Re: [Mingw-users] gcc 3.2.1 with -O2 -fomit-frame-pointer produce incorect code

[Danny S. <dan...@cl...>]

----- Original Message -----
From: "Danny Smith" <dan...@cl...>
To: "Vadim Yegorov" <zg...@bm...>; <min...@li...>
Sent: Tuesday, 17 December 2002 23:41
Subject: Re: [Mingw-users] gcc 3.2.1 with -O2 -fomit-frame-pointer
produce incorect code


>
> ----- Original Message -----
> From: "Danny Smith" <dan...@ya...>
> To: "Vadim Yegorov" <zg...@bm...>; <min...@li...>
> Sent: Tuesday, 17 December 2002 20:16
> Subject: Re: [Mingw-users] gcc 3.2.1 with -O2 -fomit-frame-pointer
> produce incorect code
>
>
> > --- Vadim Yegorov <zg...@bm...> wrote: > Hello,
> > >
> > Thanks for the report and testcase. This is fixed with 3.3 (and
IIRC,
> not too
> > long ago).  I'll see if I can track down the related GCC bug report
> and
> > hopefully a fix that can be backported.  No promises.
> >
> > Danny
> >
>
> Your testcase works if I add either -mcpu=i686 or if I change  to -O3.

Hmm, I think it was just luck that it worked for those settings.
Playing around with different settings gives  success one time and
failure another,  so insn scheduling seems to be involved.  I believe
the bug is related to this one reported at gnats:

http://gcc.gnu.org/ml/gcc-bugs/2002-11/msg01528.html
optimization/8750: Cygwin prolog generation erroneously emitting
__alloca as regular function call.

Danny