On 03/05/2018 02:54 PM, Feimi Yu wrote:


I changed my strategy to use set(r, c, v) function to set the values so that I can use the const iterators. also called compress after every add:
for(autor =Abs_A_matrix->block(0, 0).local_range().first;
r <Abs_A_matrix->block(0, 0).local_range().second; ++r)
{
for(autoitr =Abs_A_matrix->block(0, 0).begin(r);
itr !=Abs_A_matrix->block(0, 0).end(r);
++itr)
{
Abs_A_matrix->block(0, 0).set(itr->row(), itr->column(), 
std::abs(itr->value()));
Abs_A_matrix->block(0, 0).compress(VectorOperation::add);
}
}

This time it says:
*** Error in ' ' :free(): invalid size: 0x0000556443d63e50 ***

This seems like a out of bounds access. This happens even when I run with 1 process.

This is the right approach -- you do need to restrict work to the locally owned range of rows, but if you do that on every processor, then you cover all rows.

But you can't call 'compress()' after every set operation. That's because compress() is a *collective* operation, i.e., it has to be called on all processors at the same time. Since different processors may have different numbers of entries, what you are doing here leads to different numbers of calls on different processors.

In addition, it is *very* inefficient to call the function so many times. Just move the call to after the loop. (And change the argument to VectorOperation::set -- you are *setting* elements, after all, not adding to them.)

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.
For more options, visit https://groups.google.com/d/optout.

Reply via email to