ABSTRACT
Commercial finite-element software systems are used extensively in the industrial solution of engineering analysis problems. Static and dynamic analysis of structures with finite-element systems is widespread. This chapter explores the generic principles applied in the implementation of finite element systems on supercomputers. Vectorization, parallelization, and other techniques to take advantage of the very large supercomputer memories are summarized. The vectorizing capability of vectorizer compilers is also very important, and hence they are regularly applied in the process of porting a finite-element program to a supercomputer. State-of-the-art commercial finite element programs usually run on a number of supercomputers. Major vectorized element generation approaches are commonly being used in supercomputer implementation of finite-element systems. The parallel assembly method applicable to supercomputers is a node by node assembly approach. This approach is driven by the grid connectivity information. An alternative method for processing constraints, especially advantageous on supercomputers, is based on the Lagrange multiplier method.
