Sabrina 3.1.26 released (Windows/Linux/Mac OSX)

[tpoppins]

If this helps here's the output for this system:

Code: Select all

GenuineIntel
Intel(R) Xeon(R) CPU           X5670  @ 2.93GHz
3DNOW not supported
3DNOWEXT not supported
ABM not supported
ADX not supported
AES supported
AVX not supported
AVX2 not supported
AVX512CD not supported
AVX512ER not supported
AVX512F not supported
AVX512PF not supported
BMI1 not supported
BMI2 not supported
CLFSH supported
CMPXCHG16B supported
CX8 supported
ERMS not supported
F16C not supported
FMA not supported
FSGSBASE not supported
FXSR supported
HLE not supported
INVPCID not supported
LAHF supported
LZCNT not supported
MMX supported
MMXEXT not supported
MONITOR supported
MOVBE not supported
MSR supported
OSXSAVE not supported
PCLMULQDQ supported
POPCNT supported
PREFETCHWT1 not supported
RDRAND not supported
RDSEED not supported
RDTSCP supported
RTM not supported
SEP supported
SHA not supported
SSE supported
SSE2 supported
SSE3 supported
SSE4.1 supported
SSE4.2 supported
SSE4a not supported
SSSE3 supported
SYSCALL supported
TBM not supported
XOP not supported
XSAVE not supported

Perhaps it would be simpler to make a separate exec that wouldn't use POPCNT? The Elo difference would be probably in single digits, if that.

[stegemma]

I'm focused in the wrong direction: it were not the cpuid call the problem but the calling of a function not protected by the bPopCountOk flag.

I've added a command line parameter, to exclude popcnt and other intrinsic but this is not mandatory (Sabrina must disables it automatically):

Code: Select all

Sabrina -nopopcnt

This new beta should works with or without the parameter.

If this problem has been solved, I go further with other problems with WinBoard and finally I'll release the official version for all operating systems.

https://www.linformatica.com/software/S ... 7.beta.zip

Thanks to all.

[stegemma]

If this helps here's the output for this system:

Code: Select all

GenuineIntel
Intel(R) Xeon(R) CPU           X5670  @ 2.93GHz
[...]
BMI1 not supported
BMI2 not supported
[...]
POPCNT supported
[...]

Perhaps it would be simpler to make a separate exec that wouldn't use POPCNT? The Elo difference would be probably in single digits, if that.

So your system supports popcnt but maybe not blsi. I need to use two different flags, for this, but you can try with -nopopcnt, that actually disables both functions.

Thanks.

[alessandro]

Hi Stefano,

I've looked on Sabrina 3.1.27-beta on both my PCs.
On the old Surface 3 it doesn't work. It has an Intel Atom processor. Sabrina start and then exit almost immediately.
On the Surface Pro, Intel i5, runs fine.

Nice job, anyway!

[stegemma]

Have you try with -nopopcnt?

[tpoppins]

The latest beta doesn't crash with "-nopopcnt".

[stegemma]

Thanks. Now I'm working on the final release, that would be a little stronger and with the WB interface problem corrected.

[stegemma]

This beta also crashes. Here is a ZIP containing Windoze WER error reports, including a minidump.

I ran the v3.1.26 w32 exec under Cute Chess 1.1.0 pre-release and here are some observations.

1) If the starting position is provided via FEN (as it would be with a GUI book in EPD format) Sabrina loses on time at 40/2. Below is the eng-GUI log.

Code: Select all

Sabrina 3.1.26(47): # soul: Simple5
<Sabrina 3.1.26(47): # TT 1048576 entries - 64 MiB
<Sabrina 3.1.26(47): # Sabrina 3.1.26 (c) 2012-2018 Stefano Gemma
<Sabrina 3.1.26(47): # zlib (c) 1995-2011 Mark Adler
<Sabrina 3.1.26(47): # crypt (c) Brad Conte
<Sabrina 3.1.26(47): # xorshift (c) George Marsaglia
<Sabrina 3.1.26(47): feature ping=1 setboard=1 done=1 san=0 usermove=1 time=1 draw=1 sigint=0 sigterm=0 reuse=1 analyze=0 myname="Sabrina 3.1.26" colors=0 ics=1 name=1 nps=0 debug=1 memory=1 smp=1
[...]
>Sabrina 3.1.26(47): new
>Sabrina 3.1.26(47): force
>Sabrina 3.1.26(47): setboard R7/2K5/8/2k5/8/8/8/8 w - - 0 144
>Sabrina 3.1.26(47): level 40 2 0
>Sabrina 3.1.26(47): post
>Sabrina 3.1.26(47): easy
>Sabrina 3.1.26(47): computer
>Sabrina 3.1.26(47): name Stockfish 180627 64-bit
>Sabrina 3.1.26(47): ping 10
<Sabrina 3.1.26(47): # hello my peer!
<Sabrina 3.1.26(47): Sabrina 3.1.26
<Sabrina 3.1.26(47): pong 10
>Sabrina 3.1.26(47): time 12000
otim 12000
>Sabrina 3.1.26(47): go
<Sabrina 3.1.26(47): # moves 144/40 - time 18446744073708376/120000 ms
<Sabrina 3.1.26(47): # engine's running
<Sabrina 3.1.26(47): # 19 moves to analyze
<Sabrina 3.1.26(47): 10 5.05  100 3636313 Ra8-d8
[...]
>Sabrina 3.1.26(47): ?
<Sabrina 3.1.26(47): # engine stopped
<Sabrina 3.1.26(47): 14 5.11  12500 442826628 Ra8-d8
<Sabrina 3.1.26(47): # 125002 ms 3542 Knps, TT hits: 148886770 33.62%, Rept.Hits: 0
<Sabrina 3.1.26(47): move a8d8
>Sabrina 3.1.26(47): force
>Sabrina 3.1.26(47): result 0-1 {White loses on time}
>Sabrina 3.1.26(47): ping 4
<Sabrina 3.1.26(47): # result 0-1 {White loses on time}
<Sabrina 3.1.26(47): pong 4

This issue doesn't occur with a GUI book in PGN format.

2) Sabrina sends eval in x.xx format whereas it should be in centipawns. Thus no evals are displayed by the GUI or written to a PGN file.

3) done=1 is not at the end of the "feature" command. Does that mean that the features following "done=1" are not implemented? For example, "memory 256" init string has no effect -- Task Manager still reports a memory footprint around 70MB.

4) In the log above the "Error (unknown command): accepted" and "Error (unknown command): rejected" parts look very odd and could be a symptom of some issue.

Thank you for your time, Stefan.

All points are been solved now. Point 1 was related to commands "time" and "level" with a "setboard" that starts from 144 moves. Because the slot of moves to time control has been set to 40, 144 was greater than that number and the computed ms for actual move was wrong: "time 18446744073708376/120000 ms"!

Now Sabrina computes the remaining moves to next slot (144 % 40 = 24) aad uses that value to get the time for the actual move.

Still I don't know if time/otim are referred to the time remaining to reach the next move control or not, but the program should never lose on time for this.