On Sat, Jun 27, 2020 at 01:50:48AM -0400, Michael Meissner wrote:
> On Thu, Jun 25, 2020 at 12:18:42PM -0500, Segher Boessenkool wrote:
> > > +/* { dg-do compile } */
> > > +/* { dg-require-effective-target powerpc_prefixed_addr } */
> >
> > Is this test necessary anymore, does -mcpu=power10 not guarantee that?
>
> I believe the test is necessary to prevent regressions. As I said in the
> test,
> we stumbled on this problem when building GLIBC with -mcpu=future/power10.
We now have
/* Enable -mprefixed by default on power10 systems. */
if (TARGET_POWER10 && (rs6000_isa_flags_explicit & OPTION_MASK_PREFIXED) == 0)
rs6000_isa_flags |= OPTION_MASK_PREFIXED;
so the testsuite can just assume it is enabled on all targets it is
supported for at all.
(The selector doesn't directly *hurt* of course, but it leads to cargo-
culting, future tests using this unneeded selector as well).
Segher
