- if(words < 16){
- for(i=0; i<words; i++) put_bits(pb, 16, be2me_16(srcw[i]));
- }else if(put_bits_count(pb)&7){
- for(i=0; i<words; i++) put_bits(pb, 16, be2me_16(srcw[i]));
- }else{
- for(i=0; put_bits_count(pb)&31; i++)
- put_bits(pb, 8, src[i]);
- flush_put_bits(pb);
- memcpy(pbBufPtr(pb), src+i, 2*words-i);
- skip_put_bytes(pb, 2*words-i);
- }
-
- put_bits(pb, bits, be2me_16(srcw[words])>>(16-bits));
-}
-
-static inline void copy_context_before_encode(MpegEncContext *d, MpegEncContext *s, int type){
- int i;
-
- memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
-
- /* mpeg1 */
- d->mb_skip_run= s->mb_skip_run;
- for(i=0; i<3; i++)
- d->last_dc[i]= s->last_dc[i];
-
- /* statistics */
- d->mv_bits= s->mv_bits;
- d->i_tex_bits= s->i_tex_bits;
- d->p_tex_bits= s->p_tex_bits;
- d->i_count= s->i_count;
- d->f_count= s->f_count;
- d->b_count= s->b_count;
- d->skip_count= s->skip_count;
- d->misc_bits= s->misc_bits;
- d->last_bits= 0;
-
- d->mb_skipped= 0;
- d->qscale= s->qscale;
- d->dquant= s->dquant;
-}
-
-static inline void copy_context_after_encode(MpegEncContext *d, MpegEncContext *s, int type){
- int i;
-
- memcpy(d->mv, s->mv, 2*4*2*sizeof(int));
- memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
-
- /* mpeg1 */
- d->mb_skip_run= s->mb_skip_run;
- for(i=0; i<3; i++)
- d->last_dc[i]= s->last_dc[i];
-
- /* statistics */
- d->mv_bits= s->mv_bits;
- d->i_tex_bits= s->i_tex_bits;
- d->p_tex_bits= s->p_tex_bits;
- d->i_count= s->i_count;
- d->f_count= s->f_count;
- d->b_count= s->b_count;
- d->skip_count= s->skip_count;
- d->misc_bits= s->misc_bits;
-
- d->mb_intra= s->mb_intra;
- d->mb_skipped= s->mb_skipped;
- d->mv_type= s->mv_type;
- d->mv_dir= s->mv_dir;
- d->pb= s->pb;
- if(s->data_partitioning){
- d->pb2= s->pb2;
- d->tex_pb= s->tex_pb;
- }
- d->block= s->block;
- for(i=0; i<8; i++)
- d->block_last_index[i]= s->block_last_index[i];
- d->interlaced_dct= s->interlaced_dct;
- d->qscale= s->qscale;
-}
-
-static inline void encode_mb_hq(MpegEncContext *s, MpegEncContext *backup, MpegEncContext *best, int type,
- PutBitContext pb[2], PutBitContext pb2[2], PutBitContext tex_pb[2],
- int *dmin, int *next_block, int motion_x, int motion_y)
-{
- int score;
- uint8_t *dest_backup[3];
-
- copy_context_before_encode(s, backup, type);
-
- s->block= s->blocks[*next_block];
- s->pb= pb[*next_block];
- if(s->data_partitioning){
- s->pb2 = pb2 [*next_block];
- s->tex_pb= tex_pb[*next_block];
- }
-
- if(*next_block){
- memcpy(dest_backup, s->dest, sizeof(s->dest));
- s->dest[0] = s->rd_scratchpad;
- s->dest[1] = s->rd_scratchpad + 16*s->linesize;
- s->dest[2] = s->rd_scratchpad + 16*s->linesize + 8;
- assert(s->linesize >= 32); //FIXME
- }
-
- encode_mb(s, motion_x, motion_y);
-
- score= put_bits_count(&s->pb);
- if(s->data_partitioning){
- score+= put_bits_count(&s->pb2);
- score+= put_bits_count(&s->tex_pb);
- }
-
- if(s->avctx->mb_decision == FF_MB_DECISION_RD){
- MPV_decode_mb(s, s->block);
-
- score *= s->lambda2;
- score += sse_mb(s) << FF_LAMBDA_SHIFT;
- }
-
- if(*next_block){
- memcpy(s->dest, dest_backup, sizeof(s->dest));
- }
-
- if(score<*dmin){
- *dmin= score;
- *next_block^=1;
-
- copy_context_after_encode(best, s, type);
- }
-}
-
-static int sse(MpegEncContext *s, uint8_t *src1, uint8_t *src2, int w, int h, int stride){
- uint32_t *sq = ff_squareTbl + 256;
- int acc=0;
- int x,y;
-
- if(w==16 && h==16)
- return s->dsp.sse[0](NULL, src1, src2, stride, 16);
- else if(w==8 && h==8)
- return s->dsp.sse[1](NULL, src1, src2, stride, 8);
-
- for(y=0; y<h; y++){
- for(x=0; x<w; x++){
- acc+= sq[src1[x + y*stride] - src2[x + y*stride]];
- }
- }
-
- assert(acc>=0);
-
- return acc;
-}
-
-static int sse_mb(MpegEncContext *s){
- int w= 16;
- int h= 16;
-
- if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
- if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
-
- if(w==16 && h==16)
- if(s->avctx->mb_cmp == FF_CMP_NSSE){
- return s->dsp.nsse[0](s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], s->linesize, 16)
- +s->dsp.nsse[1](s, s->new_picture.data[1] + s->mb_x*8 + s->mb_y*s->uvlinesize*8,s->dest[1], s->uvlinesize, 8)
- +s->dsp.nsse[1](s, s->new_picture.data[2] + s->mb_x*8 + s->mb_y*s->uvlinesize*8,s->dest[2], s->uvlinesize, 8);
- }else{
- return s->dsp.sse[0](NULL, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], s->linesize, 16)
- +s->dsp.sse[1](NULL, s->new_picture.data[1] + s->mb_x*8 + s->mb_y*s->uvlinesize*8,s->dest[1], s->uvlinesize, 8)
- +s->dsp.sse[1](NULL, s->new_picture.data[2] + s->mb_x*8 + s->mb_y*s->uvlinesize*8,s->dest[2], s->uvlinesize, 8);
- }
- else
- return sse(s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], w, h, s->linesize)
- +sse(s, s->new_picture.data[1] + s->mb_x*8 + s->mb_y*s->uvlinesize*8,s->dest[1], w>>1, h>>1, s->uvlinesize)
- +sse(s, s->new_picture.data[2] + s->mb_x*8 + s->mb_y*s->uvlinesize*8,s->dest[2], w>>1, h>>1, s->uvlinesize);
-}
-
-static int pre_estimate_motion_thread(AVCodecContext *c, void *arg){
- MpegEncContext *s= arg;
-
-
- s->me.pre_pass=1;
- s->me.dia_size= s->avctx->pre_dia_size;
- s->first_slice_line=1;
- for(s->mb_y= s->end_mb_y-1; s->mb_y >= s->start_mb_y; s->mb_y--) {
- for(s->mb_x=s->mb_width-1; s->mb_x >=0 ;s->mb_x--) {
- ff_pre_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
- }
- s->first_slice_line=0;
- }
-
- s->me.pre_pass=0;
-
- return 0;
-}
-
-static int estimate_motion_thread(AVCodecContext *c, void *arg){
- MpegEncContext *s= arg;
-
- ff_check_alignment();
-
- s->me.dia_size= s->avctx->dia_size;
- s->first_slice_line=1;
- for(s->mb_y= s->start_mb_y; s->mb_y < s->end_mb_y; s->mb_y++) {
- s->mb_x=0; //for block init below
- ff_init_block_index(s);
- for(s->mb_x=0; s->mb_x < s->mb_width; s->mb_x++) {
- s->block_index[0]+=2;
- s->block_index[1]+=2;
- s->block_index[2]+=2;
- s->block_index[3]+=2;
-
- /* compute motion vector & mb_type and store in context */
- if(s->pict_type==B_TYPE)
- ff_estimate_b_frame_motion(s, s->mb_x, s->mb_y);
- else
- ff_estimate_p_frame_motion(s, s->mb_x, s->mb_y);
- }
- s->first_slice_line=0;
- }
- return 0;
-}
-
-static int mb_var_thread(AVCodecContext *c, void *arg){
- MpegEncContext *s= arg;
- int mb_x, mb_y;
-
- ff_check_alignment();
-
- for(mb_y=s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
- for(mb_x=0; mb_x < s->mb_width; mb_x++) {
- int xx = mb_x * 16;
- int yy = mb_y * 16;
- uint8_t *pix = s->new_picture.data[0] + (yy * s->linesize) + xx;
- int varc;
- int sum = s->dsp.pix_sum(pix, s->linesize);
-
- varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)(sum*sum))>>8) + 500 + 128)>>8;
-
- s->current_picture.mb_var [s->mb_stride * mb_y + mb_x] = varc;
- s->current_picture.mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
- s->me.mb_var_sum_temp += varc;
- }
- }
- return 0;
-}
-
-static void write_slice_end(MpegEncContext *s){
- if(s->codec_id==CODEC_ID_MPEG4){
- if(s->partitioned_frame){
- ff_mpeg4_merge_partitions(s);
- }
-
- ff_mpeg4_stuffing(&s->pb);
- }else if(ENABLE_MJPEG_ENCODER && s->out_format == FMT_MJPEG){
- ff_mjpeg_stuffing(&s->pb);
- }
-
- align_put_bits(&s->pb);
- flush_put_bits(&s->pb);
-
- if((s->flags&CODEC_FLAG_PASS1) && !s->partitioned_frame)
- s->misc_bits+= get_bits_diff(s);
-}
-
-static int encode_thread(AVCodecContext *c, void *arg){
- MpegEncContext *s= arg;
- int mb_x, mb_y, pdif = 0;
- int i, j;
- MpegEncContext best_s, backup_s;
- uint8_t bit_buf[2][MAX_MB_BYTES];
- uint8_t bit_buf2[2][MAX_MB_BYTES];
- uint8_t bit_buf_tex[2][MAX_MB_BYTES];
- PutBitContext pb[2], pb2[2], tex_pb[2];
-//printf("%d->%d\n", s->resync_mb_y, s->end_mb_y);
-
- ff_check_alignment();
-
- for(i=0; i<2; i++){
- init_put_bits(&pb [i], bit_buf [i], MAX_MB_BYTES);
- init_put_bits(&pb2 [i], bit_buf2 [i], MAX_MB_BYTES);
- init_put_bits(&tex_pb[i], bit_buf_tex[i], MAX_MB_BYTES);
- }
-
- s->last_bits= put_bits_count(&s->pb);
- s->mv_bits=0;
- s->misc_bits=0;
- s->i_tex_bits=0;
- s->p_tex_bits=0;
- s->i_count=0;
- s->f_count=0;
- s->b_count=0;
- s->skip_count=0;
-
- for(i=0; i<3; i++){
- /* init last dc values */
- /* note: quant matrix value (8) is implied here */
- s->last_dc[i] = 128 << s->intra_dc_precision;
-
- s->current_picture.error[i] = 0;
- }
- s->mb_skip_run = 0;
- memset(s->last_mv, 0, sizeof(s->last_mv));
-
- s->last_mv_dir = 0;
-
- switch(s->codec_id){
- case CODEC_ID_H263:
- case CODEC_ID_H263P:
- case CODEC_ID_FLV1:
- s->gob_index = ff_h263_get_gob_height(s);
- break;
- case CODEC_ID_MPEG4:
- if(s->partitioned_frame)
- ff_mpeg4_init_partitions(s);
- break;
- }
-
- s->resync_mb_x=0;
- s->resync_mb_y=0;
- s->first_slice_line = 1;
- s->ptr_lastgob = s->pb.buf;
- for(mb_y= s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
-// printf("row %d at %X\n", s->mb_y, (int)s);
- s->mb_x=0;
- s->mb_y= mb_y;
-
- ff_set_qscale(s, s->qscale);
- ff_init_block_index(s);
-
- for(mb_x=0; mb_x < s->mb_width; mb_x++) {
- int xy= mb_y*s->mb_stride + mb_x; // removed const, H261 needs to adjust this
- int mb_type= s->mb_type[xy];
-// int d;
- int dmin= INT_MAX;
- int dir;
-
- if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < MAX_MB_BYTES){
- av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
- return -1;
- }
- if(s->data_partitioning){
- if( s->pb2 .buf_end - s->pb2 .buf - (put_bits_count(&s-> pb2)>>3) < MAX_MB_BYTES
- || s->tex_pb.buf_end - s->tex_pb.buf - (put_bits_count(&s->tex_pb )>>3) < MAX_MB_BYTES){
- av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
- return -1;
- }
- }
-
- s->mb_x = mb_x;
- s->mb_y = mb_y; // moved into loop, can get changed by H.261
- ff_update_block_index(s);
-
- if(ENABLE_H261_ENCODER && s->codec_id == CODEC_ID_H261){
- ff_h261_reorder_mb_index(s);
- xy= s->mb_y*s->mb_stride + s->mb_x;
- mb_type= s->mb_type[xy];
- }
-
- /* write gob / video packet header */
- if(s->rtp_mode){
- int current_packet_size, is_gob_start;
-
- current_packet_size= ((put_bits_count(&s->pb)+7)>>3) - (s->ptr_lastgob - s->pb.buf);
-
- is_gob_start= s->avctx->rtp_payload_size && current_packet_size >= s->avctx->rtp_payload_size && mb_y + mb_x>0;
-
- if(s->start_mb_y == mb_y && mb_y > 0 && mb_x==0) is_gob_start=1;
-
- switch(s->codec_id){
- case CODEC_ID_H263:
- case CODEC_ID_H263P:
- if(!s->h263_slice_structured)
- if(s->mb_x || s->mb_y%s->gob_index) is_gob_start=0;
- break;
- case CODEC_ID_MPEG2VIDEO:
- if(s->mb_x==0 && s->mb_y!=0) is_gob_start=1;
- case CODEC_ID_MPEG1VIDEO:
- if(s->mb_skip_run) is_gob_start=0;
- break;
- }
-
- if(is_gob_start){
- if(s->start_mb_y != mb_y || mb_x!=0){
- write_slice_end(s);
-
- if(s->codec_id==CODEC_ID_MPEG4 && s->partitioned_frame){
- ff_mpeg4_init_partitions(s);
- }
- }
-
- assert((put_bits_count(&s->pb)&7) == 0);
- current_packet_size= pbBufPtr(&s->pb) - s->ptr_lastgob;
-
- if(s->avctx->error_rate && s->resync_mb_x + s->resync_mb_y > 0){
- int r= put_bits_count(&s->pb)/8 + s->picture_number + 16 + s->mb_x + s->mb_y;
- int d= 100 / s->avctx->error_rate;
- if(r % d == 0){
- current_packet_size=0;
-#ifndef ALT_BITSTREAM_WRITER
- s->pb.buf_ptr= s->ptr_lastgob;
-#endif
- assert(pbBufPtr(&s->pb) == s->ptr_lastgob);
- }
- }
-
- if (s->avctx->rtp_callback){
- int number_mb = (mb_y - s->resync_mb_y)*s->mb_width + mb_x - s->resync_mb_x;
- s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, current_packet_size, number_mb);
- }
-
- switch(s->codec_id){
- case CODEC_ID_MPEG4:
- ff_mpeg4_encode_video_packet_header(s);
- ff_mpeg4_clean_buffers(s);
- break;
- case CODEC_ID_MPEG1VIDEO:
- case CODEC_ID_MPEG2VIDEO:
- ff_mpeg1_encode_slice_header(s);
- ff_mpeg1_clean_buffers(s);
- break;
- case CODEC_ID_H263:
- case CODEC_ID_H263P:
- h263_encode_gob_header(s, mb_y);
- break;
- }
-
- if(s->flags&CODEC_FLAG_PASS1){
- int bits= put_bits_count(&s->pb);
- s->misc_bits+= bits - s->last_bits;
- s->last_bits= bits;
- }
-
- s->ptr_lastgob += current_packet_size;
- s->first_slice_line=1;
- s->resync_mb_x=mb_x;
- s->resync_mb_y=mb_y;
- }
- }
-
- if( (s->resync_mb_x == s->mb_x)
- && s->resync_mb_y+1 == s->mb_y){
- s->first_slice_line=0;
- }
-
- s->mb_skipped=0;
- s->dquant=0; //only for QP_RD
-
- if(mb_type & (mb_type-1) || (s->flags & CODEC_FLAG_QP_RD)){ // more than 1 MB type possible or CODEC_FLAG_QP_RD
- int next_block=0;
- int pb_bits_count, pb2_bits_count, tex_pb_bits_count;
-
- copy_context_before_encode(&backup_s, s, -1);
- backup_s.pb= s->pb;
- best_s.data_partitioning= s->data_partitioning;
- best_s.partitioned_frame= s->partitioned_frame;
- if(s->data_partitioning){
- backup_s.pb2= s->pb2;
- backup_s.tex_pb= s->tex_pb;
- }
-
- if(mb_type&CANDIDATE_MB_TYPE_INTER){
- s->mv_dir = MV_DIR_FORWARD;
- s->mv_type = MV_TYPE_16X16;
- s->mb_intra= 0;
- s->mv[0][0][0] = s->p_mv_table[xy][0];
- s->mv[0][0][1] = s->p_mv_table[xy][1];
- encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER, pb, pb2, tex_pb,
- &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
- }
- if(mb_type&CANDIDATE_MB_TYPE_INTER_I){
- s->mv_dir = MV_DIR_FORWARD;
- s->mv_type = MV_TYPE_FIELD;
- s->mb_intra= 0;
- for(i=0; i<2; i++){
- j= s->field_select[0][i] = s->p_field_select_table[i][xy];
- s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
- s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
- }
- encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER_I, pb, pb2, tex_pb,
- &dmin, &next_block, 0, 0);
- }
- if(mb_type&CANDIDATE_MB_TYPE_SKIPPED){
- s->mv_dir = MV_DIR_FORWARD;
- s->mv_type = MV_TYPE_16X16;
- s->mb_intra= 0;
- s->mv[0][0][0] = 0;
- s->mv[0][0][1] = 0;
- encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_SKIPPED, pb, pb2, tex_pb,
- &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
- }
- if(mb_type&CANDIDATE_MB_TYPE_INTER4V){
- s->mv_dir = MV_DIR_FORWARD;
- s->mv_type = MV_TYPE_8X8;
- s->mb_intra= 0;
- for(i=0; i<4; i++){
- s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
- s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
- }
- encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER4V, pb, pb2, tex_pb,
- &dmin, &next_block, 0, 0);
- }
- if(mb_type&CANDIDATE_MB_TYPE_FORWARD){
- s->mv_dir = MV_DIR_FORWARD;
- s->mv_type = MV_TYPE_16X16;
- s->mb_intra= 0;
- s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
- s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
- encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD, pb, pb2, tex_pb,
- &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
- }
- if(mb_type&CANDIDATE_MB_TYPE_BACKWARD){
- s->mv_dir = MV_DIR_BACKWARD;
- s->mv_type = MV_TYPE_16X16;
- s->mb_intra= 0;
- s->mv[1][0][0] = s->b_back_mv_table[xy][0];
- s->mv[1][0][1] = s->b_back_mv_table[xy][1];
- encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD, pb, pb2, tex_pb,
- &dmin, &next_block, s->mv[1][0][0], s->mv[1][0][1]);
- }
- if(mb_type&CANDIDATE_MB_TYPE_BIDIR){
- s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
- s->mv_type = MV_TYPE_16X16;
- s->mb_intra= 0;
- s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
- s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
- s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
- s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
- encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR, pb, pb2, tex_pb,
- &dmin, &next_block, 0, 0);
- }
- if(mb_type&CANDIDATE_MB_TYPE_FORWARD_I){
- s->mv_dir = MV_DIR_FORWARD;
- s->mv_type = MV_TYPE_FIELD;
- s->mb_intra= 0;
- for(i=0; i<2; i++){
- j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
- s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
- s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
- }
- encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_FORWARD_I, pb, pb2, tex_pb,
- &dmin, &next_block, 0, 0);
- }
- if(mb_type&CANDIDATE_MB_TYPE_BACKWARD_I){
- s->mv_dir = MV_DIR_BACKWARD;
- s->mv_type = MV_TYPE_FIELD;
- s->mb_intra= 0;
- for(i=0; i<2; i++){
- j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
- s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
- s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
- }
- encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BACKWARD_I, pb, pb2, tex_pb,
- &dmin, &next_block, 0, 0);
- }
- if(mb_type&CANDIDATE_MB_TYPE_BIDIR_I){
- s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
- s->mv_type = MV_TYPE_FIELD;
- s->mb_intra= 0;
- for(dir=0; dir<2; dir++){
- for(i=0; i<2; i++){
- j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
- s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
- s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
- }
- }
- encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_BIDIR_I, pb, pb2, tex_pb,
- &dmin, &next_block, 0, 0);
- }
- if(mb_type&CANDIDATE_MB_TYPE_INTRA){
- s->mv_dir = 0;
- s->mv_type = MV_TYPE_16X16;
- s->mb_intra= 1;
- s->mv[0][0][0] = 0;
- s->mv[0][0][1] = 0;
- encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTRA, pb, pb2, tex_pb,
- &dmin, &next_block, 0, 0);
- if(s->h263_pred || s->h263_aic){
- if(best_s.mb_intra)
- s->mbintra_table[mb_x + mb_y*s->mb_stride]=1;
- else
- ff_clean_intra_table_entries(s); //old mode?
- }
- }
-
- if((s->flags & CODEC_FLAG_QP_RD) && dmin < INT_MAX){
- if(best_s.mv_type==MV_TYPE_16X16){ //FIXME move 4mv after QPRD
- const int last_qp= backup_s.qscale;
- int qpi, qp, dc[6];
- DCTELEM ac[6][16];
- const int mvdir= (best_s.mv_dir&MV_DIR_BACKWARD) ? 1 : 0;
- static const int dquant_tab[4]={-1,1,-2,2};
-
- assert(backup_s.dquant == 0);
-
- //FIXME intra
- s->mv_dir= best_s.mv_dir;
- s->mv_type = MV_TYPE_16X16;
- s->mb_intra= best_s.mb_intra;
- s->mv[0][0][0] = best_s.mv[0][0][0];
- s->mv[0][0][1] = best_s.mv[0][0][1];
- s->mv[1][0][0] = best_s.mv[1][0][0];
- s->mv[1][0][1] = best_s.mv[1][0][1];
-
- qpi = s->pict_type == B_TYPE ? 2 : 0;
- for(; qpi<4; qpi++){
- int dquant= dquant_tab[qpi];
- qp= last_qp + dquant;
- if(qp < s->avctx->qmin || qp > s->avctx->qmax)
- continue;
- backup_s.dquant= dquant;
- if(s->mb_intra && s->dc_val[0]){
- for(i=0; i<6; i++){
- dc[i]= s->dc_val[0][ s->block_index[i] ];
- memcpy(ac[i], s->ac_val[0][s->block_index[i]], sizeof(DCTELEM)*16);
- }
- }
-
- encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
- &dmin, &next_block, s->mv[mvdir][0][0], s->mv[mvdir][0][1]);
- if(best_s.qscale != qp){
- if(s->mb_intra && s->dc_val[0]){
- for(i=0; i<6; i++){
- s->dc_val[0][ s->block_index[i] ]= dc[i];
- memcpy(s->ac_val[0][s->block_index[i]], ac[i], sizeof(DCTELEM)*16);
- }
- }
- }
- }
- }
- }
- if(mb_type&CANDIDATE_MB_TYPE_DIRECT){
- int mx= s->b_direct_mv_table[xy][0];
- int my= s->b_direct_mv_table[xy][1];
-
- backup_s.dquant = 0;
- s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
- s->mb_intra= 0;
- ff_mpeg4_set_direct_mv(s, mx, my);
- encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
- &dmin, &next_block, mx, my);
- }
- if(mb_type&CANDIDATE_MB_TYPE_DIRECT0){
- backup_s.dquant = 0;
- s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
- s->mb_intra= 0;
- ff_mpeg4_set_direct_mv(s, 0, 0);
- encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_DIRECT, pb, pb2, tex_pb,
- &dmin, &next_block, 0, 0);
- }
- if(!best_s.mb_intra && s->flags2&CODEC_FLAG2_SKIP_RD){
- int coded=0;
- for(i=0; i<6; i++)
- coded |= s->block_last_index[i];
- if(coded){
- int mx,my;
- memcpy(s->mv, best_s.mv, sizeof(s->mv));
- if(best_s.mv_dir & MV_DIRECT){
- mx=my=0; //FIXME find the one we actually used
- ff_mpeg4_set_direct_mv(s, mx, my);
- }else if(best_s.mv_dir&MV_DIR_BACKWARD){
- mx= s->mv[1][0][0];
- my= s->mv[1][0][1];
- }else{
- mx= s->mv[0][0][0];
- my= s->mv[0][0][1];
- }
-
- s->mv_dir= best_s.mv_dir;
- s->mv_type = best_s.mv_type;
- s->mb_intra= 0;
-/* s->mv[0][0][0] = best_s.mv[0][0][0];
- s->mv[0][0][1] = best_s.mv[0][0][1];
- s->mv[1][0][0] = best_s.mv[1][0][0];
- s->mv[1][0][1] = best_s.mv[1][0][1];*/
- backup_s.dquant= 0;
- s->skipdct=1;
- encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
- &dmin, &next_block, mx, my);
- s->skipdct=0;
- }
- }
-
- s->current_picture.qscale_table[xy]= best_s.qscale;
-
- copy_context_after_encode(s, &best_s, -1);
-
- pb_bits_count= put_bits_count(&s->pb);
- flush_put_bits(&s->pb);
- ff_copy_bits(&backup_s.pb, bit_buf[next_block^1], pb_bits_count);
- s->pb= backup_s.pb;
-
- if(s->data_partitioning){
- pb2_bits_count= put_bits_count(&s->pb2);
- flush_put_bits(&s->pb2);
- ff_copy_bits(&backup_s.pb2, bit_buf2[next_block^1], pb2_bits_count);
- s->pb2= backup_s.pb2;
-
- tex_pb_bits_count= put_bits_count(&s->tex_pb);
- flush_put_bits(&s->tex_pb);
- ff_copy_bits(&backup_s.tex_pb, bit_buf_tex[next_block^1], tex_pb_bits_count);
- s->tex_pb= backup_s.tex_pb;
- }
- s->last_bits= put_bits_count(&s->pb);
-
- if (s->out_format == FMT_H263 && s->pict_type!=B_TYPE)
- ff_h263_update_motion_val(s);
-
- if(next_block==0){ //FIXME 16 vs linesize16
- s->dsp.put_pixels_tab[0][0](s->dest[0], s->rd_scratchpad , s->linesize ,16);
- s->dsp.put_pixels_tab[1][0](s->dest[1], s->rd_scratchpad + 16*s->linesize , s->uvlinesize, 8);
- s->dsp.put_pixels_tab[1][0](s->dest[2], s->rd_scratchpad + 16*s->linesize + 8, s->uvlinesize, 8);
- }
-
- if(s->avctx->mb_decision == FF_MB_DECISION_BITS)
- MPV_decode_mb(s, s->block);
- } else {
- int motion_x = 0, motion_y = 0;
- s->mv_type=MV_TYPE_16X16;
- // only one MB-Type possible
-
- switch(mb_type){
- case CANDIDATE_MB_TYPE_INTRA:
- s->mv_dir = 0;
- s->mb_intra= 1;
- motion_x= s->mv[0][0][0] = 0;
- motion_y= s->mv[0][0][1] = 0;
- break;
- case CANDIDATE_MB_TYPE_INTER:
- s->mv_dir = MV_DIR_FORWARD;
- s->mb_intra= 0;
- motion_x= s->mv[0][0][0] = s->p_mv_table[xy][0];
- motion_y= s->mv[0][0][1] = s->p_mv_table[xy][1];
- break;
- case CANDIDATE_MB_TYPE_INTER_I:
- s->mv_dir = MV_DIR_FORWARD;
- s->mv_type = MV_TYPE_FIELD;
- s->mb_intra= 0;
- for(i=0; i<2; i++){
- j= s->field_select[0][i] = s->p_field_select_table[i][xy];
- s->mv[0][i][0] = s->p_field_mv_table[i][j][xy][0];
- s->mv[0][i][1] = s->p_field_mv_table[i][j][xy][1];
- }
- break;
- case CANDIDATE_MB_TYPE_INTER4V:
- s->mv_dir = MV_DIR_FORWARD;
- s->mv_type = MV_TYPE_8X8;
- s->mb_intra= 0;
- for(i=0; i<4; i++){
- s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
- s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
- }
- break;
- case CANDIDATE_MB_TYPE_DIRECT:
- s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
- s->mb_intra= 0;
- motion_x=s->b_direct_mv_table[xy][0];
- motion_y=s->b_direct_mv_table[xy][1];
- ff_mpeg4_set_direct_mv(s, motion_x, motion_y);
- break;
- case CANDIDATE_MB_TYPE_DIRECT0:
- s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
- s->mb_intra= 0;
- ff_mpeg4_set_direct_mv(s, 0, 0);
- break;
- case CANDIDATE_MB_TYPE_BIDIR:
- s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
- s->mb_intra= 0;
- s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
- s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
- s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
- s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
- break;
- case CANDIDATE_MB_TYPE_BACKWARD:
- s->mv_dir = MV_DIR_BACKWARD;
- s->mb_intra= 0;
- motion_x= s->mv[1][0][0] = s->b_back_mv_table[xy][0];
- motion_y= s->mv[1][0][1] = s->b_back_mv_table[xy][1];
- break;
- case CANDIDATE_MB_TYPE_FORWARD:
- s->mv_dir = MV_DIR_FORWARD;
- s->mb_intra= 0;
- motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
- motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
-// printf(" %d %d ", motion_x, motion_y);
- break;
- case CANDIDATE_MB_TYPE_FORWARD_I:
- s->mv_dir = MV_DIR_FORWARD;
- s->mv_type = MV_TYPE_FIELD;
- s->mb_intra= 0;
- for(i=0; i<2; i++){
- j= s->field_select[0][i] = s->b_field_select_table[0][i][xy];
- s->mv[0][i][0] = s->b_field_mv_table[0][i][j][xy][0];
- s->mv[0][i][1] = s->b_field_mv_table[0][i][j][xy][1];
- }
- break;
- case CANDIDATE_MB_TYPE_BACKWARD_I:
- s->mv_dir = MV_DIR_BACKWARD;
- s->mv_type = MV_TYPE_FIELD;
- s->mb_intra= 0;
- for(i=0; i<2; i++){
- j= s->field_select[1][i] = s->b_field_select_table[1][i][xy];
- s->mv[1][i][0] = s->b_field_mv_table[1][i][j][xy][0];
- s->mv[1][i][1] = s->b_field_mv_table[1][i][j][xy][1];
- }
- break;
- case CANDIDATE_MB_TYPE_BIDIR_I:
- s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
- s->mv_type = MV_TYPE_FIELD;
- s->mb_intra= 0;
- for(dir=0; dir<2; dir++){
- for(i=0; i<2; i++){
- j= s->field_select[dir][i] = s->b_field_select_table[dir][i][xy];
- s->mv[dir][i][0] = s->b_field_mv_table[dir][i][j][xy][0];
- s->mv[dir][i][1] = s->b_field_mv_table[dir][i][j][xy][1];
- }
- }
- break;
- default:
- av_log(s->avctx, AV_LOG_ERROR, "illegal MB type\n");
- }
-
- encode_mb(s, motion_x, motion_y);
-
- // RAL: Update last macroblock type
- s->last_mv_dir = s->mv_dir;
-
- if (s->out_format == FMT_H263 && s->pict_type!=B_TYPE)
- ff_h263_update_motion_val(s);
-
- MPV_decode_mb(s, s->block);
- }
-
- /* clean the MV table in IPS frames for direct mode in B frames */
- if(s->mb_intra /* && I,P,S_TYPE */){
- s->p_mv_table[xy][0]=0;
- s->p_mv_table[xy][1]=0;
- }
-
- if(s->flags&CODEC_FLAG_PSNR){
- int w= 16;
- int h= 16;
-
- if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
- if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
-
- s->current_picture.error[0] += sse(
- s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16,
- s->dest[0], w, h, s->linesize);
- s->current_picture.error[1] += sse(
- s, s->new_picture.data[1] + s->mb_x*8 + s->mb_y*s->uvlinesize*8,
- s->dest[1], w>>1, h>>1, s->uvlinesize);
- s->current_picture.error[2] += sse(
- s, s->new_picture .data[2] + s->mb_x*8 + s->mb_y*s->uvlinesize*8,
- s->dest[2], w>>1, h>>1, s->uvlinesize);
- }
- if(s->loop_filter){
- if(s->out_format == FMT_H263)
- ff_h263_loop_filter(s);
- }
-//printf("MB %d %d bits\n", s->mb_x+s->mb_y*s->mb_stride, put_bits_count(&s->pb));
- }
- }
-
- //not beautiful here but we must write it before flushing so it has to be here
- if (ENABLE_MSMPEG4_ENCODER && s->msmpeg4_version && s->msmpeg4_version<4 && s->pict_type == I_TYPE)
- msmpeg4_encode_ext_header(s);
-
- write_slice_end(s);
-
- /* Send the last GOB if RTP */
- if (s->avctx->rtp_callback) {
- int number_mb = (mb_y - s->resync_mb_y)*s->mb_width - s->resync_mb_x;
- pdif = pbBufPtr(&s->pb) - s->ptr_lastgob;
- /* Call the RTP callback to send the last GOB */
- emms_c();
- s->avctx->rtp_callback(s->avctx, s->ptr_lastgob, pdif, number_mb);
- }
-
- return 0;
-}
-
-#define MERGE(field) dst->field += src->field; src->field=0
-static void merge_context_after_me(MpegEncContext *dst, MpegEncContext *src){
- MERGE(me.scene_change_score);
- MERGE(me.mc_mb_var_sum_temp);
- MERGE(me.mb_var_sum_temp);
-}
-
-static void merge_context_after_encode(MpegEncContext *dst, MpegEncContext *src){
- int i;
-
- MERGE(dct_count[0]); //note, the other dct vars are not part of the context
- MERGE(dct_count[1]);
- MERGE(mv_bits);
- MERGE(i_tex_bits);
- MERGE(p_tex_bits);
- MERGE(i_count);
- MERGE(f_count);
- MERGE(b_count);
- MERGE(skip_count);
- MERGE(misc_bits);
- MERGE(error_count);
- MERGE(padding_bug_score);
- MERGE(current_picture.error[0]);
- MERGE(current_picture.error[1]);
- MERGE(current_picture.error[2]);
-
- if(dst->avctx->noise_reduction){
- for(i=0; i<64; i++){
- MERGE(dct_error_sum[0][i]);
- MERGE(dct_error_sum[1][i]);
- }
- }
-
- assert(put_bits_count(&src->pb) % 8 ==0);
- assert(put_bits_count(&dst->pb) % 8 ==0);
- ff_copy_bits(&dst->pb, src->pb.buf, put_bits_count(&src->pb));
- flush_put_bits(&dst->pb);
-}
-
-static int estimate_qp(MpegEncContext *s, int dry_run){
- if (s->next_lambda){
- s->current_picture_ptr->quality=
- s->current_picture.quality = s->next_lambda;
- if(!dry_run) s->next_lambda= 0;
- } else if (!s->fixed_qscale) {
- s->current_picture_ptr->quality=
- s->current_picture.quality = ff_rate_estimate_qscale(s, dry_run);
- if (s->current_picture.quality < 0)
- return -1;
- }
-
- if(s->adaptive_quant){
- switch(s->codec_id){
- case CODEC_ID_MPEG4:
- ff_clean_mpeg4_qscales(s);
- break;
- case CODEC_ID_H263:
- case CODEC_ID_H263P:
- case CODEC_ID_FLV1:
- ff_clean_h263_qscales(s);
- break;
- }
-
- s->lambda= s->lambda_table[0];
- //FIXME broken
- }else
- s->lambda= s->current_picture.quality;
-//printf("%d %d\n", s->avctx->global_quality, s->current_picture.quality);
- update_qscale(s);
- return 0;
-}
-
-static int encode_picture(MpegEncContext *s, int picture_number)
-{
- int i;
- int bits;
-
- s->picture_number = picture_number;
-
- /* Reset the average MB variance */
- s->me.mb_var_sum_temp =
- s->me.mc_mb_var_sum_temp = 0;
-
- /* we need to initialize some time vars before we can encode b-frames */
- // RAL: Condition added for MPEG1VIDEO
- if (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO || (s->h263_pred && !s->h263_msmpeg4))
- ff_set_mpeg4_time(s, s->picture_number); //FIXME rename and use has_b_frames or similar
-
- s->me.scene_change_score=0;
-
-// s->lambda= s->current_picture_ptr->quality; //FIXME qscale / ... stuff for ME ratedistoration
-
- if(s->pict_type==I_TYPE){
- if(s->msmpeg4_version >= 3) s->no_rounding=1;
- else s->no_rounding=0;
- }else if(s->pict_type!=B_TYPE){
- if(s->flipflop_rounding || s->codec_id == CODEC_ID_H263P || s->codec_id == CODEC_ID_MPEG4)
- s->no_rounding ^= 1;
- }
-
- if(s->flags & CODEC_FLAG_PASS2){
- if (estimate_qp(s,1) < 0)
- return -1;
- ff_get_2pass_fcode(s);
- }else if(!(s->flags & CODEC_FLAG_QSCALE)){
- if(s->pict_type==B_TYPE)
- s->lambda= s->last_lambda_for[s->pict_type];
- else
- s->lambda= s->last_lambda_for[s->last_non_b_pict_type];
- update_qscale(s);
- }
-
- s->mb_intra=0; //for the rate distortion & bit compare functions
- for(i=1; i<s->avctx->thread_count; i++){
- ff_update_duplicate_context(s->thread_context[i], s);
- }
-
- ff_init_me(s);
-
- /* Estimate motion for every MB */
- if(s->pict_type != I_TYPE){
- s->lambda = (s->lambda * s->avctx->me_penalty_compensation + 128)>>8;
- s->lambda2= (s->lambda2* (int64_t)s->avctx->me_penalty_compensation + 128)>>8;
- if(s->pict_type != B_TYPE && s->avctx->me_threshold==0){
- if((s->avctx->pre_me && s->last_non_b_pict_type==I_TYPE) || s->avctx->pre_me==2){
- s->avctx->execute(s->avctx, pre_estimate_motion_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count);
- }
- }
-
- s->avctx->execute(s->avctx, estimate_motion_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count);
- }else /* if(s->pict_type == I_TYPE) */{
- /* I-Frame */
- for(i=0; i<s->mb_stride*s->mb_height; i++)
- s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
-
- if(!s->fixed_qscale){
- /* finding spatial complexity for I-frame rate control */
- s->avctx->execute(s->avctx, mb_var_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count);
- }
- }
- for(i=1; i<s->avctx->thread_count; i++){
- merge_context_after_me(s, s->thread_context[i]);
- }
- s->current_picture.mc_mb_var_sum= s->current_picture_ptr->mc_mb_var_sum= s->me.mc_mb_var_sum_temp;
- s->current_picture. mb_var_sum= s->current_picture_ptr-> mb_var_sum= s->me. mb_var_sum_temp;
- emms_c();
-
- if(s->me.scene_change_score > s->avctx->scenechange_threshold && s->pict_type == P_TYPE){
- s->pict_type= I_TYPE;
- for(i=0; i<s->mb_stride*s->mb_height; i++)
- s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
-//printf("Scene change detected, encoding as I Frame %d %d\n", s->current_picture.mb_var_sum, s->current_picture.mc_mb_var_sum);
- }
-
- if(!s->umvplus){
- if(s->pict_type==P_TYPE || s->pict_type==S_TYPE) {
- s->f_code= ff_get_best_fcode(s, s->p_mv_table, CANDIDATE_MB_TYPE_INTER);
-
- if(s->flags & CODEC_FLAG_INTERLACED_ME){
- int a,b;
- a= ff_get_best_fcode(s, s->p_field_mv_table[0][0], CANDIDATE_MB_TYPE_INTER_I); //FIXME field_select
- b= ff_get_best_fcode(s, s->p_field_mv_table[1][1], CANDIDATE_MB_TYPE_INTER_I);
- s->f_code= FFMAX(s->f_code, FFMAX(a,b));
- }
-
- ff_fix_long_p_mvs(s);
- ff_fix_long_mvs(s, NULL, 0, s->p_mv_table, s->f_code, CANDIDATE_MB_TYPE_INTER, 0);
- if(s->flags & CODEC_FLAG_INTERLACED_ME){
- int j;
- for(i=0; i<2; i++){
- for(j=0; j<2; j++)
- ff_fix_long_mvs(s, s->p_field_select_table[i], j,
- s->p_field_mv_table[i][j], s->f_code, CANDIDATE_MB_TYPE_INTER_I, 0);
- }
- }
- }
-
- if(s->pict_type==B_TYPE){
- int a, b;
-
- a = ff_get_best_fcode(s, s->b_forw_mv_table, CANDIDATE_MB_TYPE_FORWARD);
- b = ff_get_best_fcode(s, s->b_bidir_forw_mv_table, CANDIDATE_MB_TYPE_BIDIR);
- s->f_code = FFMAX(a, b);
-
- a = ff_get_best_fcode(s, s->b_back_mv_table, CANDIDATE_MB_TYPE_BACKWARD);
- b = ff_get_best_fcode(s, s->b_bidir_back_mv_table, CANDIDATE_MB_TYPE_BIDIR);
- s->b_code = FFMAX(a, b);
-
- ff_fix_long_mvs(s, NULL, 0, s->b_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_FORWARD, 1);
- ff_fix_long_mvs(s, NULL, 0, s->b_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BACKWARD, 1);
- ff_fix_long_mvs(s, NULL, 0, s->b_bidir_forw_mv_table, s->f_code, CANDIDATE_MB_TYPE_BIDIR, 1);
- ff_fix_long_mvs(s, NULL, 0, s->b_bidir_back_mv_table, s->b_code, CANDIDATE_MB_TYPE_BIDIR, 1);
- if(s->flags & CODEC_FLAG_INTERLACED_ME){
- int dir, j;
- for(dir=0; dir<2; dir++){
- for(i=0; i<2; i++){
- for(j=0; j<2; j++){
- int type= dir ? (CANDIDATE_MB_TYPE_BACKWARD_I|CANDIDATE_MB_TYPE_BIDIR_I)
- : (CANDIDATE_MB_TYPE_FORWARD_I |CANDIDATE_MB_TYPE_BIDIR_I);
- ff_fix_long_mvs(s, s->b_field_select_table[dir][i], j,
- s->b_field_mv_table[dir][i][j], dir ? s->b_code : s->f_code, type, 1);
- }
- }
- }
- }
- }
- }
-
- if (estimate_qp(s, 0) < 0)
- return -1;
-
- if(s->qscale < 3 && s->max_qcoeff<=128 && s->pict_type==I_TYPE && !(s->flags & CODEC_FLAG_QSCALE))
- s->qscale= 3; //reduce clipping problems
-
- if (s->out_format == FMT_MJPEG) {
- /* for mjpeg, we do include qscale in the matrix */
- s->intra_matrix[0] = ff_mpeg1_default_intra_matrix[0];
- for(i=1;i<64;i++){
- int j= s->dsp.idct_permutation[i];
-
- s->intra_matrix[j] = av_clip_uint8((ff_mpeg1_default_intra_matrix[i] * s->qscale) >> 3);
- }
- convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
- s->intra_matrix, s->intra_quant_bias, 8, 8, 1);
- s->qscale= 8;
- }
-
- //FIXME var duplication
- s->current_picture_ptr->key_frame=
- s->current_picture.key_frame= s->pict_type == I_TYPE; //FIXME pic_ptr
- s->current_picture_ptr->pict_type=
- s->current_picture.pict_type= s->pict_type;
-
- if(s->current_picture.key_frame)
- s->picture_in_gop_number=0;
-
- s->last_bits= put_bits_count(&s->pb);
- switch(s->out_format) {
- case FMT_MJPEG:
- if (ENABLE_MJPEG_ENCODER)
- mjpeg_picture_header(s);
- break;
- case FMT_H261:
- if (ENABLE_H261_ENCODER)
- ff_h261_encode_picture_header(s, picture_number);
- break;
- case FMT_H263:
- if (ENABLE_WMV2_ENCODER && s->codec_id == CODEC_ID_WMV2)
- ff_wmv2_encode_picture_header(s, picture_number);
- else if (ENABLE_MSMPEG4_ENCODER && s->h263_msmpeg4)
- msmpeg4_encode_picture_header(s, picture_number);
- else if (s->h263_pred)
- mpeg4_encode_picture_header(s, picture_number);
- else if (ENABLE_RV10_ENCODER && s->codec_id == CODEC_ID_RV10)
- rv10_encode_picture_header(s, picture_number);
- else if (ENABLE_RV20_ENCODER && s->codec_id == CODEC_ID_RV20)
- rv20_encode_picture_header(s, picture_number);
- else if (s->codec_id == CODEC_ID_FLV1)
- ff_flv_encode_picture_header(s, picture_number);
- else
- h263_encode_picture_header(s, picture_number);
- break;
- case FMT_MPEG1:
- mpeg1_encode_picture_header(s, picture_number);
- break;
- case FMT_H264:
- break;
- default:
- assert(0);
- }
- bits= put_bits_count(&s->pb);
- s->header_bits= bits - s->last_bits;
-
- for(i=1; i<s->avctx->thread_count; i++){
- update_duplicate_context_after_me(s->thread_context[i], s);
- }
- s->avctx->execute(s->avctx, encode_thread, (void**)&(s->thread_context[0]), NULL, s->avctx->thread_count);
- for(i=1; i<s->avctx->thread_count; i++){
- merge_context_after_encode(s, s->thread_context[i]);
- }
- emms_c();
- return 0;
-}
-
-static void denoise_dct_c(MpegEncContext *s, DCTELEM *block){
- const int intra= s->mb_intra;
- int i;
-
- s->dct_count[intra]++;
-
- for(i=0; i<64; i++){
- int level= block[i];
-
- if(level){
- if(level>0){
- s->dct_error_sum[intra][i] += level;
- level -= s->dct_offset[intra][i];
- if(level<0) level=0;
- }else{
- s->dct_error_sum[intra][i] -= level;
- level += s->dct_offset[intra][i];
- if(level>0) level=0;
- }
- block[i]= level;
- }
- }
-}
-
-static int dct_quantize_trellis_c(MpegEncContext *s,
- DCTELEM *block, int n,
- int qscale, int *overflow){
- const int *qmat;
- const uint8_t *scantable= s->intra_scantable.scantable;
- const uint8_t *perm_scantable= s->intra_scantable.permutated;
- int max=0;
- unsigned int threshold1, threshold2;
- int bias=0;
- int run_tab[65];
- int level_tab[65];
- int score_tab[65];
- int survivor[65];
- int survivor_count;
- int last_run=0;
- int last_level=0;
- int last_score= 0;
- int last_i;
- int coeff[2][64];
- int coeff_count[64];
- int qmul, qadd, start_i, last_non_zero, i, dc;
- const int esc_length= s->ac_esc_length;
- uint8_t * length;
- uint8_t * last_length;
- const int lambda= s->lambda2 >> (FF_LAMBDA_SHIFT - 6);
-
- s->dsp.fdct (block);
-
- if(s->dct_error_sum)
- s->denoise_dct(s, block);
- qmul= qscale*16;
- qadd= ((qscale-1)|1)*8;
-
- if (s->mb_intra) {
- int q;
- if (!s->h263_aic) {
- if (n < 4)
- q = s->y_dc_scale;
- else
- q = s->c_dc_scale;
- q = q << 3;
- } else{
- /* For AIC we skip quant/dequant of INTRADC */
- q = 1 << 3;
- qadd=0;
- }
-
- /* note: block[0] is assumed to be positive */
- block[0] = (block[0] + (q >> 1)) / q;
- start_i = 1;
- last_non_zero = 0;
- qmat = s->q_intra_matrix[qscale];
- if(s->mpeg_quant || s->out_format == FMT_MPEG1)
- bias= 1<<(QMAT_SHIFT-1);
- length = s->intra_ac_vlc_length;
- last_length= s->intra_ac_vlc_last_length;
- } else {
- start_i = 0;
- last_non_zero = -1;
- qmat = s->q_inter_matrix[qscale];
- length = s->inter_ac_vlc_length;
- last_length= s->inter_ac_vlc_last_length;
- }
- last_i= start_i;
-
- threshold1= (1<<QMAT_SHIFT) - bias - 1;
- threshold2= (threshold1<<1);
-
- for(i=63; i>=start_i; i--) {
- const int j = scantable[i];
- int level = block[j] * qmat[j];
-
- if(((unsigned)(level+threshold1))>threshold2){
- last_non_zero = i;
- break;
- }
- }
-
- for(i=start_i; i<=last_non_zero; i++) {
- const int j = scantable[i];
- int level = block[j] * qmat[j];
-
-// if( bias+level >= (1<<(QMAT_SHIFT - 3))
-// || bias-level >= (1<<(QMAT_SHIFT - 3))){
- if(((unsigned)(level+threshold1))>threshold2){
- if(level>0){
- level= (bias + level)>>QMAT_SHIFT;
- coeff[0][i]= level;
- coeff[1][i]= level-1;
-// coeff[2][k]= level-2;
- }else{
- level= (bias - level)>>QMAT_SHIFT;
- coeff[0][i]= -level;
- coeff[1][i]= -level+1;
-// coeff[2][k]= -level+2;
- }
- coeff_count[i]= FFMIN(level, 2);
- assert(coeff_count[i]);
- max |=level;
- }else{
- coeff[0][i]= (level>>31)|1;
- coeff_count[i]= 1;
- }
- }
-
- *overflow= s->max_qcoeff < max; //overflow might have happened
-
- if(last_non_zero < start_i){
- memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
- return last_non_zero;
- }
-
- score_tab[start_i]= 0;
- survivor[0]= start_i;
- survivor_count= 1;
-
- for(i=start_i; i<=last_non_zero; i++){
- int level_index, j;
- const int dct_coeff= FFABS(block[ scantable[i] ]);
- const int zero_distoration= dct_coeff*dct_coeff;
- int best_score=256*256*256*120;
- for(level_index=0; level_index < coeff_count[i]; level_index++){
- int distoration;
- int level= coeff[level_index][i];
- const int alevel= FFABS(level);
- int unquant_coeff;
-
- assert(level);
-
- if(s->out_format == FMT_H263){
- unquant_coeff= alevel*qmul + qadd;
- }else{ //MPEG1
- j= s->dsp.idct_permutation[ scantable[i] ]; //FIXME optimize
- if(s->mb_intra){
- unquant_coeff = (int)( alevel * qscale * s->intra_matrix[j]) >> 3;
- unquant_coeff = (unquant_coeff - 1) | 1;
- }else{
- unquant_coeff = ((( alevel << 1) + 1) * qscale * ((int) s->inter_matrix[j])) >> 4;
- unquant_coeff = (unquant_coeff - 1) | 1;
- }
- unquant_coeff<<= 3;
- }
-
- distoration= (unquant_coeff - dct_coeff) * (unquant_coeff - dct_coeff) - zero_distoration;
- level+=64;
- if((level&(~127)) == 0){
- for(j=survivor_count-1; j>=0; j--){
- int run= i - survivor[j];
- int score= distoration + length[UNI_AC_ENC_INDEX(run, level)]*lambda;
- score += score_tab[i-run];
-
- if(score < best_score){
- best_score= score;
- run_tab[i+1]= run;
- level_tab[i+1]= level-64;
- }
- }
-
- if(s->out_format == FMT_H263){
- for(j=survivor_count-1; j>=0; j--){
- int run= i - survivor[j];
- int score= distoration + last_length[UNI_AC_ENC_INDEX(run, level)]*lambda;
- score += score_tab[i-run];
- if(score < last_score){
- last_score= score;
- last_run= run;
- last_level= level-64;
- last_i= i+1;
- }
- }
- }
- }else{
- distoration += esc_length*lambda;
- for(j=survivor_count-1; j>=0; j--){
- int run= i - survivor[j];
- int score= distoration + score_tab[i-run];
-
- if(score < best_score){
- best_score= score;
- run_tab[i+1]= run;
- level_tab[i+1]= level-64;
- }
- }
-
- if(s->out_format == FMT_H263){
- for(j=survivor_count-1; j>=0; j--){
- int run= i - survivor[j];
- int score= distoration + score_tab[i-run];
- if(score < last_score){
- last_score= score;
- last_run= run;
- last_level= level-64;
- last_i= i+1;
- }
- }
- }
- }
- }
-
- score_tab[i+1]= best_score;
-
- //Note: there is a vlc code in mpeg4 which is 1 bit shorter then another one with a shorter run and the same level
- if(last_non_zero <= 27){
- for(; survivor_count; survivor_count--){
- if(score_tab[ survivor[survivor_count-1] ] <= best_score)
- break;
- }
- }else{
- for(; survivor_count; survivor_count--){
- if(score_tab[ survivor[survivor_count-1] ] <= best_score + lambda)
- break;
- }
- }
-
- survivor[ survivor_count++ ]= i+1;
- }
-
- if(s->out_format != FMT_H263){
- last_score= 256*256*256*120;
- for(i= survivor[0]; i<=last_non_zero + 1; i++){
- int score= score_tab[i];
- if(i) score += lambda*2; //FIXME exacter?
-
- if(score < last_score){
- last_score= score;
- last_i= i;
- last_level= level_tab[i];
- last_run= run_tab[i];
- }
- }
- }
-
- s->coded_score[n] = last_score;
-
- dc= FFABS(block[0]);
- last_non_zero= last_i - 1;
- memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
-
- if(last_non_zero < start_i)
- return last_non_zero;
-
- if(last_non_zero == 0 && start_i == 0){
- int best_level= 0;
- int best_score= dc * dc;
-
- for(i=0; i<coeff_count[0]; i++){
- int level= coeff[i][0];
- int alevel= FFABS(level);
- int unquant_coeff, score, distortion;
-
- if(s->out_format == FMT_H263){
- unquant_coeff= (alevel*qmul + qadd)>>3;
- }else{ //MPEG1
- unquant_coeff = ((( alevel << 1) + 1) * qscale * ((int) s->inter_matrix[0])) >> 4;
- unquant_coeff = (unquant_coeff - 1) | 1;
- }
- unquant_coeff = (unquant_coeff + 4) >> 3;
- unquant_coeff<<= 3 + 3;
-
- distortion= (unquant_coeff - dc) * (unquant_coeff - dc);
- level+=64;
- if((level&(~127)) == 0) score= distortion + last_length[UNI_AC_ENC_INDEX(0, level)]*lambda;
- else score= distortion + esc_length*lambda;
-
- if(score < best_score){
- best_score= score;
- best_level= level - 64;
- }
- }
- block[0]= best_level;
- s->coded_score[n] = best_score - dc*dc;
- if(best_level == 0) return -1;
- else return last_non_zero;
- }
-
- i= last_i;
- assert(last_level);
-
- block[ perm_scantable[last_non_zero] ]= last_level;
- i -= last_run + 1;
-
- for(; i>start_i; i -= run_tab[i] + 1){
- block[ perm_scantable[i-1] ]= level_tab[i];
- }
-
- return last_non_zero;
-}
-
-//#define REFINE_STATS 1
-static int16_t basis[64][64];
-
-static void build_basis(uint8_t *perm){
- int i, j, x, y;
- emms_c();
- for(i=0; i<8; i++){
- for(j=0; j<8; j++){
- for(y=0; y<8; y++){
- for(x=0; x<8; x++){
- double s= 0.25*(1<<BASIS_SHIFT);
- int index= 8*i + j;
- int perm_index= perm[index];
- if(i==0) s*= sqrt(0.5);
- if(j==0) s*= sqrt(0.5);
- basis[perm_index][8*x + y]= lrintf(s * cos((M_PI/8.0)*i*(x+0.5)) * cos((M_PI/8.0)*j*(y+0.5)));
- }
- }
- }
- }
-}
-
-static int dct_quantize_refine(MpegEncContext *s, //FIXME breaks denoise?
- DCTELEM *block, int16_t *weight, DCTELEM *orig,
- int n, int qscale){
- int16_t rem[64];
- DECLARE_ALIGNED_16(DCTELEM, d1[64]);
- const int *qmat;
- const uint8_t *scantable= s->intra_scantable.scantable;
- const uint8_t *perm_scantable= s->intra_scantable.permutated;
-// unsigned int threshold1, threshold2;
-// int bias=0;
- int run_tab[65];
- int prev_run=0;
- int prev_level=0;
- int qmul, qadd, start_i, last_non_zero, i, dc;
- uint8_t * length;
- uint8_t * last_length;
- int lambda;
- int rle_index, run, q = 1, sum; //q is only used when s->mb_intra is true
-#ifdef REFINE_STATS
-static int count=0;
-static int after_last=0;
-static int to_zero=0;
-static int from_zero=0;
-static int raise=0;
-static int lower=0;
-static int messed_sign=0;
-#endif
-
- if(basis[0][0] == 0)
- build_basis(s->dsp.idct_permutation);
-
- qmul= qscale*2;
- qadd= (qscale-1)|1;
- if (s->mb_intra) {
- if (!s->h263_aic) {
- if (n < 4)
- q = s->y_dc_scale;
- else
- q = s->c_dc_scale;
- } else{
- /* For AIC we skip quant/dequant of INTRADC */
- q = 1;
- qadd=0;
- }
- q <<= RECON_SHIFT-3;
- /* note: block[0] is assumed to be positive */
- dc= block[0]*q;
-// block[0] = (block[0] + (q >> 1)) / q;
- start_i = 1;
- qmat = s->q_intra_matrix[qscale];
-// if(s->mpeg_quant || s->out_format == FMT_MPEG1)
-// bias= 1<<(QMAT_SHIFT-1);
- length = s->intra_ac_vlc_length;
- last_length= s->intra_ac_vlc_last_length;
- } else {
- dc= 0;
- start_i = 0;
- qmat = s->q_inter_matrix[qscale];
- length = s->inter_ac_vlc_length;
- last_length= s->inter_ac_vlc_last_length;
- }
- last_non_zero = s->block_last_index[n];
-
-#ifdef REFINE_STATS
-{START_TIMER
-#endif
- dc += (1<<(RECON_SHIFT-1));
- for(i=0; i<64; i++){
- rem[i]= dc - (orig[i]<<RECON_SHIFT); //FIXME use orig dirrectly instead of copying to rem[]
- }
-#ifdef REFINE_STATS
-STOP_TIMER("memset rem[]")}
-#endif
- sum=0;
- for(i=0; i<64; i++){
- int one= 36;
- int qns=4;
- int w;
-
- w= FFABS(weight[i]) + qns*one;
- w= 15 + (48*qns*one + w/2)/w; // 16 .. 63
-
- weight[i] = w;
-// w=weight[i] = (63*qns + (w/2)) / w;
-
- assert(w>0);
- assert(w<(1<<6));
- sum += w*w;
- }
- lambda= sum*(uint64_t)s->lambda2 >> (FF_LAMBDA_SHIFT - 6 + 6 + 6 + 6);
-#ifdef REFINE_STATS
-{START_TIMER
-#endif
- run=0;
- rle_index=0;
- for(i=start_i; i<=last_non_zero; i++){
- int j= perm_scantable[i];
- const int level= block[j];
- int coeff;
-
- if(level){
- if(level<0) coeff= qmul*level - qadd;
- else coeff= qmul*level + qadd;
- run_tab[rle_index++]=run;
- run=0;
-
- s->dsp.add_8x8basis(rem, basis[j], coeff);
- }else{
- run++;
- }
- }
-#ifdef REFINE_STATS
-if(last_non_zero>0){
-STOP_TIMER("init rem[]")
-}
-}
-
-{START_TIMER
-#endif
- for(;;){
- int best_score=s->dsp.try_8x8basis(rem, weight, basis[0], 0);
- int best_coeff=0;
- int best_change=0;
- int run2, best_unquant_change=0, analyze_gradient;
-#ifdef REFINE_STATS
-{START_TIMER
-#endif
- analyze_gradient = last_non_zero > 2 || s->avctx->quantizer_noise_shaping >= 3;
-
- if(analyze_gradient){
-#ifdef REFINE_STATS
-{START_TIMER
-#endif
- for(i=0; i<64; i++){
- int w= weight[i];
-
- d1[i] = (rem[i]*w*w + (1<<(RECON_SHIFT+12-1)))>>(RECON_SHIFT+12);
- }
-#ifdef REFINE_STATS
-STOP_TIMER("rem*w*w")}
-{START_TIMER
-#endif
- s->dsp.fdct(d1);
-#ifdef REFINE_STATS
-STOP_TIMER("dct")}
-#endif
- }
-
- if(start_i){
- const int level= block[0];
- int change, old_coeff;
-
- assert(s->mb_intra);
-
- old_coeff= q*level;
-
- for(change=-1; change<=1; change+=2){
- int new_level= level + change;
- int score, new_coeff;
-
- new_coeff= q*new_level;
- if(new_coeff >= 2048 || new_coeff < 0)
- continue;
-
- score= s->dsp.try_8x8basis(rem, weight, basis[0], new_coeff - old_coeff);
- if(score<best_score){
- best_score= score;
- best_coeff= 0;
- best_change= change;
- best_unquant_change= new_coeff - old_coeff;
- }
- }
- }
-
- run=0;
- rle_index=0;
- run2= run_tab[rle_index++];
- prev_level=0;
- prev_run=0;
-
- for(i=start_i; i<64; i++){
- int j= perm_scantable[i];
- const int level= block[j];
- int change, old_coeff;
-
- if(s->avctx->quantizer_noise_shaping < 3 && i > last_non_zero + 1)
- break;
-
- if(level){
- if(level<0) old_coeff= qmul*level - qadd;
- else old_coeff= qmul*level + qadd;
- run2= run_tab[rle_index++]; //FIXME ! maybe after last
- }else{
- old_coeff=0;
- run2--;
- assert(run2>=0 || i >= last_non_zero );
- }
-
- for(change=-1; change<=1; change+=2){
- int new_level= level + change;
- int score, new_coeff, unquant_change;
-
- score=0;
- if(s->avctx->quantizer_noise_shaping < 2 && FFABS(new_level) > FFABS(level))
- continue;
-
- if(new_level){
- if(new_level<0) new_coeff= qmul*new_level - qadd;
- else new_coeff= qmul*new_level + qadd;
- if(new_coeff >= 2048 || new_coeff <= -2048)
- continue;
- //FIXME check for overflow
-
- if(level){
- if(level < 63 && level > -63){
- if(i < last_non_zero)
- score += length[UNI_AC_ENC_INDEX(run, new_level+64)]
- - length[UNI_AC_ENC_INDEX(run, level+64)];
- else
- score += last_length[UNI_AC_ENC_INDEX(run, new_level+64)]
- - last_length[UNI_AC_ENC_INDEX(run, level+64)];
- }
- }else{
- assert(FFABS(new_level)==1);
-
- if(analyze_gradient){
- int g= d1[ scantable[i] ];
- if(g && (g^new_level) >= 0)
- continue;
- }
-
- if(i < last_non_zero){
- int next_i= i + run2 + 1;
- int next_level= block[ perm_scantable[next_i] ] + 64;
-
- if(next_level&(~127))
- next_level= 0;
-
- if(next_i < last_non_zero)
- score += length[UNI_AC_ENC_INDEX(run, 65)]
- + length[UNI_AC_ENC_INDEX(run2, next_level)]
- - length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
- else
- score += length[UNI_AC_ENC_INDEX(run, 65)]
- + last_length[UNI_AC_ENC_INDEX(run2, next_level)]
- - last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)];
- }else{
- score += last_length[UNI_AC_ENC_INDEX(run, 65)];
- if(prev_level){
- score += length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
- - last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
- }
- }
- }
- }else{
- new_coeff=0;
- assert(FFABS(level)==1);
-
- if(i < last_non_zero){
- int next_i= i + run2 + 1;
- int next_level= block[ perm_scantable[next_i] ] + 64;
-
- if(next_level&(~127))
- next_level= 0;
-
- if(next_i < last_non_zero)
- score += length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]
- - length[UNI_AC_ENC_INDEX(run2, next_level)]
- - length[UNI_AC_ENC_INDEX(run, 65)];
- else
- score += last_length[UNI_AC_ENC_INDEX(run + run2 + 1, next_level)]
- - last_length[UNI_AC_ENC_INDEX(run2, next_level)]
- - length[UNI_AC_ENC_INDEX(run, 65)];
- }else{
- score += -last_length[UNI_AC_ENC_INDEX(run, 65)];
- if(prev_level){
- score += last_length[UNI_AC_ENC_INDEX(prev_run, prev_level)]
- - length[UNI_AC_ENC_INDEX(prev_run, prev_level)];
- }
- }
- }
-
- score *= lambda;
-
- unquant_change= new_coeff - old_coeff;
- assert((score < 100*lambda && score > -100*lambda) || lambda==0);
-
- score+= s->dsp.try_8x8basis(rem, weight, basis[j], unquant_change);
- if(score<best_score){
- best_score= score;
- best_coeff= i;
- best_change= change;
- best_unquant_change= unquant_change;
- }
- }
- if(level){
- prev_level= level + 64;
- if(prev_level&(~127))
- prev_level= 0;
- prev_run= run;
- run=0;
- }else{
- run++;
- }
- }
-#ifdef REFINE_STATS
-STOP_TIMER("iterative step")}
-#endif
-
- if(best_change){
- int j= perm_scantable[ best_coeff ];
-
- block[j] += best_change;
-
- if(best_coeff > last_non_zero){
- last_non_zero= best_coeff;
- assert(block[j]);
-#ifdef REFINE_STATS
-after_last++;
-#endif
- }else{
-#ifdef REFINE_STATS
-if(block[j]){
- if(block[j] - best_change){
- if(FFABS(block[j]) > FFABS(block[j] - best_change)){
- raise++;
- }else{
- lower++;
- }
- }else{
- from_zero++;
- }
-}else{
- to_zero++;
-}
-#endif
- for(; last_non_zero>=start_i; last_non_zero--){
- if(block[perm_scantable[last_non_zero]])
- break;
- }
- }
-#ifdef REFINE_STATS
-count++;
-if(256*256*256*64 % count == 0){
- printf("after_last:%d to_zero:%d from_zero:%d raise:%d lower:%d sign:%d xyp:%d/%d/%d\n", after_last, to_zero, from_zero, raise, lower, messed_sign, s->mb_x, s->mb_y, s->picture_number);
-}
-#endif
- run=0;
- rle_index=0;
- for(i=start_i; i<=last_non_zero; i++){
- int j= perm_scantable[i];
- const int level= block[j];
-
- if(level){
- run_tab[rle_index++]=run;
- run=0;
- }else{
- run++;
- }
- }
-
- s->dsp.add_8x8basis(rem, basis[j], best_unquant_change);
- }else{
- break;
- }
- }
-#ifdef REFINE_STATS
-if(last_non_zero>0){
-STOP_TIMER("iterative search")
-}
-}
-#endif
-
- return last_non_zero;
-}
-
-static int dct_quantize_c(MpegEncContext *s,
- DCTELEM *block, int n,
- int qscale, int *overflow)
-{
- int i, j, level, last_non_zero, q, start_i;
- const int *qmat;
- const uint8_t *scantable= s->intra_scantable.scantable;
- int bias;
- int max=0;
- unsigned int threshold1, threshold2;
-
- s->dsp.fdct (block);
-
- if(s->dct_error_sum)
- s->denoise_dct(s, block);
-
- if (s->mb_intra) {
- if (!s->h263_aic) {
- if (n < 4)
- q = s->y_dc_scale;
- else
- q = s->c_dc_scale;
- q = q << 3;
- } else
- /* For AIC we skip quant/dequant of INTRADC */
- q = 1 << 3;
-
- /* note: block[0] is assumed to be positive */
- block[0] = (block[0] + (q >> 1)) / q;
- start_i = 1;
- last_non_zero = 0;
- qmat = s->q_intra_matrix[qscale];
- bias= s->intra_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT);
- } else {
- start_i = 0;
- last_non_zero = -1;
- qmat = s->q_inter_matrix[qscale];
- bias= s->inter_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT);
- }
- threshold1= (1<<QMAT_SHIFT) - bias - 1;
- threshold2= (threshold1<<1);
- for(i=63;i>=start_i;i--) {
- j = scantable[i];
- level = block[j] * qmat[j];
-
- if(((unsigned)(level+threshold1))>threshold2){
- last_non_zero = i;
- break;
- }else{
- block[j]=0;
- }
- }
- for(i=start_i; i<=last_non_zero; i++) {
- j = scantable[i];
- level = block[j] * qmat[j];
-
-// if( bias+level >= (1<<QMAT_SHIFT)
-// || bias-level >= (1<<QMAT_SHIFT)){
- if(((unsigned)(level+threshold1))>threshold2){
- if(level>0){
- level= (bias + level)>>QMAT_SHIFT;
- block[j]= level;
- }else{
- level= (bias - level)>>QMAT_SHIFT;
- block[j]= -level;
- }
- max |=level;
- }else{
- block[j]=0;
- }
- }
- *overflow= s->max_qcoeff < max; //overflow might have happened
-
- /* we need this permutation so that we correct the IDCT, we only permute the !=0 elements */
- if (s->dsp.idct_permutation_type != FF_NO_IDCT_PERM)
- ff_block_permute(block, s->dsp.idct_permutation, scantable, last_non_zero);
-
- return last_non_zero;
-}
-
-#endif //CONFIG_ENCODERS
-
-static void dct_unquantize_mpeg1_intra_c(MpegEncContext *s,
- DCTELEM *block, int n, int qscale)
-{
- int i, level, nCoeffs;
- const uint16_t *quant_matrix;
-
- nCoeffs= s->block_last_index[n];
-
- if (n < 4)
- block[0] = block[0] * s->y_dc_scale;
- else
- block[0] = block[0] * s->c_dc_scale;
- /* XXX: only mpeg1 */
- quant_matrix = s->intra_matrix;
- for(i=1;i<=nCoeffs;i++) {
- int j= s->intra_scantable.permutated[i];
- level = block[j];
- if (level) {
- if (level < 0) {
- level = -level;
- level = (int)(level * qscale * quant_matrix[j]) >> 3;
- level = (level - 1) | 1;
- level = -level;
- } else {
- level = (int)(level * qscale * quant_matrix[j]) >> 3;
- level = (level - 1) | 1;
- }
- block[j] = level;
- }