Add a Numba OpFromGraph implementation

aesara/link/numba/dispatch/basic.py

@@ -17,6 +17,7 @@ from numba.cpython.unsafe.tuple import tuple_setitem  # noqa: F401
 from numba.extending import box
 
 from aesara import config
+from aesara.compile.builders import OpFromGraph
 from aesara.compile.ops import DeepCopyOp
 from aesara.graph.basic import Apply, NoParams
 from aesara.graph.fg import FunctionGraph
@@ -374,6 +375,25 @@ def numba_funcify(op, node=None, storage_map=None, **kwargs):
     return perform
 
 
+@numba_funcify.register(OpFromGraph)
+def numba_funcify_OpFromGraph(op, node=None, **kwargs):
+    fgraph_fn = numba_njit(numba_funcify(op.fgraph, **kwargs))
+
+    if len(op.fgraph.outputs) == 1:
+
+        @numba_njit
+        def opfromgraph(*inputs):
+            return fgraph_fn(*inputs)[0]
+
+    else:
+
+        @numba_njit
+        def opfromgraph(*inputs):
+            return fgraph_fn(*inputs)
+
+    return opfromgraph
+
+
 @numba_funcify.register(FunctionGraph)
 def numba_funcify_FunctionGraph(
     fgraph,

tests/link/numba/test_basic.py

@@ -12,6 +12,7 @@ import aesara.scalar.math as aesm
 import aesara.tensor as at
 import aesara.tensor.math as aem
 from aesara import config, shared
+from aesara.compile.builders import OpFromGraph
 from aesara.compile.function import function
 from aesara.compile.mode import Mode
 from aesara.compile.ops import ViewOp
@@ -1003,3 +1004,18 @@ def test_scalar_return_value_conversion():
         mode=numba_mode,
     )
     assert isinstance(x_fn(1.0), np.ndarray)
+
+
+def test_OpFromGraph():
+    x, y, z = at.matrices("xyz")
+    ofg_1 = OpFromGraph([x, y], [x + y], inline=False)
+    ofg_2 = OpFromGraph([x, y], [x * y, x - y], inline=False)
+
+    o1, o2 = ofg_2(y, z)
+    out = ofg_1(x, o1) + o2
+
+    xv = np.ones((2, 2), dtype=config.floatX)
+    yv = np.ones((2, 2), dtype=config.floatX) * 3
+    zv = np.ones((2, 2), dtype=config.floatX) * 5
+
+    compare_numba_and_py(((x, y, z), (out,)), [xv, yv, zv])