mcid.py: optimize FFT and A-weighting calculations

[deanlee]

Precompute A-weighting filter coefficients for reuse in each callback to enhance performance and reduce computational overhead.

[github-actions]

Thanks for contributing to openpilot! In order for us to review your PR as quickly as possible, check the following:

• Convert your PR to a draft unless it's ready to review
• Read the contributing docs
• Before marking as "ready for review", ensure:
  • the goal is clearly stated in the description
  • all the tests are passing
  • the change is something we merge
  • include a route or your device' dongle ID if relevant

[adeebshihadeh]

Do you have before/after on some benchmark? We need that for any optimization PR.

[deanlee]

I have not tested the reduced CPU usage of this PR on the device separately. I only measured the reduced time, and it saves approximately 1.5 ms on every update on the device.

As for CPU usage, I tested it along with another PR (not yet submitted) that replaces np.concatenate with a circular buffer to efficiently manage incoming audio samples, avoiding costly array resizing operations. Together, these two PRs can reduce CPU usage by half (about reduce 4% cpu usage)

[adeebshihadeh]

test_onroad.py is a good benchmark for this

openpilot/selfdrive/test/test_onroad.py

Line 59 in 0fa6745

"system.micd": 6.0,

[adeebshihadeh]

trigger-jenkins

[adeebshihadeh]

test_onroad says the CPU went to 3.9% from 5%, so definitely an improvement! But the memory usage check failed:

[deanlee]

Not sure what the issue is. The cached calculation result A_WEIGHTING in the global variable is only about 4KB.

update: The memory size of the array A_WEIGHTING is 32KB. Is this causing the issue?

[sshane]

jenkins passed on a subsequent run w/ no changes, flaky test? @adeebshihadeh

[adeebshihadeh]

I haven't seen that test fail randomly before. We can run a few more times to confirm though

[sshane]

Passed 5 subsequent runs across the day. @deanlee can you make it use cache?

[deanlee]

Completed. The function now uses cache.

[sshane]

trigger-jenkins