Hyper-systolic algorithms for N-body computations and parallel level-3 BLAS libraries

摘要

Hyper-systolic algorithms represent a new class of parallel computing structures. Because of their regular communication and compute patterns they are well suited for implementation on most parallel architectures, in particular, high performance SIMD machines can benefit considerably. After a short explanation of the concept of hyper-systolic algorithms, their application to N-body computations and distributed matrix multiplication is discussed. Results from real implementations are presented.