On Tuesday, 1994-07-26  Donald Macmillan wrote to David Noon about
"DA/2 Distributed Applica" as follows:

DM> I believe it was Peter Fitzsimmonds who stated that the 
DM> Watcom 9.5 Fortran/C compilers supports pipelining for only 
DM> non-FP, while the Watcom 10.0 supports pipelining for both 
DM> integer AND FP. As Watcom has not yet released version 10.0 
DM> for Fortran, we were obliged to compile the Fortran code 
DM> with 9.5. To maintain the same version, we compiled both 
DM> with 9.5

Hi Donald,

So you are using FORTRAN!! [That aroma is still the same, even after
all these years.]

My message was not referring to pipelining. It was about excessive
page faulting caused by accessing, in row-major order, arrays stored
in column-major order. This has been a source of confusion for FORTRAN
non-programmers (i.e., physicists and engineers) for a couple of
decades now. I hope you kept my message, since it contains worked
examples of how to write slow FORTRAN and how to write fast FORTRAN
with just a subtle change. On a real storage machine there will be no
difference; the effect is only on virtual storage machines.

Since you are using Watcom, you might have a profiler or execution
analyzer that will tell you which subroutines and functions are
consuming the greatest amounts of processing time. If you concentrate
your optimization efforts on these you will get greater benefits
quickly.

If you care to post some code, I would be pleased to review it for
you.

As for pipelining, your DOS programs are probably not Pentium
optimized, so that would not be part of the difference. I think the
difference is simply code that is not designed with virtual storage, or
some other environmental influence, in mind.

Regards

Dave

___
 X KWQ/2 1.2e X This is Washington: we don't have time to wait for Chicago!
--- Maximus/2 2.01wb