This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: dangerous cleverness? ppc int<->float conversions, subreg
On Fri, Oct 26, 2001 at 12:40:00PM -0700, Dale Johannesen wrote:
>
> On Friday, October 26, 2001, at 12:14 PM, Zack Weinberg wrote:
>
> >On Fri, Oct 26, 2001 at 12:55:14PM -0400, David Edelsohn wrote:
> >>Zack,
> >>
> >> Have you looked into my comment about move_by_pieces?
> >>
> >> The problem is how to discourage GCC from using FPRs for DImode.
> >>GCC only uses FPRs for DImode to move blocks of memory and float->int
> >>conversions.
>
> We've noticed this on Darwin too. I'm not sure anybody's mentioned
> this point: the alignment requirements for "double" memrefs can be
> more restrictive than those for (non-double-containing) structs. A
> misaligned FP load/store may be much slower than two int load/stores,
> or may fail completely. I don't think there's anyplace in the code
> that's checking this.
Hmm, that would be an argument for not doing this at all, which would
be easier than not doing it when inappropriate. Is this a Darwin ABI
thing or a general PPC thing?
zw