[R] Compiling R with multi-threaded BLAS math libraries - why not actually ?

[Barry Rowlingson]

On Sat, Jun 12, 2010 at 3:16 PM, Tal Galili <tal.galili at gmail.com> wrote:
> Hello Douglas,
>
> Thank you for the BLAST!=BLAS correction (I imagine my slip was due to some
> working I have done recently with an RNA analysis software called BLAST).
>
> Also, thank you for the very interesting posting here and in your reply to
> David's post.
>
> My current conclusion from this thread are that:
> 1) This should be interesting ONLY if I will be working on large matrices
> and doing "very specific
> kinds of operations". (I imagine David's examples on his post demonstrate
> those)
> 2) In case I would like to do it, I will need to go follow the actions
> detailed here (thank you for the pointer):
> http://cran.r-project.org/bin/windows/base/rw-FAQ.html#Can-I-use-a-fast-BLAS_003f
> And more or less pray that my computer specification are relevant.
> <http://cran.r-project.org/bin/windows/base/rw-FAQ.html#Can-I-use-a-fast-BLAS_003f>(Although
> I do wonder how does REvolution distribution succeeds in doing this without
> making the user do any more steps then just installing R)
>

 Another option if you discover that your algorithm benefits from
massive parallelism is computing on the GPU. There are BLAS
implementations, for examples:

http://gpgpu.org/index.php?s=BLAS&searchbutton=Search

 but integration with R is another issue (and I have a vague memory it
was covered in threads on R-dev recently, which is probably where this
thread should be anyway).

Barry