Track which direction the current node's light came from, and don't
check it again when we check the current node's adjacent blocks.
e.g. if this node was the eastern neighbour of a light source, we don't
need to check this node's western neighbour, as we already know it has
a higher light level than our own.
This improves performance of basic light spread in a void by about 6%,
which isn't a huge amount, but it's something.
I've yet to explore whether light removal could also benefit from this
change.
we can use SubChunkExplorerStatus to decide whether or not to update the local light array reference.
We also dereference some properties into local variables, because dereferencing properties is slow. Indirect property dereferences add an extra performance penalty for every layer.
I previously avoided this due to being unsure of the effects; however, it's clear that we already use typed properties on Threaded things in other places anyway, and the only known issues are with uninit properties, and arrays.
this does come at a performance cost, but is necessary for metadata expansion.
we finally concede that this is not going to happen without BC breaks, however small they might be ...
instead, use a const array of the offsets and add them to the coordinates, which avoids the allocations.
In synthetic benchmarks, this method takes 40-50% less CPU time by eliding ZEND_INIT_ARRAY and ZEND_ADD_ARRAY opcodes. In practice, the benefit will likely be much smaller (perhaps even irrelevant).
in cases like PopulationTask it makes more sense to store the coordinates separately where they can be stored more efficiently (once instead of 9 times)
In addition, PopulationTask shouldn't need to serialize an empty chunk just to copy coordinates.
I've made changes like this in other areas already in preparation for the day when chunks no longer contain their coordinates, so this brings us one step closer to that goal.
this restores SimpleChunkManager's behaviour to PM3, removing the need for GeneratorChunkManager (although I'm dubious whether SubChunkExplorer makes any sense in there any more now that we have morton in the mix).
when setBlockSkyLightArray/setBlockLightArray was used, currentLightArray would retain a reference to the old light array, which would cause false readings if SubChunkExplorer didn't move away from that subchunk and back.
This causes a small degradation of performance, but I think it can be implemented differently anyway.
This also fixes#3816.
this produces a major performance improvement for large render distances, and reduces the impact of lighting calculation to zero on servers which have random blockupdates turned off.
Various bugs existed for a while with stuff using chunk managers instead of worlds when interacting with terrain due to a behavioural inconsistency between World::getChunk() (return from cache or load from disk), and SimpleChunkManager::getChunk() (return from cache only). This change brings the two in line.
World::getOrLoadChunk() has been added as a replacement, which has the same behaviour as the old getChunk() and also makes it more obvious that there is an issue with code using it during refactoring.
this solves multiple architectural issues:
- improves reusability by avoiding having old state info stick around to fuck stuff up
- prevents access to propagation state from outside of propagation
this also reduces the latent memory usage of light-updates after they have been used.
TODO: we could probably change LightPropagationContext to LightPropagator and move all the propagation-specific code into it if we can solve the subchunk-iterator and effective light problems.
