Checkpointer validates on C_SW and D_SW (#320)

* fix variable mismatch

* reinitialize dycore to reset temp storages

* fix var name

* fvdynamics-out tracers in compute domain only

* update data version

* revert comment

* fix driver subset

* clean up translate test

* add comment

* Bump to 8.1.3

* add h5netcdf to requirement

* specify xarray engine

* switch to engine netcdf4

* Revert "switch to engine netcdf4"

This reverts commit 50ea453.

* Revert "specify xarray engine"

This reverts commit 42cfc9b.

* Revert "add h5netcdf to requirement"

This reverts commit 3e9d080.

Co-authored-by: Jeremy McGibbon <[email protected]>

Makefile.data_download

@@ -2,7 +2,7 @@
 REGRESSION_DATA_STORAGE_BUCKET = gs://vcm-fv3gfs-serialized-regression-data
 ROOT_DIR:=$(shell dirname $(realpath $(firstword $(MAKEFILE_LIST))))
 EXPERIMENT ?=c12_6ranks_standard
-FORTRAN_SERIALIZED_DATA_VERSION=8.1.1
+FORTRAN_SERIALIZED_DATA_VERSION=8.1.3
 TARGET ?=dycore
 TEST_DATA_ROOT ?=$(ROOT_DIR)/test_data/
 TEST_DATA_HOST ?=$(TEST_DATA_ROOT)/$(FORTRAN_SERIALIZED_DATA_VERSION)/$(EXPERIMENT)/$(TARGET)

fv3core/pace/fv3core/stencils/dyn_core.py

@@ -705,7 +705,8 @@ def _checkpoint_dsw_in(self, state: DycoreState):
                 ucd=state.uc,
                 vcd=state.vc,
                 wd=state.w,
-                delpcd=self.cgrid_shallow_water_lagrangian_dynamics.delpc,
+                # delpc is a temporary and not a variable in D_SW savepoint
+                delpcd=self._vt,
                 delpd=state.delp,
                 ud=state.u,
                 vd=state.v,
@@ -727,7 +728,7 @@ def _checkpoint_dsw_out(self, state: DycoreState):
                 ucd=state.uc,
                 vcd=state.vc,
                 wd=state.w,
-                delpcd=self.cgrid_shallow_water_lagrangian_dynamics.delpc,
+                delpcd=self._vt,
                 delpd=state.delp,
                 ud=state.u,
                 vd=state.v,

fv3core/pace/fv3core/testing/translate_fvdynamics.py

@@ -15,9 +15,6 @@
 from pace.util.grid import GridData
 
 
-ADVECTED_TRACER_NAMES = utils.tracer_variables[: fv_dynamics.NQ]
-
-
 class TranslateDycoreFortranData2Py(TranslateFortranData2Py):
     def __init__(
         self,
@@ -234,14 +231,14 @@ def __init__(
             "v": grid.x3d_domain_dict(),
             "w": {},
             "delz": {},
-            "qvapor": {},
-            "qliquid": {},
-            "qice": {},
-            "qrain": {},
-            "qsnow": {},
-            "qgraupel": {},
-            "qo3mr": {},
-            "qsgs_tke": {},
+            "qvapor": grid.compute_dict(),
+            "qliquid": grid.compute_dict(),
+            "qice": grid.compute_dict(),
+            "qrain": grid.compute_dict(),
+            "qsnow": grid.compute_dict(),
+            "qgraupel": grid.compute_dict(),
+            "qo3mr": grid.compute_dict(),
+            "qsgs_tke": grid.compute_dict(),
             "qcld": {},
             "ps": {},
             "pe": {
@@ -340,8 +337,6 @@ def compute_parallel(self, inputs, communicator):
         )
         self.dycore.step_dynamics(state, pace.util.NullTimer())
         outputs = self.outputs_from_state(state)
-        for name, value in outputs.items():
-            outputs[name] = self.subset_output(name, value)
         return outputs
 
     def outputs_from_state(self, state: dict):
@@ -364,42 +359,3 @@ def compute_sequential(self, *args, **kwargs):
             f"{self.__class__} only has a mpirun implementation, "
             "not running in mock-parallel"
         )
-
-    def subset_output(self, varname: str, output):
-        """
-        Given an output array, return the slice of the array which we'd
-        like to validate against reference data
-        """
-        if self.dycore is None:
-            raise RuntimeError(
-                "cannot call subset_output before calling compute_parallel "
-                "to initialize dycore"
-            )
-        if hasattr(self.dycore, "selective_names") and (
-            varname in self.dycore.selective_names  # type: ignore
-        ):
-            return_value = self.dycore.subset_output(varname, output)  # type: ignore
-
-        if varname in ADVECTED_TRACER_NAMES:
-
-            def get_compute_domain_k_interfaces(
-                instance,
-            ):
-                try:
-                    origin = instance.grid_indexing.origin_compute()
-                    domain = instance.grid_indexing.domain_compute(add=(0, 0, 1))
-                except AttributeError:
-                    origin = instance.grid.compute_origin()
-                    domain = instance.grid.domain_shape_compute(add=(0, 0, 1))
-                return origin, domain
-
-            origin, domain = get_compute_domain_k_interfaces(
-                self.dycore.tracer_advection
-            )
-            self._validation_slice = tuple(
-                slice(start, start + n) for start, n in zip(origin, domain)
-            )
-            return_value = output[self._validation_slice]
-        else:
-            return_value = output
-        return return_value

physics/tests/savepoint/translate/translate_driver.py

@@ -79,6 +79,4 @@ def compute_parallel(self, inputs, communicator):
         self.dycore = driver.dycore
 
         outputs = self.outputs_from_state(driver.state.dycore_state)
-        for name, value in outputs.items():
-            outputs[name] = self.subset_output(name, value)
         return outputs

tests/mpi/test_checkpoints.py

@@ -1,5 +1,6 @@
 import dataclasses
 import os
+from datetime import timedelta
 from typing import List, Tuple
 
 import dacite
@@ -137,6 +138,7 @@ def test_fv_dynamics(
         phis=state.phis,
         state=state,
         checkpointer=validation,
+        timestep=timedelta(seconds=dycore_config.dt_atmos),
     )
     with validation.trial():
         dycore.step_dynamics(state)
@@ -162,6 +164,7 @@ def _calibrate_thresholds(
         phis=state.phis,
         state=state,
         checkpointer=calibration,
+        timestep=timedelta(seconds=dycore_config.dt_atmos),
     )
     for i in range(n_trials):
         print(f"running calibration trial {i}")