h264.c

ffmpeg源码分析

                        if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
                            if(ref[0] == 0)
                                *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
                            if(ref[1] == 0)
                                *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
                        }
                    }
                }
            }
        }
    }else{ /* direct temporal mv pred */
        if(IS_16X16(*mb_type)){
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
            if(IS_INTRA(mb_type_col)){
                fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
                fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
                fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
            }else{
                const int ref0 = l1ref0[0] >= 0 ? h->map_col_to_list0[0][l1ref0[0]]
                                                : h->map_col_to_list0[1][l1ref1[0]];
                const int dist_scale_factor = h->dist_scale_factor[ref0];
                const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
                int mv_l0[2];
                mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
                mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
                fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
                fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
                fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]), 4);
            }
        }else{
            for(i8=0; i8<4; i8++){
                const int x8 = i8&1;
                const int y8 = i8>>1;
                int ref0, dist_scale_factor;
                const int16_t (*l1mv)[2]= l1mv0;

                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
                    continue;
                h->sub_mb_type[i8] = sub_mb_type;
                if(IS_INTRA(mb_type_col)){
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
                    fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
                    continue;
                }

                ref0 = l1ref0[x8 + y8*h->b8_stride];
                if(ref0 >= 0)
                    ref0 = h->map_col_to_list0[0][ref0];
                else{
                    ref0 = h->map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
                    l1mv= l1mv1;
                }
                dist_scale_factor = h->dist_scale_factor[ref0];

                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
                if(IS_SUB_8X8(sub_mb_type)){
                    const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
                    int mx = (dist_scale_factor * mv_col[0] + 128) >> 8;
                    int my = (dist_scale_factor * mv_col[1] + 128) >> 8;
                    fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
                    fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
                }else
                for(i4=0; i4<4; i4++){
                    const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
                    int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
                    mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
                    mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
                    *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
                }
            }
        }
    }
}

static inline void write_back_motion(H264Context *h, int mb_type){
    MpegEncContext * const s = &h->s;
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
    int list;

    if(!USES_LIST(mb_type, 0))
        fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);

    for(list=0; list<2; list++){
        int y;
        if(!USES_LIST(mb_type, list))
            continue;

        for(y=0; y<4; y++){
            *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+0 + 8*y];
            *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+2 + 8*y];
        }
        if( h->pps.cabac ) {
            for(y=0; y<4; y++){
                *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
                *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
            }
        }

        {
            uint8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
            ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
            ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
            ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
            ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
        }
    }

    if(h->slice_type == B_TYPE && h->pps.cabac){
        if(IS_8X8(mb_type)){
            uint8_t *direct_table = &h->direct_table[b8_xy];
            direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
            direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
            direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
        }
    }
}

/**
 * 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;

//    src[0]&0x80;                //forbidden bit
    h->nal_ref_idc= src[0]>>5;
    h->nal_unit_type= src[0]&0x1F;

    src++; length--;
#if 0
    for(i=0; i<length; i++)
        printf("%2X ", src[i]);
#endif
    for(i=0; i+1<length; i+=2){
        if(src[i]) continue;
        if(i>0 && src[i-1]==0) i--;
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
            if(src[i+2]!=3){
                /* startcode, so we must be past the end */
                length=i;
            }
            break;
        }
    }

    if(i>=length-1){ //no escaped 0
        *dst_length= length;
        *consumed= length+1; //+1 for the header
        return src;
    }

    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
                dst[di++]= 0;
                dst[di++]= 0;
                si+=3;
                continue;
            }else //next start code
                break;
        }

        dst[di++]= src[si++];
    }

    *dst_length= di;
    *consumed= si + 1;//+1 for the header
//FIXME store exact number of bits in the getbitcontext (its needed for decoding)
    return dst;
}

#if 0
/**
 * @param src the data which should be escaped
 * @param dst the target buffer, dst+1 == src is allowed as a special case
 * @param length the length of the src data
 * @param dst_length the length of the dst array
 * @returns length of escaped data in bytes or -1 if an error occured
 */
static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
    int i, escape_count, si, di;
    uint8_t *temp;

    assert(length>=0);
    assert(dst_length>0);

    dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;

    if(length==0) return 1;

    escape_count= 0;
    for(i=0; i<length; i+=2){
        if(src[i]) continue;
        if(i>0 && src[i-1]==0)
            i--;
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
            escape_count++;
            i+=2;
        }
    }

    if(escape_count==0){
        if(dst+1 != src)
            memcpy(dst+1, src, length);
        return length + 1;
    }

    if(length + escape_count + 1> dst_length)
        return -1;

    //this should be damn rare (hopefully)

    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
    temp= h->rbsp_buffer;
//printf("encoding esc\n");

    si= 0;
    di= 0;
    while(si < length){
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
            temp[di++]= 0; si++;
            temp[di++]= 0; si++;
            temp[di++]= 3;
            temp[di++]= src[si++];
        }
        else
            temp[di++]= src[si++];
    }
    memcpy(dst+1, temp, length+escape_count);

    assert(di == length+escape_count);

    return di + 1;
}

/**
 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
 */
static void encode_rbsp_trailing(PutBitContext *pb){
    int length;
    put_bits(pb, 1, 1);
    length= (-put_bits_count(pb))&7;
    if(length) put_bits(pb, length, 0);
}
#endif

/**
 * identifies the exact end of the bitstream
 * @return the length of the trailing, or 0 if damaged
 */
static int decode_rbsp_trailing(uint8_t *src){
    int v= *src;
    int r;

    tprintf("rbsp trailing %X\n", v);

    for(r=1; r<9; r++){
        if(v&1) return r;
        v>>=1;
    }
    return 0;
}

/**
 * idct tranforms the 16 dc values and dequantize them.
 * @param qp quantization parameter
 */
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
#define stride 16
    int i;
    int temp[16]; //FIXME check if this is a good idea
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};

//memset(block, 64, 2*256);
//return;
    for(i=0; i<4; i++){
        const int offset= y_offset[i];
        const int z0= block[offset+stride*0] + block[offset+stride*4];
        const int z1= block[offset+stride*0] - block[offset+stride*4];
        const int z2= block[offset+stride*1] - block[offset+stride*5];
        const int z3= block[offset+stride*1] + block[offset+stride*5];

        temp[4*i+0]= z0+z3;
        temp[4*i+1]= z1+z2;
        temp[4*i+2]= z1-z2;
        temp[4*i+3]= z0-z3;
    }

    for(i=0; i<4; i++){
        const int offset= x_offset[i];
        const int z0= temp[4*0+i] + temp[4*2+i];
        const int z1= temp[4*0+i] - temp[4*2+i];
        const int z2= temp[4*1+i] - temp[4*3+i];
        const int z3= temp[4*1+i] + temp[4*3+i];

        block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_resdual
        block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
        block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
        block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
    }
}

#if 0
/**
 * dct tranforms the 16 dc values.
 * @param qp quantization parameter ??? FIXME
 */
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
//    const int qmul= dequant_coeff[qp][0];
    int i;
    int temp[16]; //FIXME check if this is a good idea
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};

    for(i=0; i<4; i++){
        const int offset= y_offset[i];
        const int z0= block[offset+stride*0] + block[offset+stride*4];
        const int z1= block[offset+stride*0] - block[of