This should resolve the issue with k=18 as now k=18 is as fast as k=31 on my computer (the difference before was at my PC 1.3 times instead of 2x but anyway it should be resolved) -- the issue was probably due to hash collisions.
2 -- Using 64 bits in khash
This enables ProphAsm2 to scale for very large datasets.
Both combined have a slowdown about 5% for values where there were no hash collisions.
1 -- Using Wang-hash instead of the default one.
This should resolve the issue with k=18 as now k=18 is as fast as k=31 on my computer (the difference before was at my PC 1.3 times instead of 2x but anyway it should be resolved) -- the issue was probably due to hash collisions.
2 -- Using 64 bits in khash
This enables ProphAsm2 to scale for very large datasets.
Both combined have a slowdown about 5% for values where there were no hash collisions.