This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Dumb register allocation (PPC)
- From: Dale Johannesen <dalej at apple dot com>
- To: David Edelsohn <dje at watson dot ibm dot com>
- Cc: Dale Johannesen <dalej at apple dot com>, law at redhat dot com, Zack Weinberg <zack at codesourcery dot com>, gcc at gcc dot gnu dot org
- Date: Tue, 14 May 2002 12:09:49 -0700
- Subject: Re: Dumb register allocation (PPC)
On Sunday, May 12, 2002, at 04:54 PM, David Edelsohn wrote:
> Please see the
>
> Trivial code generation stupidity thread:
> http://gcc.gnu.org/ml/gcc/2001-12/msg00907.html
>
> which evolved into the
> cant_combine_insn_p hard_reg->reg moves thread:
> http://gcc.gnu.org/ml/gcc/2001-12/msg01283.html
>
> This is a general difficulty of GCC handling argument registers. The new
> register alloctor will allow us to start addressing it.
FWIW, I've got the check in cant_combine_insn_p
conditionalized on if(SMALL_REGISTER_CLASSES), and
everything works fine on ppc, and the code does improve.
If you look back at the thread when this patch went in:
http://gcc.gnu.org/ml/gcc-patches/2000-11/msg01466.html
you'll see there was some discussion of doing what I did,
but Jan Hubicka claimed improvements with the patch
even on non-SRC machines so it didn't go in that way;
but this patch is not a win on ppc according to my
measurements. I haven't bothered submitting this
given jh's results and the firmness of rth's opinion
in the thread you quote, but I could.