a minor x86 bug reported

[Jerry Huck]

> From: figueroaaslinky.cs.nyu.edu (Samuel A. Figueroa)
> To: dghavalidgh.com
> Subject: Re: a minor x86 bug reported
> Cc: numeric-interestavalidgh.com
> 
> Perhaps a slightly different kind of "bug" is the kind described in the second
> edition of "PA-RISC 1.1 Architecture and Instruction Set Reference Manual" on
> pages 6-32, 6-34, and 6-35: when overflow or underflow occurs during floating-
> point conversion operations, the unimplemented exception may be signaled in
> certain circumstances instead of the overflow or underflow exception, as
> required by the IEEE Standard.

"unimplemented" exceptions are a little different.  This is a
mechanism to allow software support of difficult (and hopefully
unlikely) situations.  The low level exception handler is able to
resolve all unimplemented exceptions and return to the user.
unimplemented exceptions can occur for any reason the HW chooses.
Fortunately, the IEEE 754 standard does not require that all aspects
of the standard are supported directly by the hardware.  The hardware
and the trap handler creates the effect of the full standard.

Similar mechanisms exist in other RISC architectures to reduce the
cost and complexity of FP HW.  On the other hand, some folks feel the
HW abuses this capability because exceptions should not be considered
infrequent.

Jerry