Yes, an interval BLAS standard
http://www.netlib.org/blas/blast-forum/chapter5.pdf was
proposed and discussed at the BLAST Forum. The Forum placed it
in Journal of Development due to the fact that there was not a
standard for interval computing. Hope that the 1788 committee
will release the standard soon.
>>> James Demmel
<demmel@xxxxxxxxxxxxxxxxx>
8/29/2013 11:34 AM >>>
>
> If EDP (exact dot product) is so important, why is it not
in BLAS?
> The search function in netlib seems to be broken,
> so I was not able to see if some other package in netlib
includes it.
> It is true that there are separate
> libraries available though.
Just to follow-up, LAPACK has contained routines for extra
precise
iterative refinement
for a while, that use double-double BLAS to compute residuals.
See
dgesvxx.f for example.
There was a standards committee that met to standardize extra
precise BLAS
(see
http://netlib.org/blas/blast-forum/
and
http://netlib.org/xblas),
which are used
by dgesvxx.f and related routines.
Web hits are an imprecise measure at best, but for what it's
worth the
number of
hits to xblas-related parts of netlib is a tiny fraction of the
>170M
hits to other
Sca/LAPACK related netlib pages.
The BLAST Forum also worked on standardizing other BLAS
extensions. We
approached
the interval community at the time to see if they wanted to
participate
by formulating
an interval BLAS standard. For some reason there was no
consensus on
the definitions.
Perhaps this provided some motivation for P1788 :) .
Jim