Add info on threading

[danielwe]

Thought it was worth putting this information in the readme. Perhaps @staticfloat can confirm that the wording is correct?

I can't find any official documentation on the VECLIB_MAXIMUM_THREADS variable (this search comes up dry: https://developer.apple.com/search/?q=VECLIB_MAXIMUM_THREADS&type=Documentation), but it's all over the forums, and I can confirm that it works.

Btw. is there a technical reason why BLAS.{set,get}_num_threads isn't hooked up to Accelerate's API for this? (See https://developer.apple.com/documentation/accelerate/blas?language=objc#4434515) (EDIT: I mean apart from the fact that you can only select single vs multiple threads rather than set an exact number)

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 0.84%. Comparing base (fc7bfa5) to head (72e5050).
Report is 2 commits behind head on master.

❗ There is a different number of reports uploaded between BASE (fc7bfa5) and HEAD (72e5050). Click for more details.

HEAD has 2 uploads less than BASE

Flag	BASE (fc7bfa5)	HEAD (72e5050)
	4	2

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #75       +/-   ##
==========================================
- Coverage   73.22%   0.84%   -72.39%     
==========================================
  Files           4       4               
  Lines         239     238        -1     
==========================================
- Hits          175       2      -173     
- Misses         64     236      +172     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[staticfloat]

is there a technical reason why BLAS.{set,get}_num_threads isn't hooked up to Accelerate's API for this?

The reason is that back when I wrote this code, those functions didn't exist. :) Those are only available from macOS 15.0+ , which is quite recent. That being said, I totally think we should use it if possible, since it's the only hope for dynamic thread control. I'm thinking we expose it as something like, if BLAS.set_num_threads(1) is called, we can try setting it to single-threaded, and if any other value is used, set it to multithreading. It's a little confusing, but better than what we have right now, IMO.

[staticfloat]

To be more clear about that last suggestion, that would be a PR to libblastrampoline, adding another special case at the end of set_num_threads().

[danielwe]

Yeah I think everything about this PR is really implemented (or not) in LBT, which is why I specifically pinged you to verify that my additions weren't nonsense 👍