Using MatGetRow() to extract matrix values.

[Matthew Knepley]

On Fri, Jun 13, 2008 at 10:23 AM, Stephen R Ball
<Stephen.R.Ball at awe.co.uk> wrote:
> Hi
>
> I am now using MatGetRow() to extract the matrix values and column
> numbers for each row. However, I have noticed that the ordering of the
> matrix values is sometimes different to that which was originally
> inserted. For example below are the matrix values for one row.

I assume by "column no" you mean the order in the array returned, not
the global column index. We sort the rows by global column index, and thus
have no memory of the original insertion order.

  Matt

> Original row:
>
> Column no.            Matrix Value
> 0                     185693206.9234681
> 1                     4.3135145491066673E-002
> 2                    -1.5858782157409282E-010
> 3                     8.5489320680905649E-002
> 4                    -7.5011130237855266E-011
> 5                    -3.7505565118927637E-011
> 6                    -4.3135145489119342E-002
> 7                    -8.6270290978446909E-002
>
>
> After matrix assembly, extracted via MatGetRow():-
>
> Column no.            Matrix Value
> 0                    -7.5011130237855266E-011
> 1                    -3.7505565118927637E-011
> 2                     4.3135145491066673E-002
> 3                    -8.6270290978446909E-002
> 4                     185693206.9234681
> 5                    -4.3135145489119342E-002
> 6                    -1.5858782157409282E-010
> 7                     8.5489320680905649E-002
>
>
> Is this what I should expect to happen?
>
> As I mentioned in an earlier email, my purpose for extracting the matrix
> values is so that I can output the locally owned values to file in the
> format I want rather than via use of the MatView() routine. I would
> prefer to output the values in their original ordering.
>
> The reason I extract the values after matrix assembly rather than just
> output the original CSR matrix array is because, in parallel, some
> matrix entries are partial contributions to the true values and it is
> these true values that I wish to output. I get these true values after
> inserting them via MatSetValues() using the ADD_VALUES flag and then
> using MatGetRow().
>
> Can I obtain the matrix values in their original row and column ordering
> after matrix assembly? Doesn't MatView() output the matrix with its
> original ordering?
>
> Regards
>
> Stephen
>
>
>
>
>
>
>
>
>
> -----Original Message-----
> From: owner-petsc-users at mcs.anl.gov
> [mailto:owner-petsc-users at mcs.anl.gov] On Behalf Of Barry Smith
> Sent: 04 June 2008 15:30
> To: petsc-users at mcs.anl.gov
> Subject: EXTERNAL: Re: EXTERNAL: Re: MatGetArray() not supporting Mat
> type mpiaij.
>
>
>    Stephan,
>
>    For all built in PETSc matrix types, the rows are stored
> contiguously so
> you can use MatGetOwnershipRange() to get the global indices for the
> local
> rows and thus call MatGetRow().
>
>    Barry
>
>
> On Jun 4, 2008, at 6:33 AM, Stephen R Ball wrote:
>
>> Ok, I am looking into using MatGetRow(). However this requires the
>> global row number for input. I was looking to use
>> MatGetOwnershipRange()
>> to obtain the range of global row numbers owned by each processor but
>> the documentation states that this routine assumes that the matrix is
>> laid out with the first n1 rows on the first processor, the next n2
>> rows
>> on the second, etc and that for certain parallel layouts this range
>> may
>> not be well defined.
>>
>> This is the case for me. Do you have a routine where I can specify a
>> global row number and it will tell me the rank of the processor that
>> owns it? This is to ensure that MatGetRow() only gets called by the
>> owner processor for each global row number.
>>
>> Regards
>>
>> Stephen
>>
>>
>> -----Original Message-----
>> From: owner-petsc-users at mcs.anl.gov
>> [mailto:owner-petsc-users at mcs.anl.gov] On Behalf Of Matthew Knepley
>> Sent: 03 June 2008 15:29
>> To: petsc-users at mcs.anl.gov
>> Subject: EXTERNAL: Re: MatGetArray() not supporting Mat type mpiaij.
>>
>> On Tue, Jun 3, 2008 at 8:11 AM, Stephen R Ball
>> <Stephen.R.Ball at awe.co.uk> wrote:
>>> Hi
>>>
>>> I have been trying to extract an array containing the local matrix
>>> values using MatGetArray() via the Fortran interface but get the
>>> error
>>> message that Mat type mpiaij is not supported with this routine.
>>> All I
>>> want to do is to extract the local matrix values so that I can output
>>> them to file in the format I want rather than via use of the
>>> MatView()
>>> routine. Can you suggest a way of how I can go about extracting the
>>> local matrix values?
>>
>> This is no "local matrix". The Mat interface is supposed to be data
>> structure
>> neutral so we can optimize for different architectures. If you want
>> the
>> values
>> directly, I would use MatGetRow() for each row.
>>
>>  Matt
>>
>>> Thanks
>>>
>>> Stephen
>>>
>>>
>>
>>
>>
>> --
>> 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
>>
>>
>
>



-- 
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