comparison ac3dec.c @ 7554:96d57e3b78e5 libavcodec

optimize ac3_downmix. 1.3x faster 5.1->stereo, 1.9x faster 5.1->mono.

7553:b5f8d814a206

static void set_downmix_coeffs(AC3DecodeContext *s)
{
    int i;
    float cmix = gain_levels[center_levels[s->center_mix_level]];
    float smix = gain_levels[surround_levels[s->surround_mix_level]];

    for(i=0; i<s->fbw_channels; i++) {
        s->downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
        s->downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
    }

    if(s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
        int nf = s->channel_mode - 4;
        s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf+1][1] = smix;
    }

    /* calculate adjustment needed for each channel to avoid clipping */
    s->downmix_coeff_adjust[0] = s->downmix_coeff_adjust[1] = 0.0f;
    for(i=0; i<s->fbw_channels; i++) {
        s->downmix_coeff_adjust[0] += s->downmix_coeffs[i][0];
        s->downmix_coeff_adjust[1] += s->downmix_coeffs[i][1];
    }
    s->downmix_coeff_adjust[0] = 1.0f / s->downmix_coeff_adjust[0];
    s->downmix_coeff_adjust[1] = 1.0f / s->downmix_coeff_adjust[1];
}

/**
 * Decode the grouped exponents according to exponent strategy.
 * reference: Section 7.1.3 Exponent Decoding

}

/**
 * Downmix the output to mono or stereo.
 */
static void ac3_downmix(AC3DecodeContext *s,
                        float samples[AC3_MAX_CHANNELS][256], int ch_offset)
{
    int i, j;
    float v0, v1;

    for(i=0; i<256; i++) {
        v0 = v1 = 0.0f;
        for(j=0; j<s->fbw_channels; j++) {
            v0 += samples[j+ch_offset][i] * s->downmix_coeffs[j][0];
            v1 += samples[j+ch_offset][i] * s->downmix_coeffs[j][1];
        }
        v0 *= s->downmix_coeff_adjust[0];
        v1 *= s->downmix_coeff_adjust[1];
        if(s->output_mode == AC3_CHMODE_MONO) {
            samples[ch_offset][i] = (v0 + v1) * LEVEL_MINUS_3DB;
        } else if(s->output_mode == AC3_CHMODE_STEREO) {
            samples[  ch_offset][i] = v0;
            samples[1+ch_offset][i] = v1;
        }
    }
}

/**

        }

        do_imdct(s, s->channels);

        if(downmix_output) {
            ac3_downmix(s, s->output, 0);
        }
    } else {
        if(downmix_output) {
            ac3_downmix(s, s->transform_coeffs, 1);
        }

        if(!s->downmixed) {
            s->downmixed = 1;
            // FIXME delay[] is half the size of the other downmixes
            ac3_downmix(s, s->delay, 0);
        }

        do_imdct(s, s->out_channels);
    }

7554:96d57e3b78e5

static void set_downmix_coeffs(AC3DecodeContext *s)
{
    int i;
    float cmix = gain_levels[center_levels[s->center_mix_level]];
    float smix = gain_levels[surround_levels[s->surround_mix_level]];
    float norm0, norm1;

    for(i=0; i<s->fbw_channels; i++) {
        s->downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
        s->downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
    }

    if(s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
        int nf = s->channel_mode - 4;
        s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf+1][1] = smix;
    }

    /* renormalize */
    norm0 = norm1 = 0.0;
    for(i=0; i<s->fbw_channels; i++) {
        norm0 += s->downmix_coeffs[i][0];
        norm1 += s->downmix_coeffs[i][1];
    }
    norm0 = 1.0f / norm0;
    norm1 = 1.0f / norm1;
    for(i=0; i<s->fbw_channels; i++) {
        s->downmix_coeffs[i][0] *= norm0;
        s->downmix_coeffs[i][1] *= norm1;
    }

    if(s->output_mode == AC3_CHMODE_MONO) {
        for(i=0; i<s->fbw_channels; i++)
            s->downmix_coeffs[i][0] = (s->downmix_coeffs[i][0] + s->downmix_coeffs[i][1]) * LEVEL_MINUS_3DB;
    }
}

/**
 * Decode the grouped exponents according to exponent strategy.
 * reference: Section 7.1.3 Exponent Decoding

}

/**
 * Downmix the output to mono or stereo.
 */
static av_noinline void ac3_downmix(AC3DecodeContext *s,
                                    float samples[AC3_MAX_CHANNELS][256])
{
    int i, j;
    float v0, v1;

    if(s->output_mode == AC3_CHMODE_STEREO) {
        for(i=0; i<256; i++) {
            v0 = v1 = 0.0f;
            for(j=0; j<s->fbw_channels; j++) {
                v0 += samples[j][i] * s->downmix_coeffs[j][0];
                v1 += samples[j][i] * s->downmix_coeffs[j][1];
            }
            samples[0][i] = v0;
            samples[1][i] = v1;
        }
    } else if(s->output_mode == AC3_CHMODE_MONO) {
        for(i=0; i<256; i++) {
            v0 = 0.0f;
            for(j=0; j<s->fbw_channels; j++)
                v0 += samples[j][i] * s->downmix_coeffs[j][0];
            samples[0][i] = v0;
        }
    }
}

/**

        }

        do_imdct(s, s->channels);

        if(downmix_output) {
            ac3_downmix(s, s->output);
        }
    } else {
        if(downmix_output) {
            ac3_downmix(s, s->transform_coeffs+1);
        }

        if(!s->downmixed) {
            s->downmixed = 1;
            // FIXME delay[] is half the size of the other downmixes
            ac3_downmix(s, s->delay);
        }

        do_imdct(s, s->out_channels);
    }