Open HadrienG2 opened 8 years ago
Ideally, this could be turned into more general hardware topology support, where using the library a client could...
Now that SMP code is isolated and explicitly marked as such, this task isn't so critical as to be needed for M1.
On any system with CPU caches, moving threads from one CPU core to another is expensive. For NUMA systems, the cost becomes prohibitive. We really want to avoid that, and it should be possible to do so using the new Ada 2012 features. To be investigated!