This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: PowerPC register and memory cost update
- From: Geoff Keating <geoffk at geoffk dot org>
- To: David Edelsohn <dje at watson dot ibm dot com>
- Cc: gcc-patches at gcc dot gnu dot org
- Date: 23 Oct 2002 13:52:30 -0700
- Subject: Re: PowerPC register and memory cost update
- References: <200210231723.NAA30872@makai.watson.ibm.com>
David Edelsohn <dje@watson.ibm.com> writes:
> This patch fixes the register and memory cost calculations to be
> more accurate, fixing a Java bootstrap failure caused by stupid register
> allocation choices.
Actually, the failure was because of the special meaning of '2' as a
return from REGISTER_MOVE_COST.
> The memory cost change affects performance -- both positive effect
> and negative effect, depending on the testcase. This could demonstrate
> that the cost model requires more precision or that some other heuristic
> in the compiler is wrong. It needs more study.
Most likely, it's just that GCC's register allocator isn't as good as
it could be; what happens if you use -fnew-ra?
[The patch itself looks pretty good, thanks!]
--
- Geoffrey Keating <geoffk@geoffk.org>