Laplace Equation Solver - Streamed

18 1 -------- do it=1,itmax

19 1 dumax=0.0

20 M--1v ------- do j=2,jm1

21 M 12v ------ do i=2,im1

22 M 12v du(i,j)=0.25*(u(i-1,j)+u(i+1,j)

+u(i,j-1)+u(i,j+1))-u(i,j)

23 M 12v -----> enddo

24 M--1v ------> enddo

25 12 ------- do j=2,jm1

26 12v ------ do i=2,im1

27 12v dumax=max(dumax,abs(du(i,j)))

28 12v -----> enddo

29 12 ------> enddo

30 M--12 ------- do j=2,jm1

31 M 12v ------ do i=2,im1

32 M 12v u(i,j)=u(i,j)+du(i,j)

33 M 12v -----> enddo

34 M--12 ------> enddo

35 1 write (1,*) it,dumax

36 1 -------> enddo

Previous slide

Next slide

Back to first slide

View graphic version

• First, let is revisit the Laplaces equation solver. Recall that SSP performance was about 30% of the T90 due to limited cache utilization

• This is the loopmark listing with addntl annotations

• 3 additional MSP optimization;

• 1)loop @ 20 streamed

• 2)loop @ 30 streamed

• 3)loop @ 20 vectorized

• Note that reduction operation @25 not streamed