[Rd] FLAME

[Jan de Leeuw]

Yesterday I ported FLAME to OS X  
(http://www.cs.utexas.edu/users/flame/index.html).
FLAME is a system to automatically (currently semi-automatically)  
generate
efficient and correct linear algebra routines in C or MATLAB or MPI  
from descriptions
of the algorithms. It uses code that relies heavily on partitioning of  
matrices,
and avoids indices as much as possible. There are a number of articles  
in
TOMS on this approach, and on some applications. Short intro at
http://www.cs.utexas.edu/users/flame/new_FIREsite/Examples/index.html

I was able to shave 10%-20% of the time needed to do a large QR  
decomposition,
compared to the LAPACK that is sitting in vecLib.framework. See the  
examples
in http://www.cs.utexas.edu/users/flame/FIREsite/.

Now the FLAME group (in particular Robert van der Geijn, who is also  
responsible
for PLAPACK) is interested in extending the FLAME approach to  
non-standard
LAPACK extensions. I forward a message below, which he asked me to  
forward
to "my community".
====================
If there were operations that are not (properly) supported by LAPACK,  
what are they?  In other words, what wish-list of
operations would you like to see supported by an LAPACK-like library  
(just dense linear algebra, please)?  What routines
in LAPACK were modified to get slightly different functionality?

An example we ran into was in updating the QR factorization:

LAPACK supports A -> Q R.  But what if a new set of observations comes  
in?

Given A->QR ,
/ A \ -> / Q1 \  R1
\ B /      \ Q2 /
requires a complete new factorization, which is costly.
====================
I guess R-core can perhaps help getting more statistical computation  
into FLAME
by answering these questions. I can forward answers, or you can directly
contact  rvdg at cs.utexas.edu

===
Jan de Leeuw; Professor and Chair, UCLA Department of Statistics;
Editor: Journal of Multivariate Analysis, Journal of Statistical  
Software
US mail: 8130 Math Sciences Bldg, Box 951554, Los Angeles, CA 90095-1554
phone (310)-825-9550;  fax (310)-206-5658;  email: deleeuw at stat.ucla.edu
homepage: http://gifi.stat.ucla.edu
   
------------------------------------------------------------------------ 
-------------------------
           No matter where you go, there you are. --- Buckaroo Banzai
                    http://gifi.stat.ucla.edu/sounds/nomatter.au