In the message passing model, each node has its own address space, so there is no single block that all processors can read and write to. Therefore the processors need to issue a receive (and a send not shown) in order to communicate.
This comment was marked helpful 0 times.
placebo
Question: Assuming that each row has N items, is there a reason we allocate N+2 cells for each ghost row in the code shown? Or are we just working under the assumption that each row has N+2 items?
This comment was marked helpful 0 times.
GG
@placebo. It is the latter case. The gird itself is (N+2)*(N+2). You can find the problem description in lecture 4, slide 21.
This comment was marked helpful 0 times.
martin
In a shared address space model, thread 2 would simply issue loads for the last row from thread 1 and the first row from thread 3 directly from the hardware level. In the message passing model, it has to request those two rows of data through message passing.
In the message passing model, each node has its own address space, so there is no single block that all processors can read and write to. Therefore the processors need to issue a receive (and a send not shown) in order to communicate.
This comment was marked helpful 0 times.
Question: Assuming that each row has N items, is there a reason we allocate N+2 cells for each ghost row in the code shown? Or are we just working under the assumption that each row has N+2 items?
This comment was marked helpful 0 times.
@placebo. It is the latter case. The gird itself is (N+2)*(N+2). You can find the problem description in lecture 4, slide 21.
This comment was marked helpful 0 times.
In a shared address space model, thread 2 would simply issue loads for the last row from thread 1 and the first row from thread 3 directly from the hardware level. In the message passing model, it has to request those two rows of data through message passing.
This comment was marked helpful 0 times.