880891 - IonMonkey: Only set WatchObjectStateChange when we are certain we will inline a function

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Hannes Verschore [:h4writer]]

We are setting WatchObjectStateChange to trigger recompilation of inlined scripts. But we do this already before our heuristics decide if we really want to inline that function. As a result all inlineable functions will trigger recompilation (even if it is decided not inline that function). 

I noticed this when looking at ARM Box2D. Probably more noticable because the box testing didn't do parallel compilation (one core? or parallel compilation not enabled?). But the multiple bailouts are also on x86!

This should remove the ARM 40% regression on Box2D

Comment 1

[Hannes Verschore [:h4writer]]

Attachment: Patch

Comment 2

[Kannan Vijayan [:djvj]]

Comment on attachment 760007 [details] [diff] [review]
Patch

Review of attachment 760007 [details] [diff] [review]:
-----------------------------------------------------------------

Nice find.

Comment 3

[Hannes Verschore [:h4writer]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/313e1ff9266d

Comment 4

[Ed Morley [:emorley]]

https://hg.mozilla.org/mozilla-central/rev/313e1ff9266d