Tree/flora gen optimization from Danger - somebody run with it?

[Cervator]

@TheyCallMeDanger is tweaking at code and ran into an optimization, but isn't big on GitHub, at least not yet. So he sent me this optimization suggestion - somebody world related want to try it out and toss it in a PR?

The original: (from FloraProvider.java)

public void process(GeneratingRegion region) {
    PlantFacet facet = new PlantFacet(region.getRegion(), region.getBorderForFacet(PlantFacet.class));
    SurfaceHeightFacet surface = region.getRegionFacet(SurfaceHeightFacet.class);
    DensityFacet density = region.getRegionFacet(DensityFacet.class);
    BiomeFacet biomeFacet = region.getRegionFacet(BiomeFacet.class);

    int minY = facet.getWorldRegion().minY();
    int maxY = facet.getWorldRegion().maxY();
    for (int z = facet.getRelativeRegion().minZ(); z <= facet.getRelativeRegion().maxZ(); ++z) {
        for (int x = facet.getRelativeRegion().minX(); x <= facet.getRelativeRegion().maxX(); ++x) {
            int height = TeraMath.floorToInt(surface.get(x, z));
            if (height >= minY && height < maxY) {
                CoreBiome biome = biomeFacet.get(x, z);
                height = height - minY + facet.getRelativeRegion().minY();

                if ((biome == CoreBiome.FOREST || biome == CoreBiome.PLAINS) && density.get(x, height, z) > 0  && density.get(x, height + 1, z) <= 0 && noiseTable.noise(x, z) < configuration.density) {
                    facet.set(x, height + 1, z, true);
                }
            }
        }
    }
    region.setRegionFacet(PlantFacet.class, facet);
}

The better way:
Pick a random number of trees/flora appropriate to the facet,
then for each tree/plant, pick a random x,z position in the facet,
then place the tree/plant.
The result is many many less iterations through the loop and all the associated calls.
x10 speed improvement, at least.
The original method iterates for each x for each z. For a 16x16 area, this is 256 iterations, vs probably 4 actual trees/plants.
I'm sure there are many other instances in the world generation that do similar things.

In general, instead of iterating through each x and z, iterate through the number of times you actually want to do something
(like place a tree, place lava, place a plant, etc) and generate a random x, z instead.
Even grass should do this, as it will cut CPU down by about 50%.
There may be some edge-effects where facets cross biomes, but they should be minimal and acceptable...

 

[Marcin Sciesinski]

Problem is, that with the suggested "improvement" it changes what the code wants to accomplish. There is no "number of trees/flora appropriate to the facet" for a chunk/region. Also, due to the fact that this facet may be called multiple times for different region sizes (including overlapping), choosing anything at random would not be deterministic which is a requirement for any generation. A facet has to generate the same data for a specific block, no matter what region (containing said block) it is asked to generate.