In this paper we present an analysis on a shared memory system of five self-scheduling algorithms running on top of the threads programming model to schedule the loop with cross-iteration dependence. Four of them are well-known: self-scheduling (SS), chunked self-scheduling (CSS), guided self-scheduling (GSS) and factoring. Because these schemes are all for loops without cross-iteration dependence, we study the modification of these schemes to schedule the loop with cross-iteration dependence. The fifth is our proposal: carried-dependence self-scheduling (CDSS). The experiments conducted in varying parameters clearly show that CDSS outperforms other modified self-scheduling approaches in a number of simulations. CDSS, modified SS, factoring, GSS and CSS are executed efficiently in order of execution time.
展开▼
