If Flexiblas is e.g. using MKL, we'd want it to use MKL mkl_set_num_thread_local to set the number of threads.
Approaches could include:
- Bypassing Flexiblas and calling the MKL API ourselves (but what happens when the backend is changed?)
- Asking Flexiblas to add a thread-local API that would call underlying BLAS's thread-local-size-setting API when available.
If Flexiblas is e.g. using MKL, we'd want it to use MKL
mkl_set_num_thread_localto set the number of threads.Approaches could include: