Memory errors detected by memcheck (valgrind)

[DusanJovic-NOAA]

When I use the memcheck tool (valgrind) to check for memory errors in the ufs-weather-model, it detects these three errors in FMS:

==2308066== Conditional jump or move depends on uninitialised value(s)
==2308066==    at 0x4509B34: __mpp_mod_MOD_mpp_scatter_pelist_real8_gen_3d (mpp_scatter.fh:86)
==2308066==    by 0x450C0F9: __mpp_mod_MOD_mpp_scatter_pelist_real8_gen_2d (mpp_scatter.fh:63)
==2308066==    by 0x442FEDC: __fms_netcdf_domain_io_mod_MOD_domain_read_2d (domain_read.inc:261)
==2308066==    by 0x39D4CE0: __mom_io_infra_MOD_read_field_2d (MOM_io_infra.F90:907)
==2308066==    by 0x2D09994: __mom_io_MOD_mom_read_data_2d (MOM_io.F90:2217)
==2308066==    by 0x2EF73EA: __mom_grid_initialize_MOD_set_grid_metrics_from_mosaic (MOM_grid_initialize.F90:214)
==2308066==    by 0x2F00B8C: __mom_grid_initialize_MOD_set_grid_metrics (MOM_grid_initialize.F90:92)
==2308066==    by 0x2EDE9FF: __mom_fixed_initialization_MOD_mom_initialize_fixed (MOM_fixed_initialization.F90:82)
==2308066==    by 0x26AE5CA: __mom_MOD_initialize_mom (MOM.F90:2861)
==2308066==    by 0x26328B9: __mom_ocean_model_nuopc_MOD_ocean_model_init (mom_ocean_model_nuopc.F90:290)
==2308066==    by 0x25C0C9B: __mom_cap_mod_MOD_initializeadvertise (mom_cap.F90:755)

==2308066== Conditional jump or move depends on uninitialised value(s)
==2308066==    at 0x4514485: __mpp_mod_MOD_mpp_gather_pelist_real8_gen_3d (mpp_gather.fh:164)
==2308066==    by 0x45171B8: __mpp_mod_MOD_mpp_gather_pelist_real8_gen_2d (mpp_gather.fh:139)
==2308066==    by 0x441F8CF: __fms_netcdf_domain_io_mod_MOD_domain_write_2d (domain_write.inc:248)
==2308066==    by 0x39BEA00: __mom_io_infra_MOD_write_field_2d (MOM_io_infra.F90:1770)
==2308066==    by 0x2D470FD: __mom_io_file_MOD_write_field_2d_infra (MOM_io_file.F90:1157)
==2308066==    by 0x2CF9B94: __mom_io_MOD_mom_write_field_2d (MOM_io.F90:2913)
==2308066==    by 0x2F04C86: __mom_shared_initialization_MOD_write_ocean_geometry_file (MOM_shared_initialization.F90:1430)
==2308066==    by 0x26BD221: __mom_MOD_initialize_mom (MOM.F90:3137)
==2308066==    by 0x26328B9: __mom_ocean_model_nuopc_MOD_ocean_model_init (mom_ocean_model_nuopc.F90:290)
==2308066==    by 0x25C0C9B: __mom_cap_mod_MOD_initializeadvertise (mom_cap.F90:755)

==2308066== Conditional jump or move depends on uninitialised value(s)
==2308066==    at 0x4A2E748: __diag_output_mod_MOD_write_attribute_meta (diag_output.F90:730)
==2308066==    by 0x4A346D6: __diag_output_mod_MOD_write_axis_meta_data (diag_output.F90:368)
==2308066==    by 0x4355001: __diag_util_mod_MOD_opening_file (diag_util.F90:1830)
==2308066==    by 0x4350C30: __diag_util_mod_MOD_check_and_open (diag_util.F90:2173)
==2308066==    by 0x4351320: __diag_util_mod_MOD_diag_data_out (diag_util.F90:2083)
==2308066==    by 0x43031D0: __diag_manager_mod_MOD_closing_file (diag_manager.F90:4061)
==2308066==    by 0x4303439: __diag_manager_mod_MOD_diag_manager_end (diag_manager.F90:3990)
==2308066==    by 0x3993BB0: __mom_diag_manager_infra_MOD_mom_diag_manager_end (MOM_diag_manager_infra.F90:171)
==2308066==    by 0x2C0D762: __mom_diag_mediator_MOD_diag_mediator_end (MOM_diag_mediator.F90:3860)
==2308066==    by 0x2623EA7: __mom_ocean_model_nuopc_MOD_ocean_model_end (mom_ocean_model_nuopc.F90:796)
==2308066==    by 0x25A98D4: __mom_cap_mod_MOD_ocean_model_finalize (mom_cap.F90:2326)

In all three cases, the error is caused by passing either an unassociated pointer array or an unallocated allocatable array as an actual argument to a subroutine that has a corresponding dummy argument assumed shape intent(in) array.

In first two cases (mpp_gather and mpp_scatter) the simplest solution is to allocate a zero-sized array (data3D).

diff --git a/mpp/include/mpp_gather.fh b/mpp/include/mpp_gather.fh
index 6c40dc4e..7391eb3b 100644
--- a/mpp/include/mpp_gather.fh
+++ b/mpp/include/mpp_gather.fh
@@ -132,7 +132,8 @@ subroutine MPP_GATHER_PELIST_GEN_2D_(is, ie, js, je, pelist, array_seg, gather_d
    if (is_root_pe) then
      data3D(1:size(gather_data,1),1:size(gather_data,2),1:1) => gather_data
    else
-     data3D => null()
+     ! data3D => null()
+     allocate(data3D(0,0,0))
    endif

    call mpp_gather(is, ie, js, je, 1, pelist, arr3D, data3D, [dim_order, 3], is_root_pe, &

diff --git a/mpp/include/mpp_scatter.fh b/mpp/include/mpp_scatter.fh
index b0e5c5d1..4ca0edc0 100644
--- a/mpp/include/mpp_scatter.fh
+++ b/mpp/include/mpp_scatter.fh
@@ -57,7 +57,8 @@ subroutine MPP_SCATTER_PELIST_GEN_2D_(is, ie, js, je, pelist, array_seg, input_d
    if (is_root_pe) then
      data3D(1:size(input_data,1),1:size(input_data,2),1:1) => input_data
    else
-     data3D => null()
+     ! data3D => null()
+     allocate(data3D(0,0,0))
    endif

    call mpp_scatter(is, ie, js, je, 1, pelist, arr3D, data3D, [dim_order, 3], is_root_pe)

In the third case we can just skip calling the subroutine if the number of attributes is zero.

diff --git a/diag_manager/diag_output.F90 b/diag_manager/diag_output.F90
index cf8cd4f8..5f9f924e 100644
--- a/diag_manager/diag_output.F90
+++ b/diag_manager/diag_output.F90
@@ -365,10 +365,12 @@ CONTAINS
        endif

        !> Write additional axis attributes, from diag_axis_add_attribute calls
-       CALL write_attribute_meta(file_unit, num_attributes, attributes, err_msg, varname=axis_name, fileob=fileob)
-       IF ( LEN_TRIM(err_msg) .GT. 0 ) THEN
-          CALL error_mesg('diag_output_mod::write_axis_meta_data', TRIM(err_msg), FATAL)
-       END IF
+       if (num_attributes > 0) then
+          CALL write_attribute_meta(file_unit, num_attributes, attributes, err_msg, varname=axis_name, fileob=fileob)
+          IF ( LEN_TRIM(err_msg) .GT. 0 ) THEN
+             CALL error_mesg('diag_output_mod::write_axis_meta_data', TRIM(err_msg), FATAL)
+          END IF
+       endif

        !> Write additional attribute (calendar_type) for time axis ----
        !> @note calendar attribute is compliant with CF convention

[bensonr]

@DusanJovic-NOAA - can you confirm which version of FMS you are using? We may already have found and fixed it - or have a fix ready for release.

[DusanJovic-NOAA]

@DusanJovic-NOAA - can you confirm which version of FMS you are using? We may already have found and fixed it - or have a fix ready for release.

@bensonr We currently use 2024.1 in ufs, but I tried with the top of the main branch, as of day or two ago, and found the same issue.