Updated stdpar version to be more in sync with main version. Also add…

…ed a stdpar_datadir version that is the same as stdpar but with data directives added back in. This makes is different than the main version only in array reductions were the stdpar uses the do concurrent reduce clause, while the main code still uses OpenMP target loops.

src/stdpar/README

@@ -1,10 +1,11 @@
-This version of the code has ZERO OpenACC directives.
+This version of the code has no directives other than device selection.
 
 It requires Fortran 2023 standard as it uses "reduce" on 
 do concurrent loops.
 
-In order to run this on more than one GPU, you must launch it 
-in a special way.
+To make a "pure Fortran" version of the code, one can remove the two 
+device selection directives.  If they are removed, in order to run
+on more than one GPU, it must be launched in a special way.
 
 An example of a launch script to do this is below:
 

src/stdpar/pot3d.F90

@@ -52,7 +52,7 @@ module ident
 !-----------------------------------------------------------------------
 !
       character(*), parameter :: idcode='POT3D'
-      character(*), parameter :: vers  ='4.3.1_nogpumpi'
+      character(*), parameter :: vers  ='4.3.1_stdpar'
       character(*), parameter :: update='12/05/2024'
 !
 end module
@@ -3820,7 +3820,7 @@ subroutine potfld
 ! ****** boundary conditions (i.e., the inhomogenous part).
 !
       call set_boundary_points (x_ax,one)
-      call seam_gen (x_ax,nr,nt,np)
+      call seam_hhh (x_ax)
       call delsq (x_ax,rhs_cg)
 !
 ! ****** Original rhs is zero so just use negative of boundary
@@ -3847,7 +3847,7 @@ subroutine potfld
       call unpack_scalar (phi,x_cg)
 !
       call set_boundary_points (phi,one)
-      call seam_gen (phi,nr,nt,np)
+      call seam_hhh (phi)
 !
       call dealloc_pot3d_matrix_coefs
       deallocate(rhs_cg)
@@ -3986,7 +3986,7 @@ subroutine write_validation_solution
       enddo
 !
       call set_boundary_points (phi,one)
-      call seam_gen (phi,nr,nt,np)
+      call seam_hhh (phi)
 !
 !
 ! ****** Set Br.
@@ -5020,7 +5020,7 @@ subroutine ax (x,y,N)
 !
 ! ****** Seam along edges between processors.
 !
-      call seam_gen (x_ax,nr,nt,np)
+      call seam_hhh (x_ax)
 !
 ! ****** Get the matrix-vector product.
 !
@@ -5737,8 +5737,6 @@ subroutine seam_gen (a,n1,n2,n3)
       real(r_typ), dimension(n2,n3) :: sbuf12,rbuf12
       real(r_typ), dimension(n1,n3) :: sbuf21,rbuf21
       real(r_typ), dimension(n1,n3) :: sbuf22,rbuf22
-      real(r_typ), dimension(n1,n2) :: sbuf31,rbuf31
-      real(r_typ), dimension(n1,n2) :: sbuf32,rbuf32
 !
 !-----------------------------------------------------------------------
 !
@@ -5761,34 +5759,23 @@ subroutine seam_gen (a,n1,n2,n3)
       call timer_on
 !
 ! ****** Seam the third (periodic) dimension.
-!
+! ****** Since halo data is stride-1, no need for buffers.
 !
       lbuf=n1*n2
 !
-      do concurrent (j=1:n2, i=1:n1)
-        sbuf31(i,j)=a(i,j,n3-1)
-        sbuf32(i,j)=a(i,j,   2)
-      enddo
-!
-      call MPI_Isend (sbuf31,lbuf,ntype_real,iproc_pp,tag, &
+      call MPI_Isend (a(:,:,n3-1),lbuf,ntype_real,iproc_pp,tag, &
                       comm_all,reqs(1),ierr)
 !
-      call MPI_Isend (sbuf32,lbuf,ntype_real,iproc_pm,tag, &
+      call MPI_Isend (a(:,:,   2),lbuf,ntype_real,iproc_pm,tag, &
                       comm_all,reqs(2),ierr)
 !
-      call MPI_Irecv (rbuf31,lbuf,ntype_real,iproc_pm,tag, &
+      call MPI_Irecv (a(:,:, 1),lbuf,ntype_real,iproc_pm,tag,   &
                       comm_all,reqs(3),ierr)
 !
-      call MPI_Irecv (rbuf32,lbuf,ntype_real,iproc_pp,tag, &
+      call MPI_Irecv (a(:,:,n3),lbuf,ntype_real,iproc_pp,tag,   &
                       comm_all,reqs(4),ierr)
 !
       call MPI_Waitall (4,reqs,MPI_STATUSES_IGNORE,ierr)
-!
-      do concurrent (j=1:n2, i=1:n1)
-        a(i,j, 1)=rbuf31(i,j)
-        a(i,j,n3)=rbuf32(i,j)
-      enddo
-!
 !
 ! ****** Seam the first dimension.
 !