This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Is someone working on reg-stack rewrite?
- To: Richard Henderson <rth at cygnus dot com>
- Subject: Re: Is someone working on reg-stack rewrite?
- From: Jeffrey A Law <law at cygnus dot com>
- Date: Mon, 05 Apr 1999 21:06:58 -0600
- cc: Jan Hubicka <hubicka at atrey dot karlin dot mff dot cuni dot cz>, egcs at cygnus dot com
- Reply-To: law at cygnus dot com
In message <19990405175951.A30941@cygnus.com>you write:
> I doubt it will ever be useful to move this before sched2. Either you
> have a pipeline and you schedule (in which case you need sched2 for ILP)
> or you don't have a pipeline and sched2 doesn't run.
Agreed. The process of converting to a regstack model basically serializes
FP code due to the dependencies on the stacked register file.
> When I was thinking about this late last year, I couldn't come up with
> a clean cprop mapping scheme. I only thought about it for a few days;
> perhaps you can come up with something better.
I spent a little more time thinking about this a while ago.
The basic problem (if I remember right) was that stuff like loading an
FP register changes the stack. Right?
It seems to me the way to handle that is to (internally) show the copies
that occur due to pushing something on the stack.
Then you can fall back to normal copy motion.
jeff