Christos, > Hi guys. Following up on a previous user request, I've been looking > at supporting hermitian mat-vec multiplication for the SBAIJ > implementation. I haven't played with petsc before, so please > forgive me for not understanding the internals. So, SBAIJ uses the > transpose of the upper band where access to the lower band is > needed. Am I right to believe that what is needed to be done is to > have the SBAIJ data retriever (such as MatGetValues_MPISBAIJ) > modified such that on sub-diagonal accesses the conjugate of the > original values are returned instead (possibly by updating the > contents of the "v" argument once filled in)?
Yes. Be careful in parallel implementation though, Hong
