Re: Register rename and hppa return

["John David Anglin" <dave at hiauly1 dot hia dot nrc dot ca>]

Forwarded message:
From dave Tue Jan 2 01:50:51 EST 2001
Subject: Re: Register rename and hppa return
To: law@redhat.com
Date: Tue, 2 Jan 2001 01:50:51 -0500 (EST)
From: "John David Anglin" <dave@hiauly1>
Cc: alan@linuxcare.com.au, rth@cygnus.com
In-Reply-To: <24125.978395358@upchuck> from "Jeffrey A Law" at Jan 1, 2001 05:29:18 pm
X-Mailer: ELM [version 2.4 PL25]
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Content-Length: 2520      

>   > > --- gcc/config/pa/pa.md~	Fri Dec  8 20:16:38 2000
>   > > +++ gcc/config/pa/pa.md	Mon Dec 11 00:02:44 2000
>   > > @@ -5467,7 +5467,8 @@
>   > >  ;; Unconditional and other jump instructions.
>   > >  
>   > >  (define_insn "return"
>   > > -  [(return)]
>   > > +  [(return)
>   > > +   (use (reg:SI 2))]
>   > >    "hppa_can_use_return_insn_p ()"
>   > >    "*
>   > >  {
>   > 
>   > Jeff had objections to this.  I think the following is a better solution
>   > as it doesn't require changing the return insn.  However, I haven't been
>   > able to do a complete test.
>   > 
>   > Dave
>   > -- 
>   > J. David Anglin                                  dave.anglin@nrc.ca
>   > National Research Council of Canada              (613) 990-0752 (FAX: 952-6
>   > 605)
>   > 
>   > 2000-12-08  John David Anglin  <dave@hiauly1.hia.nrc.ca>
>   > 
>   > 	* pa.h: Define RETURN_POINTER_REGNUM and EPILOGUE_USES.
> Clearly the return insns need to indicate what register they use -- the
> more closely RTL matches the reality of the machine, the better.  Thus
> I think this solution is actually a step backwards from the original
> proposed solution.
> 
> The right way to go is to find a way to properly distinguish between a
> jump to a non-trivial epilogue and a jump to a trivial epilogue since
> we can eliminate the jump to the trivial epilogue in the latter case.

Previously, you wrote: 

    On Wed, Nov 29, 2000 at 11:54:09PM -0700, Jeffrey A Law wrote:
    > Err, umm.  Isn't this going to confuse reorg & jump.c's recognition of
    > return insns?  I believe they want to be able to do something like
    > GET_CODE (thingie) == RETURN?

In looking at reorg and jump.c, it appeared that EPILOGUE_USES was
invented to resolve this problem in a manner which allows GET_CODE (thingie)
== RETURN.  The alpha and ia64 ports seem to use exactly the approach
suggested above.

I haven't noticed any functional difference between the three previously
proposed solutions.  However, I haven't checked to see if jumps to trivial
epilogues are being eliminated.

It appears from your comments that your prefered solution is the
one previously suggested by Richard in this message
<http://gcc.gnu.org/ml/gcc-bugs/2000-11/msg00377.html>:

    I've wanted for quite a while to change (return) to (return <addr>)
    so that this issue couldn't be overlooked.

Dave
-- 
J. David Anglin                                  dave.anglin@nrc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6605)


-- 
J. David Anglin                                  dave.anglin@nrc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6605)