, my, best_ref, 0);
517  memcpy(s->block_state, p_state, sizeof(s->block_state));
518  return score;
519  }
520 }
521 
522 static void encode_q_branch2(SnowContext *s, int level, int x, int y){
523  const int w= s->b_width << s->block_max_depth;
524  const int rem_depth= s->block_max_depth - level;
525  const int index= (x + y*w) << rem_depth;
526  int trx= (x+1)<<rem_depth;
527  BlockNode *b= &s->block[index];
528  const BlockNode *left = x ? &s->block[index-1] : &null_block;
529  const BlockNode *top = y ? &s->block[index-w] : &null_block;
530  const BlockNode *tl = y && x ? &s->block[index-w-1] : left;
531  const BlockNode *tr = y && trx<w && ((x&1)==0 || level==0) ? &s->block[index-w+(1<<rem_depth)] : tl; //FIXME use lt
532  int pl = left->color[0];
533  int pcb= left->color[1];
534  int pcr= left->color[2];
535  int pmx, pmy;
536  int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref);
537  int mx_context= av_log2(2*FFABS(left->mx - top->mx)) + 16*!!b->ref;
538  int my_context= av_log2(2*FFABS(left->my - top->my)) + 16*!!b->ref;
539  int s_context= 2*left->level + 2*top->level + tl->level + tr->level;
540 
541  if(s->keyframe){
542  set_blocks(s, level, x, y, pl, pcb, pcr, 0, 0, 0, BLOCK_INTRA);
543  return;
544  }
545 
546  if(level!=s->block_max_depth){
547  if(same_block(b,b+1) && same_block(b,b+w) && same_block(b,b+w+1)){
548  put_rac(&s->c, &s->block_state[4 + s_context], 1);
549  }else{
550  put_rac(&s->c, &s->block_state[4 + s_context], 0);
551  encode_q_branch2(s, level+1, 2*x+0, 2*y+0);
552  encode_q_branch2(s, level+1, 2*x+1, 2*y+0);
553  encode_q_branch2(s, level+1, 2*x+0, 2*y+1);
554  encode_q_branch2(s, level+1, 2*x+1, 2*y+1);
555  return;
556  }
557  }
558  if(b->type & BLOCK_INTRA){
559  pred_mv(s, &pmx, &pmy, 0, left, top, tr);
560  put_rac(&s->c, &s->block_state[1 + (left->type&1) + (top->type&1)], 1);
561  put_symbol(&s->c, &s->block_state[32], b->color[0]-pl , 1);
562  put_symbol(&s->c, &s->block_state[64], b->color[1]-pcb, 1);
563  put_symbol(&s->c, &s->block_state[96], b->color[2]-pcr, 1);
564  set_blocks(s, level, x, y, b->color[0], b->color[1], b->color[2], pmx, pmy, 0, BLOCK_INTRA);
565  }else{
566  pred_mv(s, &pmx, &pmy, b->ref, left, top, tr);
567  put_rac(&s->c, &s->block_state[1 + (left->type&1) + (top->type&1)], 0);
568  if(s->ref_frames > 1)
569  put_symbol(&s->c, &s->block_state[128 + 1024 + 32*ref_context], b->ref, 0);
570  put_symbol(&s->c, &s->block_state[128 + 32*mx_context], b->mx - pmx, 1);
571  put_symbol(&s->c, &s->block_state[128 + 32*my_context], b->my - pmy, 1);
572  set_blocks(s, level, x, y, pl, pcb, pcr, b->mx, b->my, b->ref, 0);
573  }
574 }
575 
576 static int get_dc(SnowContext *s, int mb_x, int mb_y, int plane_index){
577  int i, x2, y2;
578  Plane *p= &s->plane[plane_index];
579  const int block_size = MB_SIZE >> s->block_max_depth;
580  const int block_w = plane_index ? block_size/2 : block_size;
581  const uint8_t *obmc = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth];
582  const int obmc_stride= plane_index ? block_size : 2*block_size;
583  const int ref_stride= s->current_picture.linesize[plane_index];
584  uint8_t *src= s-> input_picture.data[plane_index];
585  IDWTELEM *dst= (IDWTELEM*)s->m.obmc_scratchpad + plane_index*block_size*block_size*4; //FIXME change to unsigned
586  const int b_stride = s->b_width << s->block_max_depth;
587  const int w= p->width;
588  const int h= p->height;
589  int index= mb_x + mb_y*b_stride;
590  BlockNode *b= &s->block[index];
591  BlockNode backup= *b;
592  int ab=0;
593  int aa=0;
594 
595  b->type|= BLOCK_INTRA;
596  b->color[plane_index]= 0;
597  memset(dst, 0, obmc_stride*obmc_stride*sizeof(IDWTELEM));
598 
599  for(i=0; i<4; i++){
600  int mb_x2= mb_x + (i &1) - 1;
601  int mb_y2= mb_y + (i>>1) - 1;
602  int x= block_w*mb_x2 + block_w/2;
603  int y= block_w*mb_y2 + block_w/2;
604 
605  add_yblock(s, 0, NULL, dst + ((i&1)+(i>>1)*obmc_stride)*block_w, NULL, obmc,
606  x, y, block_w, block_w, w, h, obmc_stride, ref_stride, obmc_stride, mb_x2, mb_y2, 0, 0, plane_index);
607 
608  for(y2= FFMAX(y, 0); y2<FFMIN(h, y+block_w); y2++){
609  for(x2= FFMAX(x, 0); x2<FFMIN(w, x+block_w); x2++){
610  int index= x2-(block_w*mb_x - block_w/2) + (y2-(block_w*mb_y - block_w/2))*obmc_stride;
611  int obmc_v= obmc[index];
612  int d;
613  if(y<0) obmc_v += obmc[index + block_w*obmc_stride];
614  if(x<0) obmc_v += obmc[index + block_w];
615  if(y+block_w>h) obmc_v += obmc[index - block_w*obmc_stride];
616  if(x+block_w>w) obmc_v += obmc[index - block_w];
617  //FIXME precalculate this or simplify it somehow else
618 
619  d = -dst[index] + (1<<(FRAC_BITS-1));
620  dst[index] = d;
621  ab += (src[x2 + y2*ref_stride] - (d>>FRAC_BITS)) * obmc_v;
622  aa += obmc_v * obmc_v; //FIXME precalculate this
623  }
624  }
625  }
626  *b= backup;
627 
628  return av_clip(((ab<<LOG2_OBMC_MAX) + aa/2)/aa, 0, 255); //FIXME we should not need clipping
629 }
630 
631 static inline int get_block_bits(SnowContext *s, int x, int y, int w){
632  const int b_stride = s->b_width << s->block_max_depth;
633  const int b_height = s->b_height<< s->block_max_depth;
634  int index= x + y*b_stride;
635  const BlockNode *b = &s->block[index];
636  const BlockNode *left = x ? &s->block[index-1] : &null_block;
637  const BlockNode *top = y ? &s->block[index-b_stride] : &null_block;
638  const BlockNode *tl = y && x ? &s->block[index-b_stride-1] : left;
639  const BlockNode *tr = y && x+w<b_stride ? &s->block[index-b_stride+w] : tl;
640  int dmx, dmy;
641 // int mx_context= av_log2(2*FFABS(left->mx - top->mx));
642 // int my_context= av_log2(2*FFABS(left->my - top->my));
643 
644  if(x<0 || x>=b_stride || y>=b_height)
645  return 0;
646 /*
647 1 0 0
648 01X 1-2 1
649 001XX 3-6 2-3
650 0001XXX 7-14 4-7
651 00001XXXX 15-30 8-15
652 */
653 //FIXME try accurate rate
654 //FIXME intra and inter predictors if surrounding blocks are not the same type
655  if(b->type & BLOCK_INTRA){
656  return 3+2*( av_log2(2*FFABS(left->color[0] - b->color[0]))
657  + av_log2(2*FFABS(left->color[1] - b->color[1]))
658  + av_log2(2*FFABS(left->color[2] - b->color[2])));
659  }else{
660  pred_mv(s, &dmx, &dmy, b->ref, left, top, tr);
661  dmx-= b->mx;
662  dmy-= b->my;
663  return 2*(1 + av_log2(2*FFABS(dmx)) //FIXME kill the 2* can be merged in lambda
664  + av_log2(2*FFABS(dmy))
665  + av_log2(2*b->ref));
666  }
667 }
668 
669 static int get_block_rd(SnowContext *s, int mb_x, int mb_y, int plane_index, const uint8_t *obmc_edged){
670  Plane *p= &s->plane[plane_index];
671  const int block_size = MB_SIZE >> s->block_max_depth;
672  const int block_w = plane_index ? block_size/2 : block_size;
673  const int obmc_stride= plane_index ? block_size : 2*block_size;
674  const int ref_stride= s->current_picture.linesize[plane_index];
675  uint8_t *dst= s->current_picture.data[plane_index];
676  uint8_t *src= s-> input_picture.data[plane_index];
677  IDWTELEM *pred= (IDWTELEM*)s->m.obmc_scratchpad + plane_index*block_size*block_size*4;
678  uint8_t *cur = s->scratchbuf;
679  uint8_t tmp[ref_stride*(2*MB_SIZE+HTAPS_MAX-1)];
680  const int b_stride = s->b_width << s->block_max_depth;
681  const int b_height = s->b_height<< s->block_max_depth;
682  const int w= p->width;
683  const int h= p->height;
684  int distortion;
685  int rate= 0;
686  const int penalty_factor= get_penalty_factor(s->lambda, s->lambda2, s->avctx->me_cmp);
687  int sx= block_w*mb_x - block_w/2;
688  int sy= block_w*mb_y - block_w/2;
689  int x0= FFMAX(0,-sx);
690  int y0= FFMAX(0,-sy);
691  int x1= FFMIN(block_w*2, w-sx);
692  int y1= FFMIN(block_w*2, h-sy);
693  int i,x,y;
694 
695  ff_snow_pred_block(s, cur, tmp, ref_stride, sx, sy, block_w*2, block_w*2, &s->block[mb_x + mb_y*b_stride], plane_index, w, h);
696 
697  for(y=y0; y<y1; y++){
698  const uint8_t *obmc1= obmc_edged + y*obmc_stride;
699  const IDWTELEM *pred1 = pred + y*obmc_stride;
700  uint8_t *cur1 = cur + y*ref_stride;
701  uint8_t *dst1 = dst + sx + (sy+y)*ref_stride;
702  for(x=x0; x<x1; x++){
703 #if FRAC_BITS >= LOG2_OBMC_MAX
704  int v = (cur1[x] * obmc1[x]) << (FRAC_BITS - LOG2_OBMC_MAX);
705 #else
706  int v = (cur1[x] * obmc1[x] + (1<<(LOG2_OBMC_MAX - FRAC_BITS-1))) >> (LOG2_OBMC_MAX - FRAC_BITS);
707 #endif
708  v = (v + pred1[x]) >> FRAC_BITS;
709  if(v&(~255)) v= ~(v>>31);
710  dst1[x] = v;
711  }
712  }
713 
714  /* copy the regions where obmc[] = (uint8_t)256 */
715  if(LOG2_OBMC_MAX == 8
716  && (mb_x == 0 || mb_x == b_stride-1)
717  && (mb_y == 0 || mb_y == b_height-1)){
718  if(mb_x == 0)
719  x1 = block_w;
720  else
721  x0 = block_w;
722  if(mb_y == 0)
723  y1 = block_w;
724  else
725  y0 = block_w;
726  for(y=y0; y<y1; y++)
727  memcpy(dst + sx+x0 + (sy+y)*ref_stride, cur + x0 + y*ref_stride, x1-x0);
728  }
729 
730  if(block_w==16){
731  /* FIXME rearrange dsputil to fit 32x32 cmp functions */
732  /* FIXME check alignment of the cmp wavelet vs the encoding wavelet */
733  /* FIXME cmps overlap but do not cover the wavelet's whole support.
734  * So improving the score of one block is not strictly guaranteed
735  * to improve the score of the whole frame, thus iterative motion
736  * estimation does not always converge. */
737  if(s->avctx->me_cmp == FF_CMP_W97)
738  distortion = ff_w97_32_c(&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, 32);
739  else if(s->avctx->me_cmp == FF_CMP_W53)
740  distortion = ff_w53_32_c(&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, 32);
741  else{
742  distortion = 0;
743  for(i=0; i<4; i++){
744  int off = sx+16*(i&1) + (sy+16*(i>>1))*ref_stride;
745  distortion += s->dsp.me_cmp[0](&s->m, src + off, dst + off, ref_stride, 16);
746  }
747  }
748  }else{
749  assert(block_w==8);
750  distortion = s->dsp.me_cmp[0](&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, block_w*2);
751  }
752 
753  if(plane_index==0){
754  for(i=0; i<4; i++){
755 /* ..RRr
756  * .RXx.
757  * rxx..
758  */
759  rate += get_block_bits(s, mb_x + (i&1) - (i>>1), mb_y + (i>>1), 1);
760  }
761  if(mb_x == b_stride-2)
762  rate += get_block_bits(s, mb_x + 1, mb_y + 1, 1);
763  }
764  return distortion + rate*penalty_factor;
765 }
766 
767 static int get_4block_rd(SnowContext *s, int mb_x, int mb_y, int plane_index){
768  int i, y2;
769  Plane *p= &s->plane[plane_index];
770  const int block_size = MB_SIZE >> s->block_max_depth;
771  const int block_w = plane_index ? block_size/2 : block_size;
772  const uint8_t *obmc = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth];
773  const int obmc_stride= plane_index ? block_size : 2*block_size;
774  const int ref_stride= s->current_picture.linesize[plane_index];
775  uint8_t *dst= s->current_picture.data[plane_index];
776  uint8_t *src= s-> input_picture.data[plane_index];
777  //FIXME zero_dst is const but add_yblock changes dst if add is 0 (this is never the case for dst=zero_dst
778  // const has only been removed from zero_dst to suppress a warning
779  static IDWTELEM zero_dst[4096]; //FIXME
780  const int b_stride = s->b_width << s->block_max_depth;
781  const int w= p->width;
782  const int h= p->height;
783  int distortion= 0;
784  int rate= 0;
785  const int penalty_factor= get_penalty_factor(s->lambda, s->lambda2, s->avctx->me_cmp);
786 
787  for(i=0; i<9; i++){
788  int mb_x2= mb_x + (i%3) - 1;
789  int mb_y2= mb_y + (i/3) - 1;
790  int x= block_w*mb_x2 + block_w/2;
791  int y= block_w*mb_y2 + block_w/2;
792 
793  add_yblock(s, 0, NULL, zero_dst, dst, obmc,
794  x, y, block_w, block_w, w, h, /*dst_stride*/0, ref_stride, obmc_stride, mb_x2, mb_y2, 1, 1, plane_index);
795 
796  //FIXME find a cleaner/simpler way to skip the outside stuff
797  for(y2= y; y2<0; y2++)
798  memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, block_w);
799  for(y2= h; y2<y+block_w; y2++)
800  memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, block_w);
801  if(x<0){
802  for(y2= y; y2<y+block_w; y2++)
803  memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, -x);
804  }
805  if(x+block_w > w){
806  for(y2= y; y2<y+block_w; y2++)
807  memcpy(dst + w + y2*ref_stride, src + w + y2*ref_stride, x+block_w - w);
808  }
809 
810  assert(block_w== 8 || block_w==16);
811  distortion += s->dsp.me_cmp[block_w==8](&s->m, src + x + y*ref_stride, dst + x + y*ref_stride, ref_stride, block_w);
812  }
813 
814  if(plane_index==0){
815  BlockNode *b= &s->block[mb_x+mb_y*b_stride];
816  int merged= same_block(b,b+1) && same_block(b,b+b_stride) && same_block(b,b+b_stride+1);
817 
818 /* ..RRRr
819  * .RXXx.
820  * .RXXx.
821  * rxxx.
822  */
823  if(merged)
824  rate = get_block_bits(s, mb_x, mb_y, 2);
825  for(i=merged?4:0; i<9; i++){
826  static const int dxy[9][2] = {{0,0},{1,0},{0,1},{1,1},{2,0},{2,1},{-1,2},{0,2},{1,2}};
827  rate += get_block_bits(s, mb_x + dxy[i][0], mb_y + dxy[i][1], 1);
828  }
829  }
830  return distortion + rate*penalty_factor;
831 }
832 
833 static int encode_subband_c0run(SnowContext *s, SubBand *b, IDWTELEM *src, IDWTELEM *parent, int stride, int orientation){
834  const int w= b->width;
835  const int h= b->height;
836  int x, y;
837 
838  if(1){
839  int run=0;
840  int runs[w*h];
841  int run_index=0;
842  int max_index;
843 
844  for(y=0; y<h; y++){
845  for(x=0; x<w; x++){
846  int v, p=0;
847  int /*ll=0, */l=0, lt=0, t=0, rt=0;
848  v= src[x + y*stride];
849 
850  if(y){
851  t= src[x + (y-1)*stride];
852  if(x){
853  lt= src[x - 1 + (y-1)*stride];
854  }
855  if(x + 1 < w){
856  rt= src[x + 1 + (y-1)*stride];
857  }
858  }
859  if(x){
860  l= src[x - 1 + y*stride];
861  /*if(x > 1){
862  if(orientation==1) ll= src[y + (x-2)*stride];
863  else ll= src[x - 2 + y*stride];
864  }*/
865  }
866  if(parent){
867  int px= x>>1;
868  int py= y>>1;
869  if(px<b->parent->width && py<b->parent->height)
870  p= parent[px + py*2*stride];
871  }
872  if(!(/*ll|*/l|lt|t|rt|p)){
873  if(v){
874  runs[run_index++]= run;
875  run=0;
876  }else{
877  run++;
878  }
879  }
880  }
881  }
882  max_index= run_index;
883  runs[run_index++]= run;
884  run_index=0;
885  run= runs[run_index++];
886 
887  put_symbol2(&s->c, b->state[30], max_index, 0);
888  if(run_index <= max_index)
889  put_symbol2(&s->c, b->state[1], run, 3);
890 
891  for(y=0; y<h; y++){
892  if(s->c.bytestream_end - s->c.bytestream < w*40){
893  av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
894  return -1;
895  }
896  for(x=0; x<w; x++){
897  int v, p=0;
898  int /*ll=0, */l=0, lt=0, t=0, rt=0;
899  v= src[x + y*stride];
900 
901  if(y){
902  t= src[x + (y-1)*stride];
903  if(x){
904  lt= src[x - 1 + (y-1)*stride];
905  }
906  if(x + 1 < w){
907  rt= src[x + 1 + (y-1)*stride];
908  }
909  }
910  if(x){
911  l= src[x - 1 + y*stride];
912  /*if(x > 1){
913  if(orientation==1) ll= src[y + (x-2)*stride];
914  else ll= src[x - 2 + y*stride];
915  }*/
916  }
917  if(parent){
918  int px= x>>1;
919  int py= y>>1;
920  if(px<b->parent->width && py<b->parent->height)
921  p= parent[px + py*2*stride];
922  }
923  if(/*ll|*/l|lt|t|rt|p){
924  int context= av_log2(/*FFABS(ll) + */3*FFABS(l) + FFABS(lt) + 2*FFABS(t) + FFABS(rt) + FFABS(p));
925 
926  put_rac(&s->c, &b->state[0][context], !!v);
927  }else{
928  if(!run){
929  run= runs[run_index++];
930 
931  if(run_index <= max_index)
932  put_symbol2(&s->c, b->state[1], run, 3);
933  assert(v);
934  }else{
935  run--;
936  assert(!v);
937  }
938  }
939  if(v){
940  int context= av_log2(/*FFABS(ll) + */3*FFABS(l) + FFABS(lt) + 2*FFABS(t) + FFABS(rt) + FFABS(p));
941  int l2= 2*FFABS(l) + (l<0);
942  int t2= 2*FFABS(t) + (t<0);
943 
944  put_symbol2(&s->c, b->state[context + 2], FFABS(v)-1, context-4);
945  put_rac(&s->c, &b->state[0][16 + 1 + 3 + quant3bA[l2&0xFF] + 3*quant3bA[t2&0xFF]], v<0);
946  }
947  }
948  }
949  }
950  return 0;
951 }
952 
953 static int encode_subband(SnowContext *s, SubBand *b, IDWTELEM *src, IDWTELEM *parent, int stride, int orientation){
954 // encode_subband_qtree(s, b, src, parent, stride, orientation);
955 // encode_subband_z0run(s, b, src, parent, stride, orientation);
956  return encode_subband_c0run(s, b, src, parent, stride, orientation);
957 // encode_subband_dzr(s, b, src, parent, stride, orientation);
958 }
959 
960 static av_always_inline int check_block(SnowContext *s, int mb_x, int mb_y, int p[3], int intra, const uint8_t *obmc_edged, int *best_rd){
961  const int b_stride= s->b_width << s->block_max_depth;
962  BlockNode *block= &s->block[mb_x + mb_y * b_stride];
963  BlockNode backup= *block;
964  unsigned value;
965  int rd, index;
966 
967  assert(mb_x>=0 && mb_y>=0);
968  assert(mb_x<b_stride);
969 
970  if(intra){
971  block->color[0] = p[0];
972  block->color[1] = p[1];
973  block->color[2] = p[2];
974  block->type |= BLOCK_INTRA;
975  }else{
976  index= (p[0] + 31*p[1]) & (ME_CACHE_SIZE-1);
977  value= s->me_cache_generation + (p[0]>>10) + (p[1]<<6) + (block->ref<<12);
978  if(s->me_cache[index] == value)
979  return 0;
980  s->me_cache[index]= value;
981 
982  block->mx= p[0];
983  block->my= p[1];
984  block->type &= ~BLOCK_INTRA;
985  }
986 
987  rd= get_block_rd(s, mb_x, mb_y, 0, obmc_edged);
988 
989 //FIXME chroma
990  if(rd < *best_rd){
991  *best_rd= rd;
992  return 1;
993  }else{
994  *block= backup;
995  return 0;
996  }
997 }
998 
999 /* special case for int[2] args we discard afterwards,
1000  * fixes compilation problem with gcc 2.95 */
1001 static av_always_inline int check_block_inter(SnowContext *s, int mb_x, int mb_y, int p0, int p1, const uint8_t *obmc_edged, int *best_rd){
1002  int p[2] = {p0, p1};
1003  return check_block(s, mb_x, mb_y, p, 0, obmc_edged, best_rd);
1004 }
1005 
1006 static av_always_inline int check_4block_inter(SnowContext *s, int mb_x, int mb_y, int p0, int p1, int ref, int *best_rd){
1007  const int b_stride= s->b_width << s->block_max_depth;
1008  BlockNode *block= &s->block[mb_x + mb_y * b_stride];
1009  BlockNode backup[4]= {block[0], block[1], block[b_stride], block[b_stride+1]};
1010  unsigned value;
1011  int rd, index;
1012 
1013  assert(mb_x>=0 && mb_y>=0);
1014  assert(mb_x<b_stride);
1015  assert(((mb_x|mb_y)&1) == 0);
1016 
1017  index= (p0 + 31*p1) & (ME_CACHE_SIZE-1);
1018  value= s->me_cache_generation + (p0>>10) + (p1<<6) + (block->ref<<12);
1019  if(s->me_cache[index] == value)
1020  return 0;
1021  s->me_cache[index]= value;
1022 
1023  block->mx= p0;
1024  block->my= p1;
1025  block->ref= ref;
1026  block->type &= ~BLOCK_INTRA;
1027  block[1]= block[b_stride]= block[b_stride+1]= *block;
1028 
1029  rd= get_4block_rd(s, mb_x, mb_y, 0);
1030 
1031 //FIXME chroma
1032  if(rd < *best_rd){
1033  *best_rd= rd;
1034  return 1;
1035  }else{
1036  block[0]= backup[0];
1037  block[1]= backup[1];
1038  block[b_stride]= backup[2];
1039  block[b_stride+1]= backup[3];
1040  return 0;
1041  }
1042 }
1043 
1044 static void iterative_me(SnowContext *s){
1045  int pass, mb_x, mb_y;
1046  const int b_width = s->b_width << s->block_max_depth;
1047  const int b_height= s->b_height << s->block_max_depth;
1048  const int b_stride= b_width;
1049  int color[3];
1050 
1051  {
1052  RangeCoder r = s->c;
1053  uint8_t state[sizeof(s->block_state)];
1054  memcpy(state, s->block_state, sizeof(s->block_state));
1055  for(mb_y= 0; mb_y<s->b_height; mb_y++)
1056  for(mb_x= 0; mb_x<s->b_width; mb_x++)
1057  encode_q_branch(s, 0, mb_x, mb_y);
1058  s->c = r;
1059  memcpy(s->block_state, state, sizeof(s->block_state));
1060  }
1061 
1062  for(pass=0; pass<25; pass++){
1063  int change= 0;
1064 
1065  for(mb_y= 0; mb_y<b_height; mb_y++){
1066  for(mb_x= 0; mb_x<b_width; mb_x++){
1067  int dia_change, i, j, ref;
1068  int best_rd= INT_MAX, ref_rd;
1069  BlockNode backup, ref_b;
1070  const int index= mb_x + mb_y * b_stride;
1071  BlockNode *block= &s->block[index];
1072  BlockNode *tb = mb_y ? &s->block[index-b_stride ] : NULL;
1073  BlockNode *lb = mb_x ? &s->block[index -1] : NULL;
1074  BlockNode *rb = mb_x+1<b_width ? &s->block[index +1] : NULL;
1075  BlockNode *bb = mb_y+1<b_height ? &s->block[index+b_stride ] : NULL;
1076  BlockNode *tlb= mb_x && mb_y ? &s->block[index-b_stride-1] : NULL;
1077  BlockNode *trb= mb_x+1<b_width && mb_y ? &s->block[index-b_stride+1] : NULL;
1078  BlockNode *blb= mb_x && mb_y+1<b_height ? &s->block[index+b_stride-1] : NULL;
1079  BlockNode *brb= mb_x+1<b_width && mb_y+1<b_height ? &s->block[index+b_stride+1] : NULL;
1080  const int b_w= (MB_SIZE >> s->block_max_depth);
1081  uint8_t obmc_edged[b_w*2][b_w*2];
1082 
1083  if(pass && (block->type & BLOCK_OPT))
1084  continue;
1085  block->type |= BLOCK_OPT;
1086 
1087  backup= *block;
1088 
1089  if(!s->me_cache_generation)
1090  memset(s->me_cache, 0, sizeof(s->me_cache));
1091  s->me_cache_generation += 1<<22;
1092 
1093  //FIXME precalculate
1094  {
1095  int x, y;
1096  memcpy(obmc_edged, obmc_tab[s->block_max_depth], b_w*b_w*4);
1097  if(mb_x==0)
1098  for(y=0; y<b_w*2; y++)
1099  memset(obmc_edged[y], obmc_edged[y][0] + obmc_edged[y][b_w-1], b_w);
1100  if(mb_x==b_stride-1)
1101  for(y=0; y<b_w*2; y++)
1102  memset(obmc_edged[y]+b_w, obmc_edged[y][b_w] + obmc_edged[y][b_w*2-1], b_w);
1103  if(mb_y==0){
1104  for(x=0; x<b_w*2; x++)
1105  obmc_edged[0][x] += obmc_edged[b_w-1][x];
1106  for(y=1; y<b_w; y++)
1107  memcpy(obmc_edged[y], obmc_edged[0], b_w*2);
1108  }
1109  if(mb_y==b_height-1){
1110  for(x=0; x<b_w*2; x++)
1111  obmc_edged[b_w*2-1][x] += obmc_edged[b_w][x];
1112  for(y=b_w; y<b_w*2-1; y++)
1113  memcpy(obmc_edged[y], obmc_edged[b_w*2-1], b_w*2);
1114  }
1115  }
1116 
1117  //skip stuff outside the picture
1118  if(mb_x==0 || mb_y==0 || mb_x==b_width-1 || mb_y==b_height-1){
1119  uint8_t *src= s-> input_picture.data[0];
1120  uint8_t *dst= s->current_picture.data[0];
1121  const int stride= s->current_picture.linesize[0];
1122  const int block_w= MB_SIZE >> s->block_max_depth;
1123  const int sx= block_w*mb_x - block_w/2;
1124  const int sy= block_w*mb_y - block_w/2;
1125  const int w= s->plane[0].width;
1126  const int h= s->plane[0].height;
1127  int y;
1128 
1129  for(y=sy; y<0; y++)
1130  memcpy(dst + sx + y*stride, src + sx + y*stride, block_w*2);
1131  for(y=h; y<sy+block_w*2; y++)
1132  memcpy(dst + sx + y*stride, src + sx + y*stride, block_w*2);
1133  if(sx<0){
1134  for(y=sy; y<sy+block_w*2; y++)
1135  memcpy(dst + sx + y*stride, src + sx + y*stride, -sx);
1136  }
1137  if(sx+block_w*2 > w){
1138  for(y=sy; y<sy+block_w*2; y++)
1139  memcpy(dst + w + y*stride, src + w + y*stride, sx+block_w*2 - w);
1140  }
1141  }
1142 
1143  // intra(black) = neighbors' contribution to the current block
1144  for(i=0; i<3; i++)
1145  color[i]= get_dc(s, mb_x, mb_y, i);
1146 
1147  // get previous score (cannot be cached due to OBMC)
1148  if(pass > 0 && (block->type&BLOCK_INTRA)){
1149  int color0[3]= {block->color[0], block->color[1], block->color[2]};
1150  check_block(s, mb_x, mb_y, color0, 1, *obmc_edged, &best_rd);
1151  }else
1152  check_block_inter(s, mb_x, mb_y, block->mx, block->my, *obmc_edged, &best_rd);
1153 
1154  ref_b= *block;
1155  ref_rd= best_rd;
1156  for(ref=0; ref < s->ref_frames; ref++){
1157  int16_t (*mvr)[2]= &s->ref_mvs[ref][index];
1158  if(s->ref_scores[ref][index] > s->ref_scores[ref_b.ref][index]*3/2) //FIXME tune threshold
1159  continue;
1160  block->ref= ref;
1161  best_rd= INT_MAX;
1162 
1163  check_block_inter(s, mb_x, mb_y, mvr[0][0], mvr[0][1], *obmc_edged, &best_rd);
1164  check_block_inter(s, mb_x, mb_y, 0, 0, *obmc_edged, &best_rd);
1165  if(tb)
1166  check_block_inter(s, mb_x, mb_y, mvr[-b_stride][0], mvr[-b_stride][1], *obmc_edged, &best_rd);
1167  if(lb)
1168  check_block_inter(s, mb_x, mb_y, mvr[-1][0], mvr[-1][1], *obmc_edged, &best_rd);
1169  if(rb)
1170  check_block_inter(s, mb_x, mb_y, mvr[1][0], mvr[1][1], *obmc_edged, &best_rd);
1171  if(bb)
1172  check_block_inter(s, mb_x, mb_y, mvr[b_stride][0], mvr[b_stride][1], *obmc_edged, &best_rd);
1173 
1174  /* fullpel ME */
1175  //FIXME avoid subpel interpolation / round to nearest integer
1176  do{
1177  dia_change=0;
1178  for(i=0; i<FFMAX(s->avctx->dia_size, 1); i++){
1179  for(j=0; j<i; j++){
1180  dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+4*(i-j), block->my+(4*j), *obmc_edged, &best_rd);
1181  dia_change |= check_block_inter(s, mb_x, mb_y, block->mx-4*(i-j), block->my-(4*j), *obmc_edged, &best_rd);
1182  dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+4*(i-j), block->my-(4*j), *obmc_edged, &best_rd);
1183  dia_change |= check_block_inter(s, mb_x, mb_y, block->mx-4*(i-j), block->my+(4*j), *obmc_edged, &best_rd);
1184  }
1185  }
1186  }while(dia_change);
1187  /* subpel ME */
1188  do{
1189  static const int square[8][2]= {{+1, 0},{-1, 0},{ 0,+1},{ 0,-1},{+1,+1},{-1,-1},{+1,-1},{-1,+1},};
1190  dia_change=0;
1191  for(i=0; i<8; i++)
1192  dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+square[i][0], block->my+square[i][1], *obmc_edged, &best_rd);
1193  }while(dia_change);
1194  //FIXME or try the standard 2 pass qpel or similar
1195 
1196  mvr[0][0]= block->mx;
1197  mvr[0][1]= block->my;
1198  if(ref_rd > best_rd){
1199  ref_rd= best_rd;
1200  ref_b= *block;
1201  }
1202  }
1203  best_rd= ref_rd;
1204  *block= ref_b;
1205  check_block(s, mb_x, mb_y, color, 1, *obmc_edged, &best_rd);
1206  //FIXME RD style color selection
1207  if(!same_block(block, &backup)){
1208  if(tb ) tb ->type &= ~BLOCK_OPT;
1209  if(lb ) lb ->type &= ~BLOCK_OPT;
1210  if(rb ) rb ->type &= ~BLOCK_OPT;
1211  if(bb ) bb ->type &= ~BLOCK_OPT;
1212  if(tlb) tlb->type &= ~BLOCK_OPT;
1213  if(trb) trb->type &= ~BLOCK_OPT;
1214  if(blb) blb->type &= ~BLOCK_OPT;
1215  if(brb) brb->type &= ~BLOCK_OPT;
1216  change ++;
1217  }
1218  }
1219  }
1220  av_log(s->avctx, AV_LOG_ERROR, "pass:%d changed:%d\n", pass, change);
1221  if(!change)
1222  break;
1223  }
1224 
1225  if(s->block_max_depth == 1){
1226  int change= 0;
1227  for(mb_y= 0; mb_y<b_height; mb_y+=2){
1228  for(mb_x= 0; mb_x<b_width; mb_x+=2){
1229  int i;
1230  int best_rd, init_rd;
1231  const int index= mb_x + mb_y * b_stride;
1232  BlockNode *b[4];
1233 
1234  b[0]= &s->block[index];
1235  b[1]= b[0]+1;
1236  b[2]= b[0]+b_stride;
1237  b[3]= b[2]+1;
1238  if(same_block(b[0], b[1]) &&
1239  same_block(b[0], b[2]) &&
1240  same_block(b[0], b[3]))
1241  continue;
1242 
1243  if(!s->me_cache_generation)
1244  memset(s->me_cache, 0, sizeof(s->me_cache));
1245  s->me_cache_generation += 1<<22;
1246 
1247  init_rd= best_rd= get_4block_rd(s, mb_x, mb_y, 0);
1248 
1249  //FIXME more multiref search?
1250  check_4block_inter(s, mb_x, mb_y,
1251  (b[0]->mx + b[1]->mx + b[2]->mx + b[3]->mx + 2) >> 2,
1252  (b[0]->my + b[1]->my + b[2]->my + b[3]->my + 2) >> 2, 0, &best_rd);
1253 
1254  for(i=0; i<4; i++)
1255  if(!(b[i]->type&BLOCK_INTRA))
1256  check_4block_inter(s, mb_x, mb_y, b[i]->mx, b[i]->my, b[i]->ref, &best_rd);
1257 
1258  if(init_rd != best_rd)
1259  change++;
1260  }
1261  }
1262  av_log(s->avctx, AV_LOG_ERROR, "pass:4mv changed:%d\n", change*4);
1263  }
1264 }
1265 
1266 static void encode_blocks(SnowContext *s, int search){
1267  int x, y;
1268  int w= s->b_width;
1269  int h= s->b_height;
1270 
1271  if(s->avctx->me_method == ME_ITER && !s->keyframe && search)
1272  iterative_me(s);
1273 
1274  for(y=0; y<h; y++){
1275  if(s->c.bytestream_end - s->c.bytestream < w*MB_SIZE*MB_SIZE*3){ //FIXME nicer limit
1276  av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
1277  return;
1278  }
1279  for(x=0; x<w; x++){
1280  if(s->avctx->me_method == ME_ITER || !search)
1281  encode_q_branch2(s, 0, x, y);
1282  else
1283  encode_q_branch (s, 0, x, y);
1284  }
1285  }
1286 }
1287 
1288 static void quantize(SnowContext *s, SubBand *b, IDWTELEM *dst, DWTELEM *src, int stride, int bias){
1289  const int w= b->width;
1290  const int h= b->height;
1291  const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16);
1292  const int qmul= qexp[qlog&(QROOT-1)]<<((qlog>>QSHIFT) + ENCODER_EXTRA_BITS);
1293  int x,y, thres1, thres2;
1294 
1295  if(s->qlog == LOSSLESS_QLOG){
1296  for(y=0; y<h; y++)
1297  for(x=0; x<w; x++)
1298  dst[x + y*stride]= src[x + y*stride];
1299  return;
1300  }
1301 
1302  bias= bias ? 0 : (3*qmul)>>3;
1303  thres1= ((qmul - bias)>>QEXPSHIFT) - 1;
1304  thres2= 2*thres1;
1305 
1306  if(!bias){
1307  for(y=0; y<h; y++){
1308  for(x=0; x<w; x++){
1309  int i= src[x + y*stride];
1310 
1311  if((unsigned)(i+thres1) > thres2){
1312  if(i>=0){
1313  i<<= QEXPSHIFT;
1314  i/= qmul; //FIXME optimize
1315  dst[x + y*stride]= i;
1316  }else{
1317  i= -i;
1318  i<<= QEXPSHIFT;
1319  i/= qmul; //FIXME optimize
1320  dst[x + y*stride]= -i;
1321  }
1322  }else
1323  dst[x + y*stride]= 0;
1324  }
1325  }
1326  }else{
1327  for(y=0; y<h; y++){
1328  for(x=0; x<w; x++){
1329  int i= src[x + y*stride];
1330 
1331  if((unsigned)(i+thres1) > thres2){
1332  if(i>=0){
1333  i<<= QEXPSHIFT;
1334  i= (i + bias) / qmul; //FIXME optimize
1335  dst[x + y*stride]= i;
1336  }else{
1337  i= -i;
1338  i<<= QEXPSHIFT;
1339  i= (i + bias) / qmul; //FIXME optimize
1340  dst[x + y*stride]= -i;
1341  }
1342  }else
1343  dst[x + y*stride]= 0;
1344  }
1345  }
1346  }
1347 }
1348 
1349 static void dequantize(SnowContext *s, SubBand *b, IDWTELEM *src, int stride){
1350  const int w= b->width;
1351  const int h= b->height;
1352  const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16);
1353  const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
1354  const int qadd= (s->qbias*qmul)>>QBIAS_SHIFT;
1355  int x,y;
1356 
1357  if(s->qlog == LOSSLESS_QLOG) return;
1358 
1359  for(y=0; y<h; y++){
1360  for(x=0; x<w; x++){
1361  int i= src[x + y*stride];
1362  if(i<0){
1363  src[x + y*stride]= -((-i*qmul + qadd)>>(QEXPSHIFT)); //FIXME try different bias
1364  }else if(i>0){
1365  src[x + y*stride]= (( i*qmul + qadd)>>(QEXPSHIFT));
1366  }
1367  }
1368  }
1369 }
1370 
1371 static void decorrelate(SnowContext *s, SubBand *b, IDWTELEM *src, int stride, int inverse, int use_median){
1372  const int w= b->width;
1373  const int h= b->height;
1374  int x,y;
1375 
1376  for(y=h-1; y>=0; y--){
1377  for(x=w-1; x>=0; x--){
1378  int i= x + y*stride;
1379 
1380  if(x){
1381  if(use_median){
1382  if(y && x+1<w) src[i] -= mid_pred(src[i - 1], src[i - stride], src[i - stride + 1]);
1383  else src[i] -= src[i - 1];
1384  }else{
1385  if(y) src[i] -= mid_pred(src[i - 1], src[i - stride], src[i - 1] + src[i - stride] - src[i - 1 - stride]);
1386  else src[i] -= src[i - 1];
1387  }
1388  }else{
1389  if(y) src[i] -= src[i - stride];
1390  }
1391  }
1392  }
1393 }
1394 
1395 static void correlate(SnowContext *s, SubBand *b, IDWTELEM *src, int stride, int inverse, int use_median){
1396  const int w= b->width;
1397  const int h= b->height;
1398  int x,y;
1399 
1400  for(y=0; y<h; y++){
1401  for(x=0; x<w; x++){
1402  int i= x + y*stride;
1403 
1404  if(x){
1405  if(use_median){
1406  if(y && x+1<w) src[i] += mid_pred(src[i - 1], src[i - stride], src[i - stride + 1]);
1407  else src[i] += src[i - 1];
1408  }else{
1409  if(y) src[i] += mid_pred(src[i - 1], src[i - stride], src[i - 1] + src[i - stride] - src[i - 1 - stride]);
1410  else src[i] += src[i - 1];
1411  }
1412  }else{
1413  if(y) src[i] += src[i - stride];
1414  }
1415  }
1416  }
1417 }
1418 
1419 static void encode_qlogs(SnowContext *s){
1420  int plane_index, level, orientation;
1421 
1422  for(plane_index=0; plane_index<2; plane_index++){
1423  for(level=0; level<s->spatial_decomposition_count; level++){
1424  for(orientation=level ? 1:0; orientation<4; orientation++){
1425  if(orientation==2) continue;
1426  put_symbol(&s->c, s->header_state, s->plane[plane_index].band[level][orientation].qlog, 1);
1427  }
1428  }
1429  }
1430 }
1431 
1432 static void encode_header(SnowContext *s){
1433  int plane_index, i;
1434  uint8_t kstate[32];
1435 
1436  memset(kstate, MID_STATE, sizeof(kstate));
1437 
1438  put_rac(&s->c, kstate, s->keyframe);
1439  if(s->keyframe || s->always_reset){
1442  s->last_qlog=
1443  s->last_qbias=
1444  s->last_mv_scale=
1445  s->last_block_max_depth= 0;
1446  for(plane_index=0; plane_index<2; plane_index++){
1447  Plane *p= &s->plane[plane_index];
1448  p->last_htaps=0;
1449  p->last_diag_mc=0;
1450  memset(p->last_hcoeff, 0, sizeof(p->last_hcoeff));
1451  }
1452  }
1453  if(s->keyframe){
1454  put_symbol(&s->c, s->header_state, s->version, 0);
1455  put_rac(&s->c, s->header_state, s->always_reset);
1459  put_symbol(&s->c, s->header_state, s->colorspace_type, 0);
1460  put_symbol(&s->c, s->header_state, s->chroma_h_shift, 0);
1461  put_symbol(&s->c, s->header_state, s->chroma_v_shift, 0);
1463 // put_rac(&s->c, s->header_state, s->rate_scalability);
1464  put_symbol(&s->c, s->header_state, s->max_ref_frames-1, 0);
1465 
1466  encode_qlogs(s);
1467  }
1468 
1469  if(!s->keyframe){
1470  int update_mc=0;
1471  for(plane_index=0; plane_index<2; plane_index++){
1472  Plane *p= &s->plane[plane_index];
1473  update_mc |= p->last_htaps != p->htaps;
1474  update_mc |= p->last_diag_mc != p->diag_mc;
1475  update_mc |= !!memcmp(p->last_hcoeff, p->hcoeff, sizeof(p->hcoeff));
1476  }
1477  put_rac(&s->c, s->header_state, update_mc);
1478  if(update_mc){
1479  for(plane_index=0; plane_index<2; plane_index++){
1480  Plane *p= &s->plane[plane_index];
1481  put_rac(&s->c, s->header_state, p->diag_mc);
1482  put_symbol(&s->c, s->header_state, p->htaps/2-1, 0);
1483  for(i= p->htaps/2; i; i--)
1484  put_symbol(&s->c, s->header_state, FFABS(p->hcoeff[i]), 0);
1485  }
1486  }
1488  put_rac(&s->c, s->header_state, 1);
1490  encode_qlogs(s);
1491  }else
1492  put_rac(&s->c, s->header_state, 0);
1493  }
1494 
1496  put_symbol(&s->c, s->header_state, s->qlog - s->last_qlog , 1);
1497  put_symbol(&s->c, s->header_state, s->mv_scale - s->last_mv_scale, 1);
1498  put_symbol(&s->c, s->header_state, s->qbias - s->last_qbias , 1);
1500 
1501 }
1502 
1503 static void update_last_header_values(SnowContext *s){
1504  int plane_index;
1505 
1506  if(!s->keyframe){
1507  for(plane_index=0; plane_index<2; plane_index++){
1508  Plane *p= &s->plane[plane_index];
1509  p->last_diag_mc= p->diag_mc;
1510  p->last_htaps = p->htaps;
1511  memcpy(p->last_hcoeff, p->hcoeff, sizeof(p->hcoeff));
1512  }
1513  }
1514 
1516  s->last_qlog = s->qlog;
1517  s->last_qbias = s->qbias;
1518  s->last_mv_scale = s->mv_scale;
1521 }
1522 
1523 static int qscale2qlog(int qscale){
1524  return rint(QROOT*log(qscale / (float)FF_QP2LAMBDA)/log(2))
1525  + 61*QROOT/8;
1526 }
1527 
1528 static int ratecontrol_1pass(SnowContext *s, AVFrame *pict)
1529 {
1530  /* Estimate the frame's complexity as a sum of weighted dwt coefficients.
1531  * FIXME we know exact mv bits at this point,
1532  * but ratecontrol isn't set up to include them. */
1533  uint32_t coef_sum= 0;
1534  int level, orientation, delta_qlog;
1535 
1536  for(level=0; level<s->spatial_decomposition_count; level++){
1537  for(orientation=level ? 1 : 0; orientation<4; orientation++){
1538  SubBand *b= &s->plane[0].band[level][orientation];
1539  IDWTELEM *buf= b->ibuf;
1540  const int w= b->width;
1541  const int h= b->height;
1542  const int stride= b->stride;
1543  const int qlog= av_clip(2*QROOT + b->qlog, 0, QROOT*16);
1544  const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
1545  const int qdiv= (1<<16)/qmul;
1546  int x, y;
1547  //FIXME this is ugly
1548  for(y=0; y<h; y++)
1549  for(x=0; x<w; x++)
1550  buf[x+y*stride]= b->buf[x+y*stride];
1551  if(orientation==0)
1552  decorrelate(s, b, buf, stride, 1, 0);
1553  for(y=0; y<h; y++)
1554  for(x=0; x<w; x++)
1555  coef_sum+= abs(buf[x+y*stride]) * qdiv >> 16;
1556  }
1557  }
1558 
1559  /* ugly, ratecontrol just takes a sqrt again */
1560  coef_sum = (uint64_t)coef_sum * coef_sum >> 16;
1561  assert(coef_sum < INT_MAX);
1562 
1563  if(pict->pict_type == AV_PICTURE_TYPE_I){
1564  s->m.current_picture.mb_var_sum= coef_sum;
1566  }else{
1567  s->m.current_picture.mc_mb_var_sum= coef_sum;
1569  }
1570 
1571  pict->quality= ff_rate_estimate_qscale(&s->m, 1);
1572  if (pict->quality < 0)
1573  return INT_MIN;
1574  s->lambda= pict->quality * 3/2;
1575  delta_qlog= qscale2qlog(pict->quality) - s->qlog;
1576  s->qlog+= delta_qlog;
1577  return delta_qlog;
1578 }
1579 
1580 static void calculate_visual_weight(SnowContext *s, Plane *p){
1581  int width = p->width;
1582  int height= p->height;
1583  int level, orientation, x, y;
1584 
1585  for(level=0; level<s->spatial_decomposition_count; level++){
1586  for(orientation=level ? 1 : 0; orientation<4; orientation++){
1587  SubBand *b= &p->band[level][orientation];
1588  IDWTELEM *ibuf= b->ibuf;
1589  int64_t error=0;
1590 
1591  memset(s->spatial_idwt_buffer, 0, sizeof(*s->spatial_idwt_buffer)*width*height);
1592  ibuf[b->width/2 + b->height/2*b->stride]= 256*16;
1594  for(y=0; y<height; y++){
1595  for(x=0; x<width; x++){
1596  int64_t d= s->spatial_idwt_buffer[x + y*width]*16;
1597  error += d*d;
1598  }
1599  }
1600 
1601  b->qlog= (int)(log(352256.0/sqrt(error)) / log(pow(2.0, 1.0/QROOT))+0.5);
1602  }
1603  }
1604 }
1605 
1606 static int encode_frame(AVCodecContext *avctx, unsigned char *buf, int buf_size, void *data){
1607  SnowContext *s = avctx->priv_data;
1608  RangeCoder * const c= &s->c;
1609  AVFrame *pict = data;
1610  const int width= s->avctx->width;
1611  const int height= s->avctx->height;
1612  int level, orientation, plane_index, i, y;
1613  uint8_t rc_header_bak[sizeof(s->header_state)];
1614  uint8_t rc_block_bak[sizeof(s->block_state)];
1615 
1616  ff_init_range_encoder(c, buf, buf_size);
1617  ff_build_rac_states(c, 0.05*(1LL<<32), 256-8);
1618 
1619  for(i=0; i<3; i++){
1620  int shift= !!i;
1621  for(y=0; y<(height>>shift); y++)
1622  memcpy(&s->input_picture.data[i][y * s->input_picture.linesize[i]],
1623  &pict->data[i][y * pict->linesize[i]],
1624  width>>shift);
1625  }
1626  s->new_picture = *pict;
1627 
1628  s->m.picture_number= avctx->frame_number;
1629  if(avctx->flags&CODEC_FLAG_PASS2){
1630  s->m.pict_type =
1633  if(!(avctx->flags&CODEC_FLAG_QSCALE)) {
1634  pict->quality= ff_rate_estimate_qscale(&s->m, 0);
1635  if (pict->quality < 0)
1636  return -1;
1637  }
1638  }else{
1639  s->keyframe= avctx->gop_size==0 || avctx->frame_number % avctx->gop_size == 0;
1640  s->m.pict_type=
1642  }
1643 
1644  if(s->pass1_rc && avctx->frame_number == 0)
1645  pict->quality= 2*FF_QP2LAMBDA;
1646  if(pict->quality){
1647  s->qlog= qscale2qlog(pict->quality);
1648  s->lambda = pict->quality * 3/2;
1649  }
1650  if(s->qlog < 0 || (!pict->quality && (avctx->flags & CODEC_FLAG_QSCALE))){
1651  s->qlog= LOSSLESS_QLOG;
1652  s->lambda = 0;
1653  }//else keep previous frame's qlog until after motion estimation
1654 
1656 
1659  s->m.current_picture.f.pts = pict->pts;
1660  if(pict->pict_type == AV_PICTURE_TYPE_P){
1661  int block_width = (width +15)>>4;
1662  int block_height= (height+15)>>4;
1663  int stride= s->current_picture.linesize[0];
1664 
1665  assert(s->current_picture.data[0]);
1666  assert(s->last_picture[0].data[0]);
1667 
1668  s->m.avctx= s->avctx;
1669  s->m.current_picture.f.data[0] = s->current_picture.data[0];
1670  s->m. last_picture.f.data[0] = s->last_picture[0].data[0];
1671  s->m. new_picture.f.data[0] = s-> input_picture.data[0];
1672  s->m. last_picture_ptr= &s->m. last_picture;
1673  s->m.linesize=
1674  s->m. last_picture.f.linesize[0] =
1675  s->m. new_picture.f.linesize[0] =
1676  s->m.current_picture.f.linesize[0] = stride;
1678  s->m.width = width;
1679  s->m.height= height;
1680  s->m.mb_width = block_width;
1681  s->m.mb_height= block_height;
1682  s->m.mb_stride= s->m.mb_width+1;
1683  s->m.b8_stride= 2*s->m.mb_width+1;
1684  s->m.f_code=1;
1685  s->m.pict_type= pict->pict_type;
1686  s->m.me_method= s->avctx->me_method;
1687  s->m.me.scene_change_score=0;
1688  s->m.flags= s->avctx->flags;
1689  s->m.quarter_sample= (s->avctx->flags & CODEC_FLAG_QPEL)!=0;
1690  s->m.out_format= FMT_H263;
1691  s->m.unrestricted_mv= 1;
1692 
1693  s->m.lambda = s->lambda;
1694  s->m.qscale= (s->m.lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
1695  s->lambda2= s->m.lambda2= (s->m.lambda*s->m.lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT;
1696 
1697  s->m.dsp= s->dsp; //move
1698  ff_init_me(&s->m);
1699  s->dsp= s->m.dsp;
1700  }
1701 
1702  if(s->pass1_rc){
1703  memcpy(rc_header_bak, s->header_state, sizeof(s->header_state));
1704  memcpy(rc_block_bak, s->block_state, sizeof(s->block_state));
1705  }
1706 
1707 redo_frame:
1708 
1709  if(pict->pict_type == AV_PICTURE_TYPE_I)
1711  else
1713 
1714  s->m.pict_type = pict->pict_type;
1715  s->qbias= pict->pict_type == AV_PICTURE_TYPE_P ? 2 : 0;
1716 
1718 
1720  for(plane_index=0; plane_index<3; plane_index++){
1721  calculate_visual_weight(s, &s->plane[plane_index]);
1722  }
1723  }
1724 
1725  encode_header(s);
1726  s->m.misc_bits = 8*(s->c.bytestream - s->c.bytestream_start);
1727  encode_blocks(s, 1);
1728  s->m.mv_bits = 8*(s->c.bytestream - s->c.bytestream_start) - s->m.misc_bits;
1729 
1730  for(plane_index=0; plane_index<3; plane_index++){
1731  Plane *p= &s->plane[plane_index];
1732  int w= p->width;
1733  int h= p->height;
1734  int x, y;
1735 // int bits= put_bits_count(&s->c.pb);
1736 
1737  if (!s->memc_only) {
1738  //FIXME optimize
1739  if(pict->data[plane_index]) //FIXME gray hack
1740  for(y=0; y<h; y++){
1741  for(x=0; x<w; x++){
1742  s->spatial_idwt_buffer[y*w + x]= pict->data[plane_index][y*pict->linesize[plane_index] + x]<<FRAC_BITS;
1743  }
1744  }
1745  predict_plane(s, s->spatial_idwt_buffer, plane_index, 0);
1746 
1747  if( plane_index==0
1748  && pict->pict_type == AV_PICTURE_TYPE_P
1749  && !(avctx->flags&CODEC_FLAG_PASS2)
1751  ff_init_range_encoder(c, buf, buf_size);
1752  ff_build_rac_states(c, 0.05*(1LL<<32), 256-8);
1754  s->keyframe=1;
1756  goto redo_frame;
1757  }
1758 
1759  if(s->qlog == LOSSLESS_QLOG){
1760  for(y=0; y<h; y++){
1761  for(x=0; x<w; x++){
1762  s->spatial_dwt_buffer[y*w + x]= (s->spatial_idwt_buffer[y*w + x] + (1<<(FRAC_BITS-1))-1)>>FRAC_BITS;
1763  }
1764  }
1765  }else{
1766  for(y=0; y<h; y++){
1767  for(x=0; x<w; x++){
1769  }
1770  }
1771  }
1772 
1773  /* if(QUANTIZE2)
1774  dwt_quantize(s, p, s->spatial_dwt_buffer, w, h, w, s->spatial_decomposition_type);
1775  else*/
1777 
1778  if(s->pass1_rc && plane_index==0){
1779  int delta_qlog = ratecontrol_1pass(s, pict);
1780  if (delta_qlog <= INT_MIN)
1781  return -1;
1782  if(delta_qlog){
1783  //reordering qlog in the bitstream would eliminate this reset
1784  ff_init_range_encoder(c, buf, buf_size);
1785  memcpy(s->header_state, rc_header_bak, sizeof(s->header_state));
1786  memcpy(s->block_state, rc_block_bak, sizeof(s->block_state));
1787  encode_header(s);
1788  encode_blocks(s, 0);
1789  }
1790  }
1791 
1792  for(level=0; level<s->spatial_decomposition_count; level++){
1793  for(orientation=level ? 1 : 0; orientation<4; orientation++){
1794  SubBand *b= &p->band[level][orientation];
1795 
1796  if(!QUANTIZE2)
1797  quantize(s, b, b->ibuf, b->buf, b->stride, s->qbias);
1798  if(orientation==0)
1799  decorrelate(s, b, b->ibuf, b->stride, pict->pict_type == AV_PICTURE_TYPE_P, 0);
1800  encode_subband(s, b, b->ibuf, b->parent ? b->parent->ibuf : NULL, b->stride, orientation);
1801  assert(b->parent==NULL || b->parent->stride == b->stride*2);
1802  if(orientation==0)
1803  correlate(s, b, b->ibuf, b->stride, 1, 0);
1804  }
1805  }
1806 
1807  for(level=0; level<s->spatial_decomposition_count; level++){
1808  for(orientation=level ? 1 : 0; orientation<4; orientation++){
1809  SubBand *b= &p->band[level][orientation];
1810 
1811  dequantize(s, b, b->ibuf, b->stride);
1812  }
1813  }
1814 
1816  if(s->qlog == LOSSLESS_QLOG){
1817  for(y=0; y<h; y++){
1818  for(x=0; x<w; x++){
1819  s->spatial_idwt_buffer[y*w + x]<<=FRAC_BITS;
1820  }
1821  }
1822  }
1823  predict_plane(s, s->spatial_idwt_buffer, plane_index, 1);
1824  }else{
1825  //ME/MC only
1826  if(pict->pict_type == AV_PICTURE_TYPE_I){
1827  for(y=0; y<h; y++){
1828  for(x=0; x<w; x++){
1829  s->current_picture.data[plane_index][y*s->current_picture.linesize[plane_index] + x]=
1830  pict->data[plane_index][y*pict->linesize[plane_index] + x];
1831  }
1832  }
1833  }else{
1834  memset(s->spatial_idwt_buffer, 0, sizeof(IDWTELEM)*w*h);
1835  predict_plane(s, s->spatial_idwt_buffer, plane_index, 1);
1836  }
1837  }
1838  if(s->avctx->flags&CODEC_FLAG_PSNR){
1839  int64_t error= 0;
1840 
1841  if(pict->data[plane_index]) //FIXME gray hack
1842  for(y=0; y<h; y++){
1843  for(x=0; x<w; x++){
1844  int d= s->current_picture.data[plane_index][y*s->current_picture.linesize[plane_index] + x] - pict->data[plane_index][y*pict->linesize[plane_index] + x];
1845  error += d*d;
1846  }
1847  }
1848  s->avctx->error[plane_index] += error;
1849  s->current_picture.error[plane_index] = error;
1850  }
1851 
1852  }
1853 
1854  update_last_header_values(s);
1855 
1856  ff_snow_release_buffer(avctx);
1857 
1859  s->current_picture.pict_type = pict->pict_type;
1860  s->current_picture.quality = pict->quality;
1861  s->m.frame_bits = 8*(s->c.bytestream - s->c.bytestream_start);
1862  s->m.p_tex_bits = s->m.frame_bits - s->m.misc_bits - s->m.mv_bits;
1865  s->m.current_picture.f.quality = pict->quality;
1866  s->m.total_bits += 8*(s->c.bytestream - s->c.bytestream_start);
1867  if(s->pass1_rc)
1868  if (ff_rate_estimate_qscale(&s->m, 0) < 0)
1869  return -1;
1870  if(avctx->flags&CODEC_FLAG_PASS1)
1871  ff_write_pass1_stats(&s->m);
1872  s->m.last_pict_type = s->m.pict_type;
1873  avctx->frame_bits = s->m.frame_bits;
1874  avctx->mv_bits = s->m.mv_bits;
1875  avctx->misc_bits = s->m.misc_bits;
1876  avctx->p_tex_bits = s->m.p_tex_bits;
1877 
1878  emms_c();
1879 
1880  return ff_rac_terminate(c);
1881 }
1882 
1883 static av_cold int encode_end(AVCodecContext *avctx)
1884 {
1885  SnowContext *s = avctx->priv_data;
1886 
1887  ff_snow_common_end(s);
1888  if (s->input_picture.data[0])
1889  avctx->release_buffer(avctx, &s->input_picture);
1890  av_free(avctx->stats_out);
1891 
1892  return 0;
1893 }
1894 
1895 #define OFFSET(x) offsetof(SnowContext, x)
1896 #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
1897 static const AVOption options[] = {
1898  { "memc_only", "Only do ME/MC (I frames -> ref, P frame -> ME+MC).", OFFSET(memc_only), AV_OPT_TYPE_INT, { 0 }, 0, 1, VE },
1899  { NULL },
1900 };
1901 
1902 static const AVClass snowenc_class = {
1903  .class_name = "snow encoder",
1904  .item_name = av_default_item_name,
1905  .option = options,
1906  .version = LIBAVUTIL_VERSION_INT,
1907 };
1908 
1909 AVCodec ff_snow_encoder = {
1910  .name = "snow",
1911  .type = AVMEDIA_TYPE_VIDEO,
1912  .id = CODEC_ID_SNOW,
1913  .priv_data_size = sizeof(SnowContext),
1914  .init = encode_init,
1915  .encode = encode_frame,
1916  .close = encode_end,
1917  .long_name = NULL_IF_CONFIG_SMALL("Snow"),
1918  .priv_class = &snowenc_class,
1919 };
1920 #endif