ScaLAPACK - one-dimensional block row/column distribution

Post here if you have a question about LAPACK or ScaLAPACK algorithm or data format

ScaLAPACK - one-dimensional block row/column distribution

Postby sknerd » Tue Jan 10, 2012 5:13 am

Hello,
For the 1D block row/column distribution used with ScaLAPACK to solve banded systems is it possible to for the
user to define the blocking (in FORTRAN)? (mabey im reading it badley but the documentation suggests it is a fixed/automated distribution process)

Reasoning :
I require the (band stored) matricies A(1:n,1:Mmax) and B(1:n,1:Mmax - 1) to be split across a cluster in the second dimension.
For an arbitrary number of processes and matrix dimension Mmax I need the m-blocks to coincide on each process
ie - A(:, ml : ma_u), B(:, ml : mb_u) where mb_u = MINVAL(ma_u, Mmax-1)
for some ml and ma_u, fixed to the process.

Thanks.
sknerd
 
Posts: 1
Joined: Tue Jan 10, 2012 4:50 am

Return to Algorithm / Data

Who is online

Users browsing this forum: Yahoo [Bot] and 1 guest