This is the smallest working set we can use since for each cell, we need data from its immediate neighbors.
To get more cache hits, we want to access the grid row-major, and then come to the next row with its neighbors which are above it right on top of it. Since we need data from the neighbors above and below, we would split the processor's local partition into grids of three rows.
This is the processor's entire local partition, not split as in the previous example. We want the processor to be working on a "connected" set of points so that we minimize communication and get more cache hits.
This is the smallest working set we can use since for each cell, we need data from its immediate neighbors.
To get more cache hits, we want to access the grid row-major, and then come to the next row with its neighbors which are above it right on top of it. Since we need data from the neighbors above and below, we would split the processor's local partition into grids of three rows.
This is the processor's entire local partition, not split as in the previous example. We want the processor to be working on a "connected" set of points so that we minimize communication and get more cache hits.
This comment was marked helpful 0 times.