Performance improvements for nvh analysis on cray sv1 computers

摘要

Significant improvements were made to MSC. Nastran kernels to exploit the architecture of the Cray SV1. Special attention was given to kernels involved in typical NVH analysis. The biggest performance gains were made in the decomposition kiernels used in eigenanalysis and model and direct frequency response. Improvements were also made to the FBS kernels, AXPR kernels and MPYAD matrix-multiply-add kernels. Performance on a set of input decks from industry is shown, demonstrating that these optimizations greatly reduce turnaround time for most large-scale NVH computations.