[petsc-users] PaStix is slower in pestc-dev

[Xiangdong Liang]

Is there an easy way to change these options?  From MATSOLVERPASTIX,
it seems that no command line options to control this.
http://www.mcs.anl.gov/petsc/petsc-current/docs/manualpages/Mat/MATSOLVERPASTIX.html

Do I have to modify PaStiX.py and recompile? If yes, would you like to
give me some hints on where to modify? Thanks.

Xiangdong

On Sun, Dec 11, 2011 at 12:49 PM, Barry Smith <bsmith at mcs.anl.gov> wrote:
>
>  Looks like Pastix is running with different options hence different performance.
>
>   Barry
>
> On Dec 11, 2011, at 10:08 AM, Xiangdong Liang wrote:
>
>> On Sun, Dec 11, 2011 at 9:04 AM, Matthew Knepley <knepley at gmail.com> wrote:
>>> On Sun, Dec 11, 2011 at 12:05 AM, Xiangdong Liang <xdliang at gmail.com> wrote:
>>>>
>>>> Hello PETSc team,
>>>>
>>>> I was using PaStiX within petsc 3.1-p8. Today, I am trying PaStiX
>>>> within petsc-dev. However, For the same code, solving the same linear
>>>> system takes longer time (120s vs 90s) in petsc-dev. Both are compiled
>>>> with debugging mode off. Is it possible that the newer PaStiX is
>>>> slower than old version? or due to some options in compiling?
>>>
>>>
>>> It is likely the options are not exactly the same, meaning the ordering is
>>> different,
>>> etc. Did you check everything wiht -ksp_view and -ksp_monitor?
>>
>> I  use ksp_view, ksp_monitor options and same -pc_factor_zeropivot
>> 1e-12, the dev version is still slower. One difference I see from
>> pastix_verbose is the option of Pastix:
>>
>> In pets-3.1: I have
>>
>>      DISTRIBUTED         :                  Not defined
>>       FLUIDBOX            :                  Not defined
>>       METIS               :                  Not defined
>>
>>
>> While in petsc-dev, I have
>>
>>      DISTRIBUTED         :                   Defined
>>        METIS               :                   Not defined
>>        WITH_SCOTCH         :                   Defined
>>
>>
>> However, I do not understand these options. Could that be the reason?
>>
>> Xiangdong
>>
>>>
>>>    Matt
>>>
>>>>
>>>> Best,
>>>> Xiangdong
>>>>
>>>> PS. I cannot use PaStiX in 3.2 because in the runtime, PaStiX crashed
>>>> due to missing of the option -DWITHSCOTCH during compiling. It was
>>>> reported before by one user and fixed in petsc-dev, but not in
>>>> petsc-3.2-p5.
>>>>
>>>> http://lists.mcs.anl.gov/pipermail/petsc-users/2011-March/008356.html
>>>
>>>
>>>
>>>
>>> --
>>> What most experimenters take for granted before they begin their experiments
>>> is infinitely more interesting than any results to which their experiments
>>> lead.
>>> -- Norbert Wiener
>