That can wait until after Ronald has time for Cfish-NonLazySMP.
Cfish doesn't support LazySMP?
So there is C++ code in LazySMP??? I would assume since pthreads are pure C that translation would be trivial.
You misunderstand. I want Cfish-YBW (or anything other than LazySMP).
My attempt of some time (years) ago to get that working unfortunately failed. At the time, I thought I could do better than SF's old YBW implementation by copying the approach I use in my private engine, but I had only very limited succes. The very low branching factor of Stockfish makes it difficult to find enough work for all threads. Perhaps even worse is that all threads need to synchronise after each iteration of the aspiration search.
Maybe I will give it a try again some day but no promises...
I think there is still a discrepancy in bench node counts at rather high depths.
Should someone find a quick way to produce a different node count, let me know. (Otherwise I will just have to be patient when I decide to hunt down the problem and fix it.)
If there is some obvious problem with threading or time control, let me know too. My quality ocntrol mainly consists of being careful and verifying default bench node counts
That can wait until after Ronald has time for Cfish-NonLazySMP.
Cfish doesn't support LazySMP?
So there is C++ code in LazySMP??? I would assume since pthreads are pure C that translation would be trivial.
You misunderstand. I want Cfish-YBW (or anything other than LazySMP).
My attempt of some time (years) ago to get that working unfortunately failed. At the time, I thought I could do better than SF's old YBW implementation by copying the approach I use in my private engine, but I had only very limited succes. The very low branching factor of Stockfish makes it difficult to find enough work for all threads. Perhaps even worse is that all threads need to synchronise after each iteration of the aspiration search.
Maybe I will give it a try again some day but no promises...
Understood. It's great to have Cfish back in any case.
I also saw an asmfish branch on his site.
Maybe we get two miracles.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
That can wait until after Ronald has time for Cfish-NonLazySMP.
Cfish doesn't support LazySMP?
So there is C++ code in LazySMP??? I would assume since pthreads are pure C that translation would be trivial.
You misunderstand. I want Cfish-YBW (or anything other than LazySMP).
My attempt of some time (years) ago to get that working unfortunately failed. At the time, I thought I could do better than SF's old YBW implementation by copying the approach I use in my private engine, but I had only very limited succes. The very low branching factor of Stockfish makes it difficult to find enough work for all threads. Perhaps even worse is that all threads need to synchronise after each iteration of the aspiration search.
Maybe I will give it a try again some day but no promises...
Understood. It's great to have Cfish back in any case.
Speed kills and cfish has the speed. Great to see it back!
syzygy wrote: ↑Tue Jul 07, 2020 11:55 pm
I think there is still a discrepancy in bench node counts at rather high depths.
E.g. bench 256 1 20 searches 57633211 where SF searches 57633212 nodes.
I also noticed that the reported seldepth is different (perhaps always off by one, not sure).
So at least two bugs still to be fixed.
Bench 256 1 20 now searches 57633212 nodes.
It seems I missed a small patch in May 2018.
Indeed. Nodes searched : 57633212
Good that you were able to quickly ascertain the cause of the discrepancy.
Unfortunately the node counts of longer benches are still off
I also noticed that, on my laptop, Cfish and SF are about equal in speed when I set the hash size to 1024 or higher. I haven't tried on my desktop yet, but this surprised me a bit.