Quoting Michael Niedermayer (2020-10-07 16:45:56)
> On Tue, Oct 06, 2020 at 12:08:27PM +0200, Anton Khirnov wrote:
> > Quoting Paul B Mahol (2020-10-06 10:19:13)
> > > On Tue, Oct 06, 2020 at 10:08:58AM +0200, Anton Khirnov wrote:
> > > > Quoting Paul B Mahol (2020-10-06 02:17:14)
> > > > > Signed-off-by: Paul B Mahol <one...@gmail.com>
> > > > > ---
> > > > > libavcodec/apedec.c | 10 +++-------
> > > > > 1 file changed, 3 insertions(+), 7 deletions(-)
> > > > >
> > > > > diff --git a/libavcodec/apedec.c b/libavcodec/apedec.c
> > > > > index 8fe7b5ee86..ea36247eb9 100644
> > > > > --- a/libavcodec/apedec.c
> > > > > +++ b/libavcodec/apedec.c
> > > > > @@ -575,14 +575,10 @@ static inline int
> > > > > ape_decode_value_3990(APEContext *ctx, APERice *rice)
> > > > > base = range_decode_culfreq(ctx, pivot);
> > > > > range_decode_update(ctx, 1, base);
> > > > > } else {
> > > > > - int base_hi = pivot, base_lo;
> > > > > - int bbits = 0;
> > > > > + int base_hi, base_lo;
> > > > > + int bbits = 16 - ff_clz(pivot);
> > > >
> > > > This assumes unsigned is always 32bit, no?
> > >
> > > ksum is 32 bit, from which pivot is derived.
> > >
> > > Should I use explicitly uint32_t type instead?
> > > Where is unsigned not 32bit?
> >
> > I don't think uint32_t would help, as ff_clz() can expand to a compiler
> > builtin.
> >
> > What I'm concerned about is the unstated assumption about
> > sizeof(int/unsigned) == 4 spreading through the codebase. It's already
> > present in plenty of places, so I certainly don't intend to block your
> > patch over this. But we should consider explitly documenting this, and
> > maybe testing in configure.
>
> At least in code i wrote and write i consider it a bug if it would
> assume sizeof(int/unsigned) == 4
It is my impression (not sure how justified) that there are many
overflow checks that cast a multiplication of two ints to int64,
thus assuming that int64 is strictly larger. E.g. 6b7bcd437e5
>
> We also could add a ILP64 fate target, that way we would better understand
> how much really assumes 32bit int
I suppose that would be interesting.
--
Anton Khirnov
_______________________________________________
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".
