Parallelization of Direct Linear Equation Solver on Shared Memory Computer

摘要

This paper views the parallelization of direct solver for linear systems of equation that often seen in com- putational scientific simulation on shared memory type parallel computer. Then describe as example the parallelization of Skyline solver in FEM structural analysis code SAPA and apply some idea that convert from fain grain to coarse grain parallelism. Using HP/CONVEX SPP parallel computer, I got 7.3 times speed up on 8 cpus for 11 thousand nodes and show high efficiency of parallel direct solver that was usually considered difficult.