AD Performance

[willtebbutt]

Summary

Extends work from #466 to many (nearly all) of our kernels. Identifies various performance bugs in the process, and fixes many of them.

Proposed changes

• ... Introduces the test_interface_ad_perf function in test/test_utils.jl. This utilises the allocations-invariant-to-input-size heuristic to check performance.

What alternatives have you considered?

N/A

Breaking changes

None

[codecov[bot]]

Codecov Report

Merging #467 (2e560ff) into master (1831cc6) will increase coverage by 0.61%. The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #467      +/-   ##
==========================================
+ Coverage   93.09%   93.70%   +0.61%     
==========================================
  Files          52       52              
  Lines        1275     1351      +76     
==========================================
+ Hits         1187     1266      +79     
+ Misses         88       85       -3     

Impacted Files	Coverage Δ
src/TestUtils.jl	94.73% <100.00%> (+0.61%)	:arrow_up:
src/basekernels/fbm.jl	100.00% <100.00%> (ø)
src/basekernels/nn.jl	98.00% <100.00%> (+0.38%)	:arrow_up:
src/basekernels/polynomial.jl	100.00% <100.00%> (ø)
src/basekernels/rational.jl	100.00% <100.00%> (ø)
src/kernels/normalizedkernel.jl	82.50% <100.00%> (+2.50%)	:arrow_up:
src/kernels/scaledkernel.jl	88.23% <100.00%> (ø)
src/transform/ardtransform.jl	93.33% <100.00%> (+1.02%)	:arrow_up:
src/utils.jl	91.46% <100.00%> (+0.10%)	:arrow_up:
src/chainrules.jl	87.65% <0.00%> (+3.70%)	:arrow_up:

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

[theogf]

Don't forget to bump the version.

[willtebbutt]

@devmotion thanks for the careful review -- I believe I've resolved everything that you discussed

[willtebbutt]

All CI appears to be passing at last!