fix: use pointer components in grid_axis for correct OpenMP GPU device access

Change grid_axis cb/cc/spacing from allocatable to pointer components backed
by the flat module arrays (x_cb/x_cc/dx etc.). GPU pointer attachment via
GPU_ENTER_DATA(attach=) updates the device struct's pointer fields to point
to the already-mapped device flat arrays, fixing CUDA_ERROR_ILLEGAL_ADDRESS
in m_igr.fpp inline GPU_PARALLEL_LOOP bodies on NVHPC OpenMP target offload.

Eliminates the early host sync, the duplicate GPU_UPDATE for struct members,
and the OpenACC/OpenMP split in GPU_DECLARE for x/y/z.

Author: sbryngelson

src/common/m_derived_types.fpp

@@ -15,8 +15,8 @@ module m_derived_types
 
     !> Derived type for a single spatial grid axis: cell-boundary, cell-center, per-cell spacing arrays, and minimum spacing scalar
     type grid_axis
-        real(wp), allocatable, dimension(:) :: cb, cc, spacing
-        real(wp)                            :: min_spacing = 0._wp
+        real(wp), pointer, dimension(:) :: cb => null(), cc => null(), spacing => null()
+        real(wp)                        :: min_spacing = 0._wp
     end type grid_axis
 
     !> Derived type adding the field position (fp) as an attribute

src/simulation/m_global_parameters.fpp

@@ -50,7 +50,7 @@ module m_global_parameters
 
     !> @name Cell-boundary (cb), cell-center (cc), and spacing arrays per direction
     !> @{
-    type(grid_axis), target :: x, y, z
+    type(grid_axis) :: x, y, z
     !> @}
 
     !> @name Flat GPU-accessible aliases for grid arrays (used in GPU kernels)
@@ -62,11 +62,7 @@ module m_global_parameters
     $:GPU_DECLARE(create='[dx, dy, dz, x_cc, y_cc, z_cc, x_cb, y_cb, z_cb]')
 
     real(wp) :: dt  !< Size of the time-step
-#if defined(MFC_OpenACC)
-    $:GPU_DECLARE(create='[x%cb, y%cb, z%cb, x%cc, y%cc, z%cc, x%spacing, y%spacing, z%spacing, dt, m, n, p]')
-#elif defined(MFC_OpenMP)
     $:GPU_DECLARE(create='[x, y, z, dt, m, n, p]')
-#endif
 
     !> @name Starting time-step iteration, stopping time-step iteration and the number of time-step iterations between successive
     !! solution backups, respectively
@@ -1224,37 +1220,34 @@ contains
         $:GPU_UPDATE(device='[relax, relax_model, palpha_eps, ptgalpha_eps]')
 
         ! Allocating grid variables for the x-, y- and z-directions
-        @:ALLOCATE(x%cb(-1 - buff_size:m + buff_size))
-        @:ALLOCATE(x%cc(-buff_size:m + buff_size))
-        @:ALLOCATE(x%spacing(-buff_size:m + buff_size))
-        @:PREFER_GPU(x%cb)
-        @:PREFER_GPU(x%cc)
-        @:PREFER_GPU(x%spacing)
         @:ALLOCATE(x_cb(-1 - buff_size:m + buff_size))
         @:ALLOCATE(x_cc(-buff_size:m + buff_size))
         @:ALLOCATE(dx(-buff_size:m + buff_size))
+        @:PREFER_GPU(x_cb)
+        @:PREFER_GPU(x_cc)
+        @:PREFER_GPU(dx)
+        x%cb => x_cb; x%cc => x_cc; x%spacing => dx
+        $:GPU_ENTER_DATA(attach='[x%cb, x%cc, x%spacing]')
 
         if (n == 0) return
-        @:ALLOCATE(y%cb(-1 - buff_size:n + buff_size))
-        @:ALLOCATE(y%cc(-buff_size:n + buff_size))
-        @:ALLOCATE(y%spacing(-buff_size:n + buff_size))
-        @:PREFER_GPU(y%cb)
-        @:PREFER_GPU(y%cc)
-        @:PREFER_GPU(y%spacing)
         @:ALLOCATE(y_cb(-1 - buff_size:n + buff_size))
         @:ALLOCATE(y_cc(-buff_size:n + buff_size))
         @:ALLOCATE(dy(-buff_size:n + buff_size))
+        @:PREFER_GPU(y_cb)
+        @:PREFER_GPU(y_cc)
+        @:PREFER_GPU(dy)
+        y%cb => y_cb; y%cc => y_cc; y%spacing => dy
+        $:GPU_ENTER_DATA(attach='[y%cb, y%cc, y%spacing]')
 
         if (p == 0) return
-        @:ALLOCATE(z%cb(-1 - buff_size:p + buff_size))
-        @:ALLOCATE(z%cc(-buff_size:p + buff_size))
-        @:ALLOCATE(z%spacing(-buff_size:p + buff_size))
-        @:PREFER_GPU(z%cb)
-        @:PREFER_GPU(z%cc)
-        @:PREFER_GPU(z%spacing)
         @:ALLOCATE(z_cb(-1 - buff_size:p + buff_size))
         @:ALLOCATE(z_cc(-buff_size:p + buff_size))
         @:ALLOCATE(dz(-buff_size:p + buff_size))
+        @:PREFER_GPU(z_cb)
+        @:PREFER_GPU(z_cc)
+        @:PREFER_GPU(dz)
+        z%cb => z_cb; z%cc => z_cc; z%spacing => dz
+        $:GPU_ENTER_DATA(attach='[z%cb, z%cc, z%spacing]')
 
     end subroutine s_initialize_global_parameters_module
 
@@ -1335,15 +1328,18 @@ contains
         if (ib) MPI_IO_IB_DATA%var%sf => null()
 
         ! Deallocating grid variables for the x-, y- and z-directions
-        @:DEALLOCATE(x%cb, x%cc, x%spacing)
+        $:GPU_EXIT_DATA(detach='[x%cb, x%cc, x%spacing]')
+        nullify (x%cb, x%cc, x%spacing)
         @:DEALLOCATE(x_cb, x_cc, dx)
 
         if (n == 0) return
-        @:DEALLOCATE(y%cb, y%cc, y%spacing)
+        $:GPU_EXIT_DATA(detach='[y%cb, y%cc, y%spacing]')
+        nullify (y%cb, y%cc, y%spacing)
         @:DEALLOCATE(y_cb, y_cc, dy)
 
         if (p == 0) return
-        @:DEALLOCATE(z%cb, z%cc, z%spacing)
+        $:GPU_EXIT_DATA(detach='[z%cb, z%cc, z%spacing]')
+        nullify (z%cb, z%cc, z%spacing)
         @:DEALLOCATE(z_cb, z_cc, dz)
 
     end subroutine s_finalize_global_parameters_module

src/simulation/m_start_up.fpp

@@ -924,15 +924,6 @@ contains
 
         call s_populate_grid_variables_buffers()
 
-        ! Sync flat grid aliases from struct members before module inits that use them (e.g. WENO coeff setup)
-        x_cb = x%cb; x_cc = x%cc; dx = x%spacing
-        if (n > 0) then
-            y_cb = y%cb; y_cc = y%cc; dy = y%spacing
-        end if
-        if (p > 0) then
-            z_cb = z%cb; z_cc = z%cc; dz = z%spacing
-        end if
-
         if (model_eqns == 3) call s_initialize_internal_energy_equations(q_cons_ts(1)%vf)
         if (ib) then
             if (cfl_dt .and. n_start > 0) then
@@ -1089,16 +1080,6 @@ contains
         $:GPU_UPDATE(device='[acoustic_source, num_source]')
         $:GPU_UPDATE(device='[sigma, surface_tension]')
 
-        $:GPU_UPDATE(device='[x%spacing, y%spacing, z%spacing, x%cb, x%cc, y%cb, y%cc, z%cb, z%cc]')
-
-        ! Sync flat GPU aliases from struct members
-        dx = x%spacing; x_cc = x%cc; x_cb = x%cb
-        if (n > 0) then
-            dy = y%spacing; y_cc = y%cc; y_cb = y%cb
-        end if
-        if (p > 0) then
-            dz = z%spacing; z_cc = z%cc; z_cb = z%cb
-        end if
         $:GPU_UPDATE(device='[dx, x_cc, x_cb]')
         if (n > 0) then
             $:GPU_UPDATE(device='[dy, y_cc, y_cb]')