Move precision definition to ndsl.dsl.__init__

Detect bad order of imports, e.g., gt4py import before NDSL
Adapt code to use the newly defined NDSL_GLOBAL_PRECISION
Set internal GT4Py literal precision definition

ndsl/__init__.py

@@ -1,3 +1,4 @@
+from . import dsl  # isort:skip
 from .comm.communicator import CubedSphereCommunicator, TileCommunicator
 from .comm.local_comm import LocalComm
 from .comm.mpi import MPIComm

ndsl/dsl/__init__.py

@@ -1,6 +1,23 @@
-import gt4py.cartesian.config
+# Literal precision for both GT4Py & NDSL
+import os
+import sys
 
-from ndsl.comm.mpi import MPI
+
+gt4py_config_module = "gt4py.cartesian.config"
+if gt4py_config_module in sys.modules:
+    raise RuntimeError(
+        "`GT4Py` config imported before `ndsl` imported."
+        " Please import `ndsl.dsl` or any `ndsl` module "
+        " before any `gt4py` imports."
+    )
+NDSL_GLOBAL_PRECISION = int(os.getenv("PACE_FLOAT_PRECISION", "64"))
+os.environ["GT4PY_LITERAL_PRECISION"] = str(NDSL_GLOBAL_PRECISION)
+
+
+# Set cache names for default gt backends workflow
+import gt4py.cartesian.config  # noqa: E402
+
+from ndsl.comm.mpi import MPI  # noqa: E402
 
 
 if MPI is not None:
@@ -9,5 +26,3 @@
     gt4py.cartesian.config.cache_settings["dir_name"] = os.environ.get(
         "GT_CACHE_DIR_NAME", f".gt_cache_{MPI.COMM_WORLD.Get_rank():06}"
     )
-
-__version__ = "0.2.0"

ndsl/dsl/dace/dace_config.py

@@ -8,10 +8,10 @@
 
 from ndsl.comm.communicator import Communicator
 from ndsl.comm.partitioner import Partitioner
+from ndsl.dsl import NDSL_GLOBAL_PRECISION
 from ndsl.dsl.caches.cache_location import identify_code_path
 from ndsl.dsl.caches.codepath import FV3CodePath
 from ndsl.dsl.gt4py_utils import is_gpu_backend
-from ndsl.dsl.typing import floating_point_precision
 from ndsl.optional_imports import cupy as cp
 
 
@@ -264,7 +264,7 @@ def __init__(
                 "compiler", "cuda", "syncdebug", value=dace_debug_env_var
             )
 
-            if floating_point_precision() == 32:
+            if NDSL_GLOBAL_PRECISION == 32:
                 # When using 32-bit float, we flip the default dtypes to be all
                 # C, e.g. 32 bit.
                 dace.Config.set(

ndsl/dsl/typing.py

@@ -1,9 +1,10 @@
-import os
-from typing import Tuple, Union, cast
+from typing import Tuple, TypeAlias, Union, cast
 
 import gt4py.cartesian.gtscript as gtscript
 import numpy as np
 
+from ndsl.dsl import NDSL_GLOBAL_PRECISION
+
 
 # A Field
 Field = gtscript.Field
@@ -21,36 +22,30 @@
 # Union of valid data types (from gt4py.cartesian.gtscript)
 DTypes = Union[bool, np.bool_, int, np.int32, np.int64, float, np.float32, np.float64]
 
-
-def floating_point_precision() -> int:
-    return int(os.getenv("PACE_FLOAT_PRECISION", "64"))
-
-
 # We redefine the type as a way to distinguish
 # the model definition of a float to other usage of the
 # common numpy type in the rest of the code.
-NDSL_32BIT_FLOAT_TYPE = np.float32
-NDSL_64BIT_FLOAT_TYPE = np.float64
+NDSL_32BIT_FLOAT_TYPE: TypeAlias = np.float32
+NDSL_32BIT_INT_TYPE: TypeAlias = np.int32
+NDSL_64BIT_FLOAT_TYPE: TypeAlias = np.float64
+NDSL_64BIT_INT_TYPE: TypeAlias = np.int64
 
 
-def global_set_floating_point_precision():
+def global_set_floating_point_precision() -> Tuple[TypeAlias, TypeAlias]:
     """Set the global floating point precision for all reference
     to Float in the codebase. Defaults to 64 bit."""
-    global Float
-    precision_in_bit = floating_point_precision()
-    if precision_in_bit == 64:
-        return NDSL_64BIT_FLOAT_TYPE
-    elif precision_in_bit == 32:
-        return NDSL_32BIT_FLOAT_TYPE
-    else:
-        NotImplementedError(
-            f"{precision_in_bit} bit precision not implemented or tested"
-        )
+    global Float, Int
+    if NDSL_GLOBAL_PRECISION == 64:
+        return NDSL_64BIT_FLOAT_TYPE, NDSL_64BIT_INT_TYPE
+    elif NDSL_GLOBAL_PRECISION == 32:
+        return NDSL_32BIT_FLOAT_TYPE, NDSL_32BIT_INT_TYPE
+    raise NotImplementedError(
+        f"{NDSL_GLOBAL_PRECISION} bit precision not implemented or tested"
+    )
 
 
 # Default float and int types
-Float = global_set_floating_point_precision()
-Int = np.int_
+Float, Int = global_set_floating_point_precision()
 Bool = np.bool_
 
 FloatField = Field[gtscript.IJK, Float]