Howdy Maik,
AMEN!
https://github.com/easybuilders/easybuild-easyblocks/issues/1368
I have this in our latest for R 3.5 to deal with it.
# bits of this from
https://software.intel.com/en-us/articles/build-r-301-with-intel-c-compiler-and-intel-mkl-on-linux
preconfigopts = """
export BLAS_LIBS="$LIBBLAS_MT" # make sure we use multi-threaded
BLAS/LAPACK
# however, don't set LAPACK_LIBS
per: https://github.com/easybuilders/easybuild-easyconfigs/issues/1435
# note that we can't use
--with-blas='-lmkl_rt -liomp5 -lpthread' because EB will override
...
Jack Perdue
Lead Systems Administrator
High Performance Research Computing
TAMU Division of Research
[email protected] http://hprc.tamu.edu
HPRC Helpdesk: [email protected]
On 08/03/2018 02:44 AM, Maik Schmidt wrote:
Hi all,
a user has asked how he can speed up this matrix operations in R and I
have noticed that the R in EasyBuild that uses the intel toolchain
(e.g. R-3.4.4-intel-2018a-X11-20180131.eb) is only linked against the
sequential MKL, although R itself is single-threaded, so it might make
sense to use the threaded MKL instead.
I also tried the foss version and here the matrix operations are done
in parallel because OpenBLAS uses multiple threads. But not with MKL.
Is this intentional? if yes, why so? What do I have to change to
install R with the multithreaded MKL?
Thanks,
Maik
