New J.A Compiles

[Whiskers]

Hi Jim

If it is not too much, can you also try and compile Patricia?

Some changes to the source have been made 4 days ago.

Thanks a lot for your effort.

There have only been minor changes since the last compile JA did, I'd recommend using that or simply waiting till Patricia 5.

[Jim Ablett]

https://github.com/jhonnold/berserk
Berserk 20250606 by Jay Honnold

Berserk 20250606 64 JA

Windows/Linux & Android profiled builds

Proton link:
https://drive.proton.me/urls/YJBAQYFZ6W#arXhqURJVUd4

Smash link:
https://fromsmash.com/Berserk20250606JA

Jim.

[jhonnold]

Just as a heads up Jim, I believe the profiled builds are worse when it comes to Berserk. I don't know why.

[Jim Ablett]

Just as a heads up Jim, I believe the profiled builds are worse when it comes to Berserk. I don't know why.

Hi Jay,

I did a lot of testing when compiling. I got a speed increase with pgo after experimenting to find the best compiler switch
combinations.

Jim.

[jhonnold]

Just as a heads up Jim, I believe the profiled builds are worse when it comes to Berserk. I don't know why.

Hi Jay,

I did a lot of testing when compiling. I got a speed increase with pgo after experimenting to find the best compiler switch
combinations.

Jim.

Please verify with care, because while I have seen a speedup as well, testing indicates a significant decline for some other reason, see this test - http://chess.grantnet.us/test/35196/

[Jim Ablett]

https://github.com/Ilya-Ruk/RukChess
RukChess 4.1.0 by Ilya Rukavishnikov

RukChess 4.1.0 64 JA

Ported src to GCC/Clang/Linux
Added embedded net code

Windows & Linux profiled builds

Proton link:
https://drive.proton.me/urls/83WX2KP07W#jETsFT2rg5OB

Smash link:
https://fromsmash.com/RukChess410JA

Jim.

[Werner]

Hi Jim,
I compared your version with the version from the author:
nps your version around 24000k - authors version around 600k.
+++ these are displayed inside arena GUI - different to command prompt +++

do you have any idea whats the reason?

[Jim Ablett]

Hi Jim,
I compared your version with the version from the author:
nps your version around 24000k - authors version around 600k.
do you have any idea whats the reason?

Hi Werner,

I would guess Author's build is compiled with MSVC without many optimizations. I ported the code
so I could compile with Clang+pgo.

Jim.

[Werner]

Inside the command prompt the search dephts are similar:
your bmi compile
info depth 20 seldepth 29 nodes 581481995 time 29836 score cp 28 nps 19489274 pv d2d4 d7d5 c2c4 c7c6 g1f3 g8f6 e2e3 e7e6 b2b3 b7b6 c1b2 c8b7 b1d2 b8d7 f1e2 f8d6 d1b1 e8g8 e1g1 c6c5
info nodes 726906004 nps 23.663.845 hashfull 982

and authors file:
info depth 21 seldepth 24 nodes 63246729 time 26558 score cp 24 nps 2381456 hashfull 384 pv e2e4 e7e5 g1f3 b8c6 f1c4 g8f6 d2d3 f8c5 e1g1 a7a6 c2c3 c5a7 f1e1 e8g8 b1d2 h7h6 h2h3 d7d5 c4b3 c8e6 b3c2
info nodes 69935497 nps 2.333.127 hashfull 414

maybe Arena GUI has some problems. I have to make more tests inside other GUIs.

[mar]

could it be that the other build uses multiple threads by default?