This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: help interpreting gcc 4.1.1 optimisation bug
On Mon, Jun 12, 2006 at 01:40:18PM +0100, Andrew Haley wrote:
> andrew@walrond.org writes:
> > On Mon, Jun 12, 2006 at 12:30:50PM +0000, andrew@walrond.org wrote:
> > > On Mon, Jun 12, 2006 at 02:21:21PM +0200, Richard Guenther wrote:
> > > > On 6/12/06, andrew@walrond.org <andrew@walrond.org> wrote:
> > > > >On Mon, Jun 12, 2006 at 01:01:47PM +0100, Andrew Haley wrote:
> > > > >>
> > > > >> I'm starting to be a little suspicious about host2little(). I wonder
> > > > >> if that's well-defined.
> > > > >>
> > > > >> I'm just guessing here becasue I can't see the code. Is it possible
> > > > >
> > > > > template<typename T> T swap_endian(T x)
> > > > > {
> > > > > char* a = reinterpret_cast<char*>(&x);
> > > > > char* b = a + sizeof(T) - 1;
> > > > > while (a<b) {
> > > > > char tmp = *a;
> > > > > *a++ = *b;
> > > > > *b-- = tmp;
> > > > > }
> > > > > return x;
> > > > > }
> > > > >
> > > > > template<typename T> T host2little(T x)
> > > > > {
> > > > > unsigned u = 1; return (*reinterpret_cast<char*>(&u) ? x :
> > > > > swap_endian(x));
> > > > > }
> > > >
> > > > You are violating aliasing rules. Try -fno-strict-aliasing, which
> > > > should make the code
> > > > work, or fix it.
> > >
> > > Hmmm - don't grok; I'll do some googling. Compiling now with
> > > -fno-strict-aliasing, but the manual says -O2 includes
> > > -strict-aliasing, so what gives?
> >
> > Indeed, -fno-strict-aliasing makes no difference; its still broken.
>
> Well, that is interesting! I think both Richard Guenther and I both
> thought you had an aliasing problem somewhere: all those pointer casts
> are something of a red flag...
Well, my point above was that -strict-aliasing is included in -O2 and
my code works fine at -O2. Only -O3 causes problems, so I didn't expect
-fno-strict-aliasing to make any difference.
>
> OK, well, we're stuck until you can make a test case.
>
Oh well. brute force it is then ;)
I'll be back with test case asap. Thanks for comments!
> Andrew.
>