Date:	Tue, 21 Dec 93 16:42:29 EST
From:	peprbv AT cfa0 DOT harvard DOT edu (Bob Babcock)
To:	tmoore AT pnfi DOT forestry DOT ca
Cc:	djgpp AT sun DOT soe DOT clarkson DOT edu
Subject:	EMU387 problems under windows

> The program does a lot of floating point math.  
> It compiles and works fine on my '486, although some results
> are different that what I am getting with the same program and datasets
> on my SUN IPX.  The differences are small in the least significant digits,
> but I am wondering if there is any way to get them to be identical.

Small differences are almost inevitable when doing floating point arithmetic
on machines with different architectures.  Nominally identical numbers stored
with a different number of bits become different.  I/O libraries which round
differently exaccerbate the problem.  Some calculations include operations
which amplify small differences, say by subtracting two nearly identical
numbers and then multiplying by a large factor.

As far as your problem under Windows with a 386 and no 387, I can only offer
a couple of suggestions:
1. Try both standard and 386 enhanced mode (startup commands win /s and
   win /3).  I recall that Windows can supply 387 emulation; whether it does
   this may depend on what mode you run in.
2. Look in your .INI files, probably system.ini, for anything that looks like
   a 387 emulator.

- Raw text -