[MPI3 Fortran] Proposed solution for the volatile buffer problem

Aleksandar Donev donev1 at llnl.gov
Wed Jan 14 16:21:25 CST 2009


> But that degrades performance for any procedure that calls another
> procedure!
But the converse, requiring every procedure that calls MPI_Irecv to be 
flagged, seems even worse (not good for performance and horrible to the 
programmer). Good argument against the idea of a procedure attribute it 
would seem.

> Secondly, a compiler is allowed to assume that unreferenced parts of an
> array are invariant, and so may copy more than it needs to - which can
> sometimes be a lot faster.
OK, that's a good point in argument of explicitly marking variables that 
get changed behind the scenes instead of relying on some form of 
implicit assumptions.


More information about the mpiwg-fortran mailing list