TalkChess.com

Ronald's site on github appears to be gone now. Huh? Search me, the search is broken. Try this: https://github.com/search?l=C&q=syzygy&type=Repositories You're searching for a repository called syzygy. Ron's account is "syzygy1", and the repository name is "tb". The users tab brings up syzygy1 as t...

hgm wrote: ↑

Sun May 27, 2018 9:39 pm

Dorpsgek lost connection after 7 rounds, and had to be forfeited.

XBoard segfaulted while I was asleep. Sorry for the inconvenience.

Fail soft scores are still fairly unreliable in my experience; just because a score can be outside the window doesn't always mean it is. Hyatt suggested root move ordering by node count, and I think it's worth trying out, but I would have to test. Printing root move ordering coincidentally helped me...

32-bit. 64-bit RPi - last time I checked - lacked things like processor reclocking which is found in the RPF kernel, and the RPF aren't interested in a 64-bit version.

This got sidetracked by the forum migration and me getting ill over the weekend. So, I ran a Perft benchmark (which admittedly isn't the whole story of a search, but it magnifies the important part). It turns out that a monochrome move generator is slightly faster: AMD FX-6300 @ 3.5GHz: Double-colou...

I don't have data on quad-bitboard versus octa-bitboard, but quad-bitboard engines require a monochrome board representation, and my own testing does show that those are slightly slower than double colour board representations.

I can provide more accurate data in a few hours.

This line looks ripe for an OpenMP pragma, and the code looks to be thread-safe. You might find that helps. I think I spoke too soon; `refs` and `magic` are shared, and some bookkeeping that seems irrelevant is done. Still, your method of GNU Parallel works fine. EDIT: you could add `-flto -fwhole-...

This line looks ripe for an OpenMP pragma, and the code looks to be thread-safe. You might find that helps.

But the expected improvement is outside the noise, which is what I'm worried about.

This is strange. The patches all tested as improvements, but the net result is a slight regression, even with the bugfix I found.