This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: patch: vector register arguments
On Fri, 2001-11-02 at 21:17, Richard Henderson wrote:
> On Fri, Nov 02, 2001 at 01:48:38PM -0500, Aldy Hernandez wrote:
> > Without wider mode information, gcc is unable to determine how to
> > store a given vector register if the narrowest vector mode (V2SF or
> > V2QI) does not exist in the given target. Gcc will try to iterate to
> > get a wider mode, and unable to find one that fits, aborts.
>
> I don't think it makes sense to have wider mode for vectors.
> What does that mean, exactly? If it just means "here's a way
yeah, it doesn't make much sense from a "wider" perspective. i looked
at other places where we use GET_MODE_WIDER_MODE et al, and never do we
handle vector modes. so i was thinking it wouldn't hurt since we don't
do any operations vectors (apart from moving them). this could change
in the future though.
i see no clear way of implementing apply_*_size() without having to add
other fields in machmode.def, or perhaps another table altogether with
pointers so we can iterate through vectors.
> to iterate over the lot of them", then I think that should be
> clearly documented somewhere.
can i add documentation in the DEF_MACHMODE docs at the head of
machmode.def?
>
>
> r~
--
Aldy Hernandez E-mail: aldyh@redhat.com
Professional Gypsy
Red Hat, Inc.