Two hash functions for distributed transposition table
Posted: Wed Dec 17, 2014 12:43 am
I wanted to test distributed transposition table in a message passing environement. The idea is to share only the upper parts of the tree, say > depth=10. I didn't want to have separate tables for parts that are shared and not. We do a modulo to get the processor that holds the TT entry and then modify the hashkey accordingly. So basically a different hash function is used for the shared part of the table.
For depth >= 10
For depth < 10, it is simply
Do you see potential problems with using same tables ? In the depth preferred scheme, the shared part of the TT should survive longer.
For depth >= 10
Code: Select all
processor_number = hashkey % n_processors;
hashkey = hashkey / n_processors
Ger(processor_number, hashkey)
Code: Select all
Get(hashkey)