Hi there, Just for heck of it - I am trying to come up with a ConcurrentCounter that does not suffer memory sharing or the cost of mutex lock/unlock.
The idea as described in the following Java code snippet is quite straightforward: - Pin every thread to a unique int32 variable. - Put these int32 variable in an array with sufficient 'dummy' elements in between to eliminate false sharing. - In the inc() call, find the appropriate location using the thread/core id and then increment it atomically. https://github.com/ashkrit/blog/blob/master/src/main/java/counter/PaddedAtomicCounter.java However, being new to Go, I am facing a few obstacles: I thought of allocating a[numCores*16]uint32 and then use atomic.AddUint32() to atomically increment the individual element. However there are few basic things missing: - How do I get to know the Thread (or CpuCore) id on which the goroutine is running? - In the get() call of this counter, I need to sum up array values; how does one ensure that the latest values written to the array are visible to the goroutine calling the get() method? Are the values written using atomic package made visible to all cores immediately? Or else, does atomic.LoadInt32()guarantee to return the latest values? I think the problem is that atomic package does not say anything about visibility of memory updates across cores. -- cheers, gaurav -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
