OpenMM 6.0.1

[Peter Eastman]

Version 6.0.1 of OpenMM is now available for download. This is a patch release to fix a small number of bugs. In particular, it has fixes for the following:

2005: Missing information from serialized systems
2008: CPU platform segfaults with very few water molecules
2009: Multiple CustomCompoundBondForces on Reference platform produce
wrong results
2010: Integer overflow in long range dispersion correction
2011: Exception in StateDataReporter when reporting speed
2013: PME incorrect when using multiple CUDA or OpenCL devices

I'd encourage anyone using 6.0 (or earlier versions for that matter) to upgrade. Some of these are potentially serious in that they can produce wrong results, although they only occur in what are hopefully uncommon situations.

Peter

[luke czapla]

Could you possibly elaborate on issues with PME on multiple devices and if that is a bug applicable to 6.0? I'd set up some calculations with the 6.0 routines but we didn't use them in publication-intended calculations yet and will upgrade to 6.0.1. We're really interested in using the multiple GPUs w/ PME in the future (perhaps the biggest application of OpenMM) and so if there is any more testing to be done let me know. I'll do what I can from here to continue to verify the methods in the C++ code using the libraries, I've got a 2 GPU (GTX 580) workstation and a couple 4 GPU (GTX680 usually) workstations.

[Peter Eastman]

Hi Luke,

You can find details here:

https://simtk.org/tracker/index.php?fun ... 1&atid=435

If all three of the conditions described in the bug report were met, the forces and energy would be incorrect. So if you're doing multi-GPU PME simulations, you should definitely upgrade to 6.0.1 right away.

Peter

[luke czapla]

Thanks so much for the link and I'm updating the systems. I will let you know if I run into any problems and will proceed cautiously and validate some test runs.

[luke czapla]

I have installed the OpenMM distribution from libraries, but with the source I get a huge string of errors like this (I just copied a few and there's a ton associated with each folder):



CMake Error at tests/CMakeLists.txt:12 (ADD_EXECUTABLE):
add_executable called with incorrect number of arguments


CMake Error at tests/CMakeLists.txt:15 (ADD_TEST):
add_test called with incorrect number of arguments



I was just wondering if there's something wrong with the setup of the CMake files here? The one on "git" builds fine but the OpenMM6.0.1-Source.tar.gz doesn't seem to be set up right.

[Peter Eastman]

There haven't been any changes to that file in a long time. The versions in 6.0, 6.0.1, and the repository head should all be identical. I'd guess it's something to do with how you configured the CMake build?

Peter

[luke czapla]

Thanks Peter, I'm not sure because I unpack the clean OpenMM6.0.1-source.tar.gz and get that string of errors from default "cmake ." in the folder, this is on a Linux system. However, when I grab a clean "git" with the clone and type the same thing everything proceeds fine, and then I just change a couple of options and can build it.

So the current git version has all the updates I'd need to be up to date on the bug fixes? I'm just interested in setting up a test environment for some runs where I verify the PME calculation on multiple GPUs against established code (CHARMM) with single CPU and MPI. My work here is just limited to checking the OpenMM calculations for future development.


(The problem I had mentioned with XML seems solved so I edited this post here.)

Thanks,
Luke

[Peter Eastman]

So the current git version has all the updates I'd need to be up to date on the bug fixes?

Correct.

Peter