On Mon, 16 Jun 2025 13:19:57 +0200 Niklas Haas <ffm...@haasn.xyz> wrote:
> From: Niklas Haas <g...@haasn.dev>
>
> Easier to read, less convoluted, and ~30% faster. Most importantly, this
> avoids repeating the redundant recalculation of the true peak on every single
> sample, by moving the FIND_PEAK() loop out of the main loop. (Note that
> FIND_PEAK() does not depend on the current sample index at all, so there is no
> reason for it to ever be recomputed here)
Fixed the commit message formatting locally.
> ---
> libavfilter/f_ebur128.c | 23 ++++++++++++++---------
> 1 file changed, 14 insertions(+), 9 deletions(-)
>
> diff --git a/libavfilter/f_ebur128.c b/libavfilter/f_ebur128.c
> index 2e1eedd855..23092b597f 100644
> --- a/libavfilter/f_ebur128.c
> +++ b/libavfilter/f_ebur128.c
> @@ -668,16 +668,22 @@ static int filter_frame(AVFilterLink *inlink, AVFrame
> *insamples)
> (const uint8_t **)insamples->data, nb_samples);
> if (ret < 0)
> return ret;
> - for (ch = 0; ch < nb_channels; ch++)
> - ebur128->true_peaks_per_frame[ch] = 0.0;
> - for (idx_insample = 0; idx_insample < ret; idx_insample++) {
> - for (ch = 0; ch < nb_channels; ch++) {
> - ebur128->true_peaks[ch] = FFMAX(ebur128->true_peaks[ch],
> fabs(*swr_samples));
> - ebur128->true_peaks_per_frame[ch] =
> FFMAX(ebur128->true_peaks_per_frame[ch],
> -
> fabs(*swr_samples));
> - swr_samples++;
> +
> + double maxpeak = 0.0;
> + for (int ch = 0; ch < nb_channels; ch++) {
> + double tp = ebur128->true_peaks[ch];
> + double tppf = 0.0;
> + for (int i = 0; i < ret; i++) {
> + const double sample = fabs(swr_samples[i * nb_channels]);
> + tp = FFMAX(tp, sample);
> + tppf = FFMAX(tppf, sample);
> }
> + maxpeak = FFMAX(maxpeak, tp);
> + ebur128->true_peaks[ch] = tp;
> + ebur128->true_peaks_per_frame[ch] = tppf;
> }
> +
> + ebur128->true_peak = DBFS(maxpeak);
> }
> #endif
>
> @@ -720,7 +726,6 @@ static int filter_frame(AVFilterLink *inlink, AVFrame
> *insamples)
> }
>
> FIND_PEAK(ebur128->sample_peak, ebur128->sample_peaks, SAMPLES);
> - FIND_PEAK(ebur128->true_peak, ebur128->true_peaks, TRUE);
>
> /* For integrated loudness, gating blocks are 400ms long with 75%
> * overlap (see BS.1770-2 p5), so a re-computation is needed each
> 100ms
> --
> 2.49.0
>
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
