You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Previously, measure() only recorded timings for indices in the middle
range [DROP_SIZE, N_MEASURES - DROP_SIZE), while update_statistics()
assumed all entries were starting available from index 10. This
mismatch allowed zero-valued exec_times from unmeasured head and tail
indices to be included in the t-test, reducing sample means, inflating
variances, and suppressing the t-value, which may lead to
incorrect results or prevent detection thresholds from being reached.
After the fix, all samples are measured and
update_statistics() discards DROP_SIZE samples at both ends. This
ensures correct sample accounting, prevents overestimating the number
of measurements required, and avoids false negatives due to
uninitialized timing data.
Change-Id: Ibb1515043da5f56d72fe34fd5c78e2283df9a993
0 commit comments