IFIP WG 2.5 Business meeting
George Washington
University
800 21st St. NW
08:15 Continental breakfast
Social (accompanying persons invited to meet and discuss plan for week)
Opening ...
09:15 Opening and welcome
Review of the agenda
Minutes of last meeting
09:30 Review of WG 2.5 Handbook project (Einarsson)
10:30 Status of membership
Nominations of new members
10:45 Break
Technical Presentations ...
11:00 Ian Reid (NAG
Ltd), Experiences developing basic math libraries for the AMD64
architecture (see abstract
below)
12:00 Lunch
No formal lunch planned. Members patronize local restaurants.
WG Business …
14:00 Review of WG 2.5 Aims and Scope statement
14:30 Review of WG 2.5 Technical Projects
15:30 Plans for 2005 WG 2.5 Meeting/Workshop in Hong Kong (Mu)
16:30 Adjourn
19:00 Informal group dinner at local restaurant
08:15 Continental Breakfast
Working Group Business ...
09:00 Proposal for WoCo9 (Pool)
10:00 Membership
- Vote on new members
- Discussion of future nominees for membership
10:15 Break
10:45 Discussion of Future Meetings
- Future meetings of WG 2.5: 2006 and beyond
- Other future meetings of interest
- Sessions to be organized by WG 2.5
11:00 New business
11:30 Adjournment
13:00 Excursion
Meet bus at Marvin Center
Experiences developing basic math libraries for the AMD64
architecture
AMD's departure from the Intel model for 64-bit (i.e. IA64) computing necessitated the building of a whole “eco-system” to support developers. NAG has been working with AMD over recent years to deliver the math infrastructure for the AMD64 architecture. This has been an extremely successful collaboration which is ongoing and which continues to deliver exceptional results in terms of both accuracy and performance.
This talk looks at the experiences
and results of the work to produce two math libraries: libm, which includes a
range of low-level routines such as the trigonometric functions; and ACML (AMD
Core Math Library) which currently includes the BLAS, LAPACK and a range of
FFTs. We look at the approaches taken, tools employed and the various lessons
learned along the way. We also look at some of the results, both in terms of
accuracy and performance with comparisons where appropriate.
[The speaker, although not directly involved in the development, has
been the NAG project coordinator throughout, has been involved in many of the
technical discussions and has even contributed occasionally! He also holds a
PhD in computer arithmetic.]