EfficientHashing replaced hashing algorithm of Vec3i (and its subclasses like BlockPos) with one that's much much
more collision resistant, providing performance improvement in basically wherever its hashcode is used.
We can prove this by doing a simple test. Test is perform by hashing every combination of BlockPos between
new BlockPos(-100, -20, -100) and new BlockPos(100, 50, 100) (inclusive), with the total object count reaching
2,800,000+.
Vanilla algorithm produced 194,571 unique hashcode in total. This means that at least 93% of all BlockPos hashcode are colliding with each other.
And PhiMix, the algorithm used by EfficientHashing, produced 2868471 unique hashcode in total. This means that every unique BlockPos object gets its unique hashcode, and NONE of all BlockPos hashcode are colliding.
| Unique Hashcode | Collision | Collision Rate | |
|---|---|---|---|
| Vanilla | 194,571 | 2,673,900 | 93.2% |
| EfficientHashing | 2,868,471 | 0 | 0% |
If you're curious about the actual test code, see: here
Benchmark time:
Benchmark Mode Cnt Score Error Units
VecHashingBenchmark.mixin thrpt 5 47.012 ± 1.029 ops/s
VecHashingBenchmark.vanilla thrpt 5 49.326 ± 1.507 ops/s
PhiMix is about 95.3% the performance of original hashing algorithm. Yes, it's slower, but this performance gap is relatively tiny compared to other hashing solution, and can be easily outweighed by the significant improvement in hash collision resistance.
If you're curious about the actual benchmark code, see: here
In vanilla hashing algorithm, if you hash the "default" instance of BlockPos/Vec3i, aka new BlockPos(0, 0, 0), the
produced hashcode will be 0, which happens to be the "default" instance of integer. This special behavior is also
present in PhiMix, providing the best possible vanilla compatibility.
