possible race fix in Worker

[eaxeax]

The variable m_count is being read and written concurrently, in theory it is necessary to fix this

[SChernykh]

Not needed. Each worker runs in a separate thread, so there's only 1 thread writing to each m_count instance. Reading in parallel with writing will return either old or new value which is ok for hash counting (next value will be read on the next iteration).

[eaxeax]

This is UB, maybe then it's better to use memory_order_relaxed for store and load? I think that must be correct for cpuWorker, but its overhead for gpu mining. What do you think ?

[SChernykh]

There's a good rule in programming don't fix what's not broken. GPU worker also writes and reads m_count from different threads by the way. I explained already in my previous message what "UB" really means in this case, it works perfectly fine on both x86 and ARM. The only place that maybe needs fixing is CpuWorker<N>::hashrateData because it reads m_count twice and has 0.0000...0001% chance of reading different values. But even if it happens, it'll be fixed on the very next iteration, and it's only used to display hashrate. Atomic is not needed here. It's the main mining loop and it's better not to add unnecessary overhead.