! transposemat.f90 --
!     Transpose a big matrix, check CPU usage
 program Console4

    implicit none
    double precision,allocatable,dimension(:,:):: P,PP
    integer(kind=8) :: tclock1, tclock2, clock_rate
    real(kind=8) :: elapsed_time, cputime1, cputime2
    integer nstate
    integer :: i, j
!   nstate=60720
    nstate=60720
    allocate(P(nstate,nstate),PP(nstate,nstate))

! Trick to get the physical memory close to the processor
!$omp parallel do
    do j = 1,nstate
        do i = 1,nstate
            pp(i,j) = 1.0d0
        enddo
    enddo
!$omp end parallel do
!$omp parallel do
    do j = 1,nstate
        do i = 1,nstate
            p(i,j)  = 0.0d0
        enddo
    enddo
!$omp end parallel do

    write(*,*) "1"
    call system_clock(tclock1)
    call cpu_time(cputime1)

!$omp parallel do
    do j = 1,nstate
        do i = 1,nstate
            p(i,j) = pp(j,i)
        enddo
    enddo
!$omp end parallel do
    call system_clock(tclock2, clock_rate)
    call cpu_time(cputime2)
    elapsed_time = float(tclock2 - tclock1) / float(clock_rate)
    write(*,*) elapsed_time
    write(*,*) cputime2-cputime1
    end program Console4