Re: [AMBER] 'undefined reference to MPI::...' error messages when compiling Amber 12 on CentOS 6 for "-mpi -cuda"

From: Tru Huynh <>
Date: Fri, 5 Jul 2013 14:31:45 +0200

On Fri, Jul 05, 2013 at 11:45:07AM +0200, Frank Thommen wrote:
> Hi,
> when doing `make install` (after `./configure -mpi -cuda gnu`) on our
> cluster (LSF running on CentOS 6.2), I'm getting the following error
> (only the lines for the last compilation block are shown):

If the LSF MPI version is based on OpenMPI, you can try to fix your config.h
PMEMD_CU_LIBS=./cuda/cuda.a -L$(CUDA_HOME)/lib64 -L$(CUDA_HOME)/lib -lcurand -lcufft -lcudart
changed to:
PMEMD_CU_LIBS=./cuda/cuda.a -L$(CUDA_HOME)/lib64 -L$(CUDA_HOME)/lib -lcurand -lcufft -lcudart -lmpi_cxx

and run make once more.

> A better readable version of the output is available on
> This is on CentOS 6.2 (kernel 2.6.32-220.el6.x86_64) on an LSF 7.0.6
> cluster using the MPI implementation which comes with LSF. I had to add
> "-I/opt/platform_mpi/include" to the NVCC line in config.h to make it
> work at all.
> The compilation of serial, MPI and CUDA versions alone work find, it's
> only the combined CUDA/MPI compilation that fails.
the mpi_cxx trick should have been required for the plain MPI version, no other idea
except rebuilding your own mpi version.


Dr Tru Huynh          | | tel/fax +33 1 45 68 87 37/19
Institut Pasteur, 25-28 rue du Docteur Roux, 75724 Paris CEDEX 15 France  
AMBER mailing list
Received on Fri Jul 05 2013 - 06:00:04 PDT
Custom Search