The geomean computation that was incorrectly suggested by @bollu failed to take into account that the TotalRunTime counter was adding the real runtimes, not multiplying them as one should for geomean.
We change the geomean computation algorithm, to accumulate the log of the runtimes, and then divide by n followed by exponentiation as a final step for better numerical accuracy.
Description:
The
geomeancomputation that was incorrectly suggested by @bollu failed to take into account that theTotalRunTimecounter was adding the real runtimes, not multiplying them as one should for geomean.We change the
geomeancomputation algorithm, to accumulate thelogof the runtimes, and then divide bynfollowed by exponentiation as a final step for better numerical accuracy.$$ g \equiv (t_1 \times t_2 \times \dots \times t_n)^{1/n} = \exp \bigg(\frac{\log t_1 + \log t_2 + \dots \log t_n}{n} \bigg) $$
Testing:
What tests have been run? Did
make allsucceed for your changes? Was conformance testing successful on an Aarch64 machine?Conformance succeeds.
License:
By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.