Comment (by Roland Haas):

Oh, right. So that won’t quite prevent cache line sharing. Unfortunately I cannot check alignment with a static assert since both types are allocated dynamically (lc_thread_info_t via a new the other as part of a std::vector<lc_fine_thread_comm_t> (and I’d have no idea how that one would have interacted with the CCTK_ATTRIBUTE_ALIGNED(128)).

So I guess it would have to be using 2*128 or at least 2 times the cache line size if that one is know at compile time.

--
Ticket URL: https://bitbucket.org/einsteintoolkit/tickets/issues/2832/possible-race-condition-in-loopcontrol