The Community for Technology Leaders
SC Conference (1999)
Portland, Oregon, USA
Nov. 13, 1999 to Nov. 18, 1999
ISBN: 1-58113-091-0
pp: 25
Larry Carter , University of California, San Diego
Kang Su Gatlin , University of California, San Diego
Divide and conquer programs can achieve good performance on parallel computers and computers with deep memory hierarchies. We introduce architecture-cognizant divide and conquer algorithms, and explore how they can achieve even better performance.<div></div> An architecture-cognizant algorithm has functionally-equivalent variants of the divide and/or combine functions, and a variant policy that specifies which variant to use at each level of recursion. An optimal variant policy is chosen for each target computer via experimentation. With h levels of recursion, an exhaustive search requires \theta(v<sup>h</sup>) experiments (where v is the number of variants). We present a method based on dynamic programming that reduces this to \theta(v<sup>c</sup>) (where c is typically a small constant) experiments for a class of architecture-cognizant programs.<div></div> We verify our technique on two kernels (matrix multiply and 2-D Point Jacobi) using three architectures. Our technique improves performance by up to a factor of two, compared to architecture-oblivious divide and conquer implementations. Further our dynamic programming approach succeeds in selecting the optimal variant policy.
Larry Carter, Kang Su Gatlin, "Architecture-Cognizant Divide and Conquer Algorithms", SC Conference, vol. 00, no. , pp. 25, 1999, doi:10.1109/SC.1999.10068
104 ms
(Ver 3.3 (11022016))