This is a very popular way of reducing contention by giving each thread its own thread private space, and then combining all partial results in a single step. Other than reducing the synchronization costs, it also helps in reducing cache line invalidations, since each thread works on unique addresses. Thus, this method automatically gives us two different levels of improved performance.
This is a very popular way of reducing contention by giving each thread its own thread private space, and then combining all partial results in a single step. Other than reducing the synchronization costs, it also helps in reducing cache line invalidations, since each thread works on unique addresses. Thus, this method automatically gives us two different levels of improved performance.