From: Peter Smulders (p.j.m.smulders@phys.rug.nl)
To: templon@studbolt.physast.uga.edu (Jeffrey Templon)
Subject: Re: Your personal silly benchmark
Date: Sat, 21 Mar 1998 11:45:09 +0100 (MET)

Ik heb intussen jouw benchmark ook op mijn Linux computer gedraaid,
resulttat zie onder. Begrijp jij waarom ik die 189 seconden niet
kan reproduceren?

Groeten,

-- 
Peter

================= brem benchmark on Pentium 233MMX ================
Intel Pentium 233MMX 512kb cache 32Mb memory
Linux slackware  2.0.29 
absoft (precompiled brem_absof from distribution)     : 182.6 seconds
pg77   (precompiled brem_pg from distribution)        : 189.4 seconds
 g77 -O3 -malign-double -fno-f2c (egcs-1.0.1 release) : 203.4 seconds
pg77 -O6 -malign-double -fno-f2c (egcs-1.0.1 release) : 204.8 seconds
pg77 -O6 -malign-double -mstack-align-double \
         -marg-align-double -funroll-all-loops        : 200.7 seconds
-- 
Note: I cannot reproduce the 189.4 s from the precompiled pg77
 version. Did pg77 become slower, or am I not using the proper
 compiler flags?

================== brem benchmark on HP C180 ======================
HP-UX 10.20 HP C180 workstation
what /opt/fortran/lbin/f77pass1:
	 HP-UX SLLIC/OPTIMIZER UX.10.20.351 (DAVIS): 05/21/96
	 Ucode Code Generator - UX.10.20.05 (BCK_UX10.DAVIS_960521)
	HP-UX f77pass1 960522 (083850)  B3906AA/B3908AA B.10.20.01
	 High Level Optimizer - UX.10.20.960521 (DAVIS) [-DHLO_RELEASE +O3] - 22-May-96.08:16
	 /usr/lib/libc: $Revision: 76.3 $

"aggressive" optimization ( +Oall ) : 116.85 seconds

"profile based" optimization :        105.5  seconds
-- 
Note: This is a two-stage process where the program is compiled in
 a special mode, then during run time it obtains data about the
 program flow, which are used in the second compilation stage.
 See script below

=================== script to obtain the last result ========
: Bourne shell script
# make 'profile optimized' version of brem
# hpux version 10.20 specific

rm -f *.o flow.data thin.out opt.thin.out brem brem.opt

FFLAGS="+E1 +E4 +I +Oall" 
LFLAGS="+I -Wl,-a,archive +Oall"
fort77 -c $FFLAGS brem_f2c.f gammln_f2c.f
fort77 $LFLAGS brem_f2c.o gammln_f2c.o -o brem

brem thin.out

LFLAGS="+P -Wl,-a,archive -s +Oall +pgm brem"
fort77 $LFLAGS brem_f2c.o gammln_f2c.o -o brem.opt

time brem.opt opt.thin.out