comparison h264.c @ 2748:135aa7d0bd17 libavcodec

avoid one transpose (730->680 dezicycles on duron)

2747:6eded34ab57b

    int16_t     (*mvd_table[2])[2];
    int16_t     mvd_cache[2][5*8][2] __align8;
    uint8_t     *direct_table;
    uint8_t     direct_cache[5*8];

}H264Context;

static VLC coeff_token_vlc[4];
static VLC chroma_dc_coeff_token_vlc;

    // set defaults
//    s->decode_mb= ff_h263_decode_mb;
    s->low_delay= 1;
    avctx->pix_fmt= PIX_FMT_YUV420P;

    decode_init_vlc(h);
    
    if(avctx->extradata_size > 0 && avctx->extradata &&
       *(char *)avctx->extradata == 1){
        h->is_avc = 1;

        const uint8_t *scan, *dc_scan;
        
//        fill_non_zero_count_cache(h);

        if(IS_INTERLACED(mb_type)){
            scan= field_scan;
            dc_scan= luma_dc_field_scan;
        }else{
            scan= zigzag_scan;
            dc_scan= luma_dc_zigzag_scan;
        }

        dquant= get_se_golomb(&s->gb);

    if( cbp || IS_INTRA16x16( mb_type ) ) {
        const uint8_t *scan, *dc_scan;
        int dqp;

        if(IS_INTERLACED(mb_type)){
            scan= field_scan;
            dc_scan= luma_dc_field_scan;
        }else{
            scan= zigzag_scan;
            dc_scan= luma_dc_zigzag_scan;
        }

        h->last_qscale_diff = dqp = decode_cabac_mb_dqp( h );
        s->qscale += dqp;

2748:135aa7d0bd17

    int16_t     (*mvd_table[2])[2];
    int16_t     mvd_cache[2][5*8][2] __align8;
    uint8_t     *direct_table;
    uint8_t     direct_cache[5*8];

    uint8_t zigzag_scan[16];
    uint8_t field_scan[16];
}H264Context;

static VLC coeff_token_vlc[4];
static VLC chroma_dc_coeff_token_vlc;

    // set defaults
//    s->decode_mb= ff_h263_decode_mb;
    s->low_delay= 1;
    avctx->pix_fmt= PIX_FMT_YUV420P;

    if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
        memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
        memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
    }else{
        int i;
        for(i=0; i<16; i++){
#define T(x) (x>>2) | ((x<<2) & 0xF)
            h->zigzag_scan[i] = T(zigzag_scan[i]);
            h-> field_scan[i] = T( field_scan[i]);
        }
    }

    decode_init_vlc(h);
    
    if(avctx->extradata_size > 0 && avctx->extradata &&
       *(char *)avctx->extradata == 1){
        h->is_avc = 1;

        const uint8_t *scan, *dc_scan;
        
//        fill_non_zero_count_cache(h);

        if(IS_INTERLACED(mb_type)){
            scan= h->field_scan;
            dc_scan= luma_dc_field_scan;
        }else{
            scan= h->zigzag_scan;
            dc_scan= luma_dc_zigzag_scan;
        }

        dquant= get_se_golomb(&s->gb);

    if( cbp || IS_INTRA16x16( mb_type ) ) {
        const uint8_t *scan, *dc_scan;
        int dqp;

        if(IS_INTERLACED(mb_type)){
            scan= h->field_scan;
            dc_scan= luma_dc_field_scan;
        }else{
            scan= h->zigzag_scan;
            dc_scan= luma_dc_zigzag_scan;
        }

        h->last_qscale_diff = dqp = decode_cabac_mb_dqp( h );
        s->qscale += dqp;