# JavaScriptCore on Haiku riscv64 (QEMU TCG, -smp 4): interpreter vs baseline JIT
# image=/opt/brain/workspaces/scrumai/34d00b67-488e-43fd-a238-6c75dac3f561/upstream/haiku/generated.riscv64/haiku-mmc.image generated=2026-09-22T07:27:38Z script=evidence/webkit/jit-bench.js
# Each phase self-calibrates to ~1 s on the engine running it, so compare ms per unit of work,
# not the raw ms. proploop: property access + float arithmetic; sort: Array.prototype.sort.

engine                        proploop iters  proploop ms   ns/iter    sort n   sort ms   us/elem
interpreter (JIT=OFF build)           910716         2977    3268.9    106701      2841     26.63
JIT build, --useJIT=false             292121         1001    3426.7     38187       962     25.19
JIT build (baseline JIT)             3394720          991     291.9     37800       910     24.07

speedup proploop (interpreter ns/iter / JIT ns/iter) = 11.20x
speedup sort     (interpreter us/elem / JIT us/elem) = 1.11x
fixed-size checksums identical on interpreter / JIT / JIT --useJIT=false: yes  (proploop 2148901.350499, sort_hash 3995133556)
functions the baseline JIT compiled on the guest (--reportBaselineCompileTimes): propLoop, makeArray, now
sanity (print through each engine): {'int': True, 'jit': True}
raw outputs: evidence/webkit/jit-bench/*.txt, report.json, serial.log