This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: [RFC] failures found while pounding on GCC trunk
On 23 Sep 2007 09:13:59 -0700, Ian Lance Taylor <iant@google.com> wrote:
> "Richard Guenther" <richard.guenther@gmail.com> writes:
>
> > On 9/23/07, Daniel Berlin <dberlin@dberlin.org> wrote:
> > > ># all tests fail with ICE in verify_curr_properties; -m32/-m64
> > > >-O2 -fno-tree-salias
> > > >-Os -fno-tree-salias
> > >
> > >
> > > Known. This is normally the first place we build aliasing info.
> > > Unless we have a compelling reason to have this flag, we should simply
> > > remove the flag.
> >
> > I agree with that, just remove the flag. (well, make it a no-op and
> > deprecate it)
>
> We've got a test case with gcc 4.2 for which compilation time goes
> from nine minutes to 30 seconds when we use that option. I know the
> code is much better in mainline. Still, I would prefer to keep the
> option unless we are certain that similar compilation time blowouts
> can not occur. Lots of code can profit from optimization even with
> struct aliasing turned off.
Well, ATM the flag doesn't work. And honestly we have enough "kinds"
of alias bugs - adding another knob to get different wrong-code bugs
doesn't look too appealing to me ;)
Richard.