Pass depletion process time to deplete.Results.save

Method openmc.deplete.Results.save now accepts a new argument,
proc_time, to represent the time each process spent depleting
all materials. This value is summed across all processes to
get the total cpu time depleting. Time in transport simulations
should not be included.

The value written to depletion_results.h5 is
proc_time / (n_proc * n_burn_materials)
to get a look at the average time required to deplete a material.
A new dataset, "depletion time" is created in this file, and has
one fewer rows than other data set, like eigenvalues. This corresponds
to there being one fewer round of depletion simulations than transport.

The proc_time is also read in from the depletion file in the
class method Results.from_hdf5(), with one notable exception.
If data from the last step is requested, then the proc_time
attribute will be a numpy array of nans, reflecting the lack
of depletion data.

openmc/deplete/__init__.py

@@ -10,9 +10,11 @@ try:
     from mpi4py import MPI
     comm = MPI.COMM_WORLD
     have_mpi = True
+    mpi_sum = MPI.SUM
 except ImportError:
     comm = DummyCommunicator()
     have_mpi = False
+    mpi_sum = lambda x: x
 
 from .nuclide import *
 from .chain import *

openmc/deplete/results.py

@@ -10,7 +10,7 @@ from warnings import warn
 import numpy as np
 import h5py
 
-from . import comm, have_mpi
+from . import comm, have_mpi, mpi_sum
 from .reaction_rates import ReactionRates
 
 _VERSION_RESULTS = (1, 0)
@@ -55,6 +55,7 @@ class Results(object):
         self.power = None
         self.rates = None
         self.volume = None
+        self.proc_time = None
 
         self.mat_to_ind = None
         self.nuc_to_ind = None
@@ -245,6 +246,10 @@ class Results(object):
         handle.create_dataset("power", (1, n_stages), maxshape=(None, n_stages),
                               dtype='float64')
 
+        handle.create_dataset(
+            "depletion time", (1, 1), maxshape=(None, 1),
+            dtype="float64")
+
     def _to_hdf5(self, handle, index):
         """Converts results object into an hdf5 object.
 
@@ -268,6 +273,7 @@ class Results(object):
         eigenvalues_dset = handle["/eigenvalues"]
         time_dset = handle["/time"]
         power_dset = handle["/power"]
+        proc_time_dset = handle["/depletion time"]
 
         # Get number of results stored
         number_shape = list(number_dset.shape)
@@ -296,6 +302,13 @@ class Results(object):
             power_shape[0] = new_shape
             power_dset.resize(power_shape)
 
+            # keep depletion time matrix with one less index
+            # corresponding to one fewer calculation
+            if self.proc_time:
+                proc_shape = list(proc_time_dset.shape)
+                proc_shape[0] = new_shape - 1
+                proc_time_dset.resize(proc_shape)
+
         # If nothing to write, just return
         if len(self.mat_to_ind) == 0:
             return
@@ -314,6 +327,10 @@ class Results(object):
         if comm.rank == 0:
             time_dset[index, :] = self.time
             power_dset[index, :] = self.power
+            if self.proc_time is not None:
+                proc_time_dset[index - 1] = (
+                    self.proc_time / (comm.size * self.n_hdf5_mats)
+                )
 
     @classmethod
     def from_hdf5(cls, handle, step):
@@ -324,8 +341,7 @@ class Results(object):
         handle : h5py.File or h5py.Group
             An HDF5 file or group type to load from.
         step : int
-            What step is this?
-
+            Index for depletion step
         """
         results = cls()
 
@@ -334,12 +350,20 @@ class Results(object):
         eigenvalues_dset = handle["/eigenvalues"]
         time_dset = handle["/time"]
         power_dset = handle["/power"]
+        proc_time_dset = handle["/depletion time"]
 
         results.data = number_dset[step, :, :, :]
         results.k = eigenvalues_dset[step, :]
         results.time = time_dset[step, :]
         results.power = power_dset[step, :]
 
+        # depletion time data has one fewer index, corresponding
+        # to one fewer calculation stage
+        if step < proc_time_dset.shape[0]:
+            results.proc_time = proc_time_dset[step, :]
+        else:
+            results.proc_time = np.array([np.nan])
+
         # Reconstruct dictionaries
         results.volume = OrderedDict()
         results.mat_to_ind = OrderedDict()
@@ -375,7 +399,7 @@ class Results(object):
         return results
 
     @staticmethod
-    def save(op, x, op_results, t, power, step_ind):
+    def save(op, x, op_results, t, power, step_ind, proc_time):
         """Creates and writes depletion results to disk
 
         Parameters
@@ -392,6 +416,8 @@ class Results(object):
             Power during time step
         step_ind : int
             Step index.
+        proc_time: float or None
+            time spent depleting materials.
 
         """
         # Get indexing terms
@@ -422,6 +448,9 @@ class Results(object):
         results.rates = [r.rates for r in op_results]
         results.time = t
         results.power = power
+        results.proc_time = proc_time
+        if step_ind and results.proc_time is not None:
+            results.proc_time = comm.reduce(proc_time, op=mpi_sum)
 
         results.export_to_hdf5("depletion_results.h5", step_ind)