[petsc-users] MatMPIAIJSetPreallocation question

[Barry Smith]

> On Feb 15, 2015, at 9:17 PM, Gideon Simpson <gideon.simpson at gmail.com> wrote:
> 
> Got it, one follow up question:
> 
> When calling MatMPIAIJSetPreallocation, is there a reason why the number of nonzero entries in the off-diagonal sub matrix cannot be zero?

It could be zero; that means there is not coupling between processes however so essentially each process has its own independent problem; thus normally it is most definitely not zero.

  Barry

> 
> -gideon
> 
>> On Feb 15, 2015, at 9:31 PM, Barry Smith <bsmith at mcs.anl.gov> wrote:
>> 
>> 
>>> On Feb 15, 2015, at 8:22 PM, Gideon Simpson <gideon.simpson at gmail.com> wrote:
>>> 
>>> I’m trying to get a handle on the different ways of constructing matrices.  Currently, I have:
>>> 
>>> MatCreate(...)
>>> MatSetSizes(...)
>>> MatSetFromOptions(...)
>>> MatSetUp(…)
>>> 
>>> but I gather from reading the manual that, by not preallocating, I’m losing out in performance.  If I assume that my matrix will either by SeqAIJ or MPIAIJ, depending on the number of processors available, how would I go about doing this.  I see some of the example codes with:
>>> 
>>> MatSeqAIJSetPreallocation(…)
>>> MatMPIAIJSetPreallocation(…)
>>> 
>>> as successive commands.  Should I interpret this as saying that PETSc will just ignore the one that is not the active one in the current instance? 
>> 
>>  Exactly*.
>> 
>>  You can also use MatXAIJSetPreallocation() which sets the possible preallocation forms for all the matrix types without requiring you to set one each individuallly.
>> 
>>  Barry
>> 
>> * We support this in many places because we hate requiring users to put a bunch of if (type == xxx) kind of code in their applications. For example KSPGMRESSetRestart() is ignored if the KSP type is not GMRES.
>> 
>> 
>>> 
>>> -gideon
>>> 
>> 
>