fix NaN issue without skipping and retry each test 3 times in single precision

Author: Archith Iyer

src/common/m_checker_common.fpp

@@ -18,7 +18,7 @@ module m_checker_common
 
     implicit none
 
-    private; public :: s_check_inputs_common
+    private; public :: s_check_inputs_common, wp
 
 contains
 

src/common/m_precision_select.f90

@@ -16,6 +16,9 @@ module m_precision_select
     integer, parameter :: single_precision = selected_real_kind(6, 37)
     integer, parameter :: double_precision = selected_real_kind(15, 307)
 
+    integer, parameter :: sp = single_precision
+    integer, parameter :: dp = double_precision
+
     ! Set the working precision (wp) to single or double precision
 #ifdef MFC_SINGLE_PRECISION
     integer, parameter :: wp = single_precision  ! Change to single_precision if needed

src/post_process/m_data_output.fpp

@@ -65,6 +65,7 @@ module m_global_parameters
     !> @name Cell-center locations in the x-, y- and z-coordinate directions
     !> @{
     real(wp), allocatable, dimension(:) :: x_cc, x_root_cc, y_cc, z_cc
+    real(sp), allocatable, dimension(:) :: x_root_cc_s, x_cc_s
     !> @}
 
     !> Cell-width distributions in the x-, y- and z-coordinate directions
@@ -700,6 +701,8 @@ contains
                     allocate (z_cb_s(-1 - offset_x%beg:m + offset_x%end))
                 end if
             end if
+        else
+            allocate (x_cc_s(-buff_size:m + buff_size))
         end if
 
         ! Allocating the grid variables in the x-coordinate direction
@@ -727,6 +730,10 @@ contains
             allocate (x_root_cb(-1:m_root))
             allocate (x_root_cc(0:m_root))
 
+            if (precision == 1) then
+                allocate (x_root_cc_s(0:m_root))
+            end if
+
         end if
 
         allocate (adv(num_fluids))

src/post_process/m_global_parameters.fpp

@@ -1414,7 +1414,7 @@ contains
         !!  the second dimension corresponds to the processor rank.
     subroutine s_mpi_gather_spatial_extents(spatial_extents)
 
-        real(wp), dimension(1:, 0:), intent(inout) :: spatial_extents
+        real(kind(0d0)), dimension(1:, 0:), intent(INOUT) :: spatial_extents
 
 #ifdef MFC_MPI
 
@@ -1600,20 +1600,20 @@ contains
     subroutine s_mpi_defragment_1d_flow_variable(q_sf, q_root_sf)
 
         real(wp), &
-            dimension(0:m, 0:0, 0:0), &
+            dimension(0:m), &
             intent(in) :: q_sf
 
         real(wp), &
-            dimension(0:m_root, 0:0, 0:0), &
+            dimension(0:m), &
             intent(inout) :: q_root_sf
 
 #ifdef MFC_MPI
 
         ! Gathering the sub-domain flow variable data from all the processes
         ! and putting it back together for the entire computational domain
         ! on the process with rank 0
-        call MPI_GATHERV(q_sf(0, 0, 0), m + 1, mpi_p, &
-                         q_root_sf(0, 0, 0), recvcounts, displs, &
+        call MPI_GATHERV(q_sf(0), m + 1, mpi_p, &
+                         q_root_sf(0), recvcounts, displs, &
                          mpi_p, 0, MPI_COMM_WORLD, ierr)
 
 #endif

src/post_process/m_mpi_proxy.fpp

@@ -212,6 +212,7 @@ def create_directory(self):
     def __str__(self) -> str:
         return f"tests/[bold magenta]{self.get_uuid()}[/bold magenta]: {self.trace}"
 
+    # pylint: disable=global-statement, global-variable-not-assigned, too-many-return-statements
     def compute_tolerance(self) -> float:
         if ARG("single"):
             return 1e-1

toolchain/mfc/test/case.py

@@ -133,8 +133,6 @@ def test():
 # pylint: disable=too-many-locals, too-many-branches, too-many-statements, trailing-whitespace
 def _handle_case(case: TestCase, devices: typing.Set[int]):
     # pylint: disable=global-statement, global-variable-not-assigned
-    global nSKIP
-
     start_time = time.time()
 
     tol = case.compute_tolerance()      
@@ -202,11 +200,7 @@ def _handle_case(case: TestCase, devices: typing.Set[int]):
                 raise MFCException(f"Test {case}: Failed to run h5dump. You can find the run's output in {out_filepath}, and the case dictionary in {case.get_filepath()}.")
 
             if "nan," in output:
-                if not ARG("single"):
-                    raise MFCException(f"Test {case}: Post Process has detected a NaN. You can find the run's output in {out_filepath}, and the case dictionary in {case.get_filepath()}.")
-                cons.print(f"Test {case}: Skipping this test case as it cannot be run in single precision, please build MFC in double precision to run this test")
-                nSKIP += 1
-                return
+                raise MFCException(f"Test {case}: Post Process has detected a NaN. You can find the run's output in {out_filepath}, and the case dictionary in {case.get_filepath()}.")
 
             if "inf," in output:
                 raise MFCException(f"Test {case}: Post Process has detected an Infinity. You can find the run's output in {out_filepath}, and the case dictionary in {case.get_filepath()}.")
@@ -224,6 +218,10 @@ def handle_case(case: TestCase, devices: typing.Set[int]):
     global nFAIL, nPASS, nSKIP
 
     nAttempts = 0
+    if ARG('single'):
+        max_attempts = max(ARG('max_attempts'), 3)
+    else:
+        max_attempts = ARG('max_attempts')
 
     while True:
         nAttempts += 1
@@ -232,8 +230,7 @@ def handle_case(case: TestCase, devices: typing.Set[int]):
             _handle_case(case, devices)
             nPASS += 1
         except Exception as exc:
-            if nAttempts < ARG("max_attempts"):
-                cons.print(f"[bold yellow] Attempt {nAttempts}: Failed test {case.get_uuid()}. Retrying...[/bold yellow]")
+            if nAttempts < max_attempts:
                 continue
             nFAIL += 1
             cons.print(f"[bold red]Failed test {case} after {nAttempts} attempt(s).[/bold red]")