In our blocking assignment, we must communicate the top and bottom row of each assignment which is where we get our 2PN term. There are 2*N communications per block, and P blocks/processors in this example. On the interleaved assignment, each row must be communicating twice since there is a dependency on the rows above and below it. Thus, the interleaved assignment has a total of ~2N^2 elements communicated, and a total of N^2 computed which gives us the ratio of 2.
In our blocking assignment, we must communicate the top and bottom row of each assignment which is where we get our 2PN term. There are 2*N communications per block, and P blocks/processors in this example. On the interleaved assignment, each row must be communicating twice since there is a dependency on the rows above and below it. Thus, the interleaved assignment has a total of ~2N^2 elements communicated, and a total of N^2 computed which gives us the ratio of 2.
This comment was marked helpful 0 times.