I am new user of ScaLAPACK. I have a very basic question:
Given a matrix 'A", we can build a process grid and distribute the matrix data onto the grids and then call some ScalAPACK routines.
In the variable list of the routine, say, eigensolver:
pdsyev (jobz, uplo, n, a, ia, ja, desc_a, w, z, iz, jz, desc_z, work, lwork, info);
what is the matrix variable "a" on the entry actually? Is it using the whole global matrix "A" or just local distributed submatrix?
If it's just the local distributed submatrix, then what if now, my matrix has already been distributed, stored in local matrix "a", and I know exactly the block cyclic distribution mapping between the global and local indices. So, without gathering back all the data to form a single matrix "A", can I call the ScaLAPACK routines ?
I hope I am expressing myself clear....
Thank you very much !