Highlights from discussions with W. Kahan on 28 June 1991

[Bob Alverson]

David Hough says:
>Complex multiply: Kahan thinks this should be a hardware instruction, because
>it's so hard to get the critical cases for conformal mapping correct and
>fast in software.  In particular infinite operands lead to trouble with
>gratuitous NaN results, while zero results often have the wrong signs
>which lead to problems in conformal mapping involving slits in function
>definition domains.
>

Is there any reference on what the critical cases should return for
answers?  Can someone cook some up if not?  It sure would be nice if
the answers matched a vanilla multiply when the imaginary parts were
zero.

Bob