Resolves #317.

Any way to test that the change is actually working?

From #317

As noted on this SO comment, Accelerate is slightly slower on Intel macOS than OpenBLAS, which is the current version used. Using MKL BLAS on Intel macOS platforms would improve their performance considerably.

Then maybe the change should be conditional on the platform (i.e. Accelerate only enabled on Apple silicon)?

I can do some timing on a simulation I'm working on, however it is with 2312, which will have to built separately.

Baseline OpenFOAM v2312 simulation: 15m23s

Now to rebuild OpenFOAM v2312 with blas[accelerate] and retest.

With blas[accelerate] I get 8m36s.

That speedup was too substantial, so I re-ran the simulation 3 times with OpenBLAS and 3 times with Accelerate BLAS:

Time reduction: 5.85%

I have updated the PR to use platform-specific BLAS implementations. Accelerate is used on the Apple Silicon platforms and BLIS was selected for Intel as conda-forge does not have a MKL distribution of BLAS for macOS.

Time reduction: 5.85%

Thanks for the work! I'm fine with doing the change then.

BLIS was selected for Intel

Any reason why not to leave it to the whatever is the default in conda on that platform?

The rationale is that in the SO comment referenced above there was a performance improvement of 20.9% when selecting BLIS over OpenBLAS.

The rationale is that in the SO comment referenced above there was a performance improvement of 20.9% when selecting BLIS over OpenBLAS.

Great. Then it makes sense to me that we make that change too.

@oursland Thanks for the work!