Filter cache

[bmcfee]

This PR fixes #97 and adds functionality to reset the cache. (This should never be necessary, but it can be helpful for debugging.)

In implementing this, I had to rewrite an in-place multiply of the filter coefficients https://github.com/bmcfee/resampy/blob/56415951d25a539a6fb6f0065dfcf1379f6a3c8a/resampy/core.py#L117-L120 with a copy-multiply. This will cost us a little when downsampling, but is free for upsampling. I don't think the cost is any more than what we were already paying for redundant loading from disk each time, so it should be a net win. In my informal benchmarks, I'm seeing improvements on the order of 10ms.

I've put in a few more tests to bring up the coverage while I'm at it. In doing so, I found what I think to be a dead code path introduced in #82 , and I'd like @avalentino to confirm:

https://github.com/bmcfee/resampy/blob/56415951d25a539a6fb6f0065dfcf1379f6a3c8a/resampy/core.py#L206-L219

Failing on this last check would require len(t_out) < 1 (i.e., len(t_out) == 0), but this would fail earlier when trying to compute the min and max in the second check. (Numpy min or max of an empty array triggers a ValueError.) On the surface, we can safely remove this check, but I'm wondering if it's a bug and is intended to be checking something other than the domain check?

[avalentino]

@bmcfee , I agree, the last check (if shape[axis] < 1: ...) can be removed

[bmcfee]

Great, thanks for the quick response!

[codecov-commenter]

Codecov Report

Merging #99 (f559573) into main (5641595) will increase coverage by 0.56%. The diff coverage is 81.81%.

@@            Coverage Diff             @@
##             main      #99      +/-   ##
==========================================
+ Coverage   81.25%   81.81%   +0.56%     
==========================================
  Files           4        4              
  Lines          96       99       +3     
==========================================
+ Hits           78       81       +3     
  Misses         18       18              

Flag	Coverage Δ
unittests	81.81% <81.81%> (+0.56%)	:arrow_up:

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
resampy/core.py	82.00% <50.00%> (+1.23%)	:arrow_up:
resampy/filters.py	78.57% <88.88%> (+0.19%)	:arrow_up:

---

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update 5641595...f559573. Read the comment docs.