[ONNX] Added BiasAdd from com.microsoft domain (#28713)

Details:
Microsoft Contrib Operator "BiasAdd" for ONNX RT

Tickets:
N/A

src/frontends/onnx/frontend/src/op/com.microsoft/bias_add.cpp

@@ -0,0 +1,80 @@
+// Copyright (C) 2018-2025 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+//
+
+#include "core/operator_set.hpp"
+#include "exceptions.hpp"
+#include "openvino/frontend/exception.hpp"
+#include "openvino/op/add.hpp"
+#include "openvino/op/constant.hpp"
+#include "utils/common.hpp"
+
+using namespace ov::op;
+
+namespace ov {
+namespace frontend {
+namespace onnx {
+namespace com_microsoft {
+namespace opset_1 {
+
+ov::OutputVector bias_add(const ov::frontend::onnx::Node& node) {
+    // Documentation: BiasAdd computes Y = X + bias + skip
+    // https://github.com/microsoft/onnxruntime/blob/main/docs/ContribOperators.md#commicrosoftbiasadd
+
+    common::default_op_checks(node, 3);
+
+    const auto inputs = node.get_ov_inputs();
+    const auto& X = inputs[0];
+    const auto& bias = inputs[1];
+    const auto& skip = inputs[2];
+
+    auto element_type = X.get_element_type();
+    CHECK_VALID_NODE(node,
+                     element_type == ov::element::f16 || element_type == ov::element::f32,
+                     "Unsupported input data type for X, expected FP16 or FP32 but got: ",
+                     element_type);
+
+    const auto& X_shape = X.get_partial_shape();
+    const auto& bias_shape = bias.get_partial_shape();
+    const auto& skip_shape = skip.get_partial_shape();
+
+    CHECK_VALID_NODE(node,
+                     X_shape.rank().is_static() && X_shape.rank().get_length() == 3,
+                     "Input X must have rank 3 (N, S, C), but got: ",
+                     X_shape);
+    CHECK_VALID_NODE(node,
+                     skip_shape.rank().is_static() && skip_shape.rank().get_length() == 3,
+                     "Input skip must have rank 3 (N, S, C), but got: ",
+                     skip_shape);
+
+    CHECK_VALID_NODE(node,
+                     X_shape.compatible(skip_shape),
+                     "Input X and skip must have the same shape, but got: X=",
+                     X_shape,
+                     ", skip=",
+                     skip_shape);
+
+    CHECK_VALID_NODE(node,
+                     bias_shape.rank().is_static() && bias_shape.rank().get_length() == 1,
+                     "Input bias must have rank 1 (C), but got: ",
+                     bias_shape);
+
+    CHECK_VALID_NODE(node,
+                     X_shape[2].compatible(bias_shape[0]),
+                     "Input bias shape must match the channel dimension (C) of X and skip, but got: bias=",
+                     bias_shape,
+                     ", X=",
+                     X_shape);
+
+    auto X_plus_bias = std::make_shared<v1::Add>(X, bias);
+
+    return {std::make_shared<v1::Add>(X_plus_bias, skip)};
+}
+
+ONNX_OP("BiasAdd", OPSET_SINCE(1), com_microsoft::opset_1::bias_add, MICROSOFT_DOMAIN);
+
+}  // namespace opset_1
+}  // namespace com_microsoft
+}  // namespace onnx
+}  // namespace frontend
+}  // namespace ov

src/frontends/onnx/tests/models/com.microsoft/bias_add.prototxt

@@ -0,0 +1,86 @@
+ir_version: 7
+producer_name: "OpenVINO ONNX Frontend"
+graph {
+  node {
+    input: "X"
+    input: "bias"
+    input: "skip"
+    output: "Y"
+    op_type: "BiasAdd"
+    domain: "com.microsoft"
+  }
+  name: "test_bias_add_float"
+  input {
+    name: "X"
+    type {
+      tensor_type {
+        elem_type: 1
+        shape {
+          dim {
+            dim_value: 2
+          }
+          dim {
+            dim_value: 4
+          }
+          dim {
+            dim_value: 3
+          }
+        }
+      }
+    }
+  }
+  input {
+    name: "bias"
+    type {
+      tensor_type {
+        elem_type: 1
+        shape {
+          dim {
+            dim_value: 3
+          }
+        }
+      }
+    }
+  }
+  input {
+    name: "skip"
+    type {
+      tensor_type {
+        elem_type: 1
+        shape {
+          dim {
+            dim_value: 2
+          }
+          dim {
+            dim_value: 4
+          }
+          dim {
+            dim_value: 3
+          }
+        }
+      }
+    }
+  }
+  output {
+    name: "Y"
+    type {
+      tensor_type {
+        elem_type: 1
+        shape {
+          dim {
+            dim_value: 2
+          }
+          dim {
+            dim_value: 4
+          }
+          dim {
+            dim_value: 3
+          }
+        }
+      }
+    }
+  }
+}
+opset_import {
+  version: 1
+}

src/frontends/onnx/tests/onnx_import_com_microsoft.in.cpp

@@ -1682,3 +1682,24 @@ OPENVINO_TEST(${BACKEND_NAME}, onnx_com_microsoft_qlinear_mul) {
     test_case.add_expected_output<int8_t>(Shape{2, 2}, expected_output);
     test_case.run();
 }
+
+OPENVINO_TEST(${BACKEND_NAME}, onnx_com_microsoft_bias_add) {
+    const auto model = convert_model("com.microsoft/bias_add.onnx");
+    auto test_case = ov::test::TestCase(model, s_device);
+
+    const std::vector<float> X = {1,  2,  3,  4,  5,  6,  7,  8,  9,  10, 11, 12,
+                                  13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24};
+    const std::vector<float> bias = {2, 0, -1};
+    const std::vector<float> skip = {0, 0, 0, 1, 1, 1, 2, 2, 2, 3, 3, 3, 4, 4, 4, 5, 5, 5, 6, 6, 6, 7, 7, 7};
+
+    const std::vector<float> expected_output = {3,  2,  2,  7,  6,  6,  11, 10, 10, 15, 14, 14,
+                                                19, 18, 18, 23, 22, 22, 27, 26, 26, 31, 30, 30};
+
+    test_case.add_input<float>(Shape{2, 4, 3}, X);
+    test_case.add_input<float>(Shape{3}, bias);
+    test_case.add_input<float>(Shape{2, 4, 3}, skip);
+
+    test_case.add_expected_output<float>(Shape{2, 4, 3}, expected_output);
+
+    test_case.run();
+}