Splitting up the changes to the polymorphic inlining code into a number of bugs, so that the final review isn't too crazy, and so that I don't have rebase hell along the way. Hopefully, it's even easier to review! This patch fixes a bunch of stylistic nits without affecting logic: 1) Cleans up getInlineableGetPropertyCache(). 2) Uses CallInfo in makeInliningDecision(). 3) Moves InlinePropertyTable::trimToAndMaybePatchTargets() out of headers. 4) Writes patchInlinedReturns() in terms of a helper function for clarity. 5) Cleans up CodeGenerator::visitPolyInlineDispatch(). 6) Fixes various nits in MIR.h.

Comment on attachment 712054 [details] [diff] [review] patch Hannes is away for a bit -- flipping review.

Comment on attachment 712054 [details] [diff] [review] patch Review of attachment 712054 [details] [diff] [review]: ----------------------------------------------------------------- Cool. Refactors and cleanups look pretty good. One concern I have is that the inlining spew seems to have been removed along with that. It'd be nice to have something which indicates that the optimization is being performed - as it represents some reasonably major abstract stack surgery... and for perf tracking purposes.

Thanks! The spew was moved into trimToAndMaybePatchTargets(). https://hg.mozilla.org/integration/mozilla-inbound/rev/cd80e0af25cd

Nice work Sean!