[petsc-users] MatPermute for MPIBAIJ matrix Error

[Rongliang Chen]

Hi Jed,

Many thanks for your suggestions.

Best regards,
Rongliang


On 06/20/2016 11:33 AM, Jed Brown wrote:
> Rongliang Chen <rl.chen at siat.ac.cn> writes:
>
>> Many thanks for your suggestions. We are using a parallel SOR as the
>> local solver and in the SOR, we want to ignore the nonzeros not on this
>> thread, but we hope to keep the nonzeros as much as possible, so reduce
>> the bandwidth is a possible way to keep the nonzeros.
> Clearly you're asking for a partitioner.  But a low-bandwidth ordering
> makes a really crappy partitioner (it's asymptotically like a 1D
> partition).  If you want a partitioner, I recommend using a partitioner.
> And, as Barry said, do the partitioning on your mesh instead of
> redistributing after assembling a matrix.