[Pass] Support lift constants to initializers pass

onnxscript/ir/_core.py

@@ -1924,6 +1924,8 @@ def __init__(
         # Be sure the initialize the name authority before extending the nodes
         # because it is used to name the nodes and their outputs
         self._name_authority = _name_authority.NameAuthority()
+        # TODO(justinchuby): Trigger again if inputs or initializers are modified.
+        self._set_input_and_initializer_value_names_into_name_authority()
         # Call self.extend not self._nodes.extend so the graph reference is added to the nodes
         self.extend(nodes)
 
@@ -1999,6 +2001,12 @@ def __iter__(self) -> Iterator[Node]:
     def __reversed__(self) -> Iterator[Node]:
         return reversed(self._nodes)
 
+    def _set_input_and_initializer_value_names_into_name_authority(self):
+        for value in self.inputs:
+            self._name_authority.register_or_name_value(value)
+        for value in self.initializers.values():
+            self._name_authority.register_or_name_value(value)
+
     def _set_node_graph_to_self_and_assign_names(self, node: Node) -> Node:
         """Set the graph reference for the node and assign names to it and its outputs if they don't have one."""
         if node.graph is not None and node.graph is not self:

onnxscript/ir/passes/common/constant_manipulation.py

@@ -0,0 +1,100 @@
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT License.
+"""Lift constants to initializers."""
+
+from __future__ import annotations
+
+__all__ = [
+    "LiftConstantsToInitializersPass",
+]
+
+import logging
+
+import numpy as np
+
+from onnxscript import ir
+
+logger = logging.getLogger(__name__)
+
+
+class LiftConstantsToInitializersPass(ir.passes.InPlacePass):
+    def call(self, model: ir.Model) -> ir.passes.PassResult:
+        """Convert constant nodes in main graph to initializers."""
+        count = 0
+        for node in model.graph:
+            if node.op_type != "Constant" or node.domain not in ("", "onnx.ai"):
+                continue
+
+            allowed_constant_attributes = {
+                "value",
+                "value_int",
+                "value_ints",
+                "value_float",
+                "value_floats",
+                "value_string",
+                "value_strings",
+            }
+            constant_node_attribute = set(node.attributes.keys())
+            if len(constant_node_attribute) != 1:
+                logger.debug(
+                    "Invalid constant node '%s' has more than one attribute", node.name
+                )
+                continue
+            if constant_node_attribute not in allowed_constant_attributes:
+                logger.debug("Invalid constant node '%s' has unsupported attribute", node.name)
+                continue
+
+            initializer_name = node.outputs[0].name
+            assert initializer_name is not None
+            # The value of attribute can only be ir.Attr, as
+            # ir.RefAttr is only defined in Functions.
+            attr_value = node.attributes[constant_node_attribute]
+            if constant_node_attribute == "value":
+                tensor = attr_value.as_tensor()  # type: ignore[union-attr]
+            elif constant_node_attribute == "value_int":
+                tensor = ir.Tensor(
+                    np.array(attr_value.as_int(), dtype=np.int64), name=initializer_name
+                )
+            elif constant_node_attribute == "value_ints":
+                tensor = ir.Tensor(
+                    np.array(attr_value.as_ints(), dtype=np.int64), name=initializer_name
+                )
+            elif constant_node_attribute == "value_float":
+                tensor = ir.Tensor(
+                    np.array(attr_value.as_float(), dtype=np.float32), name=initializer_name
+                )
+            elif constant_node_attribute == "value_floats":
+                tensor = ir.Tensor(
+                    np.array(attr_value.as_floats(), dtype=np.float32), name=initializer_name
+                )
+            elif constant_node_attribute == "value_string":
+                tensor = ir.Tensor(
+                    np.array(attr_value.as_string(), dtype=np.object_), name=initializer_name
+                )
+            elif constant_node_attribute == "value_strings":
+                tensor = ir.Tensor(
+                    np.array(attr_value.as_strings(), dtype=np.object_), name=initializer_name
+                )
+            else:
+                logger.debug("Invalid constant node '%s' has unsupported attribute", node.name)
+                continue
+            # Register an initializer with the tensor value
+            initializer = ir.Value(
+                name=initializer_name,
+                shape=tensor.shape,  # type: ignore[arg-type]
+                type=ir.TensorType(tensor.dtype),
+                const_value=tensor,
+            )
+            # TODO(titaiwang): Is it possible that the initializer name has
+            # been taken?
+            model.graph.register_initializer(initializer)
+            # Replace the constant node with the initilizer
+            ir.convenience.replace_all_uses_with(node.outputs[0], initializer)
+            model.graph.remove(node, safe=True)
+            count += 1
+            logger.info(
+                "Converted constant node '%s' to initializer '%s'", node.name, initializer_name
+            )
+        if count:
+            logger.info("Lifted %s constants to initializers", count)
+        return ir.passes.PassResult(model, modified=bool(count))

onnxscript/ir/passes/common/constant_manipulation_test.py

@@ -0,0 +1,62 @@
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT License.
+from __future__ import annotations
+
+import unittest
+
+import numpy as np
+
+from onnxscript import ir
+from onnxscript.ir.passes.common import constant_manipulation
+
+
+class TestLiftConstantsToInitializersPass(unittest.TestCase):
+    def test_pass_with_lifting_constants_to_initializers(self):
+        inputs = [
+            ir.Value(
+                name="input_a", type=ir.TensorType(ir.DataType.FLOAT), shape=ir.Shape((2, 3))
+            ),
+            ir.Value(
+                name="input_b",
+                type=ir.TensorType(ir.DataType.FLOAT),
+                shape=ir.Shape((2, 3)),
+            ),
+        ]
+
+        constant_tensor = ir.tensor(np.random.rand(2, 3).astype(np.float32))
+        attribute = ir.convenience.convert_attributes({"value": constant_tensor})
+        const_node = ir.Node("", "Constant", inputs=[], attributes=attribute, num_outputs=1)
+        add_node = ir.Node("", "Add", inputs=[inputs[0], const_node.outputs[0]])
+        mul_node = ir.Node("", "Mul", inputs=[add_node.outputs[0], inputs[1]])
+
+        model = ir.Model(
+            graph=ir.Graph(
+                inputs=inputs,
+                outputs=mul_node.outputs,
+                nodes=[const_node, add_node, mul_node],
+                opset_imports={"": 20},
+            ),
+            ir_version=10,
+        )
+
+        # Check that the initializer is not in the graph yet
+        self.assertEqual(len(model.graph.initializers), 0)
+        # And 1 constant node
+        self.assertEqual(len([node for node in model.graph if node.op_type == "Constant"]), 1)
+
+        # Perform lift constants to initializers
+        result = constant_manipulation.LiftConstantsToInitializersPass()(model)
+        self.assertTrue(result.modified)
+        # Check that the constant node is lifted to an initializer
+        self.assertEqual(len(result.model.graph.initializers), 1)
+        # Check the value
+        self.assertEqual(
+            result.model.graph.initializers[
+                "val_0"
+            ].const_value,  # name created by name_authority
+            constant_tensor,
+        )
+        # And 0 constant node
+        self.assertEqual(
+            len([node for node in result.model.graph if node.op_type == "Constant"]), 0
+        )

onnxscript/optimizer/_optimizer.py

@@ -4,6 +4,7 @@
 
 import logging
 
+import onnxscript.ir.passes.common.constant_manipulation
 import onnxscript.ir.passes.common.unused_removal
 import onnxscript.optimizer
 from onnxscript import ir, rewriter
@@ -70,6 +71,7 @@ def optimize_ir(
             early_stop=stop_if_no_change,
         ),
         onnxscript.ir.passes.common.unused_removal.RemoveUnusedNodesPass(),
+        onnxscript.ir.passes.common.constant_manipulation.LiftConstantsToInitializersPass(),
     )
     assert optimizer_pass.in_place
     result = optimizer_pass(model)