Re: Futexes ain't fast

On 9/26/2024 10:44 AM, Bonita Montero wrote:

Am 26.09.2024 um 19:40 schrieb Chris M. Thomasson:
 

Basically right. Think of hitting a slow path (before resorting to the kernel), then trying to spin a couple of times in user space before we have to hit a kernel call. Akin to a adaptive mutex. If we can avoid a kernel call, well, go ahead and try a couple of times...

 Adaptive mutexes often don't make any sense. The locking-frequency has
to be high and the locked-interval needs to be very short to make spin-
ning likely to succeed. That's both not often.

Adaptive mutexes with the right spin counts do make a difference!

 

Actually, there is a "method" to do this even with a simple mutex via try_lock. ...

 I don't know what try_lock is good for. I've never seen someone using
try_lock. If you need to lock sth. you usually haven't an alternative
to locking.
 

I just showed you one use of try_lock. This is just one out of many. try_lock is useful.