Date: Sun, 2 Mar 1997 12:29:36 -0800 (PST) From: "Eric J. Schwertfeger" <ejs@bfd.com> To: freebsd-chat@freebsd.org Subject: GCC options for PPro, was Re: RSA 56-bit key challenge Message-ID: <Pine.BSF.3.95.970302121755.25325E-100000@harlie.bfd.com> In-Reply-To: <199703021950.LAA28043@freefall.freebsd.org>
next in thread | previous in thread | raw e-mail | index | archive | help
Actually, this has brought up something very interesting. I stated that
compiling your own client was faster for a Pentium Pro, and was
contradicted by Charles Henrich. Well, it turns out that GCC 2.6.3 from
2.1.6-RELEASE turns out a much faster PentiumPro executable than does
2.7.2 from 2.2 Beta. I tested this by compiling a static client under
both, then shipping them off to a 2.1.7 based PPro. I suspect that one of
the options below is the cause of the problem, but why the dramatic
difference? (281Kkeys/sec vs 160Kkeys/sec). My first guess is that one of
the options wasn't implemented in 2.6.3, but that should give an error.
So maybe one of them is implemented differently. -O6 changed, but I don't
know about any of the others.
Now, this concerns me just a little, as I'm expecting delivery of my PPro
for my home machine, which is running 2.2, tomorrow. Does anyone have any
experience at tuning gcc 2.7.2 for PPro CPUs?
Here's a list of what options the rc5 client uses. Which ones are wrong
for the P6?
OPTS_GCC=\
-fcaller-saves -fcse-follow-jumps -fcse-skip-blocks \
-fexpensive-optimizations -fomit-frame-pointer \
-frerun-cse-after-loop -fstrength-reduce -fthread-jumps \
-funroll-loops -O6 -Wall
Furthermore, which list would this tangent be more appropriate for?
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.BSF.3.95.970302121755.25325E-100000>