opp-repl-speed-tests

You can detect performance regressions in CI by measuring CPU instruction counts instead of wall time. Instruction counts provide deterministic performance verification across builds, overcoming the inconsistency of noisy environments.

You measure CPU instruction counts using Linux perf counters. By tracking these counters during a simulation, you get a deterministic metric to compare against a stored baseline for reliable regression detection.

CPU instruction counts are used instead of wall time because they offer higher determinism. Wall time fluctuates in noisy environments, while instruction counts reliably quantify performance regressions across different builds.

Yes, you need Linux perf support to run these speed tests. The skill relies on Linux perf counters to measure the CPU instruction counts required for its deterministic performance regression detection.

To set up a baseline for speed tests, you seed it using the update_speed_test_results interface. Once the baseline is stored in speed_store, you run subsequent tests with run_speed_tests to compare against it.

Yes, you can verify performance across different build configurations in CI. The skill compares CPU instruction counts against a stored baseline, providing deterministic regression checks across various builds.