This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [tuples] Some vectorizer fixes
On Wed, Jul 16, 2008 at 06:22, Jakub Jelinek <jakub@redhat.com> wrote:
> I guess this is similar to how gimple_build_assign doesn't set
> SSA_NAME_DEF_STMT for the lhs, only gimple_assign_set_lhs does.
Ah, OK.
>> Should we have an
>> alternate way of setting LHS of GIMPLE_CALL and/or GIMPLE_ASSIGN that
>> doesn't update SSA information?
>
> Isn't that *gimple_call_lhs_ptr that way?
Well, yes, but it looks hideous, don't you think? Not that I mind too
much. This is a corner case, after all. We used to force callers to
set SSA_NAME_DEF_STMT on their own, but the vast majority of users
need SSA_NAME_DEF_STMT set, so it's better to do it in the helper.
Another set of interfaces I'd like to narrow down are the creation of
temporaries and such. There are 2-3 bookkeeping things that need to
be done when creating temporaries and I think we can simplify that.
More generally, we tend to provide more than one way of doing the same
thing. That's confusing (though unavoidable to some extent).
> Anyway, the current vectorizable_function only calls
> gimple_call_fndecl and gimple_call_flags on the stmt, so perhaps
> we could get away even without setting lhs at all before the
> vectorizable_function call. I.e. just
Sounds like a good idea.
Diego.