R devel - Mar 2011 - R and ATLAS and distributing across boxes

Hi,

 	If I have a cluster of heterogenous machines, each with their own
self-optimized ATLAS, do I need to compile R on each machine to tell it to
take advantage of the local ATLAS? Or is it sufficient to compile R once
with the appropriate --with-blas and --with-lapack flags, and then trust
that, once installed on a certain box, R will look in the right place and
find and use the local library? Or will I need to recompile R on each box?

 	Googling around has shown me how to have R use ATLAS instead of
native BLAS, but none of them note whether changing the BLAS library out
from under an R binary will change the code that R executes.

Thanks in advance,
Adam D. I. Kramer

On 9 March 2011 at 14:34, Adam D. I. Kramer wrote:
|  	If I have a cluster of heterogenous machines, each with their own
| self-optimized ATLAS, do I need to compile R on each machine to tell it to
| take advantage of the local ATLAS? Or is it sufficient to compile R once
| with the appropriate --with-blas and --with-lapack flags, and then trust
| that, once installed on a certain box, R will look in the right place and
| find and use the local library? 

Correct. Shared libraries, well-defined interfaces, hence plug-and-play.

| Or will I need to recompile R on each box?

No. If they all have libblas and liblapack somewhere were ld.so finds them
you should be fine/
 
|  	Googling around has shown me how to have R use ATLAS instead of
| native BLAS, but none of them note whether changing the BLAS library out
| from under an R binary will change the code that R executes.

Blas etc use is discussed in Appendix A.3 Linear algebra of the R
Installation and Administration manual.

Dirk

-- 
Dirk Eddelbuettel | edd at debian.org | http://dirk.eddelbuettel.com