[Q-e-developers] 'mpif.h' vs. 'include mpi' and regression tests

[Davide Ceresoli]

On 11/22/2016 02:10 PM, Paul Kapinos wrote:
> Dear Quantum Espresso developer,
> We (try to) install QE on our computers in a great variety of version (Intel MPI
> and Open MPI; Intel, GCC (and PGI and Oracle Studio) compilers; different
> versions of) Doing this we see at least two issues we do let you know about:
>
> I. we cannot compile using Intel compiler version 17 and appropriate Open MPI
> version. (older versions can compile but raise warnings; using Intel MPI neiter
> error nor warnings are raised - likely due to very different 'mpi.h' file).
>
> Affected file:
> qe-6.0/FFTXlib/stick_base.f90
Dear all,
     I also ran into this problem yesterday. What was a 'warning' in
Intel 2015 and 2016, has become an 'error' in Intel 2017. I've tried
to turn off this specific error on the command line, but it didn't
work. Switching include 'mpif.h' to use mpi solves the problem.

> It turned out that this is likely a header clash due to multiple include's, cf.
> https://www.mail-archive.com/devel@lists.open-mpi.org//msg19768.html
>
> The recommendation of  Gilles Gouaillardet is to go to 'use mpi' syntax, cf.
> https://www.mail-archive.com/devel@lists.open-mpi.org//msg19776.html
> http://mpi-forum.org/docs/mpi-3.1/mpi31-report/node411.htm
>

Best,
     Davide