[petsc-users] Custom vector owenrship ranges

[Matthew Knepley]

On Wed, Mar 26, 2014 at 8:28 AM, Mohammad Bahaa <m.bahaa.eldein at gmail.com>wrote:

> Actually I tried your suggestion, and it works fine, but it's slightly
> different from what I need, since each process should have access to other
> processes' values, since there's some sort of interaction, so I need
> process 0 (for instance) to be capable of accessing (just reading) value
> out of its ownership range
>

Random access defeats the purpose of parallel computing. If the access is
structured, like a halo region, then
you can use a VecScatter to map between Vecs with and without a halo. See
the manual chapter on the DMDA
object for a discussion of this in the case of structured meshes.

   Thanks,

      Matt


> On Wed, Mar 19, 2014 at 1:16 AM, Matthew Knepley <knepley at gmail.com>wrote:
>
>> On Tue, Mar 18, 2014 at 11:00 AM, Mohammad Bahaa <
>> m.bahaa.eldein at gmail.com> wrote:
>>
>>> I used
>>> call VecCreateMPIWithArray(PETSC_COMM_WORLD,1,nc,ncall,myx,xall,ierr)
>>>
>>> however, when I use process 0 to write a file containing the combined
>>> values (the xall vector), the values seem not to be updated by some
>>> processes, eventhough I use PetscBarrier, in other words, values locally
>>> owned by processes 0 and 2 are ok, but those owned by process 1 & 3 aren't !
>>>
>>
>> For collective writes, use VecView() or -vec_view
>>
>>    Matt
>>
>>
>>> On Tue, Mar 18, 2014 at 3:43 PM, Mohammad Bahaa <
>>> m.bahaa.eldein at gmail.com> wrote:
>>>
>>>> the second approach of the MPI vector did it for me, thanks
>>>>
>>>>
>>>> On Tue, Mar 18, 2014 at 3:20 PM, Mohammad Bahaa <
>>>> m.bahaa.eldein at gmail.com> wrote:
>>>>
>>>>> Forgive me as my expression "sum up" was misguiding or misplaced, I
>>>>> didn't mean to literally sum the values in the vectors, I meant I want to
>>>>> put all values from each local vector into one global vector that can be
>>>>> accessed by all processes, "COMM_WORLD" communicator for instance
>>>>>
>>>>>
>>>>> On Tue, Mar 18, 2014 at 3:09 PM, Matthew Knepley <knepley at gmail.com>wrote:
>>>>>
>>>>>> On Tue, Mar 18, 2014 at 7:53 AM, Mohammad Bahaa <
>>>>>> m.bahaa.eldein at gmail.com> wrote:
>>>>>>
>>>>>>> I'm using "PETSC_COMM_SELF" communicator for running n serial
>>>>>>> independent processes, I need to sum up a certain vector from the n
>>>>>>> processes in one vector, however, vectors involved in each process vary in
>>>>>>> size, and I couldn't find any function to define custom ownership ranges,
>>>>>>> so assuming I have a 4 processes run with each computing an "x" vector as
>>>>>>> follows:
>>>>>>>
>>>>>>> 1. process (1) with x of length 51
>>>>>>> 2. process (2) with x of length 49
>>>>>>> 3. process (3) with x of length 52
>>>>>>> 4. process (4) with x of length 48
>>>>>>>
>>>>>>
>>>>>> Let your local length be n, so that on proc 3 n== 52. Then
>>>>>>
>>>>>>   VecCreate(comm, &v);
>>>>>>   VecSetSizes(v, n, PETSC_DETERMINE);
>>>>>>   VecSetFromOptions(v);
>>>>>>   <fill up v>
>>>>>>   VecSum(v, &sum);
>>>>>>
>>>>>> You could also make a parallel Vec from your Seq vecs:
>>>>>>
>>>>>>   VecGetArray(lv, &array);
>>>>>>   VecCreateMPIWithArray(comm, 1, n, PETSC_DETERMINE, array, &v);
>>>>>>
>>>>>>   Thanks,
>>>>>>
>>>>>>      Matt
>>>>>>
>>>>>>
>>>>>>> The processes sum up to 100 elements, when I define a vector "x_all"
>>>>>>> of size "100" with "PETSC_COMM_WORLD" communicator, the ownership
>>>>>>> ranges are equal, which isn't the case, how to customize them ?
>>>>>>>
>>>>>>> --
>>>>>>> Mohamamd Bahaa ElDin
>>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>> --
>>>>>> 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
>>>>>>
>>>>>
>>>>>
>>>>>
>>>>> --
>>>>> Mohamamd Bahaa ElDin
>>>>>
>>>>
>>>>
>>>>
>>>> --
>>>> Mohamamd Bahaa ElDin
>>>>
>>>
>>>
>>>
>>> --
>>> Mohamamd Bahaa ElDin
>>>
>>
>>
>>
>> --
>> 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
>>
>
>
>
> --
> Mohamamd Bahaa ElDin
>



-- 
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
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20140326/4081d293/attachment.html>