Bug 161741
| Summary: | GetByIdWithThis/GetByValWithThis should be faster when monomorphic than when polymorphic | ||
|---|---|---|---|
| Product: | WebKit | Reporter: | JF Bastien <jfbastien> |
| Component: | JavaScriptCore | Assignee: | Nobody <webkit-unassigned> |
| Status: | NEW | ||
| Severity: | Normal | ||
| Priority: | P2 | ||
| Version: | WebKit Nightly Build | ||
| Hardware: | Unspecified | ||
| OS: | Unspecified | ||
JF Bastien
Follow-up to https://bugs.webkit.org/show_bug.cgi?id=160922
The above patch added value profiling to the two opcodes, they got faster, but the new tests have odd behavior:
$ ./Tools/Scripts/run-jsc-benchmarks old:./before/bin/jsc new:./current/bin/jsc --benchmarks "super-get-by-" --microbenchmarks --outer 400
old new
super-get-by-id-with-this-monomorphic 29.3852+-0.2030 ^ 27.1366+-0.1896 ^ definitely 1.0829x faster
super-get-by-id-with-this-polymorphic 27.0021+-0.2115 ^ 25.6205+-0.1965 ^ definitely 1.0539x faster
super-get-by-val-with-this-monomorphic 29.8851+-0.1339 ^ 28.9029+-0.1612 ^ definitely 1.0340x faster
super-get-by-val-with-this-polymorphic 27.6204+-0.1616 ? 27.8924+-0.2388 ?
<geometric> 28.4117+-0.0838 ^ 27.3187+-0.0971 ^ definitely 1.0400x faster
I'd expect monomorphic to be faster than polymorphic, for both test, before as well as after the patch. The patch isn't a regression (poly was faster than mono before it), but we should investigate why this is happening. The difference seems to be that there were many more exits in mono than poly as shown by the profile from that patch:
https://bugs.webkit.org/attachment.cgi?id=287636
| Attachments | ||
|---|---|---|
| Add attachment proposed patch, testcase, etc. |