An Efficient Algorithm for Parallel Stiffness Matrix Assembling
Andreas Unterkircher, Josef Reissner · Civil-comp proceedings · 2009
We present an algorithm for stiffness matrix assembling on shared memory multiprocessor architectures. The method has been developed for our 3D bulk forming FE code. After generating the dual graph of the FE mesh a recursive maximal independent set algorithm delivers indices of elements which do not share any nodes. These elements are being assembled by different processors and the element matrices can be directly added to the global stiffness matrix. The method is independent of dimension, model shape and element type. Furthermore it can be easily incorporated into an existing FE code.