Performance comparison with Artemis about ashley HOT 8 CLOSED

I adapted the existing artemis-odb benchmarks to work against ashley and gdx-artemis a couple of days ago: https://github.com/junkdog/entity-system-benchmarks - it still requires a little more work; benchmarking entreri, pooling in ashley, some more specialized benchmarks etc, but it's a start.

from ashley.

@junkdog looks promising! I am impressed with ashley

from ashley.

Oh that's very useful @junkdog!

Correct me if I'm wrong, the higher the score the faster the system is? The fact that is not measured in time hints that such is the case. However, I don't want to get too excited with Ashley just yet!

from ashley.

No, sorry, it measures average execution in microseconds; so per world/engine update cycle. Raw logs are in the results folder.

I'm guessing it's due to the heavy usage of IntMaps (unless I made a major screw-up while writing the benchmarks). Gdx-artemis uses Array<Array<T>> and it performed similarly to artemis-odb: it might be that the array-of-arrays approach is a necessary pattern in regards to performance, unless IntMap can be optimized somehow. I haven't profiled it though, so it's possible the bottleneck lies elsewhere.

from ashley.

I'm going to profile a game made with Ashley to see what's taking so long. It could be all the IntMaps. Thanks for the clarification and the hint!

from ashley.

Not sure about that @siondream, I did a quick profile of ashley jumper and the bottle neck seemed to be com.badlogic.gdx.utils.Array.sort( Comparator ) but I did not see anything wrong with IntMaps, again it was a quick profile

from ashley.

I did the same check and, indeed, the most prominent thing was the sort. Maybe if we got rid of that, something else would show up.

I'm going to try play around with Caliper and see what happens if I change some stuff.

from ashley.

Added in d6fec98

from ashley.