snow.c

从FFMPEG转换而来的H264解码程序,VC下编译..

        }

        (xc++)->x= w+1; //end marker
    }
}

static inline void decode_subband_slice_buffered(SnowContext *s, SubBand *b, slice_buffer * sb, int start_y, int h, int save_state[1]){
    const int w= b->width;
    int y;
    const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16);
    int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
    int qadd= (s->qbias*qmul)>>QBIAS_SHIFT;
    int new_index = 0;

    if(b->buf == s->spatial_dwt_buffer || s->qlog == LOSSLESS_QLOG){
        qadd= 0;
        qmul= 1<<QEXPSHIFT;
    }

    /* If we are on the second or later slice, restore our index. */
    if (start_y != 0)
        new_index = save_state[0];


    for(y=start_y; y<h; y++){
        int x = 0;
        int v;
        DWTELEM * line = slice_buffer_get_line(sb, y * b->stride_line + b->buf_y_offset) + b->buf_x_offset;
        memset(line, 0, b->width*sizeof(DWTELEM));
        v = b->x_coeff[new_index].coeff;
        x = b->x_coeff[new_index++].x;
        while(x < w)
        {
            register int t= ( (v>>1)*qmul + qadd)>>QEXPSHIFT;
            register int u= -(v&1);
            line[x] = (t^u) - u;

            v = b->x_coeff[new_index].coeff;
            x = b->x_coeff[new_index++].x;
        }
    }

    /* Save our variables for the next slice. */
    save_state[0] = new_index;

    return;
}

static void reset_contexts(SnowContext *s){ //FIXME better initial contexts
    int plane_index, level, orientation;

    for(plane_index=0; plane_index<3; plane_index++){
        for(level=0; level<s->spatial_decomposition_count; level++){
            for(orientation=level ? 1:0; orientation<4; orientation++){
                memset(s->plane[plane_index].band[level][orientation].state, MID_STATE, sizeof(s->plane[plane_index].band[level][orientation].state));
            }
        }
    }
    memset(s->header_state, MID_STATE, sizeof(s->header_state));
    memset(s->block_state, MID_STATE, sizeof(s->block_state));
}

static int alloc_blocks(SnowContext *s){
    int w= -((-s->avctx->width )>>LOG2_MB_SIZE);
    int h= -((-s->avctx->height)>>LOG2_MB_SIZE);

    s->b_width = w;
    s->b_height= h;

    s->block= av_mallocz(w * h * sizeof(BlockNode) << (s->block_max_depth*2));
    return 0;
}

static inline void copy_rac_state(RangeCoder *d, RangeCoder *s){
    uint8_t *bytestream= d->bytestream;
    uint8_t *bytestream_start= d->bytestream_start;
    *d= *s;
    d->bytestream= bytestream;
    d->bytestream_start= bytestream_start;
}

//near copy & paste from dsputil, FIXME
static int pix_sum(uint8_t * pix, int line_size, int w)
{
    int s, i, j;

    s = 0;
    for (i = 0; i < w; i++) {
        for (j = 0; j < w; j++) {
            s += pix[0];
            pix ++;
        }
        pix += line_size - w;
    }
    return s;
}

//near copy & paste from dsputil, FIXME
static int pix_norm1(uint8_t * pix, int line_size, int w)
{
    int s, i, j;
    uint32_t *sq = ff_squareTbl + 256;

    s = 0;
    for (i = 0; i < w; i++) {
        for (j = 0; j < w; j ++) {
            s += sq[pix[0]];
            pix ++;
        }
        pix += line_size - w;
    }
    return s;
}

static inline void set_blocks(SnowContext *s, int level, int x, int y, int l, int cb, int cr, int mx, int my, int ref, int type){
    const int w= s->b_width << s->block_max_depth;
    const int rem_depth= s->block_max_depth - level;
    const int index= (x + y*w) << rem_depth;
    const int block_w= 1<<rem_depth;
    BlockNode block;
    int i,j;

    block.color[0]= l;
    block.color[1]= cb;
    block.color[2]= cr;
    block.mx= mx;
    block.my= my;
    block.ref= ref;
    block.type= type;
    block.level= level;

    for(j=0; j<block_w; j++){
        for(i=0; i<block_w; i++){
            s->block[index + i + j*w]= block;
        }
    }
}

static inline void init_ref(MotionEstContext *c, uint8_t *src[3], uint8_t *ref[3], uint8_t *ref2[3], int x, int y, int ref_index){
    const int offset[3]= {
          y*c->  stride + x,
        ((y*c->uvstride + x)>>1),
        ((y*c->uvstride + x)>>1),
    };
    int i;
    for(i=0; i<3; i++){
        c->src[0][i]= src [i];
        c->ref[0][i]= ref [i] + offset[i];
    }
    assert(!ref_index);
}

static inline void pred_mv(SnowContext *s, int *mx, int *my, int ref,
                           const BlockNode *left, const BlockNode *top, const BlockNode *tr){
    if(s->ref_frames == 1){
        *mx = mid_pred(left->mx, top->mx, tr->mx);
        *my = mid_pred(left->my, top->my, tr->my);
    }else{
        const int *scale = scale_mv_ref[ref];
        *mx = mid_pred((left->mx * scale[left->ref] + 128) >>8,
                       (top ->mx * scale[top ->ref] + 128) >>8,
                       (tr  ->mx * scale[tr  ->ref] + 128) >>8);
        *my = mid_pred((left->my * scale[left->ref] + 128) >>8,
                       (top ->my * scale[top ->ref] + 128) >>8,
                       (tr  ->my * scale[tr  ->ref] + 128) >>8);
    }
}

//FIXME copy&paste
#define P_LEFT P[1]
#define P_TOP P[2]
#define P_TOPRIGHT P[3]
#define P_MEDIAN P[4]
#define P_MV1 P[9]
#define FLAG_QPEL   1 //must be 1

static int encode_q_branch(SnowContext *s, int level, int x, int y){
    uint8_t p_buffer[1024];
    uint8_t i_buffer[1024];
    uint8_t p_state[sizeof(s->block_state)];
    uint8_t i_state[sizeof(s->block_state)];
    RangeCoder pc, ic;
    uint8_t *pbbak= s->c.bytestream;
    uint8_t *pbbak_start= s->c.bytestream_start;
    int score, score2, iscore, i_len, p_len, block_s, sum, base_bits;
    const int w= s->b_width  << s->block_max_depth;
    const int h= s->b_height << s->block_max_depth;
    const int rem_depth= s->block_max_depth - level;
    const int index= (x + y*w) << rem_depth;
    const int block_w= 1<<(LOG2_MB_SIZE - level);
    int trx= (x+1)<<rem_depth;
    int try= (y+1)<<rem_depth;
    const BlockNode *left  = x ? &s->block[index-1] : &null_block;
    const BlockNode *top   = y ? &s->block[index-w] : &null_block;
    const BlockNode *right = trx<w ? &s->block[index+1] : &null_block;
    const BlockNode *bottom= try<h ? &s->block[index+w] : &null_block;
    const BlockNode *tl    = y && x ? &s->block[index-w-1] : left;
    const BlockNode *tr    = y && trx<w && ((x&1)==0 || level==0) ? &s->block[index-w+(1<<rem_depth)] : tl; //FIXME use lt
    int pl = left->color[0];
    int pcb= left->color[1];
    int pcr= left->color[2];
    int pmx, pmy;
    int mx=0, my=0;
    int l,cr,cb;
    const int stride= s->current_picture.linesize[0];
    const int uvstride= s->current_picture.linesize[1];
    uint8_t *current_data[3]= { s->input_picture.data[0] + (x + y*  stride)*block_w,
                                s->input_picture.data[1] + (x + y*uvstride)*block_w/2,
                                s->input_picture.data[2] + (x + y*uvstride)*block_w/2};
    int P[10][2];
    int16_t last_mv[3][2];
    int qpel= !!(s->avctx->flags & CODEC_FLAG_QPEL); //unused
    const int shift= 1+qpel;
    MotionEstContext *c= &s->m.me;
    int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref);
    int mx_context= av_log2(2*FFABS(left->mx - top->mx));
    int my_context= av_log2(2*FFABS(left->my - top->my));
    int s_context= 2*left->level + 2*top->level + tl->level + tr->level;
    int ref, best_ref, ref_score, ref_mx, ref_my;

    assert(sizeof(s->block_state) >= 256);
    if(s->keyframe){
        set_blocks(s, level, x, y, pl, pcb, pcr, 0, 0, 0, BLOCK_INTRA);
        return 0;
    }

//    clip predictors / edge ?

    P_LEFT[0]= left->mx;
    P_LEFT[1]= left->my;
    P_TOP [0]= top->mx;
    P_TOP [1]= top->my;
    P_TOPRIGHT[0]= tr->mx;
    P_TOPRIGHT[1]= tr->my;

    last_mv[0][0]= s->block[index].mx;
    last_mv[0][1]= s->block[index].my;
    last_mv[1][0]= right->mx;
    last_mv[1][1]= right->my;
    last_mv[2][0]= bottom->mx;
    last_mv[2][1]= bottom->my;

    s->m.mb_stride=2;
    s->m.mb_x=
    s->m.mb_y= 0;
    c->skip= 0;

    assert(c->  stride ==   stride);
    assert(c->uvstride == uvstride);

    c->penalty_factor    = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp);
    c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp);
    c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp);
    c->current_mv_penalty= c->mv_penalty[s->m.f_code=1] + MAX_MV;

    c->xmin = - x*block_w - 16+2;
    c->ymin = - y*block_w - 16+2;
    c->xmax = - (x+1)*block_w + (w<<(LOG2_MB_SIZE - s->block_max_depth)) + 16-2;
    c->ymax = - (y+1)*block_w + (h<<(LOG2_MB_SIZE - s->block_max_depth)) + 16-2;

    if(P_LEFT[0]     > (c->xmax<<shift)) P_LEFT[0]    = (c->xmax<<shift);
    if(P_LEFT[1]     > (c->ymax<<shift)) P_LEFT[1]    = (c->ymax<<shift);
    if(P_TOP[0]      > (c->xmax<<shift)) P_TOP[0]     = (c->xmax<<shift);
    if(P_TOP[1]      > (c->ymax<<shift)) P_TOP[1]     = (c->ymax<<shift);
    if(P_TOPRIGHT[0] < (c->xmin<<shift)) P_TOPRIGHT[0]= (c->xmin<<shift);
    if(P_TOPRIGHT[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<shift); //due to pmx no clip
    if(P_TOPRIGHT[1] > (c->ymax<<shift)) P_TOPRIGHT[1]= (c->ymax<<shift);

    P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
    P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);

    if (!y) {
        c->pred_x= P_LEFT[0];
        c->pred_y= P_LEFT[1];
    } else {
        c->pred_x = P_MEDIAN[0];
        c->pred_y = P_MEDIAN[1];
    }

    score= INT_MAX;
    best_ref= 0;
    for(ref=0; ref<s->ref_frames; ref++){
        init_ref(c, current_data, s->last_picture[ref].data, NULL, block_w*x, block_w*y, 0);

        ref_score= ff_epzs_motion_search(&s->m, &ref_mx, &ref_my, P, 0, /*ref_index*/ 0, last_mv,
                                         (1<<16)>>shift, level-LOG2_MB_SIZE+4, block_w);

        assert(ref_mx >= c->xmin);
        assert(ref_mx <= c->xmax);
        assert(ref_my >= c->ymin);
        assert(ref_my <= c->ymax);

        ref_score= c->sub_motion_search(&s->m, &ref_mx, &ref_my, ref_score, 0, 0, level-LOG2_MB_SIZE+4, block_w);
        ref_score= ff_get_mb_score(&s->m, ref_mx, ref_my, 0, 0, level-LOG2_MB_SIZE+4, block_w, 0);
        ref_score+= 2*av_log2(2*ref)*c->penalty_factor;
        if(s->ref_mvs[ref]){
            s->ref_mvs[ref][index][0]= ref_mx;
            s->ref_mvs[ref][index][1]= ref_my;
            s->ref_scores[ref][index]= ref_score;
        }
        if(score > ref_score){
            score= ref_score;
            best_ref= ref;
            mx= ref_mx;
            my= ref_my;
        }
    }
    //FIXME if mb_cmp != SSE then intra cant be compared currently and mb_penalty vs. lambda2

  //  subpel search
    base_bits= get_rac_count(&s->c) - 8*(s->c.bytestream - s->c.bytestream_start);
    pc= s->c;
    pc.bytestream_start=
    pc.bytestream= p_buffer; //FIXME end/start? and at the other stoo
    memcpy(p_state, s->block_state, sizeof(s->block_state));

    if(level!=s->block_max_depth)
        put_rac(&pc, &p_state[4 + s_context], 1);
    put_rac(&pc, &p_state[1 + left->type + top->type], 0);
    if(s->ref_frames > 1)
        put_symbol(&pc, &p_state[128 + 1024 + 32*ref_context], best_ref, 0);
    pred_mv(s, &pmx, &pmy, best_ref, left, top, tr);
    put_symbol(&pc, &p_state[128 + 32*(mx_context + 16*!!best_ref)], mx - pmx, 1);
    put_symbol(&pc, &p_state[128 + 32*(my_context + 16*!!best_ref)], my - pmy, 1);
    p_len= pc.bytestream - pc.bytestream_start;
    score += (s->lambda2*(get_rac_count(&pc)-base_bits))>>FF_LAMBDA_SHIFT;

    block_s= block_w*block_w;
    sum = pix_sum(current_data[0], stride, block_w);
    l= (sum + block_s/2)/block_s;
    iscore = pix_norm1(current_data[0], stride, block_w) - 2*l*sum + l*l*block_s;

    block_s= block_w*block_w>>2;
    sum = pix_sum(current_data[1], uvstride, block_w>>1);
    cb= (sum + block_s/2)/block_s;
    sum = pix_sum(current_data[2], uvstride, block_w>>1);
    cr= (sum + block_s/2)/block_s;

    ic= s->c;
    ic.bytestream_start=
    ic.bytestream= i_buffer; //FIXME end/start? and at the other stoo
    memcpy(i_state, s->block_state, sizeof(s->block_state));
    if(level!=s->block_max_depth)
        put_rac(&ic, &i_state[4 + s_context], 1);
    put_rac(&ic, &i_state[1 + left->type + top->type], 1);
    put_symbol(&ic, &i_state[32],  l-pl , 1);
    put_symbol(&ic, &i_state[64], cb-pcb, 1);
    put_symbol(&ic, &i_state[96], cr-pcr, 1);
    i_len= ic.bytestream - ic.bytestream_start;
    iscore += (s->lambda2*(get_rac_count(&ic)-base_bits))>>FF_LAMBDA_SHIFT;

    assert(iscore < 255*255*256 + s->lambda2*10);
    assert(iscore >= 0);
    assert(l>=0 && l<=255);
    assert(pl>=0 && pl<=255);

    if(level==0){
        int varc= iscore >> 8;
        int vard= score >> 8;
        if (vard <= 64 || vard < varc)
            c->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);
        else
            c->scene_change_score+= s->m.qscale;
    }

    if(level!=s->block_max_depth){
        put_rac(&s->c, &s->block_state[4 + s_context], 0);
        score2 = encode_q_branch(s, level+1, 2*x+0, 2*y+0);
        score2+= enco