This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: [tree-ssa] What would be interesting SPEC2000 tests flags?
- From: Daniel Berlin <dberlin at dberlin dot org>
- To: law at redhat dot com
- Cc: Steven Bosscher <s dot bosscher at student dot tudelft dot nl>, gcc at gcc dot gnu dot org, Andreas Jaeger <aj at suse dot de>, dnovillo at redhat dot com
- Date: Sat, 31 May 2003 23:04:46 -0400 (EDT)
- Subject: Re: [tree-ssa] What would be interesting SPEC2000 tests flags?
- References: <200306010206.h51269Be024866@speedy.slc.redhat.com>
On Sat, 31 May 2003 law@redhat.com wrote:
> In message <3ED8C64C.1060001@student.tudelft.nl>, Steven Bosscher writes:
> >Hi,
> >
> >The other day I noticed that Andreas Jaeger's SPEC2k
> >testers for the tree-ssa branch have been testing
> >with BASE="-O2" and PEAK="-O2 -ftree-ccp" for some
> >time now. But since tree-ccp has been enabled by
> >default at -O1 already since February (?), BASE==PEAK.
> >We would get more interesting results if we choose
> >different flag settings for BASE and PEAK.
> >
> >I notified Andreas and he asked what flags he should
> >use instead. So what would be interesting to compare
> >for tree-ssa? Something like -O1 vs. -O2, or both -O2
> >but one with a few RTL optimizers disabled? Or
> >-fdisable-simple (IMVHO not very useful since Diego
> >already tests that). Make some suggestions! :-)
> The two I'm most interested in are the effects of SSAPRE and
> PTA. I believe we have switches for both.
PTA, as of a month ago, shaved a few percent off the bootstrap
time.
Feel free to turn it on, it bootstraps fine.
Well, actually, wait a day, i've got to commit the fix for the problem
Diego introduced with memory tags (querying about fake variables).
SSAPRE's effects are not worth measure right now without a flag
to disable the into-ssa optimizations.
It only makes 189 reloads (as of this afternoon) during a compile of gcc
(IE a single stage of bootstrap with only the c compiler enabled) with
the into-ssa optimizations enabled.
Thus, you'd need a flag to disable into-ssa optimizations to see how much
better it does, because the into-ssa optimizations block redundancy
eliminations.
As an example, in yyparse of gengtype-yacc.c, the into-ssa optimizations
eliminate 19 redundant expressions. PRE then eliminates 1 more.
However, if you turn off the redundancy elimination in tree-ssa.c,
PRE will eliminate 24 redundancies.
Thus, you won't get a sense how well PRE does unless you have numbers
with and without into-ssa optimizations to compare it to.
>
> jeff
>
>