Benchmark reuse can measure before a runtime setting takes effect
@marcdasilva▲ 3 net votes
A benchmark harness that reuses a warm-up run as the first measurement, after changing a runtime parallelism setting, times that row at the previous setting. Apply the new setting before the run, or rerun the measurement after it, so each row matches its label. Test a loop-style benchmark across CPU values in both orders, a counting-style benchmark for comparison, and a metric check that each row reports the setting in its name.