C++11 atomic locks

[ZirconiumX]

There are several locks used for chess programs, spinlocks, mutexes etc.

So, as I am currently implementing SMP into FruitFly, would it be worth the hassle of converting the Toga mutexes to C++11 atomic locks?

Matthew:out

[mcostalba]

There are several locks used for chess programs, spinlocks, mutexes etc.

So, as I am currently implementing SMP into FruitFly, would it be worth the hassle of converting the Toga mutexes to C++11 atomic locks?

Matthew:out

Considering that you failed to understand that C++1 atomic library does not cover lock functionality I'd strongly suggest to stay with what you know better (or at least you think to know).

[bob]

There are several locks used for chess programs, spinlocks, mutexes etc.

So, as I am currently implementing SMP into FruitFly, would it be worth the hassle of converting the Toga mutexes to C++11 atomic locks?

Matthew:out

First question is, what is your goal? Max performance on a dedicated machine? Use nothing but spinlocks. If you are going to compete for CPU resources with other running processes (non-chess processes) then a traditional mutex that spins and then blocks is the way to go... I only care about the former myself, not the latter.

[Daniel Shawul]

OpenMP locks have an interesting property. They act as spinlocks for a certain time and then as a mutex in which the thread is put to sleep. They will probably act fully as spin locks when it is our turn so no loss there. http://www.thinkingparallel.com/2006/09 ... roduction/

[Daniel Shawul]

That was the conclusion of a recent Dr Dobb's article after discussing concurrency in C11. http://drdobbs.com/cpp/232800444?pgno=1