Download Here

Transcript
Figure B–5 LU Decomposition with (*, BLOCK) Distribution
MLO-011936
Computation is done on progressively smaller submatrices in the lower
right-hand corner of the array. The first panel of the figure shows the first
iteration of the DO loop in which the entire array is worked on by all four
processors. The second panel shows the seventh iteration, by which time Peer
0 is completely idle because none of the elements of the submatrix are stored
in its memory. The third panel of the figure shows the eleventh iteration of
the DO loop, by which time both Peer 0 and Peer 1 are idle. The fourth panel
shows the fifteenth iteration, where only Peer 3 is working, with the other
three processors idle. For most of time spent in the DO loop, one or more
processors are left idle.
B–16 HPF Tutorial: LU Decomposition