Closed computablee closed 10 months ago
All modified and coverable lines are covered by tests :white_check_mark:
Comparison is base (
20c89b1
) 99.19% compared to head (8444e6b
) 99.20%.
:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.
Which issue are you addressing?
Closes #108
How have you addressed the issue?
The approach taken before to optimize
collapse(2)
andcollapse(3)
was implemented forcollapse(4)
and higher.How have you tested your patch?
All unit tests pass. Performance is greatly improved. Tests were performed with BenchmarkDotNet running an
NxNxNxN
parallel kernel with an empty body and withN
being 20, 40, 80, and 160. Static scheduling was used with the default chunk size.This is the performance pre-optimizations:
And post-optimizations:
Not only is there a >2x performance improvement, but the error and standard deviation is down significantly, making performance more predictable.