Go backward to
Shared Memory Matrix Multiplication
Go up to
Top
Go forward to
Design Goals
Nested Loop Parallelization
Parallelization of inner loop.
O(n)
processes.
n
synchronizations (one per iteration).
grain size
O(n
2
)
.
Parallelization of outer loop.
O(n)
processes.
1 synchronization.
grain size
O(n
3
)
.
Parallelization of both loops.
O(n
2
)
processes.
1 synchronization.
grain size
O(n)
.
Which strategy to choose?
Author:
Wolfgang Schreiner
Last modification: November 15, 1996