This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [RFC][PATCH] Preferred rename register in regrename pass
- From: Jeff Law <law at redhat dot com>
- To: Robert Suchanek <Robert dot Suchanek at imgtec dot com>, "gcc-patches at gcc dot gnu dot org" <gcc-patches at gcc dot gnu dot org>
- Date: Thu, 17 Sep 2015 11:01:37 -0600
- Subject: Re: [RFC][PATCH] Preferred rename register in regrename pass
- Authentication-results: sourceware.org; auth=none
- References: <B5E67142681B53468FAF6B7C31356562441B7622 at hhmail02 dot hh dot imgtec dot org>
On 09/17/2015 08:38 AM, Robert Suchanek wrote:
Hi,
We came across a situation for MIPS64 where moves for sign-extension were
not converted into a nop because of IRA spilled some of the allocnos and
assigned different hard register for the output operand in the move.
LRA is not fixing this up as most likely the move was not introduced by
the LRA itself. I found it hard to fix this in LRA and looked at
an alternative solution where regrename pass appeared to be the best candidate.
Yea, we've never been great at tying the source & destination of
sign/zero extensions. The inherently different modes often caused the
old allocator (and pre-allocator bits like regmove, and post-allocator
bits like reload) to 'give up'.
I'm not sure if this is something that should be enabled by default for everyone
or a target hook should be added. Any other comments/suggestions?
I'll let Bernd comment on the patch itself. But I would say that if
you're setting up cases where we can tie the source/dest of an extension
together, then it's a good thing. It'll cause more of them to turn into
NOPs and it'll make the redundant extension elimination pass more
effective as well. This would be something that I'd expect to benefit
most architectures (obviously to varying degrees).
Jeff