CODEWITHSUNDEEP
rblas
Leigh K
Leigh K

Reputation: 603

Linking R with AMD BLIS Library (in Ubuntu)

I would like to link the AMD BLIS library with R to perform some benchmarking/profiling. It seems that R does not need to be compiled specifically against other BLAS implementations. For example, other BLAS libraries (OpenBLAS, Intel MKL, Atlas) simply need to be installed and the BLAS implementation being used can be easily changed via update-alternatives. For example, this web page walks through several BLAS implementations:

https://csantill.github.io/RPerformanceWBLAS/

And the BLAS library used in R can be verified via sessionInfo(). However, the BLIS library is not an option that comes up once it is installed via update-alternatives. I cannot find a guide for linking AMD BLIS with R anywhere. The only other post that I can find mentioning it does not explain how it was achieved:

R performance on Ryzen+Ubuntu: openBLAS/MKL, Rcpp and other improvements?

I am working on an AMD Ryzen threadripper 3990x 64-core processor × 128 and I would like to see how the BLIS library performs against the other options. Any assistance in linking R with BLIS would certainly be appreciated.

Upvotes: 3

Views: 693

Answers (1)

DBScan
DBScan

Reputation: 46

I ran into the same problem as you last week. From my understanding, R needs to get compiled with BLIS. I did the configuration step like this: ./configure --with-blas="-lblis -I{PATH_TO_BLIS}/include -L{PATH_TO_BLIS/lib" --with-lapack="llapack". I think it is necessary to specifiy -with-lapack="llapack", because BLIS doesn't include a LAPACK implementation.

Upvotes: 3

Related Questions