9 Dec
2011
9 Dec
'11
6:15 p.m.
El 09/12/2011 11:51, Tim Blechmann escribió:
btw, helge was planning to provide a separate implementation of boost.atomic for interprocess, because his blocking emulation of atomics depends on a per-process spinlock pool.
Ok, for Interprocess internal purposes I just need 32 bit integer atomics (maybe 64 bit ones for 64 bit platforms), and I think every platform compatible with Interprocess (UNIX and Windows) will support them without spinlocks. I guess the solution for interprocess would be to use a spinlock per atomic variable, making sizeof(T) != sizeof(atomic<T>). I'm just thinking about ABI issues. Best, Ion