This is the mail archive of the
mailing list for the GCC project.
Re: [tree-ssa] Re: PATCH: [gcc3.5 improvement branch] Very Simpleconstant propagation
- From: Roger Sayle <roger at eyesopen dot com>
- To: Caroline Tice <ctice at apple dot com>
- Cc: gcc-patches at gcc dot gnu dot org, Geoffrey Keating <geoffk at apple dot com>
- Date: Sun, 18 Jan 2004 10:45:22 -0700 (MST)
- Subject: Re: [tree-ssa] Re: PATCH: [gcc3.5 improvement branch] Very Simpleconstant propagation
On Fri, 16 Jan 2004, Caroline Tice wrote:
> So...while people are deciding whether and/or how to implement this in
> the tree-ssa branch, should I commit my version to the 3.5 integration
> branch, then deal with merging it with whatever happens in tree-ssa
> later, or what? I am sorry, but it is unclear what the status of my
> patch is now...Thanks in advance for your clarifications...
Could you please confirm that my proposed alternative patch to this
problem addresses your optimization problem?
This patch fixes the example given in your post, but clearly there's
a larger underlying benchmark or application that has motivated this
change. Clearly, if your testing can demonstrate that it fixes your
issue, and can satisfy Geoff K that it really does work, then the
much less invasive improvement to GCC's combine would clearly be
preferrable to several source files and a new RTL optimization pass.
Especially, if this particular optimization (which should really be a
variant of loop strength reduction) is to be eventually implemented in
the SSA passes. However, the alternative improvement to combine, which
might also cure 20020720-1.c and other unrelated optimization failures
through the use of REG_EQUAL notes, would complement rather than
duplicate tree-ssa functionality.
In all fairness, if you can convince me your solution is technically
superior, I'll be happy to approve it. In the meantime, I'll hold off
applying my version, until this issue is settled between you, me and
Geoff (and any other maintainer). Does this sound reasonable?
Roger Sayle, E-mail: firstname.lastname@example.org
OpenEye Scientific Software, WWW: http://www.eyesopen.com/
Suite 1107, 3600 Cerrillos Road, Tel: (+1) 505-473-7385
Santa Fe, New Mexico, 87507. Fax: (+1) 505-473-0833