cavs.c

linux下的MPEG1

    const int pic_width  = 16*h->mb_width;
    const int pic_height = 16*h->mb_height;

    if(!pic->data[0])
        return;
    if(mx&7) extra_width -= 3;
    if(my&7) extra_height -= 3;

    if(   full_mx < 0-extra_width
          || full_my < 0-extra_height
          || full_mx + 16/*FIXME*/ > pic_width + extra_width
          || full_my + 16/*FIXME*/ > pic_height + extra_height){
        ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->l_stride, h->l_stride,
                            16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
        src_y= s->edge_emu_buffer + 2 + 2*h->l_stride;
        emu=1;
    }

    qpix_op[luma_xy](dest_y, src_y, h->l_stride); //FIXME try variable height perhaps?
    if(!square){
        qpix_op[luma_xy](dest_y + delta, src_y + delta, h->l_stride);
    }

    if(emu){
        ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, h->c_stride,
                            9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
        src_cb= s->edge_emu_buffer;
    }
    chroma_op(dest_cb, src_cb, h->c_stride, chroma_height, mx&7, my&7);

    if(emu){
        ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, h->c_stride,
                            9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
        src_cr= s->edge_emu_buffer;
    }
    chroma_op(dest_cr, src_cr, h->c_stride, chroma_height, mx&7, my&7);
}

static inline void mc_part_std(AVSContext *h,int square,int chroma_height,int delta,
                        uint8_t *dest_y,uint8_t *dest_cb,uint8_t *dest_cr,
                        int x_offset, int y_offset,qpel_mc_func *qpix_put,
                        h264_chroma_mc_func chroma_put,qpel_mc_func *qpix_avg,
                        h264_chroma_mc_func chroma_avg, vector_t *mv){
    qpel_mc_func *qpix_op=  qpix_put;
    h264_chroma_mc_func chroma_op= chroma_put;

    dest_y  += 2*x_offset + 2*y_offset*h->l_stride;
    dest_cb +=   x_offset +   y_offset*h->c_stride;
    dest_cr +=   x_offset +   y_offset*h->c_stride;
    x_offset += 8*h->mbx;
    y_offset += 8*h->mby;

    if(mv->ref >= 0){
        Picture *ref= &h->DPB[mv->ref];
        mc_dir_part(h, ref, square, chroma_height, delta, 0,
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
                    qpix_op, chroma_op, mv);

        qpix_op=  qpix_avg;
        chroma_op= chroma_avg;
    }

    if((mv+MV_BWD_OFFS)->ref >= 0){
        Picture *ref= &h->DPB[0];
        mc_dir_part(h, ref, square, chroma_height, delta, 1,
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
                    qpix_op, chroma_op, mv+MV_BWD_OFFS);
    }
}

static void inter_pred(AVSContext *h, enum mb_t mb_type) {
    if(partition_flags[mb_type] == 0){ // 16x16
        mc_part_std(h, 1, 8, 0, h->cy, h->cu, h->cv, 0, 0,
                h->s.dsp.put_cavs_qpel_pixels_tab[0],
                h->s.dsp.put_h264_chroma_pixels_tab[0],
                h->s.dsp.avg_cavs_qpel_pixels_tab[0],
                h->s.dsp.avg_h264_chroma_pixels_tab[0],&h->mv[MV_FWD_X0]);
    }else{
        mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 0, 0,
                h->s.dsp.put_cavs_qpel_pixels_tab[1],
                h->s.dsp.put_h264_chroma_pixels_tab[1],
                h->s.dsp.avg_cavs_qpel_pixels_tab[1],
                h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X0]);
        mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 4, 0,
                h->s.dsp.put_cavs_qpel_pixels_tab[1],
                h->s.dsp.put_h264_chroma_pixels_tab[1],
                h->s.dsp.avg_cavs_qpel_pixels_tab[1],
                h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X1]);
        mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 0, 4,
                h->s.dsp.put_cavs_qpel_pixels_tab[1],
                h->s.dsp.put_h264_chroma_pixels_tab[1],
                h->s.dsp.avg_cavs_qpel_pixels_tab[1],
                h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X2]);
        mc_part_std(h, 1, 4, 0, h->cy, h->cu, h->cv, 4, 4,
                h->s.dsp.put_cavs_qpel_pixels_tab[1],
                h->s.dsp.put_h264_chroma_pixels_tab[1],
                h->s.dsp.avg_cavs_qpel_pixels_tab[1],
                h->s.dsp.avg_h264_chroma_pixels_tab[1],&h->mv[MV_FWD_X3]);
    }
    /* set intra prediction modes to default values */
    h->pred_mode_Y[3] =  h->pred_mode_Y[6] = INTRA_L_LP;
    h->top_pred_Y[h->mbx*2+0] = h->top_pred_Y[h->mbx*2+1] = INTRA_L_LP;
}

/*****************************************************************************
 *
 * motion vector prediction
 *
 ****************************************************************************/

static inline void set_mvs(vector_t *mv, enum block_t size) {
    switch(size) {
    case BLK_16X16:
        mv[MV_STRIDE  ] = mv[0];
        mv[MV_STRIDE+1] = mv[0];
    case BLK_16X8:
        mv[1] = mv[0];
        break;
    case BLK_8X16:
        mv[MV_STRIDE] = mv[0];
        break;
    }
}

static inline void store_mvs(AVSContext *h) {
    h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 0] = h->mv[MV_FWD_X0];
    h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 1] = h->mv[MV_FWD_X1];
    h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 2] = h->mv[MV_FWD_X2];
    h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + 3] = h->mv[MV_FWD_X3];
}

static inline void scale_mv(AVSContext *h, int *d_x, int *d_y, vector_t *src, int distp) {
    int den = h->scale_den[src->ref];

    *d_x = (src->x*distp*den + 256 + (src->x>>31)) >> 9;
    *d_y = (src->y*distp*den + 256 + (src->y>>31)) >> 9;
}

static inline void mv_pred_median(AVSContext *h, vector_t *mvP, vector_t *mvA, vector_t *mvB, vector_t *mvC) {
    int ax, ay, bx, by, cx, cy;
    int len_ab, len_bc, len_ca, len_mid;

    /* scale candidates according to their temporal span */
    scale_mv(h, &ax, &ay, mvA, mvP->dist);
    scale_mv(h, &bx, &by, mvB, mvP->dist);
    scale_mv(h, &cx, &cy, mvC, mvP->dist);
    /* find the geometrical median of the three candidates */
    len_ab = abs(ax - bx) + abs(ay - by);
    len_bc = abs(bx - cx) + abs(by - cy);
    len_ca = abs(cx - ax) + abs(cy - ay);
    len_mid = mid_pred(len_ab, len_bc, len_ca);
    if(len_mid == len_ab) {
        mvP->x = cx;
        mvP->y = cy;
    } else if(len_mid == len_bc) {
        mvP->x = ax;
        mvP->y = ay;
    } else {
        mvP->x = bx;
        mvP->y = by;
    }
}

static inline void mv_pred_direct(AVSContext *h, vector_t *pmv_fw,
                                  vector_t *col_mv) {
    vector_t *pmv_bw = pmv_fw + MV_BWD_OFFS;
    int den = h->direct_den[col_mv->ref];
    int m = col_mv->x >> 31;

    pmv_fw->dist = h->dist[1];
    pmv_bw->dist = h->dist[0];
    pmv_fw->ref = 1;
    pmv_bw->ref = 0;
    /* scale the co-located motion vector according to its temporal span */
    pmv_fw->x = (((den+(den*col_mv->x*pmv_fw->dist^m)-m-1)>>14)^m)-m;
    pmv_bw->x = m-(((den+(den*col_mv->x*pmv_bw->dist^m)-m-1)>>14)^m);
    m = col_mv->y >> 31;
    pmv_fw->y = (((den+(den*col_mv->y*pmv_fw->dist^m)-m-1)>>14)^m)-m;
    pmv_bw->y = m-(((den+(den*col_mv->y*pmv_bw->dist^m)-m-1)>>14)^m);
}

static inline void mv_pred_sym(AVSContext *h, vector_t *src, enum block_t size) {
    vector_t *dst = src + MV_BWD_OFFS;

    /* backward mv is the scaled and negated forward mv */
    dst->x = -((src->x * h->sym_factor + 256) >> 9);
    dst->y = -((src->y * h->sym_factor + 256) >> 9);
    dst->ref = 0;
    dst->dist = h->dist[0];
    set_mvs(dst, size);
}

static void mv_pred(AVSContext *h, enum mv_loc_t nP, enum mv_loc_t nC,
                    enum mv_pred_t mode, enum block_t size, int ref) {
    vector_t *mvP = &h->mv[nP];
    vector_t *mvA = &h->mv[nP-1];
    vector_t *mvB = &h->mv[nP-4];
    vector_t *mvC = &h->mv[nC];
    const vector_t *mvP2 = NULL;

    mvP->ref = ref;
    mvP->dist = h->dist[mvP->ref];
    if(mvC->ref == NOT_AVAIL)
        mvC = &h->mv[nP-5]; // set to top-left (mvD)
    if((mode == MV_PRED_PSKIP) &&
       ((mvA->ref == NOT_AVAIL) || (mvB->ref == NOT_AVAIL) ||
           ((mvA->x | mvA->y | mvA->ref) == 0)  ||
           ((mvB->x | mvB->y | mvB->ref) == 0) )) {
        mvP2 = &un_mv;
    /* if there is only one suitable candidate, take it */
    } else if((mvA->ref >= 0) && (mvB->ref < 0) && (mvC->ref < 0)) {
        mvP2= mvA;
    } else if((mvA->ref < 0) && (mvB->ref >= 0) && (mvC->ref < 0)) {
        mvP2= mvB;
    } else if((mvA->ref < 0) && (mvB->ref < 0) && (mvC->ref >= 0)) {
        mvP2= mvC;
    } else if(mode == MV_PRED_LEFT     && mvA->ref == ref){
        mvP2= mvA;
    } else if(mode == MV_PRED_TOP      && mvB->ref == ref){
        mvP2= mvB;
    } else if(mode == MV_PRED_TOPRIGHT && mvC->ref == ref){
        mvP2= mvC;
    }
    if(mvP2){
        mvP->x = mvP2->x;
        mvP->y = mvP2->y;
    }else
        mv_pred_median(h, mvP, mvA, mvB, mvC);

    if(mode < MV_PRED_PSKIP) {
        mvP->x += get_se_golomb(&h->s.gb);
        mvP->y += get_se_golomb(&h->s.gb);
    }
    set_mvs(mvP,size);
}

/*****************************************************************************
 *
 * residual data decoding
 *
 ****************************************************************************/

/** kth-order exponential golomb code */
static inline int get_ue_code(GetBitContext *gb, int order) {
    if(order) {
        int ret = get_ue_golomb(gb) << order;
        return ret + get_bits(gb,order);
    }
    return get_ue_golomb(gb);
}

/**
 * decode coefficients from one 8x8 block, dequantize, inverse transform
 *  and add them to sample block
 * @param r pointer to 2D VLC table
 * @param esc_golomb_order escape codes are k-golomb with this order k
 * @param qp quantizer
 * @param dst location of sample block
 * @param stride line stride in frame buffer
 */
static int decode_residual_block(AVSContext *h, GetBitContext *gb,
                                 const residual_vlc_t *r, int esc_golomb_order,
                                 int qp, uint8_t *dst, int stride) {
    int i,pos = -1;
    int level_code, esc_code, level, run, mask;
    int level_buf[64];
    int run_buf[64];
    int dqm = dequant_mul[qp];
    int dqs = dequant_shift[qp];
    int dqa = 1 << (dqs - 1);
    const uint8_t *scantab = h->scantable.permutated;
    DCTELEM *block = h->block;

    for(i=0;i<65;i++) {
        level_code = get_ue_code(gb,r->golomb_order);
        if(level_code >= ESCAPE_CODE) {
            run = ((level_code - ESCAPE_CODE) >> 1) + 1;
            esc_code = get_ue_code(gb,esc_golomb_order);
            level = esc_code + (run > r->max_run ? 1 : r->level_add[run]);
            while(level > r->inc_limit)
                r++;
            mask = -(level_code & 1);
            level = (level^mask) - mask;
        } else {
            level = r->rltab[level_code][0];
            if(!level) //end of block signal
                break;
            run   = r->rltab[level_code][1];
            r += r->rltab[level_code][2];
        }
        level_buf[i] = level;
        run_buf[i] = run;
    }
    /* inverse scan and dequantization */
    while(--i >= 0){
        pos += run_buf[i];
        if(pos > 63) {
            av_log(h->s.avctx, AV_LOG_ERROR,
                   "position out of block bounds at pic %d MB(%d,%d)\n",
                   h->picture.poc, h->mbx, h->mby);
            return -1;
        }
        block[scantab[pos]] = (level_buf[i]*dqm + dqa) >> dqs;
    }
    h->s.dsp.cavs_idct8_add(dst,block,stride);
    return 0;
}


static inline void decode_residual_chroma(AVSContext *h) {
    if(h->cbp & (1<<4))
        decode_residual_block(h,&h->s.gb,chroma_2dvlc,0, chroma_qp[h->qp],
                              h->cu,h->c_stride);
    if(h->cbp & (1<<5))
        decode_residual_block(h,&h->s.gb,chroma_2dvlc,0, chroma_qp[h->qp],
                              h->cv,h->c_stride);
}

static inline int decode_residual_inter(AVSContext *h) {
    int block;

    /* get coded block pattern */
    int cbp= get_ue_golomb(&h->s.gb);
    if(cbp > 63){
        av_log(h->s.avctx, AV_LOG_ERROR, "illegal inter cbp\n");
        return -1;
    }
    h->cbp = cbp_tab[cbp][1];

    /* get quantizer */
    if(h->cbp && !h->qp_fixed)
        h->qp = (h->qp + get_se_golomb(&h->s.gb)) & 63;
    for(block=0;block<4;block++)
        if(h->cbp & (1<<block))
            decode_residual_block(h,&h->s.gb,inter_2dvlc,0,h->qp,
                                  h->cy + h->luma_scan[block], h->l_stride);
    decode_residual_chroma(h);

    return 0;
}

/*****************************************************************************
 *
 * macroblock level
 *
 ****************************************************************************/

/**
 * initialise predictors for motion vectors and intra prediction
 */
static inline void init_mb(AVSContext *h) {
    int i;

    /* copy predictors from top line (MB B and C) into cache */
    for(i=0;i<3;i++) {
        h->mv[MV_FWD_B2+i] = h->top_mv[0][h->mbx*2+i];
        h->mv[MV_BWD_B2+i] = h->top_mv[1][h->mbx*2+i];
    }
    h->pred_mode_Y[1] = h->top_pred_Y[h->mbx*2+0];
    h->pred_mode_Y[2] = h->top_pred_Y[h->mbx*2+1];
    /* clear top predictors if MB B is not available */
    if(!(h->flags & B_AVAIL)) {
        h->mv[MV_FWD_B2] = un_mv;
        h->mv[MV_FWD_B3] = un_mv;
        h->mv[MV_BWD_B2] = un_mv;
        h->mv[MV_BWD_B3] = un_mv;
        h->pred_mode_Y[1] = h->pred_mode_Y[2] = NOT_AVAIL;
        h->flags &= ~(C_AVAIL|D_AVAIL);
    } else if(h->mbx) {
        h->flags |= D_AVAIL;
    }
    if(h->mbx == h->mb_width-1) //MB C not available
        h->flags &= ~C_AVAIL;
    /* clear top-right predictors if MB C is not available */
    if(!(h->flags & C_AVAIL)) {
        h->mv[MV_FWD_C2] = un_mv;
        h->mv[MV_BWD_C2] = un_mv;
    }
    /* clear top-left predictors if MB D is not available */
    if(!(h->flags & D_AVAIL)) {
        h->mv[MV_FWD_D3] = un_mv;
        h->mv[MV_BWD_D3] = un_mv;
    }
    /* set pointer for co-located macroblock type */
    h->col_type = &h->col_type_base[h->mby*h->mb_width + h->mbx];
}