Performance Tuning of N-Body Codes on Modern Microprocessors: I. Direct Integration with a Hermite Scheme on x86_64 Architecture

摘要

The main performance bottleneck of gravitational N-body codes is the forcecalculation between two particles. We have succeeded in speeding up thispair-wise force calculation by factors between two and ten, depending on thecode and the processor on which the code is run. These speedups were obtainedby writing highly fine-tuned code for x86_64 microprocessors. Any existingN-body code, running on these chips, can easily incorporate our assembly codeprograms. In the current paper, we present an outline of our overall approach, which weillustrate with one specific example: the use of a Hermite scheme for a directN^2 type integration on a single 2.0 GHz Athlon 64 processor, for which weobtain an effective performance of 4.05 Gflops, for double precision accuracy.In subsequent papers, we will discuss other variations, including thecombinations of N log N codes, single precision implementations, andperformance on other microprocessors.