[petsc-dev] New implementation of PtAP based on all-at-once algorithm

[Fande Kong]

On Mon, Apr 15, 2019 at 1:41 PM Mark Adams <mfadams at lbl.gov> wrote:

>
>
>>> I wonder if the their symbolic setup is getting called every time. You
>>> do 50 solves it looks like and that should be enough to amortize a one time
>>> setup cost.
>>>
>>
>> Hypre does not have concept called symbolic. They do everything from
>> scratch, and won't reuse any data.
>>
>
> Really, Hypre does not cache the maps and non-zero structure, etc, that is
> generated in RAP?
>
> I suspect that that is contributing to Hypres poor performance, but it is
> not the whole story as you are only doing 5 solves.
>

Do not know.  The combined approach does not mean slow.  I have no
conclusion here.   Symbolic PtAP won't take much more time than the numeric
PtAP.   Might be just a tiny bug somewhere that triggered by my
applications, and that does not happen to other people?!

Thanks,

Fande,
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20190415/0afeb376/attachment.html>