Re: [Bug-gnubg] Re: Bug? Pruning On/Pruning Off gives significantly different results at 4ply

[Christian Anthon]

Another thing - what do you get if you set up the resulting positions
after the moves and then do a 4-ply eval.

Christian.

On Tue, Sep 1, 2009 at 12:35 AM, Christian
Anthon<address@hidden> wrote:
> On my single core athlon the results are perfectly normal. I assume
> that we are seeing another compiler bug. The one we found recently
> couldn't be reproduced on this machine either.
>
> Linux anthon-desktop 2.6.29.6-217.2.16.fc11.i686.PAE #1 SMP Mon Aug 24
> 17:16:21 EDT 2009 i686 athlon i386 GNU/Linux
>
> gcc (GCC) 4.4.1 20090725 (Red Hat 4.4.1-2)
>
> Multiple threads supported.
> SSE/SSE2 supported.
>
>
>    1. Cubeful 4-ply    10/6 5/2                     Eq.:  +0.558
>       0.810 0.002 0.000 - 0.190 0.000 0.000
>        4-ply cubeful prune
>    2. Cubeful 4-ply    10/3                         Eq.:  +0.555 ( -0.003)
>       0.809 0.002 0.000 - 0.191 0.000 0.000
>        4-ply cubeful prune
>    3. Cubeful 4-ply    10/6 4/1                     Eq.:  +0.551 ( -0.007)
>       0.807 0.002 0.000 - 0.193 0.000 0.000
>        4-ply cubeful prune
>    4. Cubeful 4-ply    9/6 9/5                      Eq.:  +0.542 ( -0.017)
>       0.801 0.009 0.000 - 0.199 0.002 0.000
>        4-ply cubeful prune
>
>
>    1. Cubeful 4-ply    10/6 5/2                     Eq.:  +0.560
>       0.810 0.002 0.000 - 0.190 0.000 0.000
>        4-ply cubeful
>    2. Cubeful 4-ply    10/3                         Eq.:  +0.558 ( -0.002)
>       0.810 0.002 0.000 - 0.190 0.000 0.000
>        4-ply cubeful
>    3. Cubeful 4-ply    10/6 4/1                     Eq.:  +0.552 ( -0.008)
>       0.807 0.002 0.000 - 0.193 0.000 0.000
>        4-ply cubeful
>    4. Cubeful 4-ply    9/6 9/5                      Eq.:  +0.540 ( -0.019)
>       0.800 0.009 0.000 - 0.200 0.002 0.000
>        4-ply cubeful
>
>
> 2009/8/31 Michael Petch <address@hidden>:
>>
>> On 31/08/09 3:33 PM, "Øystein Johansen" <address@hidden> wrote:
>>
>> * /Should/ means that it should give the same answer. I have tested a lot of
>> inputs and got the desired output, however..... there can be surprises.
>>
>> Unless there is a bug or there is a precision change there is a difference
>> between the outputs pre-patch and post patch versions. I am not the only one
>> to have seen it. Now what I haven’t ruled out is that an unintended side
>> effect of something else (optimizations, sse2 compiler bug etc).
>>
>> I am running some eval tests today., When finished I will take out the
>> latest CVS, and make 2 builds. No Multithreading with SSE only, and No
>> Multithreading with SSE2 only (both builds with no compiler optimizations so
>> no –O2 or -O3), I have a few thousand matches and I’ll analyze a random
>> subset of the files on both versions (At 0 ply) and see if I can find any
>> differences.
>>
>> _______________________________________________
>> Bug-gnubg mailing list
>> address@hidden
>> http://lists.gnu.org/mailman/listinfo/bug-gnubg
>>
>>
>