> Given that g77 doesn't include /usr/include on the default search path > either: > > [EMAIL PROTECTED]:/tmp$ g77 tst.f > tst.f:1: > include 'mpif.h' > ^ > Unable to open INCLUDE file `mpif.h' at (^),
Hm, that's surprising. However, despite g77 not being able to include
mpif.h, both lam and mpich mpif77 can (on the same system):
~/tmp> mpif77.lam test.f && echo fine
fine
~/tmp> mpif77.mpich test.f && echo fine
fine
~/tmp>
This is inconvenient, since it makes altering the MPI implementation
tricky. Also, it is against (de-facto) standard. The following compilers
all compile the code I tested this without even a warning: IBM's mpxlf,
Sun's mpf77, PGI's pgf77, Pathscale's pathf90 (they don't have f77 as its
own binary) and all versions of mpif77.{lam,mpich} since around year
2000.
I wonder how this can pass unnoticed, do people usually write full paths
to include files (which is *very* inconvenient when you compile and run
the programs on five totally different systems) or is fortran never
supposed to look for includes anywhere or have all the other compilers
just been very friendly in being able to find mpif.h by themselves?
> this is definitely not Severity: Grave and I'm not even sure
> whether it should be altered; that's up to the normal maintainer.
Fine, our people'll stay away from openmpi then. =(
-Juha
--
-----------------------------------------------
| Juha Jäykkä, [EMAIL PROTECTED] |
| Laboratory of Theoretical Physics |
| Department of Physics, University of Turku |
| home: http://www.utu.fi/~juolja/ |
-----------------------------------------------
signature.asc
Description: PGP signature
