IEEE Double extended

[James Demmel]

(Response to query from D. Knaak supplied by W. Kahan and forwarded
by J. Demmel).

1) 1-bit sign, 15-bit exponent and 112-bit fraction is appropriate
for double extended. It was included in the earliest draft of
IEEE p754, but dropped to "simplify" the proposal. It also
approximates DEC's H format on Vaxen, and majorizes IBM's "extended
format" on 370s.

2) No manufacturer is doing anything different anymore, but see
   the next response.

3) "Quad" is a good name. IEEE p1596.5 suggests alternate terminology,
like "binary real hexlet". Anyway, take care to distinguish "quad" from
"doubled double" used by IBM RS/6000 and, therefore, by the
IBM/Apple/Taligent "Power PC" line. This uses two IEEE doubles to
represent a number with twice the precision but the same exponent range.