Vectorization of a classical trajectory code on a floating point systems, Inc. Model 164 attached processor.

Area of Science:

• Computational chemistry
• Chemical physics
• High-performance computing

Background:

• Classical trajectory simulations are crucial for understanding chemical reaction dynamics.
• Optimizing computational performance is essential for complex molecular simulations.
• Vectorization is a technique to speed up computations by performing operations on arrays of data.

Purpose of the Study:

• To assess the performance improvements gained by vectorizing a triatomic classical trajectory code.
• To compare the vectorized code's efficiency on different hardware platforms (FPS 164 and VAX 11/780).

Main Methods:

• A triatomic classical trajectory code was modified using extensive algorithm vectorization.
• Performance timings were conducted on an FPS 164 attached processor and a VAX 11/780 with a floating-point accelerator.
• Timing tests utilized a London-Eyring-Polanyi-Sato (LEPS) potential energy surface and 1000 time steps per trajectory.
• Simulations were run with varying numbers of simultaneously executed trajectories.

Main Results:

• Vectorization led to notable timing improvements on both the VAX and FPS hardware.
• Significant speedups were observed as the number of simultaneously run trajectories increased.
• Up to a 25-fold improvement in computational speed was achieved between the vectorized VAX and FPS codes for large numbers of trajectories.

Conclusions:

• Extensive vectorization of classical trajectory codes provides substantial performance enhancements.
• The vectorized code demonstrates significant speed improvements, particularly on specialized hardware like the FPS 164.
• This optimization is critical for advancing the feasibility of complex chemical dynamics simulations.