next up previous
Next: BPKIT2 Up: BlockSolve95 Previous: Contents

Parallelism and data layout

The user has two ways to pass the distributed matrix to BlockSolve.

  1. Since the definition of the data structure is given in the manual, the user can explicitly construct it.
  2. The user can supply a compressed row storage matrix, with column indices in the local numbering (section 2.2.2), to the routine BSeasy_A, which yields the matrix structure.

In the second case, the matrix rows need to be consecutively numbered. In the first case the assignment of rows over the processors can be arbitrary; the user has to construct the mapping functions between local and global numberings. There are example codes illustrating this.

Victor Eijkhout
Mon Aug 25 17:46:10 PDT 1997