runaway uopt with f77 -O2 (was: Re: FORTRAN, format statement

arritt at kuhub.cc.ukans.edu arritt at kuhub.cc.ukans.edu
Tue Nov 27 04:52:07 AEST 1990


In article <4869 at pavo.concordia.ca>, DEMOVMS at pavo.concordia.ca (Concordia University Computing Services) writes:

>      I have the following problem with the fortran compiler.

[details omitted...]
 
>     When I have this inside a program, and complile with 'f77 -O2'
> At execution I get a 'Segmentation fault' (or a bus error if the
> output is a variable). Without the '-O2' option the program runs fine.

I've been having a different problem with f77 -O2, after conversion to
IRIX 3.3.    There is some kind of odd runaway behavior of -O2 
with one of my codes: the uopt process grows to a gigantic 
size (the biggest I recall is about 20 MB) and then just sits 
there and thrashes the disk, getting about 2 minutes of CPU in half an hour
of wall-clock time (with nothing else competing for the CPU).

This particular code is a setup for NCAR Graphics 3.0.  It is about 1000
lines including blank and comment lines (perhaps 500 executable statements).
Under IRIX 3.2, f77 -O2 worked fine.  Under 3.3, f77 -O1 also does 
just dandy.  There are no compiler error messages in any event.

I have not observed this behavior with any other codes, including some that
are considerably longer and more complicated in structure (i.e. more badly
written).

This is not directly related to the original posting but since someone else
has been having problems with f77 -O2, I thought to mention it.

Any ideas?
________________________________________________________________________
Raymond W. Arritt                     | 
Assistant Professor                   |
Dept. of Physics and Astronomy        |  "everyone knew that as time went
Univ. of Kansas                       |   by they'd get a little bit older
Lawrence, KS  66045                   |   and a little bit slower..."
arritt at kuhub.cc.ukans.edu             |               
arritt at ukanvax.bitnet                 |
                               



More information about the Comp.sys.sgi mailing list