numpy, blas

[Greg Troxel <gdt%lexort.com@localhost>]

As part of Home Assistant I am building numpy 1.26.0 (within a venv,
because HA is like that), and I discovered that not only does it object
to gcc7, but to regular blas, erroring out that it wants openblas.

1) When someone updates it, they'll have to tweak

  BLAS_ACCEPTED=         ${_BLAS_TYPES} accelerate.framework


2) As a random person who has a linear algebra textbook but has not been
studying the 12 flavors of BLAS, what shoudl I choose?  It seems the
choices are

  openblas
  openblas_pthread
  openblas_openmp

and they are all the same codebase, but built

  not to try parallelism
  to use multiple threads
  to use openmp, which presumably uses threads, but perhaps more smartly


My case is that I don't even know why HA needs numpy, and that I lean to
simpler, and I have picked openblas.  I'm guessing that  if my world
were cpu-bound in blas calls, I'd want pthread or openmp, and that
openmp is probably better, but I'm really guessing.