Procedure Calls in Parallelized loops

!HPF$ independent, new(evnp, evnd, strk, gckik,

!HPF$& vlik, wvfnbk, gkiks), reduction(evnt)

do ib = 1,mxband

evnp = 0d0

evnd = 0d0

do j=1,npw_g

wvfnbk(1,j)=wvfn(1,j,ib,ik)

wvfnbk(2,j)=wvfn(2,j,ib,ik)

end do

!HPF$ on home(wvfn(:,:,ib,ik)),resident

call evnlpd(wvfnbk,strk,

& vnlik(1,1),vnlik(1,2),

& gckik(1,1),gckik(1,2),gckik(1,3),

& gkiks,npw_g,ngk,evnp,evnd)

evnt = evnt + (5d0*evnd/vnn(2) + 3d0*evnp/vnn(1))

end do

1) independent

parameters -> new

reduction

(copy dammy argument)

2) copy to new variables

3) on home

4) resident

Parallelize the loop by copying distributed arrays into NEW variables + RESIDENT

- Extra copy needed HPF/JA Extension

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