Parallelization with Hyperplane(Pipelining)

!hpf$ processors Proc(Npe1,Npe2)

!hpf$ distribute(block,block,*) onto Proc::a0,b,r, a1,a2,x,p,q, a3,d

!hpf$ shadow(1:1,1:1,:) :: a1,a2,a3,d,x,p,q

・・・・・・・・・・

do 730 k = 1,Nz

do 720 j = 1,Ny

do 710 i = 1,Nx

q(i,j,k) = d(i,j,k)*(r(i,j,k)

& - a1(i,j,k)*q(i-1,j,k)

& - a2(i,j,k)*q(i,j-1,k)

& - a3(i,j,k)*q(i,j,k-1) )

710 continue

720 continue

730 continue

This simple case can be handled by compiler.

More complecated case: ???

前のｽﾗｲﾄﾞ

次のｽﾗｲﾄﾞ

最初のｽﾗｲﾄﾞに戻る

ｸﾞﾗﾌｨｯｸｽの表示