Crafty 19.15

[Bryan Hofmann]

Geschrieben von:/Posted by: Bryan Hofmann at 13 July 2004 23:23:15:

Crafty 19.15 is ready for download at http://home.earthlink.net/~bkhofmann/ should be the fastest Crafty yet....



http://home.earthlink.net/~bkhofmann/

[Dann Corbit]

Geschrieben von:/Posted by: Dann Corbit at 14 July 2004 00:06:29:
Als Antwort auf:/In reply to: Crafty 19.15 geschrieben von:/posted by: Bryan Hofmann at 13 July 2004 23:23:15:

Crafty 19.15 is ready for download at http://home.earthlink.net/~bkhofmann/ should be the fastest Crafty yet....

That SSE stuff is not maintained, you know.



my ftp site {remove http:// unless you like error messages}

[Dann Corbit]

Geschrieben von:/Posted by: Dann Corbit at 14 July 2004 00:07:56:
Als Antwort auf:/In reply to: Re: Crafty 19.15 geschrieben von:/posted by: Dann Corbit at 14 July 2004 00:06:29:

Crafty 19.15 is ready for download at http://home.earthlink.net/~bkhofmann/ should be the fastest Crafty yet....

That SSE stuff is not maintained, you know.

Recommend:
do builds that do not have the extensions. Dr. Hyatt has not worked on that stuff for a very long time, and I would not be surprised if it were to break all over the place.



my ftp site {remove http:// unless you like error messages}

[Bryan Hofmann]

Geschrieben von:/Posted by: Bryan Hofmann at 14 July 2004 01:19:16:
Als Antwort auf:/In reply to: Re: Crafty 19.15 geschrieben von:/posted by: Dann Corbit at 14 July 2004 00:07:56:

Crafty 19.15 is ready for download at http://home.earthlink.net/~bkhofmann/ should be the fastest Crafty yet....

That SSE stuff is not maintained, you know.

Recommend:
do builds that do not have the extensions. Dr. Hyatt has not worked on that stuff for a very long time, and I would not be surprised if it were to break all over the place.

I have tested these versions and they all work; in addition I run them on FICS and have done so for quite some time. I am a bit suprised by your comment stating it will break all over the place. Have you even tried any of these compiles and found problems? The SSE/SSE2 helps on Intel CPU speeds in my testing and there has never been a problem with them.

[Dann Corbit]

Geschrieben von:/Posted by: Dann Corbit at 14 July 2004 01:55:56:
Als Antwort auf:/In reply to: Re: Crafty 19.15 geschrieben von:/posted by: Bryan Hofmann at 14 July 2004 01:19:16:

Crafty 19.15 is ready for download at http://home.earthlink.net/~bkhofmann/ should be the fastest Crafty yet....

That SSE stuff is not maintained, you know.

Recommend:
do builds that do not have the extensions. Dr. Hyatt has not worked on that stuff for a very long time, and I would not be surprised if it were to break all over the place.

I have tested these versions and they all work; in addition I run them on FICS and have done so for quite some time. I am a bit suprised by your comment stating it will break all over the place.
Have you even tried any of these compiles and found problems? The SSE/SSE2 helps on Intel CPU speeds in my testing and there has never been a problem with them.

I would not be surprised if it is broken, since he has not used that pathway in a long time.
I did search extension compiles before and I always had to do a lot of fiddling to ensure that things were not broken.
Do you run lint against the code base with your proposed flags defined?



my ftp site {remove http:// unless you like error messages}

[Bryan Hofmann]

Geschrieben von:/Posted by: Bryan Hofmann at 14 July 2004 04:11:11:
Als Antwort auf:/In reply to: Re: Crafty 19.15 geschrieben von:/posted by: Dann Corbit at 14 July 2004 01:55:56:

Crafty 19.15 is ready for download at http://home.earthlink.net/~bkhofmann/ should be the fastest Crafty yet....

That SSE stuff is not maintained, you know.

Recommend:
do builds that do not have the extensions. Dr. Hyatt has not worked on that stuff for a very long time, and I would not be surprised if it were to break all over the place.

I have tested these versions and they all work; in addition I run them on FICS and have done so for quite some time. I am a bit suprised by your comment stating it will break all over the place.
Have you even tried any of these compiles and found problems? The SSE/SSE2 helps on Intel CPU speeds in my testing and there has never been a problem with them.

I would not be surprised if it is broken, since he has not used that pathway in a long time.
I did search extension compiles before and I always had to do a lot of fiddling to ensure that things were not broken.
Do you run lint against the code base with your proposed flags defined?

Then maybe you need to contact the author of the program and let him know that he does not know how to compile his own program. If you look at the default make file for the icc-profile (this is what Hyatt uses for the ICC) it has the compile time option of -xN which mean optimize for the pentium 4. Guess what this does? It uses SSE/SSE2. So I guess if it is good enough for the author of the program to use should it not be offered to the other Crafty users.

[Dann Corbit]

Geschrieben von:/Posted by: Dann Corbit at 14 July 2004 05:06:57:
Als Antwort auf:/In reply to: Re: Crafty 19.15 geschrieben von:/posted by: Bryan Hofmann at 14 July 2004 04:11:11:

Crafty 19.15 is ready for download at http://home.earthlink.net/~bkhofmann/ should be the fastest Crafty yet....

That SSE stuff is not maintained, you know.

Recommend:
do builds that do not have the extensions. Dr. Hyatt has not worked on that stuff for a very long time, and I would not be surprised if it were to break all over the place.

I have tested these versions and they all work; in addition I run them on FICS and have done so for quite some time. I am a bit suprised by your comment stating it will break all over the place.
Have you even tried any of these compiles and found problems? The SSE/SSE2 helps on Intel CPU speeds in my testing and there has never been a problem with them.

I would not be surprised if it is broken, since he has not used that pathway in a long time.
I did search extension compiles before and I always had to do a lot of fiddling to ensure that things were not broken.
Do you run lint against the code base with your proposed flags defined?

Then maybe you need to contact the author of the program and let him know that he does not know how to compile his own program. If you look at the default make file for the icc-profile (this is what Hyatt uses for the ICC) it has the compile time option of -xN which mean optimize for the pentium 4. Guess what this does? It uses SSE/SSE2. So I guess if it is good enough for the author of the program to use should it not be offered to the other Crafty users.

That does not put SSE/SSE2 instructions into the program. It just chooses an instruction mix that is better for pentium 4. Which is neither here nor there. What I was trying to say is that all of these builds are questionable because they use aging code that is no longer tested. Not the comiler generated instructions, but the code patches for extensions.
Of course, I have not fiddled with extensions for a couple years. So maybe it is rock solid now. Of course, I use a less-tested extension too (DETECTDRAW) that could well introduce bugs.



my ftp site {remove http:// unless you like error messages}

[Bryan Hofmann]

Geschrieben von:/Posted by: Bryan Hofmann at 14 July 2004 14:44:19:
Als Antwort auf:/In reply to: Re: Crafty 19.15 geschrieben von:/posted by: Dann Corbit at 14 July 2004 05:06:57:

Crafty 19.15 is ready for download at http://home.earthlink.net/~bkhofmann/ should be the fastest Crafty yet....

That SSE stuff is not maintained, you know.

Recommend:
do builds that do not have the extensions. Dr. Hyatt has not worked on that stuff for a very long time, and I would not be surprised if it were to break all over the place.

I have tested these versions and they all work; in addition I run them on FICS and have done so for quite some time. I am a bit suprised by your comment stating it will break all over the place.
Have you even tried any of these compiles and found problems? The SSE/SSE2 helps on Intel CPU speeds in my testing and there has never been a problem with them.

I would not be surprised if it is broken, since he has not used that pathway in a long time.
I did search extension compiles before and I always had to do a lot of fiddling to ensure that things were not broken.
Do you run lint against the code base with your proposed flags defined?

Then maybe you need to contact the author of the program and let him know that he does not know how to compile his own program. If you look at the default make file for the icc-profile (this is what Hyatt uses for the ICC) it has the compile time option of -xN which mean optimize for the pentium 4. Guess what this does? It uses SSE/SSE2. So I guess if it is good enough for the author of the program to use should it not be offered to the other Crafty users.

That does not put SSE/SSE2 instructions into the program. It just chooses an instruction mix that is better for pentium 4.
Which is neither here nor there. What I was trying to say is that all of these >builds are questionable because they use aging code that is no longer tested. >Not the comiler generated instructions, but the code patches for extensions.
Of course, I have not fiddled with extensions for a couple years. So maybe it is rock solid now. Of course, I use a less-tested extension too (DETECTDRAW) that could well introduce bugs.

Not correct. When the -xN is specified the Vectoring component of the Intel compiler is enabled. This automaticly uses SIMD instructions (MMX SSE SSE2) as per the manual that comes with the Intel Compiler.

The only Crafty flag that is used in my builds are the FUTILITY. Here again if you look at the way the author of the program compiles his, he uses -DFUTILITY on the ICC system builds.

The author of the program stated he stopped using this option due to the slow down in performance. The cost was higher then the benefit.
The builds I have created are as close as possible to the builds that Hyatt uses himself. They are compiled for speed and the only mods I make are with this interest in mind and they do not effect the search, eval or other aspects of how the program was intended to play.

[David Dahlem]

Geschrieben von:/Posted by: David Dahlem at 14 July 2004 18:10:56:
Als Antwort auf:/In reply to: Re: Crafty 19.15 geschrieben von:/posted by: Bryan Hofmann at 14 July 2004 14:44:19:

Crafty 19.15 is ready for download at http://home.earthlink.net/~bkhofmann/ should be the fastest Crafty yet....

That SSE stuff is not maintained, you know.

Recommend:
do builds that do not have the extensions. Dr. Hyatt has not worked on that stuff for a very long time, and I would not be surprised if it were to break all over the place.

I have tested these versions and they all work; in addition I run them on FICS and have done so for quite some time. I am a bit suprised by your comment stating it will break all over the place.
Have you even tried any of these compiles and found problems? The SSE/SSE2 helps on Intel CPU speeds in my testing and there has never been a problem with them.

I would not be surprised if it is broken, since he has not used that pathway in a long time.
I did search extension compiles before and I always had to do a lot of fiddling to ensure that things were not broken.
Do you run lint against the code base with your proposed flags defined?

Then maybe you need to contact the author of the program and let him know that he does not know how to compile his own program. If you look at the default make file for the icc-profile (this is what Hyatt uses for the ICC) it has the compile time option of -xN which mean optimize for the pentium 4. Guess what this does? It uses SSE/SSE2. So I guess if it is good enough for the author of the program to use should it not be offered to the other Crafty users.

That does not put SSE/SSE2 instructions into the program. It just chooses an instruction mix that is better for pentium 4.
Which is neither here nor there. What I was trying to say is that all of these >builds are questionable because they use aging code that is no longer tested. >Not the comiler generated instructions, but the code patches for extensions.
Of course, I have not fiddled with extensions for a couple years. So maybe it is rock solid now. Of course, I use a less-tested extension too (DETECTDRAW) that could well introduce bugs.

Not correct. When the -xN is specified the Vectoring component of the Intel compiler is enabled. This automaticly uses SIMD instructions (MMX SSE SSE2) as per the manual that comes with the Intel Compiler.

The only Crafty flag that is used in my builds are the FUTILITY. Here again if you look at the way the author of the program compiles his, he uses -DFUTILITY on the ICC system builds.
The author of the program stated he stopped using this option due to the slow down in performance. The cost was higher then the benefit.
The builds I have created are as close as possible to the builds that Hyatt uses himself. They are compiled for speed and the only mods I make are with this interest in mind and they do not effect the search, eval or other aspects of how the program was intended to play.

I'm sure all this is confusing to the average Crafty user, at least it is to me. How to decide which version to use? Begging for help from Prof. Hyatt.
Dave

[Bryan Hofmann]

Geschrieben von:/Posted by: Bryan Hofmann at 14 July 2004 18:59:03:
Als Antwort auf:/In reply to: Re: Crafty 19.15 geschrieben von:/posted by: David Dahlem at 14 July 2004 18:10:56:

Crafty 19.15 is ready for download at http://home.earthlink.net/~bkhofmann/ should be the fastest Crafty yet....

That SSE stuff is not maintained, you know.

Recommend:
do builds that do not have the extensions. Dr. Hyatt has not worked on that stuff for a very long time, and I would not be surprised if it were to break all over the place.

I have tested these versions and they all work; in addition I run them on FICS and have done so for quite some time. I am a bit suprised by your comment stating it will break all over the place.
Have you even tried any of these compiles and found problems? The SSE/SSE2 helps on Intel CPU speeds in my testing and there has never been a problem with them.

I would not be surprised if it is broken, since he has not used that pathway in a long time.
I did search extension compiles before and I always had to do a lot of fiddling to ensure that things were not broken.
Do you run lint against the code base with your proposed flags defined?

Then maybe you need to contact the author of the program and let him know that he does not know how to compile his own program. If you look at the default make file for the icc-profile (this is what Hyatt uses for the ICC) it has the compile time option of -xN which mean optimize for the pentium 4. Guess what this does? It uses SSE/SSE2. So I guess if it is good enough for the author of the program to use should it not be offered to the other Crafty users.

That does not put SSE/SSE2 instructions into the program. It just chooses an instruction mix that is better for pentium 4.
Which is neither here nor there. What I was trying to say is that all of these >builds are questionable because they use aging code that is no longer tested. >Not the comiler generated instructions, but the code patches for extensions.
Of course, I have not fiddled with extensions for a couple years. So maybe it is rock solid now. Of course, I use a less-tested extension too (DETECTDRAW) that could well introduce bugs.

Not correct. When the -xN is specified the Vectoring component of the Intel compiler is enabled. This automaticly uses SIMD instructions (MMX SSE SSE2) as per the manual that comes with the Intel Compiler.

The only Crafty flag that is used in my builds are the FUTILITY. Here again if you look at the way the author of the program compiles his, he uses -DFUTILITY on the ICC system builds.
The author of the program stated he stopped using this option due to the slow down in performance. The cost was higher then the benefit.
The builds I have created are as close as possible to the builds that Hyatt uses himself. They are compiled for speed and the only mods I make are with this interest in mind and they do not effect the search, eval or other aspects of how the program was intended to play.

I'm sure all this is confusing to the average Crafty user, at least it is to me. How to decide which version to use? Begging for help from Prof. Hyatt.
Dave

Dr Hyatt does not visit this site as far as I know. As to which one to use ... Well that's the great thing about choices ... Pick one or more then run it if you like it keep if you don't try another. The Crafty any X86 on my site should run with any x86 CPU and uses the same Crafty flags as Dr. Hyatt does with the version he runs on the ICC.
Bryan

[David Dahlem]

Geschrieben von:/Posted by: David Dahlem at 14 July 2004 19:55:36:
Als Antwort auf:/In reply to: Re: Crafty 19.15 geschrieben von:/posted by: Bryan Hofmann at 14 July 2004 18:59:03:

Crafty 19.15 is ready for download at http://home.earthlink.net/~bkhofmann/ should be the fastest Crafty yet....

That SSE stuff is not maintained, you know.

Recommend:
do builds that do not have the extensions. Dr. Hyatt has not worked on that stuff for a very long time, and I would not be surprised if it were to break all over the place.

I have tested these versions and they all work; in addition I run them on FICS and have done so for quite some time. I am a bit suprised by your comment stating it will break all over the place.
Have you even tried any of these compiles and found problems? The SSE/SSE2 helps on Intel CPU speeds in my testing and there has never been a problem with them.

I would not be surprised if it is broken, since he has not used that pathway in a long time.
I did search extension compiles before and I always had to do a lot of fiddling to ensure that things were not broken.
Do you run lint against the code base with your proposed flags defined?

Then maybe you need to contact the author of the program and let him know that he does not know how to compile his own program. If you look at the default make file for the icc-profile (this is what Hyatt uses for the ICC) it has the compile time option of -xN which mean optimize for the pentium 4. Guess what this does? It uses SSE/SSE2. So I guess if it is good enough for the author of the program to use should it not be offered to the other Crafty users.

That does not put SSE/SSE2 instructions into the program. It just chooses an instruction mix that is better for pentium 4.
Which is neither here nor there. What I was trying to say is that all of these >builds are questionable because they use aging code that is no longer tested. >Not the comiler generated instructions, but the code patches for extensions.
Of course, I have not fiddled with extensions for a couple years. So maybe it is rock solid now. Of course, I use a less-tested extension too (DETECTDRAW) that could well introduce bugs.

Not correct. When the -xN is specified the Vectoring component of the Intel compiler is enabled. This automaticly uses SIMD instructions (MMX SSE SSE2) as per the manual that comes with the Intel Compiler.

The only Crafty flag that is used in my builds are the FUTILITY. Here again if you look at the way the author of the program compiles his, he uses -DFUTILITY on the ICC system builds.
The author of the program stated he stopped using this option due to the slow down in performance. The cost was higher then the benefit.
The builds I have created are as close as possible to the builds that Hyatt uses himself. They are compiled for speed and the only mods I make are with this interest in mind and they do not effect the search, eval or other aspects of how the program was intended to play.

I'm sure all this is confusing to the average Crafty user, at least it is to me. How to decide which version to use? Begging for help from Prof. Hyatt.
Dave

Dr Hyatt does not visit this site as far as I know. As to which one to use ... Well that's the great thing about choices ... Pick one or more then run it if you like it keep if you don't try another. The Crafty any X86 on my site should run with any x86 CPU and uses the same Crafty flags as Dr. Hyatt does with the version he runs on the ICC.
Bryan

Hi Bryan
Yes, choices are great, but ... with all this talk about "broken", i sure don't want to spend lots of time testing a version only to find out it was broken. I bench tested your version and Danns version. Your version is slightly better on NPS and slightly worse on time to ply. So bench results are a toss-up. So i must choose a version based on compile options, which i know nothing about, or spend lots of time testing both versions. By then, version 19.16 will probably be released.
Regards
Dave

[Bryan Hofmann]

Geschrieben von:/Posted by: Bryan Hofmann at 14 July 2004 20:15:47:
Als Antwort auf:/In reply to: Re: Crafty 19.15 geschrieben von:/posted by: David Dahlem at 14 July 2004 19:55:36:

Crafty 19.15 is ready for download at http://home.earthlink.net/~bkhofmann/ should be the fastest Crafty yet....

That SSE stuff is not maintained, you know.

Recommend:
do builds that do not have the extensions. Dr. Hyatt has not worked on that stuff for a very long time, and I would not be surprised if it were to break all over the place.

I have tested these versions and they all work; in addition I run them on FICS and have done so for quite some time. I am a bit suprised by your comment stating it will break all over the place.
Have you even tried any of these compiles and found problems? The SSE/SSE2 helps on Intel CPU speeds in my testing and there has never been a problem with them.

I would not be surprised if it is broken, since he has not used that pathway in a long time.
I did search extension compiles before and I always had to do a lot of fiddling to ensure that things were not broken.
Do you run lint against the code base with your proposed flags defined?

Then maybe you need to contact the author of the program and let him know that he does not know how to compile his own program. If you look at the default make file for the icc-profile (this is what Hyatt uses for the ICC) it has the compile time option of -xN which mean optimize for the pentium 4. Guess what this does? It uses SSE/SSE2. So I guess if it is good enough for the author of the program to use should it not be offered to the other Crafty users.

That does not put SSE/SSE2 instructions into the program. It just chooses an instruction mix that is better for pentium 4.
Which is neither here nor there. What I was trying to say is that all of these >builds are questionable because they use aging code that is no longer tested. >Not the comiler generated instructions, but the code patches for extensions.
Of course, I have not fiddled with extensions for a couple years. So maybe it is rock solid now. Of course, I use a less-tested extension too (DETECTDRAW) that could well introduce bugs.

Not correct. When the -xN is specified the Vectoring component of the Intel compiler is enabled. This automaticly uses SIMD instructions (MMX SSE SSE2) as per the manual that comes with the Intel Compiler.

The only Crafty flag that is used in my builds are the FUTILITY. Here again if you look at the way the author of the program compiles his, he uses -DFUTILITY on the ICC system builds.
The author of the program stated he stopped using this option due to the slow down in performance. The cost was higher then the benefit.
The builds I have created are as close as possible to the builds that Hyatt uses himself. They are compiled for speed and the only mods I make are with this interest in mind and they do not effect the search, eval or other aspects of how the program was intended to play.

I'm sure all this is confusing to the average Crafty user, at least it is to me. How to decide which version to use? Begging for help from Prof. Hyatt.
Dave

Dr Hyatt does not visit this site as far as I know. As to which one to use ... Well that's the great thing about choices ... Pick one or more then run it if you like it keep if you don't try another. The Crafty any X86 on my site should run with any x86 CPU and uses the same Crafty flags as Dr. Hyatt does with the version he runs on the ICC.
Bryan

Hi Bryan
Yes, choices are great, but ... with all this talk about "broken", i sure don't want to spend lots of time testing a version only to find out it was broken. I bench tested your version and Danns version. Your version is slightly better on NPS and slightly worse on time to ply. So bench results are a toss-up. So i must choose a version based on compile options, which i know nothing about, or spend lots of time testing both versions. By then, version 19.16 will probably be released.
Regards
Dave

Maybe this will help. As far as the Crafty flag options my compiles use the exact same as Hyatt's (FUTILITY). Dan's compile does not use FUTILITY and he uses DETECTDRAW. The FUTILITY will speed things up but there is a chance that it may prune something out that is important. So there is a draw back to using the FUTILITY pruning. The DETECTDRAW has advantages in some positions but the draw back is it makes the eval slower. I have seen results of testing for the FUTILITY pruning posted on the CCC with 1000's of games played and it preformed better then the version without FUTILITY pruning. I have not seen any testing results with the DETECTDRAW so I can not say one way or another.
So there are pros and cons for all of the Crafty flags and nothing is broken. It is simply sometimes one works better then the other.

[Norm Pollock]

Geschrieben von:/Posted by: Norm Pollock at 14 July 2004 23:34:27:
Als Antwort auf:/In reply to: Re: Crafty 19.15 geschrieben von:/posted by: Bryan Hofmann at 14 July 2004 20:15:47:

For the non-techie users like myself, how should I read the bench tests results? I assume higher nodes per second is better. What about time to ply and other measurements? Are higher numbers better, or are lower better?

[Peter Berger]

Geschrieben von:/Posted by: Peter Berger at 14 July 2004 23:43:56:
Als Antwort auf:/In reply to: Re: Crafty 19.15 geschrieben von:/posted by: Norm Pollock at 14 July 2004 23:34:27:

For the non-techie users like myself, how should I read the bench tests results? I assume higher nodes per second is better. What about time to ply and other measurements? Are higher numbers better, or are lower better?

If you compair different versions on your own computer it is probably easiest to just look at the time needed - the less time the better.
For the other measurements higher is better.
Actually I think there is even some bug in the bench stuff, so at least on parallel hardware you should do the bench twice and go with the second number (the first being not too trustable).
As a non-techie I found a simply solution for myself. I just wait until Mike Byrne releases his SE version, that invariably proves to be fastest for me
Cheers
Peter

[Bryan Hofmann]

Geschrieben von:/Posted by: Bryan Hofmann at 15 July 2004 00:11:56:
Als Antwort auf:/In reply to: Re: Crafty 19.15 geschrieben von:/posted by: Norm Pollock at 14 July 2004 23:34:27:

For the non-techie users like myself, how should I read the bench tests results? I assume higher nodes per second is better. What about time to ply and other measurements? Are higher numbers better, or are lower better?

No easy answer here but for comparing the same version compiled differently the NPS is a sign of the speed. The Total time is also a factor in that it "x" maybe faster in time then "y" but "x" has a slower NPS. Normally you see a difference in a lower time when something in the seach or eval has been modified and a faster time is good. Also another item that will blur things is if one version was compiled with FUTILITY and another was not. The one compiled with FUTILITY will normall have a faster time due to the fact that less nodes were shearched due to prunning. The bench that is built into Crafty it is only looking at 6 positions from the BK test suite to a set depth and cares not about the correct solution so it is strictly a speed test. As for the SMP Time to PLY, all I can say here is the formua used to calucate it is (640.0 / (total_time_used / 100) and on faster system it is higher then slower. What Hyatt is acutally measuring with this I have no idea.
I don't want to start a war on which is better to look for, all I can say is in the Crafty newsgroup Hyatt has indicated that NPS is a sign of a faster crafty but only for the same version. I will leave you with a quote from a message from Dr. Hyatt in the crafty news group;
----Dr. Hyatt email quote-------
You can't pay much attention to the _time_ on the bench command. A
couple of evaluation changes were made and that is enough to make the
time for a particular position to go up or down. The NPS should not
change much however, and that is all you can really compare between
different versions, and even though you are looking at 19.9, it is
getting changed regularly...