R SIG Fedora - Sep 2014 - Compiling R on Fedora with threaded openblas

Dear List,

I am trying to build R on Fedora against a threaded openblas installed from
Fedora's yum repos, or alternatively swap out libRblas.so with a threaded
openblas built locally.

I have installed

    $ yum list installed | grep openblas
    openblas.x86_64             0.2.11-1.fc20           @updates

    openblas-devel.x86_64       0.2.11-1.fc20           @updates

    openblas-openmp.x86_64      0.2.11-1.fc20           @updates

    openblas-openmp64.x86_64    0.2.11-1.fc20           @updates

    openblas-serial64.x86_64    0.2.11-1.fc20           @updates

    openblas-threads.x86_64     0.2.11-1.fc20           @updates

    openblas-threads64.x86_64   0.2.11-1.fc20           @updates

The locally built openblas returned the following information

    OpenBLAS build complete. (BLAS CBLAS LAPACK LAPACKE)

      OS               ... Linux
      Architecture     ... x86_64
      BINARY           ... 64bit
      C compiler       ... GCC  (command line : gcc)
      Fortran compiler ... GFORTRAN  (command line : gfortran)
      Library Name     ... libopenblas_haswellp-r0.2.11.a (Multi threaded;
Max num-threads is 4)

    To install the library, you can run "make
PREFIX=/path/to/your/installation install".

which suggests to me I successfully built a multi threaded openblas.

Whilst trying to build against Fedora's openblas I used the following
configure option

    BLAS_LIBS="-L/usr/include/openblas -lopenblas"

or

    BLAS_LIBS="-L/usr/lib64 -lopenblas"

and

    ./configure --with-blas --enable-R-shlib --enable-BLAS-shlib

Whilst trying to swap out the default libRblas.so I built simply with

    ./configure --enable-R-shlib --enable-BLAS-shlib

then I symlinked my locally built libopenblas.so to RHOME/lib/libRblas.so

Neither approach is yielding an R that uses multiple threads when doing
linear algebra, e.g.

    a <- matrix(4, 10000, 10000)
    eigen(a)

What am I doing wrong? or alternatively, how should I approach building a
version of R with a threaded openblas on Fedora.

Thanks in advance,

Gavin
-- 
Gavin Simpson, PhD

	[[alternative HTML version deleted]]