In the sequential code, a single processor updates in turn (16x16=256-60=196) internal points. Each update is j --> 0.25 *(j up + jdown + jleft + jright ) involves 4 floating point operations for each point |
In the parallel case, each processor updates in turn the points for which it is responsible - this is "owner computes rule" |
A corner processor updates nine points |
(the small internal points) |
A "general" central processor |
updates sixteen points |