[Bug c++/54427] Expose more vector extensions

[glisse at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54427

--- Comment #6 from Marc Glisse <glisse at gcc dot gnu.org> 2012-09-20 19:46:57 UTC ---
(In reply to comment #4)
> This test is now failing on powerpc64-unknown-linux-gnu due to warning messages
> being emitted for 32-bit.

I was kind of expecting that (mentioned in the patch submission) but couldn't
see what I was changing by moving the test.

> This is due to the fact that in the old location the compile command included
> -w which suppressed warnings, whereas the new location does not.

Ah, thanks a lot! I didn't realize there was an automatic -w in the old
location.

> The test can
> be fixed for powerpc by adding the following line (as is present in
> c-c++-common/vector-compare-2.c).
> 
> /* { dg-options "-mabi=altivec" { target { { powerpc*-*-linux* } && ilp32 } } }
> */
> 
> But it looks like this same test started failing on i686-pc-linux-gnu
> (http://gcc.gnu.org/ml/gcc-testresults/2012-09/msg01842.html), guessing for the
> same reason. Not sure if there's some similar type option that can be added for
> that target or whether '-w' should just be added to the testcase.

Yes, I think adding -w is the easiest solution (after all that's just reverting
to the old behavior). Otherwise the test could be changed to pass arguments by
pointer like the others, but I didn't dare do that in case it was specifically
testing passing by value.

I'll submit the -w ASAP, thanks again.