3 ms·
Huh, I didn't know about this new primitive. Thanks! It seems to have been introduced into Windows 8 and later, as a competitor against Linux's futex. I did a
by dragontamer 8y ago
Huh, I didn't know about this new primitive. Thanks! It seems to have been introduced into Windows 8 and later, as a competitor against Linux's futex.
I did a brief search online, and Raymond Chen had this to say:
https://blogs.msdn.microsoft.com/oldnewthing/20160825-00/?p=94165 https://blogs.msdn.microsoft.com/oldnewthing/20160825-00/?p=...
> [snip] in real life, you should just use EnterCriticalSection because it has stuff like spin counts and lock convoy resistance.
Is there anything that pthread_mutex_lock does that EnterCriticalSection doesn't do? It seems like a good "translation" to me. I'd only use the WaitOnAddress thing if pthread_mutex_lock had some features that a raw EnterCriticalSection wouldn't do.
- caf 8y agoYou need to be able to link your pthread condition variable implementation to your pthread mutex implementation such that a pthread_cond_wait() does an atomic wait-on-condition-variable-and-unlock-mutex, noting that another thread can signal the condition variable without holding any mutex. In addition, pthread mutexes can be statically initialised which requires some kind of initialise-on-first-use for those if you implement them using Critical Section objects. I believe WaitOnAddress(), rather than being a "competitor" for futex was actually added (or at least, the underlying infrastructure was) in order to be able to implement futex() in the WSL.
- dragontamer 8y agoAh yes, I forgot about condition variables. That makes sense then.