Re: [AMBER] MPI cuda compilation error

From: case <>
Date: Sat, 2 Jul 2011 09:35:51 -0400

On Sat, Jul 02, 2011, Ross Walker wrote:

> >> In file included from gpu.h:15,
> >>
> >> from
> >> gputypes.h:24:17: error: mpi.h: No such file or directory

> Have you tried this with AmberTools 1.4? If not can you try that please
> and let me know if it works?
> I suspect someone broke the configure script in AmberTools 1.5 and did
> not test it properly for complete builds of AMBER.
> I still use AmberTools1.4 and have never seen this problem.

This is not that helpful a suggestion:

1. We don't distribute old versions of AmberTools any more (except for version
1.2, needed for Amber10).

2. The problem report did not report the actual command that led to
the error, nor the value of the PMEMD_CU_INCLUDE variable. Failure
to find mpi.h generally means that there is some problem with the MPI
installation. I still recommend that people having problems with MPI use
the "configure_mpich2" script in $AMBERHOME/AmberTools/src. That ensures
that the same compiler is used as for Amber, that the include, binary
and library files are in the correct place, and that good configuration
options have been used.

3. The current git repo works fine, and AmberTools 1.5 does as well (at
least did so when we were testing it back before the release). Amber
developers are *strongly* encouraged to use up-to-date tools, as that is
the best way to find and correct problems.

Note that there was this update from Jason:

commit 26db9611defd2f7fd36044ecc051d62619813a11
Author: Jason Swails <>
Date: Tue Mar 29 20:31:38 2011 -0700

    cuda_parallel builds break unless nvcc knows where to find the mpi.h.
    Turns out MPI_HOME is still necessary. However, in the spirit of maintainin
    its eradication, I'll 'fake' it by looking at the location of mpif90.
    Those that still wish to set it can do so.

This could cause a problem if MPI_HOME is not set and mpif90 is in an odd
location. But the best thing is to get more information from the original
poster...just having a few more lines of the error message would have
clarified a lot.


AMBER mailing list
Received on Sat Jul 02 2011 - 07:00:03 PDT
Custom Search