This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH, testsuite] Fix PR34168 - fortran vectorized tests on x86 without support of SSE2


On Wed, Mar 19, 2008 at 12:45 PM, Victor Kaplansky <VICTORK@il.ibm.com> wrote:

>  > You don't need to set -msse2 to use cpuid.
>
>  OK, thanks, I'll remove it.
>
>
>  >
>  > I think you have the logic reversed ;)
>  >
>  > You can perhaps code:
>  >
>  > if (__get_cpuid (...))
>  >   return edx & bit_SSE2;
>  >
>  > return 0;
>
>  No, in this case return is used to pass exit status of the main()
>  program, so the zero value means "true".

Ah, I see. I got confused by the comment above this part. But please
check your code with -Wunintialized, for possible "may be used
uninitialized" warning.

>  > Please also put this check in gcc.dg/vect/vect.exp and
>  > g++.dg/vect/vect.exp. In fact, it looks that all these three files
>  > should be synchronized w.r.t to unsupported target handling. Perhaps
>  > we need a global function in testsuite/lib/target-supports.exp (lets
>  > name it something like check_vector_hw_available) to be shared between
>  > these three vect.exp files, so we don't have to change this part in
>  > three separate places.
>
>  I agree. Do we want to make this reorganization as part of this PR or as
>  a separate patch? I prefer to fix PR and reorganize vect.exp later.

At the moment, it would be nice if you synchronize target handling
stuff in all three vect.exp files (just cut-n-paste job). Looking a
bit deeper into the code, I see that this code also sets target
dependent vector flags, so it is questionable, if we want to
substitute this functionality with a new function here. IMO, possible
cleanup should be part of another follow-up patch.

Uros.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]