[Trilinos-Users] [EXTERNAL] Shadow number reference

Heroux, Michael A maherou at sandia.gov
Mon Apr 9 11:20:53 MDT 2012


The first public forum for my discussion of "Float With Shadow Double" and
the general concept of shadowing was at the 2009 SIAM CSE meeting.  The
first ever forum was a project review for the Institute for Advanced
Architectures and Algorithms in December 2008 in Santa Fe, NM.

The CSE09 talk description is in the abstracts
(http://www.siam.org/meetings/cse09/CSE09abstracts.pdf).  The title info

		Software Design Issues for Scalable Multicore Computers
Michael A. Heroux and Christopher G. Baker, Sandia National Laboratories
SIAM Conference on Computational Science and Engineering
Miami, FL, March 2-6, 2009

I have not published these ideas, but have spoken about them in other
venues.  I am unaware of previous work, but have not done an exhaustive
literature search either.

If you find related work (especially earlier work), I would appreciate
knowing about it.

If you have further questions or comments, please let me know.

If you are interested in the FloatShadowDouble data type that is
compatible with Trilinos, please let me know.  I haven't used it in a
while, so it might need some modification.

I hope this helps.



On 4/9/12 9:05 AM, "Rhys Ulerich" <rhys.ulerich at gmail.com> wrote:

>I have seen Michael Heroux give a talk where he suggest packing a
>float and a double into a single C++ type, performing all computations
>on both values, and then examining the difference to gauge truncation
>errors.  Around me this has been called a "shadow number".  Does
>anyone have a reference with where the idea originated?  I'm having no
>luck digging for it.
>Trilinos-Users mailing list
>Trilinos-Users at software.sandia.gov

More information about the Trilinos-Users mailing list