add nvhpc compiler flags to cmake

CMakeLists.txt

@@ -18,15 +18,15 @@
 
 # Copyright (c) GFDL, @underwoo
 
-cmake_minimum_required(VERSION 3.12 FATAL_ERROR)
+cmake_minimum_required(VERSION 3.22 FATAL_ERROR)
 
 # Define the CMake project
 project(FMS
   VERSION 2025.04.0
   DESCRIPTION  "GFDL FMS Library"
   HOMEPAGE_URL "https://www.gfdl.noaa.gov/fms"
   LANGUAGES C Fortran)
-
+set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
 include(GNUInstallDirs)
 
 if(NOT CMAKE_BUILD_TYPE MATCHES "^(Debug|Release|RelWithDebInfo|MinSizeRel|DebugUFS|ReleaseUFS)$")
@@ -66,6 +66,11 @@ option(LARGEFILE             "Enable compiler definition -Duse_LARGEFILE"
 option(WITH_YAML             "Enable compiler definition -Duse_yaml"              OFF)
 option(USE_DEPRECATED_IO     "THIS OPTION HAS NO EFFECT AND WILL BE REMOVED IN A FUTURE RELEASE"   OFF)
 
+if(CMAKE_Fortran_COMPILER_ID MATCHES "NVHPC")
+  set(ENABLE_QUAD_PRECISION OFF)
+  message(WARNING "NVHPC does not support building FMS with QUAD_PRECISION!")
+endif()
+
 if(32BIT)
   list(APPEND kinds "r4")
 endif()
@@ -332,6 +337,7 @@ foreach(kind ${kinds})
                                                  mosaic2/include
                                                  constants
                                                  astronomy/include
+                                                 field_manager/
                                                  field_manager/include
                                                  time_interp/include
                                                  tracer_manager/include
@@ -747,9 +753,13 @@ list(APPEND TEST_SRC_SINGLE_TARGET
   test_fms/parser/parser_demo2.F90
   test_fms/parser/test_output_yaml.F90
   test_fms/parser/generic_blocks.F90
-  test_fms/string_utils/test_string_utils.F90
   test_fms/time_manager/test_time_manager.F90
   )
+if(NOT ${CMAKE_Fortran_COMPILER_ID} STREQUAL "NVHPC")
+list(APPEND TEST_SRC_SINGLE_TARGET
+test_fms/string_utils/test_string_utils.F90
+)
+endif()
 # list any test sources that are used for two executables (r4 and r8)
 list(APPEND TEST_SRC_TWO_TARGETS
   test_fms/astronomy/test_astronomy.F90

cmake/compiler_flags_NVHPC_C.cmake

@@ -0,0 +1,6 @@
+# NVHPC C
+set(CMAKE_C_FLAGS_RELEASE "-O3")
+
+set(CMAKE_C_FLAGS_DEBUG "-O0 -g")
+
+set(CMAKE_C_LINK_FLAGS "")

cmake/compiler_flags_NVHPC_Fortran.cmake

@@ -0,0 +1,13 @@
+# Precision-based Fortran compiler flags
+set(r8_flags "-r8") # Fortran flags for 64BIT precision
+
+# NVHPC Fortan
+set(CMAKE_Fortran_FLAGS "${CMAKE_Fortran_FLAGS} ")
+
+set(CMAKE_Fortran_FLAGS_RELEASE "-O3")
+
+set(CMAKE_Fortran_FLAGS_DEBUG "-O0" )
+# -g can cause bugs, see: https://forums.developer.nvidia.com/t/bug-compiling-with-g-o0-produces-a-compute-sanitizer-error-removing-g-removes-the-error/341478
+# not sure if this is only on GPUs
+
+set(CMAKE_Fortran_LINK_FLAGS "" )

cmake/fms_compiler_flags.cmake

@@ -8,6 +8,8 @@ elseif(CMAKE_Fortran_COMPILER_ID STREQUAL "Intel")
   include(compiler_flags_Intel_Fortran)
 elseif(CMAKE_Fortran_COMPILER_ID STREQUAL "IntelLLVM")
   include(compiler_flags_IntelLLVM_Fortran)
+elseif(CMAKE_Fortran_COMPILER_ID STREQUAL "NVHPC")
+  include(compiler_flags_NVHPC_Fortran)
 else()
   message(WARNING "Fortran compiler with ID ${CMAKE_Fortran_COMPILER_ID} will be used with CMake default options")
 endif()
@@ -24,6 +26,8 @@ elseif(CMAKE_C_COMPILER_ID STREQUAL "IntelLLVM")
   include(compiler_flags_IntelLLVM_C)
 elseif(CMAKE_C_COMPILER_ID MATCHES "Clang")
   include(compiler_flags_Clang_C)
+elseif(CMAKE_C_COMPILER_ID MATCHES "NVHPC")
+  include(compiler_flags_NVHPC_C)
 else()
   message(WARNING "C compiler with ID ${CMAKE_C_COMPILER_ID} will be used with CMake default options")
 endif()

exchange/xgrid.F90

@@ -4023,9 +4023,12 @@ subroutine get_1_from_xgrid_repro(d_addrs, x_addrs, xmap, xsize, lsize)
   real(r8_kind), pointer                      :: x(:)
   real(r8_kind), pointer, contiguous          :: tmpptr(:,:)
   integer :: shape_d(2)
+  integer :: i_off, j_off
 
   call mpp_clock_begin(id_get_1_from_xgrid_repro)
   shape_d = [xmap%grids(1)%ie_me-xmap%grids(1)%is_me+1, xmap%grids(1)%je_me-xmap%grids(1)%js_me+1]
+  i_off = xmap%grids(1)%is_me - 1
+  j_off = xmap%grids(1)%js_me - 1
   comm => xmap%get1_repro
   !--- pre-post receiving
   do p = 1, comm%nrecv
@@ -4071,25 +4074,25 @@ subroutine get_1_from_xgrid_repro(d_addrs, x_addrs, xmap, xsize, lsize)
 
   do l = 1, lsize
      call c_f_pointer(d_addrs(l), tmpptr, shape=shape_d)
-     d(xmap%grids(1)%is_me:xmap%grids(1)%ie_me, xmap%grids(1)%js_me:xmap%grids(1)%je_me) => tmpptr
+     d => tmpptr
      d = 0
   enddo
 
   call mpp_sync_self(check=EVENT_RECV)
 
-!$OMP parallel do default(none) shared(lsize,shape_d,d_addrs,xmap,recv_buffer,pl,ml) &
+!$OMP parallel do default(none) shared(lsize,shape_d,d_addrs,xmap,recv_buffer,pl,ml,i_off,j_off) &
 !$OMP                          private(d,tmpptr,grid,i,j,p,pos)
   do l = 1, lsize
      call c_f_pointer(d_addrs(l), tmpptr, shape=shape_d)
-     d(xmap%grids(1)%is_me:xmap%grids(1)%ie_me, xmap%grids(1)%js_me:xmap%grids(1)%je_me) => tmpptr
+     d => tmpptr
      do g=2,size(xmap%grids(:))
         grid => xmap%grids(g)
         do l3=1,grid%size_repro ! index into side1 grid's patterns
            i = grid%x_repro(l3)%i1
            j = grid%x_repro(l3)%j1
            p = grid%x_repro(l3)%pe-xmap%root_pe
            pos = pl(p) + (l-1)*ml(p) + grid%x_repro(l3)%recv_pos
-           d(i,j) = d(i,j) + recv_buffer(pos)
+           d(i - i_off, j - j_off) = d(i - i_off, j - j_off) + recv_buffer(pos)
         end do
      end do
      ! normalize with side 1 grid cell areas