! compile as: gfortran -O2 -fopenmp summ.f90 
! try also the option -march=native 
        program calcsum
        use omp_lib
        implicit none        
! rkind=4  single precision (32-bit)
! rkind=8  double precision (64-bit)
! rkind=10 double precision (80-bit)
        integer, parameter :: rkind=4
        integer(8),parameter:: two=2,p0=10,p1=35
        real(rkind) :: S(0:1),x
        integer:: ntmax,nproc,prl,nth
        integer(8) :: m,N,power
        real(8) :: t0(0:1),time(0:1),tick
        character(100) :: of

! Preparations
! Set OpenMP options (see also produce_test.sh)
! set number of thereads for parallel region
!       ntmax=omp_get_max_threads()
!       nproc=omp_get_num_procs()
!       nth=nproc
!       call omp_set_num_threads(nth)
!       tick=omp_get_wtick()
!       call omp_set_dynamic(.FALSE.)
!       call omp_set_nested(.false.)

! Get file name for output from shell script. if not define use default
       call getenv('TEST_PREC_OF',of)
       if(len_trim(adjustl(of))==0) of='local_sum.dat'

!!!!!!! Start calculations
        
        open(1,file=of)
!  write(*,11)'# p','N=2**p  ','bit','S(seq)','Time(seq)','S(par)','Time(par)' 
  write(1,11)'# p','N=2**p  ','bit','S(seq)','Time(seq)','S(par)','Time(par)'

        do power= p0,p1
           N= two**power
             do prl=0,1 !prl=0/1 for sequential/parallel OpenMP code
               x= real(0.,rkind)
               t0(prl)=omp_get_wtime()
!!!!$omp parallel if(prl==1) num_threads(nth) private(m) shared(N) 
!$omp parallel if(prl==1) private(m) shared(N) 
!$omp do reduction(+:x)
                 do m= -N,N
                   x=x + real(m,rkind)
                 end do
!$omp end parallel
               S(prl)=x
               time(prl)=omp_get_wtime()-t0(prl)
             enddo
           write(1,12) power,N,rkind*8,S(0),time(0),S(1),time(1)
!           write(*,12) power,N,rkind*8,S(0),time(0),S(1),time(1)
        enddo
        close(1)
!!!!!!!!
11  format(a3,a15,a4,2(a30,a15))
12  format(i3,i15,i4,2(e30.21,e15.6))
        end program