This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH] Re-fix PR target/47755 on powerpc VSX
On Tue, Mar 08, 2011 at 07:12:27PM -0500, David Edelsohn wrote:
> On Tue, Mar 8, 2011 at 6:56 PM, Michael Meissner
> <meissner@linux.vnet.ibm.com> wrote:
> > PR 47755 fixed some bugs on VSX with V2DI constants, but the patch itself had
> > problems. ?In particular, the easy_altivec_constant support assumed the largest
> > int size was SImode. ?This would cause the compiler to generate VSPLTI{W,S,B}
> > to load up a constant instead of loading it up from memory.
> >
> > This patch only allows (vector long long) { 0, 0 } and
> > (vector long long) { -1, -1 } as easy V2DI constants. ?There are some other
> > constants that could be generated using the Altivec instructions in the future.
> >
> > I bootstrapped this patch and had no regressions with make check. ?Is it ok to
> > install?
> >
> > [gcc]
> > 2011-03-08 ?Michael Meissner ?<meissner@linux.vnet.ibm.com>
> >
> > ? ? ? ?PR target/47755
> > ? ? ? ?* config/rs6000/rs6000.c (easy_altivec_constant): Correctly handle
> > ? ? ? ?V2DI/V2DF constants. ?Only all 0's or all 1's are easy.
> > ? ? ? ?(output_vec_const_move): Ditto.
> >
> > [gcc/testsuite]
> > 2011-03-08 ?Michael Meissner ?<meissner@linux.vnet.ibm.com>
> >
> > ? ? ? ?PR target/47755
> > ? ? ? ?* gcc.target/powerpc/pr47755-2.c: New file.
>
> Should the CONST_INT test be wrapped in #ifdef HOST_BITS_PER_WIDE_INT?
I dunno. Given that it is harmless on 64-bit, I would tend to not put the
#ifdef there. But I would have no objection to adding it.
> Okay.
>
> Thanks, David
--
Michael Meissner, IBM
5 Technology Place Drive, M/S 2757, Westford, MA 01886-3141, USA
meissner@linux.vnet.ibm.com fax +1 (978) 399-6899