平行折叠 + 减少不起作用

问：

我正在尝试并行化两个嵌套循环，但折叠子句失败了。

嘿，我正在尝试并行化这两个嵌套循环，以计算两个积分（int_coulomb 和 int_overlap）。当我应用 collapse 子句时，我的问题就来了，然后我的代码无法从 in 序列中恢复结果。

此外，我还添加了使用所选数量的处理器启动 fortran 程序（并行编译）的可能性。如果我做 ./program input.inp -omp 1，我的代码都不起作用。

另一方面，当我消除 collapse 子句时，我的代码比串行慢，但给了我正确的结果。

This is my code:

!$omp parallel do private(i,j,r,dist,invdst,sf,sf0,screen_pot) &
!$omp collapse(2) & 
!$omp reduction(+:int_coulomb,int_overlap) 
!    aceptor
     do i = 1, aceptor%n_points_reduced
        do j = 1, donor%n_points_reduced

           r(1) = (aceptor%xyz(1,i)-donor%xyz(1,j))
           r(2) = (aceptor%xyz(2,i)-donor%xyz(2,j))
           r(3) = (aceptor%xyz(3,i)-donor%xyz(3,j))
!
           dist = dsqrt(DOT_PRODUCT(r,r))
!
!          Skip when grid points are coincident to avoid instabilities
           if (dist.le.1.0e-14) then
              int_coulomb = int_coulomb + zero
              int_overlap = int_overlap + zero
           else
              invdst = one/dist
!
!             Screening function 
              sf  = dist / QMscrnFact

              sf0        = erf(sf)
              screen_pot = sf0
!   
!             Integrate charges as rho_aceptor * rho_donor * (1/dist) * screening
!               --> the density has been already weigthed by the cube volume
              int_coulomb = int_coulomb +&
                            aceptor%rho_reduced(i) * donor%rho_reduced(j) * invdst * screen_pot
!   
!             Aceptor-donor overlap 
              int_overlap = int_overlap +&
                            aceptor%rho_reduced(i) * donor%rho_reduced(j)
!   
           endif      
!
        enddo
     enddo
!$omp end parallel do`

有什么想法吗？