Phase 18. Performance gate
| Field | Value |
|---|---|
| MEP | MEP-45 §Phases · Phase 18 |
| Status | NOT STARTED |
| Started | — |
| Landed | — |
| Tracking issue | — |
| Tracking PR | — |
Gate
Median fixture wall-clock time on the BG corpus is within 2x of the equivalent Go-backend build, on x86_64-linux-gnu and aarch64-darwin.
Goal-alignment audit
To be written before sub-phase 18.0 starts. Performance gate exists so a regression cannot ship silently; the user-facing payoff is "your native build is at least as fast as the Go-embedded one". Aligns.
Sub-phases
| # | Scope | Status | Commit | PR |
|---|---|---|---|---|
| 18.0 | Benchmark harness: tests/transpiler3/c/bench/ with BG kernels (sum_loop, fib_iter, hello_world, ...) | NOT STARTED | — | — |
| 18.1 | Wall-clock, peak RSS, binary size (release/strip), compile time recorded per fixture | NOT STARTED | — | — |
| 18.2 | Per-release report published to a static page | NOT STARTED | — | — |
| 18.3 | Regression alert: > 10% wall-clock regression vs previous main posts a comment on the PR | NOT STARTED | — | — |
Decisions made
Fill in along the way.
Deferred work
Tighter (1.5x) gate: revisit after Phase 19 with measured data.
Closeout notes
Fill in after gate green.