Modem dialects of Fortran enjoy wide use and good support on high-performance computers as performance-oriented programming languages. By providing the ability to express nested data parallelism, modern Fortran dialects enable irreg- ular computations to be incorporated into existing applica- tions with minimal rewriting and without sacrificing perfor- mance within the regular portions of the application. Since performance of nested data-parallel computation is unpre- dictable and often poor using current compilers, we inves- tigate threading and flattening, two source-to-source trans- formation techniques that can improve performance and per- formance stability. For experimental validation of these tech- niques, we explore nested data-parallel implementations of the sparse matrix-vector product and the Bames-Hut n-body algorithm by hand-coding thread-based (using OpenMP di- rectives) and flattening-based versions of these algorithms and evaluating their performance on an SGI Origin 2000 and an NEC SX-4, two shared-memory machines.
展开▼
