On Thu, Oct 22, 2015 at 07:04:41AM -0400, Ganesh Ajjanagadde wrote: > On Thu, Oct 22, 2015 at 5:49 AM, Carl Eugen Hoyos <ceho...@ag.or.at> wrote: > > Ganesh Ajjanagadde <gajjanagadde <at> gmail.com> writes: > > > >> > This broke fate with -ftrapv > > > >> > The problem seems to be in av_gcd() and not the De-Bruijn version of > >> > ff_ctzll since the gcc builtin is being used. > >> > >> Don't have the time to investigate right now - revert for now unless > >> someone can fix it quickly. > > > > Ping? > > > > Please revert if nobody can fix this. > > "Fixing" this at the moment is highly dubious - read the thread. It > does not even occur in a reproducible way, does not trigger on ubsan, > I can't even see it on ftrapv. Show me why it is not a toolchain > issue.
IMO never just assume that a problem that appears after a change you do is not caused by your code. Because if you do and you are wrong the bug might never be fixed while OTOH, if you assume its in your code you will sooner or later by debuging find the bug or more details, and that may be a code generation issue in the toolchain outside your code in case you where wrong. Its the safer default assumtation as the failure path if you are wrong is less bad. either way, bug fixed the reason for non reproduceablility likely was that the x86 asm overrode the C code in which the bug was [...] -- Michael GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB I have often repented speaking, but never of holding my tongue. -- Xenocrates
signature.asc
Description: Digital signature
_______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel