Re: tree-ssa branch: Call for preliminary testing

[Andrew MacLeod <amacleod@redhat.com>]

On Thu, 2003-06-12 at 12:02, Diego Novillo wrote:
> On Mon, 2003-05-12 at 09:01, Richard Guenther wrote:
> 
> > Got it compile on a 2Gig machine - .ii file can be fetched from
> > 
> > http://www.tat.physik.uni-tuebingen.de/~rguenth/FieldCentering.cmpl.ii.gz
> > 
> > the test uses max. 1.5GB ram on this machine.
> > 
> > Richard.
> > 
> Seems partially fixed now.  It takes 19 minutes to compile, though. But!
> it doesn't use more than 300Mb of core now.
> 
> $ /usr/bin/time gcc/cc1plus -ftemplate-depth-80 -fno-exceptions -Drestrict=__restrict__ -DNOPAssert -DNOCTAssert -O2 -march=athlon -funroll-loops -fomit-frame-pointer --param min-inline-insns=200 -quiet FieldCentering.cmpl.ii               1106.46user 1.19system
> 18:55.17elapsed 97%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (1205major+142535minor)pagefaults 0swaps
> 
> In contrast, this uses 200Mb of core in mainline and takes 17 seconds to
> compile.  I'm thinking this and Gerald's PR8361 are good test cases :)
> 
> $ /usr/bin/time ~/merge-tree-ssa/mainline/native/bld.tobiano/gcc/cc1plus -ftemplate-depth-80 -fno-exceptions -Drestrict=__restrict__ -DNOPAssert -DNOCTAssert -O2 -march=athlon -funroll-loops -fomit-frame-pointer --param min-inline-insns=200 -quiet FieldCentering.cmpl.ii
> 16.29user 0.75system 0:17.13elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
> 0inputs+0outputs (1134major+109771minor)pagefaults 0swaps
> 

OK, I just posted a patch to speed up DCE, which was the culprit here.
On my machine, DCE went from using 1678 seconds to 3.5 seconds. 

This'll probably help all of POOMA, and most C++ programs.

Andrew

(PS, the patch is checked in.  tree-ssa-dce.c)