Parallelizing computation of elastodynamic response on arbitrary domains using GPU

The cellular automata (CA) approach for modeling wave propagation in elastic media presented by Georgia Tech researchers is very promising. The software package (JCA for Java cellular automata) they developed to simulate the wave propagation, takes a mesh file, runs a simulation, and outputs results. However, the processing time for large mesh file is extremely high. CUDA assisted parallel processing of finite element method (FEM) shows potential to analyze wave propagation in computational solid mechanics. In this work, we propose GPU assisted CA approach for computing elastodynamic response on different fields. The elastodynamic CA method is discontinuous like FEM and thus accurately captures propagating wave fronts. The CA technique lends itself to parallelization because there is no need to store a large sparse matrix. Experiments show that for 26K mesh size and 10000 iterations, GPU assisted solution gains speed up more than 573x over the JCA implementation.