Comment (by Erik Schnetter):

We don’t need new or std::vector. We replace the call to new with posix_memalign, and the call to delete by free. Then each thread has one cache line as intended.

Alternatively, there could be one call to posix_memalign allocating memory for all threads, and each thread uses pointer arithmetic to find its place.

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