MPI multiple sendrecv calls fortran - fortran

Question

I am working on a finite volume code and want to make MPI work. Now I am struggling with sending corners from one domain to the other. With 8 ranks the domain is divided as follow in coordinates (not matrix representation):
|1|3|5|7|
----------
|0|2|4|6|
Each domain is a subarray whose elements are equal to its rank. For example, since I am solely focused on sending corners :
I am to send 1 from the process 1 to process 2 and
send 2 from the process 2 to process 1. The following code, only works for even (resp.odd) ranks if u comment the first (resp.last) two sendrecv. Which is normal. But if the four sendrecv are uncommented only the odd ranks will receive the values. The attended result is as follow :
5 | | 2 5 5 5 5
5 | | 5 5 5 5 5
5 | | 5 5 5 5 5
5 | | 6 5 5 5 5
6 | | 6 6 6 6 5
6 | | 6 6 6 6 6
6 | | 6 6 6 6 6
7 | | 4 7 7 7 7
7 | | 7 7 7 7 7
7 | | 7 7 7 7 7
0 | | 0 0 0 0 0
0 | | 0 0 0 0 0
0 | | 0 0 0 0 3
1 | | 1 1 1 1 1
1 | | 1 1 1 1 1
1 | | 2 1 1 1 1
2 | | 2 2 2 2 1
2 | | 2 2 2 2 2
2 | | 2 2 2 2 2
2 | | 2 2 2 2 5
3 | | 0 3 3 3 3
3 | | 3 3 3 3 3
3 | | 3 3 3 3 3
3 | | 4 3 3 3 3
4 | | 4 4 4 4 3
4 | | 4 4 4 4 4
4 | | 4 4 4 4 4
4 | | 4 4 4 4 7
The print is in matrix representation (so the equivalent in coordinates/real notation is to transpose then flip the matrix according to the horizontal axis).
The code :
program scatter
use mpi
implicit none
integer,parameter :: nrows = 8,ncols = 8
integer, dimension(8,8) :: global
integer, dimension(8) :: adj
integer, dimension(2) :: dims_orig,dims,coo,sizes,subsizes,coo_init
integer, dimension(2) :: coordsnw,coordsne,coordssw,coordsse
integer, dimension(2) :: size_glob, size_loc
logical,dimension(2) :: periods
integer, dimension(:,:),allocatable :: local
integer, dimension(:),allocatable :: displ,counts
integer :: info,code,rank,n_procs,comm2d
integer :: ndi,ndj,nni,nnj,nni_last,nnj_last
integer :: i1,i2,j1,j2,ii1,ii2,jj1,jj2,row,col
integer :: type0,type_subarray,sizeofint,i,j,ii,kk
integer :: imax,jmax,loc_size,ndd,tag=1
integer :: iminR,imaxR,jminR,jmaxR
integer :: type_corner,type_base,sizeofdouble
integer, dimension(mpi_status_size) :: stats
integer(kind=mpi_address_kind) :: start,extend
logical :: reorder
imax = nrows
jmax = ncols
global = -10
call mpi_init(code)
call mpi_comm_rank(mpi_comm_world,rank,code)
call mpi_comm_size(mpi_comm_world,n_procs,code)
dims_orig = 0
CALL MPI_DIMS_CREATE(n_procs ,2,dims_orig,code)
! define number of subdomain in each direction
ndi = dims_orig(1)
ndj = dims_orig(2)
! define size of subdomain in each direction
nni = nrows/ndi
nnj = ncols/ndj
nni_last = nrows-(ndi-1)*nrows/ndi
nnj_last = ncols-(ndj-1)*ncols/ndj
dims(1) = ndi
dims(2) = ndj
periods = .false.
reorder = .true.
! order the rank according to the domain separation
call mpi_cart_create(MPI_COMM_WORLD,2,dims,periods,reorder,comm2d,code)
! get rank in the new comm_world
call mpi_comm_rank(comm2d,rank,code)
! get coordinates (for example (1,0)) in the new comm_world
call mpi_cart_get(comm2d,2,dims,periods,coo,code)
! defining the indices of each subdomains
! the indices are the real indices in the global array 8x8
if(coo(1)==ndi-1) then
nni = nni_last
i2 = imax
i1 = i2-nni+1
else
i1 = rank/ndj*nni+1
i2 = i1+nni-1
endif
if(coo(2)==ndj-1) then
nnj = nnj_last
j2 = jmax
j1 = j2-nnj+1
else
j1 = MOD(rank,ndj)*nnj+1
j2 = j1+nnj-1
endif
! get the neighboors
adj = MPI_PROC_NULL
call MPI_CART_SHIFT(comm2d,1,1,adj(1),adj(2),code)
call MPI_CART_SHIFT(comm2d,0,1,adj(3),adj(4),code)
IF(adj(2)/=-2) THEN
coordsnw(1) = coo(1)-1
coordsnw(2) = coo(2)+1
IF(coordsnw(2)/=ndj.and.coordsnw(1)/=-1) THEN
call MPI_CART_RANK(comm2d,coordsnw,adj(5),code)
ENDIF
coordsne(1) = coo(1)+1
coordsne(2) = coo(2)+1
IF(coordsne(2)/=ndj.and.coordsne(1)/=ndi) THEN
call MPI_CART_RANK(comm2d,coordsne,adj(6),code)
ENDIF
ENDIF
IF(adj(1)/=-2) THEN
coordssw(1) = coo(1)-1
coordssw(2) = coo(2)-1
IF(coordssw(1)/=-1.and.coordssw(2)/=-1) THEN
call MPI_CART_RANK(comm2d,coordssw,adj(7),code)
ENDIF
coordsse(1) = coo(1)+1
coordsse(2) = coo(2)-1
IF(coordsse(1)/=ndi.and.coordsse(2)/=-1) THEN
call MPI_CART_RANK(comm2d,coordsse,adj(8),code)
ENDIF
ENDIF
! indices of the subdomain for each rank
! bigger than the actual since I need the interface
! for flux calculation (Finite volume)
iminR = MAX0(i1-1,1)
imaxR = MIN0(i2+1,imax)
jminR = MAX0(j1-1,1)
jmaxR = MIN0(j2+1,jmax)
allocate(local(iminR:imaxR,jminR:jmaxR))
local = rank
! FIN DEFINITION TABLEAU LOCAUX
! ENVOI DE MESSAGES
! new type
! call mpi_type_extent(mpi_double_precision,sizeofdouble,code)
!
! ! type coin /
! ! call mpi_type_vector(1,1,1,mpi_double_precision,type_base,code)
! size_glob = [imaxR-iminR+1, jmaxR-jminR+1]
! size_loc = [1, 1]
! if(rank==0) coo_init = [imaxR, jmaxR]
! coo_init = [iminR, jmaxR]
! call mpi_type_create_subarray(2,size_glob,size_loc,coo_init,mpi_order_fortran&
! ,mpi_double_precision,type_base,code)
! call mpi_type_commit(type_base,code)
! adj 1 -> S
! adj 2 -> N
! adj 3 -> W
! adj 4 -> E
! adj 5 -> NW
! adj 6 -> NE
! adj 7 -> SW
! adj 8 -> SE
! NW = North West
! SE = South East ...
! send NW / recv SE
call MPI_SENDRECV(local(iminR,jmaxR),1,MPI_INTEGER,adj(5),&
tag,local(imaxR,jminR),1,MPI_INTEGER,adj(8),&
tag,mpi_comm_world,stats,info)
! send NE / recv SW
call MPI_SENDRECV(local(imaxR,jmaxR),1,MPI_INTEGER,adj(6),&
tag,local(iminR,jminR),1,MPI_INTEGER,adj(7),&
tag,MPI_COMM_WORLD,stats,info)
! send SW / recv NE
call MPI_SENDRECV(local(iminR,jminR),1,MPI_INTEGER,adj(7),&
tag,local(imaxR,jmaxR),1,MPI_INTEGER,adj(6),&
tag,MPI_COMM_WORLD,stats,info)
! send SE / recv NW
call MPI_SENDRECV(local(imaxR,jminR),1,MPI_INTEGER,adj(8),&
tag,local(iminR,jmaxR),1,MPI_INTEGER,adj(5),&
tag,mpi_comm_world,stats,info)
call mpi_barrier(mpi_comm_world,code)
do i = iminR,imaxR
! print*, rank,"|",iminR,imaxR,jminR,jmaxR,"|",local(i,:)
print*, rank,"| ","|",local(i,:)
enddo
print*, ""
call mpi_finalize(code)
endprogram scatter
Edit : each sendrecv works by itself but using multiple sendrecv doesnt work.