annotate cavs.c @ 3995:b00c06477dff libavcodec

write cabac low and range variables as early as possible to prevent stalls from reading them before they where written, the P4 is said to disslike that alot, on P3 its 2% faster (START/STOP_TIMER over decode_residual)

/*
 * Chinese AVS video (AVS1-P2, JiZhun profile) decoder.
 * Copyright (c) 2006  Stefan Gehrer <[e-mail]>
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * FFmpeg is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 */

/**
 * @file cavs.c
 * Chinese AVS video (AVS1-P2, JiZhun profile) decoder
 * @author Stefan Gehrer <[e-mail]>
 */

#include "avcodec.h"
#include "bitstream.h"
#include "golomb.h"
#include "mpegvideo.h"
#include "cavsdata.h"

typedef struct {
    MpegEncContext s;
    Picture picture; ///< currently decoded frame
    Picture DPB[2];  ///< reference frames
    int dist[2];     ///< temporal distances from current frame to ref frames
    int profile, level;
    int aspect_ratio;
    int mb_width, mb_height;
    int pic_type;
    int progressive;
    int pic_structure;
    int skip_mode_flag; ///< select between skip_count or one skip_flag per MB
    int loop_filter_disable;
    int alpha_offset, beta_offset;
    int ref_flag;
    int mbx, mby;      ///< macroblock coordinates
    int flags;         ///< availability flags of neighbouring macroblocks
    int stc;           ///< last start code
    uint8_t *cy, *cu, *cv; ///< current MB sample pointers
    int left_qp;
    uint8_t *top_qp;

    /** mv motion vector cache
       0:    D3  B2  B3  C2
       4:    A1  X0  X1   -
       8:    A3  X2  X3   -

       X are the vectors in the current macroblock (5,6,9,10)
       A is the macroblock to the left (4,8)
       B is the macroblock to the top (1,2)
       C is the macroblock to the top-right (3)
       D is the macroblock to the top-left (0)

       the same is repeated for backward motion vectors */
    vector_t mv[2*4*3];
    vector_t *top_mv[2];
    vector_t *col_mv;

    /** luma pred mode cache
       0:    --  B2  B3
       3:    A1  X0  X1
       6:    A3  X2  X3   */
    int pred_mode_Y[3*3];
    int *top_pred_Y;
    int l_stride, c_stride;
    int luma_scan[4];
    int qp;
    int qp_fixed;
    int cbp;
    ScanTable scantable;

    /** intra prediction is done with un-deblocked samples
     they are saved here before deblocking the MB  */
    uint8_t *top_border_y, *top_border_u, *top_border_v;
    uint8_t left_border_y[26], left_border_u[10], left_border_v[10];
    uint8_t intern_border_y[26];
    uint8_t topleft_border_y, topleft_border_u, topleft_border_v;

    void (*intra_pred_l[8])(uint8_t *d,uint8_t *top,uint8_t *left,int stride);
    void (*intra_pred_c[7])(uint8_t *d,uint8_t *top,uint8_t *left,int stride);
    uint8_t *col_type_base;
    uint8_t *col_type;

    /* scaling factors for MV prediction */
    int sym_factor;    ///< for scaling in symmetrical B block
    int direct_den[2]; ///< for scaling in direct B block
    int scale_den[2];  ///< for scaling neighbouring MVs

    int got_keyframe;
    DCTELEM *block;
} AVSContext;

/*****************************************************************************
 *
 * in-loop deblocking filter
 *
 ****************************************************************************/

static inline int get_bs(vector_t *mvP, vector_t *mvQ, int b) {
    if((mvP->ref == REF_INTRA) || (mvQ->ref == REF_INTRA))
        return 2;
    if( (abs(mvP->x - mvQ->x) >= 4) ||  (abs(mvP->y - mvQ->y) >= 4) )
        return 1;
    if(b){
        mvP += MV_BWD_OFFS;
        mvQ += MV_BWD_OFFS;
        if( (abs(mvP->x - mvQ->x) >= 4) ||  (abs(mvP->y - mvQ->y) >= 4) )
            return 1;
    }else{
        if(mvP->ref != mvQ->ref)
            return 1;
    }
    return 0;
}

#define SET_PARAMS                                            \
    alpha = alpha_tab[clip(qp_avg + h->alpha_offset,0,63)];   \
    beta  =  beta_tab[clip(qp_avg + h->beta_offset, 0,63)];   \
    tc    =    tc_tab[clip(qp_avg + h->alpha_offset,0,63)];

/**
 * in-loop deblocking filter for a single macroblock
 *
 * boundary strength (bs) mapping:
 *
 * --4---5--
 * 0   2   |
 * | 6 | 7 |
 * 1   3   |
 * ---------
 *
 */
static void filter_mb(AVSContext *h, enum mb_t mb_type) {
    DECLARE_ALIGNED_8(uint8_t, bs[8]);
    int qp_avg, alpha, beta, tc;
    int i;

    /* save un-deblocked lines */
    h->topleft_border_y = h->top_border_y[h->mbx*16+15];
    h->topleft_border_u = h->top_border_u[h->mbx*10+8];
    h->topleft_border_v = h->top_border_v[h->mbx*10+8];
    memcpy(&h->top_border_y[h->mbx*16], h->cy + 15* h->l_stride,16);
    memcpy(&h->top_border_u[h->mbx*10+1], h->cu +  7* h->c_stride,8);
    memcpy(&h->top_border_v[h->mbx*10+1], h->cv +  7* h->c_stride,8);
    for(i=0;i<8;i++) {
        h->left_border_y[i*2+1] = *(h->cy + 15 + (i*2+0)*h->l_stride);
        h->left_border_y[i*2+2] = *(h->cy + 15 + (i*2+1)*h->l_stride);
        h->left_border_u[i+1] = *(h->cu + 7 + i*h->c_stride);
        h->left_border_v[i+1] = *(h->cv + 7 + i*h->c_stride);
    }
    if(!h->loop_filter_disable) {
        /* determine bs */
        if(mb_type == I_8X8)
            *((uint64_t *)bs) = 0x0202020202020202ULL;
        else{
            *((uint64_t *)bs) = 0;
            if(partition_flags[mb_type] & SPLITV){
                bs[2] = get_bs(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X1], mb_type > P_8X8);
                bs[3] = get_bs(&h->mv[MV_FWD_X2], &h->mv[MV_FWD_X3], mb_type > P_8X8);
            }
            if(partition_flags[mb_type] & SPLITH){
                bs[6] = get_bs(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X2], mb_type > P_8X8);
                bs[7] = get_bs(&h->mv[MV_FWD_X1], &h->mv[MV_FWD_X3], mb_type > P_8X8);
            }
            bs[0] = get_bs(&h->mv[MV_FWD_A1], &h->mv[MV_FWD_X0], mb_type > P_8X8);
            bs[1] = get_bs(&h->mv[MV_FWD_A3], &h->mv[MV_FWD_X2], mb_type > P_8X8);
            bs[4] = get_bs(&h->mv[MV_FWD_B2], &h->mv[MV_FWD_X0], mb_type > P_8X8);
            bs[5] = get_bs(&h->mv[MV_FWD_B3], &h->mv[MV_FWD_X1], mb_type > P_8X8);
        }
        if( *((uint64_t *)bs) ) {
            if(h->flags & A_AVAIL) {
                qp_avg = (h->qp + h->left_qp + 1) >> 1;
                SET_PARAMS;
                h->s.dsp.cavs_filter_lv(h->cy,h->l_stride,alpha,beta,tc,bs[0],bs[1]);
                h->s.dsp.cavs_filter_cv(h->cu,h->c_stride,alpha,beta,tc,bs[0],bs[1]);
                h->s.dsp.cavs_filter_cv(h->cv,h->c_stride,alpha,beta,tc,bs[0],bs[1]);
            }
            qp_avg = h->qp;
            SET_PARAMS;
            h->s.dsp.cavs_filter_lv(h->cy + 8,h->l_stride,alpha,beta,tc,bs[2],bs[3]);
            h->s.dsp.cavs_filter_lh(h->cy + 8*h->l_stride,h->l_stride,alpha,beta,tc,
                           bs[6],bs[7]);

            if(h->flags & B_AVAIL) {
                qp_avg = (h->qp + h->top_qp[h->mbx] + 1) >> 1;
                SET_PARAMS;
                h->s.dsp.cavs_filter_lh(h->cy,h->l_stride,alpha,beta,tc,bs[4],bs[5]);
                h->s.dsp.cavs_filter_ch(h->cu,h->c_stride,alpha,beta,tc,bs[4],bs[5]);
                h->s.dsp.cavs_filter_ch(h->cv,h->c_stride,alpha,beta,tc,bs[4],bs[5]);
            }
        }
    }
    h->left_qp = h->qp;
    h->top_qp[h->mbx] = h->qp;
}

#undef SET_PARAMS

/*****************************************************************************
 *
 * spatial intra prediction
 *
 ****************************************************************************/

static inline void load_intra_pred_luma(AVSContext *h, uint8_t *top,
                                        uint8_t **left, int block) {
    int i;

    switch(block) {
    case 0:
        *left = h->left_border_y;
        h->left_border_y[0] = h->left_border_y[1];
        memset(&h->left_border_y[17],h->left_border_y[16],9);
        memcpy(&top[1],&h->top_border_y[h->mbx*16],16);
        top[17] = top[16];
        top[0] = top[1];
        if((h->flags & A_AVAIL) && (h->flags & B_AVAIL))
            h->left_border_y[0] = top[0] = h->topleft_border_y;
        break;
    case 1:
        *left = h->intern_border_y;
        for(i=0;i<8;i++)
            h->intern_border_y[i+1] = *(h->cy + 7 + i*h->l_stride);
        memset(&h->intern_border_y[9],h->intern_border_y[8],9);
        h->intern_border_y[0] = h->intern_border_y[1];
        memcpy(&top[1],&h->top_border_y[h->mbx*16+8],8);
        if(h->flags & C_AVAIL)
            memcpy(&top[9],&h->top_border_y[(h->mbx + 1)*16],8);
        else
            memset(&top[9],top[8],9);
        top[17] = top[16];
        top[0] = top[1];
        if(h->flags & B_AVAIL)
            h->intern_border_y[0] = top[0] = h->top_border_y[h->mbx*16+7];
        break;
    case 2:
        *left = &h->left_border_y[8];
        memcpy(&top[1],h->cy + 7*h->l_stride,16);
        top[17] = top[16];
        top[0] = top[1];
        if(h->flags & A_AVAIL)
            top[0] = h->left_border_y[8];
        break;
    case 3:
        *left = &h->intern_border_y[8];
        for(i=0;i<8;i++)
            h->intern_border_y[i+9] = *(h->cy + 7 + (i+8)*h->l_stride);
        memset(&h->intern_border_y[17],h->intern_border_y[16],9);
        memcpy(&top[0],h->cy + 7 + 7*h->l_stride,9);
        memset(&top[9],top[8],9);
        break;
    }
}

static void intra_pred_vert(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
    int y;
    uint64_t a = unaligned64(&top[1]);
    for(y=0;y<8;y++) {
        *((uint64_t *)(d+y*stride)) = a;
    }
}

static void intra_pred_horiz(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
    int y;
    uint64_t a;
    for(y=0;y<8;y++) {
        a = left[y+1] * 0x0101010101010101ULL;
        *((uint64_t *)(d+y*stride)) = a;
    }
}

static void intra_pred_dc_128(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
    int y;
    uint64_t a = 0x8080808080808080ULL;
    for(y=0;y<8;y++)
        *((uint64_t *)(d+y*stride)) = a;
}

static void intra_pred_plane(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
    int x,y,ia;
    int ih = 0;
    int iv = 0;
    uint8_t *cm = cropTbl + MAX_NEG_CROP;

    for(x=0; x<4; x++) {
        ih += (x+1)*(top[5+x]-top[3-x]);
        iv += (x+1)*(left[5+x]-left[3-x]);
    }
    ia = (top[8]+left[8])<<4;
    ih = (17*ih+16)>>5;
    iv = (17*iv+16)>>5;
    for(y=0; y<8; y++)
        for(x=0; x<8; x++)
            d[y*stride+x] = cm[(ia+(x-3)*ih+(y-3)*iv+16)>>5];
}

#define LOWPASS(ARRAY,INDEX)                                            \
    (( ARRAY[(INDEX)-1] + 2*ARRAY[(INDEX)] + ARRAY[(INDEX)+1] + 2) >> 2)

static void intra_pred_lp(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
    int x,y;
    for(y=0; y<8; y++)
        for(x=0; x<8; x++)
            d[y*stride+x] = (LOWPASS(top,x+1) + LOWPASS(left,y+1)) >> 1;
}

static void intra_pred_down_left(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
    int x,y;
    for(y=0; y<8; y++)
        for(x=0; x<8; x++)
            d[y*stride+x] = (LOWPASS(top,x+y+2) + LOWPASS(left,x+y+2)) >> 1;
}

static void intra_pred_down_right(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
    int x,y;
    for(y=0; y<8; y++)
        for(x=0; x<8; x++)
            if(x==y)
                d[y*stride+x] = (left[1]+2*top[0]+top[1]+2)>>2;
            else if(x>y)
                d[y*stride+x] = LOWPASS(top,x-y);
            else
                d[y*stride+x] = LOWPASS(left,y-x);
}

static void intra_pred_lp_left(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
    int x,y;
    for(y=0; y<8; y++)
        for(x=0; x<8; x++)
            d[y*stride+x] = LOWPASS(left,y+1);
}

static void intra_pred_lp_top(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
    int x,y;
    for(y=0; y<8; y++)
        for(x=0; x<8; x++)
            d[y*stride+x] = LOWPASS(top,x+1);
}

#undef LOWPASS

static inline void modify_pred(const int_fast8_t *mod_table, int *mode) {
    *mode = mod_table[*mode];
    if(*mode < 0) {
        av_log(NULL, AV_LOG_ERROR, "Illegal intra prediction mode\n");
        *mode = 0;
    }
}

/*****************************************************************************
 *
 * motion compensation
 *
 ****************************************************************************/

static inline void mc_dir_part(AVSContext *h,Picture *pic,int square,
                        int chroma_height,int delta,int list,uint8_t *dest_y,
                        uint8_t *dest_cb,uint8_t *dest_cr,int src_x_offset,
                        int src_y_offset,qpel_mc_func *qpix_op,
                        h264_chroma_mc_func chroma_op,vector_t *mv){
    MpegEncContext * const s = &h->s;
    const int mx= mv->x + src_x_offset*8;
    const int my= mv->y + src_y_offset*8;
    const int luma_xy= (mx&3) + ((my&3)<<2);
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->l_stride;
    uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->c_stride;
    uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->c_stride;
    int extra_width= 0; //(s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16;
    int extra_height= extra_width;
    int emu=0;
    const int full_mx= mx>>2;
    const int full_my= my>>2;
    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];
}

static inline void check_for_slice(AVSContext *h);

/**
 * save predictors for later macroblocks and increase
 * macroblock address
 * @returns 0 if end of frame is reached, 1 otherwise
 */
static inline int next_mb(AVSContext *h) {
    int i;

    h->flags |= A_AVAIL;
    h->cy += 16;
    h->cu += 8;
    h->cv += 8;
    /* copy mvs as predictors to the left */
    for(i=0;i<=20;i+=4)
        h->mv[i] = h->mv[i+2];
    /* copy bottom mvs from cache to top line */
    h->top_mv[0][h->mbx*2+0] = h->mv[MV_FWD_X2];
    h->top_mv[0][h->mbx*2+1] = h->mv[MV_FWD_X3];
    h->top_mv[1][h->mbx*2+0] = h->mv[MV_BWD_X2];
    h->top_mv[1][h->mbx*2+1] = h->mv[MV_BWD_X3];
    /* next MB address */
    h->mbx++;
    if(h->mbx == h->mb_width) { //new mb line
        h->flags = B_AVAIL|C_AVAIL;
        /* clear left pred_modes */
        h->pred_mode_Y[3] = h->pred_mode_Y[6] = NOT_AVAIL;
        /* clear left mv predictors */
        for(i=0;i<=20;i+=4)
            h->mv[i] = un_mv;
        h->mbx = 0;
        h->mby++;
        /* re-calculate sample pointers */
        h->cy = h->picture.data[0] + h->mby*16*h->l_stride;
        h->cu = h->picture.data[1] + h->mby*8*h->c_stride;
        h->cv = h->picture.data[2] + h->mby*8*h->c_stride;
        if(h->mby == h->mb_height) { //frame end
            return 0;
        } else {
            //check_for_slice(h);
        }
    }
    return 1;
}

static int decode_mb_i(AVSContext *h, int cbp_code) {
    GetBitContext *gb = &h->s.gb;
    int block, pred_mode_uv;
    uint8_t top[18];
    uint8_t *left = NULL;
    uint8_t *d;

    init_mb(h);

    /* get intra prediction modes from stream */
    for(block=0;block<4;block++) {
        int nA,nB,predpred;
        int pos = scan3x3[block];

        nA = h->pred_mode_Y[pos-1];
        nB = h->pred_mode_Y[pos-3];
        predpred = FFMIN(nA,nB);
        if(predpred == NOT_AVAIL) // if either is not available
            predpred = INTRA_L_LP;
        if(!get_bits1(gb)){
            int rem_mode= get_bits(gb, 2);
            predpred = rem_mode + (rem_mode >= predpred);
        }
        h->pred_mode_Y[pos] = predpred;
    }
    pred_mode_uv = get_ue_golomb(gb);
    if(pred_mode_uv > 6) {
        av_log(h->s.avctx, AV_LOG_ERROR, "illegal intra chroma pred mode\n");
        return -1;
    }

    /* save pred modes before they get modified */
    h->pred_mode_Y[3] =  h->pred_mode_Y[5];
    h->pred_mode_Y[6] =  h->pred_mode_Y[8];
    h->top_pred_Y[h->mbx*2+0] = h->pred_mode_Y[7];
    h->top_pred_Y[h->mbx*2+1] = h->pred_mode_Y[8];

    /* modify pred modes according to availability of neighbour samples */
    if(!(h->flags & A_AVAIL)) {
        modify_pred(left_modifier_l, &h->pred_mode_Y[4] );
        modify_pred(left_modifier_l, &h->pred_mode_Y[7] );
        modify_pred(left_modifier_c, &pred_mode_uv );
    }
    if(!(h->flags & B_AVAIL)) {
        modify_pred(top_modifier_l, &h->pred_mode_Y[4] );
        modify_pred(top_modifier_l, &h->pred_mode_Y[5] );
        modify_pred(top_modifier_c, &pred_mode_uv );
    }

    /* get coded block pattern */
    if(h->pic_type == FF_I_TYPE)
        cbp_code = get_ue_golomb(gb);
    if(cbp_code > 63){
        av_log(h->s.avctx, AV_LOG_ERROR, "illegal intra cbp\n");
        return -1;
    }
    h->cbp = cbp_tab[cbp_code][0];
    if(h->cbp && !h->qp_fixed)
        h->qp = (h->qp + get_se_golomb(gb)) & 63; //qp_delta

    /* luma intra prediction interleaved with residual decode/transform/add */
    for(block=0;block<4;block++) {
        d = h->cy + h->luma_scan[block];
        load_intra_pred_luma(h, top, &left, block);
        h->intra_pred_l[h->pred_mode_Y[scan3x3[block]]]
            (d, top, left, h->l_stride);
        if(h->cbp & (1<<block))
            decode_residual_block(h,gb,intra_2dvlc,1,h->qp,d,h->l_stride);
    }

    /* chroma intra prediction */
    /* extend borders by one pixel */
    h->left_border_u[9] = h->left_border_u[8];
    h->left_border_v[9] = h->left_border_v[8];
    h->top_border_u[h->mbx*10+9] = h->top_border_u[h->mbx*10+8];
    h->top_border_v[h->mbx*10+9] = h->top_border_v[h->mbx*10+8];
    if(h->mbx && h->mby) {
        h->top_border_u[h->mbx*10] = h->left_border_u[0] = h->topleft_border_u;
        h->top_border_v[h->mbx*10] = h->left_border_v[0] = h->topleft_border_v;
    } else {
        h->left_border_u[0] = h->left_border_u[1];
        h->left_border_v[0] = h->left_border_v[1];
        h->top_border_u[h->mbx*10] = h->top_border_u[h->mbx*10+1];
        h->top_border_v[h->mbx*10] = h->top_border_v[h->mbx*10+1];
    }
    h->intra_pred_c[pred_mode_uv](h->cu, &h->top_border_u[h->mbx*10],
                                  h->left_border_u, h->c_stride);
    h->intra_pred_c[pred_mode_uv](h->cv, &h->top_border_v[h->mbx*10],
                                  h->left_border_v, h->c_stride);

    decode_residual_chroma(h);
    filter_mb(h,I_8X8);

    /* mark motion vectors as intra */
    h->mv[MV_FWD_X0] = intra_mv;
    set_mvs(&h->mv[MV_FWD_X0], BLK_16X16);
    h->mv[MV_BWD_X0] = intra_mv;
    set_mvs(&h->mv[MV_BWD_X0], BLK_16X16);
    if(h->pic_type != FF_B_TYPE)
        *h->col_type = I_8X8;

    return 0;
}

static void decode_mb_p(AVSContext *h, enum mb_t mb_type) {
    GetBitContext *gb = &h->s.gb;
    int ref[4];

    init_mb(h);
    switch(mb_type) {
    case P_SKIP:
        mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_PSKIP, BLK_16X16, 0);
        break;
    case P_16X16:
        ref[0] = h->ref_flag ? 0 : get_bits1(gb);
        mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_MEDIAN,   BLK_16X16,ref[0]);
        break;
    case P_16X8:
        ref[0] = h->ref_flag ? 0 : get_bits1(gb);
        ref[2] = h->ref_flag ? 0 : get_bits1(gb);
        mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_TOP,      BLK_16X8, ref[0]);
        mv_pred(h, MV_FWD_X2, MV_FWD_A1, MV_PRED_LEFT,     BLK_16X8, ref[2]);
        break;
    case P_8X16:
        ref[0] = h->ref_flag ? 0 : get_bits1(gb);
        ref[1] = h->ref_flag ? 0 : get_bits1(gb);
        mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_LEFT,     BLK_8X16, ref[0]);
        mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_TOPRIGHT, BLK_8X16, ref[1]);
        break;
    case P_8X8:
        ref[0] = h->ref_flag ? 0 : get_bits1(gb);
        ref[1] = h->ref_flag ? 0 : get_bits1(gb);
        ref[2] = h->ref_flag ? 0 : get_bits1(gb);
        ref[3] = h->ref_flag ? 0 : get_bits1(gb);
        mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_MEDIAN,   BLK_8X8, ref[0]);
        mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_MEDIAN,   BLK_8X8, ref[1]);
        mv_pred(h, MV_FWD_X2, MV_FWD_X1, MV_PRED_MEDIAN,   BLK_8X8, ref[2]);
        mv_pred(h, MV_FWD_X3, MV_FWD_X0, MV_PRED_MEDIAN,   BLK_8X8, ref[3]);
    }
    inter_pred(h, mb_type);
    store_mvs(h);
    if(mb_type != P_SKIP)
        decode_residual_inter(h);
    filter_mb(h,mb_type);
    *h->col_type = mb_type;
}

static void decode_mb_b(AVSContext *h, enum mb_t mb_type) {
    int block;
    enum sub_mb_t sub_type[4];
    int flags;

    init_mb(h);

    /* reset all MVs */
    h->mv[MV_FWD_X0] = dir_mv;
    set_mvs(&h->mv[MV_FWD_X0], BLK_16X16);
    h->mv[MV_BWD_X0] = dir_mv;
    set_mvs(&h->mv[MV_BWD_X0], BLK_16X16);
    switch(mb_type) {
    case B_SKIP:
    case B_DIRECT:
        if(!(*h->col_type)) {
            /* intra MB at co-location, do in-plane prediction */
            mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_BSKIP, BLK_16X16, 1);
            mv_pred(h, MV_BWD_X0, MV_BWD_C2, MV_PRED_BSKIP, BLK_16X16, 0);
        } else
            /* direct prediction from co-located P MB, block-wise */
            for(block=0;block<4;block++)
                mv_pred_direct(h,&h->mv[mv_scan[block]],
                            &h->col_mv[(h->mby*h->mb_width+h->mbx)*4 + block]);
        break;
    case B_FWD_16X16:
        mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_MEDIAN, BLK_16X16, 1);
        break;
    case B_SYM_16X16:
        mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_MEDIAN, BLK_16X16, 1);
        mv_pred_sym(h, &h->mv[MV_FWD_X0], BLK_16X16);
        break;
    case B_BWD_16X16:
        mv_pred(h, MV_BWD_X0, MV_BWD_C2, MV_PRED_MEDIAN, BLK_16X16, 0);
        break;
    case B_8X8:
        for(block=0;block<4;block++)
            sub_type[block] = get_bits(&h->s.gb,2);
        for(block=0;block<4;block++) {
            switch(sub_type[block]) {
            case B_SUB_DIRECT:
                if(!(*h->col_type)) {
                    /* intra MB at co-location, do in-plane prediction */
                    mv_pred(h, mv_scan[block], mv_scan[block]-3,
                            MV_PRED_BSKIP, BLK_8X8, 1);
                    mv_pred(h, mv_scan[block]+MV_BWD_OFFS,
                            mv_scan[block]-3+MV_BWD_OFFS,
                            MV_PRED_BSKIP, BLK_8X8, 0);
                } else
                    mv_pred_direct(h,&h->mv[mv_scan[block]],
                                   &h->col_mv[(h->mby*h->mb_width + h->mbx)*4 + block]);
                break;
            case B_SUB_FWD:
                mv_pred(h, mv_scan[block], mv_scan[block]-3,
                        MV_PRED_MEDIAN, BLK_8X8, 1);
                break;
            case B_SUB_SYM:
                mv_pred(h, mv_scan[block], mv_scan[block]-3,
                        MV_PRED_MEDIAN, BLK_8X8, 1);
                mv_pred_sym(h, &h->mv[mv_scan[block]], BLK_8X8);
                break;
            }
        }
        for(block=0;block<4;block++) {
            if(sub_type[block] == B_SUB_BWD)
                mv_pred(h, mv_scan[block]+MV_BWD_OFFS,
                        mv_scan[block]+MV_BWD_OFFS-3,
                        MV_PRED_MEDIAN, BLK_8X8, 0);
        }
        break;
    default:
        assert((mb_type > B_SYM_16X16) && (mb_type < B_8X8));
        flags = partition_flags[mb_type];
        if(mb_type & 1) { /* 16x8 macroblock types */
            if(flags & FWD0)
                mv_pred(h, MV_FWD_X0, MV_FWD_C2, MV_PRED_TOP,  BLK_16X8, 1);
            if(flags & SYM0)
                mv_pred_sym(h, &h->mv[MV_FWD_X0], BLK_16X8);
            if(flags & FWD1)
                mv_pred(h, MV_FWD_X2, MV_FWD_A1, MV_PRED_LEFT, BLK_16X8, 1);
            if(flags & SYM1)
                mv_pred_sym(h, &h->mv[MV_FWD_X2], BLK_16X8);
            if(flags & BWD0)
                mv_pred(h, MV_BWD_X0, MV_BWD_C2, MV_PRED_TOP,  BLK_16X8, 0);
            if(flags & BWD1)
                mv_pred(h, MV_BWD_X2, MV_BWD_A1, MV_PRED_LEFT, BLK_16X8, 0);
        } else {          /* 8x16 macroblock types */
            if(flags & FWD0)
                mv_pred(h, MV_FWD_X0, MV_FWD_B3, MV_PRED_LEFT, BLK_8X16, 1);
            if(flags & SYM0)
                mv_pred_sym(h, &h->mv[MV_FWD_X0], BLK_8X16);
            if(flags & FWD1)
                mv_pred(h, MV_FWD_X1, MV_FWD_C2, MV_PRED_TOPRIGHT,BLK_8X16, 1);
            if(flags & SYM1)
                mv_pred_sym(h, &h->mv[MV_FWD_X1], BLK_8X16);
            if(flags & BWD0)
                mv_pred(h, MV_BWD_X0, MV_BWD_B3, MV_PRED_LEFT, BLK_8X16, 0);
            if(flags & BWD1)
                mv_pred(h, MV_BWD_X1, MV_BWD_C2, MV_PRED_TOPRIGHT,BLK_8X16, 0);
        }
    }
    inter_pred(h, mb_type);
    if(mb_type != B_SKIP)
        decode_residual_inter(h);
    filter_mb(h,mb_type);
}

/*****************************************************************************
 *
 * slice level
 *
 ****************************************************************************/

static inline int decode_slice_header(AVSContext *h, GetBitContext *gb) {
    if(h->stc > 0xAF)
        av_log(h->s.avctx, AV_LOG_ERROR, "unexpected start code 0x%02x\n", h->stc);
    h->mby = h->stc;
    if((h->mby == 0) && (!h->qp_fixed)){
        h->qp_fixed = get_bits1(gb);
        h->qp = get_bits(gb,6);
    }
    /* inter frame or second slice can have weighting params */
    if((h->pic_type != FF_I_TYPE) || (!h->pic_structure && h->mby >= h->mb_width/2))
        if(get_bits1(gb)) { //slice_weighting_flag
            av_log(h->s.avctx, AV_LOG_ERROR,
                   "weighted prediction not yet supported\n");
        }
    return 0;
}

static inline void check_for_slice(AVSContext *h) {
    GetBitContext *gb = &h->s.gb;
    int align;
    align = (-get_bits_count(gb)) & 7;
    if((show_bits_long(gb,24+align) & 0xFFFFFF) == 0x000001) {
        get_bits_long(gb,24+align);
        h->stc = get_bits(gb,8);
        decode_slice_header(h,gb);
    }
}

/*****************************************************************************
 *
 * frame level
 *
 ****************************************************************************/

static void init_pic(AVSContext *h) {
    int i;

    /* clear some predictors */
    for(i=0;i<=20;i+=4)
        h->mv[i] = un_mv;
    h->mv[MV_BWD_X0] = dir_mv;
    set_mvs(&h->mv[MV_BWD_X0], BLK_16X16);
    h->mv[MV_FWD_X0] = dir_mv;
    set_mvs(&h->mv[MV_FWD_X0], BLK_16X16);
    h->pred_mode_Y[3] = h->pred_mode_Y[6] = NOT_AVAIL;
    h->cy = h->picture.data[0];
    h->cu = h->picture.data[1];
    h->cv = h->picture.data[2];
    h->l_stride = h->picture.linesize[0];
    h->c_stride = h->picture.linesize[1];
    h->luma_scan[2] = 8*h->l_stride;
    h->luma_scan[3] = 8*h->l_stride+8;
    h->mbx = h->mby = 0;
    h->flags = 0;
}

static int decode_pic(AVSContext *h) {
    MpegEncContext *s = &h->s;
    int skip_count;
    enum mb_t mb_type;

    if (!s->context_initialized) {
        s->avctx->idct_algo = FF_IDCT_CAVS;
        if (MPV_common_init(s) < 0)
            return -1;
        ff_init_scantable(s->dsp.idct_permutation,&h->scantable,ff_zigzag_direct);
    }
    get_bits(&s->gb,16);//bbv_dwlay
    if(h->stc == PIC_PB_START_CODE) {
        h->pic_type = get_bits(&s->gb,2) + FF_I_TYPE;
        if(h->pic_type > FF_B_TYPE) {
            av_log(s->avctx, AV_LOG_ERROR, "illegal picture type\n");
            return -1;
        }
        /* make sure we have the reference frames we need */
        if(!h->DPB[0].data[0] ||
          (!h->DPB[1].data[0] && h->pic_type == FF_B_TYPE))
            return -1;
    } else {
        h->pic_type = FF_I_TYPE;
        if(get_bits1(&s->gb))
            get_bits(&s->gb,16);//time_code
    }
    /* release last B frame */
    if(h->picture.data[0])
        s->avctx->release_buffer(s->avctx, (AVFrame *)&h->picture);

    s->avctx->get_buffer(s->avctx, (AVFrame *)&h->picture);
    init_pic(h);
    h->picture.poc = get_bits(&s->gb,8)*2;

    /* get temporal distances and MV scaling factors */
    if(h->pic_type != FF_B_TYPE) {
        h->dist[0] = (h->picture.poc - h->DPB[0].poc  + 512) % 512;
    } else {
        h->dist[0] = (h->DPB[0].poc  - h->picture.poc + 512) % 512;
    }
    h->dist[1] = (h->picture.poc - h->DPB[1].poc  + 512) % 512;
    h->scale_den[0] = h->dist[0] ? 512/h->dist[0] : 0;
    h->scale_den[1] = h->dist[1] ? 512/h->dist[1] : 0;
    if(h->pic_type == FF_B_TYPE) {
        h->sym_factor = h->dist[0]*h->scale_den[1];
    } else {
        h->direct_den[0] = h->dist[0] ? 16384/h->dist[0] : 0;
        h->direct_den[1] = h->dist[1] ? 16384/h->dist[1] : 0;
    }

    if(s->low_delay)
        get_ue_golomb(&s->gb); //bbv_check_times
    h->progressive             = get_bits1(&s->gb);
    if(h->progressive)
        h->pic_structure = 1;
    else if(!(h->pic_structure = get_bits1(&s->gb) && (h->stc == PIC_PB_START_CODE)) )
        get_bits1(&s->gb);     //advanced_pred_mode_disable
    skip_bits1(&s->gb);        //top_field_first
    skip_bits1(&s->gb);        //repeat_first_field
    h->qp_fixed                = get_bits1(&s->gb);
    h->qp                      = get_bits(&s->gb,6);
    if(h->pic_type == FF_I_TYPE) {
        if(!h->progressive && !h->pic_structure)
            skip_bits1(&s->gb);//what is this?
        skip_bits(&s->gb,4);   //reserved bits
    } else {
        if(!(h->pic_type == FF_B_TYPE && h->pic_structure == 1))
            h->ref_flag        = get_bits1(&s->gb);
        skip_bits(&s->gb,4);   //reserved bits
        h->skip_mode_flag      = get_bits1(&s->gb);
    }
    h->loop_filter_disable     = get_bits1(&s->gb);
    if(!h->loop_filter_disable && get_bits1(&s->gb)) {
        h->alpha_offset        = get_se_golomb(&s->gb);
        h->beta_offset         = get_se_golomb(&s->gb);
    } else {
        h->alpha_offset = h->beta_offset  = 0;
    }
    check_for_slice(h);
    if(h->pic_type == FF_I_TYPE) {
        do {
            decode_mb_i(h, 0);
        } while(next_mb(h));
    } else if(h->pic_type == FF_P_TYPE) {
        do {
            if(h->skip_mode_flag) {
                skip_count = get_ue_golomb(&s->gb);
                while(skip_count--) {
                    decode_mb_p(h,P_SKIP);
                    if(!next_mb(h))
                        goto done;
                }
                mb_type = get_ue_golomb(&s->gb) + P_16X16;
            } else
                mb_type = get_ue_golomb(&s->gb) + P_SKIP;
            if(mb_type > P_8X8) {
                decode_mb_i(h, mb_type - P_8X8 - 1);
            } else
                decode_mb_p(h,mb_type);
        } while(next_mb(h));
    } else { /* FF_B_TYPE */
        do {
            if(h->skip_mode_flag) {
                skip_count = get_ue_golomb(&s->gb);
                while(skip_count--) {
                    decode_mb_b(h,B_SKIP);
                    if(!next_mb(h))
                        goto done;
                }
                mb_type = get_ue_golomb(&s->gb) + B_DIRECT;
            } else
                mb_type = get_ue_golomb(&s->gb) + B_SKIP;
            if(mb_type > B_8X8) {
                decode_mb_i(h, mb_type - B_8X8 - 1);
            } else
                decode_mb_b(h,mb_type);
        } while(next_mb(h));
    }
 done:
    if(h->pic_type != FF_B_TYPE) {
        if(h->DPB[1].data[0])
            s->avctx->release_buffer(s->avctx, (AVFrame *)&h->DPB[1]);
        memcpy(&h->DPB[1], &h->DPB[0], sizeof(Picture));
        memcpy(&h->DPB[0], &h->picture, sizeof(Picture));
        memset(&h->picture,0,sizeof(Picture));
    }
    return 0;
}

/*****************************************************************************
 *
 * headers and interface
 *
 ****************************************************************************/

/**
 * some predictions require data from the top-neighbouring macroblock.
 * this data has to be stored for one complete row of macroblocks
 * and this storage space is allocated here
 */
static void init_top_lines(AVSContext *h) {
    /* alloc top line of predictors */
    h->top_qp       = av_malloc( h->mb_width);
    h->top_mv[0]    = av_malloc((h->mb_width*2+1)*sizeof(vector_t));
    h->top_mv[1]    = av_malloc((h->mb_width*2+1)*sizeof(vector_t));
    h->top_pred_Y   = av_malloc( h->mb_width*2*sizeof(*h->top_pred_Y));
    h->top_border_y = av_malloc((h->mb_width+1)*16);
    h->top_border_u = av_malloc((h->mb_width)*10);
    h->top_border_v = av_malloc((h->mb_width)*10);

    /* alloc space for co-located MVs and types */
    h->col_mv       = av_malloc( h->mb_width*h->mb_height*4*sizeof(vector_t));
    h->col_type_base = av_malloc(h->mb_width*h->mb_height);
    h->block        = av_mallocz(64*sizeof(DCTELEM));
}

static int decode_seq_header(AVSContext *h) {
    MpegEncContext *s = &h->s;
    extern const AVRational ff_frame_rate_tab[];
    int frame_rate_code;

    h->profile =         get_bits(&s->gb,8);
    h->level =           get_bits(&s->gb,8);
    skip_bits1(&s->gb); //progressive sequence
    s->width =           get_bits(&s->gb,14);
    s->height =          get_bits(&s->gb,14);
    skip_bits(&s->gb,2); //chroma format
    skip_bits(&s->gb,3); //sample_precision
    h->aspect_ratio =    get_bits(&s->gb,4);
    frame_rate_code =    get_bits(&s->gb,4);
    skip_bits(&s->gb,18);//bit_rate_lower
    skip_bits1(&s->gb);  //marker_bit
    skip_bits(&s->gb,12);//bit_rate_upper
    s->low_delay =       get_bits1(&s->gb);
    h->mb_width  = (s->width  + 15) >> 4;
    h->mb_height = (s->height + 15) >> 4;
    h->s.avctx->time_base.den = ff_frame_rate_tab[frame_rate_code].num;
    h->s.avctx->time_base.num = ff_frame_rate_tab[frame_rate_code].den;
    h->s.avctx->width  = s->width;
    h->s.avctx->height = s->height;
    if(!h->top_qp)
        init_top_lines(h);
    return 0;
}

/**
 * finds the end of the current frame in the bitstream.
 * @return the position of the first byte of the next frame, or -1
 */
int ff_cavs_find_frame_end(ParseContext *pc, const uint8_t *buf, int buf_size) {
    int pic_found, i;
    uint32_t state;

    pic_found= pc->frame_start_found;
    state= pc->state;

    i=0;
    if(!pic_found){
        for(i=0; i<buf_size; i++){
            state= (state<<8) | buf[i];
            if(state == PIC_I_START_CODE || state == PIC_PB_START_CODE){
                i++;
                pic_found=1;
                break;
            }
        }
    }

    if(pic_found){
        /* EOF considered as end of frame */
        if (buf_size == 0)
            return 0;
        for(; i<buf_size; i++){
            state= (state<<8) | buf[i];
            if((state&0xFFFFFF00) == 0x100){
                if(state < SLICE_MIN_START_CODE || state > SLICE_MAX_START_CODE){
                    pc->frame_start_found=0;
                    pc->state=-1;
                    return i-3;
                }
            }
        }
    }
    pc->frame_start_found= pic_found;
    pc->state= state;
    return END_NOT_FOUND;
}

void ff_cavs_flush(AVCodecContext * avctx) {
    AVSContext *h = avctx->priv_data;
    h->got_keyframe = 0;
}

static int cavs_decode_frame(AVCodecContext * avctx,void *data, int *data_size,
                             uint8_t * buf, int buf_size) {
    AVSContext *h = avctx->priv_data;
    MpegEncContext *s = &h->s;
    int input_size;
    const uint8_t *buf_end;
    const uint8_t *buf_ptr;
    AVFrame *picture = data;
    uint32_t stc;

    s->avctx = avctx;

    if (buf_size == 0) {
        if(!s->low_delay && h->DPB[0].data[0]) {
            *data_size = sizeof(AVPicture);
            *picture = *(AVFrame *) &h->DPB[0];
        }
        return 0;
    }

    buf_ptr = buf;
    buf_end = buf + buf_size;
    for(;;) {
        buf_ptr = ff_find_start_code(buf_ptr,buf_end, &stc);
        if(stc & 0xFFFFFE00)
            return FFMAX(0, buf_ptr - buf - s->parse_context.last_index);
        input_size = (buf_end - buf_ptr)*8;
        switch(stc) {
        case SEQ_START_CODE:
            init_get_bits(&s->gb, buf_ptr, input_size);
            decode_seq_header(h);
            break;
        case PIC_I_START_CODE:
            if(!h->got_keyframe) {
                if(h->DPB[0].data[0])
                    avctx->release_buffer(avctx, (AVFrame *)&h->DPB[0]);
                if(h->DPB[1].data[0])
                    avctx->release_buffer(avctx, (AVFrame *)&h->DPB[1]);
                h->got_keyframe = 1;
            }
        case PIC_PB_START_CODE:
            *data_size = 0;
            if(!h->got_keyframe)
                break;
            init_get_bits(&s->gb, buf_ptr, input_size);
            h->stc = stc;
            if(decode_pic(h))
                break;
            *data_size = sizeof(AVPicture);
            if(h->pic_type != FF_B_TYPE) {
                if(h->DPB[1].data[0]) {
                    *picture = *(AVFrame *) &h->DPB[1];
                } else {
                    *data_size = 0;
                }
            } else
                *picture = *(AVFrame *) &h->picture;
            break;
        case EXT_START_CODE:
            //mpeg_decode_extension(avctx,buf_ptr, input_size);
            break;
        case USER_START_CODE:
            //mpeg_decode_user_data(avctx,buf_ptr, input_size);
            break;
        default:
            if (stc >= SLICE_MIN_START_CODE &&
                stc <= SLICE_MAX_START_CODE) {
                init_get_bits(&s->gb, buf_ptr, input_size);
                decode_slice_header(h, &s->gb);
            }
            break;
        }
    }
}

static int cavs_decode_init(AVCodecContext * avctx) {
    AVSContext *h = avctx->priv_data;
    MpegEncContext * const s = &h->s;

    MPV_decode_defaults(s);
    s->avctx = avctx;

    avctx->pix_fmt= PIX_FMT_YUV420P;

    h->luma_scan[0] = 0;
    h->luma_scan[1] = 8;
    h->intra_pred_l[      INTRA_L_VERT] = intra_pred_vert;
    h->intra_pred_l[     INTRA_L_HORIZ] = intra_pred_horiz;
    h->intra_pred_l[        INTRA_L_LP] = intra_pred_lp;
    h->intra_pred_l[ INTRA_L_DOWN_LEFT] = intra_pred_down_left;
    h->intra_pred_l[INTRA_L_DOWN_RIGHT] = intra_pred_down_right;
    h->intra_pred_l[   INTRA_L_LP_LEFT] = intra_pred_lp_left;
    h->intra_pred_l[    INTRA_L_LP_TOP] = intra_pred_lp_top;
    h->intra_pred_l[    INTRA_L_DC_128] = intra_pred_dc_128;
    h->intra_pred_c[        INTRA_C_LP] = intra_pred_lp;
    h->intra_pred_c[     INTRA_C_HORIZ] = intra_pred_horiz;
    h->intra_pred_c[      INTRA_C_VERT] = intra_pred_vert;
    h->intra_pred_c[     INTRA_C_PLANE] = intra_pred_plane;
    h->intra_pred_c[   INTRA_C_LP_LEFT] = intra_pred_lp_left;
    h->intra_pred_c[    INTRA_C_LP_TOP] = intra_pred_lp_top;
    h->intra_pred_c[    INTRA_C_DC_128] = intra_pred_dc_128;
    h->mv[ 7] = un_mv;
    h->mv[19] = un_mv;
    return 0;
}

static int cavs_decode_end(AVCodecContext * avctx) {
    AVSContext *h = avctx->priv_data;

    av_free(h->top_qp);
    av_free(h->top_mv[0]);
    av_free(h->top_mv[1]);
    av_free(h->top_pred_Y);
    av_free(h->top_border_y);
    av_free(h->top_border_u);
    av_free(h->top_border_v);
    av_free(h->col_mv);
    av_free(h->col_type_base);
    av_free(h->block);
    return 0;
}

AVCodec cavs_decoder = {
    "cavs",
    CODEC_TYPE_VIDEO,
    CODEC_ID_CAVS,
    sizeof(AVSContext),
    cavs_decode_init,
    NULL,
    cavs_decode_end,
    cavs_decode_frame,
    CODEC_CAP_DR1 | CODEC_CAP_DELAY,
    .flush= ff_cavs_flush,
};