MAPL2 OpenMP Test fails with NAG on Linux

[mathomp4]

With the introduction of the OpenMP tests from @aoloso in #3401, my NAG MAPL builds on discover are failing with:

The following tests FAILED:
	 75 - GetHorzIJIndex_NO_OMP (Failed)
	 76 - GetHorzIJIndex_OMP_1_thread (Failed)
	 77 - GetHorzIJIndex_OMP_4_thread (Failed)
Errors while running CTest

The NoOMP seems to die with:

    Start 75: GetHorzIJIndex_NO_OMP
1/3 Test #75: GetHorzIJIndex_NO_OMP ............***Failed    1.23 sec
      MAPL: No configure file specified for logging layer.  Using defaults.
 Starting pFIO input server on Clients
 Starting pFIO output server on Clients
      MAPL: Running with MOAB library for ESMF Mesh: F
     SHMEM: NumCores per Node = 6
     SHMEM: NumNodes in use   = 1
     SHMEM: Total PEs         = 6
     SHMEM: NumNodes in use  = 1
 Integer*4 Resource Parameter: BEG_DATE:20070801 0
 Integer*4 Resource Parameter: END_DATE:29990302 210000
pe=00005 FAIL at line=00318    MAPL_CapGridComp.F90                     <status=51>
pe=00005 FAIL at line=00856    MAPL_CapGridComp.F90                     <status=51>
pe=00005 FAIL at line=00349    MAPL_Cap.F90                             <status=51>
pe=00005 FAIL at line=00258    MAPL_Cap.F90                             <status=51>
pe=00005 FAIL at line=00192    MAPL_Cap.F90                             <status=51>
pe=00005 FAIL at line=00169    MAPL_Cap.F90                             <status=51>
pe=00005 FAIL at line=00018    driver.F90                               <status=51>

and the other seem to fail on the same line.

I'm testing now on Bucy to see if it happens there.

[mathomp4]

Odd thing is. That line it's not happy with is "find HEARTBEAT_DT"

[mathomp4]

Confirmed that it fails on Bucy as well:

The following tests FAILED:
	 75 - GetHorzIJIndex_NO_OMP (Failed)
	 76 - GetHorzIJIndex_OMP_1_thread (Failed)
	 77 - GetHorzIJIndex_OMP_4_thread (Failed)

So this is at least NAG-on-Linux.

[mathomp4]

Discussed with @bena-nasa and it looks like NAG is somehow getting the wrong cap pointer. Will definitely need @tclune help with this.

[tclune]

My first thought is that we compile several bits of Baselibs (GFE) without OpenMP for NAG. So maybe NAG fundamentally cannot handle this test. OTOH, the NO_OMP case ought to work even then, right?