comparison h264.c @ 2645:42528c1f0246 libavcodec

More spelling errors patch by (Kevin Baragona (kevinmb500 gawab com>)

2644:6ff5dc0dbaf0

    
    PPS pps_buffer[MAX_PPS_COUNT];
    /**
     * current pps
     */
    PPS pps; //FIXME move tp Picture perhaps? (->no) do we need that?

    int slice_num;
    uint8_t *slice_table_base;
    uint8_t *slice_table;      ///< slice_table_base + mb_stride + 1
    int slice_type;

    
    w      *= size;
    stride *= size;
    
    assert((((int)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
//FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it
    if(w==2 && h==2){
        *(uint16_t*)(p + 0)=
        *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
    }else if(w==2 && h==4){
        *(uint16_t*)(p + 0*stride)=

    // the actual condition is whether we're on the edge of a slice,
    // and even then the intra and nnz parts are unnecessary.
    if(for_deblock && h->slice_num == 1)
        return;

    //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it 
    
    top_xy     = mb_xy  - s->mb_stride;
    topleft_xy = top_xy - 1;
    topright_xy= top_xy + 1;
    left_xy[1] = left_xy[0] = mb_xy-1;

            }
            

            h->ref_cache[list][scan8[5 ]+1] = 
            h->ref_cache[list][scan8[7 ]+1] = 
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewher else)
            h->ref_cache[list][scan8[4 ]] = 
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;

            if( h->pps.cabac ) {
                /* XXX beurk, Load mvd */

                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
                }
                *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
                *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
                *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
                *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
                *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;

                if(h->slice_type == B_TYPE){
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);

    int list;

    for(list=0; list<2; list++){
        int y;
        if(!USES_LIST(mb_type, list)){
            if(1){ //FIXME skip or never read if mb_type doesnt use it
                for(y=0; y<4; y++){
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
                }
                if( h->pps.cabac ) {

/**
 * Decodes a network abstraction layer unit.
 * @param consumed is the number of bytes used as input
 * @param length is the length of the array
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
 * @returns decoded bytes, might be src+1 if no escapes 
 */
static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
    int i, si, di;
    uint8_t *dst;

    }

    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
    dst= h->rbsp_buffer;

//printf("deoding esc\n");
    si=di=0;
    while(si<length){ 
        //remove escapes (very rare 1:2^22)
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
            if(src[si+2]==3){ //escape

        block[3*4 + i]=   z3 - 2*z2;
    }
}
#endif

//FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
//FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
    int i;
    const int * const quant_table= quant_coeff[qscale];
    const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;

    av_freep(&h->s.obmc_scratchpad);
}

/**
 * allocates tables.
 * needs widzh/height
 */
static int alloc_tables(H264Context *h){
    MpegEncContext * const s = &h->s;
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
    int x,y;

    
    src_y  -=   linesize;
    src_cb -= uvlinesize;
    src_cr -= uvlinesize;

    // There is two lines saved, the line above the the top macroblock of a pair,
    // and the line above the bottom macroblock
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
    for(i=1; i<17; i++){
        h->left_border[i]= src_y[15+i*  linesize];
    }

    
    src_y  -= 2 *   linesize;
    src_cb -= 2 * uvlinesize;
    src_cr -= 2 * uvlinesize;

    // There is two lines saved, the line above the the top macroblock of a pair,
    // and the line above the bottom macroblock
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
    h->left_border[1]= h->top_borders[1][s->mb_x][15];
    for(i=2; i<34; i++){
        h->left_border[i]= src_y[15+i*  linesize];

            backup_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize);
            // TODO deblock a pair
            // top 
            s->mb_y--;
            tprintf("call mbaff filter_mb mb_x:%d mb_y:%d pair_dest_y = %p, dest_y = %p\n", mb_x, mb_y, pair_dest_y, dest_y);
            fill_caches(h, mb_type_top, 1); //FIXME dont fill stuff which isnt used by filter_mb
            filter_mb(h, mb_x, mb_y, pair_dest_y, pair_dest_cb, pair_dest_cr, linesize, uvlinesize);
            if (tmp != s->current_picture.data[1][384]) {
                tprintf("modified pixel 8,1 (1)\n");
            }
            // bottom
            s->mb_y++;
            tprintf("call mbaff filter_mb\n");
            fill_caches(h, mb_type_bottom, 1); //FIXME dont fill stuff which isnt used by filter_mb
            filter_mb(h, mb_x, mb_y+1, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
            if (tmp != s->current_picture.data[1][384]) {
                tprintf("modified pixel 8,1 (2)\n");
            }
        } else {
            tprintf("call filter_mb\n");
            backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
            fill_caches(h, mb_type, 1); //FIXME dont fill stuff which isnt used by filter_mb
            filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
        }
    }
}

    MpegEncContext * const s = &h->s;
    int list;
    
    print_short_term(h);
    print_long_term(h);
    if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func
    
    for(list=0; list<2; list++){
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);

        if(get_bits1(&s->gb)){

    idr(h);
}

/**
 *
 * @return the removed picture or NULL if an error occures
 */
static Picture * remove_short(H264Context *h, int frame_num){
    MpegEncContext * const s = &h->s;
    int i;
    

    return NULL;
}

/**
 *
 * @return the removed picture or NULL if an error occures
 */
static Picture * remove_long(H264Context *h, int i){
    Picture *pic;

    pic= h->long_ref[i];

        || (h->slice_num != 0 && slice_type == h->slice_type) ) {
        default_ref_list_done = 1;
    }
    h->slice_type= slice_type;

    s->pict_type= h->slice_type; // to make a few old func happy, its wrong though
        
    pps_id= get_ue_golomb(&s->gb);
    if(pps_id>255){
        av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
        return -1;

                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
            else
                level_code= prefix + get_bits(gb, 4); //part
        }else if(prefix==15){
            level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
            if(suffix_length==0) level_code+=15; //FIXME doesnt make (much)sense
        }else{
            av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
            return -1;
        }

static int decode_mb_cavlc(H264Context *h){
    MpegEncContext * const s = &h->s;
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
    int mb_type, partition_count, cbp;

    s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?    

    tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
    cbp = 0; /* avoid warning. FIXME: find a solution without slowing
                down the code */
    if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){

                tprintf("CHROMA V ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));
                h->mb[index + (x&3) + 16*(x>>2)]= get_bits(&s->gb, 8);
            }
        }
    
        // In deblocking, the quantiser is 0
        s->current_picture.qscale_table[mb_xy]= 0;
        h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, 0);
        // All coeffs are presents
        memset(h->non_zero_count[mb_xy], 16, 16);
        
        return 0;
    }
        

        }
        
        h->chroma_qp= chroma_qp= get_chroma_qp(h->pps.chroma_qp_index_offset, s->qscale);
        if(IS_INTRA16x16(mb_type)){
            if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, s->qscale, 16) < 0){
                return -1; //FIXME continue if partotioned and other retirn -1 too
            }

            assert((cbp&15) == 0 || (cbp&15) == 15);

            if(cbp&15){

static int decode_mb_cabac(H264Context *h) {
    MpegEncContext * const s = &h->s;
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
    int mb_type, partition_count, cbp = 0;

    s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?)

    tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
    if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE ) {
        /* read skip flags */
        if( decode_cabac_mb_skip( h ) ) {

            }
        }

        ff_init_cabac_decoder(&h->cabac, ptr, h->cabac.bytestream_end - ptr);

        // All blocks are presents
        h->cbp_table[mb_xy] = 0x1ef;
        h->chroma_pred_mode_table[mb_xy] = 0;
        // In deblocking, the quantiser is 0
        s->current_picture.qscale_table[mb_xy]= 0;
        h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, 0);
        // All coeffs are presents
        memset(h->non_zero_count[mb_xy], 16, 16);
        return 0;
    }

    fill_caches(h, mb_type, 0);

            // and current and left pair do not have the same interlaced type
            && (IS_INTERLACED(s->current_picture.mb_type[mb_xy]) != IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]))
            // and left mb is in the same slice if deblocking_filter == 2
            && (h->deblocking_filter!=2 || h->slice_table[mb_xy-1] == h->slice_table[mb_xy])) {
        /* First vertical edge is different in MBAFF frames
         * There are 8 differents bS to compute and 2 differents Qp
         */
        int bS[8];
        int qp[2];
        int chroma_qp[2];

                    }
                }
            }
        }
        if(bS[0]+bS[1]+bS[2]+bS[3] != 0) {
            // Do not use s->qscale as luma quantiser because it has not the same
            // value in IPCM macroblocks.
            qp[0] = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[h->left_mb_xy[0]] + 1 ) >> 1;
            chroma_qp[0] = ( get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy] ) +
                             get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[h->left_mb_xy[0]] ) + 1 ) >> 1;
            qp[1] = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[h->left_mb_xy[1]] + 1 ) >> 1;

        if (h->deblocking_filter==2 && h->slice_table[mbm_xy] != h->slice_table[mb_xy])
            start = 1;

        /* Calculate bS */
        for( edge = start; edge < 4; edge++ ) {
            /* mbn_xy: neighbour macroblock */
            int mbn_xy = edge > 0 ? mb_xy : mbm_xy;
            int bS[4];
            int qp;

            if (h->mb_aff_frame && (dir == 1) && (edge == 0) && ((mb_y & 1) == 0)

                } else {
                    // TODO
                    assert(0);
                }
                /* Filter edge */
                // Do not use s->qscale as luma quantiser because it has not the same
                // value in IPCM macroblocks.
                qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
                tprintf("filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, tmp_linesize, tmp_uvlinesize);
                { int i; for (i = 0; i < 4; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); }
                filter_mb_edgeh( h, &img_y[0], tmp_linesize, bS, qp );

                } else {
                    // TODO
                    assert(0);
                }
                /* Filter edge */
                // Do not use s->qscale as luma quantiser because it has not the same
                // value in IPCM macroblocks.
                qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
                tprintf("filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, tmp_linesize, tmp_uvlinesize);
                { int i; for (i = 0; i < 4; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); }
                filter_mb_edgeh( h, &img_y[linesize], tmp_linesize, bS, qp );

                if(bS[0]+bS[1]+bS[2]+bS[3] == 0)
                    continue;
            }

            /* Filter edge */
            // Do not use s->qscale as luma quantiser because it has not the same
            // value in IPCM macroblocks.
            qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
            //tprintf("filter mb:%d/%d dir:%d edge:%d, QPy:%d, QPc:%d, QPcn:%d\n", mb_x, mb_y, dir, edge, qp, h->chroma_qp, s->current_picture.qscale_table[mbn_xy]);
            tprintf("filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, linesize, uvlinesize);
            { int i; for (i = 0; i < 4; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); }

        if( s->hurry_up == 1 && h->nal_ref_idc  == 0 )
            continue;
        
        switch(h->nal_unit_type){
        case NAL_IDR_SLICE:
            idr(h); //FIXME ensure we dont loose some frames if there is reordering
        case NAL_SLICE:
            init_get_bits(&s->gb, ptr, bit_length);
            h->intra_gb_ptr=
            h->inter_gb_ptr= &s->gb;
            s->data_partitioning = 0;

    return buf_index;
}

/**
 * retunrs the number of bytes consumed for building the current frame
 */
static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
    if(s->flags&CODEC_FLAG_TRUNCATED){
        pos -= s->parse_context.last_index;
        if(pos<0) pos=0; // FIXME remove (uneeded?)
        
        return pos;
    }else{
        if(pos==0) pos=1; //avoid infinite loops (i doubt thats needed but ...)
        if(pos+10>buf_size) pos=buf_size; // oops ;)

2645:42528c1f0246

    
    PPS pps_buffer[MAX_PPS_COUNT];
    /**
     * current pps
     */
    PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?

    int slice_num;
    uint8_t *slice_table_base;
    uint8_t *slice_table;      ///< slice_table_base + mb_stride + 1
    int slice_type;

    
    w      *= size;
    stride *= size;
    
    assert((((int)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
//FIXME check what gcc generates for 64 bit on x86 and possibly write a 32 bit ver of it
    if(w==2 && h==2){
        *(uint16_t*)(p + 0)=
        *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
    }else if(w==2 && h==4){
        *(uint16_t*)(p + 0*stride)=

    // the actual condition is whether we're on the edge of a slice,
    // and even then the intra and nnz parts are unnecessary.
    if(for_deblock && h->slice_num == 1)
        return;

    //wow what a mess, why didn't they simplify the interlacing&intra stuff, i can't imagine that these complex rules are worth it 
    
    top_xy     = mb_xy  - s->mb_stride;
    topleft_xy = top_xy - 1;
    topright_xy= top_xy + 1;
    left_xy[1] = left_xy[0] = mb_xy-1;

            }
            

            h->ref_cache[list][scan8[5 ]+1] = 
            h->ref_cache[list][scan8[7 ]+1] = 
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewhere else)
            h->ref_cache[list][scan8[4 ]] = 
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;

            if( h->pps.cabac ) {
                /* XXX beurk, Load mvd */

                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
                }
                *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
                *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
                *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
                *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
                *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;

                if(h->slice_type == B_TYPE){
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);

    int list;

    for(list=0; list<2; list++){
        int y;
        if(!USES_LIST(mb_type, list)){
            if(1){ //FIXME skip or never read if mb_type doesn't use it
                for(y=0; y<4; y++){
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
                }
                if( h->pps.cabac ) {

/**
 * Decodes a network abstraction layer unit.
 * @param consumed is the number of bytes used as input
 * @param length is the length of the array
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
 * @returns decoded bytes, might be src+1 if no escapes 
 */
static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
    int i, si, di;
    uint8_t *dst;

    }

    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
    dst= h->rbsp_buffer;

//printf("decoding esc\n");
    si=di=0;
    while(si<length){ 
        //remove escapes (very rare 1:2^22)
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
            if(src[si+2]==3){ //escape

        block[3*4 + i]=   z3 - 2*z2;
    }
}
#endif

//FIXME need to check that this doesnt overflow signed 32 bit for low qp, i am not sure, it's very close
//FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
    int i;
    const int * const quant_table= quant_coeff[qscale];
    const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;

    av_freep(&h->s.obmc_scratchpad);
}

/**
 * allocates tables.
 * needs width/height
 */
static int alloc_tables(H264Context *h){
    MpegEncContext * const s = &h->s;
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
    int x,y;

    
    src_y  -=   linesize;
    src_cb -= uvlinesize;
    src_cr -= uvlinesize;

    // There are two lines saved, the line above the the top macroblock of a pair,
    // and the line above the bottom macroblock
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
    for(i=1; i<17; i++){
        h->left_border[i]= src_y[15+i*  linesize];
    }

    
    src_y  -= 2 *   linesize;
    src_cb -= 2 * uvlinesize;
    src_cr -= 2 * uvlinesize;

    // There are two lines saved, the line above the the top macroblock of a pair,
    // and the line above the bottom macroblock
    h->left_border[0]= h->top_borders[0][s->mb_x][15];
    h->left_border[1]= h->top_borders[1][s->mb_x][15];
    for(i=2; i<34; i++){
        h->left_border[i]= src_y[15+i*  linesize];

            backup_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize);
            // TODO deblock a pair
            // top 
            s->mb_y--;
            tprintf("call mbaff filter_mb mb_x:%d mb_y:%d pair_dest_y = %p, dest_y = %p\n", mb_x, mb_y, pair_dest_y, dest_y);
            fill_caches(h, mb_type_top, 1); //FIXME don't fill stuff which isn't used by filter_mb
            filter_mb(h, mb_x, mb_y, pair_dest_y, pair_dest_cb, pair_dest_cr, linesize, uvlinesize);
            if (tmp != s->current_picture.data[1][384]) {
                tprintf("modified pixel 8,1 (1)\n");
            }
            // bottom
            s->mb_y++;
            tprintf("call mbaff filter_mb\n");
            fill_caches(h, mb_type_bottom, 1); //FIXME don't fill stuff which isn't used by filter_mb
            filter_mb(h, mb_x, mb_y+1, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
            if (tmp != s->current_picture.data[1][384]) {
                tprintf("modified pixel 8,1 (2)\n");
            }
        } else {
            tprintf("call filter_mb\n");
            backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
            fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
            filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
        }
    }
}

    MpegEncContext * const s = &h->s;
    int list;
    
    print_short_term(h);
    print_long_term(h);
    if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move before func
    
    for(list=0; list<2; list++){
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);

        if(get_bits1(&s->gb)){

    idr(h);
}

/**
 *
 * @return the removed picture or NULL if an error occurs
 */
static Picture * remove_short(H264Context *h, int frame_num){
    MpegEncContext * const s = &h->s;
    int i;
    

    return NULL;
}

/**
 *
 * @return the removed picture or NULL if an error occurs
 */
static Picture * remove_long(H264Context *h, int i){
    Picture *pic;

    pic= h->long_ref[i];

        || (h->slice_num != 0 && slice_type == h->slice_type) ) {
        default_ref_list_done = 1;
    }
    h->slice_type= slice_type;

    s->pict_type= h->slice_type; // to make a few old func happy, it's wrong though
        
    pps_id= get_ue_golomb(&s->gb);
    if(pps_id>255){
        av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
        return -1;

                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
            else
                level_code= prefix + get_bits(gb, 4); //part
        }else if(prefix==15){
            level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
            if(suffix_length==0) level_code+=15; //FIXME doesn't make (much)sense
        }else{
            av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
            return -1;
        }

static int decode_mb_cavlc(H264Context *h){
    MpegEncContext * const s = &h->s;
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
    int mb_type, partition_count, cbp;

    s->dsp.clear_blocks(h->mb); //FIXME avoid if already clear (move after skip handlong?    

    tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
    cbp = 0; /* avoid warning. FIXME: find a solution without slowing
                down the code */
    if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){

                tprintf("CHROMA V ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));
                h->mb[index + (x&3) + 16*(x>>2)]= get_bits(&s->gb, 8);
            }
        }
    
        // In deblocking, the quantizer is 0
        s->current_picture.qscale_table[mb_xy]= 0;
        h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, 0);
        // All coeffs are present
        memset(h->non_zero_count[mb_xy], 16, 16);
        
        return 0;
    }
        

        }
        
        h->chroma_qp= chroma_qp= get_chroma_qp(h->pps.chroma_qp_index_offset, s->qscale);
        if(IS_INTRA16x16(mb_type)){
            if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, s->qscale, 16) < 0){
                return -1; //FIXME continue if partitioned and other return -1 too
            }

            assert((cbp&15) == 0 || (cbp&15) == 15);

            if(cbp&15){

static int decode_mb_cabac(H264Context *h) {
    MpegEncContext * const s = &h->s;
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
    int mb_type, partition_count, cbp = 0;

    s->dsp.clear_blocks(h->mb); //FIXME avoid if already clear (move after skip handlong?)

    tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
    if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE ) {
        /* read skip flags */
        if( decode_cabac_mb_skip( h ) ) {

            }
        }

        ff_init_cabac_decoder(&h->cabac, ptr, h->cabac.bytestream_end - ptr);

        // All blocks are present
        h->cbp_table[mb_xy] = 0x1ef;
        h->chroma_pred_mode_table[mb_xy] = 0;
        // In deblocking, the quantizer is 0
        s->current_picture.qscale_table[mb_xy]= 0;
        h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, 0);
        // All coeffs are present
        memset(h->non_zero_count[mb_xy], 16, 16);
        return 0;
    }

    fill_caches(h, mb_type, 0);

            // and current and left pair do not have the same interlaced type
            && (IS_INTERLACED(s->current_picture.mb_type[mb_xy]) != IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]))
            // and left mb is in the same slice if deblocking_filter == 2
            && (h->deblocking_filter!=2 || h->slice_table[mb_xy-1] == h->slice_table[mb_xy])) {
        /* First vertical edge is different in MBAFF frames
         * There are 8 different bS to compute and 2 different Qp
         */
        int bS[8];
        int qp[2];
        int chroma_qp[2];

                    }
                }
            }
        }
        if(bS[0]+bS[1]+bS[2]+bS[3] != 0) {
            // Do not use s->qscale as luma quantizer because it has not the same
            // value in IPCM macroblocks.
            qp[0] = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[h->left_mb_xy[0]] + 1 ) >> 1;
            chroma_qp[0] = ( get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy] ) +
                             get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[h->left_mb_xy[0]] ) + 1 ) >> 1;
            qp[1] = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[h->left_mb_xy[1]] + 1 ) >> 1;

        if (h->deblocking_filter==2 && h->slice_table[mbm_xy] != h->slice_table[mb_xy])
            start = 1;

        /* Calculate bS */
        for( edge = start; edge < 4; edge++ ) {
            /* mbn_xy: neighbor macroblock */
            int mbn_xy = edge > 0 ? mb_xy : mbm_xy;
            int bS[4];
            int qp;

            if (h->mb_aff_frame && (dir == 1) && (edge == 0) && ((mb_y & 1) == 0)

                } else {
                    // TODO
                    assert(0);
                }
                /* Filter edge */
                // Do not use s->qscale as luma quantizer because it has not the same
                // value in IPCM macroblocks.
                qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
                tprintf("filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, tmp_linesize, tmp_uvlinesize);
                { int i; for (i = 0; i < 4; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); }
                filter_mb_edgeh( h, &img_y[0], tmp_linesize, bS, qp );

                } else {
                    // TODO
                    assert(0);
                }
                /* Filter edge */
                // Do not use s->qscale as luma quantizer because it has not the same
                // value in IPCM macroblocks.
                qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
                tprintf("filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, tmp_linesize, tmp_uvlinesize);
                { int i; for (i = 0; i < 4; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); }
                filter_mb_edgeh( h, &img_y[linesize], tmp_linesize, bS, qp );

                if(bS[0]+bS[1]+bS[2]+bS[3] == 0)
                    continue;
            }

            /* Filter edge */
            // Do not use s->qscale as luma quantizer because it has not the same
            // value in IPCM macroblocks.
            qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
            //tprintf("filter mb:%d/%d dir:%d edge:%d, QPy:%d, QPc:%d, QPcn:%d\n", mb_x, mb_y, dir, edge, qp, h->chroma_qp, s->current_picture.qscale_table[mbn_xy]);
            tprintf("filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, linesize, uvlinesize);
            { int i; for (i = 0; i < 4; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); }

        if( s->hurry_up == 1 && h->nal_ref_idc  == 0 )
            continue;
        
        switch(h->nal_unit_type){
        case NAL_IDR_SLICE:
            idr(h); //FIXME ensure we don't loose some frames if there is reordering
        case NAL_SLICE:
            init_get_bits(&s->gb, ptr, bit_length);
            h->intra_gb_ptr=
            h->inter_gb_ptr= &s->gb;
            s->data_partitioning = 0;

    return buf_index;
}

/**
 * returns the number of bytes consumed for building the current frame
 */
static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
    if(s->flags&CODEC_FLAG_TRUNCATED){
        pos -= s->parse_context.last_index;
        if(pos<0) pos=0; // FIXME remove (unneeded?)
        
        return pos;
    }else{
        if(pos==0) pos=1; //avoid infinite loops (i doubt thats needed but ...)
        if(pos+10>buf_size) pos=buf_size; // oops ;)