comparison vp3.c @ 10620:972b17631531 libavcodec

Small refactoring: Instead of 4 loops for decoding AC coefficients based on their grouping, create one loop that indexes into a table of AC VLC tables. There is also a small optimization here: Do not call unpack_vlcs() if there are no fragments in the list with outstanding coefficients. My profiling indicates that this can save upwards of 1 million dezicycles per frame throughout the course of unpack_dct_coeffs().

10619:d930f99edbf9

    int dc_y_table;
    int dc_c_table;
    int ac_y_table;
    int ac_c_table;
    int residual_eob_run = 0;

    /* fetch the DC table indexes */
    dc_y_table = get_bits(gb, 4);
    dc_c_table = get_bits(gb, 4);

    /* fetch the AC table indexes */
    ac_y_table = get_bits(gb, 4);
    ac_c_table = get_bits(gb, 4);

    /* unpack the group 1 AC coefficients (coeffs 1-5) */
    for (i = 1; i <= 5; i++) {
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_1[ac_y_table], i,
            1, residual_eob_run);

        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_1[ac_c_table], i,
            0, residual_eob_run);
    }

    /* unpack the group 2 AC coefficients (coeffs 6-14) */
    for (i = 6; i <= 14; i++) {
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_2[ac_y_table], i,
            1, residual_eob_run);

        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_2[ac_c_table], i,
            0, residual_eob_run);
    }

    /* unpack the group 3 AC coefficients (coeffs 15-27) */
    for (i = 15; i <= 27; i++) {
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_3[ac_y_table], i,
            1, residual_eob_run);

        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_3[ac_c_table], i,
            0, residual_eob_run);
    }

    /* unpack the group 4 AC coefficients (coeffs 28-63) */
    for (i = 28; i <= 63; i++) {
        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_4[ac_y_table], i,
            1, residual_eob_run);

        residual_eob_run = unpack_vlcs(s, gb, &s->ac_vlc_4[ac_c_table], i,
            0, residual_eob_run);
    }

    return 0;
}

10620:972b17631531

    int dc_y_table;
    int dc_c_table;
    int ac_y_table;
    int ac_c_table;
    int residual_eob_run = 0;
    VLC *y_tables[64];
    VLC *c_tables[64];

    /* fetch the DC table indexes */
    dc_y_table = get_bits(gb, 4);
    dc_c_table = get_bits(gb, 4);

    /* fetch the AC table indexes */
    ac_y_table = get_bits(gb, 4);
    ac_c_table = get_bits(gb, 4);

    /* build tables of AC VLC tables */
    for (i = 1; i <= 5; i++) {
        y_tables[i] = &s->ac_vlc_1[ac_y_table];
        c_tables[i] = &s->ac_vlc_1[ac_c_table];
    }
    for (i = 6; i <= 14; i++) {
        y_tables[i] = &s->ac_vlc_2[ac_y_table];
        c_tables[i] = &s->ac_vlc_2[ac_c_table];
    }
    for (i = 15; i <= 27; i++) {
        y_tables[i] = &s->ac_vlc_3[ac_y_table];
        c_tables[i] = &s->ac_vlc_3[ac_c_table];
    }
    for (i = 28; i <= 63; i++) {
        y_tables[i] = &s->ac_vlc_4[ac_y_table];
        c_tables[i] = &s->ac_vlc_4[ac_c_table];
    }

    /* decode all AC coefficents */
    for (i = 1; i <= 63; i++) {
        if (s->fragment_list_y_head != -1)
            residual_eob_run = unpack_vlcs(s, gb, y_tables[i], i,
                1, residual_eob_run);

        if (s->fragment_list_c_head != -1)
            residual_eob_run = unpack_vlcs(s, gb, c_tables[i], i,
                0, residual_eob_run);
    }

    return 0;
}