This is the mail archive of the
mailing list for the GCC project.
Re: Register allocation problem?
- To: David Edelsohn <dje at watson dot ibm dot com>
- Subject: Re: Register allocation problem?
- From: Peter Bergner <bergner at brule dot borg dot umn dot edu>
- Date: Tue, 21 Aug 2001 12:47:34 -0500
- Cc: Peter Bergner <bergner at brule dot borg dot umn dot edu>, Daniel Berlin <dan at cgsoftware dot com>, Lars Brinkhoff <lars at nocrew dot org>, gcc at gcc dot gnu dot org
- References: <firstname.lastname@example.org> <200108211722.NAA29138@makai.watson.ibm.com>
David Edelsohn wrote:
: Reload primarily is necessary to handle some perverse restrictions
: of some architectures. The difficulty is finding the balance between
: leaving enough of reload in place for the difficult corner cases while not
: having reload interfere with the correct register allocation and
: materialization which the new register allocator has performed.
But it's the "perverse restrictions" that make register allocation fun! :-)
So reload is a noop for some/most architectures?
: Rice and others are helping with the design and tuning.
Good to hear that Rice is helping. I assume that means they are
willing to allow the use of their patent in this area? How goes
the same for IBM? Personally I'd like to see this released.
It means I might have a chance of getting some of my spilling
and coalescing ideas released too.
BTW, is there some other mailing list or informal list where
discussion of the new reg-alloc branch is occurring? I haven't
seen much on this list. In particular, who is actively working
on this branch?