Page 3 of 3
Re: Rybka 4 vs Stockfish 1.8 - Long TC Match - LIVE BROADCAS
Posted: Thu Jul 22, 2010 1:09 pm
by Robert Houdart
Yes, that is quite confusing, as everywhere in the SF source code there are references to POPCNT-enabled code via a <bool HasPopCnt> template.
But Tord is absolutely correct, I now see that the usage of the POPCNT routines has been disabled by default in bitcount.h (with a compiler conditional "USE_POPCNT"). To obtain the behaviour I described, you need to compile with the "USE_POPCNT" conditional on.
Interestingly Tord's estimate of about 3% speed improvement confirms my findings with Houdini, basically the POPCNT versions are hardly worth the while.
Robert
Re: Rybka 4 vs Stockfish 1.8 - Long TC Match - LIVE BROADCAS
Posted: Thu Jul 22, 2010 1:42 pm
by Martin Thoresen
Robert Houdart wrote:Yes, that is quite confusing, as everywhere in the SF source code there are references to POPCNT-enabled code via a <bool HasPopCnt> template.
But Tord is absolutely correct, I now see that the usage of the POPCNT routines has been disabled by default in bitcount.h (with a compiler conditional "USE_POPCNT"). To obtain the behaviour I described, you need to compile with the "USE_POPCNT" conditional on.
Interestingly Tord's estimate of about 3% speed improvement confirms my findings with Houdini, basically the POPCNT versions are hardly worth the while.
Robert
Well, if one has a CPU which supports SSE 4.2 then it's very little effort from the
end-user point of view (me). But I can see that it's not always easy to compile this stuff.
Robert, hopefully you will show up on
chessbomb.com for the chat. I think many people would like to see you there.
Right now in Game #1 Houdini is under pressure.
Re: Rybka 4 vs Stockfish 1.8 - Long TC Match - LIVE BROADCAS
Posted: Thu Jul 22, 2010 4:13 pm
by Martin Thoresen
Robert,
This is a quote from Tord during game #1 when Houdini uses tons of time in forced mate positions:
TordRomstad: It would have been over a lot sooner if Houdini had played instantly when there's a forced mate, as most engines do.
Is this something you can implement?
Might not be the coolest feature, but it's handy when it comes to games like this one.
Best Regards,
Martin
Re: Rybka 4 vs Stockfish 1.8 - Long TC Match - LIVE BROADCAS
Posted: Thu Jul 22, 2010 5:40 pm
by Robert Houdart
Martin Thoresen wrote:Robert,
This is a quote from Tord during game #1 when Houdini uses tons of time in forced mate positions:
TordRomstad: It would have been over a lot sooner if Houdini had played instantly when there's a forced mate, as most engines do.
Is this something you can implement?
Might not be the coolest feature, but it's handy when it comes to games like this one.
Best Regards,
Martin
A very nice attack by SF in the first game - I find king safety by far the most difficult component to improve, as this game demonstrates
.
I'll take a look at the time usage when there's a forced mate, but why didn't the GUI stop the game around move 40 when both programs were already announcing mate scores ?
Robert