Re: correctly rounded transcendentals

[Nick Maclaren <nmm1@xxxxxxxxxxxxx>]

The claim that it can be done without significant loss of performance
is dubious, to put it mildly.

If one selects the constraints and objectives used by the authors,
then that may well be true, but there are many and good reasons to
use different ones.  However, this is merely a more extreme state
to that which exists with IEEE 754, which is one of the reasons that
it is so often implemented or used in aberrant forms.

As I have posted before, it is false in at least 3 classes of use:

    It is not true when programs are computation bound and limited
by cache sizes or hardware real estate.

    It is not true for large-scale SIMD implementations, where the
speed of an operation is that of the slowest operand.

    It is not extensible to more than the simplest transcendental
functions for a realistic amount of work, or to 128-bit for many
of the existing ones.


Regards,
Nick Maclaren,
University of Cambridge Computing Service,
New Museums Site, Pembroke Street, Cambridge CB2 3QH, England.
Email:  nmm1@xxxxxxxxx
Tel.:  +44 1223 334761    Fax:  +44 1223 334679