This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Graph coloring for register allocation?
- To: Michael Matz <matzmich at cs dot tu-berlin dot de>
- Subject: Re: Graph coloring for register allocation?
- From: Daniel Berlin <dberlin at redhat dot com>
- Date: Sun, 21 Jan 2001 12:56:23 -0500 (EST)
- cc: Michael Hayes <m dot hayes at elec dot canterbury dot ac dot nz>, Daniel Berlin <dberlin at redhat dot com>, Robert Dewar <dewar at gnat dot com>, Untitled <gcc at gcc dot gnu dot org>, Stan Shebs <shebs at apple dot com>
On Sun, 21 Jan 2001, Michael Matz wrote:
> Hi,
>
> On Sat, 20 Jan 2001, Michael Hayes wrote:
> > > This was actually the largest pain in the ass for generating the
> > > interference graph. I stole two routines from combine.c and modified them
> > > (they tell you whether a register is live after a certain insn).
> >
> > I posted some generic dataflow routines that I posted to this list a
> > few weeks ago that could help with this. As well as liveness info,
> > they generate def-use and use-def
>
> Yep, I have them, but not looked at them closely due to missing time.
> Also, meanwhile I was playing with the thought of building the
> interference graph incrementally (as proposed in one of the newer
> articles), and for that u/d-d/u-chains are not the best source. But at
> least for gathering the info your routines seem very handy.
What papers show how to build it incrementally?
I couldn't find any.
As for the handiness, Yes, definitely.
Because there are so many implementations available, i've started with
Iterated Register Coalescing, and will convert it to optimistic when it's
done.
Works well so far.
Just so we are all on the same page, everyone realizes that the new
register allocator will replace the regmove, the local-alloc, and the
global-alloc passes, right?
> > > info from which you can generate webs from.
>
> Ahh, a Muchnick reader? ;-) I found nowhere else (IIRC) the term "web"
> for what is called everywhere else live range. Although I also like more
> the former one ;)
>
>
> Ciao,
> Michael.
>
>