This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: VIS2 pattern review
On 10/13/2011 12:55 PM, David Miller wrote:
> -(define_insn "<vis3_addsub_ss_insn><vbits>_vis"
> +(define_insn "<vis3_addsub_ss_patname><mode>"
Missing a "3" on the end. Otherwise these look ok.
> Unfortunately, that would involve some ABI changes for the VIS
> builtins. I'm trending towards considering just changing things
> anyways since the VIS intrinsics were next to unusable beforehand.
Why? You can do just about anything you like inside the builtin
expander, including frobbing the modes around.
E.g. for x86, we can't create anything TImode at the user level,
and thus can't expose either TImode or V2TImode operands into the
builtins. So we have the builtins use V2DI or V4DImode and do
some gen_lowpart frobbing while expanding the builtins to rtl.
I don't see why you couldn't accept DImode from the builtin and
transform it to V1DImode in the rtl. Etc.
r~