OpenBLAS on in Windows without admin privileges

OpenBLAS-Windows-user.md

If you work on a Windows computer without admin privileges to modify system files (e.g., many government employees), you can likely still install R in your user folder and replace the default BLAS library there.^* Instructions:

1. Download the latest R installer.
2. Run the installer and select a user folder for installation when prompted, e.g. C:\Users\USERNAME\SUBFOLDER.
3. In RStudio go to 'Tools' > 'Global Options' > 'General' > 'R version' to change the R installation being used to the one you just installed in your user folder.
4. Download OpenBLAS following these instructions. A summarized version of this and subsequent steps follows.
5. Extract libopenblas.dll from the OpenBLAS zip file.
6. Rename or move Rblas.dll and Rlapack.dll from the R\R-4.X.X\bin\x64 subfolder (where X.X represents version numbers). Make sure you are doing so from the R installation in your user folder that you just made.
7. Paste two copies of libopenblas.dll into the same folder as above and rename them Rblas.dll and Rlapack.dll.
8. Restart R/RStudio.

More detailed instructions from step 3 onwards are available here.

Test to make sure OpenBLAS is being used:

m <- 1e4; n <- 1e3; k <- 3e2
X <- matrix(rnorm(m*k), nrow=m); Y <- matrix(rnorm(n*k), ncol=n)
system.time(X %*% Y)

The result should look something like:

##   user  system elapsed 
##  0.169   0.011   0.165 

The standard BLAS shipped with R will likely take 1-2 seconds for the above.

_{^*Thanks to Paul Regular and Semra Yalcin for figuring this out and to Seb Pardo for drafting the step-by-step instructions.}

You may find that openBLAS uses all available cores, which may not be computationally efficient with TMB. If so, you can specify BLAS should only use 1 core with the RhpcBLASctl package:

RhpcBLASctl::blas_set_num_threads(1)
RhpcBLASctl::omp_set_num_threads(1)

You can either set that at the beginning of each script or (better yet) place it in your .Rprofile using:

usethis::edit_r_profile()