Static copy
Copied from docs/en/performance/baseline.md in https://github.com/poppy-team/aipo-lang (MIT). Pinned to revision ce47c4e070404fc41722ccd8e2b7b808b99062a8, blob c6e642b83682ac5de5d8d6268c6da56de6d0ef2f. The source repository remains canonical; this copy is refreshed through a sync pull request, not live.
Aipo Performance Baseline
Status: record (generated artifact, not a gate) Scope: wall-clock baselines for frontend stages, commands, VM workloads, the JS backend and the separate cross-language suite Environment: Linux x86_64 (4 CPU, 5 GiB RAM), rustc 1.98.1, Node v24.18.0, --release, shared runner (noisy — see MAD column) Recorded: 2026-09-20, cargo run --release -p aipo-bench (7 samples, median/MAD), git a8709881
These numbers are baselines for future comparison on a dedicated runner, not pass/fail thresholds. For the multi-runtime protocol, raw samples, checksums and fairness rules, see docs/performance/cross-language.md.
Frontend stages (median)
| stage | 858 B | 9 KiB | 99 KiB | 1023 KiB |
|---|---|---|---|---|
| bytecode/compile+verify | 34.12µs | 347.30µs | 3.91ms | 45.94ms |
| formatter/format | 50.16µs | 561.38µs | 9.55ms | 78.44ms |
| hir/lower | 24.52µs | 224.03µs | 4.18ms | 41.71ms |
| ir/lower | 19.97µs | 239.95µs | 4.27ms | 39.82ms |
| lexer/tokenize | 24.13µs | 275.36µs | 5.36ms | 58.56ms |
| sema/check | 14.65µs | 105.30µs | 2.43ms | 24.32ms |
| source/new | 1.90µs | 15.45µs | 179.48µs | 2.23ms |
| syntax/parse | 77.95µs | 861.32µs | 11.42ms | 102.74ms |
Commands and VM workloads (median)
| workload | input | median | MAD |
|---|---|---|---|
| cmd/check | 858 B | 188.04µs | 7.59µs |
| cmd/run | 858 B | 349.31µs | 6.43µs |
| cmd/build-emit | 858 B | 483.31µs | 4.75µs |
| cmd/fmt-check | 858 B | 94.61µs | 2.92µs |
| vm/int-arith | 43 B | 217.06µs | 1.62µs |
| vm/float-arith | 46 B | 206.48µs | 1.16µs |
| vm/call-overhead | 61 B | 223.84µs | 625ns |
| vm/recursion | 75 B | 13.79ms | 315.07µs |
| vm/closure-shared-var | 94 B | 274.68µs | 12.55µs |
| vm/global-lookup | 45 B | 218.07µs | 7.66µs |
| vm/list-build-iter | 74 B | 503.85µs | 33.07µs |
| vm/dict-insert-lookup | 78 B | 226.62µs | 2.02µs |
| vm/string-concat | 39 B | 477.75µs | 2.80µs |
| vm/interpolation | 43 B | 222.32µs | 2.16µs |
| vm/bytes-alloc | 49 B | 77.50µs | 611ns |
| vm/pipeline | 40 B | 142.71µs | 389ns |
| vm/failure-propagate | 112 B | 175.88µs | 180ns |
| vm/attempt | 68 B | 106.75µs | 262ns |
| vm/contracts | 71 B | 179.25µs | 1.07µs |
| vm/invariant-commit | 167 B | 212.84µs | 1.34µs |
JS backend (frontend + emit + node spawn + run)
| workload | input | median | MAD |
|---|---|---|---|
| js/emit | 858 B | 288.64µs | 13.93µs |
| js/emit | 9 KiB | 4.00ms | 178.16µs |
| js/emit | 99 KiB | 57.60ms | 4.30ms |
| js/emit | 1023 KiB | 576.11ms | 16.64ms |
| js/hello | 17 B | 57.64ms | 5.16ms |
| js/integrated | 858 B | 72.06ms | 6.68ms |
Node process spawn dominates small workloads (~50–70 ms total). A medição cross-language separa Aipo VM in-process, Aipo CLI/VM e Aipo→JavaScript; comparações de steady-state ainda exigem runner dedicado.
Scaling (list build/iterate, dict insert, N=200..1600)
All ratios near 2× per doubling on this runner — linear behavior; verdicts report noise honestly. Raw ratios:
list-build: N=200..1600 ratios [1.73x, 2.17x, 1.66x] verdict=noisy — rerun on dedicated runnerlist-iterate: N=200..1600 ratios [1.88x, 2.07x, 1.86x] verdict=noisy — rerun on dedicated runnerdict-insert: N=200..1600 ratios [1.75x, 1.87x, 2.02x] verdict=noisy — rerun on dedicated runner
Notable finding during this gauntlet
Format-string parsing scaled 68× per 10× input (measured 23.4 ms for 400 f-strings over 8.4 KiB). Root cause: per-placeholder " ".repeat(offset) padding plus full rescans. Fixed by parsing slices unpadded and shifting spans back (crates/aipo-syntax/src/shift.rs, ADP-005 addendum): 1.77 ms after, offset-proportional cost eliminated, corpus snapshots byte-identical.