The Enhanced Virginia Hydrodynamics \#1 (EVH1) code represents an important application kernel in the ``TeraScale Simulations of Neutrino-Driven Supernovae and Their Nucleosynthesis'' SciDAC project. EVH1 is an MPI code that uses second-order operator splitting and 1D Lagrangian hydrodynamics in each coordinate direction.

For these experiments, the code was compiled using "-C hopt" optimization. Profiling and performance counters were also used to identify routines that were not performing well on the SX-6. Two routines (riemann.f and parabola.f) were modified to improve vector performance.

The EVH1 benchmark is a fixed-size problem specified by the application scientists, Sedov-Taylor blast-wave solution in a 2D spherical geometry using a 128 X 128 computational grid. Results are presented in terms of simulations seconds per day.

From these results, SX-6 performance using the modified code is twice that of the p690 for the same number of processors. Code scalability is good up to 64 processors, and the full 32-way p690 node is 60% faster than the 8-way SX-6 node for solving this fixed size problem.

From the hardware performance counters, EVH1 is achieving approximately 15% of peak performance on the SX-6. Exploiting additional vectorization is feasible, but would require a significant restructuring of the code. A larger problem size would also increase vector lengths (and vector performance).