Implement vectorize_node dispatch for some forms of Join

Author: ricardoV94

pytensor/tensor/basic.py

@@ -23,7 +23,7 @@
 from pytensor import scalar as ps
 from pytensor.gradient import DisconnectedType, grad_undefined
 from pytensor.graph import RewriteDatabaseQuery
-from pytensor.graph.basic import Apply, Constant, Variable
+from pytensor.graph.basic import Apply, Constant, Variable, equal_computations
 from pytensor.graph.fg import FunctionGraph
 from pytensor.graph.op import Op
 from pytensor.graph.replace import _vectorize_node
@@ -42,7 +42,7 @@
     as_tensor_variable,
     get_vector_length,
 )
-from pytensor.tensor.blockwise import Blockwise
+from pytensor.tensor.blockwise import Blockwise, vectorize_node_fallback
 from pytensor.tensor.elemwise import (
     DimShuffle,
     Elemwise,
@@ -2662,6 +2662,36 @@ def join(axis, *tensors_list):
         return join_(axis, *tensors_list)
 
 
+@_vectorize_node.register(Join)
+def vectorize_join(op: Join, node, batch_axis, *batch_inputs):
+    original_axis, *old_inputs = node.inputs
+    # We can vectorize join as a shifted axis on the batch inputs if:
+    # 1. The batch axis is a constant and has not changed
+    # 2. All inputs are batched with the same broadcastable pattern
+    if (
+        original_axis.type.ndim == 0
+        and isinstance(original_axis, Constant)
+        and equal_computations([original_axis], [batch_axis])
+    ):
+        batch_ndims = {
+            batch_input.type.ndim - old_input.type.ndim
+            for batch_input, old_input in zip(batch_inputs, old_inputs)
+        }
+        if len(batch_ndims) == 1:
+            [batch_ndim] = batch_ndims
+            batch_bcast = batch_inputs[0].type.broadcastable[:batch_ndim]
+            if all(
+                batch_input.type.broadcastable[:batch_ndim] == batch_bcast
+                for batch_input in batch_inputs[1:]
+            ):
+                original_ndim = node.outputs[0].type.ndim
+                original_axis = normalize_axis_index(original_axis.data, original_ndim)
+                batch_axis = original_axis + batch_ndim
+                return op.make_node(batch_axis, *batch_inputs)
+
+    return vectorize_node_fallback(op, node, batch_axis, *batch_inputs)
+
+
 def roll(x, shift, axis=None):
     """
     Convenience function to roll TensorTypes along the given axis.

tests/tensor/test_basic.py

@@ -10,6 +10,7 @@
 import pytensor.tensor.basic as ptb
 import pytensor.tensor.math as ptm
 from pytensor import compile, config, function, shared
+from pytensor.compile import SharedVariable
 from pytensor.compile.io import In, Out
 from pytensor.compile.mode import Mode, get_default_mode
 from pytensor.compile.ops import DeepCopyOp
@@ -4565,3 +4566,37 @@ def core_np(x):
         vectorize_pt(x_test),
         vectorize_np(x_test),
     )
+
+
+@pytest.mark.parametrize("axis", [constant(1), constant(-2), shared(1)])
+@pytest.mark.parametrize("broadcasting_y", ["none", "implicit", "explicit"])
+@config.change_flags(cxx="")  # C code not needed
+def test_vectorize_join(axis, broadcasting_y):
+    # Signature for join along intermediate axis
+    signature = "(a,b1,c),(a,b2,c)->(a,b,c)"
+
+    def core_pt(x, y):
+        return join(axis, x, y)
+
+    def core_np(x, y):
+        return np.concatenate([x, y], axis=axis.eval())
+
+    x = tensor(shape=(4, 2, 3, 5))
+    y_shape = {"none": (4, 2, 3, 5), "implicit": (2, 3, 5), "explicit": (1, 2, 3, 5)}
+    y = tensor(shape=y_shape[broadcasting_y])
+
+    vectorize_pt = function([x, y], vectorize(core_pt, signature=signature)(x, y))
+
+    blockwise_needed = isinstance(axis, SharedVariable) or broadcasting_y != "none"
+    has_blockwise = any(
+        isinstance(node.op, Blockwise) for node in vectorize_pt.maker.fgraph.apply_nodes
+    )
+    assert has_blockwise == blockwise_needed
+
+    x_test = np.random.normal(size=x.type.shape).astype(x.type.dtype)
+    y_test = np.random.normal(size=y.type.shape).astype(y.type.dtype)
+    vectorize_np = np.vectorize(core_np, signature=signature)
+    np.testing.assert_allclose(
+        vectorize_pt(x_test, y_test),
+        vectorize_np(x_test, y_test),
+    )