Optimization of finite element codes
A general algorithm for constructing finite element matrices within a multiple input, multiple data stream (MIMD) vector-processing environment is presented. Efficiency of the vectorized code is determined by the number of elements which differs from the more intuitive algorithms based on the number of quadrature points or shape functions. Performance is evaluated analytically and then verified by timings obtained by a series of experimental runs on a Cray Y-MP. A speedup factor of 25 is observed. © 1990 Plenum Publishing Corporation.