Ulf Zibis (12019-07-19): > >From f14142a22d340cba48b3e2352a33026590dec3a5 Mon Sep 17 00:00:00 2001 > From: Ulf Zibis <[email protected]> > Date: 19.07.2019, 18:27:51 >
> avfilter/select: avoid twice rounding
Maybe I a missing something, I see no change in rounding in this code. I
suspect your diagnosis is wrong.
>
> diff --git a/libavfilter/f_select.c b/libavfilter/f_select.c
> index 1132375..39cc004 100644
> --- a/libavfilter/f_select.c
> +++ b/libavfilter/f_select.c
> @@ -297,6 +297,7 @@
>
> #define D2TS(d) (isnan(d) ? AV_NOPTS_VALUE : (int64_t)(d))
> #define TS2D(ts) ((ts) == AV_NOPTS_VALUE ? NAN : (double)(ts))
> +#define TS2DT(ts, tb) (TS2D((ts) * (tb).num / (double)(tb).den))
You are multiplying two possibly large integers: it can overflow. I am
not sure if it is really a concern, but it must not be done blindly.
>
> static void select_frame(AVFilterContext *ctx, AVFrame *frame)
> {
> @@ -307,11 +308,11 @@
> if (isnan(select->var_values[VAR_START_PTS]))
> select->var_values[VAR_START_PTS] = TS2D(frame->pts);
> if (isnan(select->var_values[VAR_START_T]))
> - select->var_values[VAR_START_T] = TS2D(frame->pts) *
> av_q2d(inlink->time_base);
> + select->var_values[VAR_START_T] = TS2DT(frame->pts,
> inlink->time_base);
>
> select->var_values[VAR_N ] = inlink->frame_count_out;
> select->var_values[VAR_PTS] = TS2D(frame->pts);
> - select->var_values[VAR_T ] = TS2D(frame->pts) *
> av_q2d(inlink->time_base);
> + select->var_values[VAR_T ] = TS2DT(frame->pts, inlink->time_base);
> select->var_values[VAR_POS] = frame->pkt_pos == -1 ? NAN :
> frame->pkt_pos;
> select->var_values[VAR_KEY] = frame->key_frame;
> select->var_values[VAR_CONCATDEC_SELECT] = get_concatdec_select(frame,
> av_rescale_q(frame->pts, inlink->time_base, AV_TIME_BASE_Q));
Regards,
--
Nicolas George
signature.asc
Description: PGP signature
_______________________________________________ ffmpeg-devel mailing list [email protected] https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email [email protected] with subject "unsubscribe".
