I want to verify something. In the following link (http://www.netlib.org/scalapack/slug/node35.html)
it is said that distribution of the matrices is users responsibility. However in this example code (http://www.netlib.org/scalapack/slug/no ... l#example2)
- Code: Select all
* Read from file and distribute matrices A and B
*
CALL PDLAREAD( 'SCAEXMAT.dat', MEM( IPA ), DESCA, 0, 0,
$ MEM( IPW ) )
CALL PDLAREAD( 'SCAEXRHS.dat', MEM( IPB ), DESCB, 0, 0,
$ MEM( IPW ) )
the part about the distibution is just calls to PDLAREAD functions. No distribution handling made in this example right?
Ofcourse when the "A" matrix is placed in all nodes as a whole matrix it can be possible somehow.
In my case I have the whole matrix partially at each node with their mappings on whole matrix. So for the time being I am distributing the data among the nodes. But if there is any easier way of doing this, it will be greatly welcome. because I know that in mumps you are not forced to distribute data, it is handling it in its nature. But in scalapack?
