Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: g++ 3.4 and -fprofile-generate

Author: Andrew Williams

Date: 13:10:49 02/16/05

Go up one level in this thread


On February 16, 2005 at 16:05:37, Frank Phillips wrote:

>On February 16, 2005 at 15:45:49, Andrew Williams wrote:
>
>>On February 16, 2005 at 15:11:13, Frank Phillips wrote:
>>
>>>On February 16, 2005 at 14:27:49, Andrew Williams wrote:
>>>
>>>>On February 16, 2005 at 13:09:14, Frank Phillips wrote:
>>>>
>>>>>Has anybody got any experience with g++ 3.4 for amd64 (x86_64) - for Linux?
>>>>>
>>>>>I have been using the profile generated optimisation option, but the code it
>>>>>produces is no faster then with simple -O3.
>>>>>
>>>>>I simply compile with -fprofile-generate
>>>>>then run,
>>>>
>>>>What do you run?
>>>
>>>My chess program (searcher).  I have a mode for calcuating to fixed depth, which
>>>I often use to check that the total have not changed after making changes to the
>>>program that should not affect the tree.  Not an extensive test suite, but I
>>>expected to see some difference.
>>>
>>>Frank
>>>
>>
>>Is this several positions? How long do you run it for? I'm using gcc 3.3.5 at
>>the moment, and I'd expect at least a 10% speedup as a result of using
>>profiling. I might use 10 positions and run them for about 2 minutes each.
>>
>>Andrew
>
>Six positions to fixed depth (10 ply in this case.).  The kilonodes per second
>are exactly the same.
>
>There was a boost with gcc 3.3.?  using -fprofile-arcs / -fbranch-probabilities
>in Mandrake 10.0 - about 10% IIRC.  I am using Mandrake 10.1 at the moment,
>which has a later gcc.
>
>
Ah. I see. I haven't tried 3.4.x yet, and I was planning at some point to
install it and test it out. I'm glad I haven't tried it yet!

Andrew



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.