comparison cabac.h @ 3956:0910f2844f9a libavcodec

branchless renormalization (1% faster get_cabac) old branchless renormalization wasnt faster because gcc was scared of the shift variable (missusing bit variable now)

3955:051a9c8b257f

#endif
    c->low -= CABAC_MASK;
    c->bytestream+= CABAC_BITS/8;
}

#if 0 /* all use commented */
static void refill2(CABACContext *c){
    int i, x;

    x= c->low ^ (c->low-1);
    i= 8 - ff_h264_norm_shift[x>>(CABAC_BITS+1)];

    x= -CABAC_MASK;

    if(c->bytestream < c->bytestream_end)
#if CABAC_BITS == 16
        x+= (c->bytestream[0]<<9) + (c->bytestream[1]<<1);
#else
        x+= c->bytestream[0]<<1;
#endif

        if(s<128)
            *state= s;
#endif
        renorm_cabac_decoder_once(c);
    }else{
//        int shift= ff_h264_norm_shift[RangeLPS>>17];
        bit= (s&1)^1;
        c->low -= c->range;
        *state= c->lps_state[s];
        c->range = RangeLPS;
        renorm_cabac_decoder(c);
/*        c->range = RangeLPS<<shift;
        c->low <<= shift;
        if(!(c->low & 0xFFFF)){
            refill2(c);
        }*/
    }
#else
    lps_mask= (c->range - c->low)>>31;

    c->low -= c->range & lps_mask;

3956:0910f2844f9a

#endif
    c->low -= CABAC_MASK;
    c->bytestream+= CABAC_BITS/8;
}

#if 1 /* all use commented */
static void refill2(CABACContext *c){
    int i, x;

    x= c->low ^ (c->low-1);
    i= 9 - ff_h264_norm_shift[x>>(CABAC_BITS+1)];

    x= -CABAC_MASK;

    if(c->bytestream <= c->bytestream_end)
#if CABAC_BITS == 16
        x+= (c->bytestream[0]<<9) + (c->bytestream[1]<<1);
#else
        x+= c->bytestream[0]<<1;
#endif

        if(s<128)
            *state= s;
#endif
        renorm_cabac_decoder_once(c);
    }else{
        bit= ff_h264_norm_shift[RangeLPS>>17];
        c->low -= c->range;
        *state= c->lps_state[s];
//        c->range = RangeLPS;
//        renorm_cabac_decoder(c);
        c->range = RangeLPS<<bit;
        c->low <<= bit;
        bit= (s&1)^1;

        if(!(c->low & 0xFFFF)){
            refill2(c);
        }
    }
#else
    lps_mask= (c->range - c->low)>>31;

    c->low -= c->range & lps_mask;