Re: [PATCH], PR target/81193, Add warning for using __builtin_cpu_* on old PowerPC GLIBC's

[Michael Meissner <meissner at linux dot vnet dot ibm dot com>]

On Wed, Jul 12, 2017 at 04:07:42PM -0500, Segher Boessenkool wrote:
> Hi Mike,
> 
> On Wed, Jul 12, 2017 at 11:38:11AM -0400, Michael Meissner wrote:
> > I also changed the current warning in target_clones handling to be an error
> > instead of a warning, since it really makes no sense to use target_clones if we
> > can't generate a resolver function.
> 
> Okay.  Another option is to then always use the default, but this
> certainly is easier; let's hope we can get away with it.

We can always change it, but it needs to make sure the user doesn't get two
warnings then (one for target_clones, and a second one for
__builtin_cpu_supports).

> > 	(rs6000_get_function_versions_dispatcher): Change the warnging
> 
> Typo.

Thanks.

> > --- gcc/config/rs6000/rs6000-c.c	(.../svn+ssh://meissner@gcc.gnu.org/svn/gcc/trunk/gcc/config/rs6000)	(revision 250063)
> > +++ gcc/config/rs6000/rs6000-c.c	(.../gcc/config/rs6000)	(working copy)
> > @@ -644,6 +644,10 @@ rs6000_cpu_cpp_builtins (cpp_reader *pfi
> >      builtin_define ("__FLOAT128_HARDWARE__");
> >    if (TARGET_LONG_DOUBLE_128 && FLOAT128_IBM_P (TFmode))
> >      builtin_define ("__ibm128=long double");
> > +#ifdef TARGET_LIBC_PROVIDES_HWCAP_IN_TCB
> > +  builtin_define ("__BUILTIN_CPU_SUPPORTS__");
> > +  builtin_define ("__BUILTIN_CPU_IS__");
> > +#endif
> 
> Is it useful to have two defines?  They always are enabled at the same
> time afaics?  Thinking of a good name isn't so easy of course, but maybe
> just __BUILTIN_CPU_SUPPORTS__ is fine always.

Yeah, I was wondering about that.  I'll delete __BUILTIN_CPU_IS__, since I find
that to be less useful.

> 
> Or do other architectures already have both?

The only other architecture that supports __builtin_cpu_{init,is,supports} is
the X86, and they don't have the macros.  But then they also don't have the
issue that you can't do a lot in the ifunc resolver.

> > +The @code{__builtin_cpu_is} function requires GLIBC 2.23 or newer in
> > +order to export the hardware capability bits.  The
> > +@code{__builtin_cpu_is} function will return 0 if the compiler was
> > +configured for an earlier GCC.  GCC defines the macro
> > +@code{__BUILTIN_CPU_IS__} if you can use the @code{__builtin_cpu_is}
> > +built-in function.
> 
> "if you can use it" is not very clear (the line before you said you
> always can use it, it just returns 0).

I will reword it.

> Okay for trunk and 7 with those things taken care of.  Thanks!

Thanks.

-- 
Michael Meissner, IBM
IBM, M/S 2506R, 550 King Street, Littleton, MA 01460-6245, USA
email: meissner@linux.vnet.ibm.com, phone: +1 (978) 899-4797