This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: removing ctr branches from rs6000
- From: Spundun <spundun at ISI dot EDU>
- To: Dale Johannesen <dalej at apple dot com>
- Cc: David Edelsohn <dje at watson dot ibm dot com>, gcc <gcc at gcc dot gnu dot org>
- Date: 26 Nov 2002 11:51:30 -0800
- Subject: Re: removing ctr branches from rs6000
- References: <A7690EF7-016A-11D7-A6D6-000393D76DAA@apple.com>
On Tue, 2002-11-26 at 10:12, Dale Johannesen wrote:
>
> On Tuesday, November 26, 2002, at 09:56 AM, David Edelsohn wrote:
>
> >>>>>> Spundun writes:
> > Spundun> grep through the source). Can anyone suggest a clean way of
> > removing
> > Spundun> these "barnch on count" support?
> >
> > Those macros are defined automatically based on the existence of
> > particular named patterns in the machine description.
>
> Yes. In this particular case, decrement_and_branch_on_count is already
> gone
> from rs6000.md, as of July 2000 (but see *ctrsi*). You might want to
> use a
> more recent version as a starting point.
Thanx for your replies.
I am working on 2.95.3+motorola's version of altivec stuff added in
2.95.2+lots of changes already done by me :) (any comments on this are
welcomed :))
I Commented out the decrement_and_branch_on_count template, didnt work;
commented out ctrsi template, didnt work; commented out the offending
template that was showing up in the .greg file "ctrsi_internal1",
worked.
didnt work = The ctr instructions still remained in the O2 optimized
output.
So it works now but can somebody explain what just happened over here?
Spundun