Lawrence Berkeley National Laboratory

Global address space languages like UPC exhibit high performance and portability on a broad class of shared and distributed memory parallel architectures. The most scalable applications use bulk memory copies rather than individual reads and writes to the shared space, but finer-grained sharing can be useful for scenarios such as dynamicload balancing, event signaling, and distributed hash tables. In this paper we present three optimization techniques for global address space programs with fine-grained communication: redundancy elimination,use of split-phase communication, and communication coalescing. Parallel UPC programs are analyzed using static single assignment form and a data flow graph, which are extended to handle the various shared and private pointer types that are available in UPC. The optimizations also take advantage of UPC's relaxed memory consistency model, which reduces the need for cross thread analysis. We demonstrate the effectiveness of the analysis and optimizations using several benchmarks, which were chosen to reflect the kinds of fine-grained, communication-intensive phases that exist in some larger applications. The optimizations show speedups of upto 70 percent on three parallel systems, which represent three different types of cluster network technologies.