Previous test compiled, meaning the problem is one of the two remaining loops or both. Changing one of them now.

Author: Daniel Vickers

src/common/m_mpi_common.fpp

@@ -871,22 +871,22 @@ contains
                     $:END_GPU_PARALLEL_LOOP()
 
                     if (qbmm_comm) then
-                        #:call GPU_PARALLEL_LOOP_OLD(collapse=5,private='[r]')
-                            do i = nVar + 1, nVar + 4
-                                do l = 0, buff_size - 1
-                                    do k = -buff_size, n + buff_size
-                                        do j = -buff_size, m + buff_size
-                                            do q = 1, nb
-                                                r = (i - 1) + (q - 1)*4 + v_size* &
-                                                    ((j + buff_size) + (m + 2*buff_size + 1)* &
-                                                     ((k + buff_size) + (n + 2*buff_size + 1)*l))
-                                                buff_send(r) = real(pb_in(j, k, l + pack_offset, i - nVar, q), kind=wp)
-                                            end do
+                        $:GPU_PARALLEL_LOOP(collapse=5,private='[r]')
+                        do i = nVar + 1, nVar + 4
+                            do l = 0, buff_size - 1
+                                do k = -buff_size, n + buff_size
+                                    do j = -buff_size, m + buff_size
+                                        do q = 1, nb
+                                            r = (i - 1) + (q - 1)*4 + v_size* &
+                                                ((j + buff_size) + (m + 2*buff_size + 1)* &
+                                                 ((k + buff_size) + (n + 2*buff_size + 1)*l))
+                                            buff_send(r) = real(pb_in(j, k, l + pack_offset, i - nVar, q), kind=wp)
                                         end do
                                     end do
                                 end do
                             end do
-                        #:endcall GPU_PARALLEL_LOOP_OLD
+                        end do
+                        $:END_GPU_PARALLEL_LOOP()
 
                         #:call GPU_PARALLEL_LOOP_OLD(collapse=5,private='[r]')
                             do i = nVar + 1, nVar + 4