This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH] PowerPC multiply-add (was patch: rs6000 specific)
- From: Geoff Keating <geoffk at geoffk dot org>
- To: Dale Johannesen <dalej at apple dot com>
- Cc: gcc-patches at gcc dot gnu dot org, David Edelsohn <dje at watson dot ibm dot com>
- Date: 04 Dec 2001 17:16:17 -0800
- Subject: Re: [PATCH] PowerPC multiply-add (was patch: rs6000 specific)
- References: <4602C132-E90A-11D5-8923-003065C86F94@apple.com>
- Reply-to: Geoff Keating <geoffk at redhat dot com>
Dale Johannesen <dalej@apple.com> writes:
> On Tuesday, December 4, 2001, at 02:46 PM, David Edelsohn wrote:
> > My biggest problem with the FMA patch is why GCC does not perform
> > this transformation itself. It seems sort of sad that we need to insert
> > all of these variants instead of GCC canonicalizing the form itself.
>
> It is sad, but of course the machine-independent parts currently have
> no concept of multiply-add; these patterns arise from combine.
... however, the same is true of (for instance) the mask insert
instruction, but we still canonify combinations of NOT, AND, and OR,
because even though combine has no specific knowledge about what
instructions are available, it does know that if there is an
instruction, it'll be
(not (and (reg ...) (reg ...)))
and not
(or (not (reg ...)) (not (reg ...)))
--
- Geoffrey Keating <geoffk@geoffk.org> <geoffk@redhat.com>