[FFmpeg-cvslog] avfilter/f_ebur128: multiply is usually faster than divide

Paul B Mahol git at videolan.org
Thu Feb 24 12:18:36 EET 2022


ffmpeg | branch: master | Paul B Mahol <onemda at gmail.com> | Thu Feb 24 10:30:37 2022 +0100| [15a1104d73cbd47a67fc14ce2ddc97a92145695e] | committer: Paul B Mahol

avfilter/f_ebur128: multiply is usually faster than divide

Also guard against overflow when subtracting from unsigned.

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=15a1104d73cbd47a67fc14ce2ddc97a92145695e
---

 libavfilter/f_ebur128.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/libavfilter/f_ebur128.c b/libavfilter/f_ebur128.c
index d5ffab7ca1..f63cf77983 100644
--- a/libavfilter/f_ebur128.c
+++ b/libavfilter/f_ebur128.c
@@ -762,7 +762,7 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *insamples)
 
                     /* get lower loudness to consider */
                     n = 0;
-                    nb_pow = LRA_LOWER_PRC  * nb_powers / 100. + 0.5;
+                    nb_pow = LRA_LOWER_PRC * 0.01 * nb_powers + 0.5;
                     for (i = gate_hist_pos; i < HIST_SIZE; i++) {
                         n += ebur128->i3000.histogram[i].count;
                         if (n >= nb_pow) {
@@ -773,9 +773,9 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *insamples)
 
                     /* get higher loudness to consider */
                     n = nb_powers;
-                    nb_pow = LRA_HIGHER_PRC * nb_powers / 100. + 0.5;
+                    nb_pow = LRA_HIGHER_PRC * 0.01 * nb_powers + 0.5;
                     for (i = HIST_SIZE - 1; i >= 0; i--) {
-                        n -= ebur128->i3000.histogram[i].count;
+                        n -= FFMIN(n, ebur128->i3000.histogram[i].count);
                         if (n < nb_pow) {
                             ebur128->lra_high = ebur128->i3000.histogram[i].loudness;
                             break;



More information about the ffmpeg-cvslog mailing list