Shao-Jing Dong University of Kentucky The SIGN Function Calculation of Large Scale Sparse Matrix --- The Problem, Programming and Performance on HP/SPP2200 To calculate the SIGN function of the large scale sparse matrix is a big challenge to the numerical supercomputing today.

In this talk an iteration algorithm which combines the rational expansion, multiple shifted matrix inversion and deflation preconditioning is presented. The cache missing latency, decode latency and message passing latency are discussed. The performance on the SPP2200 is measured. The friendlyship of programming environment on the HP/SPP2200 is discussed. Numerical experiment of the chiral lattice fermion is included too.

---