On 12/29/23 11:39, Simon Wiesheier wrote:
For efficiency reasons, I decided to loop manually over the rows I am
interested in.
In a serial program, that worked smoothly. But if I get the gist of your
message, vmult and extracting the relevant elements is the only way to do this
in parallel, right?
If you only need certain rows, you still have to do the same basic steps a
parallel vmult has to do, except that you can restrict yourself to a subset of
the rows. What I meant to say is that it isn't trivial to determine which
elements you need to communicate. You might have to read through the source
code of PETSc's or Trilinos's vmult functions to understand the sequence of
events. I would not be surprised if a parallel vmult runs into several hundred
of lines of code.
It is certainly *easier* to just do the full vmult with a vector, and then
pick out the entries you actually need. Whether that's a performance problem
is something you should benchmark -- it is not worth optimizing operations for
which you do not already know that they are a bottleneck.
Best
W.
--
------------------------------------------------------------------------
Wolfgang Bangerth email: bange...@colostate.edu
www: http://www.math.colostate.edu/~bangerth/
--
The deal.II project is located at http://www.dealii.org/
For mailing list/forum options, see
https://groups.google.com/d/forum/dealii?hl=en
---
You received this message because you are subscribed to the Google Groups "deal.II User Group" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to dealii+unsubscr...@googlegroups.com.
To view this discussion on the web visit
https://groups.google.com/d/msgid/dealii/666e38ae-57d7-453b-8dbe-28848f05de0d%40colostate.edu.
