An Evaluation of High Performance Fortran Compilers Using the HPFBench Benchmark Suite
- 83 Downloads
The High Performance Fortran (HPF) benchmark suite HPFBench was designed for evaluating the HPF language and compilers on scalable architectures. The functionality of the benchmarks covers scientific software library functions and application kernels. In this paper, we report on an evaluation of two commercial HPF compilers, namely, xlhpf from IBM and pghpf from PGI, on an IBM SP2 using the linear algebra subset of the HPFBench benchmarks.
Our evaluation shows that, on a single processor, there is a significant overhead for the codes compiled under the two HPF compilers and their Fortran 90 companions, compared with the sequential versions of the codes compiled using xlf. The difference mainly comes from the difference in code segments corresponding to the communications when running in parallel. When running in parallel, codes compiled under pghpf achieve from slightly to significantly better speedups than when compiled under xlhpf. The difference is mainly from better performance of communications such as cshift, spread, sum and gather/scatter under pghpf.
KeywordsSequential Version Benchmark Suite Code Segment Scalable Architecture Tridiagonal System
- D. Bailey, et. al. The NAS parallel benchmarks. Technical Report RNR-94-007, NASA Ames Research Center, Moffett Field, California, March 1994.Google Scholar
- M. Berry et. al. The PERFECT Club benchmarks: Effective performance evaluation of supercomputers. International Journal of Supercomputer Applications, 3:5–40, 1989.Google Scholar
- L.S. Blackford, J. Choi, A. Cleary, E. D’Azevedo, J. Demmel, I. Dhillon, J. Dongarra, S. Hammarling, G. Henry, A. Petitet, K. Stanley, D. Walker, and R.C. Whaley. ScaLAPACK: A Linear Algebra Library for Message-Passing Computers. In SIAM Conference on Parallel Processing, March 1997.Google Scholar
- J.C. Cooley and J. Tukey. An algorithm for the machine computation of complex fourier series. Math. Comp, 19:291–301, 1965.Google Scholar
- G. Dahlquist, A. Björck, and N. Anderson. Numerical Methods. Series in Automatic Computation. Prentice Hall, Inc., Englewood Cliffs, NJ, 1974.Google Scholar
- J.J. Dongarra. Performance of various computers using standard linear equations software. Technical Report CS-89-85, University of Tennessee, Department of Computer Science, 1989.Google Scholar
- G. Golub and C. vanLoan. Matrix Computations. The Johns Hopkins University Press, second edition, 1989.Google Scholar
- High Performance Fortran Forum. High Performance Fortran; language specification, version 1.0. Scientific Programming, 2(1-2):1–170, 1993.Google Scholar
- R.W. Hockney and C. Jesshope. Parallel Computers 2. Adam Hilger, 1988.Google Scholar
- R. Hocney and M. Berry. Public international benchmarks for parallel computers: Parkbench committe report-1. Technical report, Netlib, Oak Ridge National Laboratory, February 1994.Google Scholar
- Y.C. Hu, G. Jin, S.L. Johnsson, D. Kehagias, and N. Shalaby. HPFBench: A High Performance Fortran benchmark. Tech. Rep. TR98-322, Computer Science Dept., Rice Univ., 1998. URL: http://www.crpc.rice.edu/HPFF/benchmarks.
- Y.C. Hu, S.L. Johnsson, and S.-H. Teng. High Performance Fortran for highly irregular problems. In Proc. of the 6th ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming, Las Vegas, Nevada, June 1997.Google Scholar
- IBM. IBM Parallel Engineering and Scientific Subroutine Library Release 2, Guide and Reference, 1996.Google Scholar
- S.L. Johnsson, T. Harris, and K.K. Mathur. Matrix multiplication on the Connection Machine. In Supercomputing 89, pages 326–332. ACM, November 1989.Google Scholar
- F. McMahon. The Livermore Fortran kernels: A test of numerical performance range. In Performance Evaluation of Supercomputers, pages 143–186. North Holland, 1988.Google Scholar