This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: m68k/regclass/regmove fun
- To: law at cygnus dot com, egcs-patches at egcs dot cygnus dot com
- Subject: Re: m68k/regclass/regmove fun
- From: Toon Moene <toon at moene dot indiv dot nluug dot nl>
- Date: Mon, 05 Apr 1999 16:57:06 +0200
- Organization: Moene Computational Physics, Maartensdijk, The Netherlands
- References: <6238.922869228@upchuck>
Jeffrey A Law wrote:
> Well, I was hoping to get some data, particularly from a benchmark other
> than daxpy that what I was doing isn't absurd :-) I'm pretty sure the daxpy
> loop will improve :-)
Well, I finally got around testing this. On a simple (750-line)
relaxation type CFD code (from someone in France) that I tend to use for
this quick-'n-dirty sort of benchmarking, I saw a whopping improvement
from 85.8 -> 85.6 seconds (using compiler options -O2 -funroll-loops
-fomit-frame-pointer).
When comparing the assembly code from both compilations, I noticed that
indeed, in about a dozen cases, the new code managed to use fmove's
instead of "move d0,(sp)-; move d1,(sp)-; fmoved (sp)+,fp0". So the
code seems to do the right thing. Alas, unless it happens in a tight
loop like daxpy's, the win is small.
So I still think the prize for this year's egcs improvement still goes
to Jim :-)
--
Toon Moene (toon@moene.indiv.nluug.nl)
Saturnushof 14, 3738 XG Maartensdijk, The Netherlands
Phone: +31 346 214290; Fax: +31 346 214286
g77 Support: fortran@gnu.org; egcs: egcs-bugs@cygnus.com