[MPI3 Fortran] MPI draft 2011-04-21

Hubert Ritzdorf hritzdorf at hpce.nec.com
Fri Apr 29 12:38:10 CDT 2011


I reviewed the draft and I propose only some small corrections to the draft


(*) Page 15 Line 1:
This is done by having two bindings where one has to optional argument and one does not.

This sentence is not really correct since at least functions MPI_Dist_graph_create_adjacent and MPI_Dist_graph_neighbors have
3 bindings since MPI_UNWEIGHTED can be used for 2 arguments.

I would propose to simply remove the sentence. This is an implementation

(*) Page 546, Line 17 and Line 21:
The font of _NOT_ in MPI_SUBARRAYS_NOT_SUPPORTED is too large.

(*) Page 552, Line 23:

Remove "The" after "the" in "In Example, 16.12, the The compiler"

(*) Page 553, Line 17:
applictions -> applications

(*) Page 562, Line 9:
gerneric -> generic

(*) Page 545, Lines 36-40:
MPI will work operate only on the elements s(1), s(6) s(11). Therefore, remove the "etc.," in Line 36 and ",..., s(96)" in Line 40.

(*) Page 558, Line 34:
autamatic -> automatic

(*) Page 559, Line 7
Add a newline before subsubsection "Fortran Support Though the mpif.h Include File"

(*) Page 564, Line 38:
interfces -> interfaces

In addition, I have 2 questions to 

(*) Section Fortran Support Through the mpi_f08 Module and
    Section Fortran Support Through the mpi Module
    "Provide Interfaces for all MPI routines, ..."

Does all include the deprecated MPI routines ?

When looking for aim "compile-time checking", they should be included.
Since no Fortran 2008 interface is defined, it is not required that
they are included. What is right ?

(*) Section 16.2.15 "Fortran Support Through the mpi_f08 Module"

What should item "Declare each argument with an INTENT=IN, OUT, or INOUT as appropriate" mean ?

The MPI 3.0 draft defines all Intents. I would propose to remove this item entirely or to change it into
"Declare each argument with an INTENT as defined in this standard" and remove the rational.

The following 4 items are also defined by each individual MPI interface and could also be removed.

Best regards


More information about the mpiwg-fortran mailing list