Data decomposition - divide the data into pieces with equal amounts of processing required. |
Distribute data blocks among processors - shown here on a MIMD distributed memory or SIMD machine. |
Each processor runs a traditional sequential program implementing the Fortran90 program, looping over the subarray in its own memory. |
If an array element A(i) is used with an array element B(j) which happens to be on another processor, communication will be used. |
001 Introduction to High Performance Fortran (HPF) 002 Parallel Implementations of Fortran90 003 Example Program: a PDE solver, the solution of LaPlace's Equation 004 Grid Decomposition 005 Implementation of the Gauss-Jordan Program 006 Load Imbalance arises after i iterations over columns: 007 Load Balancing for Regular Gaussian Elimination during Forward Reduction