[petsc-users] Memory for matrix assembly

[Andrej Mesaros]

Jed Brown wrote:
> On Mon, Nov 21, 2011 at 23:46, Andrej Mesaros <andrej.mesaros at bc.edu 
> <mailto:andrej.mesaros at bc.edu>> wrote:
> 
>     My code has a function which, when given a fixed matrix row index,
>     calculates one by one values of all non-zero matrix elements in this
>     row, while also returning the column index of each of these
>     elements. So, all I need to do is put that the 1st process has a
>     loop for row index going from 1 to 60k, the 2nd process has the loop
>     going from 60k+1 to 120k, etc. Inside the loops, the row index is
>     given, so it finds the non-zero elements and their column indices.
> 
> 
> That is fine, but the error circumnstance indicated that a huge number 
> of entries were being computed by a different process. It is possible 
> that memory was corrupted earlier. You can try smaller problems with 
> valgrind or -malloc_debug -malloc_dump, but if these don't work, it 
> could be difficult to track down.


Given that I indeed call MatSetValues exclusively with row indices 
within the range determined by MatGetOwnershipRange, it should be 
impossible to generate entries on the wrong process, right? In such a 
case, could the corruption you mention be somehow due to the way I call 
other PETSc functions? Or is it at all possible that too small 
preallocation is making a problem?

Also, what is the meaning of the memories in the report: "allocated", 
"used by process" and "requested"? Still don't understand, and couldn't 
find in the manual.



>         To clarify, is "n" the matrix dimension? So that's memory for
>         100 vectors (the Krylov space) plus the memory already taken by
>         PETSc when assembly is done?
> 
> 
> Yeah, roughly that used by PETSc plus those additional vectors needed by 
> SLEPc.