diff --git a/src/pass/onnx_rewrite.td b/src/pass/onnx_rewrite.td
index c3b5490..22a3cd1 100644
--- a/src/pass/onnx_rewrite.td
+++ b/src/pass/onnx_rewrite.td
@@ -33,6 +33,11 @@ class StringAttrOfValue<string val>:
 class FloatAttrOfValue<int val>:
   NativeCodeCall<"FloatAttr::get($0.getType().cast<TensorType>().getElementType(), " # val # ")">;
 
+// Create a FloatAttr for the negative infinity.
+def FloatAttrOfNegativeInfinity:
+  NativeCodeCall<"FloatAttr::get($0.getType().cast<TensorType>().getElementType(), "
+                                "-std::numeric_limits<double>::infinity())">;
+
 // Create an ArrayAttr of IntergerAttr(s) of zero values.
 // This function is used for padding attribute in MaxPoolSingleOut.
 def createArrayAttrOfZerosFrom:
@@ -80,7 +85,7 @@ def MaxPoolSingleOutOpPaddingPattern: Pat<
   (ONNXMaxPoolSingleOutOp
      (ONNXPadConstantValuePadOp $x,
         (insertZerosForNonPaddedDims<2> $pads),
-        (FloatAttrOfValue<0> $res),
+        (FloatAttrOfNegativeInfinity $res),
         (StringAttrOfValue<"constant">)),
      $auto_pad, $ceil_mode, $dilation, $kernel_shape,
      (createArrayAttrOfZerosFrom $pads),
diff --git a/test/mlir/onnx/onnx_canonicalization.mlir b/test/mlir/onnx/onnx_canonicalization.mlir
index 78193cb..3e27d7e 100644
--- a/test/mlir/onnx/onnx_canonicalization.mlir
+++ b/test/mlir/onnx/onnx_canonicalization.mlir
@@ -82,7 +82,7 @@ func @test_maxpoolsingleout_split(%arg0: tensor<5x5x32x32xf32>) -> tensor<5x8x32
   %0 = "onnx.MaxPoolSingleOut"(%arg0) {auto_pad = "NOTSET", ceil_mode = 0, kernel_shape = [5,3], pads = [1, 2, 3, 4] } : (tensor<5x5x32x32xf32>) -> tensor<5x8x32x39xf32>
   "std.return"(%0) : (tensor<5x8x32x39xf32>) -> ()
 
-  // CHECK-NEXT: %0 = "onnx.PadConstantValuePad"(%arg0) {constant_value = 0.000000e+00 : f32, mode = "constant", pads = [0, 0, 1, 2, 0, 0, 3, 4]} : (tensor<5x5x32x32xf32>) -> tensor<5x8x32x39xf32>
+  // CHECK-NEXT: %0 = "onnx.PadConstantValuePad"(%arg0) {constant_value = 0xFF800000 : f32, mode = "constant", pads = [0, 0, 1, 2, 0, 0, 3, 4]} : (tensor<5x5x32x32xf32>) -> tensor<5x8x32x39xf32>
   // CHECK-NEXT: %1 = "onnx.MaxPoolSingleOut"(%0) {auto_pad = "NOTSET", ceil_mode = 0 : i64, kernel_shape = [5, 3], pads = [0, 0, 0, 0], storage_order = 0 : i64} : (tensor<5x8x32x39xf32>) -> tensor<5x8x32x39xf32>
   // CHECK-NEXT: return %1 : tensor<5x8x32x39xf32>
 }
@@ -92,7 +92,7 @@ func @test_maxpoolsingleout_split_unknown_dims(%arg0: tensor<*xf32>) -> tensor<*
   %0 = "onnx.MaxPoolSingleOut"(%arg0) {auto_pad = "NOTSET", ceil_mode = 0, kernel_shape = [5,3], pads = [1, 2, 3, 4] } : (tensor<*xf32>) -> tensor<*xf32>
   "std.return"(%0) : (tensor<*xf32>) -> ()
 
-  // CHECK-NEXT: %0 = "onnx.PadConstantValuePad"(%arg0) {constant_value = 0.000000e+00 : f32, mode = "constant", pads = [0, 0, 1, 2, 0, 0, 3, 4]} : (tensor<*xf32>) -> tensor<*xf32>
+  // CHECK-NEXT: %0 = "onnx.PadConstantValuePad"(%arg0) {constant_value = 0xFF800000 : f32, mode = "constant", pads = [0, 0, 1, 2, 0, 0, 3, 4]} : (tensor<*xf32>) -> tensor<*xf32>
   // CHECK-NEXT: %1 = "onnx.MaxPoolSingleOut"(%0) {auto_pad = "NOTSET", ceil_mode = 0 : i64, kernel_shape = [5, 3], pads = [0, 0, 0, 0], storage_order = 0 : i64} : (tensor<*xf32>) -> tensor<*xf32>
   // CHECK-NEXT: return %1 : tensor<*xf32>
 }