[petsc-users] Understanding preallocation for MPI

[Matthew Knepley]

On Fri, Jul 7, 2017 at 2:49 AM, Dave May <dave.mayhem23 at gmail.com> wrote:

> On Fri, 7 Jul 2017 at 11:31, Florian Lindner <mailinglists at xgm.de> wrote:
>
>> Hello,
>>
>> I'm having some struggle understanding the preallocation for MPIAIJ
>> matrices, especially when a value is in off-diagonal
>> vs. diagonal block.
>>
>> The small example program is at https://pastebin.com/67dXnGm3
>>
>> In general it should be parallel, but right now I just run it in serial.
>
>
> When you run this code in serial, the mat type will be MATSEQAIJ. Hence,
> the call to MatMPIAIJSetPreallocation() will have no effect because the mat
> type does not match MPIAIJ. As a result, your code doesn't perform any
> preallocation for SEQAIJ matrices.
>
> In addition to calling MatMPIAIJSetPreallocation(), add a call to
> MatSEQAIJSetPreallocation.
>

To make it easier we now provide


http://www.mcs.anl.gov/petsc/petsc-current/docs/manualpages/Mat/MatXAIJSetPreallocation.html

    Matt


> Thanks,
>   Dave
>
>
>>
>> According to my understanding of
>>
>> http://www.mcs.anl.gov/petsc/petsc-3.7/docs/manualpages/
>> Mat/MatMPIAIJSetPreallocation.html
>>
>> a entry is in the diagonal submatrix, if its row is in the OwnershipRange
>> and its column is in OwnershipRangeColumn.
>> That also means that in a serial run, there is only a diagonal submatrix.
>>
>> However, having MAT_NEW_NONZERO_ALLOCATION_ERR set, I get an error when
>>
>> Inserting 6 elements in row 2, though I have exactly
>>
>> 2 o_nnz = 0, d_nnz = 6 (means 6 elements allocated in the diagonal
>> submatrix of row 2)
>>
>> Error is:
>>
>> [0]PETSC ERROR: Argument out of range
>> [0]PETSC ERROR: New nonzero at (2,5) caused a malloc
>> Use MatSetOption(A, MAT_NEW_NONZERO_ALLOCATION_ERR, PETSC_FALSE) to turn
>> off this check
>>
>>
>> What is wrong with my understanding?
>>
>> Thanks,
>> Florian
>>
>


-- 
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

http://www.caam.rice.edu/~mk51/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20170707/d065fb0b/attachment.html>