[R-SIG-Mac] vecLib-based BLAS not working with R 4.5.0

[Simon Urbanek]

TL;DR latest R 4.5.0 Patched (r88147 or higher) from https://mac.R-project.org should work.

Ulrich,

Thanks for the report. Yes, the whole purpose of the R shared BLAS is that you can point it to any BLAS implementation as long as it is complete enough for R to use. Unfortunately, R 4.5.0 now requires entry points which are not supplied by vecLib/Accelerate - which has not been noticed by those managing LAPACK/BLAS in R. This means vecLib can no longer be simply plugged in as it has in the past. In fact, very few implementations provide the required entry points (not even MKL), but OpenBLAS is one of the few that does.

I have now added --with-2025blas to the nightly builds which adds those entry points to R's LAPACK so they not required in the shared BLAS and thus the latest binaries should work again with vecLib.

Cheers,
Simon


> On Apr 16, 2025, at 1:17 AM, Ulrich KELLER <ulrich.keller using uni.lu> wrote:
> 
> Thanks a lot. Installing OpenBLAS through Homebrew and symlinking it worked fine (to my surprise) and it's almost as fast as vecLib.
> 
>> On 15 Apr 2025, at 13:49, Prof Brian Ripley <ripleybd using icloud.com> wrote:
>> 
>> This is covered in NEWS (and in detail in the R-admin manual).  See in
>> particular the NEWS section 'BLAS and LAPACK:'.
>> 
>> It seems no one who cares about this tested the nightly builds (and I
>> believe using OpenBLAS, which is complete and is documented, is to be
>> preferred).  Ideally Dr Urbanek would add the missing routines to
>> libRblas.vecLib.dylib: he does not publicise how he builds that.  Or R
>> could be built with --with-2025blas.
>> 
>> --
>> Brian D. Ripley,                  ripley using stats.ox.ac.uk
>> Emeritus Professor of Applied Statistics, University of Oxford
> 
> _______________________________________________
> R-SIG-Mac mailing list
> R-SIG-Mac using r-project.org
> https://stat.ethz.ch/mailman/listinfo/r-sig-mac
>