This is the mail archive of the
mailing list for the GCC project.
Re: IRA improvements 0/4
- From: Jan Hubicka <hubicka at ucw dot cz>
- To: Vladimir Makarov <vmakarov at redhat dot com>, gcc-patches at gcc dot gnu dot org
- Date: Wed, 7 Jul 2010 16:08:38 +0200
- Subject: Re: IRA improvements 0/4
- References: <4C34818E.email@example.com>
> The following patches are from ira-improv branch. I believe that
> the work on the branch reached the state when it can be merged into
> the trunk.
> The branch was tested on x86/x86_64/ppc64/arm and benchmarked on
> SPEC2000 and SPEC2006 on x86_64. For x86_64 SPEC20006 (Core i7) the
> branch creates smaller size code (about 0.2%), needs less 1 sec for
> compilation, and generates the same INT score and 1% better FP score.
> The result may vary for other targets (e.g. a bit worse SPECInt2000
> score but bigger improvements for SPECFP2000 score, or bigger
> compilation time which is still less 1%).
> Removing cover classes makes coloring algorithm more complicated and
> it slows down considerably the code for trivial colorability and the
> code for
> cost caclulation (because more reg classes should be tried) therefore a
> lot of work has been done to speedup IRA to compensate the slowing
I am impressed you got computations without cover classes without slowdowns!
Will this allow us to simplify constraints in i386.md where we currently
lie about instructions that can be done both in integer registers or
in SSE but can't mix the classes?