Authors: Pascal Iselin & Matthias Untergassmair Lecture: ETH Zürich HighPerformanceComputing II Project, spring semester 2015

Note: this public repository only holds a reduced version of the original project

• make run in the code directory for compiling and running the program with default parameters
• System requirements: GCC compiler, support for AVX, OpenMP, MPI
• Setting parallelization parameters
  • MPI: run the program prefixed by mpirun -np P, where P is the number of MPI processes to run
  • OpenMP: set the corresponding environment variable via export OMP_NUM_THREADS=N where N is the number of OpenMP threads
  • Vectorization: the macro VECTORIZE must be set at compile time via -D VECTORIZE=0 (SIMD off) or -D VECTORIZE=1 (SIMD on).
• visualize graph and solution: after having run a simulation, make viz draws the partitioned graph using graphviz ('www.graphviz.org'). Make sure it is installed on the system.
• Input Arguments:
  • First input argument N: must be an integer between 2 and 32 which determines the number of nodes in the graph NOTE: runtime and memory scale exponentially with the number of nodes!
  • Second input argument R (optional): is the random seed for the graph. In the special case of R equal -1 the graph consists of two maximally connected subgraphs with only one connection. In the special case of R equal -2 the graph is a circular graph. Default random seed is 42.
• benchmark.db is a sqlite3 database that summarizes the benchmark results obtained from timing our runs on the Piz Daint Supercomputer at the CSCS Lugano, the plots directory holds the associated graphs