NetNews Usenet Archive 1992 #27/NN_1992_27.iso/spool/comp/sys/sgi/16534

Internet Message Format | 1992-11-17 | 1.8 KB

Path: sparky!uunet!usc!zaphod.mps.ohio-state.edu!uwm.edu!ogicse!psgrain!hippo!shannon!news
From: [e-mail] (Philip Machanick)
Newsgroups: comp.sys.sgi
Subject: Re: Weird CPU time ratio between R4000 and R3000
Message-ID: <[e-mail]>
Date: 17 Nov 92 10:30:06 GMT
Article-I.D.: shannon.1992Nov17.103006.3667
References: <[e-mail]>
Sender: [e-mail]
Organization: Computer Science Dept, Wits University
Lines: 26
X-Xxdate: Tue, 17 Nov 92 11:27:39 GMT
X-Useragent: Nuntius v1.1.1d12

In article <[e-mail]> Dave Olson,
[e-mail] writes:
>It is possible that compiling with -mips2 on the r4k will help, if
>you are memory traffic bound, and are using double precision.  Of course,
>that binary won't run on r3k machines any more.
>
>Running osview during the execution may be interesting, as might
>timex -s program.  pixie may also help spot the bottlenecks.

Another R4000 wierdness.

I was helping someone work out the cause of high system time on FORTRAN
R4000 code. I was only in on a small part of the entire exercise but here
are some observations. For some reason loops containing a series of very
complex array references and function calls all in one big assignment
statement many lines long caused a very large amount of system time.
3 such loops contributed to system time that was nearly 30% of user CPU
time. Simplifying the loops by splitting the big assignment apparently
radically reduced system time.

This behaviour was mainfested on an R4000 Indigo and a Crimson, but not
an R3000 machine.
--
Philip Machanick
Computer Science Dept, Univ of the Witwatersrand, 2050 Wits, South Africa
[e-mail]  phone: [phone]  fax: [phone]