LCOV - code coverage report
Current view: top level - libavcodec - snowenc.c (source / functions) Hit Total Coverage
Test: coverage.info Lines: 994 1223 81.3 %
Date: 2017-12-16 13:57:32 Functions: 26 29 89.7 %

          Line data    Source code
       1             : /*
       2             :  * Copyright (C) 2004 Michael Niedermayer <michaelni@gmx.at>
       3             :  *
       4             :  * This file is part of FFmpeg.
       5             :  *
       6             :  * FFmpeg 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             :  * FFmpeg 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 FFmpeg; 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/libm.h"
      23             : #include "libavutil/log.h"
      24             : #include "libavutil/opt.h"
      25             : #include "libavutil/pixdesc.h"
      26             : #include "avcodec.h"
      27             : #include "internal.h"
      28             : #include "snow_dwt.h"
      29             : #include "snow.h"
      30             : 
      31             : #include "rangecoder.h"
      32             : #include "mathops.h"
      33             : 
      34             : #include "mpegvideo.h"
      35             : #include "h263.h"
      36             : 
      37          11 : static av_cold int encode_init(AVCodecContext *avctx)
      38             : {
      39          11 :     SnowContext *s = avctx->priv_data;
      40             :     int plane_index, ret;
      41             :     int i;
      42             : 
      43             : #if FF_API_PRIVATE_OPT
      44             : FF_DISABLE_DEPRECATION_WARNINGS
      45          11 :     if (avctx->prediction_method)
      46           0 :         s->pred = avctx->prediction_method;
      47             : FF_ENABLE_DEPRECATION_WARNINGS
      48             : #endif
      49             : 
      50          11 :     if(s->pred == DWT_97
      51           8 :        && (avctx->flags & AV_CODEC_FLAG_QSCALE)
      52           8 :        && avctx->global_quality == 0){
      53           0 :         av_log(avctx, AV_LOG_ERROR, "The 9/7 wavelet is incompatible with lossless mode.\n");
      54           0 :         return AVERROR(EINVAL);
      55             :     }
      56             : 
      57          11 :     s->spatial_decomposition_type= s->pred; //FIXME add decorrelator type r transform_type
      58             : 
      59          11 :     s->mv_scale       = (avctx->flags & AV_CODEC_FLAG_QPEL) ? 2 : 4;
      60          11 :     s->block_max_depth= (avctx->flags & AV_CODEC_FLAG_4MV ) ? 1 : 0;
      61             : 
      62          44 :     for(plane_index=0; plane_index<3; plane_index++){
      63          33 :         s->plane[plane_index].diag_mc= 1;
      64          33 :         s->plane[plane_index].htaps= 6;
      65          33 :         s->plane[plane_index].hcoeff[0]=  40;
      66          33 :         s->plane[plane_index].hcoeff[1]= -10;
      67          33 :         s->plane[plane_index].hcoeff[2]=   2;
      68          33 :         s->plane[plane_index].fast_mc= 1;
      69             :     }
      70             : 
      71          11 :     if ((ret = ff_snow_common_init(avctx)) < 0) {
      72           0 :         return ret;
      73             :     }
      74          11 :     ff_mpegvideoencdsp_init(&s->mpvencdsp, avctx);
      75             : 
      76          11 :     ff_snow_alloc_blocks(s);
      77             : 
      78          11 :     s->version=0;
      79             : 
      80          11 :     s->m.avctx   = avctx;
      81          11 :     s->m.bit_rate= avctx->bit_rate;
      82          11 :     s->m.lmin    = avctx->mb_lmin;
      83          11 :     s->m.lmax    = avctx->mb_lmax;
      84             : 
      85          11 :     s->m.me.temp      =
      86          11 :     s->m.me.scratchpad= av_mallocz_array((avctx->width+64), 2*16*2*sizeof(uint8_t));
      87          11 :     s->m.me.map       = av_mallocz(ME_MAP_SIZE*sizeof(uint32_t));
      88          11 :     s->m.me.score_map = av_mallocz(ME_MAP_SIZE*sizeof(uint32_t));
      89          11 :     s->m.sc.obmc_scratchpad= av_mallocz(MB_SIZE*MB_SIZE*12*sizeof(uint32_t));
      90          11 :     if (!s->m.me.scratchpad || !s->m.me.map || !s->m.me.score_map || !s->m.sc.obmc_scratchpad)
      91           0 :         return AVERROR(ENOMEM);
      92             : 
      93          11 :     ff_h263_encode_init(&s->m); //mv_penalty
      94             : 
      95          11 :     s->max_ref_frames = av_clip(avctx->refs, 1, MAX_REF_FRAMES);
      96             : 
      97          11 :     if(avctx->flags&AV_CODEC_FLAG_PASS1){
      98           0 :         if(!avctx->stats_out)
      99           0 :             avctx->stats_out = av_mallocz(256);
     100             : 
     101           0 :         if (!avctx->stats_out)
     102           0 :             return AVERROR(ENOMEM);
     103             :     }
     104          11 :     if((avctx->flags&AV_CODEC_FLAG_PASS2) || !(avctx->flags&AV_CODEC_FLAG_QSCALE)){
     105           0 :         ret = ff_rate_control_init(&s->m);
     106           0 :         if(ret < 0)
     107           0 :             return ret;
     108             :     }
     109          11 :     s->pass1_rc= !(avctx->flags & (AV_CODEC_FLAG_QSCALE|AV_CODEC_FLAG_PASS2));
     110             : 
     111          11 :     switch(avctx->pix_fmt){
     112          11 :     case AV_PIX_FMT_YUV444P:
     113             : //    case AV_PIX_FMT_YUV422P:
     114             :     case AV_PIX_FMT_YUV420P:
     115             : //    case AV_PIX_FMT_YUV411P:
     116             :     case AV_PIX_FMT_YUV410P:
     117          11 :         s->nb_planes = 3;
     118          11 :         s->colorspace_type= 0;
     119          11 :         break;
     120           0 :     case AV_PIX_FMT_GRAY8:
     121           0 :         s->nb_planes = 1;
     122           0 :         s->colorspace_type = 1;
     123           0 :         break;
     124             : /*    case AV_PIX_FMT_RGB32:
     125             :         s->colorspace= 1;
     126             :         break;*/
     127           0 :     default:
     128           0 :         av_log(avctx, AV_LOG_ERROR, "pixel format not supported\n");
     129           0 :         return AVERROR_PATCHWELCOME;
     130             :     }
     131             : 
     132          11 :     ret = av_pix_fmt_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_h_shift,
     133             :                                            &s->chroma_v_shift);
     134          11 :     if (ret) {
     135           0 :         av_log(avctx, AV_LOG_ERROR, "pixel format invalid or unknown\n");
     136           0 :         return ret;
     137             :     }
     138             : 
     139          11 :     ff_set_cmp(&s->mecc, s->mecc.me_cmp, s->avctx->me_cmp);
     140          11 :     ff_set_cmp(&s->mecc, s->mecc.me_sub_cmp, s->avctx->me_sub_cmp);
     141             : 
     142          11 :     s->input_picture = av_frame_alloc();
     143          11 :     if (!s->input_picture)
     144           0 :         return AVERROR(ENOMEM);
     145             : 
     146          11 :     if ((ret = ff_snow_get_buffer(s, s->input_picture)) < 0)
     147           0 :         return ret;
     148             : 
     149          11 :     if(s->motion_est == FF_ME_ITER){
     150           6 :         int size= s->b_width * s->b_height << 2*s->block_max_depth;
     151          12 :         for(i=0; i<s->max_ref_frames; i++){
     152           6 :             s->ref_mvs[i]= av_mallocz_array(size, sizeof(int16_t[2]));
     153           6 :             s->ref_scores[i]= av_mallocz_array(size, sizeof(uint32_t));
     154           6 :             if (!s->ref_mvs[i] || !s->ref_scores[i])
     155           0 :                 return AVERROR(ENOMEM);
     156             :         }
     157             :     }
     158             : 
     159          11 :     return 0;
     160             : }
     161             : 
     162             : //near copy & paste from dsputil, FIXME
     163     1673460 : static int pix_sum(uint8_t * pix, int line_size, int w, int h)
     164             : {
     165             :     int s, i, j;
     166             : 
     167     1673460 :     s = 0;
     168    13095540 :     for (i = 0; i < h; i++) {
     169   109918080 :         for (j = 0; j < w; j++) {
     170    98496000 :             s += pix[0];
     171    98496000 :             pix ++;
     172             :         }
     173    11422080 :         pix += line_size - w;
     174             :     }
     175     1673460 :     return s;
     176             : }
     177             : 
     178             : //near copy & paste from dsputil, FIXME
     179      557820 : static int pix_norm1(uint8_t * pix, int line_size, int w)
     180             : {
     181             :     int s, i, j;
     182      557820 :     uint32_t *sq = ff_square_tab + 256;
     183             : 
     184      557820 :     s = 0;
     185     6268860 :     for (i = 0; i < w; i++) {
     186    71375040 :         for (j = 0; j < w; j ++) {
     187    65664000 :             s += sq[pix[0]];
     188    65664000 :             pix ++;
     189             :         }
     190     5711040 :         pix += line_size - w;
     191             :     }
     192      557820 :     return s;
     193             : }
     194             : 
     195     1989788 : static inline int get_penalty_factor(int lambda, int lambda2, int type){
     196     1989788 :     switch(type&0xFF){
     197     1374300 :     default:
     198             :     case FF_CMP_SAD:
     199     1374300 :         return lambda>>FF_LAMBDA_SHIFT;
     200           0 :     case FF_CMP_DCT:
     201           0 :         return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
     202           0 :     case FF_CMP_W53:
     203           0 :         return (4*lambda)>>(FF_LAMBDA_SHIFT);
     204      333608 :     case FF_CMP_W97:
     205      333608 :         return (2*lambda)>>(FF_LAMBDA_SHIFT);
     206           0 :     case FF_CMP_SATD:
     207             :     case FF_CMP_DCT264:
     208           0 :         return (2*lambda)>>FF_LAMBDA_SHIFT;
     209      281880 :     case FF_CMP_RD:
     210             :     case FF_CMP_PSNR:
     211             :     case FF_CMP_SSE:
     212             :     case FF_CMP_NSSE:
     213      281880 :         return lambda2>>FF_LAMBDA_SHIFT;
     214           0 :     case FF_CMP_BIT:
     215           0 :         return 1;
     216             :     }
     217             : }
     218             : 
     219             : //FIXME copy&paste
     220             : #define P_LEFT P[1]
     221             : #define P_TOP P[2]
     222             : #define P_TOPRIGHT P[3]
     223             : #define P_MEDIAN P[4]
     224             : #define P_MV1 P[9]
     225             : #define FLAG_QPEL   1 //must be 1
     226             : 
     227      567000 : static int encode_q_branch(SnowContext *s, int level, int x, int y){
     228             :     uint8_t p_buffer[1024];
     229             :     uint8_t i_buffer[1024];
     230             :     uint8_t p_state[sizeof(s->block_state)];
     231             :     uint8_t i_state[sizeof(s->block_state)];
     232             :     RangeCoder pc, ic;
     233      567000 :     uint8_t *pbbak= s->c.bytestream;
     234      567000 :     uint8_t *pbbak_start= s->c.bytestream_start;
     235             :     int score, score2, iscore, i_len, p_len, block_s, sum, base_bits;
     236      567000 :     const int w= s->b_width  << s->block_max_depth;
     237      567000 :     const int h= s->b_height << s->block_max_depth;
     238      567000 :     const int rem_depth= s->block_max_depth - level;
     239      567000 :     const int index= (x + y*w) << rem_depth;
     240      567000 :     const int block_w= 1<<(LOG2_MB_SIZE - level);
     241      567000 :     int trx= (x+1)<<rem_depth;
     242      567000 :     int try= (y+1)<<rem_depth;
     243      567000 :     const BlockNode *left  = x ? &s->block[index-1] : &null_block;
     244      567000 :     const BlockNode *top   = y ? &s->block[index-w] : &null_block;
     245      567000 :     const BlockNode *right = trx<w ? &s->block[index+1] : &null_block;
     246      567000 :     const BlockNode *bottom= try<h ? &s->block[index+w] : &null_block;
     247      567000 :     const BlockNode *tl    = y && x ? &s->block[index-w-1] : left;
     248      567000 :     const BlockNode *tr    = y && trx<w && ((x&1)==0 || level==0) ? &s->block[index-w+(1<<rem_depth)] : tl; //FIXME use lt
     249      567000 :     int pl = left->color[0];
     250      567000 :     int pcb= left->color[1];
     251      567000 :     int pcr= left->color[2];
     252             :     int pmx, pmy;
     253      567000 :     int mx=0, my=0;
     254             :     int l,cr,cb;
     255      567000 :     const int stride= s->current_picture->linesize[0];
     256      567000 :     const int uvstride= s->current_picture->linesize[1];
     257     1701000 :     uint8_t *current_data[3]= { s->input_picture->data[0] + (x + y*  stride)*block_w,
     258      567000 :                                 s->input_picture->data[1] + ((x*block_w)>>s->chroma_h_shift) + ((y*uvstride*block_w)>>s->chroma_v_shift),
     259      567000 :                                 s->input_picture->data[2] + ((x*block_w)>>s->chroma_h_shift) + ((y*uvstride*block_w)>>s->chroma_v_shift)};
     260             :     int P[10][2];
     261             :     int16_t last_mv[3][2];
     262      567000 :     int qpel= !!(s->avctx->flags & AV_CODEC_FLAG_QPEL); //unused
     263      567000 :     const int shift= 1+qpel;
     264      567000 :     MotionEstContext *c= &s->m.me;
     265      567000 :     int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref);
     266      567000 :     int mx_context= av_log2(2*FFABS(left->mx - top->mx));
     267      567000 :     int my_context= av_log2(2*FFABS(left->my - top->my));
     268      567000 :     int s_context= 2*left->level + 2*top->level + tl->level + tr->level;
     269             :     int ref, best_ref, ref_score, ref_mx, ref_my;
     270             : 
     271             :     av_assert0(sizeof(s->block_state) >= 256);
     272      567000 :     if(s->keyframe){
     273        9180 :         set_blocks(s, level, x, y, pl, pcb, pcr, 0, 0, 0, BLOCK_INTRA);
     274        9180 :         return 0;
     275             :     }
     276             : 
     277             : //    clip predictors / edge ?
     278             : 
     279      557820 :     P_LEFT[0]= left->mx;
     280      557820 :     P_LEFT[1]= left->my;
     281      557820 :     P_TOP [0]= top->mx;
     282      557820 :     P_TOP [1]= top->my;
     283      557820 :     P_TOPRIGHT[0]= tr->mx;
     284      557820 :     P_TOPRIGHT[1]= tr->my;
     285             : 
     286      557820 :     last_mv[0][0]= s->block[index].mx;
     287      557820 :     last_mv[0][1]= s->block[index].my;
     288      557820 :     last_mv[1][0]= right->mx;
     289      557820 :     last_mv[1][1]= right->my;
     290      557820 :     last_mv[2][0]= bottom->mx;
     291      557820 :     last_mv[2][1]= bottom->my;
     292             : 
     293      557820 :     s->m.mb_stride=2;
     294      557820 :     s->m.mb_x=
     295      557820 :     s->m.mb_y= 0;
     296      557820 :     c->skip= 0;
     297             : 
     298             :     av_assert1(c->  stride ==   stride);
     299             :     av_assert1(c->uvstride == uvstride);
     300             : 
     301      557820 :     c->penalty_factor    = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp);
     302      557820 :     c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp);
     303      557820 :     c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp);
     304      557820 :     c->current_mv_penalty= c->mv_penalty[s->m.f_code=1] + MAX_DMV;
     305             : 
     306      557820 :     c->xmin = - x*block_w - 16+3;
     307      557820 :     c->ymin = - y*block_w - 16+3;
     308      557820 :     c->xmax = - (x+1)*block_w + (w<<(LOG2_MB_SIZE - s->block_max_depth)) + 16-3;
     309      557820 :     c->ymax = - (y+1)*block_w + (h<<(LOG2_MB_SIZE - s->block_max_depth)) + 16-3;
     310             : 
     311      557820 :     if(P_LEFT[0]     > (c->xmax<<shift)) P_LEFT[0]    = (c->xmax<<shift);
     312      557820 :     if(P_LEFT[1]     > (c->ymax<<shift)) P_LEFT[1]    = (c->ymax<<shift);
     313      557820 :     if(P_TOP[0]      > (c->xmax<<shift)) P_TOP[0]     = (c->xmax<<shift);
     314      557820 :     if(P_TOP[1]      > (c->ymax<<shift)) P_TOP[1]     = (c->ymax<<shift);
     315      557820 :     if(P_TOPRIGHT[0] < (c->xmin<<shift)) P_TOPRIGHT[0]= (c->xmin<<shift);
     316      557820 :     if(P_TOPRIGHT[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<shift); //due to pmx no clip
     317      557820 :     if(P_TOPRIGHT[1] > (c->ymax<<shift)) P_TOPRIGHT[1]= (c->ymax<<shift);
     318             : 
     319      557820 :     P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
     320      557820 :     P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
     321             : 
     322      557820 :     if (!y) {
     323       16290 :         c->pred_x= P_LEFT[0];
     324       16290 :         c->pred_y= P_LEFT[1];
     325             :     } else {
     326      541530 :         c->pred_x = P_MEDIAN[0];
     327      541530 :         c->pred_y = P_MEDIAN[1];
     328             :     }
     329             : 
     330      557820 :     score= INT_MAX;
     331      557820 :     best_ref= 0;
     332     1115640 :     for(ref=0; ref<s->ref_frames; ref++){
     333      557820 :         init_ref(c, current_data, s->last_picture[ref]->data, NULL, block_w*x, block_w*y, 0);
     334             : 
     335      557820 :         ref_score= ff_epzs_motion_search(&s->m, &ref_mx, &ref_my, P, 0, /*ref_index*/ 0, last_mv,
     336             :                                          (1<<16)>>shift, level-LOG2_MB_SIZE+4, block_w);
     337             : 
     338             :         av_assert2(ref_mx >= c->xmin);
     339             :         av_assert2(ref_mx <= c->xmax);
     340             :         av_assert2(ref_my >= c->ymin);
     341             :         av_assert2(ref_my <= c->ymax);
     342             : 
     343      557820 :         ref_score= c->sub_motion_search(&s->m, &ref_mx, &ref_my, ref_score, 0, 0, level-LOG2_MB_SIZE+4, block_w);
     344      557820 :         ref_score= ff_get_mb_score(&s->m, ref_mx, ref_my, 0, 0, level-LOG2_MB_SIZE+4, block_w, 0);
     345      557820 :         ref_score+= 2*av_log2(2*ref)*c->penalty_factor;
     346      557820 :         if(s->ref_mvs[ref]){
     347        8640 :             s->ref_mvs[ref][index][0]= ref_mx;
     348        8640 :             s->ref_mvs[ref][index][1]= ref_my;
     349        8640 :             s->ref_scores[ref][index]= ref_score;
     350             :         }
     351      557820 :         if(score > ref_score){
     352      557820 :             score= ref_score;
     353      557820 :             best_ref= ref;
     354      557820 :             mx= ref_mx;
     355      557820 :             my= ref_my;
     356             :         }
     357             :     }
     358             :     //FIXME if mb_cmp != SSE then intra cannot be compared currently and mb_penalty vs. lambda2
     359             : 
     360             :   //  subpel search
     361      557820 :     base_bits= get_rac_count(&s->c) - 8*(s->c.bytestream - s->c.bytestream_start);
     362      557820 :     pc= s->c;
     363      557820 :     pc.bytestream_start=
     364      557820 :     pc.bytestream= p_buffer; //FIXME end/start? and at the other stoo
     365      557820 :     memcpy(p_state, s->block_state, sizeof(s->block_state));
     366             : 
     367      557820 :     if(level!=s->block_max_depth)
     368      100440 :         put_rac(&pc, &p_state[4 + s_context], 1);
     369      557820 :     put_rac(&pc, &p_state[1 + left->type + top->type], 0);
     370      557820 :     if(s->ref_frames > 1)
     371           0 :         put_symbol(&pc, &p_state[128 + 1024 + 32*ref_context], best_ref, 0);
     372      557820 :     pred_mv(s, &pmx, &pmy, best_ref, left, top, tr);
     373      557820 :     put_symbol(&pc, &p_state[128 + 32*(mx_context + 16*!!best_ref)], mx - pmx, 1);
     374      557820 :     put_symbol(&pc, &p_state[128 + 32*(my_context + 16*!!best_ref)], my - pmy, 1);
     375      557820 :     p_len= pc.bytestream - pc.bytestream_start;
     376      557820 :     score += (s->lambda2*(get_rac_count(&pc)-base_bits))>>FF_LAMBDA_SHIFT;
     377             : 
     378      557820 :     block_s= block_w*block_w;
     379      557820 :     sum = pix_sum(current_data[0], stride, block_w, block_w);
     380      557820 :     l= (sum + block_s/2)/block_s;
     381      557820 :     iscore = pix_norm1(current_data[0], stride, block_w) - 2*l*sum + l*l*block_s;
     382             : 
     383      557820 :     if (s->nb_planes > 2) {
     384      557820 :         block_s= block_w*block_w>>(s->chroma_h_shift + s->chroma_v_shift);
     385      557820 :         sum = pix_sum(current_data[1], uvstride, block_w>>s->chroma_h_shift, block_w>>s->chroma_v_shift);
     386      557820 :         cb= (sum + block_s/2)/block_s;
     387             :     //    iscore += pix_norm1(&current_mb[1][0], uvstride, block_w>>1) - 2*cb*sum + cb*cb*block_s;
     388      557820 :         sum = pix_sum(current_data[2], uvstride, block_w>>s->chroma_h_shift, block_w>>s->chroma_v_shift);
     389      557820 :         cr= (sum + block_s/2)/block_s;
     390             :     //    iscore += pix_norm1(&current_mb[2][0], uvstride, block_w>>1) - 2*cr*sum + cr*cr*block_s;
     391             :     }else
     392           0 :         cb = cr = 0;
     393             : 
     394      557820 :     ic= s->c;
     395      557820 :     ic.bytestream_start=
     396      557820 :     ic.bytestream= i_buffer; //FIXME end/start? and at the other stoo
     397      557820 :     memcpy(i_state, s->block_state, sizeof(s->block_state));
     398      557820 :     if(level!=s->block_max_depth)
     399      100440 :         put_rac(&ic, &i_state[4 + s_context], 1);
     400      557820 :     put_rac(&ic, &i_state[1 + left->type + top->type], 1);
     401      557820 :     put_symbol(&ic, &i_state[32],  l-pl , 1);
     402      557820 :     if (s->nb_planes > 2) {
     403      557820 :         put_symbol(&ic, &i_state[64], cb-pcb, 1);
     404      557820 :         put_symbol(&ic, &i_state[96], cr-pcr, 1);
     405             :     }
     406      557820 :     i_len= ic.bytestream - ic.bytestream_start;
     407      557820 :     iscore += (s->lambda2*(get_rac_count(&ic)-base_bits))>>FF_LAMBDA_SHIFT;
     408             : 
     409             :     av_assert1(iscore < 255*255*256 + s->lambda2*10);
     410             :     av_assert1(iscore >= 0);
     411             :     av_assert1(l>=0 && l<=255);
     412             :     av_assert1(pl>=0 && pl<=255);
     413             : 
     414      557820 :     if(level==0){
     415      156060 :         int varc= iscore >> 8;
     416      156060 :         int vard= score >> 8;
     417      156060 :         if (vard <= 64 || vard < varc)
     418      134321 :             c->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);
     419             :         else
     420       21739 :             c->scene_change_score+= s->m.qscale;
     421             :     }
     422             : 
     423      557820 :     if(level!=s->block_max_depth){
     424      100440 :         put_rac(&s->c, &s->block_state[4 + s_context], 0);
     425      100440 :         score2 = encode_q_branch(s, level+1, 2*x+0, 2*y+0);
     426      100440 :         score2+= encode_q_branch(s, level+1, 2*x+1, 2*y+0);
     427      100440 :         score2+= encode_q_branch(s, level+1, 2*x+0, 2*y+1);
     428      100440 :         score2+= encode_q_branch(s, level+1, 2*x+1, 2*y+1);
     429      100440 :         score2+= s->lambda2>>FF_LAMBDA_SHIFT; //FIXME exact split overhead
     430             : 
     431      100440 :         if(score2 < score && score2 < iscore)
     432       88607 :             return score2;
     433             :     }
     434             : 
     435      469213 :     if(iscore < score){
     436       82013 :         pred_mv(s, &pmx, &pmy, 0, left, top, tr);
     437       82013 :         memcpy(pbbak, i_buffer, i_len);
     438       82013 :         s->c= ic;
     439       82013 :         s->c.bytestream_start= pbbak_start;
     440       82013 :         s->c.bytestream= pbbak + i_len;
     441       82013 :         set_blocks(s, level, x, y, l, cb, cr, pmx, pmy, 0, BLOCK_INTRA);
     442       82013 :         memcpy(s->block_state, i_state, sizeof(s->block_state));
     443       82013 :         return iscore;
     444             :     }else{
     445      387200 :         memcpy(pbbak, p_buffer, p_len);
     446      387200 :         s->c= pc;
     447      387200 :         s->c.bytestream_start= pbbak_start;
     448      387200 :         s->c.bytestream= pbbak + p_len;
     449      387200 :         set_blocks(s, level, x, y, pl, pcb, pcr, mx, my, best_ref, 0);
     450      387200 :         memcpy(s->block_state, p_state, sizeof(s->block_state));
     451      387200 :         return score;
     452             :     }
     453             : }
     454             : 
     455        9600 : static void encode_q_branch2(SnowContext *s, int level, int x, int y){
     456        9600 :     const int w= s->b_width  << s->block_max_depth;
     457        9600 :     const int rem_depth= s->block_max_depth - level;
     458        9600 :     const int index= (x + y*w) << rem_depth;
     459        9600 :     int trx= (x+1)<<rem_depth;
     460        9600 :     BlockNode *b= &s->block[index];
     461        9600 :     const BlockNode *left  = x ? &s->block[index-1] : &null_block;
     462        9600 :     const BlockNode *top   = y ? &s->block[index-w] : &null_block;
     463        9600 :     const BlockNode *tl    = y && x ? &s->block[index-w-1] : left;
     464        9600 :     const BlockNode *tr    = y && trx<w && ((x&1)==0 || level==0) ? &s->block[index-w+(1<<rem_depth)] : tl; //FIXME use lt
     465        9600 :     int pl = left->color[0];
     466        9600 :     int pcb= left->color[1];
     467        9600 :     int pcr= left->color[2];
     468             :     int pmx, pmy;
     469        9600 :     int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref);
     470        9600 :     int mx_context= av_log2(2*FFABS(left->mx - top->mx)) + 16*!!b->ref;
     471        9600 :     int my_context= av_log2(2*FFABS(left->my - top->my)) + 16*!!b->ref;
     472        9600 :     int s_context= 2*left->level + 2*top->level + tl->level + tr->level;
     473             : 
     474        9600 :     if(s->keyframe){
     475         960 :         set_blocks(s, level, x, y, pl, pcb, pcr, 0, 0, 0, BLOCK_INTRA);
     476         960 :         return;
     477             :     }
     478             : 
     479        8640 :     if(level!=s->block_max_depth){
     480           0 :         if(same_block(b,b+1) && same_block(b,b+w) && same_block(b,b+w+1)){
     481           0 :             put_rac(&s->c, &s->block_state[4 + s_context], 1);
     482             :         }else{
     483           0 :             put_rac(&s->c, &s->block_state[4 + s_context], 0);
     484           0 :             encode_q_branch2(s, level+1, 2*x+0, 2*y+0);
     485           0 :             encode_q_branch2(s, level+1, 2*x+1, 2*y+0);
     486           0 :             encode_q_branch2(s, level+1, 2*x+0, 2*y+1);
     487           0 :             encode_q_branch2(s, level+1, 2*x+1, 2*y+1);
     488           0 :             return;
     489             :         }
     490             :     }
     491        8640 :     if(b->type & BLOCK_INTRA){
     492          40 :         pred_mv(s, &pmx, &pmy, 0, left, top, tr);
     493          40 :         put_rac(&s->c, &s->block_state[1 + (left->type&1) + (top->type&1)], 1);
     494          40 :         put_symbol(&s->c, &s->block_state[32], b->color[0]-pl , 1);
     495          40 :         if (s->nb_planes > 2) {
     496          40 :             put_symbol(&s->c, &s->block_state[64], b->color[1]-pcb, 1);
     497          40 :             put_symbol(&s->c, &s->block_state[96], b->color[2]-pcr, 1);
     498             :         }
     499          40 :         set_blocks(s, level, x, y, b->color[0], b->color[1], b->color[2], pmx, pmy, 0, BLOCK_INTRA);
     500             :     }else{
     501        8600 :         pred_mv(s, &pmx, &pmy, b->ref, left, top, tr);
     502        8600 :         put_rac(&s->c, &s->block_state[1 + (left->type&1) + (top->type&1)], 0);
     503        8600 :         if(s->ref_frames > 1)
     504           0 :             put_symbol(&s->c, &s->block_state[128 + 1024 + 32*ref_context], b->ref, 0);
     505        8600 :         put_symbol(&s->c, &s->block_state[128 + 32*mx_context], b->mx - pmx, 1);
     506        8600 :         put_symbol(&s->c, &s->block_state[128 + 32*my_context], b->my - pmy, 1);
     507        8600 :         set_blocks(s, level, x, y, pl, pcb, pcr, b->mx, b->my, b->ref, 0);
     508             :     }
     509             : }
     510             : 
     511       56355 : static int get_dc(SnowContext *s, int mb_x, int mb_y, int plane_index){
     512             :     int i, x2, y2;
     513       56355 :     Plane *p= &s->plane[plane_index];
     514       56355 :     const int block_size = MB_SIZE >> s->block_max_depth;
     515       56355 :     const int block_w    = plane_index ? block_size>>s->chroma_h_shift : block_size;
     516       56355 :     const int block_h    = plane_index ? block_size>>s->chroma_v_shift : block_size;
     517       56355 :     const uint8_t *obmc  = plane_index ? ff_obmc_tab[s->block_max_depth+s->chroma_h_shift] : ff_obmc_tab[s->block_max_depth];
     518       56355 :     const int obmc_stride= plane_index ? (2*block_size)>>s->chroma_h_shift : 2*block_size;
     519       56355 :     const int ref_stride= s->current_picture->linesize[plane_index];
     520       56355 :     uint8_t *src= s-> input_picture->data[plane_index];
     521       56355 :     IDWTELEM *dst= (IDWTELEM*)s->m.sc.obmc_scratchpad + plane_index*block_size*block_size*4; //FIXME change to unsigned
     522       56355 :     const int b_stride = s->b_width << s->block_max_depth;
     523       56355 :     const int w= p->width;
     524       56355 :     const int h= p->height;
     525       56355 :     int index= mb_x + mb_y*b_stride;
     526       56355 :     BlockNode *b= &s->block[index];
     527       56355 :     BlockNode backup= *b;
     528       56355 :     int ab=0;
     529       56355 :     int aa=0;
     530             : 
     531             :     av_assert2(s->chroma_h_shift == s->chroma_v_shift); //obmc stuff above
     532             : 
     533       56355 :     b->type|= BLOCK_INTRA;
     534       56355 :     b->color[plane_index]= 0;
     535       56355 :     memset(dst, 0, obmc_stride*obmc_stride*sizeof(IDWTELEM));
     536             : 
     537      281775 :     for(i=0; i<4; i++){
     538      225420 :         int mb_x2= mb_x + (i &1) - 1;
     539      225420 :         int mb_y2= mb_y + (i>>1) - 1;
     540      225420 :         int x= block_w*mb_x2 + block_w/2;
     541      225420 :         int y= block_h*mb_y2 + block_h/2;
     542             : 
     543      225420 :         add_yblock(s, 0, NULL, dst + (i&1)*block_w + (i>>1)*obmc_stride*block_h, NULL, obmc,
     544             :                     x, y, block_w, block_h, w, h, obmc_stride, ref_stride, obmc_stride, mb_x2, mb_y2, 0, 0, plane_index);
     545             : 
     546     2352140 :         for(y2= FFMAX(y, 0); y2<FFMIN(h, y+block_h); y2++){
     547    26255072 :             for(x2= FFMAX(x, 0); x2<FFMIN(w, x+block_w); x2++){
     548    24128352 :                 int index= x2-(block_w*mb_x - block_w/2) + (y2-(block_h*mb_y - block_h/2))*obmc_stride;
     549    24128352 :                 int obmc_v= obmc[index];
     550             :                 int d;
     551    24128352 :                 if(y<0) obmc_v += obmc[index + block_h*obmc_stride];
     552    24128352 :                 if(x<0) obmc_v += obmc[index + block_w];
     553    24128352 :                 if(y+block_h>h) obmc_v += obmc[index - block_h*obmc_stride];
     554    24128352 :                 if(x+block_w>w) obmc_v += obmc[index - block_w];
     555             :                 //FIXME precalculate this or simplify it somehow else
     556             : 
     557    24128352 :                 d = -dst[index] + (1<<(FRAC_BITS-1));
     558    24128352 :                 dst[index] = d;
     559    24128352 :                 ab += (src[x2 + y2*ref_stride] - (d>>FRAC_BITS)) * obmc_v;
     560    24128352 :                 aa += obmc_v * obmc_v; //FIXME precalculate this
     561             :             }
     562             :         }
     563             :     }
     564       56355 :     *b= backup;
     565             : 
     566       56355 :     return av_clip_uint8( ROUNDED_DIV(ab<<LOG2_OBMC_MAX, aa) ); //FIXME we should not need clipping
     567             : }
     568             : 
     569     1304392 : static inline int get_block_bits(SnowContext *s, int x, int y, int w){
     570     1304392 :     const int b_stride = s->b_width << s->block_max_depth;
     571     1304392 :     const int b_height = s->b_height<< s->block_max_depth;
     572     1304392 :     int index= x + y*b_stride;
     573     1304392 :     const BlockNode *b     = &s->block[index];
     574     1304392 :     const BlockNode *left  = x ? &s->block[index-1] : &null_block;
     575     1304392 :     const BlockNode *top   = y ? &s->block[index-b_stride] : &null_block;
     576     1304392 :     const BlockNode *tl    = y && x ? &s->block[index-b_stride-1] : left;
     577     1304392 :     const BlockNode *tr    = y && x+w<b_stride ? &s->block[index-b_stride+w] : tl;
     578             :     int dmx, dmy;
     579             : //  int mx_context= av_log2(2*FFABS(left->mx - top->mx));
     580             : //  int my_context= av_log2(2*FFABS(left->my - top->my));
     581             : 
     582     1304392 :     if(x<0 || x>=b_stride || y>=b_height)
     583      197515 :         return 0;
     584             : /*
     585             : 1            0      0
     586             : 01X          1-2    1
     587             : 001XX        3-6    2-3
     588             : 0001XXX      7-14   4-7
     589             : 00001XXXX   15-30   8-15
     590             : */
     591             : //FIXME try accurate rate
     592             : //FIXME intra and inter predictors if surrounding blocks are not the same type
     593     1106877 :     if(b->type & BLOCK_INTRA){
     594       46510 :         return 3+2*( av_log2(2*FFABS(left->color[0] - b->color[0]))
     595       23255 :                    + av_log2(2*FFABS(left->color[1] - b->color[1]))
     596       23255 :                    + av_log2(2*FFABS(left->color[2] - b->color[2])));
     597             :     }else{
     598     1083622 :         pred_mv(s, &dmx, &dmy, b->ref, left, top, tr);
     599     1083622 :         dmx-= b->mx;
     600     1083622 :         dmy-= b->my;
     601     2167244 :         return 2*(1 + av_log2(2*FFABS(dmx)) //FIXME kill the 2* can be merged in lambda
     602     1083622 :                     + av_log2(2*FFABS(dmy))
     603     1083622 :                     + av_log2(2*b->ref));
     604             :     }
     605             : }
     606             : 
     607      316328 : static int get_block_rd(SnowContext *s, int mb_x, int mb_y, int plane_index, uint8_t (*obmc_edged)[MB_SIZE * 2]){
     608      316328 :     Plane *p= &s->plane[plane_index];
     609      316328 :     const int block_size = MB_SIZE >> s->block_max_depth;
     610      316328 :     const int block_w    = plane_index ? block_size>>s->chroma_h_shift : block_size;
     611      316328 :     const int block_h    = plane_index ? block_size>>s->chroma_v_shift : block_size;
     612      316328 :     const int obmc_stride= plane_index ? (2*block_size)>>s->chroma_h_shift : 2*block_size;
     613      316328 :     const int ref_stride= s->current_picture->linesize[plane_index];
     614      316328 :     uint8_t *dst= s->current_picture->data[plane_index];
     615      316328 :     uint8_t *src= s->  input_picture->data[plane_index];
     616      316328 :     IDWTELEM *pred= (IDWTELEM*)s->m.sc.obmc_scratchpad + plane_index*block_size*block_size*4;
     617      316328 :     uint8_t *cur = s->scratchbuf;
     618      316328 :     uint8_t *tmp = s->emu_edge_buffer;
     619      316328 :     const int b_stride = s->b_width << s->block_max_depth;
     620      316328 :     const int b_height = s->b_height<< s->block_max_depth;
     621      316328 :     const int w= p->width;
     622      316328 :     const int h= p->height;
     623             :     int distortion;
     624      316328 :     int rate= 0;
     625      316328 :     const int penalty_factor= get_penalty_factor(s->lambda, s->lambda2, s->avctx->me_cmp);
     626      316328 :     int sx= block_w*mb_x - block_w/2;
     627      316328 :     int sy= block_h*mb_y - block_h/2;
     628      316328 :     int x0= FFMAX(0,-sx);
     629      316328 :     int y0= FFMAX(0,-sy);
     630      316328 :     int x1= FFMIN(block_w*2, w-sx);
     631      316328 :     int y1= FFMIN(block_h*2, h-sy);
     632             :     int i,x,y;
     633             : 
     634             :     av_assert2(s->chroma_h_shift == s->chroma_v_shift); //obmc and square assumtions below chckinhg only block_w
     635             : 
     636      316328 :     ff_snow_pred_block(s, cur, tmp, ref_stride, sx, sy, block_w*2, block_h*2, &s->block[mb_x + mb_y*b_stride], plane_index, w, h);
     637             : 
     638     9270952 :     for(y=y0; y<y1; y++){
     639     8954624 :         const uint8_t *obmc1= obmc_edged[y];
     640     8954624 :         const IDWTELEM *pred1 = pred + y*obmc_stride;
     641     8954624 :         uint8_t *cur1 = cur + y*ref_stride;
     642     8954624 :         uint8_t *dst1 = dst + sx + (sy+y)*ref_stride;
     643   279555200 :         for(x=x0; x<x1; x++){
     644             : #if FRAC_BITS >= LOG2_OBMC_MAX
     645             :             int v = (cur1[x] * obmc1[x]) << (FRAC_BITS - LOG2_OBMC_MAX);
     646             : #else
     647   270600576 :             int v = (cur1[x] * obmc1[x] + (1<<(LOG2_OBMC_MAX - FRAC_BITS-1))) >> (LOG2_OBMC_MAX - FRAC_BITS);
     648             : #endif
     649   270600576 :             v = (v + pred1[x]) >> FRAC_BITS;
     650   270600576 :             if(v&(~255)) v= ~(v>>31);
     651   270600576 :             dst1[x] = v;
     652             :         }
     653             :     }
     654             : 
     655             :     /* copy the regions where obmc[] = (uint8_t)256 */
     656      316328 :     if(LOG2_OBMC_MAX == 8
     657      277774 :         && (mb_x == 0 || mb_x == b_stride-1)
     658       70256 :         && (mb_y == 0 || mb_y == b_height-1)){
     659       31846 :         if(mb_x == 0)
     660       17372 :             x1 = block_w;
     661             :         else
     662       14474 :             x0 = block_w;
     663       31846 :         if(mb_y == 0)
     664       18284 :             y1 = block_h;
     665             :         else
     666       13562 :             y0 = block_h;
     667      286614 :         for(y=y0; y<y1; y++)
     668      254768 :             memcpy(dst + sx+x0 + (sy+y)*ref_stride, cur + x0 + y*ref_stride, x1-x0);
     669             :     }
     670             : 
     671      316328 :     if(block_w==16){
     672             :         /* FIXME rearrange dsputil to fit 32x32 cmp functions */
     673             :         /* FIXME check alignment of the cmp wavelet vs the encoding wavelet */
     674             :         /* FIXME cmps overlap but do not cover the wavelet's whole support.
     675             :          * So improving the score of one block is not strictly guaranteed
     676             :          * to improve the score of the whole frame, thus iterative motion
     677             :          * estimation does not always converge. */
     678      316328 :         if(s->avctx->me_cmp == FF_CMP_W97)
     679      316328 :             distortion = ff_w97_32_c(&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, 32);
     680           0 :         else if(s->avctx->me_cmp == FF_CMP_W53)
     681           0 :             distortion = ff_w53_32_c(&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, 32);
     682             :         else{
     683           0 :             distortion = 0;
     684           0 :             for(i=0; i<4; i++){
     685           0 :                 int off = sx+16*(i&1) + (sy+16*(i>>1))*ref_stride;
     686           0 :                 distortion += s->mecc.me_cmp[0](&s->m, src + off, dst + off, ref_stride, 16);
     687             :             }
     688             :         }
     689             :     }else{
     690             :         av_assert2(block_w==8);
     691           0 :         distortion = s->mecc.me_cmp[0](&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, block_w*2);
     692             :     }
     693             : 
     694      316328 :     if(plane_index==0){
     695     1581640 :         for(i=0; i<4; i++){
     696             : /* ..RRr
     697             :  * .RXx.
     698             :  * rxx..
     699             :  */
     700     1265312 :             rate += get_block_bits(s, mb_x + (i&1) - (i>>1), mb_y + (i>>1), 1);
     701             :         }
     702      316328 :         if(mb_x == b_stride-2)
     703       39080 :             rate += get_block_bits(s, mb_x + 1, mb_y + 1, 1);
     704             :     }
     705      316328 :     return distortion + rate*penalty_factor;
     706             : }
     707             : 
     708           0 : static int get_4block_rd(SnowContext *s, int mb_x, int mb_y, int plane_index){
     709             :     int i, y2;
     710           0 :     Plane *p= &s->plane[plane_index];
     711           0 :     const int block_size = MB_SIZE >> s->block_max_depth;
     712           0 :     const int block_w    = plane_index ? block_size>>s->chroma_h_shift : block_size;
     713           0 :     const int block_h    = plane_index ? block_size>>s->chroma_v_shift : block_size;
     714           0 :     const uint8_t *obmc  = plane_index ? ff_obmc_tab[s->block_max_depth+s->chroma_h_shift] : ff_obmc_tab[s->block_max_depth];
     715           0 :     const int obmc_stride= plane_index ? (2*block_size)>>s->chroma_h_shift : 2*block_size;
     716           0 :     const int ref_stride= s->current_picture->linesize[plane_index];
     717           0 :     uint8_t *dst= s->current_picture->data[plane_index];
     718           0 :     uint8_t *src= s-> input_picture->data[plane_index];
     719             :     //FIXME zero_dst is const but add_yblock changes dst if add is 0 (this is never the case for dst=zero_dst
     720             :     // const has only been removed from zero_dst to suppress a warning
     721             :     static IDWTELEM zero_dst[4096]; //FIXME
     722           0 :     const int b_stride = s->b_width << s->block_max_depth;
     723           0 :     const int w= p->width;
     724           0 :     const int h= p->height;
     725           0 :     int distortion= 0;
     726           0 :     int rate= 0;
     727           0 :     const int penalty_factor= get_penalty_factor(s->lambda, s->lambda2, s->avctx->me_cmp);
     728             : 
     729             :     av_assert2(s->chroma_h_shift == s->chroma_v_shift); //obmc and square assumtions below
     730             : 
     731           0 :     for(i=0; i<9; i++){
     732           0 :         int mb_x2= mb_x + (i%3) - 1;
     733           0 :         int mb_y2= mb_y + (i/3) - 1;
     734           0 :         int x= block_w*mb_x2 + block_w/2;
     735           0 :         int y= block_h*mb_y2 + block_h/2;
     736             : 
     737           0 :         add_yblock(s, 0, NULL, zero_dst, dst, obmc,
     738             :                    x, y, block_w, block_h, w, h, /*dst_stride*/0, ref_stride, obmc_stride, mb_x2, mb_y2, 1, 1, plane_index);
     739             : 
     740             :         //FIXME find a cleaner/simpler way to skip the outside stuff
     741           0 :         for(y2= y; y2<0; y2++)
     742           0 :             memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, block_w);
     743           0 :         for(y2= h; y2<y+block_h; y2++)
     744           0 :             memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, block_w);
     745           0 :         if(x<0){
     746           0 :             for(y2= y; y2<y+block_h; y2++)
     747           0 :                 memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, -x);
     748             :         }
     749           0 :         if(x+block_w > w){
     750           0 :             for(y2= y; y2<y+block_h; y2++)
     751           0 :                 memcpy(dst + w + y2*ref_stride, src + w + y2*ref_stride, x+block_w - w);
     752             :         }
     753             : 
     754             :         av_assert1(block_w== 8 || block_w==16);
     755           0 :         distortion += s->mecc.me_cmp[block_w==8](&s->m, src + x + y*ref_stride, dst + x + y*ref_stride, ref_stride, block_h);
     756             :     }
     757             : 
     758           0 :     if(plane_index==0){
     759           0 :         BlockNode *b= &s->block[mb_x+mb_y*b_stride];
     760           0 :         int merged= same_block(b,b+1) && same_block(b,b+b_stride) && same_block(b,b+b_stride+1);
     761             : 
     762             : /* ..RRRr
     763             :  * .RXXx.
     764             :  * .RXXx.
     765             :  * rxxx.
     766             :  */
     767           0 :         if(merged)
     768           0 :             rate = get_block_bits(s, mb_x, mb_y, 2);
     769           0 :         for(i=merged?4:0; i<9; i++){
     770             :             static const int dxy[9][2] = {{0,0},{1,0},{0,1},{1,1},{2,0},{2,1},{-1,2},{0,2},{1,2}};
     771           0 :             rate += get_block_bits(s, mb_x + dxy[i][0], mb_y + dxy[i][1], 1);
     772             :         }
     773             :     }
     774           0 :     return distortion + rate*penalty_factor;
     775             : }
     776             : 
     777       21600 : static int encode_subband_c0run(SnowContext *s, SubBand *b, const IDWTELEM *src, const IDWTELEM *parent, int stride, int orientation){
     778       21600 :     const int w= b->width;
     779       21600 :     const int h= b->height;
     780             :     int x, y;
     781             : 
     782             :     if(1){
     783       21600 :         int run=0;
     784       21600 :         int *runs = s->run_buffer;
     785       21600 :         int run_index=0;
     786             :         int max_index;
     787             : 
     788      388200 :         for(y=0; y<h; y++){
     789    26862600 :             for(x=0; x<w; x++){
     790    26496000 :                 int v, p=0;
     791    26496000 :                 int /*ll=0, */l=0, lt=0, t=0, rt=0;
     792    26496000 :                 v= src[x + y*stride];
     793             : 
     794    26496000 :                 if(y){
     795    25960200 :                     t= src[x + (y-1)*stride];
     796    25960200 :                     if(x){
     797    25615200 :                         lt= src[x - 1 + (y-1)*stride];
     798             :                     }
     799    25960200 :                     if(x + 1 < w){
     800    25615200 :                         rt= src[x + 1 + (y-1)*stride];
     801             :                     }
     802             :                 }
     803    26496000 :                 if(x){
     804    26129400 :                     l= src[x - 1 + y*stride];
     805             :                     /*if(x > 1){
     806             :                         if(orientation==1) ll= src[y + (x-2)*stride];
     807             :                         else               ll= src[x - 2 + y*stride];
     808             :                     }*/
     809             :                 }
     810    26496000 :                 if(parent){
     811    26392500 :                     int px= x>>1;
     812    26392500 :                     int py= y>>1;
     813    26392500 :                     if(px<b->parent->width && py<b->parent->height)
     814    26380800 :                         p= parent[px + py*2*stride];
     815             :                 }
     816    26496000 :                 if(!(/*ll|*/l|lt|t|rt|p)){
     817     2820927 :                     if(v){
     818      291329 :                         runs[run_index++]= run;
     819      291329 :                         run=0;
     820             :                     }else{
     821     2529598 :                         run++;
     822             :                     }
     823             :                 }
     824             :             }
     825             :         }
     826       21600 :         max_index= run_index;
     827       21600 :         runs[run_index++]= run;
     828       21600 :         run_index=0;
     829       21600 :         run= runs[run_index++];
     830             : 
     831       21600 :         put_symbol2(&s->c, b->state[30], max_index, 0);
     832       21600 :         if(run_index <= max_index)
     833       16916 :             put_symbol2(&s->c, b->state[1], run, 3);
     834             : 
     835      388200 :         for(y=0; y<h; y++){
     836      366600 :             if(s->c.bytestream_end - s->c.bytestream < w*40){
     837           0 :                 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
     838           0 :                 return AVERROR(ENOMEM);
     839             :             }
     840    26862600 :             for(x=0; x<w; x++){
     841    26496000 :                 int v, p=0;
     842    26496000 :                 int /*ll=0, */l=0, lt=0, t=0, rt=0;
     843    26496000 :                 v= src[x + y*stride];
     844             : 
     845    26496000 :                 if(y){
     846    25960200 :                     t= src[x + (y-1)*stride];
     847    25960200 :                     if(x){
     848    25615200 :                         lt= src[x - 1 + (y-1)*stride];
     849             :                     }
     850    25960200 :                     if(x + 1 < w){
     851    25615200 :                         rt= src[x + 1 + (y-1)*stride];
     852             :                     }
     853             :                 }
     854    26496000 :                 if(x){
     855    26129400 :                     l= src[x - 1 + y*stride];
     856             :                     /*if(x > 1){
     857             :                         if(orientation==1) ll= src[y + (x-2)*stride];
     858             :                         else               ll= src[x - 2 + y*stride];
     859             :                     }*/
     860             :                 }
     861    26496000 :                 if(parent){
     862    26392500 :                     int px= x>>1;
     863    26392500 :                     int py= y>>1;
     864    26392500 :                     if(px<b->parent->width && py<b->parent->height)
     865    26380800 :                         p= parent[px + py*2*stride];
     866             :                 }
     867    26496000 :                 if(/*ll|*/l|lt|t|rt|p){
     868    23675073 :                     int context= av_log2(/*FFABS(ll) + */3*FFABS(l) + FFABS(lt) + 2*FFABS(t) + FFABS(rt) + FFABS(p));
     869             : 
     870    23675073 :                     put_rac(&s->c, &b->state[0][context], !!v);
     871             :                 }else{
     872     2820927 :                     if(!run){
     873      291329 :                         run= runs[run_index++];
     874             : 
     875      291329 :                         if(run_index <= max_index)
     876      274413 :                             put_symbol2(&s->c, b->state[1], run, 3);
     877             :                         av_assert2(v);
     878             :                     }else{
     879     2529598 :                         run--;
     880             :                         av_assert2(!v);
     881             :                     }
     882             :                 }
     883    26496000 :                 if(v){
     884    15993468 :                     int context= av_log2(/*FFABS(ll) + */3*FFABS(l) + FFABS(lt) + 2*FFABS(t) + FFABS(rt) + FFABS(p));
     885    15993468 :                     int l2= 2*FFABS(l) + (l<0);
     886    15993468 :                     int t2= 2*FFABS(t) + (t<0);
     887             : 
     888    15993468 :                     put_symbol2(&s->c, b->state[context + 2], FFABS(v)-1, context-4);
     889    15993468 :                     put_rac(&s->c, &b->state[0][16 + 1 + 3 + ff_quant3bA[l2&0xFF] + 3*ff_quant3bA[t2&0xFF]], v<0);
     890             :                 }
     891             :             }
     892             :         }
     893             :     }
     894       21600 :     return 0;
     895             : }
     896             : 
     897       21600 : static int encode_subband(SnowContext *s, SubBand *b, const IDWTELEM *src, const IDWTELEM *parent, int stride, int orientation){
     898             : //    encode_subband_qtree(s, b, src, parent, stride, orientation);
     899             : //    encode_subband_z0run(s, b, src, parent, stride, orientation);
     900       21600 :     return encode_subband_c0run(s, b, src, parent, stride, orientation);
     901             : //    encode_subband_dzr(s, b, src, parent, stride, orientation);
     902             : }
     903             : 
     904      435268 : static av_always_inline int check_block(SnowContext *s, int mb_x, int mb_y, int p[3], int intra, uint8_t (*obmc_edged)[MB_SIZE * 2], int *best_rd){
     905      435268 :     const int b_stride= s->b_width << s->block_max_depth;
     906      435268 :     BlockNode *block= &s->block[mb_x + mb_y * b_stride];
     907      435268 :     BlockNode backup= *block;
     908             :     unsigned value;
     909             :     int rd, index;
     910             : 
     911             :     av_assert2(mb_x>=0 && mb_y>=0);
     912             :     av_assert2(mb_x<b_stride);
     913             : 
     914      435268 :     if(intra){
     915       18870 :         block->color[0] = p[0];
     916       18870 :         block->color[1] = p[1];
     917       18870 :         block->color[2] = p[2];
     918       18870 :         block->type |= BLOCK_INTRA;
     919             :     }else{
     920      416398 :         index= (p[0] + 31*p[1]) & (ME_CACHE_SIZE-1);
     921      416398 :         value= s->me_cache_generation + (p[0]>>10) + (p[1]<<6) + (block->ref<<12);
     922      416398 :         if(s->me_cache[index] == value)
     923      118940 :             return 0;
     924      297458 :         s->me_cache[index]= value;
     925             : 
     926      297458 :         block->mx= p[0];
     927      297458 :         block->my= p[1];
     928      297458 :         block->type &= ~BLOCK_INTRA;
     929             :     }
     930             : 
     931      316328 :     rd= get_block_rd(s, mb_x, mb_y, 0, obmc_edged) + s->intra_penalty * !!intra;
     932             : 
     933             : //FIXME chroma
     934      316328 :     if(rd < *best_rd){
     935       65411 :         *best_rd= rd;
     936       65411 :         return 1;
     937             :     }else{
     938      250917 :         *block= backup;
     939      250917 :         return 0;
     940             :     }
     941             : }
     942             : 
     943             : /* special case for int[2] args we discard afterwards,
     944             :  * fixes compilation problem with gcc 2.95 */
     945      416398 : static av_always_inline int check_block_inter(SnowContext *s, int mb_x, int mb_y, int p0, int p1, uint8_t (*obmc_edged)[MB_SIZE * 2], int *best_rd){
     946      416398 :     int p[2] = {p0, p1};
     947      416398 :     return check_block(s, mb_x, mb_y, p, 0, obmc_edged, best_rd);
     948             : }
     949             : 
     950           0 : 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){
     951           0 :     const int b_stride= s->b_width << s->block_max_depth;
     952           0 :     BlockNode *block= &s->block[mb_x + mb_y * b_stride];
     953             :     BlockNode backup[4];
     954             :     unsigned value;
     955             :     int rd, index;
     956             : 
     957             :     /* We don't initialize backup[] during variable declaration, because
     958             :      * that fails to compile on MSVC: "cannot convert from 'BlockNode' to
     959             :      * 'int16_t'". */
     960           0 :     backup[0] = block[0];
     961           0 :     backup[1] = block[1];
     962           0 :     backup[2] = block[b_stride];
     963           0 :     backup[3] = block[b_stride + 1];
     964             : 
     965             :     av_assert2(mb_x>=0 && mb_y>=0);
     966             :     av_assert2(mb_x<b_stride);
     967             :     av_assert2(((mb_x|mb_y)&1) == 0);
     968             : 
     969           0 :     index= (p0 + 31*p1) & (ME_CACHE_SIZE-1);
     970           0 :     value= s->me_cache_generation + (p0>>10) + (p1<<6) + (block->ref<<12);
     971           0 :     if(s->me_cache[index] == value)
     972           0 :         return 0;
     973           0 :     s->me_cache[index]= value;
     974             : 
     975           0 :     block->mx= p0;
     976           0 :     block->my= p1;
     977           0 :     block->ref= ref;
     978           0 :     block->type &= ~BLOCK_INTRA;
     979           0 :     block[1]= block[b_stride]= block[b_stride+1]= *block;
     980             : 
     981           0 :     rd= get_4block_rd(s, mb_x, mb_y, 0);
     982             : 
     983             : //FIXME chroma
     984           0 :     if(rd < *best_rd){
     985           0 :         *best_rd= rd;
     986           0 :         return 1;
     987             :     }else{
     988           0 :         block[0]= backup[0];
     989           0 :         block[1]= backup[1];
     990           0 :         block[b_stride]= backup[2];
     991           0 :         block[b_stride+1]= backup[3];
     992           0 :         return 0;
     993             :     }
     994             : }
     995             : 
     996         270 : static void iterative_me(SnowContext *s){
     997             :     int pass, mb_x, mb_y;
     998         270 :     const int b_width = s->b_width  << s->block_max_depth;
     999         270 :     const int b_height= s->b_height << s->block_max_depth;
    1000         270 :     const int b_stride= b_width;
    1001             :     int color[3];
    1002             : 
    1003             :     {
    1004         270 :         RangeCoder r = s->c;
    1005             :         uint8_t state[sizeof(s->block_state)];
    1006         270 :         memcpy(state, s->block_state, sizeof(s->block_state));
    1007        1350 :         for(mb_y= 0; mb_y<s->b_height; mb_y++)
    1008        9720 :             for(mb_x= 0; mb_x<s->b_width; mb_x++)
    1009        8640 :                 encode_q_branch(s, 0, mb_x, mb_y);
    1010         270 :         s->c = r;
    1011         270 :         memcpy(s->block_state, state, sizeof(s->block_state));
    1012             :     }
    1013             : 
    1014         964 :     for(pass=0; pass<25; pass++){
    1015         964 :         int change= 0;
    1016             : 
    1017        4820 :         for(mb_y= 0; mb_y<b_height; mb_y++){
    1018       34704 :             for(mb_x= 0; mb_x<b_width; mb_x++){
    1019             :                 int dia_change, i, j, ref;
    1020       30848 :                 int best_rd= INT_MAX, ref_rd;
    1021             :                 BlockNode backup, ref_b;
    1022       30848 :                 const int index= mb_x + mb_y * b_stride;
    1023       30848 :                 BlockNode *block= &s->block[index];
    1024       30848 :                 BlockNode *tb =                   mb_y            ? &s->block[index-b_stride  ] : NULL;
    1025       30848 :                 BlockNode *lb = mb_x                              ? &s->block[index         -1] : NULL;
    1026       30848 :                 BlockNode *rb = mb_x+1<b_width                    ? &s->block[index         +1] : NULL;
    1027       30848 :                 BlockNode *bb =                   mb_y+1<b_height ? &s->block[index+b_stride  ] : NULL;
    1028       30848 :                 BlockNode *tlb= mb_x           && mb_y            ? &s->block[index-b_stride-1] : NULL;
    1029       30848 :                 BlockNode *trb= mb_x+1<b_width && mb_y            ? &s->block[index-b_stride+1] : NULL;
    1030       30848 :                 BlockNode *blb= mb_x           && mb_y+1<b_height ? &s->block[index+b_stride-1] : NULL;
    1031       30848 :                 BlockNode *brb= mb_x+1<b_width && mb_y+1<b_height ? &s->block[index+b_stride+1] : NULL;
    1032       30848 :                 const int b_w= (MB_SIZE >> s->block_max_depth);
    1033             :                 uint8_t obmc_edged[MB_SIZE * 2][MB_SIZE * 2];
    1034             : 
    1035       30848 :                 if(pass && (block->type & BLOCK_OPT))
    1036       12063 :                     continue;
    1037       18785 :                 block->type |= BLOCK_OPT;
    1038             : 
    1039       18785 :                 backup= *block;
    1040             : 
    1041       18785 :                 if(!s->me_cache_generation)
    1042          21 :                     memset(s->me_cache, 0, sizeof(s->me_cache));
    1043       18785 :                 s->me_cache_generation += 1<<22;
    1044             : 
    1045             :                 //FIXME precalculate
    1046             :                 {
    1047             :                     int x, y;
    1048      619905 :                     for (y = 0; y < b_w * 2; y++)
    1049      601120 :                         memcpy(obmc_edged[y], ff_obmc_tab[s->block_max_depth] + y * b_w * 2, b_w * 2);
    1050       18785 :                     if(mb_x==0)
    1051       73953 :                         for(y=0; y<b_w*2; y++)
    1052       71712 :                             memset(obmc_edged[y], obmc_edged[y][0] + obmc_edged[y][b_w-1], b_w);
    1053       18785 :                     if(mb_x==b_stride-1)
    1054       60951 :                         for(y=0; y<b_w*2; y++)
    1055       59104 :                             memset(obmc_edged[y]+b_w, obmc_edged[y][b_w] + obmc_edged[y][b_w*2-1], b_w);
    1056       18785 :                     if(mb_y==0){
    1057      164670 :                         for(x=0; x<b_w*2; x++)
    1058      159680 :                             obmc_edged[0][x] += obmc_edged[b_w-1][x];
    1059       79840 :                         for(y=1; y<b_w; y++)
    1060       74850 :                             memcpy(obmc_edged[y], obmc_edged[0], b_w*2);
    1061             :                     }
    1062       18785 :                     if(mb_y==b_height-1){
    1063      121770 :                         for(x=0; x<b_w*2; x++)
    1064      118080 :                             obmc_edged[b_w*2-1][x] += obmc_edged[b_w][x];
    1065       59040 :                         for(y=b_w; y<b_w*2-1; y++)
    1066       55350 :                             memcpy(obmc_edged[y], obmc_edged[b_w*2-1], b_w*2);
    1067             :                     }
    1068             :                 }
    1069             : 
    1070             :                 //skip stuff outside the picture
    1071       18785 :                 if(mb_x==0 || mb_y==0 || mb_x==b_width-1 || mb_y==b_height-1){
    1072       10919 :                     uint8_t *src= s->  input_picture->data[0];
    1073       10919 :                     uint8_t *dst= s->current_picture->data[0];
    1074       10919 :                     const int stride= s->current_picture->linesize[0];
    1075       10919 :                     const int block_w= MB_SIZE >> s->block_max_depth;
    1076       10919 :                     const int block_h= MB_SIZE >> s->block_max_depth;
    1077       10919 :                     const int sx= block_w*mb_x - block_w/2;
    1078       10919 :                     const int sy= block_h*mb_y - block_h/2;
    1079       10919 :                     const int w= s->plane[0].width;
    1080       10919 :                     const int h= s->plane[0].height;
    1081             :                     int y;
    1082             : 
    1083       50839 :                     for(y=sy; y<0; y++)
    1084       39920 :                         memcpy(dst + sx + y*stride, src + sx + y*stride, block_w*2);
    1085       40439 :                     for(y=h; y<sy+block_h*2; y++)
    1086       29520 :                         memcpy(dst + sx + y*stride, src + sx + y*stride, block_w*2);
    1087       10919 :                     if(sx<0){
    1088       73953 :                         for(y=sy; y<sy+block_h*2; y++)
    1089       71712 :                             memcpy(dst + sx + y*stride, src + sx + y*stride, -sx);
    1090             :                     }
    1091       10919 :                     if(sx+block_w*2 > w){
    1092       60951 :                         for(y=sy; y<sy+block_h*2; y++)
    1093       59104 :                             memcpy(dst + w + y*stride, src + w + y*stride, sx+block_w*2 - w);
    1094             :                     }
    1095             :                 }
    1096             : 
    1097             :                 // intra(black) = neighbors' contribution to the current block
    1098       75140 :                 for(i=0; i < s->nb_planes; i++)
    1099       56355 :                     color[i]= get_dc(s, mb_x, mb_y, i);
    1100             : 
    1101             :                 // get previous score (cannot be cached due to OBMC)
    1102       18870 :                 if(pass > 0 && (block->type&BLOCK_INTRA)){
    1103          85 :                     int color0[3]= {block->color[0], block->color[1], block->color[2]};
    1104          85 :                     check_block(s, mb_x, mb_y, color0, 1, obmc_edged, &best_rd);
    1105             :                 }else
    1106       18700 :                     check_block_inter(s, mb_x, mb_y, block->mx, block->my, obmc_edged, &best_rd);
    1107             : 
    1108       18785 :                 ref_b= *block;
    1109       18785 :                 ref_rd= best_rd;
    1110       37570 :                 for(ref=0; ref < s->ref_frames; ref++){
    1111       18785 :                     int16_t (*mvr)[2]= &s->ref_mvs[ref][index];
    1112       18785 :                     if(s->ref_scores[ref][index] > s->ref_scores[ref_b.ref][index]*3/2) //FIXME tune threshold
    1113           0 :                         continue;
    1114       18785 :                     block->ref= ref;
    1115       18785 :                     best_rd= INT_MAX;
    1116             : 
    1117       18785 :                     check_block_inter(s, mb_x, mb_y, mvr[0][0], mvr[0][1], obmc_edged, &best_rd);
    1118       18785 :                     check_block_inter(s, mb_x, mb_y, 0, 0, obmc_edged, &best_rd);
    1119       18785 :                     if(tb)
    1120       13795 :                         check_block_inter(s, mb_x, mb_y, mvr[-b_stride][0], mvr[-b_stride][1], obmc_edged, &best_rd);
    1121       18785 :                     if(lb)
    1122       16544 :                         check_block_inter(s, mb_x, mb_y, mvr[-1][0], mvr[-1][1], obmc_edged, &best_rd);
    1123       18785 :                     if(rb)
    1124       16938 :                         check_block_inter(s, mb_x, mb_y, mvr[1][0], mvr[1][1], obmc_edged, &best_rd);
    1125       18785 :                     if(bb)
    1126       15095 :                         check_block_inter(s, mb_x, mb_y, mvr[b_stride][0], mvr[b_stride][1], obmc_edged, &best_rd);
    1127             : 
    1128             :                     /* fullpel ME */
    1129             :                     //FIXME avoid subpel interpolation / round to nearest integer
    1130             :                     do{
    1131       19209 :                         int newx = block->mx;
    1132       19209 :                         int newy = block->my;
    1133       19209 :                         int dia_size = s->iterative_dia_size ? s->iterative_dia_size : FFMAX(s->avctx->dia_size, 1);
    1134       19209 :                         dia_change=0;
    1135       57627 :                         for(i=0; i < dia_size; i++){
    1136       57627 :                             for(j=0; j<i; j++){
    1137       19209 :                                 dia_change |= check_block_inter(s, mb_x, mb_y, newx+4*(i-j), newy+(4*j), obmc_edged, &best_rd);
    1138       19209 :                                 dia_change |= check_block_inter(s, mb_x, mb_y, newx-4*(i-j), newy-(4*j), obmc_edged, &best_rd);
    1139       19209 :                                 dia_change |= check_block_inter(s, mb_x, mb_y, newx-(4*j), newy+4*(i-j), obmc_edged, &best_rd);
    1140       19209 :                                 dia_change |= check_block_inter(s, mb_x, mb_y, newx+(4*j), newy-4*(i-j), obmc_edged, &best_rd);
    1141             :                             }
    1142             :                         }
    1143       19209 :                     }while(dia_change);
    1144             :                     /* subpel ME */
    1145             :                     do{
    1146             :                         static const int square[8][2]= {{+1, 0},{-1, 0},{ 0,+1},{ 0,-1},{+1,+1},{-1,-1},{+1,-1},{-1,+1},};
    1147       27615 :                         dia_change=0;
    1148      248535 :                         for(i=0; i<8; i++)
    1149      220920 :                             dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+square[i][0], block->my+square[i][1], obmc_edged, &best_rd);
    1150       27615 :                     }while(dia_change);
    1151             :                     //FIXME or try the standard 2 pass qpel or similar
    1152             : 
    1153       18785 :                     mvr[0][0]= block->mx;
    1154       18785 :                     mvr[0][1]= block->my;
    1155       18785 :                     if(ref_rd > best_rd){
    1156        4553 :                         ref_rd= best_rd;
    1157        4553 :                         ref_b= *block;
    1158             :                     }
    1159             :                 }
    1160       18785 :                 best_rd= ref_rd;
    1161       18785 :                 *block= ref_b;
    1162       18785 :                 check_block(s, mb_x, mb_y, color, 1, obmc_edged, &best_rd);
    1163             :                 //FIXME RD style color selection
    1164       18785 :                 if(!same_block(block, &backup)){
    1165        4573 :                     if(tb ) tb ->type &= ~BLOCK_OPT;
    1166        4573 :                     if(lb ) lb ->type &= ~BLOCK_OPT;
    1167        4573 :                     if(rb ) rb ->type &= ~BLOCK_OPT;
    1168        4573 :                     if(bb ) bb ->type &= ~BLOCK_OPT;
    1169        4573 :                     if(tlb) tlb->type &= ~BLOCK_OPT;
    1170        4573 :                     if(trb) trb->type &= ~BLOCK_OPT;
    1171        4573 :                     if(blb) blb->type &= ~BLOCK_OPT;
    1172        4573 :                     if(brb) brb->type &= ~BLOCK_OPT;
    1173        4573 :                     change ++;
    1174             :                 }
    1175             :             }
    1176             :         }
    1177         964 :         av_log(s->avctx, AV_LOG_DEBUG, "pass:%d changed:%d\n", pass, change);
    1178         964 :         if(!change)
    1179         270 :             break;
    1180             :     }
    1181             : 
    1182         270 :     if(s->block_max_depth == 1){
    1183           0 :         int change= 0;
    1184           0 :         for(mb_y= 0; mb_y<b_height; mb_y+=2){
    1185           0 :             for(mb_x= 0; mb_x<b_width; mb_x+=2){
    1186             :                 int i;
    1187             :                 int best_rd, init_rd;
    1188           0 :                 const int index= mb_x + mb_y * b_stride;
    1189             :                 BlockNode *b[4];
    1190             : 
    1191           0 :                 b[0]= &s->block[index];
    1192           0 :                 b[1]= b[0]+1;
    1193           0 :                 b[2]= b[0]+b_stride;
    1194           0 :                 b[3]= b[2]+1;
    1195           0 :                 if(same_block(b[0], b[1]) &&
    1196           0 :                    same_block(b[0], b[2]) &&
    1197           0 :                    same_block(b[0], b[3]))
    1198           0 :                     continue;
    1199             : 
    1200           0 :                 if(!s->me_cache_generation)
    1201           0 :                     memset(s->me_cache, 0, sizeof(s->me_cache));
    1202           0 :                 s->me_cache_generation += 1<<22;
    1203             : 
    1204           0 :                 init_rd= best_rd= get_4block_rd(s, mb_x, mb_y, 0);
    1205             : 
    1206             :                 //FIXME more multiref search?
    1207           0 :                 check_4block_inter(s, mb_x, mb_y,
    1208           0 :                                    (b[0]->mx + b[1]->mx + b[2]->mx + b[3]->mx + 2) >> 2,
    1209           0 :                                    (b[0]->my + b[1]->my + b[2]->my + b[3]->my + 2) >> 2, 0, &best_rd);
    1210             : 
    1211           0 :                 for(i=0; i<4; i++)
    1212           0 :                     if(!(b[i]->type&BLOCK_INTRA))
    1213           0 :                         check_4block_inter(s, mb_x, mb_y, b[i]->mx, b[i]->my, b[i]->ref, &best_rd);
    1214             : 
    1215           0 :                 if(init_rd != best_rd)
    1216           0 :                     change++;
    1217             :             }
    1218             :         }
    1219           0 :         av_log(s->avctx, AV_LOG_ERROR, "pass:4mv changed:%d\n", change*4);
    1220             :     }
    1221         270 : }
    1222             : 
    1223         510 : static void encode_blocks(SnowContext *s, int search){
    1224             :     int x, y;
    1225         510 :     int w= s->b_width;
    1226         510 :     int h= s->b_height;
    1227             : 
    1228         510 :     if(s->motion_est == FF_ME_ITER && !s->keyframe && search)
    1229         270 :         iterative_me(s);
    1230             : 
    1231        6570 :     for(y=0; y<h; y++){
    1232        6060 :         if(s->c.bytestream_end - s->c.bytestream < w*MB_SIZE*MB_SIZE*3){ //FIXME nicer limit
    1233           0 :             av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
    1234           0 :             return;
    1235             :         }
    1236      172260 :         for(x=0; x<w; x++){
    1237      166200 :             if(s->motion_est == FF_ME_ITER || !search)
    1238        9600 :                 encode_q_branch2(s, 0, x, y);
    1239             :             else
    1240      156600 :                 encode_q_branch (s, 0, x, y);
    1241             :         }
    1242             :     }
    1243             : }
    1244             : 
    1245       21600 : static void quantize(SnowContext *s, SubBand *b, IDWTELEM *dst, DWTELEM *src, int stride, int bias){
    1246       21600 :     const int w= b->width;
    1247       21600 :     const int h= b->height;
    1248       21600 :     const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16);
    1249       21600 :     const int qmul= ff_qexp[qlog&(QROOT-1)]<<((qlog>>QSHIFT) + ENCODER_EXTRA_BITS);
    1250             :     int x,y, thres1, thres2;
    1251             : 
    1252       21600 :     if(s->qlog == LOSSLESS_QLOG){
    1253      261000 :         for(y=0; y<h; y++)
    1254    23063400 :             for(x=0; x<w; x++)
    1255    22809600 :                 dst[x + y*stride]= src[x + y*stride];
    1256        7200 :         return;
    1257             :     }
    1258             : 
    1259       14400 :     bias= bias ? 0 : (3*qmul)>>3;
    1260       14400 :     thres1= ((qmul - bias)>>QEXPSHIFT) - 1;
    1261       14400 :     thres2= 2*thres1;
    1262             : 
    1263       14400 :     if(!bias){
    1264      114480 :         for(y=0; y<h; y++){
    1265     3419280 :             for(x=0; x<w; x++){
    1266     3317760 :                 int i= src[x + y*stride];
    1267             : 
    1268     3317760 :                 if((unsigned)(i+thres1) > thres2){
    1269      550125 :                     if(i>=0){
    1270      275359 :                         i<<= QEXPSHIFT;
    1271      275359 :                         i/= qmul; //FIXME optimize
    1272      275359 :                         dst[x + y*stride]=  i;
    1273             :                     }else{
    1274      274766 :                         i= -i;
    1275      274766 :                         i<<= QEXPSHIFT;
    1276      274766 :                         i/= qmul; //FIXME optimize
    1277      274766 :                         dst[x + y*stride]= -i;
    1278             :                     }
    1279             :                 }else
    1280     2767635 :                     dst[x + y*stride]= 0;
    1281             :             }
    1282             :         }
    1283             :     }else{
    1284       12720 :         for(y=0; y<h; y++){
    1285      379920 :             for(x=0; x<w; x++){
    1286      368640 :                 int i= src[x + y*stride];
    1287             : 
    1288      368640 :                 if((unsigned)(i+thres1) > thres2){
    1289      202458 :                     if(i>=0){
    1290      101688 :                         i<<= QEXPSHIFT;
    1291      101688 :                         i= (i + bias) / qmul; //FIXME optimize
    1292      101688 :                         dst[x + y*stride]=  i;
    1293             :                     }else{
    1294      100770 :                         i= -i;
    1295      100770 :                         i<<= QEXPSHIFT;
    1296      100770 :                         i= (i + bias) / qmul; //FIXME optimize
    1297      100770 :                         dst[x + y*stride]= -i;
    1298             :                     }
    1299             :                 }else
    1300      166182 :                     dst[x + y*stride]= 0;
    1301             :             }
    1302             :         }
    1303             :     }
    1304             : }
    1305             : 
    1306       21600 : static void dequantize(SnowContext *s, SubBand *b, IDWTELEM *src, int stride){
    1307       21600 :     const int w= b->width;
    1308       21600 :     const int h= b->height;
    1309       21600 :     const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16);
    1310       21600 :     const int qmul= ff_qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
    1311       21600 :     const int qadd= (s->qbias*qmul)>>QBIAS_SHIFT;
    1312             :     int x,y;
    1313             : 
    1314       21600 :     if(s->qlog == LOSSLESS_QLOG) return;
    1315             : 
    1316      127200 :     for(y=0; y<h; y++){
    1317     3799200 :         for(x=0; x<w; x++){
    1318     3686400 :             int i= src[x + y*stride];
    1319     3686400 :             if(i<0){
    1320      375486 :                 src[x + y*stride]= -((-i*qmul + qadd)>>(QEXPSHIFT)); //FIXME try different bias
    1321     3310914 :             }else if(i>0){
    1322      377019 :                 src[x + y*stride]=  (( i*qmul + qadd)>>(QEXPSHIFT));
    1323             :             }
    1324             :         }
    1325             :     }
    1326             : }
    1327             : 
    1328        1350 : static void decorrelate(SnowContext *s, SubBand *b, IDWTELEM *src, int stride, int inverse, int use_median){
    1329        1350 :     const int w= b->width;
    1330        1350 :     const int h= b->height;
    1331             :     int x,y;
    1332             : 
    1333        5400 :     for(y=h-1; y>=0; y--){
    1334       31500 :         for(x=w-1; x>=0; x--){
    1335       27450 :             int i= x + y*stride;
    1336             : 
    1337       27450 :             if(x){
    1338       23400 :                 if(use_median){
    1339           0 :                     if(y && x+1<w) src[i] -= mid_pred(src[i - 1], src[i - stride], src[i - stride + 1]);
    1340           0 :                     else  src[i] -= src[i - 1];
    1341             :                 }else{
    1342       23400 :                     if(y) src[i] -= mid_pred(src[i - 1], src[i - stride], src[i - 1] + src[i - stride] - src[i - 1 - stride]);
    1343        4500 :                     else  src[i] -= src[i - 1];
    1344             :                 }
    1345             :             }else{
    1346        4050 :                 if(y) src[i] -= src[i - stride];
    1347             :             }
    1348             :         }
    1349             :     }
    1350        1350 : }
    1351             : 
    1352        1350 : static void correlate(SnowContext *s, SubBand *b, IDWTELEM *src, int stride, int inverse, int use_median){
    1353        1350 :     const int w= b->width;
    1354        1350 :     const int h= b->height;
    1355             :     int x,y;
    1356             : 
    1357        5400 :     for(y=0; y<h; y++){
    1358       31500 :         for(x=0; x<w; x++){
    1359       27450 :             int i= x + y*stride;
    1360             : 
    1361       27450 :             if(x){
    1362       23400 :                 if(use_median){
    1363           0 :                     if(y && x+1<w) src[i] += mid_pred(src[i - 1], src[i - stride], src[i - stride + 1]);
    1364           0 :                     else  src[i] += src[i - 1];
    1365             :                 }else{
    1366       23400 :                     if(y) src[i] += mid_pred(src[i - 1], src[i - stride], src[i - 1] + src[i - stride] - src[i - 1 - stride]);
    1367        4500 :                     else  src[i] += src[i - 1];
    1368             :                 }
    1369             :             }else{
    1370        4050 :                 if(y) src[i] += src[i - stride];
    1371             :             }
    1372             :         }
    1373             :     }
    1374        1350 : }
    1375             : 
    1376          47 : static void encode_qlogs(SnowContext *s){
    1377             :     int plane_index, level, orientation;
    1378             : 
    1379         141 :     for(plane_index=0; plane_index<FFMIN(s->nb_planes, 2); plane_index++){
    1380         564 :         for(level=0; level<s->spatial_decomposition_count; level++){
    1381        1974 :             for(orientation=level ? 1:0; orientation<4; orientation++){
    1382        1504 :                 if(orientation==2) continue;
    1383        1034 :                 put_symbol(&s->c, s->header_state, s->plane[plane_index].band[level][orientation].qlog, 1);
    1384             :             }
    1385             :         }
    1386             :     }
    1387          47 : }
    1388             : 
    1389         510 : static void encode_header(SnowContext *s){
    1390             :     int plane_index, i;
    1391             :     uint8_t kstate[32];
    1392             : 
    1393         510 :     memset(kstate, MID_STATE, sizeof(kstate));
    1394             : 
    1395         510 :     put_rac(&s->c, kstate, s->keyframe);
    1396         510 :     if(s->keyframe || s->always_reset){
    1397          47 :         ff_snow_reset_contexts(s);
    1398          47 :         s->last_spatial_decomposition_type=
    1399          47 :         s->last_qlog=
    1400          47 :         s->last_qbias=
    1401          47 :         s->last_mv_scale=
    1402          47 :         s->last_block_max_depth= 0;
    1403         141 :         for(plane_index=0; plane_index<2; plane_index++){
    1404          94 :             Plane *p= &s->plane[plane_index];
    1405          94 :             p->last_htaps=0;
    1406          94 :             p->last_diag_mc=0;
    1407          94 :             memset(p->last_hcoeff, 0, sizeof(p->last_hcoeff));
    1408             :         }
    1409             :     }
    1410         510 :     if(s->keyframe){
    1411          47 :         put_symbol(&s->c, s->header_state, s->version, 0);
    1412          47 :         put_rac(&s->c, s->header_state, s->always_reset);
    1413          47 :         put_symbol(&s->c, s->header_state, s->temporal_decomposition_type, 0);
    1414          47 :         put_symbol(&s->c, s->header_state, s->temporal_decomposition_count, 0);
    1415          47 :         put_symbol(&s->c, s->header_state, s->spatial_decomposition_count, 0);
    1416          47 :         put_symbol(&s->c, s->header_state, s->colorspace_type, 0);
    1417          47 :         if (s->nb_planes > 2) {
    1418          47 :             put_symbol(&s->c, s->header_state, s->chroma_h_shift, 0);
    1419          47 :             put_symbol(&s->c, s->header_state, s->chroma_v_shift, 0);
    1420             :         }
    1421          47 :         put_rac(&s->c, s->header_state, s->spatial_scalability);
    1422             : //        put_rac(&s->c, s->header_state, s->rate_scalability);
    1423          47 :         put_symbol(&s->c, s->header_state, s->max_ref_frames-1, 0);
    1424             : 
    1425          47 :         encode_qlogs(s);
    1426             :     }
    1427             : 
    1428         510 :     if(!s->keyframe){
    1429         463 :         int update_mc=0;
    1430        1389 :         for(plane_index=0; plane_index<FFMIN(s->nb_planes, 2); plane_index++){
    1431         926 :             Plane *p= &s->plane[plane_index];
    1432         926 :             update_mc |= p->last_htaps   != p->htaps;
    1433         926 :             update_mc |= p->last_diag_mc != p->diag_mc;
    1434         926 :             update_mc |= !!memcmp(p->last_hcoeff, p->hcoeff, sizeof(p->hcoeff));
    1435             :         }
    1436         463 :         put_rac(&s->c, s->header_state, update_mc);
    1437         463 :         if(update_mc){
    1438         141 :             for(plane_index=0; plane_index<FFMIN(s->nb_planes, 2); plane_index++){
    1439          94 :                 Plane *p= &s->plane[plane_index];
    1440          94 :                 put_rac(&s->c, s->header_state, p->diag_mc);
    1441          94 :                 put_symbol(&s->c, s->header_state, p->htaps/2-1, 0);
    1442         376 :                 for(i= p->htaps/2; i; i--)
    1443         282 :                     put_symbol(&s->c, s->header_state, FFABS(p->hcoeff[i]), 0);
    1444             :             }
    1445             :         }
    1446         463 :         if(s->last_spatial_decomposition_count != s->spatial_decomposition_count){
    1447           0 :             put_rac(&s->c, s->header_state, 1);
    1448           0 :             put_symbol(&s->c, s->header_state, s->spatial_decomposition_count, 0);
    1449           0 :             encode_qlogs(s);
    1450             :         }else
    1451         463 :             put_rac(&s->c, s->header_state, 0);
    1452             :     }
    1453             : 
    1454         510 :     put_symbol(&s->c, s->header_state, s->spatial_decomposition_type - s->last_spatial_decomposition_type, 1);
    1455         510 :     put_symbol(&s->c, s->header_state, s->qlog            - s->last_qlog    , 1);
    1456         510 :     put_symbol(&s->c, s->header_state, s->mv_scale        - s->last_mv_scale, 1);
    1457         510 :     put_symbol(&s->c, s->header_state, s->qbias           - s->last_qbias   , 1);
    1458         510 :     put_symbol(&s->c, s->header_state, s->block_max_depth - s->last_block_max_depth, 1);
    1459             : 
    1460         510 : }
    1461             : 
    1462         510 : static void update_last_header_values(SnowContext *s){
    1463             :     int plane_index;
    1464             : 
    1465         510 :     if(!s->keyframe){
    1466        1389 :         for(plane_index=0; plane_index<2; plane_index++){
    1467         926 :             Plane *p= &s->plane[plane_index];
    1468         926 :             p->last_diag_mc= p->diag_mc;
    1469         926 :             p->last_htaps  = p->htaps;
    1470         926 :             memcpy(p->last_hcoeff, p->hcoeff, sizeof(p->hcoeff));
    1471             :         }
    1472             :     }
    1473             : 
    1474         510 :     s->last_spatial_decomposition_type  = s->spatial_decomposition_type;
    1475         510 :     s->last_qlog                        = s->qlog;
    1476         510 :     s->last_qbias                       = s->qbias;
    1477         510 :     s->last_mv_scale                    = s->mv_scale;
    1478         510 :     s->last_block_max_depth             = s->block_max_depth;
    1479         510 :     s->last_spatial_decomposition_count = s->spatial_decomposition_count;
    1480         510 : }
    1481             : 
    1482         360 : static int qscale2qlog(int qscale){
    1483         360 :     return lrint(QROOT*log2(qscale / (float)FF_QP2LAMBDA))
    1484         360 :            + 61*QROOT/8; ///< 64 > 60
    1485             : }
    1486             : 
    1487           0 : static int ratecontrol_1pass(SnowContext *s, AVFrame *pict)
    1488             : {
    1489             :     /* Estimate the frame's complexity as a sum of weighted dwt coefficients.
    1490             :      * FIXME we know exact mv bits at this point,
    1491             :      * but ratecontrol isn't set up to include them. */
    1492           0 :     uint32_t coef_sum= 0;
    1493             :     int level, orientation, delta_qlog;
    1494             : 
    1495           0 :     for(level=0; level<s->spatial_decomposition_count; level++){
    1496           0 :         for(orientation=level ? 1 : 0; orientation<4; orientation++){
    1497           0 :             SubBand *b= &s->plane[0].band[level][orientation];
    1498           0 :             IDWTELEM *buf= b->ibuf;
    1499           0 :             const int w= b->width;
    1500           0 :             const int h= b->height;
    1501           0 :             const int stride= b->stride;
    1502           0 :             const int qlog= av_clip(2*QROOT + b->qlog, 0, QROOT*16);
    1503           0 :             const int qmul= ff_qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
    1504           0 :             const int qdiv= (1<<16)/qmul;
    1505             :             int x, y;
    1506             :             //FIXME this is ugly
    1507           0 :             for(y=0; y<h; y++)
    1508           0 :                 for(x=0; x<w; x++)
    1509           0 :                     buf[x+y*stride]= b->buf[x+y*stride];
    1510           0 :             if(orientation==0)
    1511           0 :                 decorrelate(s, b, buf, stride, 1, 0);
    1512           0 :             for(y=0; y<h; y++)
    1513           0 :                 for(x=0; x<w; x++)
    1514           0 :                     coef_sum+= abs(buf[x+y*stride]) * qdiv >> 16;
    1515             :         }
    1516             :     }
    1517             : 
    1518             :     /* ugly, ratecontrol just takes a sqrt again */
    1519           0 :     av_assert0(coef_sum < INT_MAX);
    1520           0 :     coef_sum = (uint64_t)coef_sum * coef_sum >> 16;
    1521             : 
    1522           0 :     if(pict->pict_type == AV_PICTURE_TYPE_I){
    1523           0 :         s->m.current_picture.mb_var_sum= coef_sum;
    1524           0 :         s->m.current_picture.mc_mb_var_sum= 0;
    1525             :     }else{
    1526           0 :         s->m.current_picture.mc_mb_var_sum= coef_sum;
    1527           0 :         s->m.current_picture.mb_var_sum= 0;
    1528             :     }
    1529             : 
    1530           0 :     pict->quality= ff_rate_estimate_qscale(&s->m, 1);
    1531           0 :     if (pict->quality < 0)
    1532           0 :         return INT_MIN;
    1533           0 :     s->lambda= pict->quality * 3/2;
    1534           0 :     delta_qlog= qscale2qlog(pict->quality) - s->qlog;
    1535           0 :     s->qlog+= delta_qlog;
    1536           0 :     return delta_qlog;
    1537             : }
    1538             : 
    1539          33 : static void calculate_visual_weight(SnowContext *s, Plane *p){
    1540          33 :     int width = p->width;
    1541          33 :     int height= p->height;
    1542             :     int level, orientation, x, y;
    1543             : 
    1544         198 :     for(level=0; level<s->spatial_decomposition_count; level++){
    1545         693 :         for(orientation=level ? 1 : 0; orientation<4; orientation++){
    1546         528 :             SubBand *b= &p->band[level][orientation];
    1547         528 :             IDWTELEM *ibuf= b->ibuf;
    1548         528 :             int64_t error=0;
    1549             : 
    1550         528 :             memset(s->spatial_idwt_buffer, 0, sizeof(*s->spatial_idwt_buffer)*width*height);
    1551         528 :             ibuf[b->width/2 + b->height/2*b->stride]= 256*16;
    1552         528 :             ff_spatial_idwt(s->spatial_idwt_buffer, s->temp_idwt_buffer, width, height, width, s->spatial_decomposition_type, s->spatial_decomposition_count);
    1553       77328 :             for(y=0; y<height; y++){
    1554    28462080 :                 for(x=0; x<width; x++){
    1555    28385280 :                     int64_t d= s->spatial_idwt_buffer[x + y*width]*16;
    1556    28385280 :                     error += d*d;
    1557             :                 }
    1558             :             }
    1559             : 
    1560         528 :             b->qlog= (int)(QROOT * log2(352256.0/sqrt(error)) + 0.5);
    1561             :         }
    1562             :     }
    1563          33 : }
    1564             : 
    1565         510 : static int encode_frame(AVCodecContext *avctx, AVPacket *pkt,
    1566             :                         const AVFrame *pict, int *got_packet)
    1567             : {
    1568         510 :     SnowContext *s = avctx->priv_data;
    1569         510 :     RangeCoder * const c= &s->c;
    1570             :     AVFrame *pic;
    1571         510 :     const int width= s->avctx->width;
    1572         510 :     const int height= s->avctx->height;
    1573             :     int level, orientation, plane_index, i, y, ret;
    1574             :     uint8_t rc_header_bak[sizeof(s->header_state)];
    1575             :     uint8_t rc_block_bak[sizeof(s->block_state)];
    1576             : 
    1577         510 :     if ((ret = ff_alloc_packet2(avctx, pkt, s->b_width*s->b_height*MB_SIZE*MB_SIZE*3 + AV_INPUT_BUFFER_MIN_SIZE, 0)) < 0)
    1578           0 :         return ret;
    1579             : 
    1580         510 :     ff_init_range_encoder(c, pkt->data, pkt->size);
    1581         510 :     ff_build_rac_states(c, (1LL<<32)/20, 256-8);
    1582             : 
    1583        2040 :     for(i=0; i < s->nb_planes; i++){
    1584        1530 :         int hshift= i ? s->chroma_h_shift : 0;
    1585        1530 :         int vshift= i ? s->chroma_v_shift : 0;
    1586      195450 :         for(y=0; y<AV_CEIL_RSHIFT(height, vshift); y++)
    1587      387840 :             memcpy(&s->input_picture->data[i][y * s->input_picture->linesize[i]],
    1588      193920 :                    &pict->data[i][y * pict->linesize[i]],
    1589      193920 :                    AV_CEIL_RSHIFT(width, hshift));
    1590        4590 :         s->mpvencdsp.draw_edges(s->input_picture->data[i], s->input_picture->linesize[i],
    1591        3060 :                                 AV_CEIL_RSHIFT(width, hshift), AV_CEIL_RSHIFT(height, vshift),
    1592             :                                 EDGE_WIDTH >> hshift, EDGE_WIDTH >> vshift,
    1593             :                                 EDGE_TOP | EDGE_BOTTOM);
    1594             : 
    1595             :     }
    1596         510 :     emms_c();
    1597         510 :     pic = s->input_picture;
    1598         510 :     pic->pict_type = pict->pict_type;
    1599         510 :     pic->quality = pict->quality;
    1600             : 
    1601         510 :     s->m.picture_number= avctx->frame_number;
    1602         510 :     if(avctx->flags&AV_CODEC_FLAG_PASS2){
    1603           0 :         s->m.pict_type = pic->pict_type = s->m.rc_context.entry[avctx->frame_number].new_pict_type;
    1604           0 :         s->keyframe = pic->pict_type == AV_PICTURE_TYPE_I;
    1605           0 :         if(!(avctx->flags&AV_CODEC_FLAG_QSCALE)) {
    1606           0 :             pic->quality = ff_rate_estimate_qscale(&s->m, 0);
    1607           0 :             if (pic->quality < 0)
    1608           0 :                 return -1;
    1609             :         }
    1610             :     }else{
    1611         510 :         s->keyframe= avctx->gop_size==0 || avctx->frame_number % avctx->gop_size == 0;
    1612         510 :         s->m.pict_type = pic->pict_type = s->keyframe ? AV_PICTURE_TYPE_I : AV_PICTURE_TYPE_P;
    1613             :     }
    1614             : 
    1615         510 :     if(s->pass1_rc && avctx->frame_number == 0)
    1616           0 :         pic->quality = 2*FF_QP2LAMBDA;
    1617         510 :     if (pic->quality) {
    1618         360 :         s->qlog   = qscale2qlog(pic->quality);
    1619         360 :         s->lambda = pic->quality * 3/2;
    1620             :     }
    1621         510 :     if (s->qlog < 0 || (!pic->quality && (avctx->flags & AV_CODEC_FLAG_QSCALE))) {
    1622         150 :         s->qlog= LOSSLESS_QLOG;
    1623         150 :         s->lambda = 0;
    1624             :     }//else keep previous frame's qlog until after motion estimation
    1625             : 
    1626         510 :     if (s->current_picture->data[0]) {
    1627         499 :         int w = s->avctx->width;
    1628         499 :         int h = s->avctx->height;
    1629             : 
    1630         998 :         s->mpvencdsp.draw_edges(s->current_picture->data[0],
    1631         499 :                                 s->current_picture->linesize[0], w   , h   ,
    1632             :                                 EDGE_WIDTH  , EDGE_WIDTH  , EDGE_TOP | EDGE_BOTTOM);
    1633         499 :         if (s->current_picture->data[2]) {
    1634        2994 :             s->mpvencdsp.draw_edges(s->current_picture->data[1],
    1635        1497 :                                     s->current_picture->linesize[1], w>>s->chroma_h_shift, h>>s->chroma_v_shift,
    1636         998 :                                     EDGE_WIDTH>>s->chroma_h_shift, EDGE_WIDTH>>s->chroma_v_shift, EDGE_TOP | EDGE_BOTTOM);
    1637        2994 :             s->mpvencdsp.draw_edges(s->current_picture->data[2],
    1638        1497 :                                     s->current_picture->linesize[2], w>>s->chroma_h_shift, h>>s->chroma_v_shift,
    1639         998 :                                     EDGE_WIDTH>>s->chroma_h_shift, EDGE_WIDTH>>s->chroma_v_shift, EDGE_TOP | EDGE_BOTTOM);
    1640             :         }
    1641         499 :         emms_c();
    1642             :     }
    1643             : 
    1644         510 :     ff_snow_frame_start(s);
    1645             : #if FF_API_CODED_FRAME
    1646             : FF_DISABLE_DEPRECATION_WARNINGS
    1647         510 :     av_frame_unref(avctx->coded_frame);
    1648         510 :     ret = av_frame_ref(avctx->coded_frame, s->current_picture);
    1649             : FF_ENABLE_DEPRECATION_WARNINGS
    1650             : #endif
    1651         510 :     if (ret < 0)
    1652           0 :         return ret;
    1653             : 
    1654         510 :     s->m.current_picture_ptr= &s->m.current_picture;
    1655         510 :     s->m.current_picture.f = s->current_picture;
    1656         510 :     s->m.current_picture.f->pts = pict->pts;
    1657         510 :     if(pic->pict_type == AV_PICTURE_TYPE_P){
    1658         463 :         int block_width = (width +15)>>4;
    1659         463 :         int block_height= (height+15)>>4;
    1660         463 :         int stride= s->current_picture->linesize[0];
    1661             : 
    1662         463 :         av_assert0(s->current_picture->data[0]);
    1663         463 :         av_assert0(s->last_picture[0]->data[0]);
    1664             : 
    1665         463 :         s->m.avctx= s->avctx;
    1666         463 :         s->m.   last_picture.f = s->last_picture[0];
    1667         463 :         s->m.    new_picture.f = s->input_picture;
    1668         463 :         s->m.   last_picture_ptr= &s->m.   last_picture;
    1669         463 :         s->m.linesize = stride;
    1670         463 :         s->m.uvlinesize= s->current_picture->linesize[1];
    1671         463 :         s->m.width = width;
    1672         463 :         s->m.height= height;
    1673         463 :         s->m.mb_width = block_width;
    1674         463 :         s->m.mb_height= block_height;
    1675         463 :         s->m.mb_stride=   s->m.mb_width+1;
    1676         463 :         s->m.b8_stride= 2*s->m.mb_width+1;
    1677         463 :         s->m.f_code=1;
    1678         463 :         s->m.pict_type = pic->pict_type;
    1679         463 :         s->m.motion_est= s->motion_est;
    1680         463 :         s->m.me.scene_change_score=0;
    1681         463 :         s->m.me.dia_size = avctx->dia_size;
    1682         463 :         s->m.quarter_sample= (s->avctx->flags & AV_CODEC_FLAG_QPEL)!=0;
    1683         463 :         s->m.out_format= FMT_H263;
    1684         463 :         s->m.unrestricted_mv= 1;
    1685             : 
    1686         463 :         s->m.lambda = s->lambda;
    1687         463 :         s->m.qscale= (s->m.lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
    1688         463 :         s->lambda2= s->m.lambda2= (s->m.lambda*s->m.lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT;
    1689             : 
    1690         463 :         s->m.mecc= s->mecc; //move
    1691         463 :         s->m.qdsp= s->qdsp; //move
    1692         463 :         s->m.hdsp = s->hdsp;
    1693         463 :         ff_init_me(&s->m);
    1694         463 :         s->hdsp = s->m.hdsp;
    1695         463 :         s->mecc= s->m.mecc;
    1696             :     }
    1697             : 
    1698         510 :     if(s->pass1_rc){
    1699           0 :         memcpy(rc_header_bak, s->header_state, sizeof(s->header_state));
    1700           0 :         memcpy(rc_block_bak, s->block_state, sizeof(s->block_state));
    1701             :     }
    1702             : 
    1703        1020 : redo_frame:
    1704             : 
    1705         510 :     s->spatial_decomposition_count= 5;
    1706             : 
    1707        1020 :     while(   !(width >>(s->chroma_h_shift + s->spatial_decomposition_count))
    1708         510 :           || !(height>>(s->chroma_v_shift + s->spatial_decomposition_count)))
    1709           0 :         s->spatial_decomposition_count--;
    1710             : 
    1711         510 :     if (s->spatial_decomposition_count <= 0) {
    1712           0 :         av_log(avctx, AV_LOG_ERROR, "Resolution too low\n");
    1713           0 :         return AVERROR(EINVAL);
    1714             :     }
    1715             : 
    1716         510 :     s->m.pict_type = pic->pict_type;
    1717         510 :     s->qbias = pic->pict_type == AV_PICTURE_TYPE_P ? 2 : 0;
    1718             : 
    1719         510 :     ff_snow_common_init_after_header(avctx);
    1720             : 
    1721         510 :     if(s->last_spatial_decomposition_count != s->spatial_decomposition_count){
    1722          44 :         for(plane_index=0; plane_index < s->nb_planes; plane_index++){
    1723          33 :             calculate_visual_weight(s, &s->plane[plane_index]);
    1724             :         }
    1725             :     }
    1726             : 
    1727         510 :     encode_header(s);
    1728         510 :     s->m.misc_bits = 8*(s->c.bytestream - s->c.bytestream_start);
    1729         510 :     encode_blocks(s, 1);
    1730         510 :     s->m.mv_bits = 8*(s->c.bytestream - s->c.bytestream_start) - s->m.misc_bits;
    1731             : 
    1732        2040 :     for(plane_index=0; plane_index < s->nb_planes; plane_index++){
    1733        1530 :         Plane *p= &s->plane[plane_index];
    1734        1530 :         int w= p->width;
    1735        1530 :         int h= p->height;
    1736             :         int x, y;
    1737             : //        int bits= put_bits_count(&s->c.pb);
    1738             : 
    1739        1530 :         if (!s->memc_only) {
    1740             :             //FIXME optimize
    1741        1350 :             if(pict->data[plane_index]) //FIXME gray hack
    1742      126150 :                 for(y=0; y<h; y++){
    1743    26620800 :                     for(x=0; x<w; x++){
    1744    26496000 :                         s->spatial_idwt_buffer[y*w + x]= pict->data[plane_index][y*pict->linesize[plane_index] + x]<<FRAC_BITS;
    1745             :                     }
    1746             :                 }
    1747        1350 :             predict_plane(s, s->spatial_idwt_buffer, plane_index, 0);
    1748             : 
    1749             : #if FF_API_PRIVATE_OPT
    1750             : FF_DISABLE_DEPRECATION_WARNINGS
    1751        1350 :             if(s->avctx->scenechange_threshold)
    1752           0 :                 s->scenechange_threshold = s->avctx->scenechange_threshold;
    1753             : FF_ENABLE_DEPRECATION_WARNINGS
    1754             : #endif
    1755             : 
    1756        1350 :             if(   plane_index==0
    1757         450 :                && pic->pict_type == AV_PICTURE_TYPE_P
    1758         405 :                && !(avctx->flags&AV_CODEC_FLAG_PASS2)
    1759         405 :                && s->m.me.scene_change_score > s->scenechange_threshold){
    1760           0 :                 ff_init_range_encoder(c, pkt->data, pkt->size);
    1761           0 :                 ff_build_rac_states(c, (1LL<<32)/20, 256-8);
    1762           0 :                 pic->pict_type= AV_PICTURE_TYPE_I;
    1763           0 :                 s->keyframe=1;
    1764           0 :                 s->current_picture->key_frame=1;
    1765           0 :                 goto redo_frame;
    1766             :             }
    1767             : 
    1768        1350 :             if(s->qlog == LOSSLESS_QLOG){
    1769       86850 :                 for(y=0; y<h; y++){
    1770    22896000 :                     for(x=0; x<w; x++){
    1771    22809600 :                         s->spatial_dwt_buffer[y*w + x]= (s->spatial_idwt_buffer[y*w + x] + (1<<(FRAC_BITS-1))-1)>>FRAC_BITS;
    1772             :                     }
    1773             :                 }
    1774             :             }else{
    1775       39300 :                 for(y=0; y<h; y++){
    1776     3724800 :                     for(x=0; x<w; x++){
    1777     3686400 :                         s->spatial_dwt_buffer[y*w + x]=s->spatial_idwt_buffer[y*w + x]<<ENCODER_EXTRA_BITS;
    1778             :                     }
    1779             :                 }
    1780             :             }
    1781             : 
    1782        1350 :             ff_spatial_dwt(s->spatial_dwt_buffer, s->temp_dwt_buffer, w, h, w, s->spatial_decomposition_type, s->spatial_decomposition_count);
    1783             : 
    1784        1350 :             if(s->pass1_rc && plane_index==0){
    1785           0 :                 int delta_qlog = ratecontrol_1pass(s, pic);
    1786           0 :                 if (delta_qlog <= INT_MIN)
    1787           0 :                     return -1;
    1788           0 :                 if(delta_qlog){
    1789             :                     //reordering qlog in the bitstream would eliminate this reset
    1790           0 :                     ff_init_range_encoder(c, pkt->data, pkt->size);
    1791           0 :                     memcpy(s->header_state, rc_header_bak, sizeof(s->header_state));
    1792           0 :                     memcpy(s->block_state, rc_block_bak, sizeof(s->block_state));
    1793           0 :                     encode_header(s);
    1794           0 :                     encode_blocks(s, 0);
    1795             :                 }
    1796             :             }
    1797             : 
    1798        8100 :             for(level=0; level<s->spatial_decomposition_count; level++){
    1799       28350 :                 for(orientation=level ? 1 : 0; orientation<4; orientation++){
    1800       21600 :                     SubBand *b= &p->band[level][orientation];
    1801             : 
    1802       21600 :                     quantize(s, b, b->ibuf, b->buf, b->stride, s->qbias);
    1803       21600 :                     if(orientation==0)
    1804        1350 :                         decorrelate(s, b, b->ibuf, b->stride, pic->pict_type == AV_PICTURE_TYPE_P, 0);
    1805       21600 :                     if (!s->no_bitstream)
    1806       21600 :                     encode_subband(s, b, b->ibuf, b->parent ? b->parent->ibuf : NULL, b->stride, orientation);
    1807       21600 :                     av_assert0(b->parent==NULL || b->parent->stride == b->stride*2);
    1808       21600 :                     if(orientation==0)
    1809        1350 :                         correlate(s, b, b->ibuf, b->stride, 1, 0);
    1810             :                 }
    1811             :             }
    1812             : 
    1813        8100 :             for(level=0; level<s->spatial_decomposition_count; level++){
    1814       28350 :                 for(orientation=level ? 1 : 0; orientation<4; orientation++){
    1815       21600 :                     SubBand *b= &p->band[level][orientation];
    1816             : 
    1817       21600 :                     dequantize(s, b, b->ibuf, b->stride);
    1818             :                 }
    1819             :             }
    1820             : 
    1821        1350 :             ff_spatial_idwt(s->spatial_idwt_buffer, s->temp_idwt_buffer, w, h, w, s->spatial_decomposition_type, s->spatial_decomposition_count);
    1822        1350 :             if(s->qlog == LOSSLESS_QLOG){
    1823       86850 :                 for(y=0; y<h; y++){
    1824    22896000 :                     for(x=0; x<w; x++){
    1825    22809600 :                         s->spatial_idwt_buffer[y*w + x]<<=FRAC_BITS;
    1826             :                     }
    1827             :                 }
    1828             :             }
    1829        1350 :             predict_plane(s, s->spatial_idwt_buffer, plane_index, 1);
    1830             :         }else{
    1831             :             //ME/MC only
    1832         180 :             if(pic->pict_type == AV_PICTURE_TYPE_I){
    1833        2310 :                 for(y=0; y<h; y++){
    1834     1246464 :                     for(x=0; x<w; x++){
    1835     2488320 :                         s->current_picture->data[plane_index][y*s->current_picture->linesize[plane_index] + x]=
    1836     2488320 :                             pict->data[plane_index][y*pict->linesize[plane_index] + x];
    1837             :                     }
    1838             :                 }
    1839             :             }else{
    1840         174 :                 memset(s->spatial_idwt_buffer, 0, sizeof(IDWTELEM)*w*h);
    1841         174 :                 predict_plane(s, s->spatial_idwt_buffer, plane_index, 1);
    1842             :             }
    1843             :         }
    1844        1530 :         if(s->avctx->flags&AV_CODEC_FLAG_PSNR){
    1845           0 :             int64_t error= 0;
    1846             : 
    1847           0 :             if(pict->data[plane_index]) //FIXME gray hack
    1848           0 :                 for(y=0; y<h; y++){
    1849           0 :                     for(x=0; x<w; x++){
    1850           0 :                         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];
    1851           0 :                         error += d*d;
    1852             :                     }
    1853             :                 }
    1854           0 :             s->avctx->error[plane_index] += error;
    1855           0 :             s->encoding_error[plane_index] = error;
    1856             :         }
    1857             : 
    1858             :     }
    1859         510 :     emms_c();
    1860             : 
    1861         510 :     update_last_header_values(s);
    1862             : 
    1863         510 :     ff_snow_release_buffer(avctx);
    1864             : 
    1865         510 :     s->current_picture->coded_picture_number = avctx->frame_number;
    1866         510 :     s->current_picture->pict_type = pic->pict_type;
    1867         510 :     s->current_picture->quality = pic->quality;
    1868         510 :     s->m.frame_bits = 8*(s->c.bytestream - s->c.bytestream_start);
    1869         510 :     s->m.p_tex_bits = s->m.frame_bits - s->m.misc_bits - s->m.mv_bits;
    1870        1020 :     s->m.current_picture.f->display_picture_number =
    1871        1020 :     s->m.current_picture.f->coded_picture_number   = avctx->frame_number;
    1872         510 :     s->m.current_picture.f->quality                = pic->quality;
    1873         510 :     s->m.total_bits += 8*(s->c.bytestream - s->c.bytestream_start);
    1874         510 :     if(s->pass1_rc)
    1875           0 :         if (ff_rate_estimate_qscale(&s->m, 0) < 0)
    1876           0 :             return -1;
    1877         510 :     if(avctx->flags&AV_CODEC_FLAG_PASS1)
    1878           0 :         ff_write_pass1_stats(&s->m);
    1879         510 :     s->m.last_pict_type = s->m.pict_type;
    1880             : #if FF_API_STAT_BITS
    1881             : FF_DISABLE_DEPRECATION_WARNINGS
    1882         510 :     avctx->frame_bits = s->m.frame_bits;
    1883         510 :     avctx->mv_bits = s->m.mv_bits;
    1884         510 :     avctx->misc_bits = s->m.misc_bits;
    1885         510 :     avctx->p_tex_bits = s->m.p_tex_bits;
    1886             : FF_ENABLE_DEPRECATION_WARNINGS
    1887             : #endif
    1888             : 
    1889         510 :     emms_c();
    1890             : 
    1891        1530 :     ff_side_data_set_encoder_stats(pkt, s->current_picture->quality,
    1892         510 :                                    s->encoding_error,
    1893         510 :                                    (s->avctx->flags&AV_CODEC_FLAG_PSNR) ? 4 : 0,
    1894         510 :                                    s->current_picture->pict_type);
    1895             : 
    1896             : #if FF_API_ERROR_FRAME
    1897             : FF_DISABLE_DEPRECATION_WARNINGS
    1898         510 :     memcpy(s->current_picture->error, s->encoding_error, sizeof(s->encoding_error));
    1899             : FF_ENABLE_DEPRECATION_WARNINGS
    1900             : #endif
    1901             : 
    1902         510 :     pkt->size = ff_rac_terminate(c);
    1903         510 :     if (s->current_picture->key_frame)
    1904          47 :         pkt->flags |= AV_PKT_FLAG_KEY;
    1905         510 :     *got_packet = 1;
    1906             : 
    1907         510 :     return 0;
    1908             : }
    1909             : 
    1910          11 : static av_cold int encode_end(AVCodecContext *avctx)
    1911             : {
    1912          11 :     SnowContext *s = avctx->priv_data;
    1913             : 
    1914          11 :     ff_snow_common_end(s);
    1915          11 :     ff_rate_control_uninit(&s->m);
    1916          11 :     av_frame_free(&s->input_picture);
    1917          11 :     av_freep(&avctx->stats_out);
    1918             : 
    1919          11 :     return 0;
    1920             : }
    1921             : 
    1922             : #define OFFSET(x) offsetof(SnowContext, x)
    1923             : #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
    1924             : static const AVOption options[] = {
    1925             :     {"motion_est", "motion estimation algorithm", OFFSET(motion_est), AV_OPT_TYPE_INT, {.i64 = FF_ME_EPZS }, FF_ME_ZERO, FF_ME_ITER, VE, "motion_est" },
    1926             :     { "zero", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FF_ME_ZERO }, 0, 0, VE, "motion_est" },
    1927             :     { "epzs", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FF_ME_EPZS }, 0, 0, VE, "motion_est" },
    1928             :     { "xone", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FF_ME_XONE }, 0, 0, VE, "motion_est" },
    1929             :     { "iter", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FF_ME_ITER }, 0, 0, VE, "motion_est" },
    1930             :     { "memc_only",      "Only do ME/MC (I frames -> ref, P frame -> ME+MC).",   OFFSET(memc_only), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, VE },
    1931             :     { "no_bitstream",   "Skip final bitstream writeout.",                    OFFSET(no_bitstream), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, VE },
    1932             :     { "intra_penalty",  "Penalty for intra blocks in block decission",      OFFSET(intra_penalty), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, INT_MAX, VE },
    1933             :     { "iterative_dia_size",  "Dia size for the iterative ME",          OFFSET(iterative_dia_size), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, INT_MAX, VE },
    1934             :     { "sc_threshold",   "Scene change threshold",                   OFFSET(scenechange_threshold), AV_OPT_TYPE_INT, { .i64 = 0 }, INT_MIN, INT_MAX, VE },
    1935             :     { "pred",           "Spatial decomposition type",                                OFFSET(pred), AV_OPT_TYPE_INT, { .i64 = 0 }, DWT_97, DWT_53, VE, "pred" },
    1936             :         { "dwt97", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 0 }, INT_MIN, INT_MAX, VE, "pred" },
    1937             :         { "dwt53", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 1 }, INT_MIN, INT_MAX, VE, "pred" },
    1938             :     { NULL },
    1939             : };
    1940             : 
    1941             : static const AVClass snowenc_class = {
    1942             :     .class_name = "snow encoder",
    1943             :     .item_name  = av_default_item_name,
    1944             :     .option     = options,
    1945             :     .version    = LIBAVUTIL_VERSION_INT,
    1946             : };
    1947             : 
    1948             : AVCodec ff_snow_encoder = {
    1949             :     .name           = "snow",
    1950             :     .long_name      = NULL_IF_CONFIG_SMALL("Snow"),
    1951             :     .type           = AVMEDIA_TYPE_VIDEO,
    1952             :     .id             = AV_CODEC_ID_SNOW,
    1953             :     .priv_data_size = sizeof(SnowContext),
    1954             :     .init           = encode_init,
    1955             :     .encode2        = encode_frame,
    1956             :     .close          = encode_end,
    1957             :     .pix_fmts       = (const enum AVPixelFormat[]){
    1958             :         AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV410P, AV_PIX_FMT_YUV444P,
    1959             :         AV_PIX_FMT_GRAY8,
    1960             :         AV_PIX_FMT_NONE
    1961             :     },
    1962             :     .priv_class     = &snowenc_class,
    1963             :     .caps_internal  = FF_CODEC_CAP_INIT_THREADSAFE |
    1964             :                       FF_CODEC_CAP_INIT_CLEANUP,
    1965             : };

Generated by: LCOV version 1.13