Hi Folks, I am still slogging away at debugging my AIJMKL matrix class using the PETSc test suite. In some cases, such as mat/examples/tests/ex160.c, I am seeing a failure of the mkl_sparse_optimize() call inside MatAssembly. Digging into this, it appears that the problem is that the A array (the nonzero values) for the matrix is NULL -- understandably, mkl_sparse_optimize() does not like this. I've verified that this isn't due to some bug my matrix class has introduced, as running the same example with just plain SeqAIJ matrices and stepping through it with the debugger, I still see the same NULL value for A.
I am assuming that this is intended behavior, and what one might expect if one, say, calls MatAssemblyBegin/End without ever doing an MatSetValues? Looks like I need to simply check to see if A is NULL and not do the mkl_sparse_optimize() step if so. I already had a check in there for a matrix with no columns, as I believe I encountered that case (but with A pointing to allocated memory). (And maybe I should check for a matrix with no rows?) Are these the things I should be testing for, or is there something nicer that I should be checking to see if the matrix to be "assembled" doesn't actually consist of any matrix? --Richard
