• Fix up various references of 'tint' to 'dawn' post-migration.
• Round the CPU speed to 100MHz. This appears to fluctuate on reboots
for some machines.
• Fix git.Log() - the Count was trimming from the wrong end, leading
to perfmon spamming the github repo with repeat results.
• Instead of using google-benchmark's `--benchmark_repetitions` flag
for repeating benchmarks, calculate averages by re-running the
benchmark executable. Use '--benchmark_enable_random_interleaving'
to randomize the order in which benchmarks are run, which greatly
helps reduce noise in the averaged results.
• If the host machine supports CPU temperature sensors, wait for
thermals to stabilise before running the benchmarks. Further helps
reduce result noise.
• Breakout of the historic benchmarking loop every 15 minutes to check
for new Gerrit changes to benchmark.
• When idle, attempt to re-benchmark historic results that are 'spiky'
to reduce noise in the graphs.
• Specify more CMake flags to avoid building non-benchmark executables.
• Update the default base change for historic results. Attempting to
use the old hash, which was prior to the tint -> dawn merge makes,
the git log go back to T0 on the dawn branch. We don't want to
benchmark nearly 3k changes.
Change-Id: I8e59c7838720eb8bd11f217e9bd3104ba1eb51c3
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/87642
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>
Commit-Queue: Ben Clayton <bclayton@google.com>
368b3eaae4