IvanHoe compile comparison (as promised)

[Gusev]

I recompiled IvanHoe code kindly provided by Velmarin. I used all SGVAD settings (discussed earlier), with a notable exception: No profile-guided optimization (PGO), because the instrumented executable crashed under Arena when I tried to perform PGO.

Engine Score Iv Iv S-B
1: IvanHoe.x64.VE.f 101.0/200 ········································································································································································································ =1=00======0==1===0=1==00==0==============1=00==1=100=1=1=10=1========1=11===0==0=====0========
===========1=1=1=1==1====1=====0==1==0=01=====0==0======1====1=========1============0=====1=
==========0=0 9999.0
2: IvanHoe_999945_SGVAD-_x64 99.0/200 =0=11======1==0===1=0==11==1==============0=11==0=011=0=0=01=0========0=00===1==1=====1========
===========0=0=0=0==0====0=====1==0==1=10=====1==1======0====0=========0============1=====0=
==========1=1 ········································································································································································································ 9999.0

200 games played / Tournament is finished
Name of the tournament: IvanHoe_test01
Site/ Country: DMITRI-ASUS, United States
Level: Blitz 2/2
Hardware: Intel(R) Core(TM) i7 CPU Q 720 @ 1.60GHz with 5.9 GB Memory
Operating system: Windows 7 Professional Professional Service Pack 1 (Build 7601) 64 bit

The result corresponds to almost non-measurable 3.5 Elo difference, can be a statistical fluke. Designation 999945 of my compile should be disregarded, at least for the time being. I simply didn't know which IvanHoe Velmarin had modified. I found more data on his IvanHoe.x64.VE.f here: http://www.chess2u.com/t4670p75-ivanhoe

1 Houdini 1.5a XXXXXXXXXXX / 67.0 - 33.0 / 60.5 - 39.5 / 61.0 - 39.0 / 74.0 - 26.0 262.5/400
2 IvanHoe B46fc (PP compile) 33.0 - 67.0 / XXXXXXXXXXX / 49.0 - 51.0 / 58.5 - 41.5 / 65.5 - 34.5 206.0/400
3 IvanHoe 46f (KLO compile) 39.5 - 60.5 / 51.0 - 49.0 / XXXXXXXXXXX / 50.5 - 49.5 / 54.0 - 46.0 195.0/400
4 IvanHoe 47c+GH (GH compile) 39.0 - 61.0 / 41.5 - 58.5 / 49.5 - 50.5 / XXXXXXXXXXX / 60.5 - 39.5 190.5/400
5 IvanHoe .x64.VE.f (VE compile) 26.0 - 74.0 / 34.5 - 65.5 / 46.0 - 54.0 / 39.5 - 60.5 / XXXXXXXXXXX 146.0/400

Predictably enough, IvanHoe compile performances vary, too. (I don't mean to argue with George here. On the grand scale, he is absolutely right. This compile variability aspect is a nuisance and a distraction. Just something to still have in mind, though.)

[lucasart]

These comparisons are meaningless. You need to specify whic IvanHoe's are *functionally equivalent* and which are not. And to verify it, you need to setup a test where you search a set of positions at a given depth. The functionally equivalent ones will give the same number of nodes (given the same hash size and 1 search thread only, as SMP introduces non determinism).

If you want to compare "compiles", that is to say *compiler's performance* at compiling the same code, then compare them along the above explained test. Besides, I'm pretty convinced that GCC 4.8 beats all your Microsoft and even Intel compilers by a margin. So I would never use any of your compiles (nor KLO, nor Peterpan, or Capitain Hook), but compile it myself if I really wanted to use IvanHoe.

All these numbers you posted are within "statistical fluke", except perhaps the Velmarin version which is a regression as usual: only the original authors have been capable to improve IvanHoe, improvised cloners have never done anything but introduce regressions in IvanHoe as demonstrated by Stefan Pohl's very precise LS rating list.

[velmarin]

I recompiled IvanHoe code kindly provided by Velmarin.

Dmitri, you complain of closed projects.
That's why we provide these codes, so that looked, perhaps find something you can serve, not to compile, without more.
The post how ideas, nothing more.

Although I think it is useless, if you want you can look at the file "mios.h"
find some "define".
You will also find a file "evaluation.cc" I think there are some changes.
Also in "control.c" has six ways to control Time Magnament with ideas by Kinglivenson according OpenChess displayed.
But I think it is useless effort.
You just compile and recompile.

Besides giving meat to the Lucas's dog.

[Gusev]

Dmitri, you complain of closed projects.
That's why we provide these codes, so that looked, perhaps find something you can serve, not to compile, without more.

Jose,

I am deeply grateful to you for the code and the executable! Thanks!! Before making any changes to the code, it is necessary to establish a baseline. This is just engineering (as opposed to hacking in a bad sense of the word). In this case, I am reporting a useful result. Admittedly, it may look trivial. I did not mean to criticize you. (I see that someone else did.)

The first question was, if I compile your version of the IvanHoe code, will I get a comparable result? It turns out, yes, this much can be done. This is good news. This is also important to know from the Firenzina/Fire standpoint.

Dmitri

[geots]

Dmitri, you complain of closed projects.
That's why we provide these codes, so that looked, perhaps find something you can serve, not to compile, without more.

Jose,

I am deeply grateful to you for the code and the executable! Thanks!! Before making any changes to the code, it is necessary to establish a baseline. This is just engineering (as opposed to hacking in a bad sense of the word). In this case, I am reporting a useful result. Admittedly, it may look trivial. I did not mean to criticize you. (I see that someone else did.)

The first question was, if I compile your version of the IvanHoe code, will I get a comparable result? It turns out, yes, this much can be done. This is good news. This is also important to know from the Firenzina/Fire standpoint.

Dmitri

Dmitri, it is not my intention to step in and cause problems. I just want to tell you that when Lucas speaks, you would do well to listen- and get as much advice from him as you can. Besides being an excellent programmer, he is one of few who when I asked Jim Ablett this last time if he was going to compile a version of DiscoCheck, he said "no, no use going behind Lucas' compile trying to gain a speed-up- he has it covered." Not often a person gets a compliment like that. And when you look at the resume of the person who paid the compliment- I'm just sayin'..............................



Best,

[Gusev]

George,

This makes sense to me. For example, Lucas says,

only the original authors have been capable to improve IvanHoe

That seems to imply that he knows who the original authors of IvanHoe are. Naturally, I don't know that, and it's unclear just how many people do. I will heed his advice.

Dmitri

[velmarin]

Two or three gossips...

The world's best tester spent several months making a "world championship freeware"
http://www.talkchess.com/forum/viewtopi ... =&start=10
My thanks, great effort, results:

Code: Select all

Group 1 XPS-Core-i5  2012

                                1    2    3    4    5    6    7    8    9    0    1    2    
1   Critter 1.6a 64-bit         **** ½010 1½½½ ½½½½ 11½1 1½½½ ½01½ 0101 ½½½1 ½1½1 ½1½1 1½1½  27.5/44
2   Bouquet _1.6 pop64          ½101 **** ½½½½ ½½1½ ½½0½ ½0½1 01½½ ½11½ ½1½0 1½1½ ½1½½ ½1½½  25.5/44
3   Ivanhoe 46h x64             0½½½ ½½½½ **** 1½½½ ½½½0 1½½½ ½½1½ ½½½½ 1½1½ 0½½½ 1½01 11½½  24.5/44
4   Strelka 5.5 x64             ½½½½ ½½0½ 0½½½ **** ½½0½ 1½1½ ½11½ 1½½½ ½1½0 0½01 ½½½½ 1½11  24.0/44
5   RobboLito 0.21Q x64 SSE4.2  00½0 ½½1½ ½½½1 ½½1½ **** ½½½0 ½0½0 ½1½½ 1½1½ ½1½½ ½½½1 0½½½  22.5/44  488.25
6   IvanHoe 50kQ x64 SSE4.2     0½½½ ½1½0 0½½½ 0½0½ ½½½1 **** 0½½½ 10½½ ½111 ½½½1 ½1½½ ½½½½  22.5/44  484.00
7   Stockfish 2.3.1 pcnt x64    ½10½ 10½½ ½½0½ ½00½ ½1½1 1½½½ **** ½½½0 0½0½ ½110 1½10 0001  20.5/44  455.50
8   Gull II beta2 x64           1010 ½00½ ½½½½ 0½½½ ½0½½ 01½½ ½½½1 **** 0½½0 ½½10 1½½½ ½½½1  20.5/44  445.75
9   Sting SF 2_VE 64bit         ½½½0 ½0½1 0½0½ ½0½1 0½0½ ½000 1½1½ 1½½1 **** ½½1½ 1001 ½½½½  20.5/44  443.75
10  BlackMamba 1.0 x64 SSE42    ½0½0 0½0½ 1½½½ 1½10 ½0½½ ½½½0 ½001 ½½01 ½½0½ **** 1½1½ ½1½½  20.5/44  442.00
11  Fire 2.2b xTreme GH x64     ½0½0 ½0½½ 0½10 ½½½½ ½½½0 ½0½½ 0½01 0½½½ 0110 0½0½ **** 1½½1  18.0/44
12  DeepSaros ver.3.2 pcnt x64  0½0½ ½0½½ 00½½ 0½00 1½½½ ½½½½ 1110 ½½½0 ½½½½ ½0½½ 0½½0 ****  17.5/44

In the end, hats off to Critter and Bouquet- and we shall see them a bit later in the finals.


george

I hope Richard received his trophy and send us a photo.

regarding the code I sent, that was a first executable that you may want to try is called
George.FemFatale
http://www.mediafire.com/?mcv2aax6tv0d02q

It was a joke and a tribute to someone who registered forum n immortal as
FemFemale, made ​​a thorough examination as a woman, asked things, information, etc, to become the first great connoisseur of all IvanHoes.

Then he asked the name change and introduced himself as the great tester CCR *.
From birth to one engine immortalized.
Basically it was fun.
George has a great sense of humor ..


No more gossip. Too many to count.

[Gusev]

Bouquet is a known good product, CCRL-tested, that's why I was surprised to see criticism. I did not criticize. IvanHoe (9999)46h is a good product, too (presumably, the KLO compile).

[Ajedrecista]

Hello Chema:

Two or three gossips...

The world's best tester spent several months making a "world championship freeware"
http://www.talkchess.com/forum/viewtopi ... =&start=10
My thanks, great effort, results:

Code: Select all

Group 1 XPS-Core-i5  2012

                                1    2    3    4    5    6    7    8    9    0    1    2    
1   Critter 1.6a 64-bit         **** ½010 1½½½ ½½½½ 11½1 1½½½ ½01½ 0101 ½½½1 ½1½1 ½1½1 1½1½  27.5/44
2   Bouquet _1.6 pop64          ½101 **** ½½½½ ½½1½ ½½0½ ½0½1 01½½ ½11½ ½1½0 1½1½ ½1½½ ½1½½  25.5/44
3   Ivanhoe 46h x64             0½½½ ½½½½ **** 1½½½ ½½½0 1½½½ ½½1½ ½½½½ 1½1½ 0½½½ 1½01 11½½  24.5/44
4   Strelka 5.5 x64             ½½½½ ½½0½ 0½½½ **** ½½0½ 1½1½ ½11½ 1½½½ ½1½0 0½01 ½½½½ 1½11  24.0/44
5   RobboLito 0.21Q x64 SSE4.2  00½0 ½½1½ ½½½1 ½½1½ **** ½½½0 ½0½0 ½1½½ 1½1½ ½1½½ ½½½1 0½½½  22.5/44  488.25
6   IvanHoe 50kQ x64 SSE4.2     0½½½ ½1½0 0½½½ 0½0½ ½½½1 **** 0½½½ 10½½ ½111 ½½½1 ½1½½ ½½½½  22.5/44  484.00
7   Stockfish 2.3.1 pcnt x64    ½10½ 10½½ ½½0½ ½00½ ½1½1 1½½½ **** ½½½0 0½0½ ½110 1½10 0001  20.5/44  455.50
8   Gull II beta2 x64           1010 ½00½ ½½½½ 0½½½ ½0½½ 01½½ ½½½1 **** 0½½0 ½½10 1½½½ ½½½1  20.5/44  445.75
9   Sting SF 2_VE 64bit         ½½½0 ½0½1 0½0½ ½0½1 0½0½ ½000 1½1½ 1½½1 **** ½½1½ 1001 ½½½½  20.5/44  443.75
10  BlackMamba 1.0 x64 SSE42    ½0½0 0½0½ 1½½½ 1½10 ½0½½ ½½½0 ½001 ½½01 ½½0½ **** 1½1½ ½1½½  20.5/44  442.00
11  Fire 2.2b xTreme GH x64     ½0½0 ½0½½ 0½10 ½½½½ ½½½0 ½0½½ 0½01 0½½½ 0110 0½0½ **** 1½½1  18.0/44
12  DeepSaros ver.3.2 pcnt x64  0½0½ ½0½½ 00½½ 0½00 1½½½ ½½½½ 1110 ½½½0 ½½½½ ½0½½ 0½½0 ****  17.5/44

In the end, hats off to Critter and Bouquet- and we shall see them a bit later in the finals.


george

I hope Richard received his trophy and send us a photo.

regarding the code I sent, that was a first executable that you may want to try is called
George.FemFatale
http://www.mediafire.com/?mcv2aax6tv0d02q

It was a joke and a tribute to someone who registered forum n immortal as
FemFemale, made ​​a thorough examination as a woman, asked things, information, etc, to become the first great connoisseur of all IvanHoes.

Then he asked the name change and introduced himself as the great tester CCR *.
From birth to one engine immortalized.
Basically it was fun.
George has a great sense of humor ..


No more gossip. Too many to count.

Knowing that it is a Round Robin, I can provide rating performances similar to EloSTAT ratings but without PGN files:

Code: Select all

Rating_performances_for_Round_Robin_tournaments, ® 2012-2013.

Write down the full name of the Notepad (including .txt), up to 64 characters:

Group_1.txt

Write down the number of engines of the Round Robin tournament (up to 64):

12

Write down your desired mean of ratings between -5000 and 5000:

3000

Write down the clock rate of the CPU (in GHz), only for timing the elapsed time of the calculations:

3

Rating performances have been saved into List_of_rating_performances.txt file.

Approximated elapsed time:   42 ms.

Thanks for using Rating_performances_for_Round_Robin_tournaments. Press Enter to exit.

Code: Select all

Round Robin with 12 engines and     44 games per engine.
Total number of games:       264 games.
 
 Engines:     Performance:     Score:
 
Engine 01:      3080.62       62.50 %
Engine 02:      3050.62       57.95 %
Engine 03:      3035.99       55.68 %
Engine 04:      3028.74       54.55 %
Engine 05:      3007.12       51.14 %
Engine 06:      3007.12       51.14 %
Engine 07:      2978.38       46.59 %
Engine 08:      2978.38       46.59 %
Engine 09:      2978.38       46.59 %
Engine 10:      2978.38       46.59 %
Engine 11:      2941.87       40.91 %
Engine 12:      2934.41       39.77 %
 
Mean of ratings:  3000.00 Elo.

I randomly choosed 3000 as the average rating of these twelve engines.

Regards from Spain.

Ajedrecista.

[velmarin]

Bouquet is a known good product, CCRL-tested, that's why I was surprised to see criticism. I did not criticize. IvanHoe (9999)46h is a good product, too (presumably, the KLO compile).

You should understand why I get angry.

Now has a similar project, open source.
soon see the difficulties of taking performance, when things do go wrong will give back.
I chose it closed, working every day, in Bouquet, JukeBox and more projects, the Spanish crisis has left me a long time and this has become a one of my hobbies.

You can see attacks Lucas, Norman, in engine origins.
George always joins the party.
George is funny, yes.