Re: why -fno-globals at -O3 with g77

[Toon Moene <toon at moene dot indiv dot nluug dot nl>]

Jack Howarth wrote:
>    A molecular dynamics package, which extensively uses
> shared libs and COMMON blocks, has apparently needed to
> compile certain files with -fno-globals to build at -O3
> with g77 for sometime now. The current g77 3.4 snapshot
> of 101503 seems to reduce these down to about 6 fortran
> source files. These files will only compile at -O3
> when -fno-globals is used. 
>    I haven't seen this issue mentioned and was wondering
> what the origin of it typically was and if there was
> an easy way to avoid tickling this glitch. I ask because
> when I removed -fno-globals from all the uneffected source
> file compilations the resulting binaries ran about 10%
> faster than before.
>    Thanks in advance for any insights into this issue.

Until now, I was mostly silent on your bug reports, because I didn't 
understand most of them (OS X, Apple version of gcc/binutils - lets face 
it: Debian runs quite well on Apple hardware).

However, this one I think I have a handle on: If you need -fno-globals 
it means that you are trying to inline routines that have arguments that 
differ (in type) between the call and the definition site.

This often happens if you have a "generic container" which is INTEGER in 
the routine itself, but in the call might be REAL or DOUBLE PRECISION or 
even COMPLEX if space allows.

There's no easy solution here, because Fortran 77 doesn't allow you to 
write generic routines to handle different type arguments - you need 
Fortran 90/95 for this.

Unfortunately, g95 (see my signature) isn't ready yet ...

--
Toon Moene - mailto:toon@moene.indiv.nluug.nl - phoneto: +31 346 214290
Saturnushof 14, 3738 XG  Maartensdijk, The Netherlands
Maintainer, GNU Fortran 77: http://gcc.gnu.org/onlinedocs/g77_news.html
GNU Fortran 95: http://gcc.gnu.org/fortran/ (under construction)