Provide complex and real type matrices regardless of the underlying PETSc type about gridpack HOT 5 CLOSED

This is nearly complete. The only thing left are diagonal operations (extraction, add, multiply, etc.).

from gridpack.

Some MatrixT<> methods for complex value manipulation are not correct: real(), imaginary(), conjugate(). These should not work with a real-valued PETSc. Review unit tests to make sure these are tested.
Also, there is a similar problem with MatrixT<>::norm2().

from gridpack.

I think the only thing left is an implementation of MatrixT<>::scale() for complex matrices based on a real-valued library. I could not figure out a way to implement this at a high level (fallback namespace). It will probably have to be done at a pretty low level. An exception is thrown if MatrixT<ComplexType>::scale() is called when the underlying library is real-valued.

from gridpack.

I need to re-open this because there are several complex matrix methods not yet implemented for a real-based library: real(), imaginary(), and conjugate() (maybe others). Apparently, these need better unit tests too. These methods should be implemented using the operator strategy used by the vector implementation, but using MatGetRow() is in PETScMatrixImplementation<>::p_scale().

from gridpack.

I think this is done (again).

from gridpack.