Minic version 3

[AlexChess]

I've played 500 chess960 games so far, and it is unfortunately -95 Elo to version 3.06. I'll check the engine set-up, but I don't think there is any issue there.

This is very possible, this net was not trained with FRC. But 95Elo seems a lot, I'll check that !

Hi Vivien!
When you likes, I'm ready to test also Minic 3.0.x on Windows 10 | 11 ARM64.

http://talkchess.com/forum3/viewtopic.p ... 10#p896907

You could ask to V. Cherbyna and Andreas Matthies how they have compiled their engines for it.

Best regards! AlexChess
C'est un plaisir, ma mère est belge francophone

[xr_a_y]

Hi Vivien!
When you likes, I'm ready to test also Minic 3.0.x on Windows 10 | 11 ARM64.

http://talkchess.com/forum3/viewtopic.p ... 10#p896907

You could ask to V. Cherbyna and Andreas Matthies how they have compiled their engines for it.

Best regards! AlexChess
C'est un plaisir, ma mère est belge francophone

This is a good question. I'd love to compile for ARM64 Windows, and for other systems also but I'd really appreciate to be able to do so from Linux.

For Windows x32/x64 I have mingw, for Android arm/arm64/linux I have the android NDK, but for Windows arm64 I have nothing yet under linux.

[AlexChess]

Hi Vivien!
When you likes, I'm ready to test also Minic 3.0.x on Windows 10 | 11 ARM64.

http://talkchess.com/forum3/viewtopic.p ... 10#p896907

You could ask to V. Cherbyna and Andreas Matthies how they have compiled their engines for it.

Best regards! AlexChess
C'est un plaisir, ma mère est belge francophone

This is a good question. I'd love to compile for ARM64 Windows, and for other systems also but I'd really appreciate to be able to do so from Linux.

For Windows x32/x64 I have mingw, for Android arm/arm64/linux I have the android NDK, but for Windows arm64 I have nothing yet under linux.

Hi!

Ok, I'll wait
PS: Just testing Minic 3.0.8 ARM64 (taken from your github repository) on Ubuntu, but it doesn't start (also after chmod +x) . The net is embedded. Minic 3.0.6 ARM64 without embedded net works fine.

Regards, AlexChess

[xr_a_y]

Oh ok, i'd love to debug that, I have to find a more secure way to cross-compiler and test all this ... let's try qemu

mmm indeed :

Code: Select all

qemu-aarch64 ./Dist/Minic3/minic_dev_linux_x64_armv8 bench 10
info string This is Minic version 3.09 (NNUE available)
terminate called after throwing an instance of 'std::system_error'
  what():  Unknown error 167851120
qemu: uncaught target signal 6 (Aborted) - core dumped
Abandon (core dumped)

Can be a threading thing maybe, but not sure you encounter the same issue.

[CMCanavessi]

I'm having loses on time / crashes with 3.08 (previous versions were ok).

Using FRC, any TC (tried from 1 minute to 30 minutes), ponder ON (maybe here's the issue), 1 thread.

[xr_a_y]

I'm having loses on time / crashes with 3.08 (previous versions were ok).

Using FRC, any TC (tried from 1 minute to 30 minutes), ponder ON (maybe here's the issue), 1 thread.

Indeed, pondering was refactored and not much tested. I'll have a look at that asap.

[xr_a_y]

I'm having loses on time / crashes with 3.08 (previous versions were ok).

Using FRC, any TC (tried from 1 minute to 30 minutes), ponder ON (maybe here's the issue), 1 thread.

Indeed, pondering was refactored and not much tested. I'll have a look at that asap.

Weird situation, it seems possible that when pondering, Minic is receiving a "stop" command "too soon" to be properly handled.

Here is a simplified log

Code: Select all

readyok
info string Info  2021-07-05 21:08:02-138: Waiting for input ...
]info string Info  2021-07-05 21:08:02-138: Received command : "go ponder wtime 910 btime 331 winc 100 binc 100"     <<<<=== ponder ask
info string Info  2021-07-05 21:08:02-138: uci received command go
info string Info  2021-07-05 21:08:02-138: received parameter ponder
[......]
info string Info  2021-07-05 21:08:02-138: Search Sync   <<<<=== launching a search
info string Info  2021-07-05 21:08:02-138: Thread waiting 0
info string Info  2021-07-05 21:08:02-138: state set to 0
info string Info  2021-07-05 21:08:02-138: Locking other threads
info string Info  2021-07-05 21:08:02-138: Calling main thread startThread
info string Info  2021-07-05 21:08:02-138: Starting worker 0
info string Info  2021-07-05 21:08:02-138: uci async started
info string Info  2021-07-05 21:08:02-138: Waiting for input ...
info string Info  2021-07-05 21:08:02-139: Received command : "stop" <<<<=== stop ask while lauching search
info string Info  2021-07-05 21:08:02-139: uci received command stop
info string Info  2021-07-05 21:08:02-139: stop requested
info string Info  2021-07-05 21:08:02-139: Stopping previous search
info string Info  2021-07-05 21:08:02-139: Setting stopflag on every threads  <<<<=== setting stopfalg to true in respond to the requested stop
info string Info  2021-07-05 21:08:02-139: Waiting for input ...
info string Info  2021-07-05 21:08:02-141: Search launched for thread 0 <<<<=== setting stopflag to false in respond to thread start !
info string Info  2021-07-05 21:08:02-141: Init stat
info string Info  2021-07-05 21:08:02-141: Init killers
info string Info  2021-07-05 21:08:02-141: Init history
info string Info  2021-07-05 21:08:02-141: Init counter
info string Info  2021-07-05 21:08:02-141: Search params :
info string Info  2021-07-05 21:08:02-141: requested time  2592000000 (138)  <<<<=== using no time control for now (because pondering)
info string Info  2021-07-05 21:08:02-141: requested depth 127
info string Info  2021-07-05 21:08:02-141: MultiPV 1
info string Info  2021-07-05 21:08:02-141: Thread 0 searching depth 1
info string Info  2021-07-05 21:08:02-141: Dynamic contempt 46 23
info depth 1 score cp -131 time 1 nodes 17 nps 17000 seldepth 1 tbhits 0 pv e4c3
[.....]

Search will never end because Minic is expecting stop or ponderhint but GUI already send stop and will thus never send ponderhit.

A classic race condition. I have to find a better way to handle this. Very sorry.

[CMCanavessi]

What GUI is that? From what I've seen (using Banksia), is that Minic starts "thinking" (can see high cpu usage) and then suddenly stops thinking (can see 0% cpu usage), but the process is still there, so it's not a crash "per-se", more like a loss of sync with the GUI. GUI is waiting for Minic, and Minic is waiting for the GUI.

[xr_a_y]

What GUI is that? From what I've seen (using Banksia), is that Minic starts "thinking" (can see high cpu usage) and then suddenly stops thinking (can see 0% cpu usage), but the process is still there, so it's not a crash "per-se", more like a loss of sync with the GUI. GUI is waiting for Minic, and Minic is waiting for the GUI.

Was with cutechess. I ll try banksia today.

[xr_a_y]

What GUI is that? From what I've seen (using Banksia), is that Minic starts "thinking" (can see high cpu usage) and then suddenly stops thinking (can see 0% cpu usage), but the process is still there, so it's not a crash "per-se", more like a loss of sync with the GUI. GUI is waiting for Minic, and Minic is waiting for the GUI.

Was with cutechess. I ll try banksia today.

Yeah same issue with Banksia with a little more complicated outcome as Banksia seems to repeat the stop command when something is looks wrong,
Minic sees

Code: Select all

info string Info  2021-07-06 09:41:07-075: Received command : "go ponder wtime 20084 btime 15726 winc 100 binc 100"
[....] <<<<===  launching the search
info string Info  2021-07-06 09:41:07-075: Received command : "stop"    <<<<==== but receive a stop command (should set stopflag to true
[....]
info string Info  2021-07-06 09:41:07-075: Thread 0 searching depth 1 <<<<<=== but set later stopflag to false
[...]  search for a while
info depth 21 score cp 67 time 13054 nodes 8388091 nps 642568 seldepth 32 tbhits 0 hashfull 984 pv d6f8 f2f4 a3e7 g1h1 b8d7 e1g3 e7d8 g3f3 d7b6 g2g4 b6c4 d2e1 c4a3 b1b3 f6d7 c3e2 d7e5 f4e5 a3c4 
info string Info  2021-07-06 09:41:23-414: Received command : "stop"   <<<<<==== banksia re-send stop again (much later)
[....]
bestmove d6f8 ponder f2f4  <<<<<=== Minic is sending move to GUI
[....]
info string Info  2021-07-06 09:41:23-418: Received command : "position startpos moves e2e4 e7e6 d2d4 d7d5 e4d5 e6d5 g1f3 g8f6 f1d3 f8d6 e1g1 e8g8 b1c3 c7c6 h2h3 f8e8 f1e1 e8e1 d1e1 h7h6 f3e5 d8b6 c1e3 b6b2 a1b1 b2a3 e3h6"   <<<<<=== banksia send next position
[.....]
info string Info  2021-07-06 09:41:23-418: Received command : "go wtime 20074 btime 15826 winc 100 binc 100"   <<<<==== and ask for a search
info string Info  2021-07-06 09:41:23-913: Received command : "stop"  <<<<<==== and stop it quit fast, probably GUI 

I'm pretty sure the root cause is the same. I'll try to fix this asap.