[Rewriter] Implement zero bias removal for Conv operations and related rules

onnxscript/rewriter/__init__.py

@@ -19,8 +19,8 @@
 ]
 
 import onnx
-import onnx_ir.passes.common as common_passes
 
+import onnxscript.ir.passes.common as common_passes
 from onnxscript import ir
 from onnxscript.rewriter import pattern
 from onnxscript.rewriter._basics import MatchContext, MatchingTracer, MatchResult, MatchStatus
@@ -41,6 +41,7 @@
     _min_max_to_clip,
     _no_op,
     _redundant_scatter_nd,
+    _remove_zero_bias,
 )
 
 _ModelProtoOrIr = TypeVar("_ModelProtoOrIr", onnx.ModelProto, ir.Model)
@@ -55,6 +56,7 @@
     *_redundant_scatter_nd.rules,
     *_fuse_pad_into_conv.rules,
     *_fuse_batchnorm.rules,
+    *_remove_zero_bias.rules,
 )
 
 

onnxscript/rewriter/ort_fusions/_core.py

@@ -33,14 +33,23 @@
     fuse_skip_layer_normalization,
     fuse_skip_rms_normalization,
 )
-from onnxscript.rewriter.rules.common import _gemm_to_matmul_add
+from onnxscript.rewriter.rules.common import (
+    _fuse_batchnorm,
+    _fuse_pad_into_conv,
+    _gemm_to_matmul_add,
+    _remove_zero_bias,
+)
 
 ORT_PATTERN_REWRITE_RULES = [
     *softmax.rules.rules,
     *instance_to_group_normalization.rules.rules,
     # NOTE: group normalization merge silu should be applied after instance to group normalization
     # *group_normalization_merge_silu.rules.rules,
     *fused_matmul_rule_sets.fused_matmul_rule_sets(),
+    # Add Conv fusion rules for better ORT optimization
+    *_fuse_batchnorm.rules.rules,
+    *_fuse_pad_into_conv.rules.rules,
+    *_remove_zero_bias.rules.rules,
 ]
 
 

onnxscript/rewriter/rules/common/__init__.py

@@ -34,6 +34,10 @@
     "normalize_pad_format_conv_integer_rule",
     "normalize_pad_format_conv_rule",
     "one_reshape_matmul_reshape_rule",
+    "remove_zero_bias_from_conv_rule",
+    "remove_zero_bias_from_conv_transpose_rule",
+    "remove_zero_bias_from_qlinear_conv_rule",
+    "remove_zero_bias_from_gemm_rule",
     "reshape_reshape_rule",
     "slice_split_rule",
     "squeeze_reshape_1d_rule",
@@ -121,3 +125,9 @@
     no_op_dynamic_scatter_nd_rule,
     no_op_static_scatter_nd_rule,
 )
+from onnxscript.rewriter.rules.common._remove_zero_bias import (
+    remove_zero_bias_from_conv_rule,
+    remove_zero_bias_from_conv_transpose_rule,
+    remove_zero_bias_from_gemm_rule,
+    remove_zero_bias_from_qlinear_conv_rule,
+)

onnxscript/rewriter/rules/common/_remove_zero_bias.py

@@ -0,0 +1,133 @@
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT License.
+"""Remove optional bias when it is all zero from Conv and related operations."""
+
+from __future__ import annotations
+
+from typing import ClassVar
+
+import numpy as np
+
+from onnxscript import ir
+from onnxscript.ir import convenience
+from onnxscript.rewriter._basics import MatchResult
+from onnxscript.rewriter._rewrite_rule import RewriteRuleClassBase, RewriteRuleSet
+
+
+class _RemoveZeroBiasBase(RewriteRuleClassBase):
+    """Base class for removing zero bias from operations."""
+
+    def rewrite(self, op: ir.tape.Tape, out: ir.Value, **_) -> ir.Value:
+        """Remove the bias input from the operation."""
+        node = out.producer()
+
+        original_inputs = list(node.inputs)
+        inputs_without_bias = original_inputs[:-1]
+
+        return op.op(
+            self.op_type,
+            inputs=inputs_without_bias,
+            attributes=node.attributes,
+            domain=node.domain,
+        )
+
+    def _check_bias_is_zero(self, bias_value: ir.Value) -> MatchResult:
+        """Check if the bias value is present and is all zeros."""
+        check_result = MatchResult()
+
+        # Check if bias is a constant/initializer
+        bias_tensor = convenience.get_const_tensor(bias_value)
+        if bias_tensor is None:
+            return check_result.fail("Bias is not a constant/initializer.")
+
+        # Check if bias is all zeros
+        bias_array = bias_tensor.numpy()
+        if not np.allclose(bias_array, 0.0, atol=1e-8):
+            return check_result.fail("Bias is not all zeros.")
+
+        return check_result
+
+    def check(self, context, x: ir.Value, w: ir.Value, b: ir.Value, **_) -> MatchResult:
+        """Check if the bias is present and is all zeros."""
+        del context  # Unused
+        return self._check_bias_is_zero(b)
+
+
+class RemoveZeroBiasFromConv(_RemoveZeroBiasBase):
+    """Remove zero bias from Conv operations."""
+
+    op_type: ClassVar = "Conv"
+
+    def pattern(self, op: ir.tape.Tape, x: ir.Value, w: ir.Value, b: ir.Value) -> ir.Value:
+        return op.Conv(x, w, b, _outputs=["out"])
+
+
+class RemoveZeroBiasFromConvTranspose(_RemoveZeroBiasBase):
+    """Remove zero bias from ConvTranspose operations."""
+
+    op_type: ClassVar = "ConvTranspose"
+
+    def pattern(self, op: ir.tape.Tape, x: ir.Value, w: ir.Value, b: ir.Value) -> ir.Value:
+        return op.ConvTranspose(x, w, b, _outputs=["out"])
+
+
+class RemoveZeroBiasFromQLinearConv(_RemoveZeroBiasBase):
+    """Remove zero bias from QLinearConv operations."""
+
+    op_type: ClassVar = "QLinearConv"
+
+    def pattern(
+        self,
+        op: ir.tape.Tape,
+        x,
+        x_scale,
+        x_zero_point,
+        w,
+        w_scale,
+        w_zero_point,
+        y_scale,
+        y_zero_point,
+        b: ir.Value,
+    ) -> ir.Value:
+        return op.QLinearConv(
+            x,
+            x_scale,
+            x_zero_point,
+            w,
+            w_scale,
+            w_zero_point,
+            y_scale,
+            y_zero_point,
+            b,
+            _outputs=["out"],
+        )
+
+
+class RemoveZeroBiasFromGemm(_RemoveZeroBiasBase):
+    """Remove zero bias from Gemm operations."""
+
+    op_type: ClassVar = "Gemm"
+
+    def pattern(self, op: ir.tape.Tape, a: ir.Value, b: ir.Value, c: ir.Value) -> ir.Value:
+        return op.Gemm(a, b, c, _outputs=["out"])
+
+    def check(self, context, a: ir.Value, b: ir.Value, c: ir.Value, **_) -> MatchResult:
+        """Check if the bias (c parameter) is present and is all zeros."""
+        del context  # Unused
+        return self._check_bias_is_zero(c)
+
+
+# Create rule instances
+remove_zero_bias_from_conv_rule = RemoveZeroBiasFromConv().rule()
+remove_zero_bias_from_conv_transpose_rule = RemoveZeroBiasFromConvTranspose().rule()
+remove_zero_bias_from_qlinear_conv_rule = RemoveZeroBiasFromQLinearConv().rule()
+remove_zero_bias_from_gemm_rule = RemoveZeroBiasFromGemm().rule()
+
+rules = RewriteRuleSet(
+    [
+        remove_zero_bias_from_conv_rule,
+        remove_zero_bias_from_conv_transpose_rule,
+        remove_zero_bias_from_qlinear_conv_rule,
+        remove_zero_bias_from_gemm_rule,
+    ]
+)