Use less time on recaptures #5189

xoto10 · 2024-04-24T03:06:59Z

Credit for the idea goes to peregrine in Discord.

Passed STC 10+0.1:
LLR: 2.95 (-2.94,2.94) <0.00,2.00>
Total: 75712 W: 19793 L: 19423 D: 36496
Ptnml(0-2): 258, 8487, 20023, 8803, 285
https://tests.stockfishchess.org/tests/view/662652623fe04ce4cefc48cf

Passed LTC 60+0.6:
LLR: 2.94 (-2.94,2.94) <0.50,2.50>
Total: 49788 W: 12743 L: 12404 D: 24641
Ptnml(0-2): 29, 5141, 14215, 5480, 29
https://tests.stockfishchess.org/tests/view/6627495e3fe04ce4cefc59b6

The code was updated slightly and tested for non-regression against the original code at STC:
LLR: 2.94 (-2.94,2.94) <-1.75,0.25>
Total: 41952 W: 10912 L: 10698 D: 20342
Ptnml(0-2): 133, 4825, 10835, 5061, 122
https://tests.stockfishchess.org/tests/view/662d84f56115ff6764c7e438

Bench 1479416

peregrineshahin · 2024-04-24T04:23:48Z

running this https://tests.stockfishchess.org/tests/view/6628859d3fe04ce4cefc6f37 with only the base change to verify it's not the responsible for the gain
Edit: paused STC
here is LTC https://tests.stockfishchess.org/tests/view/66288f5a3fe04ce4cefc6fd1

xoto10 · 2024-04-24T04:41:59Z

Yeah, less time seems to be helping stc, no doubt about that from the tests I've been doing. Whether it helps ltc is the real question ...
This was why I tried the timetime4 tests - trying to get stc and ltc about right

src/engine.h

Disservin

uci is the wrong place to implement it, since it's supposed to be a small wrapper around the engine and this leaks internals to uci.

xoto10 · 2024-04-25T09:30:39Z

this leaks internals to uci.

If we get a fen with moves via UCI, and the last move is a capture, I want to store the capture square and then pass it into search - UCI seemed the right place to do this to me and I don't see how else I could do it??
The moves are parsed and applied in Engine::set_position(), so I need to derive the square in there, and pass it back out.

Happy to change it if there's a better way ofc.

Disservin · 2024-04-25T09:48:22Z

The point is that we are slowly making it possible to use stockfish as a library, in which case the dev (the library user) should only include the engine code and use the engine functions to get the desired behaviour, so implementing uci would be left up to him. This patch would require him to implement the logic of the capSq manually and then set it in the limits struct, which is also perhaps the wrong data structure for this as the comment states LimitsType struct stores information sent by GUI.

Another solution would be to save the previous capture square in the StateInfo, but this perhaps a bit overkill, you could also add a new data field to the SharedState and pass the data like that to search

dubslow · 2024-04-27T21:46:49Z

the simple LTC has proven to be considerably less good (but it was good to have tested it ofc), so i vote we stop that second LTC and approve this functionality to be merged (once the abstraction parts are figured out ofc)

Disservin · 2024-04-27T22:11:53Z

@xoto10 do you need some other input from me? or do you want me to take a look how we can implement this differently?

xoto10 · 2024-04-27T23:09:32Z

@Disservin
I've coded recap2b, see requested test https://tests.stockfishchess.org/tests/view/662d84f56115ff6764c7e438
If you want me to revise that further I can of course.

The huge influx of cores to fishtest caught me by surprise (very welcome of course!).

xoto10 · 2024-04-27T23:16:54Z

I like the limits struct, I was wondering if we could rename it, maybe Analysis,Request, AnalysisRequest, something like that and it could hold the position requested, my capSq and any other inputs from the controlling program, as well as the go command params.
Maybe make it part of Engine so that it exists as soon as the Engine is defined.

xoto10 · 2024-04-27T23:51:31Z

The recap2b non-regression passed (much quicker than I expected) so it looks like it's a valid rework of the recap2 code.
https://tests.stockfishchess.org/tests/view/662d84f56115ff6764c7e438

With hindsight I don't think I should have adjusted the EvalLevel constants so that the recapture multiplier was above/below 1.0 but it's done now so I'll leave it.

Let me know if I need to modify the recap2b code further ...

Disservin · 2024-04-28T14:55:46Z

src/search.cpp

@@ -446,9 +446,11 @@ void Search::Worker::iterative_deepening() {
            double reduction = (1.48 + mainThread->previousTimeReduction) / (2.17 * timeReduction);
            double bestMoveInstability = 1 + 1.88 * totBestMoveChanges / threads.size();
            int    el                  = std::clamp((bestValue + 750) / 150, 0, 9);
+            double recapture =
+              limits.capSq != SQ_NONE && limits.capSq == rootMoves[0].pv[0].to_sq() ? 0.955 : 1.005;


dont think we need the limits.capSq != SQ_NONE check, right? It is already implied by limits.capSq == rootMoves[0].pv[0].to_sq() and here we should always have a valid pv move.

Yes, good spot. Testing for the second element (ponder) makes sense, but I guess the first one has to be there!

…dea. Bench 1836777

Disservin

i think this is fine for now, modifying the struct Search::LimitsType from the go function is perhaps a bit weird but that part will likely be changed in the future

commit 0fe6428 Author: Stefan Geschwentner <stgeschwentner@gmail.com> Date: Sun Apr 28 16:53:47 2024 +0200 More reduction at cut nodes which are not a former PV node But the tt move and first killer are excluded. This idea is based on following LMR condition tuning https://tests.stockfishchess.org/tests/view/66228bed3fe04ce4cefc0c71 by using only the two largest terms P[0] and P[1]. Passed STC: LLR: 2.93 (-2.94,2.94) <0.00,2.00> Total: 173248 W: 45091 L: 44565 D: 83592 Ptnml(0-2): 693, 20534, 43673, 21002, 722 https://tests.stockfishchess.org/tests/view/6629603b3fe04ce4cefc7d37 Passed LTC: LLR: 2.94 (-2.94,2.94) <0.50,2.50> Total: 722394 W: 183231 L: 181487 D: 357676 Ptnml(0-2): 462, 80650, 197252, 82348, 485 https://tests.stockfishchess.org/tests/view/662cbe45d46f72253dcff7bf closes official-stockfish#5199 Bench: 1619613 commit 48a3b7c Author: Stefan Geschwentner <stgeschwentner@gmail.com> Date: Sun Apr 28 16:04:28 2024 +0200 Simplify non-pawn material divisor to a constant Passed STC: https://tests.stockfishchess.org/tests/view/662942603fe04ce4cefc7aba LLR: 2.93 (-2.94,2.94) <-1.75,0.25> Total: 272832 W: 70456 L: 70497 D: 131879 Ptnml(0-2): 1020, 32619, 69154, 32628, 995 Passed LTC: https://tests.stockfishchess.org/tests/view/662dfe3b6115ff6764c829eb LLR: 2.94 (-2.94,2.94) <-1.75,0.25> Total: 100254 W: 25446 L: 25303 D: 49505 Ptnml(0-2): 121, 11292, 27166, 11419, 129 closes official-stockfish#5198 Bench: 1544645 commit 834e8ff Author: cj5716 <125858804+cj5716@users.noreply.github.com> Date: Sun Apr 28 08:53:28 2024 +0800 Penalise the TT move in multicut Passed STC: LLR: 2.99 (-2.94,2.94) <0.00,2.00> Total: 185504 W: 48079 L: 47533 D: 89892 Ptnml(0-2): 716, 21866, 46988, 22520, 662 https://tests.stockfishchess.org/tests/view/662d9e1d6115ff6764c7f83d Passed LTC: LLR: 2.94 (-2.94,2.94) <0.50,2.50> Total: 75612 W: 19351 L: 18948 D: 37313 Ptnml(0-2): 46, 8363, 20592, 8752, 53 https://tests.stockfishchess.org/tests/view/662dc9dc6115ff6764c80fea closes official-stockfish#5195 Bench: 1415435 commit a129c06 Author: mstembera <m_stembera@yahoo.com> Date: Sun Apr 28 10:28:25 2024 -0700 Combine remove and add in update_accumulator_refresh_cache() Combine remove and add in update_accumulator_refresh_cache(). Move remove before add to match other parts of the code. STC: https://tests.stockfishchess.org/tests/view/662d96dc6115ff6764c7f4ca LLR: 2.95 (-2.94,2.94) <0.00,2.00> Total: 364032 W: 94421 L: 93624 D: 175987 Ptnml(0-2): 1261, 41983, 94811, 42620, 1341 closes official-stockfish#5194 Bench: 1836777 commit 940a3a7 Author: mstembera <m_stembera@yahoo.com> Date: Thu Apr 25 18:20:08 2024 -0700 Cache small net w/ psqtOnly support Caching the small net in the same way as the big net allows them to share the same code path and completely removes update_accumulator_refresh(). STC: https://tests.stockfishchess.org/tests/view/662bfb5ed46f72253dcfed85 LLR: 2.94 (-2.94,2.94) <-1.75,0.25> Total: 151712 W: 39252 L: 39158 D: 73302 Ptnml(0-2): 565, 17474, 39683, 17570, 564 closes official-stockfish#5194 Bench: 1836777 commit bc45cbc Author: Joost VandeVondele <Joost.VandeVondele@gmail.com> Date: Sat Apr 27 18:09:45 2024 +0200 Output some basic info about the used networks Adds size in memory as well as layer sizes as in info string NNUE evaluation using nn-ae6a388e4a1a.nnue (132MiB, (22528, 3072, 15, 32, 1)) info string NNUE evaluation using nn-baff1ede1f90.nnue (6MiB, (22528, 128, 15, 32, 1)) For example, the size in MiB is useful to keep the fishtest memory sizes up-to-date, the L1-L3 sizes give a useful hint about the architecture used. closes official-stockfish#5193 No functional change commit 3502c8a Author: Disservin <disservin.social@gmail.com> Date: Thu Apr 25 19:20:57 2024 +0200 Fix missing initialization of AccumulatorCaches in Eval::trace Add a constructor to `AccumulatorCaches` instead of just calling `clear(networks)` to prevent similar issues from appearing in the future. fixes official-stockfish#5190 closes official-stockfish#5191 No functional change commit 886ed90 Author: xoto10 <23479932+xoto10@users.noreply.github.com> Date: Sun Apr 28 16:27:40 2024 +0100 Use less time on recaptures Credit for the idea goes to peregrine on discord. Passed STC 10+0.1: https://tests.stockfishchess.org/tests/view/662652623fe04ce4cefc48cf LLR: 2.95 (-2.94,2.94) <0.00,2.00> Total: 75712 W: 19793 L: 19423 D: 36496 Ptnml(0-2): 258, 8487, 20023, 8803, 285 Passed LTC 60+0.6: https://tests.stockfishchess.org/tests/view/6627495e3fe04ce4cefc59b6 LLR: 2.94 (-2.94,2.94) <0.50,2.50> Total: 49788 W: 12743 L: 12404 D: 24641 Ptnml(0-2): 29, 5141, 14215, 5480, 29 The code was updated slightly and tested for non-regression against the original code at STC: LLR: 2.94 (-2.94,2.94) <-1.75,0.25> Total: 41952 W: 10912 L: 10698 D: 20342 Ptnml(0-2): 133, 4825, 10835, 5061, 122 https://tests.stockfishchess.org/tests/view/662d84f56115ff6764c7e438 closes official-stockfish#5189 Bench: 1836777

xoto10 force-pushed the recap2 branch from 2e4e5f1 to 2a8743e Compare April 24, 2024 03:21

XInTheDark reviewed Apr 25, 2024

View reviewed changes

src/engine.h Outdated Show resolved Hide resolved

Disservin requested changes Apr 25, 2024

View reviewed changes

xoto10 force-pushed the recap2 branch from 2a8743e to b77fc33 Compare April 25, 2024 10:17

xoto10 force-pushed the recap2 branch from b77fc33 to bdf7ba1 Compare April 27, 2024 23:04

xoto10 force-pushed the recap2 branch 3 times, most recently from 3e3951a to 391985a Compare April 28, 2024 02:11

Disservin added 🚀 gainer Gains elo functional-change no-bench-change labels Apr 28, 2024

Disservin reviewed Apr 28, 2024

View reviewed changes

xoto10 force-pushed the recap2 branch from 391985a to 3d1a768 Compare April 28, 2024 15:25

Use less time on recaptures, credit to peregrine in Discord for the i…

d773fa7

…dea. Bench 1836777

xoto10 force-pushed the recap2 branch from 3d1a768 to d773fa7 Compare April 28, 2024 15:28

Disservin approved these changes Apr 28, 2024

View reviewed changes

fix iwyu

0e64c14

Disservin added the to be merged Will be merged shortly label Apr 28, 2024

Disservin closed this in 886ed90 Apr 28, 2024

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Use less time on recaptures #5189

Use less time on recaptures #5189

xoto10 commented Apr 24, 2024 •

edited

peregrineshahin commented Apr 24, 2024 •

edited

xoto10 commented Apr 24, 2024 •

edited

Disservin left a comment

xoto10 commented Apr 25, 2024

Disservin commented Apr 25, 2024

dubslow commented Apr 27, 2024

Disservin commented Apr 27, 2024

xoto10 commented Apr 27, 2024

xoto10 commented Apr 27, 2024

xoto10 commented Apr 27, 2024

Disservin Apr 28, 2024 •

edited

xoto10 Apr 28, 2024

Disservin left a comment

Use less time on recaptures #5189

Use less time on recaptures #5189

Conversation

xoto10 commented Apr 24, 2024 • edited

peregrineshahin commented Apr 24, 2024 • edited

xoto10 commented Apr 24, 2024 • edited

Disservin left a comment

Choose a reason for hiding this comment

xoto10 commented Apr 25, 2024

Disservin commented Apr 25, 2024

dubslow commented Apr 27, 2024

Disservin commented Apr 27, 2024

xoto10 commented Apr 27, 2024

xoto10 commented Apr 27, 2024

xoto10 commented Apr 27, 2024

Disservin Apr 28, 2024 • edited

Choose a reason for hiding this comment

xoto10 Apr 28, 2024

Choose a reason for hiding this comment

Disservin left a comment

Choose a reason for hiding this comment

xoto10 commented Apr 24, 2024 •

edited

peregrineshahin commented Apr 24, 2024 •

edited

xoto10 commented Apr 24, 2024 •

edited

Disservin Apr 28, 2024 •

edited