Create and use dpnp specific ufunc db

numba_dpex/core/targets/dpjit_target.py

@@ -14,6 +14,8 @@
 from numba.core.imputils import Registry
 from numba.core.target_extension import CPU, target_registry
 
+from numba_dpex.dpnp_iface import dpnp_ufunc_db
+
 
 class Dpex(CPU):
     pass
@@ -57,3 +59,7 @@ def load_additional_registries(self):
 
         # loading CPU specific registries
         super().load_additional_registries()
+
+    # TODO: do we need it?
+    def get_ufunc_info(self, ufunc_key):
+        return dpnp_ufunc_db.get_ufunc_info(ufunc_key)

numba_dpex/core/targets/kernel_target.py

@@ -251,21 +251,20 @@ def init(self):
         self._target_data = ll.create_target_data(
             codegen.SPIR_DATA_LAYOUT[utils.MACHINE_BITS]
         )
-        # Override data model manager to SPIR model
-        import numba.cpython.unicode
 
+        # Override data model manager to SPIR model
         self.data_model_manager = _init_data_model_manager()
         self.extra_compile_options = dict()
 
-        import copy
+        from numba_dpex.dpnp_iface.dpnp_ufunc_db import _lazy_init_dpnp_db
 
-        from numba.np.ufunc_db import _lazy_init_db
+        _lazy_init_dpnp_db()
 
-        _lazy_init_db()
-        from numba.np.ufunc_db import _ufunc_db as ufunc_db
+        # we need to import it after, because before init it is None and
+        # variable is passed by value
+        from numba_dpex.dpnp_iface.dpnp_ufunc_db import _dpnp_ufunc_db
 
-        self.ufunc_db = copy.deepcopy(ufunc_db)
-        self.cpu_context = cpu_target.target_context
+        self.ufunc_db = _dpnp_ufunc_db
 
     def create_module(self, name):
         return self._internal_codegen._create_empty_module(name)

numba_dpex/dpnp_iface/dpnp_ufunc_db.py

@@ -3,6 +3,8 @@
 # SPDX-License-Identifier: Apache-2.0
 
 
+import copy
+
 import dpnp
 import numpy as np
 from numba.core import types
@@ -11,23 +13,50 @@
 
 from ..ocl import mathimpl
 
+# A global instance of dpnp ufuncs that are supported by numba-dpex
+_dpnp_ufunc_db = None
+
+
+def _lazy_init_dpnp_db():
+    global _dpnp_ufunc_db
+
+    if _dpnp_ufunc_db is None:
+        _dpnp_ufunc_db = {}
+        _fill_ufunc_db_with_dpnp_ufuncs(_dpnp_ufunc_db)
+
 
 def get_ufuncs():
-    """obtain a list of supported ufuncs in the db"""
+    """Returns the list of supported dpnp ufuncs in the _dpnp_ufunc_db"""
 
-    from numba.np.ufunc_db import _lazy_init_db
+    _lazy_init_dpnp_db()
 
-    _lazy_init_db()
-    from numba.np.ufunc_db import _ufunc_db
+    return _dpnp_ufunc_db.keys()
 
-    _fill_ufunc_db_with_dpnp_ufuncs(_ufunc_db)
 
-    return _ufunc_db.keys()
+def get_ufunc_info(ufunc_key):
+    """get the lowering information for the ufunc with key ufunc_key.
+
+    The lowering information is a dictionary that maps from a numpy
+    loop string (as given by the ufunc types attribute) to a function
+    that handles code generation for a scalar version of the ufunc
+    (that is, generates the "per element" operation").
+
+    raises a KeyError if the ufunc is not in the ufunc_db
+    """
+    _lazy_init_dpnp_db()
+    return _dpnp_ufunc_db[ufunc_key]
 
 
 def _fill_ufunc_db_with_dpnp_ufuncs(ufunc_db):
-    """Monkey patching dpnp for missing attributes."""
-    # FIXME: add more docstring
+    """Populates the _dpnp_ufunc_db from Numba's NumPy ufunc_db"""
+
+    from numba.np.ufunc_db import _lazy_init_db
+
+    _lazy_init_db()
+
+    # we need to import it after, because before init it is None and
+    # variable is passed by value
+    from numba.np.ufunc_db import _ufunc_db
 
     for ufuncop in dpnpdecl.supported_ufuncs:
         if ufuncop == "erf":
@@ -52,7 +81,12 @@ def _fill_ufunc_db_with_dpnp_ufuncs(ufunc_db):
             op.nargs = npop.nargs
             op.types = npop.types
             op.is_dpnp_ufunc = True
-            ufunc_db.update({op: ufunc_db[npop]})
+            cp = copy.copy(_ufunc_db[npop])
+            if "'divide'" in str(npop):
+                # TODO: why do we need to do it only for divide?
+                # https://github.com/IntelPython/numba-dpex/issues/1270
+                ufunc_db.update({npop: cp})
+            ufunc_db.update({op: cp})
             for key in list(ufunc_db[op].keys()):
                 if (
                     "FF->" in key