PGHPFPortland Group High Performance Fortran
Copyright 1988-2018, All rights reserved.
References in periodicals archive ?
Figure 2 shows that, in general, communications cshift, scatter, spread, and sum scale well under pghpf in most benchmarks, except for cshift in benchmark pcr:inst_coef.
Table XVII compares the performance of the application kernel benchmark codes compiled using pgf90 and pghpf, respectively, versus that of the sequential versions of the same codes compiled using xlf, on a single node of the SP2.
Single-Node Performance of the Application Kernels xlf pgf90 pghpf Time FLOP Rate Time Time Code (sec.) (Mflops/s) (vs.
For most applications, communications cshift, gather/ scatter, sort, and sum scale well under pghpf till eight nodes.