Funny, I just experimented with __builtin_prefetch this week, and got no speedup. Does anyone know whether the kernel list.h always triggers hardware prefetching when going over a list?
BTW, my case wasn't the problem the kernel maintainers encountered with small lists, detailed here:
BTW, my case wasn't the problem the kernel maintainers encountered with small lists, detailed here:
http://lwn.net/Articles/444336/