Influence of the Sparse Matrix Structure on Automatic Parallelisation Efficiency

The simulated models and requirements of engineering programs like computational fluids dynamics and structural mechanics grow more rapidly than single processor performance. Automatic parallelisation seem to be the obvious approach for huge and historic packages like PERMAS. In this paper we evaluate how preparatory steps on the big input matrices can improve the performance of the parallelisation. We show that a preparatory blocking of the matrix saves storage and decreases the critical path length of the task graph when it is done with variable sized blocks. Also, a data distribution step is proposed that drives the modified dynamic scheduler. Results of this combination show an efficient parallelisation of the programs even on slow multiprocessor networks. Finally, the last step proposed is to interleave the array blocks that are distributed to different processors with post-ordering algorithm. This step is essential to expose the parallelism to the scheduler.