Seed is now modified on each rank

This was not necessary before since SITERANDOM was used,
but SITERANDOM does not take into account the 4D parallelisation.

Also, the seed change would invalidate anything done with SITERANDOM,
so the main program must not be compiled with SITERANDOM.

Also, some fprettifying.

dwf3d_lib.F90

@@ -92,11 +92,11 @@ subroutine dwf3d_main
     wait_on_master = .true.
     if (ip_global .eq. 0) then
       do while (wait_on_master)
-         print*,'Waiting on master for intervention with GDB...'
-         call sleep(1)
+        print *, 'Waiting on master for intervention with GDB...'
+        call sleep(1)
       enddo
     endif
-    call MPI_Barrier(MPI_COMM_WORLD,ierr)
+    call MPI_Barrier(MPI_COMM_WORLD, ierr)
 #endif
     call timeinit
 #endif
@@ -134,6 +134,11 @@ subroutine dwf3d_main
     if (ip_global .eq. 0) then
       write (7, *) 'seed: ', seed
     end if
+
+#if defined(MPI)
+    call seed_rank(seed)
+#endif
+
     call init_random(seed)
 
     if (ip_global .eq. 0) then
@@ -614,13 +619,13 @@ subroutine hamilton(Phi, h, hg, hp, s, res2, isweep, iflag, am, imass)
 
       call qmrherm(R, Xresult, res2, itercg, One, 1, anum4, aden4, ndiag, 0)
       ancghpv = ancghpv + float(itercg)
-      call check_qmr_iterations(niterations = itercg, abort_on_max_reached = .true. ) 
+      call check_qmr_iterations(niterations=itercg, abort_on_max_reached=.true.)
 !
       R = Xresult
 !
       call qmrherm(R, Xresult, res2, itercg, am, imass, bnum2, bden2, ndiag, 0)
       ancgh = ancgh + float(itercg)
-      call check_qmr_iterations(niterations = itercg, abort_on_max_reached = .true. ) 
+      call check_qmr_iterations(niterations=itercg, abort_on_max_reached=.true.)
 !
       hf = hf + sum(real(conjg(R(:, 1:ksizex_l, 1:ksizey_l, 1:ksizet_l, :)) &
                         &        *Xresult(:, 1:ksizex_l, 1:ksizey_l, 1:ksizet_l, :)))
@@ -788,6 +793,28 @@ subroutine saveseed(seedtosave)
 #endif
   end subroutine saveseed
 !
+#ifdef MPI
+! This subroutine takes the global seed and changes it so that it is
+! different for each rank.
+! Makes also sure that is positive and greater than 1.
+  subroutine seed_rank(seed)
+    use comms_common, only: ip_global
+    real(dp), intent(inout) :: seed
+#ifdef SITERANDOM
+    if (NP_THIRD .ne. 1) then
+      print *, "SITERANDOM should not be used with NP_THIRD>1"
+      stop
+    endif
+#endif
+
+    seed = seed + ip_global
+
+    if (seed .lt. 1) then
+      seed = abs(seed) + 1
+    endif
+  end subroutine seed_rank
+#endif
+
   subroutine init(nc)
     use gammamatrices, only: init_gammas
     use gauge
@@ -887,28 +914,28 @@ subroutine coef(u, theta)
     return
   end subroutine coef
 
-subroutine check_qmr_iterations(niterations, abort_on_max_reached)
-  use params, only: max_qmr_iters
+  subroutine check_qmr_iterations(niterations, abort_on_max_reached)
+    use params, only: max_qmr_iters
 #ifdef MPI
-  use mpi
-  integer :: ierr
+    use mpi
+    integer :: ierr
 #endif
-  integer, intent(in) :: niterations
-  logical, intent(in) :: abort_on_max_reached
+    integer, intent(in) :: niterations
+    logical, intent(in) :: abort_on_max_reached
 
-  if ( niterations .eq. max_qmr_iters ) then
-    if (abort_on_max_reached) then 
-      print*,"ERROR: Max QMR iterations reached."
+    if (niterations .eq. max_qmr_iters) then
+      if (abort_on_max_reached) then
+        print *, "ERROR: Max QMR iterations reached."
 #ifdef MPI
-      call MPI_Abort(MPI_COMM_WORLD,1,ierr)
+        call MPI_Abort(MPI_COMM_WORLD, 1, ierr)
 #else
-      stop
+        stop
 #endif
-    else
-      print*,"WARNING: Max QMR iterations reached."
-    endif 
-  endif
-end subroutine check_qmr_iterations
-  
+      else
+        print *, "WARNING: Max QMR iterations reached."
+      endif
+    endif
+  end subroutine check_qmr_iterations
+
 end module dwf3d_lib