Reduce big time spikes by reducing PV re-searches above original depth.

[xoto10]

Save time by reducing PV re-searches above original depth. Instead use 5% extra time on every move.

STC 10+0.1 th 1 :
LLR: 2.93 (-2.94,2.94) {-0.25,1.25}
Total: 90688 W: 9702 L: 9436 D: 71550
Ptnml(0-2): 408, 7252, 29792, 7450, 442
https://tests.stockfishchess.org/tests/view/5f8df807bacb75a4f9a47223

LTC 60+0.6 th 1 :
LLR: 2.97 (-2.94,2.94) {0.25,1.25}
Total: 97856 W: 4602 L: 4303 D: 88951
Ptnml(0-2): 53, 3757, 41057, 3960, 101
https://tests.stockfishchess.org/tests/view/5f8ec4872c92c7fe3a8c602d

Bench 3943959

[vondele]

question on the implementation. Can't you use a local variable maxNextDepth computed as depth + 1 on function entry, instead of adding this as an argument to search?

[xoto10]

I'm not sure. My idea was that the top level call uses (... adjustedDepth, adjustedDepth ...), and the PV call uses min(maxNextDepth, newDepth) so preventing the PV re-search from being deeper than the original iteration depth. This suggests puttng adjustedDepth into the thread structure to me instead of using the extra fn parameter, and using min(thisThread->adjustedDepth, newDepth) in the PV call, but I think I tried that and it didn't work. Perhaps I did it wrong / differently somehow?

[vondele]

I see the difference between the call at the root and the others now. I need a closer look first.. (something like rootNode ? ... : ... + 1), which will have to wait a bit.

[xoto10]

Perhaps I had the right idea with my last test on the idpv2 branch, but 10% extra time was too much? Might be worth trying that with only 5% extra time.

[vondele]

@xoto10 vondele@cc7e03d is the variant I had in mind. Gives the same bench. What do you think?

[xoto10]

@vondele
Yeah, that makes sense. When I thought about it earlier on, I didn't think it would work for some reason, but now I look at it again it looks good. You could just test rootNode in the min() and then it's just a 1 line change (plus the time change). Nice :)

You've probably seen my test with adjustedDepth in the thread structure. Seems pointless now. If you start a test for yours I'll stop mine ...

[vondele]

since my variant is really equivalent to this PR (same bench), it needs no test IMO.

[xoto10]

fair enough!

[xoto10]

How do you want to progress, update this pr or will you create a new one?

[vondele]

I'll merge & close this PR using my cleanup. No need for a new PR, just leave this open.