Yes, I am currently starting half the threads with the current iteration depth and the other half with iteration depth + 1. As I said in an earlier post, there may be a more optimum configuration than this, but this is what I am doing now (perhaps starting all extra threads with different depths? +1, +2, +3, etc... but that may be too much difference between them for the shared hash to do its magic)jd1 wrote:Thanks Dan.
Just to make sure I understand your method correctly, at every iteration you start the even threads at depth+1, and stop them when the iteration finishes.
This is then quite different to the usual shared hash table smp implementation, which launches x threads at the beginning of the search, as http://chessprogramming.wikispaces.com/ ... sh%20Table
I am testing right now.
Jerry
From reading the wikispace paragraph (thanks for the link!), my implementation is not that different... the only detailed difference is that I am restarting each thread with each new iteration and some of my threads are searching a somewhat deeper iteration depth (+1).
- Dan