Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!usc!apple!mips!hal!mark
From: mark@mips.COM (Mark G. Johnson)
Newsgroups: comp.benchmarks
Subject: Re: (Useless)  X  (Meaningless)   {digits of PI}
Message-ID: <45474@mips.mips.COM>
Date: 1 Feb 91 23:42:39 GMT
References: <45469@mips.mips.COM>
Sender: news@mips.COM
Reply-To: mark@mips.COM (Mark G. Johnson)
Lines: 24


  > Sun     SPARCStation 1+        9.5 real     9.5 user     0.0 sys
  > Sun     SPARCStation 2        29.4 real    29.3 user     0.0 sys <*** huh?
  > MIPS    Magnum workstation     3.0 real     2.8 user     0.0 sys
  >
  >Note the very anomalous result on SS-II; I tried the code on 3 different
  >machines and kept getting the same answer.  Replication/refutation of
  >this result would be much appreciated.


Many thanks to Earl Killian; this has been explained.  It is a function
of the way the mpy and div subroutines are linked.  With command line
options you can alter the linking.  Unfortunately, the most straightforward
way also gave the longest runtime :-(.  The third trial below takes
advantage of the fact that the C source code #include's math.h but
doesn't actually use it.

  Compiler invocation:  cc -O prog.c -lm            29.3 user seconds
  Compiler invocation:  cc -Bstatic -O prog.c -lm    5.2 user seconds
  Compiler invocation:  cc -O prog.c                 5.5 user seconds
-- 
 -- Mark Johnson	
 	MIPS Computer Systems, 930 E. Arques M/S 2-02, Sunnyvale, CA 94086
	(408) 524-8308    mark@mips.com  {or ...!decwrl!mips!mark}