comparison flac.c @ 5741:154b02065699 libavcodec

20% faster lpc, 6% overall flac decoding

5740:4d62c0055174

static int decode_subframe_lpc(FLACContext *s, int channel, int pred_order)
{
    int i, j;
    int coeff_prec, qlevel;
    int coeffs[pred_order];

//    av_log(s->avctx, AV_LOG_DEBUG, "  SUBFRAME LPC\n");

    /* warm up samples */
//    av_log(s->avctx, AV_LOG_DEBUG, "   warm up samples: %d\n", pred_order);

    for (i = 0; i < pred_order; i++)
    {
        s->decoded[channel][i] = get_sbits(&s->gb, s->curr_bps);
//        av_log(s->avctx, AV_LOG_DEBUG, "    %d: %d\n", i, s->decoded[channel][i]);
    }

    coeff_prec = get_bits(&s->gb, 4) + 1;
    if (coeff_prec == 16)
    {

        int64_t sum;
        for (i = pred_order; i < s->blocksize; i++)
        {
            sum = 0;
            for (j = 0; j < pred_order; j++)
                sum += (int64_t)coeffs[j] * s->decoded[channel][i-j-1];
            s->decoded[channel][i] += sum >> qlevel;
        }
    } else {
        for (i = pred_order; i < s->blocksize-1; i += 2)
        {
            int c = coeffs[pred_order-1];
            int s0 = c * s->decoded[channel][i-pred_order];
            int s1 = 0;
            for (j = pred_order-1; j > 0; j--)
            {
                int d = s->decoded[channel][i-j];
                s1 += c*d;
                c = coeffs[j-1];
                s0 += c*d;
            }
            s0 = s->decoded[channel][i] += s0 >> qlevel;
            s1 += c * s0;
            s->decoded[channel][i+1] += s1 >> qlevel;
        }
        if (i < s->blocksize)
        {
            int sum = 0;
            for (j = 0; j < pred_order; j++)
                sum += coeffs[j] * s->decoded[channel][i-j-1];
            s->decoded[channel][i] += sum >> qlevel;
        }
    }

    return 0;
}

5741:154b02065699

static int decode_subframe_lpc(FLACContext *s, int channel, int pred_order)
{
    int i, j;
    int coeff_prec, qlevel;
    int coeffs[pred_order];
    int32_t *decoded = s->decoded[channel];

//    av_log(s->avctx, AV_LOG_DEBUG, "  SUBFRAME LPC\n");

    /* warm up samples */
//    av_log(s->avctx, AV_LOG_DEBUG, "   warm up samples: %d\n", pred_order);

    for (i = 0; i < pred_order; i++)
    {
        decoded[i] = get_sbits(&s->gb, s->curr_bps);
//        av_log(s->avctx, AV_LOG_DEBUG, "    %d: %d\n", i, decoded[i]);
    }

    coeff_prec = get_bits(&s->gb, 4) + 1;
    if (coeff_prec == 16)
    {

        int64_t sum;
        for (i = pred_order; i < s->blocksize; i++)
        {
            sum = 0;
            for (j = 0; j < pred_order; j++)
                sum += (int64_t)coeffs[j] * decoded[i-j-1];
            decoded[i] += sum >> qlevel;
        }
    } else {
        for (i = pred_order; i < s->blocksize-1; i += 2)
        {
            int c;
            int d = decoded[i-pred_order];
            int s0 = 0, s1 = 0;
            for (j = pred_order-1; j > 0; j--)
            {
                c = coeffs[j];
                s0 += c*d;
                d = decoded[i-j];
                s1 += c*d;
            }
            c = coeffs[0];
            s0 += c*d;
            d = decoded[i] += s0 >> qlevel;
            s1 += c*d;
            decoded[i+1] += s1 >> qlevel;
        }
        if (i < s->blocksize)
        {
            int sum = 0;
            for (j = 0; j < pred_order; j++)
                sum += coeffs[j] * decoded[i-j-1];
            decoded[i] += sum >> qlevel;
        }
    }

    return 0;
}