Skip to content

bench: report minimum + gcsample heavy benchmarks to cut CI variance #18

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
oameye merged 1 commit into from
Jun 24, 2026
Merged

bench: report minimum + gcsample heavy benchmarks to cut CI variance #18

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 5 additions & 1 deletion benchmarks/correlations.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -59,8 +59,12 @@ function benchmark_correlations!(SUITE)

SUITE["Correlations"]["two-time"] = BenchmarkGroup()

# Allocation-heavy: a full master-equation solve that rebuilds operators every
# ODE step (~0.5 GB / call). `gcsample=true` runs a GC before each sample so the
# timing isn't dominated by non-deterministic GC pauses carried over between
# samples, which otherwise gives this benchmark a ~4x run-to-run spread.
SUITE["Correlations"]["two-time"]["single photon cavity"] =
@benchmarkable correlation_matrix($T, $ρt, $input_output, $Ls)
@benchmarkable correlation_matrix($T, $ρt, $input_output, $Ls) gcsample = true

return nothing
end
4 changes: 3 additions & 1 deletion benchmarks/interaction_picture.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,8 +40,10 @@ function benchmark_interaction_picture!(SUITE)

SUITE["Interaction Picture"]["coefficient matrix M"] = BenchmarkGroup()

# Allocation-heavy ODE solve: GC before each sample to stabilize the timing
# (see the note in runbenchmarks.jl).
SUITE["Interaction Picture"]["coefficient matrix M"]["numerical (ODE)"] =
@benchmarkable solve_mode_evolution($A_uv, $T)
@benchmarkable solve_mode_evolution($A_uv, $T) gcsample = true

SUITE["Interaction Picture"]["coefficient matrix M"]["analytical (2 equal modes)"] =
@benchmarkable solve_mode_evolution_symmetric($u, $T)
Expand Down
12 changes: 10 additions & 2 deletions benchmarks/runbenchmarks.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,6 +24,14 @@ benchmark_correlations!(SUITE)

BenchmarkTools.tune!(SUITE)
results = BenchmarkTools.run(SUITE; verbose = true)
display(median(results))

BenchmarkTools.save("benchmarks_output.json", median(results))
# Report the minimum rather than the median. The minimum is BenchmarkTools'
# recommended estimator for tracking: measurement noise (GC pauses, scheduler
# preemption, frequency scaling) is strictly additive, so the minimum is the
# most reproducible estimate of the underlying cost and the least sensitive to
# cross-runner variance. Allocation-heavy benchmarks additionally set
# `gcsample=true` so each sample starts from a clean heap (see the individual
# `@benchmarkable`s).
display(minimum(results))

BenchmarkTools.save("benchmarks_output.json", minimum(results))
Loading