Running the Test Program

The source code distribution comes with several simple test programs that will run the DPMTA code with a variety of parameters and randomly generated points. Timing results are provided as output, along with optional output of the particle and force data. This is useful for benchmarking the code performance on a variety of platforms, as well as providing examples of the DPMTA calling structure.

The user is responsible for installing and invoking the PVM system prior to attempting to execute the DPMTA program.

The program has the following command line options

dpmta_test Procs Levels Parts FFT Terms Theta Iterations PBC

Procs

The number of slave processes that should be created. As of DPMTA version 2.5, his value may be any number. In versions previous to 2.5 this number was limited to a power of two.

Levels

The number of levels used for spatial decomposition. Reasonable values are from four to six.

Parts

The number of particles to be generated. Currently the code generates a uniform random particle distribution. You have no choice right now.

FFT

A flag to indicate if Fast Fourier Transforms should be used to improve performance. The FFT option introduces restriction on legal values on the Terms parameter, as described in Section 3.5.

Terms

the number of terms in the multipole expansion that should be retained.

Theta

is the separation ratio for the multipole acceptance criteria (MAC). It should be between 0 and 1, although optimal values lie between 0.5 and 0.75.

Iterations

the number of iterations of the complete force calculations to perform on the data.

PBC

is a flag to indicate if the particle interactions should include periodic boundary conditions.

There is also a test program, dpmta_direct that performs the N-body computation directly. It uses the output particle file created by dpmta_test when the DATADUMP compile flag is turned on.

The program dpmta_error can be used to compute the resulting errors between the output files generated by the DPMTA test and the direct computation.