beaucham@uiucuxc.UUCP (06/21/84)
#N:uiucuxc:32400001:000:652
uiucuxc!beaucham Jun 20 21:45:00 1984
eee In looking at an article entitled "Performance of Various Computers
Using Standard Linear Equations Software in a Fortran Environment" by
Jack J. Dongarra of Argonne National Lab. I note that they tested the
VAX 780 with floating pt. accelerator using the "xf77" Fortran compiler.
His benchmark results show that xf77 under Unix is only slightly slower
than the Fortran which runs under VMS with the same hardware.
However, I have heard from several sources that f77 under Unix runs
much slower than VMS Fortran . Question: does anyone know what xf77 is? (we e don't have it).
Jim Beauchamp uiucdcs!uiucuxc!beaucham grandi@noao.UUCP (06/22/84)
*
My understanding of the LINPACK Benchmark is that xf77
represents the BSD 4.2 f77 compiler (with optimization on).
However, as regular readers of this newsgroup know, the 4.2 f77
optimizer is FULL of bugs. Thank the Gods of Fortran for Donn Seeley
and his bugfixes that at least seem to make the compiler reliable with
the optimizer off. SO, if all the bugs can be rooted out, the 4.2 f77
compiler is at least close to the VMS compiler in execution speed. But
it is sure slow when compiling! HOWEVER, DEC has just announced that the f77
compiler that will come with version 4.0 of VMS will increase execution
performance by 10 - 50%.
For those of you interested in the actual benchmarks, they are
taken from "Performance of Various Computers Using Standard Linear
Equation Software in a Fortran Environment," Jack J. Dougarra,
Mathematics and Computer Science Division, Argonne National
Laboratory, 10 Jan. 1984. Unfortunately, I have misplaced my copy of
this report and can only quote from a bowderlized version reprinted in
the Astronomical Image Processing Circular (#10). Linpack routines SGEFA
and SGESL were used for problems of order 100.
Computer/Compiler Double-Precision Mflops Single-Precision Mflops
--------------------------------------------------------------------------------
Cray X-MP/CFT 21.0
CDC Cyber 205/FTN 8.4
Amdahl 5860/H enhan (opt=3) 3.9 5.5
IBM 3081K/H enhan (opt=3) 2.1 2.4
DG MV-10000/f77 (opt=3) 0.30 0.39
VAX 11/780 + FPA/VMS-f77 0.13 0.21
VAX 11/780 + FPA/UNIX(4.2)-f77 opt 0.11 0.20
Ridge 32/f77 0.12 0.20
VAX 11/750 + FPA/VMS-f77 0.10 0.14
IBM 4331/H (opt=3) 0.038 0.088
VAX 11/730 + FPA/VMS-f77 0.036 0.047
IBM PC-XT/370/H (opt=3) 0.031 0.040
Sun 2 + FPA/UNIX(4.2)-f77 opt 0.022 0.039
--
Steve Grandi, National Optical Astronomy Observatories, Tucson, AZ, 602-325-9228
{allegra,arizona,astrovax,decvax,hao,ihnp4,lbl-csam} !noao!grandi