ITP – addendum

As promised, I performed some tests of my thread pool on quad-core machine. Difference between lock-based and lock-free version is bigger (than on dual core), but definitelly it’s not as significant as one could guess looking at results in Intel Thread Profiler (higher concurrency, smaller contention). The most important speed-up can be gained by letting main thread help workers when it’s waiting anyway (50 fps ‘> 61 fps), but this one is independent from other modifications and can be applied to lock-based version as well.

Anyway, this has been interesting, but I guess I can move to other items on my ‘to experiment with’ list now’