Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Intel compiler for Linux

Author: Matt Taylor

Date: 07:00:14 03/01/03

Go up one level in this thread


On March 01, 2003 at 09:57:28, Robert Hyatt wrote:

>On March 01, 2003 at 03:26:19, Alessandro Damiani wrote:
>
>>On March 01, 2003 at 00:24:41, Robert Hyatt wrote:
>>
>>>On February 28, 2003 at 17:39:55, Dann Corbit wrote:
>>>
>>>>On February 28, 2003 at 17:34:23, Matthew Hull wrote:
>>>>
>>>>>On February 28, 2003 at 14:37:53, Robert Hyatt wrote:
>>>>>
>>>>>>On February 28, 2003 at 12:21:26, Russell Reagan wrote:
>>>>>>
>>>>>>>On February 28, 2003 at 12:15:22, Robert Hyatt wrote:
>>>>>>>
>>>>>>>>I've been using it since version 6.0, and for Crafty it
>>>>>>>>produces the
>>>>>>>>fastest executable of any compiler I have tried.
>>>>>>>
>>>>>>>Fastest linux executable, or fastest executable period, including the MS
>>>>>>>compilers?
>>>>>>
>>>>>>
>>>>>>fastest linux is all I can say with any degree of confidence.  I don't have any
>>>>>>easy way
>>>>>>to compare to MSVC and windows since I don't have any "equal" machines here yet.
>>>>>>
>>>>>>We are expecting a group of dual xeons that will be half linux half windows XP
>>>>>>within
>>>>>>a week or two so I might get to compare there, but I'd personally suspect that
>>>>>>MSVC
>>>>>>is going to be better (faster) based on past experience.  Intel probably has
>>>>>>closed the gap,
>>>>>>but not completely I suspect.
>>>>>
>>>>>
>>>>>I thought Dan Corbit had said his binaries are done with Intel for max
>>>>>performance on Windows, not MSVC.
>>>>
>>>>Usually, Intel makes better binaries than MS VC++ 6.0 (with all relevant
>>>>patches)
>>>>
>>>>However, the MS VC++ .NET compiler frequently beats the Intel compiler.
>>>>
>>>>In addition, the latest MINGW GCC will sometimes pull a surprise with the right
>>>>combination of compiler flags.
>>>
>>>I try the latest gcc from time to time.  I tried the most recent (non-beta)
>>>version today and discovered the profile-based optimization is DOA.  Compiles
>>>fine, produces the profile files fine, but re-compiling causes the compiler to
>>>go into a royal snit complaing about corrupted profile data with impossible
>>>branch addresses and the like.  And without profiling Intel doesn't just beat
>>>it, it destroys it.  Profiling closes the gap, but it doesn't work in the
>>>current gcc 3 compiler (for Crafty it doesn't work, I have not tried it on
>>>other programs).
>>
>>Dann is talking about the MINGW variant of GCC (www.mingw.org). Did you mean
>>this one or the normal GCC? I guess the MINGW is faster than the normal GCC on
>>Windows maschines.
>>
>>Alessandro
>
>
>No.  I was talking about gcc 3.x as released for unix platforms....

I don't think there would be a major difference, anyway. The MinGW version is
going to spit out a PE binary, but it's still GCC generating all the important
code.

-Matt



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.