The Wayback Machine - https://web.archive.org/web/20170706053605/http://grouper.ieee.org/groups/754/email/msg03842.html
[
Date Prev][
Date Next][
Thread Prev][
Thread Next][
Date Index][
Thread Index]
Re: Missing functions tanPi, asinPi and acosPi
On Apr 11, 2008, at 8:15 AM, Michel Hack wrote:
I checked several 754R drafts all the way back to P754/D0.15.3 of
October 2005, where the transcendental function section was still
marked "New". Initially, the set of trig functions were those
where correctly-rounded versions existed (at least for Binary64),
namely sinPi, cosPi, atanPi and sin, cos, atan. There was a
comment from Vincent that the inverses asin, acos and tan were
also "feasibly correctly rounded", so by April 2006 (possibly
earlier) those were added to the list.
Nobody ever mentioned tanPi, asinPi and acosPi since then, until
now. Could it be that they were thought to be easily-derivable
from the others? Perhaps they are -- except for single-rounding
of the result...
So we need input from Vincent, Jean-Michel etc.
Michel.
Sent: 2008-04-11 15:13:39 UTC
tanPi was removed at some point. As I recall it introduced an annoying
case where if a format had an unusually large precision for the
exponent range
it could overflow near the singularities, but not for typical formats.
Note that there is nothing preventing someone from including tanPi,
asinPi, etc from a language. The feeling on the BRC was that there
were enough cases where sinPi and cosPi would be appropriate
that the simpler reduction and added accuracy would be a nice feature.
This seemed less compelling for the inverse functions, and seemed
less commonly needed for tan.
Overall, there was a desire to keep the list reasonably short.
There are quite a lot of other functions that one could argue for
and the line needed to be drawn somewhere.
Cheers,
Jeff Kidder
