Fortran中的OpenMP矩阵向量乘法

Question

我正在编写Fortran代码来执行矩阵向量乘法。在未预先存储矩阵元素的意义上，乘法是即时执行的。代码在串行情况下运行良好，但是当我使用OpenMP实现并行化的情况时，我无法获得正确的输出。可能有人建议我做错了什么？

Module module1
   integer:: N, M
   integer, allocatable:: Jeven_A(:), Jeven_B(:), Jodd_A(:), Jodd_B(:), s(:)
END module module1

subroutine lookup(a,Pswitch,sa)
    use module1
    implicit none
    integer, intent(in):: a, Pswitch
    integer, intent(out):: sa
    integer:: Ia, Ib

    Ib = ibits(a,N/2,N/2)
    Ia = ibits(a,0,N/2)
    if (Pswitch==0) then
        sa = Jeven_A(Ia)+Jeven_B(Ib)
    else if (Pswitch==1) then
        sa = Jodd_A(Ia)+Jodd_B(Ib)
    end if
end subroutine lookup

SUBROUTINE Mv_onthefly(V, vni, vno)
    USE module1
    IMPLICIT NONE
    INTEGER, INTENT(IN):: vni, vno
    complex, DIMENSION(1:M,1:3):: V
    INTEGER::sb, i,j ,k, a, b
    complex, allocatable:: Vvni(:), Vvno(:)

    allocate(Vvni(1:M),Vvno(1:M))
    Do i=1, M
        Vvni(i) = V(i,vni)
    END DO
    Vvno = 0.

!$OMP PARALLEL private(k,i,j,sb,a,b,bittest,Vvni)
!$OMP DO REDUCTION(+:Vvno)
   DO k=1,M
       a = s(k) 
       b = 0
       DO i=0, N-1
            Do j=0, N-1
                  b = IEOR(a,2**i+2**j)
                  CALL lookup(b,0,sb)  !subroutine which lookup the index sb corresponds to b in the array s
                  Vvno(sb)=Vvno(sb)-J1*Vvni(k)
            END do
        END DO
        DO i=0, N-1
            Vvno(k)=Vvno(k)-J2*Vvni(k)
        END DO
    END DO
!$OMP END DO
!$OMP END PARALLEL

   Do i =1, M
       V(i,vno)=Vvno(i)
   END DO
   deallocate(Vvni,Vvno)

END SUBROUTINE Mv_onthefly