This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: PR testsuite/20772 - further pieces
On Thu, May 26, 2005 at 11:35:25PM +0200, Andreas Jaeger wrote:
> >> >> --- i386-sse-8.c 17 Nov 2004 17:21:45 -0000 1.7
> >> >> +++ i386-sse-8.c 26 May 2005 17:44:07 -0000
> >> >> @@ -4,8 +4,8 @@
> >> >> /* The xstormy16 doesn't support V2DI. */
> >> >> /* { dg-do compile { xfail xstormy16-*-* } } */
> >> >> /* { dg-options "" } */
> >> >> -/* { dg-options "-march=pentium3" { target i?86-*-* } } */
> >> >> -/* { dg-skip-if "" { i?86-*-* } { "-m64" } { "" } } */
> >> >> +/* { dg-options "-march=pentium3" { target i?86-*-* x86_64-*-* } } */
> >> >> +/* { dg-require-effective-target ilp32 } */
> >> >
> >> > The test is for a problem using SSE, so when compiling for ilp32 x86 it
> >> > should probably use "-march-pentium3 -msse" (or is -msse implied for
> >> > -march=pentium3?). Without those options the test will use generic
> >> > vectors and presumably shouldn't fail anywhere.
> >>
> >> I just checked - this looks fine, -march=pentium3 allows the usage of
> >> SSE registers. So, I'm not going to change this.
> >
> > It shouldn't need "dg-require-effective-target ilp32", since everything
> > besides x86 ilp32 can use generic vectors. Leaving that in your patch
> > will cause this test to be skipped for targets that are currently using
> > it.
>
> I do not agree, the bug only occured with -march=pentium3 - and not
> -march=pentium4. Both imply SSE. For x86_64 -march=pentium3 is not a
> valid architecture, so I have to ask for x86 ilp32,
The test is supposed to succeed even on x86-64 -m64, you just can't use
-march=pentium3 with -m64.
So this means IMHO that:
/* { dg-do compile { xfail xstormy16-*-* } } */
/* { dg-options "" } */
/* { dg-options "-march=pentium3" { target { i?86-*-* && ilp32 } } } */
/* { dg-options "-march=pentium3" { target { x86_64-*-* && ilp32 } } } */
should be used (and dg-skip-if removed).
Jakub