vectorization: data dependence and subscript too complex

vectorization: data dependence and subscript too complex

Hi everyone - I am trying to better understand both data dependency and the subscript too complex rules.

The snippet below gives the subscript too complex error, whereas if I swap this line for the commented ones the code vectorizes (ifort 12, linux).

Can anyone explain why the compiler would balk at the one but not the other two in combination? They seem to access the same subscripts. What can be done?

Also, am I correct that I can do the two commented lines together (two accesses to zz(k) on one iteration of k)?

Thanks,

Eli

do i=1,ni
  do j=1,nj
    zz = 0.d0
!DIR$ IVDEP 
     do k=1,nk
         !zz(k)=zz(k)+trel_tmp(k,j,i)
         !zz(k)=zz(k)+trel_tmp(k,j-1,i)
         zz(k)=zz(k)+trel_tmp(k,j,i)+trel_tmp(k,j-1,i)
     end do
   end do
 end do

 

 

publicaciones de 2 / 0 nuevos
Último envío
Para obtener más información sobre las optimizaciones del compilador, consulte el aviso sobre la optimización.

I suppose it is obvious, but the loop on j would start at 2.

Inicie sesión para dejar un comentario.