[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference turris::fortran

Title:Digital Fortran
Notice:Read notes 1.* for important information
Moderator:QUARK::LIONEL
Created:Thu Jun 01 1995
Last Modified:Fri Jun 06 1997
Last Successful Update:Fri Jun 06 1997
Number of topics:1333
Total number of notes:6734

1171.0. "Help to Compile and Link an Alpha NT Image?" by CGOOA::OWONG (SKIWI in Canada (VAO)) Mon Feb 10 1997 18:12

    No response from the other conferences.  Anyone in here able to build
    me a .exe for Alpha NT from the following code.  I'll run the program
    on an AlphaStation 500/333 and do a scaling based on the SPECfp95
    ratings.
    
    	Owen.
    
     <<< WRKSYS::SYS_TOOLS:[NOTES$LIBRARY]XL_PERSONAL_WORKSTATION.NOTE;1 >>>
                          -< XL_PERSONAL_WORKSTATION >-
================================================================================
Note 258.0       Help to run FORTRAN Prog on PWS433a or PWS500a?      No replies
CGOOA::OWONG "SKIWI in Canada (VAO)"                 49 lines   6-FEB-1997 14:34
--------------------------------------------------------------------------------
    Crossposted wrksys::xl_personal_workstation, decwet::windows-nt
    
    Could somebody with access to a PWS433a and/or PWS500a try running the
    following Fortran program under Windows NT for me.  It's code that a
    customer has requested be run on the above platforms.  It should
    complete in under 1 minute.  If you could supply the name and version of
    the compiler and Windows NT used and amount of CPU time consumed that
    would be appreciated.
    
    	Owen.
    **************************************************************
C U Vic test program
c Test CPU by doing a simple autocorrelation function calculation.
c
	parameter (n=10000)
c
	real x(n), y(n)
	integer numr(1000)
	do 9 i=1,1000
9	numr(i)=0
	rseed=1234567.
c
c First load up arrays with positions ...
c
	do 1 i=1,n
	x(i)=random(rseed)
1	y(i)=random(rseed)
	write(6,*)'Arrays loaded ...'
c
c Do calculation ...
c
	do 5 i=1,n
	do 5 j=1,n
	r=sqrt((x(i)-x(j))**2+(y(i)-y(j))**2)
	irbin=int(r*100)+1
5	numr(irbin)=numr(irbin)+1
c
	write(6,*)(numr(k),k=1,100)
	write(6,*)'n=',n,' cross-correlation finished.'
	stop
	end

	function random(rseed)
	data rmax/2147483600./, pi/3.14159273/
	random=amod(rseed,rmax)/rmax
	rseed=(random+pi)*rmax
	return
	end
T.RTitleUserPersonal
Name		DateLines
1171.1Bret 400Mhz resultGEMGRP::GROVETue Feb 11 1997 16:4710
    I ran your program on my Alphastation 500/400Mhz (Bret workstation).
    It ran in 25 seconds.
    
    NT Workstation 4.0
    f77 compiler: current development version, not released
    
    One of my colleagues has a PWS433a running NT - we'll run the program
    on it tomorrow morning; she's not here this afternoon.
    
    Rich Grove, GEM compiler group
1171.2433a timesGEMGRP::GROVEWed Feb 12 1997 15:107
    Jeannie Lieb ran the program on a PWS433a (cacheless) and it ran it
    18 seconds.
    
    NT workstation 4.0
    f77 V1.2 the released compiler
    
    /Rich
1171.3Thanks for the helpCGOOA::OWONGSKIWI in Canada (VAO)Wed Feb 12 1997 16:248
    Rich/Jeannie,
    
    Thanks for the help.  I let the customer know what the times were.
    
    FYI - On an existing DEC 3000 Model 500 under OpenVMS that the customer
    currently has it took 61 secs.
    
    	Owen.
1171.4other results from customerCGOOA::OWONGSKIWI in Canada (VAO)Thu Feb 13 1997 01:4417
    Other timings from my customer for the same program.
    
    	Owen.
    
> This benchmark is a simple program that does concentrated
> real*4 number crunching, but virtually no I/O or swapping.
> 
> clam.phys.uvic.ca       SS10/41        327s
> beluga.phys.uvic.ca     SS20/50        301s
> heron.phys              Sparc 5        280s
> loon.phys               Pentium 90Mhz  214s
> phastf.phys.uvic.ca     DEC 3000/500X   66s
> makani.cfht.hawaii.edu  SparcUltra      61s
> ?                       DEC 500/400Mhz  25s
> 
>
1171.5CGOOA::OWONGSKIWI in Canada (VAO)Fri Feb 14 1997 22:516
    Just received an email from my customer.  Based upon the program timings
    they will be replacing existing Sun systems with Alpha.
    
    Thanks again for for the help.
    
    	Owen.
1171.6How about Pentium Pro performance ??BLOFLY::SMITHPBeware the knights who say &quot;NT&quot;...Sat Feb 15 1997 00:1813
    I'd like to see the results for a single processor PPro200 (kauai).  We
    field a lot of questions about the relative performance of Intel vs
    Alpha for heavy-FP-intensive operations.
    
    Does anyone have a DPW200i with Fortran and could compile/run the
    sample code ?
    
    
    Thanks,
    
    
    Peter Smith
    Workstation Sales/Australia
1171.7200i, 200i� and Alcor266CERN::HOBBSCongrats to the Ignoble Peace Prize winner! (http://www.eecs.harvard.edu/ig_nobel)Sat Feb 15 1997 09:2214
Just ran on:

	200i�		default Fortran quals		78.9  sec
	200i�		/fast /automatic /inline:all	68.92 sec
	200i  (*)	/fast /automatic /inline:all	68.88 sec

	600/266		default Fortran quals		27.7  sec
			/fast /automatic /inline:all	24.1  sec
			(tune:ev5 had no effect)

-cw

 (*) Same Kauai, just loaded uni-processor HAL and kernel to convert from
     200i� to 200i
1171.8And 1/3 faster with 4-at-a-time SQRTGEMGRP::GROVESat Feb 15 1997 11:179
    BTW, on Unix we (Fortran, Math Library, GEM) have implemented a
    4-at-time short vector sqrt function. The Unix compiler uses that
    routine in this program and reduces the execution time by about 1/3.
    [From 25 seconds to 17 seconds on the EV5 system I used].
    
    We should get the 4-at-a-time SQRT into the next math library and
    Fortran release for NT as well.
    
    /Rich