1297604 - Consider hoisting some shape guards even if we had shape guard failures

Status: NEW

(Core :: JavaScript Engine: JIT, defect, P3)

Description

[Boris Zbarsky [:bzbarsky]]

Right now a shape guard failure anywhere in a function, even in cold code, will deoptimize (force to not be loop-hoistable) shape guards everywhere in the function.  This severely hampers LICM...

Could we do something smarter to avoid the problems bug 806793 was trying to do?  dvander mentioned not hoisting shape guards out of conditionals in there, for example...

Comment 1

[Tom S [:evilpie]]

We just talked about this this week. We really want to do this for every kind of guard. We need obviously need to track which guard failed, but we also need to figure out what to when guards are folded together.