Next:
PBLAS operations
Up:
Departmental seminar at UWCC
Previous:
Basic Linear Alegbra Communication Subprograms
PBLAS
PBLAS perform Level 1, 2, and 3 BLAS operations on distributed matrices
Matrices are global objects
Matrices have a block cyclic data distribution
PBLAS are a subset of the BLAS, but
no banded and packed storage schemes
no vector rotation routines
Same calling sequence as BLAS except for each distributed matrix we have
global indices of start of matrix
descriptor array
call pdgemm (..., a, i, j, desca, ...)
Slide 40 of
Departmental seminar at UWCC
, David W. Walker, UWCC. (Updated 01/31/96)