Rework between_bb() into ray_bb() + prefab results #3397

BM123499 · 2021-03-15T19:22:24Z

In all times (except in has_game_cycle()) between_bb() uses or is indifferent if the one of the given square is added. And has shown in here that using prefab results can speed up. So combining both, I reworked between_bb() into ray_bb(), that uses prefab results and includes the last parameter square.

Build Tester: 1.4.7.0
Windows 10 (Version 10.0, Build 0, 64-bit Edition)
Intel(R) Core(TM) i7-8750H CPU @ 2.20GHz
SafeMode: No 
Running In VM: No 
HyperThreading Enabled: Yes
CPU Warmup: Yes
Command Line: bench 16 1 5 default perft
Tests per Build: 10
ANOVA: n/a

                Engine# (NPS)                               Speedup     Sp     Conf. 99.5%  S.S.
patch  (102.715.251,9 ) ---> master  (98.242.950,2  ) --->  4,552%  2.105.495,3    Yes       No

Passed STC:
LLR: 2.96 (-2.94,2.94) {-0.25,1.25}
Total: 18736 W: 1746 L: 1607 D: 15383
Ptnml(0-2): 61, 1226, 6644, 1387, 50
https://tests.stockfishchess.org/tests/view/60428c84ddcba5f0627bb6e4

Yellow LTC:
LTC:
LLR: -3.00 (-2.94,2.94) {0.25,1.25}
Total: 39144 W: 1431 L: 1413 D: 36300
Ptnml(0-2): 13, 1176, 17184, 1178, 21
https://tests.stockfishchess.org/tests/view/605128702433018de7a38ca1

No functional change

fix description

snicolet · 2021-03-16T21:49:02Z

Might be a nice idea.

It would be interesting to have a LTC test also.

BM123499 · 2021-03-16T21:51:59Z

@snicolet https://tests.stockfishchess.org/tests/view/605128702433018de7a38ca1 here. Standard bounds, right?

snicolet · 2021-03-17T08:52:13Z

The description is still wrong : if the two squares s1 and s2 are not on a same file/rank/diagonal, the comment claims that the function returns 0 but in fact it returns s2.

At this point I have spent one hour reformatting and debugging what should be a trivial patch, so I will continue tomorrow :-)

Current state of pull request: snicolet/Stockfish@d01749b

We remark that in current master, most of our use cases for between_bb() can be optimized if the second parameter of the function is added to the segment. So we change the definition of between_bb(s1, s2) such that it excludes s1 but includes s2. We also use a precomputed array for between_bb() for another small speed gain (see https://tests.stockfishchess.org/tests/view/604d09f72433018de7a389fb). Passed STC: LLR: 2.96 (-2.94,2.94) {-0.25,1.25} Total: 18736 W: 1746 L: 1607 D: 15383 Ptnml(0-2): 61, 1226, 6644, 1387, 50 https://tests.stockfishchess.org/tests/view/60428c84ddcba5f0627bb6e4 Yellow LTC: LTC: LLR: -3.00 (-2.94,2.94) {0.25,1.25} Total: 39144 W: 1431 L: 1413 D: 36300 Ptnml(0-2): 13, 1176, 17184, 1178, 21 https://tests.stockfishchess.org/tests/view/605128702433018de7a38ca1 Closes official-stockfish/Stockfish#3397 No functional change

BM123499 · 2021-03-17T12:44:28Z

@snicolet Yes, I forgot about that. My goal is ray_bb(s1, s2) = between(s1, s2) | s2. So s2 will always be included. This is because of horse attack.
Your XX is different because it didn't considered it. So when between(s1, s2) is empty, XX is also empty, but it's different of RayBB.
I believe that is the problem you're facing. Sorry if I was too imprecise.

That why I set RayBB as between_bb() was calculated and then add s2. So s2 will always be in RayBB.

BM123499 · 2021-03-17T12:47:39Z

You can do something like.

if (PseudoAttacks[pt][s1] & s2){
    LineBB[s1][s2] = (attacks_bb(pt, s1, 0) & attacks_bb(pt, s2, 0)) | s1 | s2;
    XX[s1][s2] = (attacks_bb(pt, s1, square_bb(s2)) & attacks_bb(pt, s2, square_bb(s1)));
}
XX[s1][s2] |= s2;

Remember that it's due to horse attack, you can do another workaround... :)

snicolet · 2021-03-17T21:19:22Z

Remember that it's due to horse attack, you can do another workaround... :)

I don't understand what a horse attack is, you mean a knight move?

And why exactly is it relevant to the problem that between(s1, s2) now contains s2, contrary to what the comment says when s1 and s2 are not aligned?

I think that if you could provide a complete example of how your patch work on a concrete position, that would be illuminating. Thanks :-)

BM123499 · 2021-03-17T21:30:13Z

src/movegen.cpp

            break;
        case EVASIONS:
-            target = between_bb(pos.square<KING>(Us), lsb(pos.checkers())) | pos.checkers();
+            target = ray_bb(pos.square<KING>(Us), lsb(pos.checkers()));


@snicolet here, if the checker is a knight, we have to mark it, that's the foundation on adding s2 on RayBB. it's not a line anymore, it has also trigger when the knight is attacking.

it should also be a "ray" for a knight. That's the only place where it needs to include s2 "no matter what", all other use is indifferent on s2. except on has_game_cycle, so there I did a workaround.

We remark that in current master, most of our use cases for between_bb() can be optimized if the second parameter of the function is added to the segment. So we change the definition of between_bb(s1, s2) such that it excludes s1 but includes s2. We also use a precomputed array for between_bb() for another small speed gain (see https://tests.stockfishchess.org/tests/view/604d09f72433018de7a389fb). Passed STC: LLR: 2.96 (-2.94,2.94) {-0.25,1.25} Total: 18736 W: 1746 L: 1607 D: 15383 Ptnml(0-2): 61, 1226, 6644, 1387, 50 https://tests.stockfishchess.org/tests/view/60428c84ddcba5f0627bb6e4 Yellow LTC: LTC: LLR: -3.00 (-2.94,2.94) {0.25,1.25} Total: 39144 W: 1431 L: 1413 D: 36300 Ptnml(0-2): 13, 1176, 17184, 1178, 21 https://tests.stockfishchess.org/tests/view/605128702433018de7a38ca1 Closes official-stockfish/Stockfish#3397 No functional change

snicolet · 2021-03-17T23:29:01Z

Merged via 5089061, congrats :-)

We remark that in current master, most of our use cases for between_bb() can be optimized if the second parameter of the function is added to the segment. So we change the definition of between_bb(s1, s2) such that it excludes s1 but includes s2. We also use a precomputed array for between_bb() for another small speed gain (see https://tests.stockfishchess.org/tests/view/604d09f72433018de7a389fb). Passed STC: LLR: 2.96 (-2.94,2.94) {-0.25,1.25} Total: 18736 W: 1746 L: 1607 D: 15383 Ptnml(0-2): 61, 1226, 6644, 1387, 50 https://tests.stockfishchess.org/tests/view/60428c84ddcba5f0627bb6e4 Yellow LTC: LTC: LLR: -3.00 (-2.94,2.94) {0.25,1.25} Total: 39144 W: 1431 L: 1413 D: 36300 Ptnml(0-2): 13, 1176, 17184, 1178, 21 https://tests.stockfishchess.org/tests/view/605128702433018de7a38ca1 Closes official-stockfish#3397 --------- Verified for correctness by running perft on the following position: ./stockfish position fen 4rrk1/1p1nq3/p7/2p1P1pp/3P2bp/3Q1Bn1/PPPB4/1K2R1NR w - - 40 21 go perft 6 Nodes searched: 6136386434 -------- No functional change

We remark that in current master, most of our use cases for between_bb() can be optimized if the second parameter of the function is added to the segment. So we change the definition of between_bb(s1, s2) such that it excludes s1 but includes s2. We also use a precomputed array for between_bb() for another small speed gain (see https://tests.stockfishchess.org/tests/view/604d09f72433018de7a389fb). Passed STC: LLR: 2.96 (-2.94,2.94) {-0.25,1.25} Total: 18736 W: 1746 L: 1607 D: 15383 Ptnml(0-2): 61, 1226, 6644, 1387, 50 https://tests.stockfishchess.org/tests/view/60428c84ddcba5f0627bb6e4 Yellow LTC: LTC: LLR: -3.00 (-2.94,2.94) {0.25,1.25} Total: 39144 W: 1431 L: 1413 D: 36300 Ptnml(0-2): 13, 1176, 17184, 1178, 21 https://tests.stockfishchess.org/tests/view/605128702433018de7a38ca1 Closes official-stockfish/Stockfish#3397 --------- Verified for correctness by running perft on the following position: ./stockfish position fen 4rrk1/1p1nq3/p7/2p1P1pp/3P2bp/3Q1Bn1/PPPB4/1K2R1NR w - - 40 21 go perft 6 Nodes searched: 6136386434 -------- No functional change

BM123499 added 2 commits March 15, 2021 16:46

ray_bb()

10efdd0

fix description

fix description

8505b2b

BM123499 force-pushed the RayBB branch from d3b51f6 to 8505b2b Compare March 15, 2021 19:47

BM123499 commented Mar 17, 2021

View reviewed changes

BM123499 added 2 commits March 17, 2021 18:36

new comment

48ffe95

fix

9deba5b

snicolet closed this in 5089061 Mar 17, 2021

snicolet added the to be merged Will be merged shortly label Mar 19, 2021

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Rework between_bb() into ray_bb() + prefab results #3397

Rework between_bb() into ray_bb() + prefab results #3397

Uh oh!

BM123499 commented Mar 15, 2021 •

edited

Loading

Uh oh!

snicolet commented Mar 16, 2021

Uh oh!

BM123499 commented Mar 16, 2021

Uh oh!

snicolet commented Mar 17, 2021 •

edited

Loading

Uh oh!

BM123499 commented Mar 17, 2021 •

edited

Loading

Uh oh!

BM123499 commented Mar 17, 2021

Uh oh!

snicolet commented Mar 17, 2021

Uh oh!

BM123499 Mar 17, 2021

Uh oh!

BM123499 Mar 17, 2021

Uh oh!

snicolet commented Mar 17, 2021

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Rework between_bb() into ray_bb() + prefab results #3397

Rework between_bb() into ray_bb() + prefab results #3397

Uh oh!

Conversation

BM123499 commented Mar 15, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

snicolet commented Mar 16, 2021

Uh oh!

BM123499 commented Mar 16, 2021

Uh oh!

snicolet commented Mar 17, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

BM123499 commented Mar 17, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

BM123499 commented Mar 17, 2021

Uh oh!

snicolet commented Mar 17, 2021

Uh oh!

BM123499 Mar 17, 2021

Choose a reason for hiding this comment

Uh oh!

BM123499 Mar 17, 2021

Choose a reason for hiding this comment

Uh oh!

snicolet commented Mar 17, 2021

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

BM123499 commented Mar 15, 2021 •

edited

Loading

snicolet commented Mar 17, 2021 •

edited

Loading

BM123499 commented Mar 17, 2021 •

edited

Loading