[petsc-users] Matrix preallocation

[Mark Adams]

On Sat, Feb 5, 2022 at 10:36 AM Matthew Knepley <knepley at gmail.com> wrote:

> On Fri, Feb 4, 2022 at 11:47 PM Samuel Estes <samuelestes91 at gmail.com>
> wrote:
>
>> Hi,
>>
>> I have a very basic question about matrix preallocation. I am trying to
>> use the MatCreate(), MatSetFromOptions(), MatXXXXSetPreallocation()
>> paradigm. I thought that I should use the MatXAIJSetPreallocation() routine
>> since the code may be run with a SeqAIJ or MPIAIJ matrix but I do not
>> understand all of the inputs required for the
>> MatXAIJSetPreallocation routine. In particular, the dnnzu and
>> onnzu variables don't quite make sense to me. Can these be NULL? I was
>> basically just hoping for a routine that would preallocate for either a
>> sequential or parallel matrix depending on what was given at runtime. This
>> routine seems to be what I want but I don't understand it very well and the
>> documentation hasn't helped me to figure it out.
>>
>
> The example for this is here
>
>
> https://petsc.org/main/docs/manualpages/Mat/MatMPIAIJSetPreallocation.html#MatMPIAIJSetPreallocation
>
> Maybe we should copy it to the XAIJ page as well. Does this help explain
> the arguments?
>
>
>> A related followup question: Is it good practice to use this function or
>> should I just use the other routines like MatSeqAIJSetPreallocation() and
>> MatMPIAIJSetPreallocation()?
>>
>> And finally my last question: if I were to use the
>> MatSeqAIJSetPreallocation()/MatMPIAIJSetPreallocation() routines for
>> preallocating memory, is it common to just call MatGetType() then call the
>> appropriate routine depending on whether or not the matrix is parallel or
>> not? I ask because when I have tested these routines out, it seems
>> that MatSeqAIJSetPreallocation() works even for parallel matrices which is
>> a bit confusing. I'm assuming that it just sets the diagonal part of the
>> matrix?
>>
>
> No, it definitely will not preallocate in parallel, so something else is
> happening.
>

With only MatSeqAIJSetPreallocation in parallel it would not do any
preallocation, in which case it would fall back to dynamic allocation. I'm
guessing that is what is happening and it will be very slow.


>
>   Thanks,
>
>      Matt
>
>
>> I hope that my questions were clear. Let me know if they need
>> clarification and thanks in advance for the help!
>>
>> Sam
>>
>
>
> --
> 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
>
> https://www.cse.buffalo.edu/~knepley/
> <http://www.cse.buffalo.edu/~knepley/>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20220205/0c49d755/attachment.html>