[rescue] Another fine business decision for us axp-lovers

[Scott Newell]

>I believe that Intel uses 80bits internally to prevent rounding errors,
>but only presents a 64 interface.  I think.  I don't really know or care

Nope, you can get all 80 external as well.


>If they really are 80bit viewable to the program, then MS is being really
>stupid if they don't support it.  That would be a significant edge to use
>against other competitor for users who do scientific computing.

I don't have a MS compiler handy, but Borland 5.something shows:
	sizeof(float) as 4
	sizeof(double) as 8
	sizeof(long double) as 10


Here's an interview with Dr. William Kahan that you might find interesting:


http://www.ddj.com/articles/1997/9711/9711a/9711a.htm


One of his papers with some interesting comments on Java, numerics, and the
inferior numeric precision of RISC chips:


http://www.cs.berkeley.edu/~wkahan/JAVAhurt.pdf


newell