gravishanker@eagle.wesleyan.edu (03/29/91)
Hi I have a general question on using any of the Optimization options for f77 on SG Irises. We have several programs used for molecular modelling that were written in old Fortran IV style and they all work great with -O option and work fine. We also have some recent programs written to run on a VAX. Strangley most of these programs do not run correctly if -O is used and the only way they can be made to behave the same way as on a VAX is to use -g. We know that these programs mostly assume that the variables local to subroutines are saved and used in subsequent calls, a property that is the default in VAX/FORTRAN. We use -static option to emulate this on our Irises along with -O, but no luck. We have not seriously worried about optimizing these programs because they run fast anyway. However, we are at a point now where we want to look into why is this happening. When we called the SG hotline, after various questions the person wanted to see our code. These programs have several hundred lines of code and it is not going to help much. What I would like to know is if there are any general troubleshooting guidelines one can follow to locate the problems. Also, if there are others who convert their VAX/FORTRAN programs over to the Irises who have encountered similar problems and have found out solutions. Thanks in advance for any help. Ravi
rosalia@max.physics.sunysb.edu (Mark Galassi) (04/03/91)
In article <1991Mar28.220805.40858@eagle.wesleyan.edu> gravishanker@eagle.wesleyan.edu writes: >of these programs do not run correctly if -O is used and the only way they can >be made to behave the same way as on a VAX is to use -g. We know that these You don't have to use -g. That really slows things down. You can use -O0 to avoid all optimization and still not include all the debugging stuff. One of our users here had exactly the same problem, but with a rather short program, so he bugged sgi until they conceded him a bug #. They have not yet sent us a fix after a couple of months, though. I guess we will have to call them up. Meanwhile, and this might be relevant, our MIPS computer (same chip, same compilers) got an upgrade, and the release notes point out several fixes in the f77 optimizer. One of them looked like this: #187 (or something) Serious. Wrong results in addition using f77 optimizer. That's a pretty bad error, what? SGI should really send out the same upgrade, or whatever it is that they do with new versions they get from MIPS. -- {These opinions are mine, and should be everybody else's :-)} Mark Galassi rosalia@dirac.physics.sunysb.edu rosalia@sunysbnp.BITNET