floating-point emulators on PC's, a question from Cleve Moler

[David G. Hough on validgh]

Hi.  We're trying to build a version of MATLAB that will run
on IBM-type PC's WITHOUT 80x87 chips.  (Until now, all MATLAB's
have required the math chips.)  We're using the Microsoft 6.0 C
compiler.  I had thought they did a good job of emulating the
exception handling.  But either we're not compiling with the
correct flags, or they're not handling exceptions correctly.
Even things like

              double x, Zero = 0;
              x = 1/Zero;
              printf("x = %e\n",x);

give us x = 0.00000, not infinity.

So, I have a couple of questions.

   1.  How does Microsoft C handle 1/0, 0/0, log(0), etc. in
       their default mode which is 'sposed to run with or without
       the math coprocessor?

   2.  What compile flags do we need to set to get proper handling 
       if it is available?

   3.  If we need to provide our own exception handler, is there
       a decent one somewhere that we could start from?

If anybody wants to set us straight, please respond to

     moleramathworks.com

or call me at
  
     508-653-1415 (extension 352 if you get our machine.)

Thanks a lot.

   -- Cleve Moler