The Cache Complexity of Multithreaded Cache Oblivious Algorithms

We present a technique for analyzing the number of cache misses incurred by multithreaded cache oblivious algorithms on an idealized parallel machine in which each processor has a private cache. We specialize this technique to computations executed by the Cilk work-stealing scheduler on a machine with dag-consistent shared memory. We show that a multithreaded cache oblivious matrix multiplication incurs Image (http://www.springerlink.com/content/l8946u2p54m72263/224_2007_9098_Article_IEq1.gif) is missing or otherwise invalid. cache misses when executed by the Cilk scheduler on a machine with  P processors, each with a cache of size  Z , with high probability. This bound is tighter than previously published bounds. We also present a new multithreaded cache oblivious algorithm for 1D stencil computations incurring Image (http://www.springerlink.com/content/l8946u2p54m72263/224_2007_9098_Article_IEq2.gif) is missing or otherwise invalid. cache misses with high probability, one for Gaussian elimination and back substitution, and one for the length computation part of the longest common subsequence problem incurring Image (http://www.springerlink.com/content/l8946u2p54m72263/224_2007_9098_Article_IEq3.gif) is missing or otherwise invalid. cache misses with high probability.