Scattering matrix column by column in MPI using Fortran and contiguous datatype - parallel-processing

Question

I'm trying to generate a matrix full of random numbers, of dimensions (mysize)x(Np), where Np is a given number (say,5). Mysize comes from the number of processes. And then each processor should take its own column, of size 1xNp.
My first idea was to create a vector data type but then I realized this can be done by using the contiguous data type. The code runs with no errors, but it prints only some of the numbers that were supposed to be there. I'm quite sure the problem lies on the indicators (type_column, MPI_REAL, etc.) and how should they actually be but I can't figure this out. Anyone care to help?
The code:
program main
use mpi
integer :: ierr,myrank,mysize
integer :: Np=5, type_column
integer*8, external :: seedgen
real*8, dimension(:,:), allocatable :: x_init
real*8, dimension(:), allocatable :: block_x
character(4) :: rank
integer, dimension(1) :: new_seed
call MPI_INIT(ierr)
call MPI_COMM_RANK(MPI_COMM_WORLD,myrank,ierr)
call MPI_COMM_SIZE(MPI_COMM_WORLD,mysize,ierr)
allocate(block_x(0:Np-1))
if (myrank==0) then
allocate(x_init(0:Np-1,0:mysize-1))
new_seed = seedgen(myrank)
write(*,*) new_seed !This is to check if the seed changes each time
call random_seed(put=new_seed)
call random_number(x_init)
open(unit=1111, file='./x_tot.txt')
write(1111,*) x_init
close(1111) !Up to this point, everything works
end if
call MPI_TYPE_CONTIGUOUS(Np,MPI_REAL,type_column,ierr)
call MPI_TYPE_COMMIT(type_column,ierr)
call MPI_SCATTER(x_init(:,myrank),Np,MPI_REAL,block_x,Np,type_column,0,MPI_COMM_WORLD,ierr)
write(rank,'(I4)') myrank
write(*,*) 'I, process ',myrank,' received ', block_x(0:Np-1)
open(unit=myrank, file='./x_teste_'//trim(adjustl(rank))//'.txt')
write(myrank,*) block_x
close(myrank)
call MPI_TYPE_FREE(type_column,ierr)
call MPI_FINALIZE(ierr)
end program main
!!! You may ignore what is down below, just a function to generate random seeds.
function seedgen(myrank)
use iso_fortran_env
implicit none
integer(kind=int64) :: seedgen
integer, intent(IN) :: myrank
integer :: s
call system_clock(s)
seedgen = abs( mod((s*181)*((myrank-83)*359), 104729) )
end function seedgen
The output (after running with 3 processors, thus mysize=3)
I, process 0 received 0.247272870046176
0.386432141459887 8.816221414742263E-316 0.000000000000000E+000
0.000000000000000E+000
I, process 1 received 2.737792569698344E-008
I, process 2 received 5.231628354959370E-002
0.454932876242927 3.330771999738651E-315 0.000000000000000E+000
0.000000000000000E+000
3.352416881721526E+270 5.298272496856962E-315 0.000000000000000E+000
0.000000000000000E+000
The matrix x_tot:
0.247272870046176 0.386432141459887 0.473788032900533
0.239586263133600 0.851724848335892 5.231628354959370E-002
0.454932876242927 0.702720716936168 0.559915585253771
0.605745282251549 0.253298270763062 0.809867899324171
0.590174190311136 0.125210425650182 8.138975171285148E-002
So you can see that some numbers are the good ones, the rest is garbage.

Answer 1

Using MPI_DOUBLE as suggested above and changing the writing as below worked fine. Thanks again
open(unit=myrank, file='./x_teste_'//trim(adjustl(rank))//'.txt')
do i=0,Np-1
write(myrank,*) block_x(i)
end do
close(myrank)