ilan343@violet.berkeley.edu (10/03/90)
I've been some running some benchmarks tests with f2c on a AT&T 3B2/400. I would like to hear from other people playing with f2c if these results were to be expected. As you can see from the numbers below, the Whetstone ratings are virtually identical for the native FORTRAN compiler and for the f2c translated code. However, the f2c code is much slower in the LINPACK benchmark tests. Is this reasonable? Any guesses on why the f2c LINPACK is so much slower? Did I goof somewhere? Thanks for your comments, Geraldo Veiga BENCHMARKS: Whetstones MFLOPS (linpack) Single/Double Single/Double Compiler KWhet/sec Ave. Mflops 1. 3B2 f77 F77-XLA+ Rel 1.0 339/231 4.6E-2/4.5E-2 2. f2c + 3B2 cc 343/232 1.65E-2/.78E-2 (11.0) 1/3/86 (f2c library) The benchmark programs are the current (as of 09/30/90) FORTRAN versions available in the NETLIB. System: Don't laugh at these performance numbers, the test machine is an underpowered AT&T 3B2/400 with a Floating point co- processor, running SysV 3.0. I was wondering if I could safely use f2c as a replacement FORTRAN compiler in other System V systems, so the absolute performace is not important. Compiler flags: CFLAGS= -O FFLAGS= -O LDFLAGS= -lf2c -lm -lc # libf2c.a is the replacement library provided # with f2c Flags for f2c: -R