Enable no-cpython-wrapper in numba where possible (#765)

* Enable no-cpython-wrapper in numba where possible

* Fix test with no_cpython_wrapper

* Add docstring to numba_funcify

pytensor/link/numba/dispatch/basic.py

@@ -59,6 +59,8 @@ def global_numba_func(func):
 
 def numba_njit(*args, **kwargs):
     kwargs.setdefault("cache", config.numba__cache)
+    kwargs.setdefault("no_cpython_wrapper", True)
+    kwargs.setdefault("no_cfunc_wrapper", True)
 
     # Supress caching warnings
     warnings.filterwarnings(
@@ -419,7 +421,12 @@ def perform(*inputs):
 
 @singledispatch
 def numba_funcify(op, node=None, storage_map=None, **kwargs):
-    """Generate a numba function for a given op and apply node."""
+    """Generate a numba function for a given op and apply node.
+
+    The resulting function will usually use the `no_cpython_wrapper`
+    argument in numba, so it can not be called directly from python,
+    but only from other jit functions.
+    """
     return generate_fallback_impl(op, node, storage_map, **kwargs)
 
 

pytensor/link/numba/dispatch/elemwise.py

@@ -470,7 +470,9 @@ def axis_apply_fn(x):
         "afn",  # Approximate functions
         "reassoc",
         "nsz",  # TODO Do we want this one?
-    }
+    },
+    "no_cpython_wrapper": True,
+    "no_cfunc_wrapper": True,
 }
 
 
@@ -698,7 +700,14 @@ def elemwise(*inputs):
             return tuple(outputs_summed)
         return outputs_summed[0]
 
-    @overload(elemwise)
+    @overload(
+        elemwise,
+        jit_options={
+            "fastmath": flags,
+            "no_cpython_wrapper": True,
+            "no_cfunc_wrapper": True,
+        },
+    )
     def ov_elemwise(*inputs):
         return elemwise_wrapper
 

pytensor/link/numba/linker.py

@@ -29,7 +29,7 @@ def fgraph_convert(self, fgraph, **kwargs):
     def jit_compile(self, fn):
         from pytensor.link.numba.dispatch.basic import numba_njit
 
-        jitted_fn = numba_njit(fn)
+        jitted_fn = numba_njit(fn, no_cpython_wrapper=False, no_cfunc_wrapper=False)
         return jitted_fn
 
     def create_thunk_inputs(self, storage_map):

tests/link/numba/test_tensor_basic.py

@@ -386,6 +386,8 @@ def test_ExtractDiag(val, offset):
 )
 @pytest.mark.parametrize("reverse_axis", (False, True))
 def test_ExtractDiag_exhaustive(k, axis1, axis2, reverse_axis):
+    from pytensor.link.numba.dispatch.basic import numba_njit
+
     if reverse_axis:
         axis1, axis2 = axis2, axis1
 
@@ -394,7 +396,12 @@ def test_ExtractDiag_exhaustive(k, axis1, axis2, reverse_axis):
     x_test = np.arange(np.prod(x_shape)).reshape(x_shape)
     out = pt.diagonal(x, k, axis1, axis2)
     numba_fn = numba_funcify(out.owner.op, out.owner)
-    np.testing.assert_allclose(numba_fn(x_test), np.diagonal(x_test, k, axis1, axis2))
+
+    @numba_njit(no_cpython_wrapper=False)
+    def wrap(x):
+        return numba_fn(x)
+
+    np.testing.assert_allclose(wrap(x_test), np.diagonal(x_test, k, axis1, axis2))
 
 
 @pytest.mark.parametrize(