h264.c

ffmpeg源码分析

    GetBitContext *inter_gb_ptr;

    DECLARE_ALIGNED_8(DCTELEM, mb[16*24]);

    /**
     * Cabac
     */
    CABACContext cabac;
    uint8_t      cabac_state[460];
    int          cabac_init_idc;

    /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
    uint16_t     *cbp_table;
    int top_cbp;
    int left_cbp;
    /* chroma_pred_mode for i4x4 or i16x16, else 0 */
    uint8_t     *chroma_pred_mode_table;
    int         last_qscale_diff;
    int16_t     (*mvd_table[2])[2];
    DECLARE_ALIGNED_8(int16_t, mvd_cache[2][5*8][2]);
    uint8_t     *direct_table;
    uint8_t     direct_cache[5*8];

    uint8_t zigzag_scan[16];
    uint8_t field_scan[16];
    uint8_t zigzag_scan8x8[64];
    uint8_t zigzag_scan8x8_cavlc[64];
    const uint8_t *zigzag_scan_q0;
    const uint8_t *field_scan_q0;
    const uint8_t *zigzag_scan8x8_q0;
    const uint8_t *zigzag_scan8x8_cavlc_q0;

    int x264_build;
}H264Context;

static VLC coeff_token_vlc[4];
static VLC chroma_dc_coeff_token_vlc;

static VLC total_zeros_vlc[15];
static VLC chroma_dc_total_zeros_vlc[3];

static VLC run_vlc[6];
static VLC run7_vlc;

static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize);

static always_inline uint32_t pack16to32(int a, int b){
#ifdef WORDS_BIGENDIAN
   return (b&0xFFFF) + (a<<16);
#else
   return (a&0xFFFF) + (b<<16);
#endif
}

/**
 * fill a rectangle.
 * @param h height of the rectangle, should be a constant
 * @param w width of the rectangle, should be a constant
 * @param size the size of val (1 or 4), should be a constant
 */
static always_inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){
    uint8_t *p= (uint8_t*)vp;
    assert(size==1 || size==4);

    w      *= size;
    stride *= size;

    assert((((long)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
    assert((stride&(w-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)=
        *(uint16_t*)(p + 1*stride)=
        *(uint16_t*)(p + 2*stride)=
        *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
    }else if(w==4 && h==1){
        *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
    }else if(w==4 && h==2){
        *(uint32_t*)(p + 0*stride)=
        *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
    }else if(w==4 && h==4){
        *(uint32_t*)(p + 0*stride)=
        *(uint32_t*)(p + 1*stride)=
        *(uint32_t*)(p + 2*stride)=
        *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
    }else if(w==8 && h==1){
        *(uint32_t*)(p + 0)=
        *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
    }else if(w==8 && h==2){
        *(uint32_t*)(p + 0 + 0*stride)=
        *(uint32_t*)(p + 4 + 0*stride)=
        *(uint32_t*)(p + 0 + 1*stride)=
        *(uint32_t*)(p + 4 + 1*stride)=  size==4 ? val : val*0x01010101;
    }else if(w==8 && h==4){
        *(uint64_t*)(p + 0*stride)=
        *(uint64_t*)(p + 1*stride)=
        *(uint64_t*)(p + 2*stride)=
        *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
    }else if(w==16 && h==2){
        *(uint64_t*)(p + 0+0*stride)=
        *(uint64_t*)(p + 8+0*stride)=
        *(uint64_t*)(p + 0+1*stride)=
        *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
    }else if(w==16 && h==4){
        *(uint64_t*)(p + 0+0*stride)=
        *(uint64_t*)(p + 8+0*stride)=
        *(uint64_t*)(p + 0+1*stride)=
        *(uint64_t*)(p + 8+1*stride)=
        *(uint64_t*)(p + 0+2*stride)=
        *(uint64_t*)(p + 8+2*stride)=
        *(uint64_t*)(p + 0+3*stride)=
        *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
    }else
        assert(0);
}

static void fill_caches(H264Context *h, int mb_type, int for_deblock){
    MpegEncContext * const s = &h->s;
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
    int topleft_xy, top_xy, topright_xy, left_xy[2];
    int topleft_type, top_type, topright_type, left_type[2];
    int left_block[8];
    int i;

    //FIXME deblocking can skip fill_caches much of the time with multiple slices too.
    // 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;
    left_block[0]= 0;
    left_block[1]= 1;
    left_block[2]= 2;
    left_block[3]= 3;
    left_block[4]= 7;
    left_block[5]= 10;
    left_block[6]= 8;
    left_block[7]= 11;
    if(h->mb_aff_frame){
        const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;
        const int top_pair_xy      = pair_xy     - s->mb_stride;
        const int topleft_pair_xy  = top_pair_xy - 1;
        const int topright_pair_xy = top_pair_xy + 1;
        const int topleft_mb_frame_flag  = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
        const int top_mb_frame_flag      = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
        const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
        const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
        const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
        const int bottom = (s->mb_y & 1);
        tprintf("fill_caches: curr_mb_frame_flag:%d, left_mb_frame_flag:%d, topleft_mb_frame_flag:%d, top_mb_frame_flag:%d, topright_mb_frame_flag:%d\n", curr_mb_frame_flag, left_mb_frame_flag, topleft_mb_frame_flag, top_mb_frame_flag, topright_mb_frame_flag);
        if (bottom
                ? !curr_mb_frame_flag // bottom macroblock
                : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
                ) {
            top_xy -= s->mb_stride;
        }
        if (bottom
                ? !curr_mb_frame_flag // bottom macroblock
                : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
                ) {
            topleft_xy -= s->mb_stride;
        }
        if (bottom
                ? !curr_mb_frame_flag // bottom macroblock
                : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
                ) {
            topright_xy -= s->mb_stride;
        }
        if (left_mb_frame_flag != curr_mb_frame_flag) {
            left_xy[1] = left_xy[0] = pair_xy - 1;
            if (curr_mb_frame_flag) {
                if (bottom) {
                    left_block[0]= 2;
                    left_block[1]= 2;
                    left_block[2]= 3;
                    left_block[3]= 3;
                    left_block[4]= 8;
                    left_block[5]= 11;
                    left_block[6]= 8;
                    left_block[7]= 11;
                } else {
                    left_block[0]= 0;
                    left_block[1]= 0;
                    left_block[2]= 1;
                    left_block[3]= 1;
                    left_block[4]= 7;
                    left_block[5]= 10;
                    left_block[6]= 7;
                    left_block[7]= 10;
                }
            } else {
                left_xy[1] += s->mb_stride;
                //left_block[0]= 0;
                left_block[1]= 2;
                left_block[2]= 0;
                left_block[3]= 2;
                //left_block[4]= 7;
                left_block[5]= 10;
                left_block[6]= 7;
                left_block[7]= 10;
            }
        }
    }

    h->top_mb_xy = top_xy;
    h->left_mb_xy[0] = left_xy[0];
    h->left_mb_xy[1] = left_xy[1];
    if(for_deblock){
        topleft_type = h->slice_table[topleft_xy ] < 255 ? s->current_picture.mb_type[topleft_xy] : 0;
        top_type     = h->slice_table[top_xy     ] < 255 ? s->current_picture.mb_type[top_xy]     : 0;
        topright_type= h->slice_table[topright_xy] < 255 ? s->current_picture.mb_type[topright_xy]: 0;
        left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
        left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
    }else{
        topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
        top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
        topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
        left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
        left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
    }

    if(IS_INTRA(mb_type)){
        h->topleft_samples_available=
        h->top_samples_available=
        h->left_samples_available= 0xFFFF;
        h->topright_samples_available= 0xEEEA;

        if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
            h->topleft_samples_available= 0xB3FF;
            h->top_samples_available= 0x33FF;
            h->topright_samples_available= 0x26EA;
        }
        for(i=0; i<2; i++){
            if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
                h->topleft_samples_available&= 0xDF5F;
                h->left_samples_available&= 0x5F5F;
            }
        }

        if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
            h->topleft_samples_available&= 0x7FFF;

        if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
            h->topright_samples_available&= 0xFBFF;

        if(IS_INTRA4x4(mb_type)){
            if(IS_INTRA4x4(top_type)){
                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
            }else{
                int pred;
                if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
                    pred= -1;
                else{
                    pred= 2;
                }
                h->intra4x4_pred_mode_cache[4+8*0]=
                h->intra4x4_pred_mode_cache[5+8*0]=
                h->intra4x4_pred_mode_cache[6+8*0]=
                h->intra4x4_pred_mode_cache[7+8*0]= pred;
            }
            for(i=0; i<2; i++){
                if(IS_INTRA4x4(left_type[i])){
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
                }else{
                    int pred;
                    if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
                        pred= -1;
                    else{
                        pred= 2;
                    }
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
                }
            }
        }
    }


/*
0 . T T. T T T T
1 L . .L . . . .
2 L . .L . . . .
3 . T TL . . . .
4 L . .L . . . .
5 L . .. . . . .
*/
//FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
    if(top_type){
        h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
        h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
        h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
        h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];

        h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
        h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];

        h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
        h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];

    }else{
        h->non_zero_count_cache[4+8*0]=
        h->non_zero_count_cache[5+8*0]=
        h->non_zero_count_cache[6+8*0]=
        h->non_zero_count_cache[7+8*0]=

        h->non_zero_count_cache[1+8*0]=
        h->non_zero_count_cache[2+8*0]=

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

    }

    for (i=0; i<2; i++) {
        if(left_type[i]){
            h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
            h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
            h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
            h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
        }else{