multithreading - Why do we use dynamic memory in ccNUMA systems when we talk about data distribution into locality domains by first touch policy?

Question

In many books when they talk about firch touch policy in ccNUMA systems they are using dynamic memory allocation(dynamic arrays as example) when they distribute data across locality domains.What if for example have an array in stack?Does first touch policy work in the same way?

标签： multithreadingparallel-processingmultiprocessingopenmpnuma