yading@10: /*
yading@10:  * Copyright (C) 2004 Michael Niedermayer <michaelni@gmx.at>
yading@10:  *
yading@10:  * This file is part of FFmpeg.
yading@10:  *
yading@10:  * FFmpeg is free software; you can redistribute it and/or
yading@10:  * modify it under the terms of the GNU Lesser General Public
yading@10:  * License as published by the Free Software Foundation; either
yading@10:  * version 2.1 of the License, or (at your option) any later version.
yading@10:  *
yading@10:  * FFmpeg is distributed in the hope that it will be useful,
yading@10:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
yading@10:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
yading@10:  * Lesser General Public License for more details.
yading@10:  *
yading@10:  * You should have received a copy of the GNU Lesser General Public
yading@10:  * License along with FFmpeg; if not, write to the Free Software
yading@10:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
yading@10:  */
yading@10: 
yading@10: #include "libavutil/intmath.h"
yading@10: #include "libavutil/log.h"
yading@10: #include "libavutil/opt.h"
yading@10: #include "avcodec.h"
yading@10: #include "internal.h"
yading@10: #include "dsputil.h"
yading@10: #include "internal.h"
yading@10: #include "snow_dwt.h"
yading@10: #include "snow.h"
yading@10: 
yading@10: #include "rangecoder.h"
yading@10: #include "mathops.h"
yading@10: 
yading@10: #include "mpegvideo.h"
yading@10: #include "h263.h"
yading@10: 
yading@10: #undef NDEBUG
yading@10: #include <assert.h>
yading@10: 
yading@10: #define QUANTIZE2 0
yading@10: 
yading@10: #if QUANTIZE2==1
yading@10: #define Q2_STEP 8
yading@10: 
yading@10: static void find_sse(SnowContext *s, Plane *p, int *score, int score_stride, IDWTELEM *r0, IDWTELEM *r1, int level, int orientation){
yading@10:     SubBand *b= &p->band[level][orientation];
yading@10:     int x, y;
yading@10:     int xo=0;
yading@10:     int yo=0;
yading@10:     int step= 1 << (s->spatial_decomposition_count - level);
yading@10: 
yading@10:     if(orientation&1)
yading@10:         xo= step>>1;
yading@10:     if(orientation&2)
yading@10:         yo= step>>1;
yading@10: 
yading@10:     //FIXME bias for nonzero ?
yading@10:     //FIXME optimize
yading@10:     memset(score, 0, sizeof(*score)*score_stride*((p->height + Q2_STEP-1)/Q2_STEP));
yading@10:     for(y=0; y<p->height; y++){
yading@10:         for(x=0; x<p->width; x++){
yading@10:             int sx= (x-xo + step/2) / step / Q2_STEP;
yading@10:             int sy= (y-yo + step/2) / step / Q2_STEP;
yading@10:             int v= r0[x + y*p->width] - r1[x + y*p->width];
yading@10:             assert(sx>=0 && sy>=0 && sx < score_stride);
yading@10:             v= ((v+8)>>4)<<4;
yading@10:             score[sx + sy*score_stride] += v*v;
yading@10:             assert(score[sx + sy*score_stride] >= 0);
yading@10:         }
yading@10:     }
yading@10: }
yading@10: 
yading@10: static void dequantize_all(SnowContext *s, Plane *p, IDWTELEM *buffer, int width, int height){
yading@10:     int level, orientation;
yading@10: 
yading@10:     for(level=0; level<s->spatial_decomposition_count; level++){
yading@10:         for(orientation=level ? 1 : 0; orientation<4; orientation++){
yading@10:             SubBand *b= &p->band[level][orientation];
yading@10:             IDWTELEM *dst= buffer + (b->ibuf - s->spatial_idwt_buffer);
yading@10: 
yading@10:             dequantize(s, b, dst, b->stride);
yading@10:         }
yading@10:     }
yading@10: }
yading@10: 
yading@10: static void dwt_quantize(SnowContext *s, Plane *p, DWTELEM *buffer, int width, int height, int stride, int type){
yading@10:     int level, orientation, ys, xs, x, y, pass;
yading@10:     IDWTELEM best_dequant[height * stride];
yading@10:     IDWTELEM idwt2_buffer[height * stride];
yading@10:     const int score_stride= (width + 10)/Q2_STEP;
yading@10:     int best_score[(width + 10)/Q2_STEP * (height + 10)/Q2_STEP]; //FIXME size
yading@10:     int score[(width + 10)/Q2_STEP * (height + 10)/Q2_STEP]; //FIXME size
yading@10:     int threshold= (s->m.lambda * s->m.lambda) >> 6;
yading@10: 
yading@10:     //FIXME pass the copy cleanly ?
yading@10: 
yading@10: //    memcpy(dwt_buffer, buffer, height * stride * sizeof(DWTELEM));
yading@10:     ff_spatial_dwt(buffer, s->temp_dwt_buffer, width, height, stride, type, s->spatial_decomposition_count);
yading@10: 
yading@10:     for(level=0; level<s->spatial_decomposition_count; level++){
yading@10:         for(orientation=level ? 1 : 0; orientation<4; orientation++){
yading@10:             SubBand *b= &p->band[level][orientation];
yading@10:             IDWTELEM *dst= best_dequant + (b->ibuf - s->spatial_idwt_buffer);
yading@10:              DWTELEM *src=       buffer + (b-> buf - s->spatial_dwt_buffer);
yading@10:             assert(src == b->buf); // code does not depend on this but it is true currently
yading@10: 
yading@10:             quantize(s, b, dst, src, b->stride, s->qbias);
yading@10:         }
yading@10:     }
yading@10:     for(pass=0; pass<1; pass++){
yading@10:         if(s->qbias == 0) //keyframe
yading@10:             continue;
yading@10:         for(level=0; level<s->spatial_decomposition_count; level++){
yading@10:             for(orientation=level ? 1 : 0; orientation<4; orientation++){
yading@10:                 SubBand *b= &p->band[level][orientation];
yading@10:                 IDWTELEM *dst= idwt2_buffer + (b->ibuf - s->spatial_idwt_buffer);
yading@10:                 IDWTELEM *best_dst= best_dequant + (b->ibuf - s->spatial_idwt_buffer);
yading@10: 
yading@10:                 for(ys= 0; ys<Q2_STEP; ys++){
yading@10:                     for(xs= 0; xs<Q2_STEP; xs++){
yading@10:                         memcpy(idwt2_buffer, best_dequant, height * stride * sizeof(IDWTELEM));
yading@10:                         dequantize_all(s, p, idwt2_buffer, width, height);
yading@10:                         ff_spatial_idwt(idwt2_buffer, s->temp_idwt_buffer, width, height, stride, type, s->spatial_decomposition_count);
yading@10:                         find_sse(s, p, best_score, score_stride, idwt2_buffer, s->spatial_idwt_buffer, level, orientation);
yading@10:                         memcpy(idwt2_buffer, best_dequant, height * stride * sizeof(IDWTELEM));
yading@10:                         for(y=ys; y<b->height; y+= Q2_STEP){
yading@10:                             for(x=xs; x<b->width; x+= Q2_STEP){
yading@10:                                 if(dst[x + y*b->stride]<0) dst[x + y*b->stride]++;
yading@10:                                 if(dst[x + y*b->stride]>0) dst[x + y*b->stride]--;
yading@10:                                 //FIXME try more than just --
yading@10:                             }
yading@10:                         }
yading@10:                         dequantize_all(s, p, idwt2_buffer, width, height);
yading@10:                         ff_spatial_idwt(idwt2_buffer, s->temp_idwt_buffer, width, height, stride, type, s->spatial_decomposition_count);
yading@10:                         find_sse(s, p, score, score_stride, idwt2_buffer, s->spatial_idwt_buffer, level, orientation);
yading@10:                         for(y=ys; y<b->height; y+= Q2_STEP){
yading@10:                             for(x=xs; x<b->width; x+= Q2_STEP){
yading@10:                                 int score_idx= x/Q2_STEP + (y/Q2_STEP)*score_stride;
yading@10:                                 if(score[score_idx] <= best_score[score_idx] + threshold){
yading@10:                                     best_score[score_idx]= score[score_idx];
yading@10:                                     if(best_dst[x + y*b->stride]<0) best_dst[x + y*b->stride]++;
yading@10:                                     if(best_dst[x + y*b->stride]>0) best_dst[x + y*b->stride]--;
yading@10:                                     //FIXME copy instead
yading@10:                                 }
yading@10:                             }
yading@10:                         }
yading@10:                     }
yading@10:                 }
yading@10:             }
yading@10:         }
yading@10:     }
yading@10:     memcpy(s->spatial_idwt_buffer, best_dequant, height * stride * sizeof(IDWTELEM)); //FIXME work with that directly instead of copy at the end
yading@10: }
yading@10: 
yading@10: #endif /* QUANTIZE2==1 */
yading@10: 
yading@10: static av_cold int encode_init(AVCodecContext *avctx)
yading@10: {
yading@10:     SnowContext *s = avctx->priv_data;
yading@10:     int plane_index, ret;
yading@10: 
yading@10:     if(avctx->strict_std_compliance > FF_COMPLIANCE_EXPERIMENTAL){
yading@10:         av_log(avctx, AV_LOG_ERROR, "This codec is under development, files encoded with it may not be decodable with future versions!!!\n"
yading@10:                "Use vstrict=-2 / -strict -2 to use it anyway.\n");
yading@10:         return -1;
yading@10:     }
yading@10: 
yading@10:     if(avctx->prediction_method == DWT_97
yading@10:        && (avctx->flags & CODEC_FLAG_QSCALE)
yading@10:        && avctx->global_quality == 0){
yading@10:         av_log(avctx, AV_LOG_ERROR, "The 9/7 wavelet is incompatible with lossless mode.\n");
yading@10:         return -1;
yading@10:     }
yading@10: 
yading@10:     s->spatial_decomposition_type= avctx->prediction_method; //FIXME add decorrelator type r transform_type
yading@10: 
yading@10:     s->mv_scale       = (avctx->flags & CODEC_FLAG_QPEL) ? 2 : 4;
yading@10:     s->block_max_depth= (avctx->flags & CODEC_FLAG_4MV ) ? 1 : 0;
yading@10: 
yading@10:     for(plane_index=0; plane_index<3; plane_index++){
yading@10:         s->plane[plane_index].diag_mc= 1;
yading@10:         s->plane[plane_index].htaps= 6;
yading@10:         s->plane[plane_index].hcoeff[0]=  40;
yading@10:         s->plane[plane_index].hcoeff[1]= -10;
yading@10:         s->plane[plane_index].hcoeff[2]=   2;
yading@10:         s->plane[plane_index].fast_mc= 1;
yading@10:     }
yading@10: 
yading@10:     if ((ret = ff_snow_common_init(avctx)) < 0) {
yading@10:         ff_snow_common_end(avctx->priv_data);
yading@10:         return ret;
yading@10:     }
yading@10:     ff_snow_alloc_blocks(s);
yading@10: 
yading@10:     s->version=0;
yading@10: 
yading@10:     s->m.avctx   = avctx;
yading@10:     s->m.flags   = avctx->flags;
yading@10:     s->m.bit_rate= avctx->bit_rate;
yading@10: 
yading@10:     s->m.me.temp      =
yading@10:     s->m.me.scratchpad= av_mallocz((avctx->width+64)*2*16*2*sizeof(uint8_t));
yading@10:     s->m.me.map       = av_mallocz(ME_MAP_SIZE*sizeof(uint32_t));
yading@10:     s->m.me.score_map = av_mallocz(ME_MAP_SIZE*sizeof(uint32_t));
yading@10:     s->m.obmc_scratchpad= av_mallocz(MB_SIZE*MB_SIZE*12*sizeof(uint32_t));
yading@10:     ff_h263_encode_init(&s->m); //mv_penalty
yading@10: 
yading@10:     s->max_ref_frames = FFMAX(FFMIN(avctx->refs, MAX_REF_FRAMES), 1);
yading@10: 
yading@10:     if(avctx->flags&CODEC_FLAG_PASS1){
yading@10:         if(!avctx->stats_out)
yading@10:             avctx->stats_out = av_mallocz(256);
yading@10:     }
yading@10:     if((avctx->flags&CODEC_FLAG_PASS2) || !(avctx->flags&CODEC_FLAG_QSCALE)){
yading@10:         if(ff_rate_control_init(&s->m) < 0)
yading@10:             return -1;
yading@10:     }
yading@10:     s->pass1_rc= !(avctx->flags & (CODEC_FLAG_QSCALE|CODEC_FLAG_PASS2));
yading@10: 
yading@10:     avctx->coded_frame= &s->current_picture;
yading@10:     switch(avctx->pix_fmt){
yading@10:     case AV_PIX_FMT_YUV444P:
yading@10: //    case AV_PIX_FMT_YUV422P:
yading@10:     case AV_PIX_FMT_YUV420P:
yading@10: //     case AV_PIX_FMT_GRAY8:
yading@10: //    case AV_PIX_FMT_YUV411P:
yading@10:     case AV_PIX_FMT_YUV410P:
yading@10:         s->colorspace_type= 0;
yading@10:         break;
yading@10: /*    case AV_PIX_FMT_RGB32:
yading@10:         s->colorspace= 1;
yading@10:         break;*/
yading@10:     default:
yading@10:         av_log(avctx, AV_LOG_ERROR, "pixel format not supported\n");
yading@10:         return -1;
yading@10:     }
yading@10:     avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_h_shift, &s->chroma_v_shift);
yading@10: 
yading@10:     ff_set_cmp(&s->dsp, s->dsp.me_cmp, s->avctx->me_cmp);
yading@10:     ff_set_cmp(&s->dsp, s->dsp.me_sub_cmp, s->avctx->me_sub_cmp);
yading@10: 
yading@10:     if ((ret = ff_get_buffer(s->avctx, &s->input_picture, AV_GET_BUFFER_FLAG_REF)) < 0)
yading@10:         return ret;
yading@10: 
yading@10:     if(s->avctx->me_method == ME_ITER){
yading@10:         int i;
yading@10:         int size= s->b_width * s->b_height << 2*s->block_max_depth;
yading@10:         for(i=0; i<s->max_ref_frames; i++){
yading@10:             s->ref_mvs[i]= av_mallocz(size*sizeof(int16_t[2]));
yading@10:             s->ref_scores[i]= av_mallocz(size*sizeof(uint32_t));
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: //near copy & paste from dsputil, FIXME
yading@10: static int pix_sum(uint8_t * pix, int line_size, int w, int h)
yading@10: {
yading@10:     int s, i, j;
yading@10: 
yading@10:     s = 0;
yading@10:     for (i = 0; i < h; i++) {
yading@10:         for (j = 0; j < w; j++) {
yading@10:             s += pix[0];
yading@10:             pix ++;
yading@10:         }
yading@10:         pix += line_size - w;
yading@10:     }
yading@10:     return s;
yading@10: }
yading@10: 
yading@10: //near copy & paste from dsputil, FIXME
yading@10: static int pix_norm1(uint8_t * pix, int line_size, int w)
yading@10: {
yading@10:     int s, i, j;
yading@10:     uint32_t *sq = ff_squareTbl + 256;
yading@10: 
yading@10:     s = 0;
yading@10:     for (i = 0; i < w; i++) {
yading@10:         for (j = 0; j < w; j ++) {
yading@10:             s += sq[pix[0]];
yading@10:             pix ++;
yading@10:         }
yading@10:         pix += line_size - w;
yading@10:     }
yading@10:     return s;
yading@10: }
yading@10: 
yading@10: static inline int get_penalty_factor(int lambda, int lambda2, int type){
yading@10:     switch(type&0xFF){
yading@10:     default:
yading@10:     case FF_CMP_SAD:
yading@10:         return lambda>>FF_LAMBDA_SHIFT;
yading@10:     case FF_CMP_DCT:
yading@10:         return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
yading@10:     case FF_CMP_W53:
yading@10:         return (4*lambda)>>(FF_LAMBDA_SHIFT);
yading@10:     case FF_CMP_W97:
yading@10:         return (2*lambda)>>(FF_LAMBDA_SHIFT);
yading@10:     case FF_CMP_SATD:
yading@10:     case FF_CMP_DCT264:
yading@10:         return (2*lambda)>>FF_LAMBDA_SHIFT;
yading@10:     case FF_CMP_RD:
yading@10:     case FF_CMP_PSNR:
yading@10:     case FF_CMP_SSE:
yading@10:     case FF_CMP_NSSE:
yading@10:         return lambda2>>FF_LAMBDA_SHIFT;
yading@10:     case FF_CMP_BIT:
yading@10:         return 1;
yading@10:     }
yading@10: }
yading@10: 
yading@10: //FIXME copy&paste
yading@10: #define P_LEFT P[1]
yading@10: #define P_TOP P[2]
yading@10: #define P_TOPRIGHT P[3]
yading@10: #define P_MEDIAN P[4]
yading@10: #define P_MV1 P[9]
yading@10: #define FLAG_QPEL   1 //must be 1
yading@10: 
yading@10: static int encode_q_branch(SnowContext *s, int level, int x, int y){
yading@10:     uint8_t p_buffer[1024];
yading@10:     uint8_t i_buffer[1024];
yading@10:     uint8_t p_state[sizeof(s->block_state)];
yading@10:     uint8_t i_state[sizeof(s->block_state)];
yading@10:     RangeCoder pc, ic;
yading@10:     uint8_t *pbbak= s->c.bytestream;
yading@10:     uint8_t *pbbak_start= s->c.bytestream_start;
yading@10:     int score, score2, iscore, i_len, p_len, block_s, sum, base_bits;
yading@10:     const int w= s->b_width  << s->block_max_depth;
yading@10:     const int h= s->b_height << s->block_max_depth;
yading@10:     const int rem_depth= s->block_max_depth - level;
yading@10:     const int index= (x + y*w) << rem_depth;
yading@10:     const int block_w= 1<<(LOG2_MB_SIZE - level);
yading@10:     int trx= (x+1)<<rem_depth;
yading@10:     int try= (y+1)<<rem_depth;
yading@10:     const BlockNode *left  = x ? &s->block[index-1] : &null_block;
yading@10:     const BlockNode *top   = y ? &s->block[index-w] : &null_block;
yading@10:     const BlockNode *right = trx<w ? &s->block[index+1] : &null_block;
yading@10:     const BlockNode *bottom= try<h ? &s->block[index+w] : &null_block;
yading@10:     const BlockNode *tl    = y && x ? &s->block[index-w-1] : left;
yading@10:     const BlockNode *tr    = y && trx<w && ((x&1)==0 || level==0) ? &s->block[index-w+(1<<rem_depth)] : tl; //FIXME use lt
yading@10:     int pl = left->color[0];
yading@10:     int pcb= left->color[1];
yading@10:     int pcr= left->color[2];
yading@10:     int pmx, pmy;
yading@10:     int mx=0, my=0;
yading@10:     int l,cr,cb;
yading@10:     const int stride= s->current_picture.linesize[0];
yading@10:     const int uvstride= s->current_picture.linesize[1];
yading@10:     uint8_t *current_data[3]= { s->input_picture.data[0] + (x + y*  stride)*block_w,
yading@10:                                 s->input_picture.data[1] + ((x*block_w)>>s->chroma_h_shift) + ((y*uvstride*block_w)>>s->chroma_v_shift),
yading@10:                                 s->input_picture.data[2] + ((x*block_w)>>s->chroma_h_shift) + ((y*uvstride*block_w)>>s->chroma_v_shift)};
yading@10:     int P[10][2];
yading@10:     int16_t last_mv[3][2];
yading@10:     int qpel= !!(s->avctx->flags & CODEC_FLAG_QPEL); //unused
yading@10:     const int shift= 1+qpel;
yading@10:     MotionEstContext *c= &s->m.me;
yading@10:     int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref);
yading@10:     int mx_context= av_log2(2*FFABS(left->mx - top->mx));
yading@10:     int my_context= av_log2(2*FFABS(left->my - top->my));
yading@10:     int s_context= 2*left->level + 2*top->level + tl->level + tr->level;
yading@10:     int ref, best_ref, ref_score, ref_mx, ref_my;
yading@10: 
yading@10:     assert(sizeof(s->block_state) >= 256);
yading@10:     if(s->keyframe){
yading@10:         set_blocks(s, level, x, y, pl, pcb, pcr, 0, 0, 0, BLOCK_INTRA);
yading@10:         return 0;
yading@10:     }
yading@10: 
yading@10: //    clip predictors / edge ?
yading@10: 
yading@10:     P_LEFT[0]= left->mx;
yading@10:     P_LEFT[1]= left->my;
yading@10:     P_TOP [0]= top->mx;
yading@10:     P_TOP [1]= top->my;
yading@10:     P_TOPRIGHT[0]= tr->mx;
yading@10:     P_TOPRIGHT[1]= tr->my;
yading@10: 
yading@10:     last_mv[0][0]= s->block[index].mx;
yading@10:     last_mv[0][1]= s->block[index].my;
yading@10:     last_mv[1][0]= right->mx;
yading@10:     last_mv[1][1]= right->my;
yading@10:     last_mv[2][0]= bottom->mx;
yading@10:     last_mv[2][1]= bottom->my;
yading@10: 
yading@10:     s->m.mb_stride=2;
yading@10:     s->m.mb_x=
yading@10:     s->m.mb_y= 0;
yading@10:     c->skip= 0;
yading@10: 
yading@10:     assert(c->  stride ==   stride);
yading@10:     assert(c->uvstride == uvstride);
yading@10: 
yading@10:     c->penalty_factor    = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp);
yading@10:     c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp);
yading@10:     c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp);
yading@10:     c->current_mv_penalty= c->mv_penalty[s->m.f_code=1] + MAX_MV;
yading@10: 
yading@10:     c->xmin = - x*block_w - 16+3;
yading@10:     c->ymin = - y*block_w - 16+3;
yading@10:     c->xmax = - (x+1)*block_w + (w<<(LOG2_MB_SIZE - s->block_max_depth)) + 16-3;
yading@10:     c->ymax = - (y+1)*block_w + (h<<(LOG2_MB_SIZE - s->block_max_depth)) + 16-3;
yading@10: 
yading@10:     if(P_LEFT[0]     > (c->xmax<<shift)) P_LEFT[0]    = (c->xmax<<shift);
yading@10:     if(P_LEFT[1]     > (c->ymax<<shift)) P_LEFT[1]    = (c->ymax<<shift);
yading@10:     if(P_TOP[0]      > (c->xmax<<shift)) P_TOP[0]     = (c->xmax<<shift);
yading@10:     if(P_TOP[1]      > (c->ymax<<shift)) P_TOP[1]     = (c->ymax<<shift);
yading@10:     if(P_TOPRIGHT[0] < (c->xmin<<shift)) P_TOPRIGHT[0]= (c->xmin<<shift);
yading@10:     if(P_TOPRIGHT[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<shift); //due to pmx no clip
yading@10:     if(P_TOPRIGHT[1] > (c->ymax<<shift)) P_TOPRIGHT[1]= (c->ymax<<shift);
yading@10: 
yading@10:     P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
yading@10:     P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
yading@10: 
yading@10:     if (!y) {
yading@10:         c->pred_x= P_LEFT[0];
yading@10:         c->pred_y= P_LEFT[1];
yading@10:     } else {
yading@10:         c->pred_x = P_MEDIAN[0];
yading@10:         c->pred_y = P_MEDIAN[1];
yading@10:     }
yading@10: 
yading@10:     score= INT_MAX;
yading@10:     best_ref= 0;
yading@10:     for(ref=0; ref<s->ref_frames; ref++){
yading@10:         init_ref(c, current_data, s->last_picture[ref].data, NULL, block_w*x, block_w*y, 0);
yading@10: 
yading@10:         ref_score= ff_epzs_motion_search(&s->m, &ref_mx, &ref_my, P, 0, /*ref_index*/ 0, last_mv,
yading@10:                                          (1<<16)>>shift, level-LOG2_MB_SIZE+4, block_w);
yading@10: 
yading@10:         assert(ref_mx >= c->xmin);
yading@10:         assert(ref_mx <= c->xmax);
yading@10:         assert(ref_my >= c->ymin);
yading@10:         assert(ref_my <= c->ymax);
yading@10: 
yading@10:         ref_score= c->sub_motion_search(&s->m, &ref_mx, &ref_my, ref_score, 0, 0, level-LOG2_MB_SIZE+4, block_w);
yading@10:         ref_score= ff_get_mb_score(&s->m, ref_mx, ref_my, 0, 0, level-LOG2_MB_SIZE+4, block_w, 0);
yading@10:         ref_score+= 2*av_log2(2*ref)*c->penalty_factor;
yading@10:         if(s->ref_mvs[ref]){
yading@10:             s->ref_mvs[ref][index][0]= ref_mx;
yading@10:             s->ref_mvs[ref][index][1]= ref_my;
yading@10:             s->ref_scores[ref][index]= ref_score;
yading@10:         }
yading@10:         if(score > ref_score){
yading@10:             score= ref_score;
yading@10:             best_ref= ref;
yading@10:             mx= ref_mx;
yading@10:             my= ref_my;
yading@10:         }
yading@10:     }
yading@10:     //FIXME if mb_cmp != SSE then intra cannot be compared currently and mb_penalty vs. lambda2
yading@10: 
yading@10:   //  subpel search
yading@10:     base_bits= get_rac_count(&s->c) - 8*(s->c.bytestream - s->c.bytestream_start);
yading@10:     pc= s->c;
yading@10:     pc.bytestream_start=
yading@10:     pc.bytestream= p_buffer; //FIXME end/start? and at the other stoo
yading@10:     memcpy(p_state, s->block_state, sizeof(s->block_state));
yading@10: 
yading@10:     if(level!=s->block_max_depth)
yading@10:         put_rac(&pc, &p_state[4 + s_context], 1);
yading@10:     put_rac(&pc, &p_state[1 + left->type + top->type], 0);
yading@10:     if(s->ref_frames > 1)
yading@10:         put_symbol(&pc, &p_state[128 + 1024 + 32*ref_context], best_ref, 0);
yading@10:     pred_mv(s, &pmx, &pmy, best_ref, left, top, tr);
yading@10:     put_symbol(&pc, &p_state[128 + 32*(mx_context + 16*!!best_ref)], mx - pmx, 1);
yading@10:     put_symbol(&pc, &p_state[128 + 32*(my_context + 16*!!best_ref)], my - pmy, 1);
yading@10:     p_len= pc.bytestream - pc.bytestream_start;
yading@10:     score += (s->lambda2*(get_rac_count(&pc)-base_bits))>>FF_LAMBDA_SHIFT;
yading@10: 
yading@10:     block_s= block_w*block_w;
yading@10:     sum = pix_sum(current_data[0], stride, block_w, block_w);
yading@10:     l= (sum + block_s/2)/block_s;
yading@10:     iscore = pix_norm1(current_data[0], stride, block_w) - 2*l*sum + l*l*block_s;
yading@10: 
yading@10:     block_s= block_w*block_w>>(s->chroma_h_shift + s->chroma_v_shift);
yading@10:     sum = pix_sum(current_data[1], uvstride, block_w>>s->chroma_h_shift, block_w>>s->chroma_v_shift);
yading@10:     cb= (sum + block_s/2)/block_s;
yading@10: //    iscore += pix_norm1(&current_mb[1][0], uvstride, block_w>>1) - 2*cb*sum + cb*cb*block_s;
yading@10:     sum = pix_sum(current_data[2], uvstride, block_w>>s->chroma_h_shift, block_w>>s->chroma_v_shift);
yading@10:     cr= (sum + block_s/2)/block_s;
yading@10: //    iscore += pix_norm1(&current_mb[2][0], uvstride, block_w>>1) - 2*cr*sum + cr*cr*block_s;
yading@10: 
yading@10:     ic= s->c;
yading@10:     ic.bytestream_start=
yading@10:     ic.bytestream= i_buffer; //FIXME end/start? and at the other stoo
yading@10:     memcpy(i_state, s->block_state, sizeof(s->block_state));
yading@10:     if(level!=s->block_max_depth)
yading@10:         put_rac(&ic, &i_state[4 + s_context], 1);
yading@10:     put_rac(&ic, &i_state[1 + left->type + top->type], 1);
yading@10:     put_symbol(&ic, &i_state[32],  l-pl , 1);
yading@10:     put_symbol(&ic, &i_state[64], cb-pcb, 1);
yading@10:     put_symbol(&ic, &i_state[96], cr-pcr, 1);
yading@10:     i_len= ic.bytestream - ic.bytestream_start;
yading@10:     iscore += (s->lambda2*(get_rac_count(&ic)-base_bits))>>FF_LAMBDA_SHIFT;
yading@10: 
yading@10: //    assert(score==256*256*256*64-1);
yading@10:     assert(iscore < 255*255*256 + s->lambda2*10);
yading@10:     assert(iscore >= 0);
yading@10:     assert(l>=0 && l<=255);
yading@10:     assert(pl>=0 && pl<=255);
yading@10: 
yading@10:     if(level==0){
yading@10:         int varc= iscore >> 8;
yading@10:         int vard= score >> 8;
yading@10:         if (vard <= 64 || vard < varc)
yading@10:             c->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);
yading@10:         else
yading@10:             c->scene_change_score+= s->m.qscale;
yading@10:     }
yading@10: 
yading@10:     if(level!=s->block_max_depth){
yading@10:         put_rac(&s->c, &s->block_state[4 + s_context], 0);
yading@10:         score2 = encode_q_branch(s, level+1, 2*x+0, 2*y+0);
yading@10:         score2+= encode_q_branch(s, level+1, 2*x+1, 2*y+0);
yading@10:         score2+= encode_q_branch(s, level+1, 2*x+0, 2*y+1);
yading@10:         score2+= encode_q_branch(s, level+1, 2*x+1, 2*y+1);
yading@10:         score2+= s->lambda2>>FF_LAMBDA_SHIFT; //FIXME exact split overhead
yading@10: 
yading@10:         if(score2 < score && score2 < iscore)
yading@10:             return score2;
yading@10:     }
yading@10: 
yading@10:     if(iscore < score){
yading@10:         pred_mv(s, &pmx, &pmy, 0, left, top, tr);
yading@10:         memcpy(pbbak, i_buffer, i_len);
yading@10:         s->c= ic;
yading@10:         s->c.bytestream_start= pbbak_start;
yading@10:         s->c.bytestream= pbbak + i_len;
yading@10:         set_blocks(s, level, x, y, l, cb, cr, pmx, pmy, 0, BLOCK_INTRA);
yading@10:         memcpy(s->block_state, i_state, sizeof(s->block_state));
yading@10:         return iscore;
yading@10:     }else{
yading@10:         memcpy(pbbak, p_buffer, p_len);
yading@10:         s->c= pc;
yading@10:         s->c.bytestream_start= pbbak_start;
yading@10:         s->c.bytestream= pbbak + p_len;
yading@10:         set_blocks(s, level, x, y, pl, pcb, pcr, mx, my, best_ref, 0);
yading@10:         memcpy(s->block_state, p_state, sizeof(s->block_state));
yading@10:         return score;
yading@10:     }
yading@10: }
yading@10: 
yading@10: static void encode_q_branch2(SnowContext *s, int level, int x, int y){
yading@10:     const int w= s->b_width  << s->block_max_depth;
yading@10:     const int rem_depth= s->block_max_depth - level;
yading@10:     const int index= (x + y*w) << rem_depth;
yading@10:     int trx= (x+1)<<rem_depth;
yading@10:     BlockNode *b= &s->block[index];
yading@10:     const BlockNode *left  = x ? &s->block[index-1] : &null_block;
yading@10:     const BlockNode *top   = y ? &s->block[index-w] : &null_block;
yading@10:     const BlockNode *tl    = y && x ? &s->block[index-w-1] : left;
yading@10:     const BlockNode *tr    = y && trx<w && ((x&1)==0 || level==0) ? &s->block[index-w+(1<<rem_depth)] : tl; //FIXME use lt
yading@10:     int pl = left->color[0];
yading@10:     int pcb= left->color[1];
yading@10:     int pcr= left->color[2];
yading@10:     int pmx, pmy;
yading@10:     int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref);
yading@10:     int mx_context= av_log2(2*FFABS(left->mx - top->mx)) + 16*!!b->ref;
yading@10:     int my_context= av_log2(2*FFABS(left->my - top->my)) + 16*!!b->ref;
yading@10:     int s_context= 2*left->level + 2*top->level + tl->level + tr->level;
yading@10: 
yading@10:     if(s->keyframe){
yading@10:         set_blocks(s, level, x, y, pl, pcb, pcr, 0, 0, 0, BLOCK_INTRA);
yading@10:         return;
yading@10:     }
yading@10: 
yading@10:     if(level!=s->block_max_depth){
yading@10:         if(same_block(b,b+1) && same_block(b,b+w) && same_block(b,b+w+1)){
yading@10:             put_rac(&s->c, &s->block_state[4 + s_context], 1);
yading@10:         }else{
yading@10:             put_rac(&s->c, &s->block_state[4 + s_context], 0);
yading@10:             encode_q_branch2(s, level+1, 2*x+0, 2*y+0);
yading@10:             encode_q_branch2(s, level+1, 2*x+1, 2*y+0);
yading@10:             encode_q_branch2(s, level+1, 2*x+0, 2*y+1);
yading@10:             encode_q_branch2(s, level+1, 2*x+1, 2*y+1);
yading@10:             return;
yading@10:         }
yading@10:     }
yading@10:     if(b->type & BLOCK_INTRA){
yading@10:         pred_mv(s, &pmx, &pmy, 0, left, top, tr);
yading@10:         put_rac(&s->c, &s->block_state[1 + (left->type&1) + (top->type&1)], 1);
yading@10:         put_symbol(&s->c, &s->block_state[32], b->color[0]-pl , 1);
yading@10:         put_symbol(&s->c, &s->block_state[64], b->color[1]-pcb, 1);
yading@10:         put_symbol(&s->c, &s->block_state[96], b->color[2]-pcr, 1);
yading@10:         set_blocks(s, level, x, y, b->color[0], b->color[1], b->color[2], pmx, pmy, 0, BLOCK_INTRA);
yading@10:     }else{
yading@10:         pred_mv(s, &pmx, &pmy, b->ref, left, top, tr);
yading@10:         put_rac(&s->c, &s->block_state[1 + (left->type&1) + (top->type&1)], 0);
yading@10:         if(s->ref_frames > 1)
yading@10:             put_symbol(&s->c, &s->block_state[128 + 1024 + 32*ref_context], b->ref, 0);
yading@10:         put_symbol(&s->c, &s->block_state[128 + 32*mx_context], b->mx - pmx, 1);
yading@10:         put_symbol(&s->c, &s->block_state[128 + 32*my_context], b->my - pmy, 1);
yading@10:         set_blocks(s, level, x, y, pl, pcb, pcr, b->mx, b->my, b->ref, 0);
yading@10:     }
yading@10: }
yading@10: 
yading@10: static int get_dc(SnowContext *s, int mb_x, int mb_y, int plane_index){
yading@10:     int i, x2, y2;
yading@10:     Plane *p= &s->plane[plane_index];
yading@10:     const int block_size = MB_SIZE >> s->block_max_depth;
yading@10:     const int block_w    = plane_index ? block_size>>s->chroma_h_shift : block_size;
yading@10:     const int block_h    = plane_index ? block_size>>s->chroma_v_shift : block_size;
yading@10:     const uint8_t *obmc  = plane_index ? ff_obmc_tab[s->block_max_depth+s->chroma_h_shift] : ff_obmc_tab[s->block_max_depth];
yading@10:     const int obmc_stride= plane_index ? (2*block_size)>>s->chroma_h_shift : 2*block_size;
yading@10:     const int ref_stride= s->current_picture.linesize[plane_index];
yading@10:     uint8_t *src= s-> input_picture.data[plane_index];
yading@10:     IDWTELEM *dst= (IDWTELEM*)s->m.obmc_scratchpad + plane_index*block_size*block_size*4; //FIXME change to unsigned
yading@10:     const int b_stride = s->b_width << s->block_max_depth;
yading@10:     const int w= p->width;
yading@10:     const int h= p->height;
yading@10:     int index= mb_x + mb_y*b_stride;
yading@10:     BlockNode *b= &s->block[index];
yading@10:     BlockNode backup= *b;
yading@10:     int ab=0;
yading@10:     int aa=0;
yading@10: 
yading@10:     av_assert2(s->chroma_h_shift == s->chroma_v_shift); //obmc stuff above
yading@10: 
yading@10:     b->type|= BLOCK_INTRA;
yading@10:     b->color[plane_index]= 0;
yading@10:     memset(dst, 0, obmc_stride*obmc_stride*sizeof(IDWTELEM));
yading@10: 
yading@10:     for(i=0; i<4; i++){
yading@10:         int mb_x2= mb_x + (i &1) - 1;
yading@10:         int mb_y2= mb_y + (i>>1) - 1;
yading@10:         int x= block_w*mb_x2 + block_w/2;
yading@10:         int y= block_h*mb_y2 + block_h/2;
yading@10: 
yading@10:         add_yblock(s, 0, NULL, dst + (i&1)*block_w + (i>>1)*obmc_stride*block_h, NULL, obmc,
yading@10:                     x, y, block_w, block_h, w, h, obmc_stride, ref_stride, obmc_stride, mb_x2, mb_y2, 0, 0, plane_index);
yading@10: 
yading@10:         for(y2= FFMAX(y, 0); y2<FFMIN(h, y+block_h); y2++){
yading@10:             for(x2= FFMAX(x, 0); x2<FFMIN(w, x+block_w); x2++){
yading@10:                 int index= x2-(block_w*mb_x - block_w/2) + (y2-(block_h*mb_y - block_h/2))*obmc_stride;
yading@10:                 int obmc_v= obmc[index];
yading@10:                 int d;
yading@10:                 if(y<0) obmc_v += obmc[index + block_h*obmc_stride];
yading@10:                 if(x<0) obmc_v += obmc[index + block_w];
yading@10:                 if(y+block_h>h) obmc_v += obmc[index - block_h*obmc_stride];
yading@10:                 if(x+block_w>w) obmc_v += obmc[index - block_w];
yading@10:                 //FIXME precalculate this or simplify it somehow else
yading@10: 
yading@10:                 d = -dst[index] + (1<<(FRAC_BITS-1));
yading@10:                 dst[index] = d;
yading@10:                 ab += (src[x2 + y2*ref_stride] - (d>>FRAC_BITS)) * obmc_v;
yading@10:                 aa += obmc_v * obmc_v; //FIXME precalculate this
yading@10:             }
yading@10:         }
yading@10:     }
yading@10:     *b= backup;
yading@10: 
yading@10:     return av_clip( ROUNDED_DIV(ab<<LOG2_OBMC_MAX, aa), 0, 255); //FIXME we should not need clipping
yading@10: }
yading@10: 
yading@10: static inline int get_block_bits(SnowContext *s, int x, int y, int w){
yading@10:     const int b_stride = s->b_width << s->block_max_depth;
yading@10:     const int b_height = s->b_height<< s->block_max_depth;
yading@10:     int index= x + y*b_stride;
yading@10:     const BlockNode *b     = &s->block[index];
yading@10:     const BlockNode *left  = x ? &s->block[index-1] : &null_block;
yading@10:     const BlockNode *top   = y ? &s->block[index-b_stride] : &null_block;
yading@10:     const BlockNode *tl    = y && x ? &s->block[index-b_stride-1] : left;
yading@10:     const BlockNode *tr    = y && x+w<b_stride ? &s->block[index-b_stride+w] : tl;
yading@10:     int dmx, dmy;
yading@10: //  int mx_context= av_log2(2*FFABS(left->mx - top->mx));
yading@10: //  int my_context= av_log2(2*FFABS(left->my - top->my));
yading@10: 
yading@10:     if(x<0 || x>=b_stride || y>=b_height)
yading@10:         return 0;
yading@10: /*
yading@10: 1            0      0
yading@10: 01X          1-2    1
yading@10: 001XX        3-6    2-3
yading@10: 0001XXX      7-14   4-7
yading@10: 00001XXXX   15-30   8-15
yading@10: */
yading@10: //FIXME try accurate rate
yading@10: //FIXME intra and inter predictors if surrounding blocks are not the same type
yading@10:     if(b->type & BLOCK_INTRA){
yading@10:         return 3+2*( av_log2(2*FFABS(left->color[0] - b->color[0]))
yading@10:                    + av_log2(2*FFABS(left->color[1] - b->color[1]))
yading@10:                    + av_log2(2*FFABS(left->color[2] - b->color[2])));
yading@10:     }else{
yading@10:         pred_mv(s, &dmx, &dmy, b->ref, left, top, tr);
yading@10:         dmx-= b->mx;
yading@10:         dmy-= b->my;
yading@10:         return 2*(1 + av_log2(2*FFABS(dmx)) //FIXME kill the 2* can be merged in lambda
yading@10:                     + av_log2(2*FFABS(dmy))
yading@10:                     + av_log2(2*b->ref));
yading@10:     }
yading@10: }
yading@10: 
yading@10: static int get_block_rd(SnowContext *s, int mb_x, int mb_y, int plane_index, uint8_t (*obmc_edged)[MB_SIZE * 2]){
yading@10:     Plane *p= &s->plane[plane_index];
yading@10:     const int block_size = MB_SIZE >> s->block_max_depth;
yading@10:     const int block_w    = plane_index ? block_size>>s->chroma_h_shift : block_size;
yading@10:     const int block_h    = plane_index ? block_size>>s->chroma_v_shift : block_size;
yading@10:     const int obmc_stride= plane_index ? (2*block_size)>>s->chroma_h_shift : 2*block_size;
yading@10:     const int ref_stride= s->current_picture.linesize[plane_index];
yading@10:     uint8_t *dst= s->current_picture.data[plane_index];
yading@10:     uint8_t *src= s->  input_picture.data[plane_index];
yading@10:     IDWTELEM *pred= (IDWTELEM*)s->m.obmc_scratchpad + plane_index*block_size*block_size*4;
yading@10:     uint8_t *cur = s->scratchbuf;
yading@10:     uint8_t *tmp = s->emu_edge_buffer;
yading@10:     const int b_stride = s->b_width << s->block_max_depth;
yading@10:     const int b_height = s->b_height<< s->block_max_depth;
yading@10:     const int w= p->width;
yading@10:     const int h= p->height;
yading@10:     int distortion;
yading@10:     int rate= 0;
yading@10:     const int penalty_factor= get_penalty_factor(s->lambda, s->lambda2, s->avctx->me_cmp);
yading@10:     int sx= block_w*mb_x - block_w/2;
yading@10:     int sy= block_h*mb_y - block_h/2;
yading@10:     int x0= FFMAX(0,-sx);
yading@10:     int y0= FFMAX(0,-sy);
yading@10:     int x1= FFMIN(block_w*2, w-sx);
yading@10:     int y1= FFMIN(block_h*2, h-sy);
yading@10:     int i,x,y;
yading@10: 
yading@10:     av_assert2(s->chroma_h_shift == s->chroma_v_shift); //obmc and square assumtions below chckinhg only block_w
yading@10: 
yading@10:     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);
yading@10: 
yading@10:     for(y=y0; y<y1; y++){
yading@10:         const uint8_t *obmc1= obmc_edged[y];
yading@10:         const IDWTELEM *pred1 = pred + y*obmc_stride;
yading@10:         uint8_t *cur1 = cur + y*ref_stride;
yading@10:         uint8_t *dst1 = dst + sx + (sy+y)*ref_stride;
yading@10:         for(x=x0; x<x1; x++){
yading@10: #if FRAC_BITS >= LOG2_OBMC_MAX
yading@10:             int v = (cur1[x] * obmc1[x]) << (FRAC_BITS - LOG2_OBMC_MAX);
yading@10: #else
yading@10:             int v = (cur1[x] * obmc1[x] + (1<<(LOG2_OBMC_MAX - FRAC_BITS-1))) >> (LOG2_OBMC_MAX - FRAC_BITS);
yading@10: #endif
yading@10:             v = (v + pred1[x]) >> FRAC_BITS;
yading@10:             if(v&(~255)) v= ~(v>>31);
yading@10:             dst1[x] = v;
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     /* copy the regions where obmc[] = (uint8_t)256 */
yading@10:     if(LOG2_OBMC_MAX == 8
yading@10:         && (mb_x == 0 || mb_x == b_stride-1)
yading@10:         && (mb_y == 0 || mb_y == b_height-1)){
yading@10:         if(mb_x == 0)
yading@10:             x1 = block_w;
yading@10:         else
yading@10:             x0 = block_w;
yading@10:         if(mb_y == 0)
yading@10:             y1 = block_h;
yading@10:         else
yading@10:             y0 = block_h;
yading@10:         for(y=y0; y<y1; y++)
yading@10:             memcpy(dst + sx+x0 + (sy+y)*ref_stride, cur + x0 + y*ref_stride, x1-x0);
yading@10:     }
yading@10: 
yading@10:     if(block_w==16){
yading@10:         /* FIXME rearrange dsputil to fit 32x32 cmp functions */
yading@10:         /* FIXME check alignment of the cmp wavelet vs the encoding wavelet */
yading@10:         /* FIXME cmps overlap but do not cover the wavelet's whole support.
yading@10:          * So improving the score of one block is not strictly guaranteed
yading@10:          * to improve the score of the whole frame, thus iterative motion
yading@10:          * estimation does not always converge. */
yading@10:         if(s->avctx->me_cmp == FF_CMP_W97)
yading@10:             distortion = ff_w97_32_c(&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, 32);
yading@10:         else if(s->avctx->me_cmp == FF_CMP_W53)
yading@10:             distortion = ff_w53_32_c(&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, 32);
yading@10:         else{
yading@10:             distortion = 0;
yading@10:             for(i=0; i<4; i++){
yading@10:                 int off = sx+16*(i&1) + (sy+16*(i>>1))*ref_stride;
yading@10:                 distortion += s->dsp.me_cmp[0](&s->m, src + off, dst + off, ref_stride, 16);
yading@10:             }
yading@10:         }
yading@10:     }else{
yading@10:         assert(block_w==8);
yading@10:         distortion = s->dsp.me_cmp[0](&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, block_w*2);
yading@10:     }
yading@10: 
yading@10:     if(plane_index==0){
yading@10:         for(i=0; i<4; i++){
yading@10: /* ..RRr
yading@10:  * .RXx.
yading@10:  * rxx..
yading@10:  */
yading@10:             rate += get_block_bits(s, mb_x + (i&1) - (i>>1), mb_y + (i>>1), 1);
yading@10:         }
yading@10:         if(mb_x == b_stride-2)
yading@10:             rate += get_block_bits(s, mb_x + 1, mb_y + 1, 1);
yading@10:     }
yading@10:     return distortion + rate*penalty_factor;
yading@10: }
yading@10: 
yading@10: static int get_4block_rd(SnowContext *s, int mb_x, int mb_y, int plane_index){
yading@10:     int i, y2;
yading@10:     Plane *p= &s->plane[plane_index];
yading@10:     const int block_size = MB_SIZE >> s->block_max_depth;
yading@10:     const int block_w    = plane_index ? block_size>>s->chroma_h_shift : block_size;
yading@10:     const int block_h    = plane_index ? block_size>>s->chroma_v_shift : block_size;
yading@10:     const uint8_t *obmc  = plane_index ? ff_obmc_tab[s->block_max_depth+s->chroma_h_shift] : ff_obmc_tab[s->block_max_depth];
yading@10:     const int obmc_stride= plane_index ? (2*block_size)>>s->chroma_h_shift : 2*block_size;
yading@10:     const int ref_stride= s->current_picture.linesize[plane_index];
yading@10:     uint8_t *dst= s->current_picture.data[plane_index];
yading@10:     uint8_t *src= s-> input_picture.data[plane_index];
yading@10:     //FIXME zero_dst is const but add_yblock changes dst if add is 0 (this is never the case for dst=zero_dst
yading@10:     // const has only been removed from zero_dst to suppress a warning
yading@10:     static IDWTELEM zero_dst[4096]; //FIXME
yading@10:     const int b_stride = s->b_width << s->block_max_depth;
yading@10:     const int w= p->width;
yading@10:     const int h= p->height;
yading@10:     int distortion= 0;
yading@10:     int rate= 0;
yading@10:     const int penalty_factor= get_penalty_factor(s->lambda, s->lambda2, s->avctx->me_cmp);
yading@10: 
yading@10:     av_assert2(s->chroma_h_shift == s->chroma_v_shift); //obmc and square assumtions below
yading@10: 
yading@10:     for(i=0; i<9; i++){
yading@10:         int mb_x2= mb_x + (i%3) - 1;
yading@10:         int mb_y2= mb_y + (i/3) - 1;
yading@10:         int x= block_w*mb_x2 + block_w/2;
yading@10:         int y= block_h*mb_y2 + block_h/2;
yading@10: 
yading@10:         add_yblock(s, 0, NULL, zero_dst, dst, obmc,
yading@10:                    x, y, block_w, block_h, w, h, /*dst_stride*/0, ref_stride, obmc_stride, mb_x2, mb_y2, 1, 1, plane_index);
yading@10: 
yading@10:         //FIXME find a cleaner/simpler way to skip the outside stuff
yading@10:         for(y2= y; y2<0; y2++)
yading@10:             memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, block_w);
yading@10:         for(y2= h; y2<y+block_h; y2++)
yading@10:             memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, block_w);
yading@10:         if(x<0){
yading@10:             for(y2= y; y2<y+block_h; y2++)
yading@10:                 memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, -x);
yading@10:         }
yading@10:         if(x+block_w > w){
yading@10:             for(y2= y; y2<y+block_h; y2++)
yading@10:                 memcpy(dst + w + y2*ref_stride, src + w + y2*ref_stride, x+block_w - w);
yading@10:         }
yading@10: 
yading@10:         assert(block_w== 8 || block_w==16);
yading@10:         distortion += s->dsp.me_cmp[block_w==8](&s->m, src + x + y*ref_stride, dst + x + y*ref_stride, ref_stride, block_h);
yading@10:     }
yading@10: 
yading@10:     if(plane_index==0){
yading@10:         BlockNode *b= &s->block[mb_x+mb_y*b_stride];
yading@10:         int merged= same_block(b,b+1) && same_block(b,b+b_stride) && same_block(b,b+b_stride+1);
yading@10: 
yading@10: /* ..RRRr
yading@10:  * .RXXx.
yading@10:  * .RXXx.
yading@10:  * rxxx.
yading@10:  */
yading@10:         if(merged)
yading@10:             rate = get_block_bits(s, mb_x, mb_y, 2);
yading@10:         for(i=merged?4:0; i<9; i++){
yading@10:             static const int dxy[9][2] = {{0,0},{1,0},{0,1},{1,1},{2,0},{2,1},{-1,2},{0,2},{1,2}};
yading@10:             rate += get_block_bits(s, mb_x + dxy[i][0], mb_y + dxy[i][1], 1);
yading@10:         }
yading@10:     }
yading@10:     return distortion + rate*penalty_factor;
yading@10: }
yading@10: 
yading@10: static int encode_subband_c0run(SnowContext *s, SubBand *b, const IDWTELEM *src, const IDWTELEM *parent, int stride, int orientation){
yading@10:     const int w= b->width;
yading@10:     const int h= b->height;
yading@10:     int x, y;
yading@10: 
yading@10:     if(1){
yading@10:         int run=0;
yading@10:         int *runs = s->run_buffer;
yading@10:         int run_index=0;
yading@10:         int max_index;
yading@10: 
yading@10:         for(y=0; y<h; y++){
yading@10:             for(x=0; x<w; x++){
yading@10:                 int v, p=0;
yading@10:                 int /*ll=0, */l=0, lt=0, t=0, rt=0;
yading@10:                 v= src[x + y*stride];
yading@10: 
yading@10:                 if(y){
yading@10:                     t= src[x + (y-1)*stride];
yading@10:                     if(x){
yading@10:                         lt= src[x - 1 + (y-1)*stride];
yading@10:                     }
yading@10:                     if(x + 1 < w){
yading@10:                         rt= src[x + 1 + (y-1)*stride];
yading@10:                     }
yading@10:                 }
yading@10:                 if(x){
yading@10:                     l= src[x - 1 + y*stride];
yading@10:                     /*if(x > 1){
yading@10:                         if(orientation==1) ll= src[y + (x-2)*stride];
yading@10:                         else               ll= src[x - 2 + y*stride];
yading@10:                     }*/
yading@10:                 }
yading@10:                 if(parent){
yading@10:                     int px= x>>1;
yading@10:                     int py= y>>1;
yading@10:                     if(px<b->parent->width && py<b->parent->height)
yading@10:                         p= parent[px + py*2*stride];
yading@10:                 }
yading@10:                 if(!(/*ll|*/l|lt|t|rt|p)){
yading@10:                     if(v){
yading@10:                         runs[run_index++]= run;
yading@10:                         run=0;
yading@10:                     }else{
yading@10:                         run++;
yading@10:                     }
yading@10:                 }
yading@10:             }
yading@10:         }
yading@10:         max_index= run_index;
yading@10:         runs[run_index++]= run;
yading@10:         run_index=0;
yading@10:         run= runs[run_index++];
yading@10: 
yading@10:         put_symbol2(&s->c, b->state[30], max_index, 0);
yading@10:         if(run_index <= max_index)
yading@10:             put_symbol2(&s->c, b->state[1], run, 3);
yading@10: 
yading@10:         for(y=0; y<h; y++){
yading@10:             if(s->c.bytestream_end - s->c.bytestream < w*40){
yading@10:                 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
yading@10:                 return -1;
yading@10:             }
yading@10:             for(x=0; x<w; x++){
yading@10:                 int v, p=0;
yading@10:                 int /*ll=0, */l=0, lt=0, t=0, rt=0;
yading@10:                 v= src[x + y*stride];
yading@10: 
yading@10:                 if(y){
yading@10:                     t= src[x + (y-1)*stride];
yading@10:                     if(x){
yading@10:                         lt= src[x - 1 + (y-1)*stride];
yading@10:                     }
yading@10:                     if(x + 1 < w){
yading@10:                         rt= src[x + 1 + (y-1)*stride];
yading@10:                     }
yading@10:                 }
yading@10:                 if(x){
yading@10:                     l= src[x - 1 + y*stride];
yading@10:                     /*if(x > 1){
yading@10:                         if(orientation==1) ll= src[y + (x-2)*stride];
yading@10:                         else               ll= src[x - 2 + y*stride];
yading@10:                     }*/
yading@10:                 }
yading@10:                 if(parent){
yading@10:                     int px= x>>1;
yading@10:                     int py= y>>1;
yading@10:                     if(px<b->parent->width && py<b->parent->height)
yading@10:                         p= parent[px + py*2*stride];
yading@10:                 }
yading@10:                 if(/*ll|*/l|lt|t|rt|p){
yading@10:                     int context= av_log2(/*FFABS(ll) + */3*FFABS(l) + FFABS(lt) + 2*FFABS(t) + FFABS(rt) + FFABS(p));
yading@10: 
yading@10:                     put_rac(&s->c, &b->state[0][context], !!v);
yading@10:                 }else{
yading@10:                     if(!run){
yading@10:                         run= runs[run_index++];
yading@10: 
yading@10:                         if(run_index <= max_index)
yading@10:                             put_symbol2(&s->c, b->state[1], run, 3);
yading@10:                         assert(v);
yading@10:                     }else{
yading@10:                         run--;
yading@10:                         assert(!v);
yading@10:                     }
yading@10:                 }
yading@10:                 if(v){
yading@10:                     int context= av_log2(/*FFABS(ll) + */3*FFABS(l) + FFABS(lt) + 2*FFABS(t) + FFABS(rt) + FFABS(p));
yading@10:                     int l2= 2*FFABS(l) + (l<0);
yading@10:                     int t2= 2*FFABS(t) + (t<0);
yading@10: 
yading@10:                     put_symbol2(&s->c, b->state[context + 2], FFABS(v)-1, context-4);
yading@10:                     put_rac(&s->c, &b->state[0][16 + 1 + 3 + ff_quant3bA[l2&0xFF] + 3*ff_quant3bA[t2&0xFF]], v<0);
yading@10:                 }
yading@10:             }
yading@10:         }
yading@10:     }
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: static int encode_subband(SnowContext *s, SubBand *b, const IDWTELEM *src, const IDWTELEM *parent, int stride, int orientation){
yading@10: //    encode_subband_qtree(s, b, src, parent, stride, orientation);
yading@10: //    encode_subband_z0run(s, b, src, parent, stride, orientation);
yading@10:     return encode_subband_c0run(s, b, src, parent, stride, orientation);
yading@10: //    encode_subband_dzr(s, b, src, parent, stride, orientation);
yading@10: }
yading@10: 
yading@10: 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){
yading@10:     const int b_stride= s->b_width << s->block_max_depth;
yading@10:     BlockNode *block= &s->block[mb_x + mb_y * b_stride];
yading@10:     BlockNode backup= *block;
yading@10:     unsigned value;
yading@10:     int rd, index;
yading@10: 
yading@10:     assert(mb_x>=0 && mb_y>=0);
yading@10:     assert(mb_x<b_stride);
yading@10: 
yading@10:     if(intra){
yading@10:         block->color[0] = p[0];
yading@10:         block->color[1] = p[1];
yading@10:         block->color[2] = p[2];
yading@10:         block->type |= BLOCK_INTRA;
yading@10:     }else{
yading@10:         index= (p[0] + 31*p[1]) & (ME_CACHE_SIZE-1);
yading@10:         value= s->me_cache_generation + (p[0]>>10) + (p[1]<<6) + (block->ref<<12);
yading@10:         if(s->me_cache[index] == value)
yading@10:             return 0;
yading@10:         s->me_cache[index]= value;
yading@10: 
yading@10:         block->mx= p[0];
yading@10:         block->my= p[1];
yading@10:         block->type &= ~BLOCK_INTRA;
yading@10:     }
yading@10: 
yading@10:     rd= get_block_rd(s, mb_x, mb_y, 0, obmc_edged);
yading@10: 
yading@10: //FIXME chroma
yading@10:     if(rd < *best_rd){
yading@10:         *best_rd= rd;
yading@10:         return 1;
yading@10:     }else{
yading@10:         *block= backup;
yading@10:         return 0;
yading@10:     }
yading@10: }
yading@10: 
yading@10: /* special case for int[2] args we discard afterwards,
yading@10:  * fixes compilation problem with gcc 2.95 */
yading@10: 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){
yading@10:     int p[2] = {p0, p1};
yading@10:     return check_block(s, mb_x, mb_y, p, 0, obmc_edged, best_rd);
yading@10: }
yading@10: 
yading@10: 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){
yading@10:     const int b_stride= s->b_width << s->block_max_depth;
yading@10:     BlockNode *block= &s->block[mb_x + mb_y * b_stride];
yading@10:     BlockNode backup[4];
yading@10:     unsigned value;
yading@10:     int rd, index;
yading@10: 
yading@10:     /* We don't initialize backup[] during variable declaration, because
yading@10:      * that fails to compile on MSVC: "cannot convert from 'BlockNode' to
yading@10:      * 'int16_t'". */
yading@10:     backup[0] = block[0];
yading@10:     backup[1] = block[1];
yading@10:     backup[2] = block[b_stride];
yading@10:     backup[3] = block[b_stride + 1];
yading@10: 
yading@10:     assert(mb_x>=0 && mb_y>=0);
yading@10:     assert(mb_x<b_stride);
yading@10:     assert(((mb_x|mb_y)&1) == 0);
yading@10: 
yading@10:     index= (p0 + 31*p1) & (ME_CACHE_SIZE-1);
yading@10:     value= s->me_cache_generation + (p0>>10) + (p1<<6) + (block->ref<<12);
yading@10:     if(s->me_cache[index] == value)
yading@10:         return 0;
yading@10:     s->me_cache[index]= value;
yading@10: 
yading@10:     block->mx= p0;
yading@10:     block->my= p1;
yading@10:     block->ref= ref;
yading@10:     block->type &= ~BLOCK_INTRA;
yading@10:     block[1]= block[b_stride]= block[b_stride+1]= *block;
yading@10: 
yading@10:     rd= get_4block_rd(s, mb_x, mb_y, 0);
yading@10: 
yading@10: //FIXME chroma
yading@10:     if(rd < *best_rd){
yading@10:         *best_rd= rd;
yading@10:         return 1;
yading@10:     }else{
yading@10:         block[0]= backup[0];
yading@10:         block[1]= backup[1];
yading@10:         block[b_stride]= backup[2];
yading@10:         block[b_stride+1]= backup[3];
yading@10:         return 0;
yading@10:     }
yading@10: }
yading@10: 
yading@10: static void iterative_me(SnowContext *s){
yading@10:     int pass, mb_x, mb_y;
yading@10:     const int b_width = s->b_width  << s->block_max_depth;
yading@10:     const int b_height= s->b_height << s->block_max_depth;
yading@10:     const int b_stride= b_width;
yading@10:     int color[3];
yading@10: 
yading@10:     {
yading@10:         RangeCoder r = s->c;
yading@10:         uint8_t state[sizeof(s->block_state)];
yading@10:         memcpy(state, s->block_state, sizeof(s->block_state));
yading@10:         for(mb_y= 0; mb_y<s->b_height; mb_y++)
yading@10:             for(mb_x= 0; mb_x<s->b_width; mb_x++)
yading@10:                 encode_q_branch(s, 0, mb_x, mb_y);
yading@10:         s->c = r;
yading@10:         memcpy(s->block_state, state, sizeof(s->block_state));
yading@10:     }
yading@10: 
yading@10:     for(pass=0; pass<25; pass++){
yading@10:         int change= 0;
yading@10: 
yading@10:         for(mb_y= 0; mb_y<b_height; mb_y++){
yading@10:             for(mb_x= 0; mb_x<b_width; mb_x++){
yading@10:                 int dia_change, i, j, ref;
yading@10:                 int best_rd= INT_MAX, ref_rd;
yading@10:                 BlockNode backup, ref_b;
yading@10:                 const int index= mb_x + mb_y * b_stride;
yading@10:                 BlockNode *block= &s->block[index];
yading@10:                 BlockNode *tb =                   mb_y            ? &s->block[index-b_stride  ] : NULL;
yading@10:                 BlockNode *lb = mb_x                              ? &s->block[index         -1] : NULL;
yading@10:                 BlockNode *rb = mb_x+1<b_width                    ? &s->block[index         +1] : NULL;
yading@10:                 BlockNode *bb =                   mb_y+1<b_height ? &s->block[index+b_stride  ] : NULL;
yading@10:                 BlockNode *tlb= mb_x           && mb_y            ? &s->block[index-b_stride-1] : NULL;
yading@10:                 BlockNode *trb= mb_x+1<b_width && mb_y            ? &s->block[index-b_stride+1] : NULL;
yading@10:                 BlockNode *blb= mb_x           && mb_y+1<b_height ? &s->block[index+b_stride-1] : NULL;
yading@10:                 BlockNode *brb= mb_x+1<b_width && mb_y+1<b_height ? &s->block[index+b_stride+1] : NULL;
yading@10:                 const int b_w= (MB_SIZE >> s->block_max_depth);
yading@10:                 uint8_t obmc_edged[MB_SIZE * 2][MB_SIZE * 2];
yading@10: 
yading@10:                 if(pass && (block->type & BLOCK_OPT))
yading@10:                     continue;
yading@10:                 block->type |= BLOCK_OPT;
yading@10: 
yading@10:                 backup= *block;
yading@10: 
yading@10:                 if(!s->me_cache_generation)
yading@10:                     memset(s->me_cache, 0, sizeof(s->me_cache));
yading@10:                 s->me_cache_generation += 1<<22;
yading@10: 
yading@10:                 //FIXME precalculate
yading@10:                 {
yading@10:                     int x, y;
yading@10:                     for (y = 0; y < b_w * 2; y++)
yading@10:                         memcpy(obmc_edged[y], ff_obmc_tab[s->block_max_depth] + y * b_w * 2, b_w * 2);
yading@10:                     if(mb_x==0)
yading@10:                         for(y=0; y<b_w*2; y++)
yading@10:                             memset(obmc_edged[y], obmc_edged[y][0] + obmc_edged[y][b_w-1], b_w);
yading@10:                     if(mb_x==b_stride-1)
yading@10:                         for(y=0; y<b_w*2; y++)
yading@10:                             memset(obmc_edged[y]+b_w, obmc_edged[y][b_w] + obmc_edged[y][b_w*2-1], b_w);
yading@10:                     if(mb_y==0){
yading@10:                         for(x=0; x<b_w*2; x++)
yading@10:                             obmc_edged[0][x] += obmc_edged[b_w-1][x];
yading@10:                         for(y=1; y<b_w; y++)
yading@10:                             memcpy(obmc_edged[y], obmc_edged[0], b_w*2);
yading@10:                     }
yading@10:                     if(mb_y==b_height-1){
yading@10:                         for(x=0; x<b_w*2; x++)
yading@10:                             obmc_edged[b_w*2-1][x] += obmc_edged[b_w][x];
yading@10:                         for(y=b_w; y<b_w*2-1; y++)
yading@10:                             memcpy(obmc_edged[y], obmc_edged[b_w*2-1], b_w*2);
yading@10:                     }
yading@10:                 }
yading@10: 
yading@10:                 //skip stuff outside the picture
yading@10:                 if(mb_x==0 || mb_y==0 || mb_x==b_width-1 || mb_y==b_height-1){
yading@10:                     uint8_t *src= s->  input_picture.data[0];
yading@10:                     uint8_t *dst= s->current_picture.data[0];
yading@10:                     const int stride= s->current_picture.linesize[0];
yading@10:                     const int block_w= MB_SIZE >> s->block_max_depth;
yading@10:                     const int block_h= MB_SIZE >> s->block_max_depth;
yading@10:                     const int sx= block_w*mb_x - block_w/2;
yading@10:                     const int sy= block_h*mb_y - block_h/2;
yading@10:                     const int w= s->plane[0].width;
yading@10:                     const int h= s->plane[0].height;
yading@10:                     int y;
yading@10: 
yading@10:                     for(y=sy; y<0; y++)
yading@10:                         memcpy(dst + sx + y*stride, src + sx + y*stride, block_w*2);
yading@10:                     for(y=h; y<sy+block_h*2; y++)
yading@10:                         memcpy(dst + sx + y*stride, src + sx + y*stride, block_w*2);
yading@10:                     if(sx<0){
yading@10:                         for(y=sy; y<sy+block_h*2; y++)
yading@10:                             memcpy(dst + sx + y*stride, src + sx + y*stride, -sx);
yading@10:                     }
yading@10:                     if(sx+block_w*2 > w){
yading@10:                         for(y=sy; y<sy+block_h*2; y++)
yading@10:                             memcpy(dst + w + y*stride, src + w + y*stride, sx+block_w*2 - w);
yading@10:                     }
yading@10:                 }
yading@10: 
yading@10:                 // intra(black) = neighbors' contribution to the current block
yading@10:                 for(i=0; i<3; i++)
yading@10:                     color[i]= get_dc(s, mb_x, mb_y, i);
yading@10: 
yading@10:                 // get previous score (cannot be cached due to OBMC)
yading@10:                 if(pass > 0 && (block->type&BLOCK_INTRA)){
yading@10:                     int color0[3]= {block->color[0], block->color[1], block->color[2]};
yading@10:                     check_block(s, mb_x, mb_y, color0, 1, obmc_edged, &best_rd);
yading@10:                 }else
yading@10:                     check_block_inter(s, mb_x, mb_y, block->mx, block->my, obmc_edged, &best_rd);
yading@10: 
yading@10:                 ref_b= *block;
yading@10:                 ref_rd= best_rd;
yading@10:                 for(ref=0; ref < s->ref_frames; ref++){
yading@10:                     int16_t (*mvr)[2]= &s->ref_mvs[ref][index];
yading@10:                     if(s->ref_scores[ref][index] > s->ref_scores[ref_b.ref][index]*3/2) //FIXME tune threshold
yading@10:                         continue;
yading@10:                     block->ref= ref;
yading@10:                     best_rd= INT_MAX;
yading@10: 
yading@10:                     check_block_inter(s, mb_x, mb_y, mvr[0][0], mvr[0][1], obmc_edged, &best_rd);
yading@10:                     check_block_inter(s, mb_x, mb_y, 0, 0, obmc_edged, &best_rd);
yading@10:                     if(tb)
yading@10:                         check_block_inter(s, mb_x, mb_y, mvr[-b_stride][0], mvr[-b_stride][1], obmc_edged, &best_rd);
yading@10:                     if(lb)
yading@10:                         check_block_inter(s, mb_x, mb_y, mvr[-1][0], mvr[-1][1], obmc_edged, &best_rd);
yading@10:                     if(rb)
yading@10:                         check_block_inter(s, mb_x, mb_y, mvr[1][0], mvr[1][1], obmc_edged, &best_rd);
yading@10:                     if(bb)
yading@10:                         check_block_inter(s, mb_x, mb_y, mvr[b_stride][0], mvr[b_stride][1], obmc_edged, &best_rd);
yading@10: 
yading@10:                     /* fullpel ME */
yading@10:                     //FIXME avoid subpel interpolation / round to nearest integer
yading@10:                     do{
yading@10:                         dia_change=0;
yading@10:                         for(i=0; i<FFMAX(s->avctx->dia_size, 1); i++){
yading@10:                             for(j=0; j<i; j++){
yading@10:                                 dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+4*(i-j), block->my+(4*j), obmc_edged, &best_rd);
yading@10:                                 dia_change |= check_block_inter(s, mb_x, mb_y, block->mx-4*(i-j), block->my-(4*j), obmc_edged, &best_rd);
yading@10:                                 dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+4*(i-j), block->my-(4*j), obmc_edged, &best_rd);
yading@10:                                 dia_change |= check_block_inter(s, mb_x, mb_y, block->mx-4*(i-j), block->my+(4*j), obmc_edged, &best_rd);
yading@10:                             }
yading@10:                         }
yading@10:                     }while(dia_change);
yading@10:                     /* subpel ME */
yading@10:                     do{
yading@10:                         static const int square[8][2]= {{+1, 0},{-1, 0},{ 0,+1},{ 0,-1},{+1,+1},{-1,-1},{+1,-1},{-1,+1},};
yading@10:                         dia_change=0;
yading@10:                         for(i=0; i<8; i++)
yading@10:                             dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+square[i][0], block->my+square[i][1], obmc_edged, &best_rd);
yading@10:                     }while(dia_change);
yading@10:                     //FIXME or try the standard 2 pass qpel or similar
yading@10: 
yading@10:                     mvr[0][0]= block->mx;
yading@10:                     mvr[0][1]= block->my;
yading@10:                     if(ref_rd > best_rd){
yading@10:                         ref_rd= best_rd;
yading@10:                         ref_b= *block;
yading@10:                     }
yading@10:                 }
yading@10:                 best_rd= ref_rd;
yading@10:                 *block= ref_b;
yading@10:                 check_block(s, mb_x, mb_y, color, 1, obmc_edged, &best_rd);
yading@10:                 //FIXME RD style color selection
yading@10:                 if(!same_block(block, &backup)){
yading@10:                     if(tb ) tb ->type &= ~BLOCK_OPT;
yading@10:                     if(lb ) lb ->type &= ~BLOCK_OPT;
yading@10:                     if(rb ) rb ->type &= ~BLOCK_OPT;
yading@10:                     if(bb ) bb ->type &= ~BLOCK_OPT;
yading@10:                     if(tlb) tlb->type &= ~BLOCK_OPT;
yading@10:                     if(trb) trb->type &= ~BLOCK_OPT;
yading@10:                     if(blb) blb->type &= ~BLOCK_OPT;
yading@10:                     if(brb) brb->type &= ~BLOCK_OPT;
yading@10:                     change ++;
yading@10:                 }
yading@10:             }
yading@10:         }
yading@10:         av_log(s->avctx, AV_LOG_ERROR, "pass:%d changed:%d\n", pass, change);
yading@10:         if(!change)
yading@10:             break;
yading@10:     }
yading@10: 
yading@10:     if(s->block_max_depth == 1){
yading@10:         int change= 0;
yading@10:         for(mb_y= 0; mb_y<b_height; mb_y+=2){
yading@10:             for(mb_x= 0; mb_x<b_width; mb_x+=2){
yading@10:                 int i;
yading@10:                 int best_rd, init_rd;
yading@10:                 const int index= mb_x + mb_y * b_stride;
yading@10:                 BlockNode *b[4];
yading@10: 
yading@10:                 b[0]= &s->block[index];
yading@10:                 b[1]= b[0]+1;
yading@10:                 b[2]= b[0]+b_stride;
yading@10:                 b[3]= b[2]+1;
yading@10:                 if(same_block(b[0], b[1]) &&
yading@10:                    same_block(b[0], b[2]) &&
yading@10:                    same_block(b[0], b[3]))
yading@10:                     continue;
yading@10: 
yading@10:                 if(!s->me_cache_generation)
yading@10:                     memset(s->me_cache, 0, sizeof(s->me_cache));
yading@10:                 s->me_cache_generation += 1<<22;
yading@10: 
yading@10:                 init_rd= best_rd= get_4block_rd(s, mb_x, mb_y, 0);
yading@10: 
yading@10:                 //FIXME more multiref search?
yading@10:                 check_4block_inter(s, mb_x, mb_y,
yading@10:                                    (b[0]->mx + b[1]->mx + b[2]->mx + b[3]->mx + 2) >> 2,
yading@10:                                    (b[0]->my + b[1]->my + b[2]->my + b[3]->my + 2) >> 2, 0, &best_rd);
yading@10: 
yading@10:                 for(i=0; i<4; i++)
yading@10:                     if(!(b[i]->type&BLOCK_INTRA))
yading@10:                         check_4block_inter(s, mb_x, mb_y, b[i]->mx, b[i]->my, b[i]->ref, &best_rd);
yading@10: 
yading@10:                 if(init_rd != best_rd)
yading@10:                     change++;
yading@10:             }
yading@10:         }
yading@10:         av_log(s->avctx, AV_LOG_ERROR, "pass:4mv changed:%d\n", change*4);
yading@10:     }
yading@10: }
yading@10: 
yading@10: static void encode_blocks(SnowContext *s, int search){
yading@10:     int x, y;
yading@10:     int w= s->b_width;
yading@10:     int h= s->b_height;
yading@10: 
yading@10:     if(s->avctx->me_method == ME_ITER && !s->keyframe && search)
yading@10:         iterative_me(s);
yading@10: 
yading@10:     for(y=0; y<h; y++){
yading@10:         if(s->c.bytestream_end - s->c.bytestream < w*MB_SIZE*MB_SIZE*3){ //FIXME nicer limit
yading@10:             av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
yading@10:             return;
yading@10:         }
yading@10:         for(x=0; x<w; x++){
yading@10:             if(s->avctx->me_method == ME_ITER || !search)
yading@10:                 encode_q_branch2(s, 0, x, y);
yading@10:             else
yading@10:                 encode_q_branch (s, 0, x, y);
yading@10:         }
yading@10:     }
yading@10: }
yading@10: 
yading@10: static void quantize(SnowContext *s, SubBand *b, IDWTELEM *dst, DWTELEM *src, int stride, int bias){
yading@10:     const int w= b->width;
yading@10:     const int h= b->height;
yading@10:     const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16);
yading@10:     const int qmul= ff_qexp[qlog&(QROOT-1)]<<((qlog>>QSHIFT) + ENCODER_EXTRA_BITS);
yading@10:     int x,y, thres1, thres2;
yading@10: 
yading@10:     if(s->qlog == LOSSLESS_QLOG){
yading@10:         for(y=0; y<h; y++)
yading@10:             for(x=0; x<w; x++)
yading@10:                 dst[x + y*stride]= src[x + y*stride];
yading@10:         return;
yading@10:     }
yading@10: 
yading@10:     bias= bias ? 0 : (3*qmul)>>3;
yading@10:     thres1= ((qmul - bias)>>QEXPSHIFT) - 1;
yading@10:     thres2= 2*thres1;
yading@10: 
yading@10:     if(!bias){
yading@10:         for(y=0; y<h; y++){
yading@10:             for(x=0; x<w; x++){
yading@10:                 int i= src[x + y*stride];
yading@10: 
yading@10:                 if((unsigned)(i+thres1) > thres2){
yading@10:                     if(i>=0){
yading@10:                         i<<= QEXPSHIFT;
yading@10:                         i/= qmul; //FIXME optimize
yading@10:                         dst[x + y*stride]=  i;
yading@10:                     }else{
yading@10:                         i= -i;
yading@10:                         i<<= QEXPSHIFT;
yading@10:                         i/= qmul; //FIXME optimize
yading@10:                         dst[x + y*stride]= -i;
yading@10:                     }
yading@10:                 }else
yading@10:                     dst[x + y*stride]= 0;
yading@10:             }
yading@10:         }
yading@10:     }else{
yading@10:         for(y=0; y<h; y++){
yading@10:             for(x=0; x<w; x++){
yading@10:                 int i= src[x + y*stride];
yading@10: 
yading@10:                 if((unsigned)(i+thres1) > thres2){
yading@10:                     if(i>=0){
yading@10:                         i<<= QEXPSHIFT;
yading@10:                         i= (i + bias) / qmul; //FIXME optimize
yading@10:                         dst[x + y*stride]=  i;
yading@10:                     }else{
yading@10:                         i= -i;
yading@10:                         i<<= QEXPSHIFT;
yading@10:                         i= (i + bias) / qmul; //FIXME optimize
yading@10:                         dst[x + y*stride]= -i;
yading@10:                     }
yading@10:                 }else
yading@10:                     dst[x + y*stride]= 0;
yading@10:             }
yading@10:         }
yading@10:     }
yading@10: }
yading@10: 
yading@10: static void dequantize(SnowContext *s, SubBand *b, IDWTELEM *src, int stride){
yading@10:     const int w= b->width;
yading@10:     const int h= b->height;
yading@10:     const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16);
yading@10:     const int qmul= ff_qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
yading@10:     const int qadd= (s->qbias*qmul)>>QBIAS_SHIFT;
yading@10:     int x,y;
yading@10: 
yading@10:     if(s->qlog == LOSSLESS_QLOG) return;
yading@10: 
yading@10:     for(y=0; y<h; y++){
yading@10:         for(x=0; x<w; x++){
yading@10:             int i= src[x + y*stride];
yading@10:             if(i<0){
yading@10:                 src[x + y*stride]= -((-i*qmul + qadd)>>(QEXPSHIFT)); //FIXME try different bias
yading@10:             }else if(i>0){
yading@10:                 src[x + y*stride]=  (( i*qmul + qadd)>>(QEXPSHIFT));
yading@10:             }
yading@10:         }
yading@10:     }
yading@10: }
yading@10: 
yading@10: static void decorrelate(SnowContext *s, SubBand *b, IDWTELEM *src, int stride, int inverse, int use_median){
yading@10:     const int w= b->width;
yading@10:     const int h= b->height;
yading@10:     int x,y;
yading@10: 
yading@10:     for(y=h-1; y>=0; y--){
yading@10:         for(x=w-1; x>=0; x--){
yading@10:             int i= x + y*stride;
yading@10: 
yading@10:             if(x){
yading@10:                 if(use_median){
yading@10:                     if(y && x+1<w) src[i] -= mid_pred(src[i - 1], src[i - stride], src[i - stride + 1]);
yading@10:                     else  src[i] -= src[i - 1];
yading@10:                 }else{
yading@10:                     if(y) src[i] -= mid_pred(src[i - 1], src[i - stride], src[i - 1] + src[i - stride] - src[i - 1 - stride]);
yading@10:                     else  src[i] -= src[i - 1];
yading@10:                 }
yading@10:             }else{
yading@10:                 if(y) src[i] -= src[i - stride];
yading@10:             }
yading@10:         }
yading@10:     }
yading@10: }
yading@10: 
yading@10: static void correlate(SnowContext *s, SubBand *b, IDWTELEM *src, int stride, int inverse, int use_median){
yading@10:     const int w= b->width;
yading@10:     const int h= b->height;
yading@10:     int x,y;
yading@10: 
yading@10:     for(y=0; y<h; y++){
yading@10:         for(x=0; x<w; x++){
yading@10:             int i= x + y*stride;
yading@10: 
yading@10:             if(x){
yading@10:                 if(use_median){
yading@10:                     if(y && x+1<w) src[i] += mid_pred(src[i - 1], src[i - stride], src[i - stride + 1]);
yading@10:                     else  src[i] += src[i - 1];
yading@10:                 }else{
yading@10:                     if(y) src[i] += mid_pred(src[i - 1], src[i - stride], src[i - 1] + src[i - stride] - src[i - 1 - stride]);
yading@10:                     else  src[i] += src[i - 1];
yading@10:                 }
yading@10:             }else{
yading@10:                 if(y) src[i] += src[i - stride];
yading@10:             }
yading@10:         }
yading@10:     }
yading@10: }
yading@10: 
yading@10: static void encode_qlogs(SnowContext *s){
yading@10:     int plane_index, level, orientation;
yading@10: 
yading@10:     for(plane_index=0; plane_index<2; plane_index++){
yading@10:         for(level=0; level<s->spatial_decomposition_count; level++){
yading@10:             for(orientation=level ? 1:0; orientation<4; orientation++){
yading@10:                 if(orientation==2) continue;
yading@10:                 put_symbol(&s->c, s->header_state, s->plane[plane_index].band[level][orientation].qlog, 1);
yading@10:             }
yading@10:         }
yading@10:     }
yading@10: }
yading@10: 
yading@10: static void encode_header(SnowContext *s){
yading@10:     int plane_index, i;
yading@10:     uint8_t kstate[32];
yading@10: 
yading@10:     memset(kstate, MID_STATE, sizeof(kstate));
yading@10: 
yading@10:     put_rac(&s->c, kstate, s->keyframe);
yading@10:     if(s->keyframe || s->always_reset){
yading@10:         ff_snow_reset_contexts(s);
yading@10:         s->last_spatial_decomposition_type=
yading@10:         s->last_qlog=
yading@10:         s->last_qbias=
yading@10:         s->last_mv_scale=
yading@10:         s->last_block_max_depth= 0;
yading@10:         for(plane_index=0; plane_index<2; plane_index++){
yading@10:             Plane *p= &s->plane[plane_index];
yading@10:             p->last_htaps=0;
yading@10:             p->last_diag_mc=0;
yading@10:             memset(p->last_hcoeff, 0, sizeof(p->last_hcoeff));
yading@10:         }
yading@10:     }
yading@10:     if(s->keyframe){
yading@10:         put_symbol(&s->c, s->header_state, s->version, 0);
yading@10:         put_rac(&s->c, s->header_state, s->always_reset);
yading@10:         put_symbol(&s->c, s->header_state, s->temporal_decomposition_type, 0);
yading@10:         put_symbol(&s->c, s->header_state, s->temporal_decomposition_count, 0);
yading@10:         put_symbol(&s->c, s->header_state, s->spatial_decomposition_count, 0);
yading@10:         put_symbol(&s->c, s->header_state, s->colorspace_type, 0);
yading@10:         put_symbol(&s->c, s->header_state, s->chroma_h_shift, 0);
yading@10:         put_symbol(&s->c, s->header_state, s->chroma_v_shift, 0);
yading@10:         put_rac(&s->c, s->header_state, s->spatial_scalability);
yading@10: //        put_rac(&s->c, s->header_state, s->rate_scalability);
yading@10:         put_symbol(&s->c, s->header_state, s->max_ref_frames-1, 0);
yading@10: 
yading@10:         encode_qlogs(s);
yading@10:     }
yading@10: 
yading@10:     if(!s->keyframe){
yading@10:         int update_mc=0;
yading@10:         for(plane_index=0; plane_index<2; plane_index++){
yading@10:             Plane *p= &s->plane[plane_index];
yading@10:             update_mc |= p->last_htaps   != p->htaps;
yading@10:             update_mc |= p->last_diag_mc != p->diag_mc;
yading@10:             update_mc |= !!memcmp(p->last_hcoeff, p->hcoeff, sizeof(p->hcoeff));
yading@10:         }
yading@10:         put_rac(&s->c, s->header_state, update_mc);
yading@10:         if(update_mc){
yading@10:             for(plane_index=0; plane_index<2; plane_index++){
yading@10:                 Plane *p= &s->plane[plane_index];
yading@10:                 put_rac(&s->c, s->header_state, p->diag_mc);
yading@10:                 put_symbol(&s->c, s->header_state, p->htaps/2-1, 0);
yading@10:                 for(i= p->htaps/2; i; i--)
yading@10:                     put_symbol(&s->c, s->header_state, FFABS(p->hcoeff[i]), 0);
yading@10:             }
yading@10:         }
yading@10:         if(s->last_spatial_decomposition_count != s->spatial_decomposition_count){
yading@10:             put_rac(&s->c, s->header_state, 1);
yading@10:             put_symbol(&s->c, s->header_state, s->spatial_decomposition_count, 0);
yading@10:             encode_qlogs(s);
yading@10:         }else
yading@10:             put_rac(&s->c, s->header_state, 0);
yading@10:     }
yading@10: 
yading@10:     put_symbol(&s->c, s->header_state, s->spatial_decomposition_type - s->last_spatial_decomposition_type, 1);
yading@10:     put_symbol(&s->c, s->header_state, s->qlog            - s->last_qlog    , 1);
yading@10:     put_symbol(&s->c, s->header_state, s->mv_scale        - s->last_mv_scale, 1);
yading@10:     put_symbol(&s->c, s->header_state, s->qbias           - s->last_qbias   , 1);
yading@10:     put_symbol(&s->c, s->header_state, s->block_max_depth - s->last_block_max_depth, 1);
yading@10: 
yading@10: }
yading@10: 
yading@10: static void update_last_header_values(SnowContext *s){
yading@10:     int plane_index;
yading@10: 
yading@10:     if(!s->keyframe){
yading@10:         for(plane_index=0; plane_index<2; plane_index++){
yading@10:             Plane *p= &s->plane[plane_index];
yading@10:             p->last_diag_mc= p->diag_mc;
yading@10:             p->last_htaps  = p->htaps;
yading@10:             memcpy(p->last_hcoeff, p->hcoeff, sizeof(p->hcoeff));
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     s->last_spatial_decomposition_type  = s->spatial_decomposition_type;
yading@10:     s->last_qlog                        = s->qlog;
yading@10:     s->last_qbias                       = s->qbias;
yading@10:     s->last_mv_scale                    = s->mv_scale;
yading@10:     s->last_block_max_depth             = s->block_max_depth;
yading@10:     s->last_spatial_decomposition_count = s->spatial_decomposition_count;
yading@10: }
yading@10: 
yading@10: static int qscale2qlog(int qscale){
yading@10:     return rint(QROOT*log2(qscale / (float)FF_QP2LAMBDA))
yading@10:            + 61*QROOT/8; ///< 64 > 60
yading@10: }
yading@10: 
yading@10: static int ratecontrol_1pass(SnowContext *s, AVFrame *pict)
yading@10: {
yading@10:     /* Estimate the frame's complexity as a sum of weighted dwt coefficients.
yading@10:      * FIXME we know exact mv bits at this point,
yading@10:      * but ratecontrol isn't set up to include them. */
yading@10:     uint32_t coef_sum= 0;
yading@10:     int level, orientation, delta_qlog;
yading@10: 
yading@10:     for(level=0; level<s->spatial_decomposition_count; level++){
yading@10:         for(orientation=level ? 1 : 0; orientation<4; orientation++){
yading@10:             SubBand *b= &s->plane[0].band[level][orientation];
yading@10:             IDWTELEM *buf= b->ibuf;
yading@10:             const int w= b->width;
yading@10:             const int h= b->height;
yading@10:             const int stride= b->stride;
yading@10:             const int qlog= av_clip(2*QROOT + b->qlog, 0, QROOT*16);
yading@10:             const int qmul= ff_qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
yading@10:             const int qdiv= (1<<16)/qmul;
yading@10:             int x, y;
yading@10:             //FIXME this is ugly
yading@10:             for(y=0; y<h; y++)
yading@10:                 for(x=0; x<w; x++)
yading@10:                     buf[x+y*stride]= b->buf[x+y*stride];
yading@10:             if(orientation==0)
yading@10:                 decorrelate(s, b, buf, stride, 1, 0);
yading@10:             for(y=0; y<h; y++)
yading@10:                 for(x=0; x<w; x++)
yading@10:                     coef_sum+= abs(buf[x+y*stride]) * qdiv >> 16;
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     /* ugly, ratecontrol just takes a sqrt again */
yading@10:     coef_sum = (uint64_t)coef_sum * coef_sum >> 16;
yading@10:     assert(coef_sum < INT_MAX);
yading@10: 
yading@10:     if(pict->pict_type == AV_PICTURE_TYPE_I){
yading@10:         s->m.current_picture.mb_var_sum= coef_sum;
yading@10:         s->m.current_picture.mc_mb_var_sum= 0;
yading@10:     }else{
yading@10:         s->m.current_picture.mc_mb_var_sum= coef_sum;
yading@10:         s->m.current_picture.mb_var_sum= 0;
yading@10:     }
yading@10: 
yading@10:     pict->quality= ff_rate_estimate_qscale(&s->m, 1);
yading@10:     if (pict->quality < 0)
yading@10:         return INT_MIN;
yading@10:     s->lambda= pict->quality * 3/2;
yading@10:     delta_qlog= qscale2qlog(pict->quality) - s->qlog;
yading@10:     s->qlog+= delta_qlog;
yading@10:     return delta_qlog;
yading@10: }
yading@10: 
yading@10: static void calculate_visual_weight(SnowContext *s, Plane *p){
yading@10:     int width = p->width;
yading@10:     int height= p->height;
yading@10:     int level, orientation, x, y;
yading@10: 
yading@10:     for(level=0; level<s->spatial_decomposition_count; level++){
yading@10:         for(orientation=level ? 1 : 0; orientation<4; orientation++){
yading@10:             SubBand *b= &p->band[level][orientation];
yading@10:             IDWTELEM *ibuf= b->ibuf;
yading@10:             int64_t error=0;
yading@10: 
yading@10:             memset(s->spatial_idwt_buffer, 0, sizeof(*s->spatial_idwt_buffer)*width*height);
yading@10:             ibuf[b->width/2 + b->height/2*b->stride]= 256*16;
yading@10:             ff_spatial_idwt(s->spatial_idwt_buffer, s->temp_idwt_buffer, width, height, width, s->spatial_decomposition_type, s->spatial_decomposition_count);
yading@10:             for(y=0; y<height; y++){
yading@10:                 for(x=0; x<width; x++){
yading@10:                     int64_t d= s->spatial_idwt_buffer[x + y*width]*16;
yading@10:                     error += d*d;
yading@10:                 }
yading@10:             }
yading@10: 
yading@10:             b->qlog= (int)(log(352256.0/sqrt(error)) / log(pow(2.0, 1.0/QROOT))+0.5);
yading@10:         }
yading@10:     }
yading@10: }
yading@10: 
yading@10: static int encode_frame(AVCodecContext *avctx, AVPacket *pkt,
yading@10:                         const AVFrame *pict, int *got_packet)
yading@10: {
yading@10:     SnowContext *s = avctx->priv_data;
yading@10:     RangeCoder * const c= &s->c;
yading@10:     AVFrame *pic = &s->new_picture;
yading@10:     const int width= s->avctx->width;
yading@10:     const int height= s->avctx->height;
yading@10:     int level, orientation, plane_index, i, y, ret;
yading@10:     uint8_t rc_header_bak[sizeof(s->header_state)];
yading@10:     uint8_t rc_block_bak[sizeof(s->block_state)];
yading@10: 
yading@10:     if ((ret = ff_alloc_packet2(avctx, pkt, s->b_width*s->b_height*MB_SIZE*MB_SIZE*3 + FF_MIN_BUFFER_SIZE)) < 0)
yading@10:         return ret;
yading@10: 
yading@10:     ff_init_range_encoder(c, pkt->data, pkt->size);
yading@10:     ff_build_rac_states(c, 0.05*(1LL<<32), 256-8);
yading@10: 
yading@10:     for(i=0; i<3; i++){
yading@10:         int hshift= i ? s->chroma_h_shift : 0;
yading@10:         int vshift= i ? s->chroma_v_shift : 0;
yading@10:         for(y=0; y<(height>>vshift); y++)
yading@10:             memcpy(&s->input_picture.data[i][y * s->input_picture.linesize[i]],
yading@10:                    &pict->data[i][y * pict->linesize[i]],
yading@10:                    width>>hshift);
yading@10:         s->dsp.draw_edges(s->input_picture.data[i], s->input_picture.linesize[i],
yading@10:                             width >> hshift, height >> vshift,
yading@10:                             EDGE_WIDTH >> hshift, EDGE_WIDTH >> vshift,
yading@10:                             EDGE_TOP | EDGE_BOTTOM);
yading@10: 
yading@10:     }
yading@10:     emms_c();
yading@10:     s->new_picture = *pict;
yading@10: 
yading@10:     s->m.picture_number= avctx->frame_number;
yading@10:     if(avctx->flags&CODEC_FLAG_PASS2){
yading@10:         s->m.pict_type = pic->pict_type = s->m.rc_context.entry[avctx->frame_number].new_pict_type;
yading@10:         s->keyframe = pic->pict_type == AV_PICTURE_TYPE_I;
yading@10:         if(!(avctx->flags&CODEC_FLAG_QSCALE)) {
yading@10:             pic->quality = ff_rate_estimate_qscale(&s->m, 0);
yading@10:             if (pic->quality < 0)
yading@10:                 return -1;
yading@10:         }
yading@10:     }else{
yading@10:         s->keyframe= avctx->gop_size==0 || avctx->frame_number % avctx->gop_size == 0;
yading@10:         s->m.pict_type = pic->pict_type = s->keyframe ? AV_PICTURE_TYPE_I : AV_PICTURE_TYPE_P;
yading@10:     }
yading@10: 
yading@10:     if(s->pass1_rc && avctx->frame_number == 0)
yading@10:         pic->quality = 2*FF_QP2LAMBDA;
yading@10:     if (pic->quality) {
yading@10:         s->qlog   = qscale2qlog(pic->quality);
yading@10:         s->lambda = pic->quality * 3/2;
yading@10:     }
yading@10:     if (s->qlog < 0 || (!pic->quality && (avctx->flags & CODEC_FLAG_QSCALE))) {
yading@10:         s->qlog= LOSSLESS_QLOG;
yading@10:         s->lambda = 0;
yading@10:     }//else keep previous frame's qlog until after motion estimation
yading@10: 
yading@10:     ff_snow_frame_start(s);
yading@10: 
yading@10:     s->m.current_picture_ptr= &s->m.current_picture;
yading@10:     s->m.last_picture.f.pts = s->m.current_picture.f.pts;
yading@10:     s->m.current_picture.f.pts = pict->pts;
yading@10:     if(pic->pict_type == AV_PICTURE_TYPE_P){
yading@10:         int block_width = (width +15)>>4;
yading@10:         int block_height= (height+15)>>4;
yading@10:         int stride= s->current_picture.linesize[0];
yading@10: 
yading@10:         assert(s->current_picture.data[0]);
yading@10:         assert(s->last_picture[0].data[0]);
yading@10: 
yading@10:         s->m.avctx= s->avctx;
yading@10:         s->m.current_picture.f.data[0] = s->current_picture.data[0];
yading@10:         s->m.   last_picture.f.data[0] = s->last_picture[0].data[0];
yading@10:         s->m.    new_picture.f.data[0] = s->  input_picture.data[0];
yading@10:         s->m.   last_picture_ptr= &s->m.   last_picture;
yading@10:         s->m.linesize=
yading@10:         s->m.   last_picture.f.linesize[0] =
yading@10:         s->m.    new_picture.f.linesize[0] =
yading@10:         s->m.current_picture.f.linesize[0] = stride;
yading@10:         s->m.uvlinesize= s->current_picture.linesize[1];
yading@10:         s->m.width = width;
yading@10:         s->m.height= height;
yading@10:         s->m.mb_width = block_width;
yading@10:         s->m.mb_height= block_height;
yading@10:         s->m.mb_stride=   s->m.mb_width+1;
yading@10:         s->m.b8_stride= 2*s->m.mb_width+1;
yading@10:         s->m.f_code=1;
yading@10:         s->m.pict_type = pic->pict_type;
yading@10:         s->m.me_method= s->avctx->me_method;
yading@10:         s->m.me.scene_change_score=0;
yading@10:         s->m.flags= s->avctx->flags;
yading@10:         s->m.quarter_sample= (s->avctx->flags & CODEC_FLAG_QPEL)!=0;
yading@10:         s->m.out_format= FMT_H263;
yading@10:         s->m.unrestricted_mv= 1;
yading@10: 
yading@10:         s->m.lambda = s->lambda;
yading@10:         s->m.qscale= (s->m.lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
yading@10:         s->lambda2= s->m.lambda2= (s->m.lambda*s->m.lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT;
yading@10: 
yading@10:         s->m.dsp= s->dsp; //move
yading@10:         s->m.hdsp = s->hdsp;
yading@10:         ff_init_me(&s->m);
yading@10:         s->hdsp = s->m.hdsp;
yading@10:         s->dsp= s->m.dsp;
yading@10:     }
yading@10: 
yading@10:     if(s->pass1_rc){
yading@10:         memcpy(rc_header_bak, s->header_state, sizeof(s->header_state));
yading@10:         memcpy(rc_block_bak, s->block_state, sizeof(s->block_state));
yading@10:     }
yading@10: 
yading@10: redo_frame:
yading@10: 
yading@10:     if (pic->pict_type == AV_PICTURE_TYPE_I)
yading@10:         s->spatial_decomposition_count= 5;
yading@10:     else
yading@10:         s->spatial_decomposition_count= 5;
yading@10: 
yading@10:     while(   !(width >>(s->chroma_h_shift + s->spatial_decomposition_count))
yading@10:           || !(height>>(s->chroma_v_shift + s->spatial_decomposition_count)))
yading@10:         s->spatial_decomposition_count--;
yading@10: 
yading@10:     s->m.pict_type = pic->pict_type;
yading@10:     s->qbias = pic->pict_type == AV_PICTURE_TYPE_P ? 2 : 0;
yading@10: 
yading@10:     ff_snow_common_init_after_header(avctx);
yading@10: 
yading@10:     if(s->last_spatial_decomposition_count != s->spatial_decomposition_count){
yading@10:         for(plane_index=0; plane_index<3; plane_index++){
yading@10:             calculate_visual_weight(s, &s->plane[plane_index]);
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     encode_header(s);
yading@10:     s->m.misc_bits = 8*(s->c.bytestream - s->c.bytestream_start);
yading@10:     encode_blocks(s, 1);
yading@10:     s->m.mv_bits = 8*(s->c.bytestream - s->c.bytestream_start) - s->m.misc_bits;
yading@10: 
yading@10:     for(plane_index=0; plane_index<3; plane_index++){
yading@10:         Plane *p= &s->plane[plane_index];
yading@10:         int w= p->width;
yading@10:         int h= p->height;
yading@10:         int x, y;
yading@10: //        int bits= put_bits_count(&s->c.pb);
yading@10: 
yading@10:         if (!s->memc_only) {
yading@10:             //FIXME optimize
yading@10:             if(pict->data[plane_index]) //FIXME gray hack
yading@10:                 for(y=0; y<h; y++){
yading@10:                     for(x=0; x<w; x++){
yading@10:                         s->spatial_idwt_buffer[y*w + x]= pict->data[plane_index][y*pict->linesize[plane_index] + x]<<FRAC_BITS;
yading@10:                     }
yading@10:                 }
yading@10:             predict_plane(s, s->spatial_idwt_buffer, plane_index, 0);
yading@10: 
yading@10:             if(   plane_index==0
yading@10:                && pic->pict_type == AV_PICTURE_TYPE_P
yading@10:                && !(avctx->flags&CODEC_FLAG_PASS2)
yading@10:                && s->m.me.scene_change_score > s->avctx->scenechange_threshold){
yading@10:                 ff_init_range_encoder(c, pkt->data, pkt->size);
yading@10:                 ff_build_rac_states(c, 0.05*(1LL<<32), 256-8);
yading@10:                 pic->pict_type= AV_PICTURE_TYPE_I;
yading@10:                 s->keyframe=1;
yading@10:                 s->current_picture.key_frame=1;
yading@10:                 goto redo_frame;
yading@10:             }
yading@10: 
yading@10:             if(s->qlog == LOSSLESS_QLOG){
yading@10:                 for(y=0; y<h; y++){
yading@10:                     for(x=0; x<w; x++){
yading@10:                         s->spatial_dwt_buffer[y*w + x]= (s->spatial_idwt_buffer[y*w + x] + (1<<(FRAC_BITS-1))-1)>>FRAC_BITS;
yading@10:                     }
yading@10:                 }
yading@10:             }else{
yading@10:                 for(y=0; y<h; y++){
yading@10:                     for(x=0; x<w; x++){
yading@10:                         s->spatial_dwt_buffer[y*w + x]=s->spatial_idwt_buffer[y*w + x]<<ENCODER_EXTRA_BITS;
yading@10:                     }
yading@10:                 }
yading@10:             }
yading@10: 
yading@10:             /*  if(QUANTIZE2)
yading@10:                 dwt_quantize(s, p, s->spatial_dwt_buffer, w, h, w, s->spatial_decomposition_type);
yading@10:             else*/
yading@10:                 ff_spatial_dwt(s->spatial_dwt_buffer, s->temp_dwt_buffer, w, h, w, s->spatial_decomposition_type, s->spatial_decomposition_count);
yading@10: 
yading@10:             if(s->pass1_rc && plane_index==0){
yading@10:                 int delta_qlog = ratecontrol_1pass(s, pic);
yading@10:                 if (delta_qlog <= INT_MIN)
yading@10:                     return -1;
yading@10:                 if(delta_qlog){
yading@10:                     //reordering qlog in the bitstream would eliminate this reset
yading@10:                     ff_init_range_encoder(c, pkt->data, pkt->size);
yading@10:                     memcpy(s->header_state, rc_header_bak, sizeof(s->header_state));
yading@10:                     memcpy(s->block_state, rc_block_bak, sizeof(s->block_state));
yading@10:                     encode_header(s);
yading@10:                     encode_blocks(s, 0);
yading@10:                 }
yading@10:             }
yading@10: 
yading@10:             for(level=0; level<s->spatial_decomposition_count; level++){
yading@10:                 for(orientation=level ? 1 : 0; orientation<4; orientation++){
yading@10:                     SubBand *b= &p->band[level][orientation];
yading@10: 
yading@10:                     if(!QUANTIZE2)
yading@10:                         quantize(s, b, b->ibuf, b->buf, b->stride, s->qbias);
yading@10:                     if(orientation==0)
yading@10:                         decorrelate(s, b, b->ibuf, b->stride, pic->pict_type == AV_PICTURE_TYPE_P, 0);
yading@10:                     if (!s->no_bitstream)
yading@10:                     encode_subband(s, b, b->ibuf, b->parent ? b->parent->ibuf : NULL, b->stride, orientation);
yading@10:                     assert(b->parent==NULL || b->parent->stride == b->stride*2);
yading@10:                     if(orientation==0)
yading@10:                         correlate(s, b, b->ibuf, b->stride, 1, 0);
yading@10:                 }
yading@10:             }
yading@10: 
yading@10:             for(level=0; level<s->spatial_decomposition_count; level++){
yading@10:                 for(orientation=level ? 1 : 0; orientation<4; orientation++){
yading@10:                     SubBand *b= &p->band[level][orientation];
yading@10: 
yading@10:                     dequantize(s, b, b->ibuf, b->stride);
yading@10:                 }
yading@10:             }
yading@10: 
yading@10:             ff_spatial_idwt(s->spatial_idwt_buffer, s->temp_idwt_buffer, w, h, w, s->spatial_decomposition_type, s->spatial_decomposition_count);
yading@10:             if(s->qlog == LOSSLESS_QLOG){
yading@10:                 for(y=0; y<h; y++){
yading@10:                     for(x=0; x<w; x++){
yading@10:                         s->spatial_idwt_buffer[y*w + x]<<=FRAC_BITS;
yading@10:                     }
yading@10:                 }
yading@10:             }
yading@10:             predict_plane(s, s->spatial_idwt_buffer, plane_index, 1);
yading@10:         }else{
yading@10:             //ME/MC only
yading@10:             if(pic->pict_type == AV_PICTURE_TYPE_I){
yading@10:                 for(y=0; y<h; y++){
yading@10:                     for(x=0; x<w; x++){
yading@10:                         s->current_picture.data[plane_index][y*s->current_picture.linesize[plane_index] + x]=
yading@10:                             pict->data[plane_index][y*pict->linesize[plane_index] + x];
yading@10:                     }
yading@10:                 }
yading@10:             }else{
yading@10:                 memset(s->spatial_idwt_buffer, 0, sizeof(IDWTELEM)*w*h);
yading@10:                 predict_plane(s, s->spatial_idwt_buffer, plane_index, 1);
yading@10:             }
yading@10:         }
yading@10:         if(s->avctx->flags&CODEC_FLAG_PSNR){
yading@10:             int64_t error= 0;
yading@10: 
yading@10:             if(pict->data[plane_index]) //FIXME gray hack
yading@10:                 for(y=0; y<h; y++){
yading@10:                     for(x=0; x<w; x++){
yading@10:                         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];
yading@10:                         error += d*d;
yading@10:                     }
yading@10:                 }
yading@10:             s->avctx->error[plane_index] += error;
yading@10:             s->current_picture.error[plane_index] = error;
yading@10:         }
yading@10: 
yading@10:     }
yading@10: 
yading@10:     update_last_header_values(s);
yading@10: 
yading@10:     ff_snow_release_buffer(avctx);
yading@10: 
yading@10:     s->current_picture.coded_picture_number = avctx->frame_number;
yading@10:     s->current_picture.pict_type = pict->pict_type;
yading@10:     s->current_picture.quality = pict->quality;
yading@10:     s->m.frame_bits = 8*(s->c.bytestream - s->c.bytestream_start);
yading@10:     s->m.p_tex_bits = s->m.frame_bits - s->m.misc_bits - s->m.mv_bits;
yading@10:     s->m.current_picture.f.display_picture_number =
yading@10:     s->m.current_picture.f.coded_picture_number   = avctx->frame_number;
yading@10:     s->m.current_picture.f.quality                = pic->quality;
yading@10:     s->m.total_bits += 8*(s->c.bytestream - s->c.bytestream_start);
yading@10:     if(s->pass1_rc)
yading@10:         if (ff_rate_estimate_qscale(&s->m, 0) < 0)
yading@10:             return -1;
yading@10:     if(avctx->flags&CODEC_FLAG_PASS1)
yading@10:         ff_write_pass1_stats(&s->m);
yading@10:     s->m.last_pict_type = s->m.pict_type;
yading@10:     avctx->frame_bits = s->m.frame_bits;
yading@10:     avctx->mv_bits = s->m.mv_bits;
yading@10:     avctx->misc_bits = s->m.misc_bits;
yading@10:     avctx->p_tex_bits = s->m.p_tex_bits;
yading@10: 
yading@10:     emms_c();
yading@10: 
yading@10:     pkt->size = ff_rac_terminate(c);
yading@10:     if (avctx->coded_frame->key_frame)
yading@10:         pkt->flags |= AV_PKT_FLAG_KEY;
yading@10:     *got_packet = 1;
yading@10: 
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: static av_cold int encode_end(AVCodecContext *avctx)
yading@10: {
yading@10:     SnowContext *s = avctx->priv_data;
yading@10: 
yading@10:     ff_snow_common_end(s);
yading@10:     av_frame_unref(&s->input_picture);
yading@10:     av_free(avctx->stats_out);
yading@10: 
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: #define OFFSET(x) offsetof(SnowContext, x)
yading@10: #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
yading@10: static const AVOption options[] = {
yading@10:     { "memc_only",      "Only do ME/MC (I frames -> ref, P frame -> ME+MC).",   OFFSET(memc_only), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE },
yading@10:     { "no_bitstream",   "Skip final bitstream writeout.",                    OFFSET(no_bitstream), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE },
yading@10:     { NULL },
yading@10: };
yading@10: 
yading@10: static const AVClass snowenc_class = {
yading@10:     .class_name = "snow encoder",
yading@10:     .item_name  = av_default_item_name,
yading@10:     .option     = options,
yading@10:     .version    = LIBAVUTIL_VERSION_INT,
yading@10: };
yading@10: 
yading@10: AVCodec ff_snow_encoder = {
yading@10:     .name           = "snow",
yading@10:     .type           = AVMEDIA_TYPE_VIDEO,
yading@10:     .id             = AV_CODEC_ID_SNOW,
yading@10:     .priv_data_size = sizeof(SnowContext),
yading@10:     .init           = encode_init,
yading@10:     .encode2        = encode_frame,
yading@10:     .close          = encode_end,
yading@10:     .pix_fmts       = (const enum AVPixelFormat[]){
yading@10:         AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV410P, AV_PIX_FMT_YUV444P,
yading@10:         AV_PIX_FMT_NONE
yading@10:     },
yading@10:     .long_name      = NULL_IF_CONFIG_SMALL("Snow"),
yading@10:     .priv_class     = &snowenc_class,
yading@10: };
yading@10: 
yading@10: 
yading@10: #ifdef TEST
yading@10: #undef malloc
yading@10: #undef free
yading@10: #undef printf
yading@10: 
yading@10: #include "libavutil/lfg.h"
yading@10: #include "libavutil/mathematics.h"
yading@10: 
yading@10: int main(void){
yading@10: #define width  256
yading@10: #define height 256
yading@10:     int buffer[2][width*height];
yading@10:     SnowContext s;
yading@10:     int i;
yading@10:     AVLFG prng;
yading@10:     s.spatial_decomposition_count=6;
yading@10:     s.spatial_decomposition_type=1;
yading@10: 
yading@10:     s.temp_dwt_buffer  = av_mallocz(width * sizeof(DWTELEM));
yading@10:     s.temp_idwt_buffer = av_mallocz(width * sizeof(IDWTELEM));
yading@10: 
yading@10:     av_lfg_init(&prng, 1);
yading@10: 
yading@10:     printf("testing 5/3 DWT\n");
yading@10:     for(i=0; i<width*height; i++)
yading@10:         buffer[0][i] = buffer[1][i] = av_lfg_get(&prng) % 54321 - 12345;
yading@10: 
yading@10:     ff_spatial_dwt(buffer[0], s.temp_dwt_buffer, width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
yading@10:     ff_spatial_idwt((IDWTELEM*)buffer[0], s.temp_idwt_buffer, width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
yading@10: 
yading@10:     for(i=0; i<width*height; i++)
yading@10:         if(buffer[0][i]!= buffer[1][i]) printf("fsck: %6d %12d %7d\n",i, buffer[0][i], buffer[1][i]);
yading@10: 
yading@10:     printf("testing 9/7 DWT\n");
yading@10:     s.spatial_decomposition_type=0;
yading@10:     for(i=0; i<width*height; i++)
yading@10:         buffer[0][i] = buffer[1][i] = av_lfg_get(&prng) % 54321 - 12345;
yading@10: 
yading@10:     ff_spatial_dwt(buffer[0], s.temp_dwt_buffer, width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
yading@10:     ff_spatial_idwt((IDWTELEM*)buffer[0], s.temp_idwt_buffer, width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
yading@10: 
yading@10:     for(i=0; i<width*height; i++)
yading@10:         if(FFABS(buffer[0][i] - buffer[1][i])>20) printf("fsck: %6d %12d %7d\n",i, buffer[0][i], buffer[1][i]);
yading@10: 
yading@10:     {
yading@10:     int level, orientation, x, y;
yading@10:     int64_t errors[8][4];
yading@10:     int64_t g=0;
yading@10: 
yading@10:         memset(errors, 0, sizeof(errors));
yading@10:         s.spatial_decomposition_count=3;
yading@10:         s.spatial_decomposition_type=0;
yading@10:         for(level=0; level<s.spatial_decomposition_count; level++){
yading@10:             for(orientation=level ? 1 : 0; orientation<4; orientation++){
yading@10:                 int w= width  >> (s.spatial_decomposition_count-level);
yading@10:                 int h= height >> (s.spatial_decomposition_count-level);
yading@10:                 int stride= width  << (s.spatial_decomposition_count-level);
yading@10:                 DWTELEM *buf= buffer[0];
yading@10:                 int64_t error=0;
yading@10: 
yading@10:                 if(orientation&1) buf+=w;
yading@10:                 if(orientation>1) buf+=stride>>1;
yading@10: 
yading@10:                 memset(buffer[0], 0, sizeof(int)*width*height);
yading@10:                 buf[w/2 + h/2*stride]= 256*256;
yading@10:                 ff_spatial_idwt((IDWTELEM*)buffer[0], s.temp_idwt_buffer, width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
yading@10:                 for(y=0; y<height; y++){
yading@10:                     for(x=0; x<width; x++){
yading@10:                         int64_t d= buffer[0][x + y*width];
yading@10:                         error += d*d;
yading@10:                         if(FFABS(width/2-x)<9 && FFABS(height/2-y)<9 && level==2) printf("%8"PRId64" ", d);
yading@10:                     }
yading@10:                     if(FFABS(height/2-y)<9 && level==2) printf("\n");
yading@10:                 }
yading@10:                 error= (int)(sqrt(error)+0.5);
yading@10:                 errors[level][orientation]= error;
yading@10:                 if(g) g=av_gcd(g, error);
yading@10:                 else g= error;
yading@10:             }
yading@10:         }
yading@10:         printf("static int const visual_weight[][4]={\n");
yading@10:         for(level=0; level<s.spatial_decomposition_count; level++){
yading@10:             printf("  {");
yading@10:             for(orientation=0; orientation<4; orientation++){
yading@10:                 printf("%8"PRId64",", errors[level][orientation]/g);
yading@10:             }
yading@10:             printf("},\n");
yading@10:         }
yading@10:         printf("};\n");
yading@10:         {
yading@10:             int level=2;
yading@10:             int w= width  >> (s.spatial_decomposition_count-level);
yading@10:             //int h= height >> (s.spatial_decomposition_count-level);
yading@10:             int stride= width  << (s.spatial_decomposition_count-level);
yading@10:             DWTELEM *buf= buffer[0];
yading@10:             int64_t error=0;
yading@10: 
yading@10:             buf+=w;
yading@10:             buf+=stride>>1;
yading@10: 
yading@10:             memset(buffer[0], 0, sizeof(int)*width*height);
yading@10:             for(y=0; y<height; y++){
yading@10:                 for(x=0; x<width; x++){
yading@10:                     int tab[4]={0,2,3,1};
yading@10:                     buffer[0][x+width*y]= 256*256*tab[(x&1) + 2*(y&1)];
yading@10:                 }
yading@10:             }
yading@10:             ff_spatial_dwt(buffer[0], s.temp_dwt_buffer, width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
yading@10:             for(y=0; y<height; y++){
yading@10:                 for(x=0; x<width; x++){
yading@10:                     int64_t d= buffer[0][x + y*width];
yading@10:                     error += d*d;
yading@10:                     if(FFABS(width/2-x)<9 && FFABS(height/2-y)<9) printf("%8"PRId64" ", d);
yading@10:                 }
yading@10:                 if(FFABS(height/2-y)<9) printf("\n");
yading@10:             }
yading@10:         }
yading@10: 
yading@10:     }
yading@10:     return 0;
yading@10: }
yading@10: #endif /* TEST */