Fix a bunch of small errors. #1301
Conversation
|
/style |
| mfem::out << "Energy using subspace solver from: " << base_energy << ", to: " << subspace_energy << " / " | ||
| << energy_change << ". Min eig: " << leftvals[0] << std::endl; | ||
| << energy_change << ". Min eig: " << leftval << std::endl; |
There was a problem hiding this comment.
It looks like the minimum eigenvalue estimate will be reported as 1.0 when no min eigenvalues estimates have been requested. I'm fine with this for now, since it's really an inconsistent request from the user, but it is misleading. An option to consider is to check if leftvals.size() == 0 once at the beginning of the step and echo a warning saying that the min eigenvalue will be reported as 1.0 in this case.
| DenseMat(const DenseMat& a) | ||
| { | ||
| MatDuplicate(a.A, MAT_COPY_VALUES, &A); | ||
| MatCopy(a.A, A, SAME_NONZERO_PATTERN); | ||
| MatAssemblyBegin(A, MAT_FINAL_ASSEMBLY); | ||
| MatAssemblyEnd(A, MAT_FINAL_ASSEMBLY); | ||
| } |
There was a problem hiding this comment.
You've done the testing, so I trust you on this. If you actually wanted me to double check, let me know. Otherwise mark as resolved.
There was a problem hiding this comment.
The assembly are definitely needed. The Copy, maybe not. But I'm sick of trying to guess what PETSc is going to do, best be safe. It costs nothing in practice.
Petsc was printing lots of errors. This makes it a bit more quiet. Also, a possible indexing error is fixed if no leftmost eigenvectors are asked for.