schreiber@schreiber.asd.sgi.com (Olivier Schreiber) (12/11/90)
I am a confused about what timing should be used for running the
livermore loops:
C LATEST KERNEL MODIFICATION DATE: 22/DEC/86
C LATEST FILE MODIFICATION DATE: 6/SEP/90 version mf447
Some of the directions included with the source code :
C 5. Installation includes verifying or changing the following:
C
C Second: the definition of function SECOND for CPU time only, and
C the value of TIC:= minimum cpu clock time(sec) in SIZES.
c------------------------------------------------------------
c This is a sample version of the DSECOND function.
c It differs from the SECOND function only in that
c it is double precision.
c------------------------------------------------------------
c**********************************************
double precision function second( oldsec )
c***********************************************
implicit double precision (a-h,o-z)
dimension xtime(4)
c...If no adjustment is necessary, leave 'fudge' at 1. Otherwise,
c...modify 'fudge' until the calibration program is accurate.
fudge = 1.0 ! No adjustment necessary.
xt = etime( xtime)
second = xt * fudge
return
end
seem to suggest using user cpu time but then paralellizing the benchmark
would not show improvements since at best cpu time remains the same
or increases due to overhead.
I think most system provide only an aggregate user time through
the etime function first argument.
Should the timing function then be real elapsed time?
Thanks in advance for the help.
--
Olivier Schreiber schreiber@asd.sgi.com Tel(415)335 7353
Technical Marketing MS 7L580
Silicon Graphics Inc., 2011 North Shoreline Blvd. Mountain View, Ca 94039-7311