1980910 - Optimize JIT post barriers

Status: NEW

(Core :: JavaScript: GC, enhancement, P3)

Description

[Jon Coppeard (:jonco) (PTO until Monday 30th)]

There may be some scope to optimize post barriers in the JIT.

Most of the time we use the whole cell store buffer in the JIT. When adding an edge from a tenured thing to a nursery thing we call into the VM set a bit in a bitset associated with the tenured cell's arena, allocating the bitset if necessary. We can optimize this by setting the bit directly if the bitset is already present and skipping the call.

I did previously implement this and it had no discernable effect on performance. It would be good to come up with a micro benchmark is affected by this.

The pre-barrier already has a similar optimization to check whether a full barrier is required before calling into the VM.