[Carpet] Problems with checkpoint recovery on AMD machine

[Yosef Zlochower]

The problem appears to be due to the horizon code attempting to
calculate the metric for horizons that were not found. When this
happens the radius function is zero. If I hack
BH_diagnostics::load so that
p.ghosted_gridfn(gfns::gfn__h, irho,isigma) is 1 if
ah_radius[...] is zero then the code runs. I don't know why the code
is attempting to analyze a non-existent horizon.

Yosef

Erik Schnetter wrote:
> Hi Yosef,
>
> can you produce more debugging output?  Have a look at 
> AHFinderDirect::verbose_level, and please send more context -- at 
> least all the parameters that you set.
>
> -erik
>
> On Apr 23, 2007, at 13:34:52, Yosef Zlochower wrote:
>
>> Hi,
>>
>>   The problem seems to be with the Erik tagged version. I was actually
>> using a custom
>> version. With the "Erik" version I get the error:
>> cactus_ctest:
>> /home/yosef/Cactus_New/configs/ctest/build/AHFinderDirect/patch/coords.cc:313: 
>>
>> void AHFinderDirect::local_coords::partial_xyz_wrt_r_mu_nu(fp, fp, fp,
>> fp&, fp&, fp&, fp&, fp&, fp&, fp&, fp&, fp&): Assertion
>> `jtutil::fuzzy<fp>::NE(r, 0.0)' failed.
>>
>> The recovery seems to work correctly with the standard version of
>> AHFinderDirect.
>>
>> Yosef
>>
>> Erik Schnetter wrote:
>>> On Apr 20, 2007, at 13:38:49, Yosef Zlochower wrote:
>>>
>>>> Hi,
>>>>
>>>>   I have been having problems restarting a carpet run
>>>> on a new AMD X86_64 cluster with mpich-infinipath version
>>>> of mpi.
>>>> AHFinderDirect gives the following error message on recovery:
>>>> WARNING level -1 in thorn AHFinderDirect processor 2 host n016.cluster
>>>>   (line 78 of
>>>> /home/yosef/Cactus_New/configs/carpetdt/build/AHFinderDirect/jtutil/error_exit.cc): 
>>>>
>>>>
>>>>   -> ***** row_sparse_Jacobian__UMFPACK::solve_linear_system():
>>>>         error return status=1 from umfpack_numeric() routine
>>>>
>>>> I tried compiling AHFinderDirect with Lapack and got a similar
>>>> message.
>>>>
>>>>  I used both the stable and development version of
>>>> carpet as well as the pathscale and gnu (gcc4, gfortran4, g++4)
>>>> compilers. The operating system is CentOs 4.4.
>>>>
>>>> Any ideas?
>>>
>>> Hi Yosef.
>>>
>>> What version of the horizon finder are you using?  There are two
>>> branches in CVS; "HEAD" and "Erik".  You can also increase the horizon
>>> finder verbosity to find out more about this problem.
>>>
>>> -erik
>>>
>>> --Erik Schnetter <schnetter at cct.lsu.edu>
>>>
>>> My email is as private as my paper mail.  I therefore support 
>>> encrypting
>>> and signing email messages.  Get my PGP key from www.keyserver.net.
>>>
>>>
>>>
>>> ------------------------------------------------------------------------ 
>>>
>>>
>>> _______________________________________________
>>> developers mailing list
>>> developers at lists.carpetcode.org
>>> http://lists.carpetcode.org/listinfo/developers
>>>
>>
>> _______________________________________________
>> developers mailing list
>> developers at lists.carpetcode.org
>> http://lists.carpetcode.org/listinfo/developers
>>
>
>
> --Erik Schnetter <schnetter at cct.lsu.edu>
>
> My email is as private as my paper mail.  I therefore support encrypting
> and signing email messages.  Get my PGP key from www.keyserver.net.
>
>
>
> ------------------------------------------------------------------------
>
> _______________________________________________
> developers mailing list
> developers at lists.carpetcode.org
> http://lists.carpetcode.org/listinfo/developers
>