h264.c

ffmpeg源码分析

            h->non_zero_count_cache[3+8*1 + 2*8*i]=
            h->non_zero_count_cache[3+8*2 + 2*8*i]=
            h->non_zero_count_cache[0+8*1 +   8*i]=
            h->non_zero_count_cache[0+8*4 +   8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
        }
    }

    if( h->pps.cabac ) {
        // top_cbp
        if(top_type) {
            h->top_cbp = h->cbp_table[top_xy];
        } else if(IS_INTRA(mb_type)) {
            h->top_cbp = 0x1C0;
        } else {
            h->top_cbp = 0;
        }
        // left_cbp
        if (left_type[0]) {
            h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
        } else if(IS_INTRA(mb_type)) {
            h->left_cbp = 0x1C0;
        } else {
            h->left_cbp = 0;
        }
        if (left_type[0]) {
            h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
        }
        if (left_type[1]) {
            h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
        }
    }

#if 1
    //FIXME direct mb can skip much of this
    if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
        int list;
        for(list=0; list<1+(h->slice_type==B_TYPE); list++){
            if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
                /*if(!h->mv_cache_clean[list]){
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
                    h->mv_cache_clean[list]= 1;
                }*/
                continue;
            }
            h->mv_cache_clean[list]= 0;

            if(USES_LIST(top_type, list)){
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
                *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
                *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
                *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
                *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
                h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
                h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
            }else{
                *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
                *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
                *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
                *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
            }

            //FIXME unify cleanup or sth
            if(USES_LIST(left_type[0], list)){
                const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
                const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 0*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0]];
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[1]];
                h->ref_cache[list][scan8[0] - 1 + 0*8]=
                h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
            }else{
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
                h->ref_cache[list][scan8[0] - 1 + 0*8]=
                h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
            }

            if(USES_LIST(left_type[1], list)){
                const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
                const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 2*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[2]];
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 3*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[3]];
                h->ref_cache[list][scan8[0] - 1 + 2*8]=
                h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
            }else{
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
                h->ref_cache[list][scan8[0] - 1 + 2*8]=
                h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
                assert((!left_type[0]) == (!left_type[1]));
            }

            if(for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred))
                continue;

            if(USES_LIST(topleft_type, list)){
                const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
            }else{
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
            }

            if(USES_LIST(topright_type, list)){
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
            }else{
                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
            }


            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 */
                if(USES_LIST(topleft_type, list)){
                    const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
                }else{
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
                }

                if(USES_LIST(top_type, list)){
                    const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
                }else{
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
                }
                if(USES_LIST(left_type[0], list)){
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 0*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[0]];
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[1]];
                }else{
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
                }
                if(USES_LIST(left_type[1], list)){
                    const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 2*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[2]];
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 3*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[3]];
                }else{
                    *(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);

                    if(IS_DIRECT(top_type)){
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
                    }else if(IS_8X8(top_type)){
                        int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
                    }else{
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
                    }

                    //FIXME interlacing
                    if(IS_DIRECT(left_type[0])){
                        h->direct_cache[scan8[0] - 1 + 0*8]=
                        h->direct_cache[scan8[0] - 1 + 2*8]= 1;
                    }else if(IS_8X8(left_type[0])){
                        int b8_xy = h->mb2b8_xy[left_xy[0]] + 1;
                        h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[b8_xy];
                        h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[b8_xy + h->b8_stride];
                    }else{
                        h->direct_cache[scan8[0] - 1 + 0*8]=
                        h->direct_cache[scan8[0] - 1 + 2*8]= 0;
                    }
                }
            }
        }
    }
#endif

    h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
}

static inline void write_back_intra_pred_mode(H264Context *h){
    MpegEncContext * const s = &h->s;
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;

    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
}

/**
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
 */
static inline int check_intra4x4_pred_mode(H264Context *h){
    MpegEncContext * const s = &h->s;
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
    int i;

    if(!(h->top_samples_available&0x8000)){
        for(i=0; i<4; i++){
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
            if(status<0){
                av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
                return -1;
            } else if(status){
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
            }
        }
    }

    if(!(h->left_samples_available&0x8000)){
        for(i=0; i<4; i++){
            int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
            if(status<0){
                av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
                return -1;
            } else if(status){
                h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
            }
        }
    }

    return 0;
} //FIXME cleanup like next

/**
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
 */
static inline int check_intra_pred_mode(H264Context *h, int mode){
    MpegEncContext * const s = &h->s;
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};

    if(mode < 0 || mode > 6) {
        av_log(h->s.avctx, AV_LOG_ERROR, "out of range intra chroma pred mode at %d %d\n", s->mb_x, s->mb_y);
        return -1;
    }

    if(!(h->top_samples_available&0x8000)){
        mode= top[ mode ];
        if(mode<0){
            av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
            return -1;
        }
    }

    if(!(h->left_samples_available&0x8000)){
        mode= left[ mode ];
        if(mode<0){
            av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
            return -1;
        }
    }

    return mode;
}

/**
 * gets the predicted intra4x4 prediction mode.
 */
static inline int pred_intra_mode(H264Context *h, int n){
    const int index8= scan8[n];
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
    const int min= FFMIN(left, top);

    tprintf("mode:%d %d min:%d\n", left ,top, min);

    if(min<0) return DC_PRED;
    else      return min;
}

static inline void write_back_non_zero_count(H264Context *h){
    MpegEncContext * const s = &h->s;
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;

    h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
    h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
    h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];

    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];

    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
}

/**
 * gets the predicted number of non zero coefficients.
 * @param n block index
 */
static inline int pred_non_zero_count(H264Context *h, int n){
    const int index8= scan8[n];
    const int left= h->non_zero_count_cache[index8 - 1];
    const int top = h->non_zero_count_cache[index8 - 8];
    int i= left + top;