simtk.testInstallation produces segfault

[Joe Napoli]

Hello, I am puzzled by the following error I receive when testing my OpenMM installation:

Code: Select all

There are 4 Platforms available:

1 Reference - Successfully computed forces
2 CPU - Successfully computed forces
3 CUDA - Successfully computed forces
Segmentation fault (core dumped)

I installed the latest version of OpenMM via conda, am using CUDA 7.5.18, and have set

Code: Select all

export OPENMM_CUDA_COMPILER=`which nvcc`

.

Any idea what is happening? Thank you!

-Joe

[Peter Eastman]

Could you verify what it set OPENMM_CUDA_COMPILER to? Just to make sure it's really correct.

What OS and GPU are you using? I assume this is a clean install of OpenMM? Is it possible there's a second copy of it somewhere else on the computer? Problems like this can potentially happen when there are two different copies of OpenMM on the same computer, and Python links to the wrong one.

And just to suggest the obvious, have you tried rebooting?

Peter

[Joe Napoli]

Hi Peter,

Thanks for getting back to me. OPENMM_CUDA_COMPILER is set to

Code: Select all

/global/software/Core/CUDA/7.5.18/bin/nvcc

I am running on Stanford's Xstream cluster, which I believe is configured with Nvidia Tesla K80 GPUs.

As far as I can tell, I only have one copy of openmm on the system (installed via conda), and I don't encounter any issues importing it from ipython:

Code: Select all

Python 2.7.12 |Anaconda 4.1.1 (64-bit)| (default, Jul  2 2016, 17:42:40)
Type "copyright", "credits" or "license" for more information.

IPython 4.2.0 -- An enhanced Interactive Python.
?         -> Introduction and overview of IPython's features.
%quickref -> Quick reference.
help      -> Python's own help system.
object?   -> Details about 'object', use 'object??' for extra details.

In [1]: import simtk.openmm as omm

In [2]: omm.version.full_version
Out[2]: '7.0.1.dev-5e86c4f'

In [3]: omm.version.short_version
Out[3]: '7.0.1'

Thanks,
Joe

[Peter Eastman]

That's odd, I've run OpenMM on Xstream and haven't seen this problem. Maybe try running it in gdb? See if you can get a stack trace of where the segfault is happening.

Peter