This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Performance of Integer Multiplication on PIII (Results forgcc-2.95& Athlon)
- To: Paolo Carlini <pcarlini at unitus dot it>
- Subject: Re: Performance of Integer Multiplication on PIII (Results forgcc-2.95& Athlon)
- From: pete at ltoi dot iap dot physik dot tu-darmstadt dot de
- Date: Mon, 5 Nov 2001 21:02:08 +0200 (MEST)
- Cc: jh at suse dot cz, <gcc at gcc dot gnu dot org>
In advance of the question {Which one really matters?}
Base & -fomit-frame-pointer -mpreferred-stack-boundary=2
yields:
Loop: 1.18, Code: 2.57
Clocks: 19.15
Base speed was -O2 -march=i686
Loop: 1.93, Code: 2.57
Clocks: 19.15
Oops, live is different on this box!
This P6 Model II (dual) box, BX-Chipset, uses an unmodified
SuSE-6.1, i.e. glibc-2.0.7
now, for the original read.c:
hgcc -s -o imul imul.c read_empty.c read.c -march=i686 -O2
Loop: 1.92, Code: 3.12
Clocks: 23.24
Base & -fomit-frame-pointer -mpreferred-stack-boundary=2
Loop: 1.18, Code: 3.31
Clocks: 24.66
... and the athlon setting no longer works here ...
hgcc -s -o imul imul.c read_empty.c read.c -march=i686 -O2
-fomit-frame-pointer -malign-loops=2 -malign-jumps=2
-mpreferred-stack-boundary=2
Loop: 1.18, Code: 3.30
Clocks: 24.58
... There is no way ...
Well, we may continue to fiddle with the options, but what really is
needed, is a "theory" about, how fast it should be ...
Peter