The write barrier includes a test for whether the LHS has pointers, we could remove that test if we had a specialized write barrier for leaf objects. This could also be employed for Vector<Number> write barriers. Further more if all leaf objects were guarnateed to be written with this specialized write barrier we could make the normal write barrier faster by assuming that the LHS does contain pointers. We'd have to tighten up the type system so that all write barriers were done with types from GCObject.h instead of void* but that would be a nice change too. This would be a nice step on the path towards have write barriers small enough to be inlined (in C++ and in the JIT).
Assignee: nobody → treilly
Status: NEW → ASSIGNED
OS: Mac OS X → All
Priority: -- → P2
Hardware: x86 → All
Target Milestone: --- → Q4 11 - Anza
Vector<Number> doesn't involve atoms so that idea was bunk, but it does apply to the atomWriteBarrier for Number atoms.
Target Milestone: Q4 11 - Anza → Q1 12 - Brannan
This bug is bogus, we used to look at the LHS but now we only look at the RHS so this isn't valid.
Status: ASSIGNED → RESOLVED
Last Resolved: 7 years ago
Resolution: --- → INVALID
You need to log in before you can comment on or make changes to this bug.