Exception: clCreateKernel

[Carl Benzer]

Hi,

tried my first steps with OpenMM today. Installation on Ubuntu 12.04 went fine, I can run all examples using the reference platform. To fully use all four cores of my machine, I installed the AMD OpenCL SDK. As a quick test, I ran mandelbulbGPU:

OpenCL Platform 0: Advanced Micro Devices, Inc.
OpenCL Device 0: Type = TYPE_CPU
OpenCL Device 0: Name = Intel(R) Core(TM) i5-2400 CPU @ 3.10GHz
OpenCL Device 0: Compute units = 4
OpenCL Device 0: Max. work group size = 1024

Now, if I try to run the OpenMM examples again, I get errors:

>python testInstallation.py
There are 2 Platforms available:

1 Reference - Successfully computed forces
2 OpenCL - Error computing forces

>python simulateAmber.py
Traceback (most recent call last):
File "simulateAmber.py", line 12, in <module>
simulation.minimizeEnergy()
File "/usr/local/lib/python2.7/dist-packages/simtk/openmm/app/simulation.py", line 89, in minimizeEnergy
mm.LocalEnergyMinimizer.minimize(self.context, tolerance, maxIterations)
File "/usr/local/lib/python2.7/dist-packages/simtk/openmm/openmm.py", line 9420, in minimize
return _openmm.LocalEnergyMinimizer_minimize(*args)
Exception: clCreateKernel

What am I missing?

Thanks
Carl

[Peter Eastman]

Very strange. That's basically an internal error, saying that something went wrong while trying to compile a kernel. I'm not sure what would cause that, so let's start with some high level questions and see if we can track it down.

- Which version of OpenMM are you using?
- Which version of the AMD SDK are you using?
- Are you using a binary distribution of OpenMM, or compiling from source?
- Is your Linux 32 or 64 bit?

Peter

[Carl Benzer]

- Which version of OpenMM are you using?

OpenMM 4.1.1

- Which version of the AMD SDK are you using?

I downloaded and installed AMD-APP-SDK-v2.7-RC-lnx64.tgz

- Are you using a binary distribution of OpenMM, or compiling from source?

Binary distribution.

- Is your Linux 32 or 64 bit?

64bit Ubuntu 12.04

I managed to install and successfully run OpenMM 4.1.1 binary distribution on another machine (CentOS 5.8 32bit) using the NVIDIA CUDA SDK and drivers. This worked out of the box. The machine on which OpenMM doesn't work is a Intel i5-2400 with on-board Intel graphics. I assume to use all four cores I need the OpenCL version, right?

Thanks
Carl