h264.c

手机端的H264源码

    
        h->non_zero_count_cache[1+8*3]=
        h->non_zero_count_cache[2+8*3]= 64;
    }
    
    if(left_type[0]){
        h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][6];
        h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][5];
        h->non_zero_count_cache[0+8*1]= h->non_zero_count[left_xy[0]][9]; //FIXME left_block
        h->non_zero_count_cache[0+8*4]= h->non_zero_count[left_xy[0]][12];
    }else{
        h->non_zero_count_cache[3+8*1]= 
        h->non_zero_count_cache[3+8*2]= 
        h->non_zero_count_cache[0+8*1]= 
        h->non_zero_count_cache[0+8*4]= 64;
    }
    
    if(left_type[1]){
        h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[1]][4];
        h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[1]][3];
        h->non_zero_count_cache[0+8*2]= h->non_zero_count[left_xy[1]][8];
        h->non_zero_count_cache[0+8*5]= h->non_zero_count[left_xy[1]][11];
    }else{
        h->non_zero_count_cache[3+8*3]= 
        h->non_zero_count_cache[3+8*4]= 
        h->non_zero_count_cache[0+8*2]= 
        h->non_zero_count_cache[0+8*5]= 64;
    }
    
#if 1
    if(IS_INTER(mb_type)){
        int list;
        for(list=0; list<2; list++){
            if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
                
                continue; //FIXME direct mode ...
            }
            
            if(IS_INTER(topleft_type)){
                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(IS_INTER(top_type)){
                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;
            }

            if(IS_INTER(topright_type)){
                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;
            }
            
            //FIXME unify cleanup or sth
            if(IS_INTER(left_type[0])){
                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(IS_INTER(left_type[1])){
                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;
            }

            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 */
                if(IS_INTER(topleft_type)){
                    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(IS_INTER(top_type)){
                    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(IS_INTER(left_type[0])){
                    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(IS_INTER(left_type[1])){
                    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 somewher else)
                *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
                *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
            }
        }
//FIXME
    }
#endif
}

static  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  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  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(!(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  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  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[4+8*4];
    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[5+8*4];
    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[6+8*4];
    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[7+8*3];
    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[7+8*2];
    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[7+8*1];
    
    h->non_zero_count[mb_xy][7]= 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][9]= h->non_zero_count_cache[2+8*1];

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

/**
 * gets the predicted number of non zero coefficients.
 * @param n block index
 */
static  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;
    
    if(i<64) i= (i+1)>>1;

    //tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);

    return i&31;
}

static  int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];

    if(topright_ref != PART_NOT_AVAILABLE){
        *C= h->mv_cache[list][ i - 8 + part_width ];
        return topright_ref;
    }else{
        //tprintf("topright MV not available\n");

        *C= h->mv_cache[list][ i - 8 - 1 ];
        return h->ref_cache[list][ i - 8 - 1 ];
    }
}

/**
 * gets the predicted MV.
 * @param n the block index
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
 * @param mx the x component of the predicted motion vector
 * @param my the y component of the predicted motion vector
 */
static  void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
    const int index8= scan8[n];
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
    const int16_t * C;
    int diagonal_ref, match_count;

    assert(part_width==1 || part_width==2 || part_width==4);

/* mv_cache
  B . . A T T T T 
  U . . L . . , .
  U . . L . . . .
  U . . L . . , .
  . . . L . . . .
*/

    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
    if(match_count > 1){ //most common
        *mx= mid_pred(A[0], B[0], C[0]);
        *my= mid_pred(A[1], B[1], C[1]);
    }else if(match_count==1){
        if(left_ref==ref){
            *mx= A[0];
            *my= A[1];        
        }else if(top_ref==ref){
            *mx= B[0];
            *my= B[1];        
        }else{
            *mx= C[0];
            *my= C[1];        
        }
    }else{
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
            *mx= A[0];
            *my= A[1];        
        }else{
            *mx= mid_pred(A[0], B[0], C[0]);
            *my= mid_pred(A[1], B[1], C[1]);
        }
    }
        
    //tprintf("pred_motion (%2d %2d %2d) (%2d %2d %2d) (%2d %2d %2d) -> (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1],                    diagonal_ref, C[0], C[1], left_ref, A[0], A[1], ref, *mx, *my, h->s.mb_x, h->s.mb_y, n, list);