10. Platform-Specific Properties

When creating a Context, you can specify values for properties specific to a particular Platform. This is used to control how calculations are done in ways that are outside the scope of the generic OpenMM API.

To do this, pass both the Platform object and a map of property values to the Context constructor:

Platform& platform = Platform::getPlatformByName("OpenCL");
map<string, string> properties;
properties["DeviceIndex"] = "1";
Context context(system, integrator, platform, properties);

After a Context is created, you can use the Platform’s getPropertyValue() method to query the values of properties.

10.1. OpenCL Platform

The OpenCL Platform recognizes the following Platform-specific properties:

  • Precision: This selects what numeric precision to use for calculations. The allowed values are “single”, “mixed”, and “double”. If it is set to “single”, nearly all calculations are done in single precision. This is the fastest option but also the least accurate. If it is set to “mixed”, forces are computed in single precision but integration is done in double precision. This gives much better energy conservation with only a slight decrease in speed. If it is set to “double”, all calculations are done in double precision. This is the most accurate option, but is usually much slower than the others.

  • UseCpuPme: This selects whether to use the CPU-based PME implementation. The allowed values are “true” or “false”. Depending on your hardware, this might (or might not) improve performance.

  • OpenCLPlatformIndex: When multiple OpenCL implementations are installed on your computer, this is used to select which one to use. The value is the zero-based index of the platform (in the OpenCL sense, not the OpenMM sense) to use, in the order they are returned by the OpenCL platform API. This is useful, for example, in selecting whether to use a GPU or CPU based OpenCL implementation.

  • DeviceIndex: When multiple OpenCL devices are available on your computer, this is used to select which one to use. The value is the zero-based index of the device to use, in the order they are returned by the OpenCL device API.

The OpenCL Platform also supports parallelizing a simulation across multiple GPUs. To do that, set the DeviceIndex property to a comma separated list of values. For example,

properties["DeviceIndex"] = "0,1";

This tells it to use both devices 0 and 1, splitting the work between them.

10.2. CUDA Platform

The CUDA Platform recognizes the following Platform-specific properties:

  • Precision: This selects what numeric precision to use for calculations. The allowed values are “single”, “mixed”, and “double”. If it is set to “single”, nearly all calculations are done in single precision. This is the fastest option but also the least accurate. If it is set to “mixed”, forces are computed in single precision but integration is done in double precision. This gives much better energy conservation with only a slight decrease in speed. If it is set to “double”, all calculations are done in double precision. This is the most accurate option, but is usually much slower than the others.

  • UseCpuPme: This selects whether to use the CPU-based PME implementation. The allowed values are “true” or “false”. Depending on your hardware, this might (or might not) improve performance.

  • CudaCompiler: This specifies the path to the CUDA kernel compiler. Versions of CUDA before 7.0 require a separate compiler executable. If you do not specify this, OpenMM will try to locate the compiler itself. Specify this only when you want to override the default location. The logic used to pick the default location depends on the operating system:

    • Mac/Linux: It first looks for an environment variable called OPENMM_CUDA_COMPILER. If that is set, its value is used. Otherwise, the default location is set to /usr/local/cuda/bin/nvcc.

    • Windows: It looks for an environment variable called CUDA_BIN_PATH, then appends nvcc.exe to it. That environment variable is set by the CUDA installer, so it usually is present.

  • TempDirectory: This specifies a directory where temporary files can be written while compiling kernels. OpenMM usually can locate your operating system’s temp directory automatically (for example, by looking for the TEMP environment variable), so you rarely need to specify this.

  • DeviceIndex: When multiple CUDA devices are available on your computer, this is used to select which one to use. The value is the zero-based index of the device to use, in the order they are returned by the CUDA API.

  • UseBlockingSync: This is used to control how the CUDA runtime synchronizes between the CPU and GPU. If this is set to “true” (the default), CUDA will allow the calling thread to sleep while the GPU is performing a computation, allowing the CPU to do other work. If it is set to “false”, CUDA will spin-lock while the GPU is working. Setting it to “false” can improve performance slightly, but also prevents the CPU from doing anything else while the GPU is working.

  • DeterministicForces: In some cases, the CUDA platform may compute forces in ways that are not fully deterministic (typically differing in what order a set of numbers get added together). This means that if you compute the forces twice for the same particle positions, there may be tiny differences in the results. In most cases this is not a problem, but certain algorithms depend on forces being exactly reproducible to the last bit. If you set this property to “true”, it will instead do these calculations in a way that produces fully deterministic results, at the cost of a small decrease in performance.

The CUDA Platform also supports parallelizing a simulation across multiple GPUs. To do that, set the DeviceIndex property to a comma separated list of values. For example,

properties["DeviceIndex"] = "0,1";

This tells it to use both devices 0 and 1, splitting the work between them.

10.3. CPU Platform

The CPU Platform recognizes the following Platform-specific properties:

  • Threads: This specifies the number of CPU threads to use. If you do not specify this, OpenMM will select a default number of threads as follows:

    • If an environment variable called OPENMM_CPU_THREADS is set, its value is used as the number of threads.

    • Otherwise, the number of threads is set to the number of logical CPU cores in the computer it is running on.

    Usually the default value works well. This is mainly useful when you are running something else on the computer at the same time, and you want to prevent OpenMM from monopolizing all available cores.

10.4. Determinism

Whether a simulation is deterministic will depend on what platform you run on in addition to what settings/methods you use. For instance, as of this writing, using PME on the Reference, OpenCL, and double-precision CUDA will result in deterministic simulations. Single-precision CUDA and CPU platforms are not deterministic in this case. However, none of this behavior is guaranteed in future versions. In many cases it will still result in an identical trajectory. If determinism is a critical for your needs, you should carefully check to ensure that your settings and platform allow for this.