Installing OpenBLAS for self-compiled R

How to compile, and make sure a self-compiled R uses OpenBLAS, which may be faster than the installed one.


November 1, 2023


When you compile R from scratch on Linux, it doesn’t use the system BLAS libraries, it uses one that is compiled itself. This BLAS library may be rather slow. If you want to compile and use OpenBLAS, you then tell R to use the newly compiled library.

Compiling & Installing OpenBLAS

Clone and compile OpenBLAS. If you’ve already compiled R on this machine, it should have all of the pre-requisites already installed.

# get it
cd ~/software
git clone

We are going to compile it in multi-threaded mode (USE_THREAD=1), and to use a dynamic architecture, where it will have the gotoblas symbols included (DYNAMIC_ARCH=1).

# compile it
cd OpenBLAS

You should see test output here, that indicates everything is OK.

Now we create a directory to install OpenBLAS into. For my purposes, this is still going to be in ~/software.

# install it
mkdir ~/software/libOpenBLAS
make -j PREFIX=~/software/libOpenBLAS USE_THREAD=1 DYNAMIC_ARCH=1 install

Telling R To Use It

If you’ve compiled your own version of R, it should be using it’s own internal BLAS and LAPACK libraries. You can see this using sessionInfo().

sessionInfo()[c("BLAS", "LAPACK")]

# mine shows
# > $BLAS
# > [1] "/software/R-4.3.0/lib/"
# > [1] "/software/R-4.3.0/lib/"

So, now we want to backup the BLAS library, and then link it to our compiled OpenBLAS.

cp ~/software/R-4.3.0/lib/ ~/software/R-4.3.0/lib/
ln -snf ~/software/libOpenBLAS/lib/ ~/software/R-4.3.0/lib/

When you start R again, you should now see that it is using a different BLAS & LAPACK library.

sessionInfo()[c("BLAS", "LAPACK")]

# mine shows
# > $BLAS
# > [1] "~/software/libOpenBLAS/lib/"
# > [1] "~/software/libOpenBLAS/lib/"


The default BLAS and LAPACK libraries that ship with R are not the most efficient versions. There are decent comparisons on the internet showing substantial speedups depending on the task you are performing. In my particular case, I have observed a speedup of 10X when doing a single-value-decomposition on moderately sized matrices (1600 x 1600). Given what I’m doing, that is definitely worth it.



BibTeX citation:
@online{m flight2023,
  author = {M Flight, Robert},
  title = {Installing {OpenBLAS} for Self-Compiled {R}},
  date = {2023-11-01},
  url = {},
  langid = {en}
For attribution, please cite this work as:
M Flight, Robert. 2023. “Installing OpenBLAS for Self-Compiled R.” November 1, 2023.