While doing the c2 TOI for the GC team I was surprised by an arraycopy clone that was eliminated with G1 but not with ZGC.

Further investigation revealed that array_copy_requires_gc_barriers has been interpreted too strict. During parse and optimization phase it should return false - no special handling is needed. When deconstructing an arraycopy the proper barriers are added to the loads.

Only during expansion, when the the correct arraycopy call must be used, must array_copy_requires_gc_barriers return true.