diff --git a/tf_adapter/kernels/aicpu/npu_cpu_ops.cc b/tf_adapter/kernels/aicpu/npu_cpu_ops.cc
index ff9dc6d2ffe9fbae28d9febe6bef0b50b4a5d5df..58299c2f628114fc78dc7f4d86bf048e5d58fa28 100644
--- a/tf_adapter/kernels/aicpu/npu_cpu_ops.cc
+++ b/tf_adapter/kernels/aicpu/npu_cpu_ops.cc
@@ -221,6 +221,13 @@ class ScatterElementsV2Op : public OpKernel {
   }
 };
 
+class PrecisonCompareOp : public OpKernel {
+ public:
+  explicit PrecisonCompareOp(OpKernelConstruction *context) : OpKernel(context) {}
+  ~PrecisonCompareOp() override {}
+  void Compute(OpKernelContext *context) override { ADP_LOG(INFO) << "PrecisonCompareOp Compute"; }
+};
+
 REGISTER_KERNEL_BUILDER(Name("ScatterElementsV2").Device(DEVICE_CPU), ScatterElementsV2Op);
 REGISTER_KERNEL_BUILDER(Name("EmbeddingRankId").Device(DEVICE_CPU), EmbeddingRankIdOpKernel);
 REGISTER_KERNEL_BUILDER(Name("EmbeddingLocalIndex").Device(DEVICE_CPU), EmbeddingLocalIndexOpKernel);
@@ -242,6 +249,7 @@ REGISTER_KERNEL_BUILDER(Name("NonZeroWithValueShape").Device(DEVICE_CPU), NonZer
 REGISTER_KERNEL_BUILDER(Name("NonZeroWithValueShapeV2").Device(DEVICE_CPU), NonZeroWithValueShapeV2Op);
 REGISTER_KERNEL_BUILDER(Name("WarpAffineV2").Device(DEVICE_CPU), WarpAffineV2Op);
 REGISTER_KERNEL_BUILDER(Name("ResizeV2").Device(DEVICE_CPU), ResizeV2Op);
+REGISTER_KERNEL_BUILDER(Name("PrecisonCompare").Device(DEVICE_CPU), PrecisonCompareOp);
 
 class DecodeImageV3Op : public OpKernel {
 public:
diff --git a/tf_adapter/ops/aicpu/npu_cpu_ops.cc b/tf_adapter/ops/aicpu/npu_cpu_ops.cc
index 40857b44989f707098f76c2f6a00ca1930d70969..d710815006283b1c31984a0de85c539cbf90dfd4 100644
--- a/tf_adapter/ops/aicpu/npu_cpu_ops.cc
+++ b/tf_adapter/ops/aicpu/npu_cpu_ops.cc
@@ -627,4 +627,11 @@ REGISTER_OP("ResizeV2")
     .Output("image: dtype")
     .Attr("expand_animations: bool = true")
     .SetShapeFn(DecodeImageV3ShapeFn);
+
+    REGISTER_OP("PrecisionCompare")
+        .Input("x1: T")
+        .Input("x2: T")
+        .Attr("detect_type: int32")
+        .Attr("T: {float32, float16, bfloat16}")
+        .SetShapeFn(tensorflow::shape_inference::ScalarShape);
 }  // namespace tensorflow
diff --git a/tf_adapter/python/npu_bridge/npu_cpu/npu_cpu_ops.py b/tf_adapter/python/npu_bridge/npu_cpu/npu_cpu_ops.py
index 417119584f09363575c133283ad275aab85380dd..4d6d77b19ede0a15a6ab65a7d7cbf4b0afb75047 100644
--- a/tf_adapter/python/npu_bridge/npu_cpu/npu_cpu_ops.py
+++ b/tf_adapter/python/npu_bridge/npu_cpu/npu_cpu_ops.py
@@ -558,3 +558,14 @@ def embedding_hashmap_import_v2(file_path, table_ids, table_sizes, table_names,
         file_path=file_path, table_ids=table_ids, table_sizes=table_sizes,
         table_names=table_names, global_step=global_step, embedding_dims=embedding_dims, num=num)
     return result
+
+## 提供硬件精度检测功能
+#  @param golden 支持float/float16/bfloat16类型
+#  @param realdata 支持float/float16/bfloat16类型
+#  @param detect_type int32 类型, 取值范围：0/1/2
+#  @return uint32
+def precision_compare(golden, realdata, detect_type):
+    """ precision compare. """
+    result = gen_npu_cpu_ops.PrecisonCompare(
+        x1=golden, x2=realdata, detect_type=detect_type)
+    return result
\ No newline at end of file