[MPI3 Fortran] What to do with mpif.h in MPI-3?

N.M. Maclaren nmm1 at cam.ac.uk
Tue Jan 26 03:17:09 CST 2010

On Jan 26 2010, Malcolm Cohen wrote:
> As for "INTEGER*8", even if an F77-era Fortran compiler accepted the 
> syntax (many did, but quite a few did not) there was not universal 
> agreement on semantics; plus few 16-bit machines had big integers at all, 
> and the 18-bit, 24-bit, 36-bit, 48-bit and 60-bit machines quite often 
> didn't do any power-of-two-sized integers. INTEGER*3 and INTEGER*6 were 
> fun though.

On at least some systems, REAL*3 and INTEGER*3 were the third-smallest
integer and real representations, and not the same size (or with sizes
a multiple of 3).

Jeff is, of course, right.  INTEGER(KIND=MPI_COUNT_KIND).  If it is felt to
be a good idea to add to mpif.h, then the obvious solution has already been
posted (I forget by who).  That is a separate set of names, with (say) _c
appended to the existing ones, and an implementation-defined integer type of
count argument.  Whether that should be done is a moot point.

Nick Maclaren.

More information about the mpiwg-fortran mailing list