feat(rewriter): introduce fuse batchnorm

- Fuses Batchnorm node into the following nodes (Conv, ConvTranspose,
  Gemm)

Author: AyoubMDL

onnxscript/rewriter/__init__.py

@@ -18,6 +18,7 @@
     broadcast_to_matmul,
     cast_constant_of_shape,
     collapse_slices,
+    fuse_batchnorm,
     gemm_to_matmul_add,
     llama_rule_sets,
     no_op,
@@ -28,6 +29,7 @@
 _DEFAULT_REWRITE_RULES: tuple[pattern.RewriteRule, ...] = (
     *no_op.rules.rules,  # TODO: merge this rule into constant folding?
     *broadcast_to_matmul.rules.rules,
+    *fuse_batchnorm.fuse_batchnorm_rule_set().rules,
     gemm_to_matmul_add.rule,  # type: ignore[has-type]
     *cast_constant_of_shape.rules.rules,
     *collapse_slices.rules.rules,

onnxscript/rewriter/fuse_batchnorm.py

@@ -0,0 +1,183 @@
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT License.
+"""Fuses BatchNormalization nodes into preceding nodes. Supported fusion patterns:
+- BatchNormalization + Conv         -> Conv
+- BatchNormalization + ConvTranpose -> ConvTranpose
+- BatchNormalization + Gemm         -> Gemm
+
+Approach:
+    Given an inbound operation output: Y = W * X + B
+    And a BatchNormalization outputs: Y_BN = (gamma * (Y - μ) / std) + β, where std = sqrt(var + eps)
+
+    The fusion updates the inbound weights as follows:
+        - W_fused = W * (gamma / std)
+        - B_fused = (B - μ) * (gamma / std) + β
+"""
+
+from abc import ABC, abstractmethod
+
+import numpy as np
+
+from onnxscript import ir
+from onnxscript.rewriter import pattern as orp
+
+
+class FuseBatchNormBase(orp.RewriteRuleClassBase, ABC):
+    """Interface for BatchNormalization nodes fusion."""
+
+    def __init__(
+        self,
+        op_type: str,
+        name: str | None = None,
+        remove_nodes: bool = True,
+        as_function: bool = False,
+    ) -> None:
+        super().__init__(name=name, remove_nodes=remove_nodes, as_function=as_function)
+        self.op_type = op_type
+
+    @abstractmethod
+    def get_filters_axis(self, attributes) -> int:
+        """Return the axis along which BatchNorm scale should be broadcasted."""
+
+    def _reshape_for_broadcast(self, x: np.ndarray, rank: int, axis: int = 1) -> np.ndarray:
+        # Convert axis to positive
+        if axis < 0:
+            axis += rank
+
+        # Build shape: 1s everywhere except -1 at the target axis
+        broadcast_shape = [1 if axis != i else -1 for i in range(rank)]
+        return np.reshape(x, broadcast_shape)
+
+    def rewrite(self, op, x: ir.Value, inbound_out: ir.Value, batchnorm_out: ir.Value):
+        batchnorm_node = batchnorm_out.producer()
+        # Get BatchNorm parameters
+        gamma, beta, input_mean, input_var = [
+            inp.const_value.numpy() for inp in batchnorm_node.inputs[1:]
+        ]
+
+        # 1e-5 is the default value for epsilon according to
+        # https://onnx.ai/onnx/operators/onnx__BatchNormalization.html#attributes
+        default_eps = ir.Attr("epsilon", ir.AttributeType.FLOAT, 1e-5)
+        eps = batchnorm_node.attributes.get("epsilon", default_eps).as_float()
+
+        # Compute the scale_factor to update the inbound weights and bias
+        scale_factor = gamma / np.sqrt(input_var + eps)
+
+        # Update inbound weights
+        inbound_node = inbound_out.producer()
+        weights = inbound_node.inputs[1].const_value.numpy()
+
+        # Reshape scale factor so it is broadcastable
+        axis = self.get_filters_axis(inbound_node.attributes)
+        fused_weights = ir.tensor(
+            weights * self._reshape_for_broadcast(scale_factor, weights.ndim, axis=axis)
+        )
+
+        # Update bias
+        if len(inbound_node.inputs) > 2:
+            original_bias = inbound_node.inputs[2].const_value.numpy()
+            bias_name = inbound_node.inputs[2].name
+        else:
+            original_bias = np.zeros_like(input_mean)
+            bias_name = x.name + "_bias"
+        fused_bias = ir.tensor((original_bias - input_mean) * scale_factor + beta)
+
+        return op.op(
+            self.op_type,
+            inputs=[
+                x,
+                op.initializer(fused_weights, name=inbound_node.inputs[1].name),
+                op.initializer(fused_bias, name=bias_name),
+            ],
+            attributes=inbound_node.attributes,
+        )
+
+    def check(self, context, x, inbound_out, batchnorm_out) -> orp.MatchResult:
+        del context  # Unused
+        check_result = orp.MatchResult()
+
+        inbound_node = inbound_out.producer()
+        batchnorm_node = batchnorm_out.producer()
+
+        # Check that inbound weights + (inbound bias) + batchnorm params are initializers
+        initializers = [inbound_node.inputs[1], *batchnorm_node.inputs[1:]]
+        if len(inbound_node.inputs) > 2:
+            initializers.append(inbound_node.inputs[2])
+
+        for initializer in initializers:
+            if not initializer.is_initializer() or initializer.const_value is None:
+                return check_result.fail(f"{initializer.name} is not a constant initializer")
+
+        return check_result
+
+
+class FuseBatchNormIntoConv(FuseBatchNormBase):
+    """Replaces ``BatchNormalization(Conv(x))`` with ``Conv(x)``."""
+
+    def __init__(self):
+        super().__init__("Conv")
+
+    def get_filters_axis(self, attributes) -> int:
+        return 0
+
+    def pattern(self, op, x):
+        return op.BatchNormalization(
+            op.Conv(x, _allow_other_inputs=True, _outputs=["inbound_out"]),
+            _allow_other_inputs=True,
+            _outputs=["batchnorm_out"],
+        )
+
+
+class FuseBatchNormIntoConvTranspose(FuseBatchNormBase):
+    """Replaces ``BatchNormalization(ConvTranspose(x))`` with ``ConvTranspose(x)``."""
+
+    def __init__(self):
+        super().__init__("ConvTranspose")
+
+    def get_filters_axis(self, attributes) -> int:
+        return 1
+
+    def pattern(self, op, x):
+        return op.BatchNormalization(
+            op.ConvTranspose(x, _allow_other_inputs=True, _outputs=["inbound_out"]),
+            _allow_other_inputs=True,
+            _outputs=["batchnorm_out"],
+        )
+
+
+class FuseBatchNormIntoGemm(FuseBatchNormBase):
+    """Replaces ``BatchNormalization(Gemm(x))`` with ``Gemm(x)``."""
+
+    def __init__(self):
+        super().__init__("Gemm")
+
+    def get_filters_axis(self, attributes) -> int:
+        return 0 if attributes.get("transB") is not None and attributes["transB"].value else 1
+
+    def pattern(self, op, x):
+        return op.BatchNormalization(
+            op.Gemm(x, _allow_other_inputs=True, _outputs=["inbound_out"]),
+            _allow_other_inputs=True,
+            _outputs=["batchnorm_out"],
+        )
+
+
+fuse_batchnorm_into_conv_rule = FuseBatchNormIntoConv().rule()
+fuse_batchnorm_into_convtranspose_rule = FuseBatchNormIntoConvTranspose().rule()
+fuse_batchnorm_into_gemm_rule = FuseBatchNormIntoGemm().rule()
+
+
+def fuse_batchnorm_rule_set() -> orp.RewriteRuleSet:
+    """Returns a set of rewrite rules that fuse BatchNormalization nodes
+    into preceding nodes such as Conv, ConvTranspose, and Gemm.
+
+    Returns:
+        RewriteRuleSet
+    """
+    return orp.RewriteRuleSet(
+        [
+            fuse_batchnorm_into_conv_rule,
+            fuse_batchnorm_into_convtranspose_rule,
+            fuse_batchnorm_into_gemm_rule,
+        ]
+    )