mirror of
https://github.com/NVIDIA/nvbench.git
synced 2026-07-17 09:08:01 +00:00
Permit nvbench-compare to use bulk timing samples to confirm SAME even when sample frequency sidecars are unavailable. This lets legacy --jsonbin outputs, which recorded sample times but not frequencies, avoid AMBG results when summary noise is high but timing coverage clearly supports equality. Cycle coverage is still required when frequency samples are present, and FAST/SLOW clear-gap confirmation remains cycle-based. Add regression coverage for noisy bulk samples that exceed the SAME noise ceiling but are classified SAME from bulk timing/cycle coverage, including the timing-only legacy-data path.
124 KiB
124 KiB