Currently, volatile write is implemented using a normal write and a StoreLoad barrier (lock_add [rsp - 0x40], 0x0). This can be reduced to a single locked xchg, similar to how C++ compilers are doing.