diff --git a/backends-clickhouse/src/test/scala/io/glutenproject/execution/GlutenFunctionValidateSuite.scala b/backends-clickhouse/src/test/scala/io/glutenproject/execution/GlutenFunctionValidateSuite.scala
index 833e21b0c254..57b5a09fc1a6 100644
--- a/backends-clickhouse/src/test/scala/io/glutenproject/execution/GlutenFunctionValidateSuite.scala
+++ b/backends-clickhouse/src/test/scala/io/glutenproject/execution/GlutenFunctionValidateSuite.scala
@@ -457,4 +457,14 @@ class GlutenFunctionValidateSuite extends GlutenClickHouseWholeStageTransformerS
       )(checkOperatorMatch[ProjectExecTransformer])
     }
   }
+
+  test("test round issue: https://github.com/oap-project/gluten/issues/3462") {
+    runQueryAndCompare(
+      "select round(0.41875d * id , 4) from range(10);"
+    )(checkOperatorMatch[ProjectExecTransformer])
+
+    runQueryAndCompare(
+      "select round(0.41875f * id , 4) from range(10);"
+    )(checkOperatorMatch[ProjectExecTransformer])
+  }
 }
diff --git a/cpp-ch/local-engine/Functions/SparkFunctionRoundHalfUp.h b/cpp-ch/local-engine/Functions/SparkFunctionRoundHalfUp.h
index 6bd3b04eaeee..01d628c5c63a 100644
--- a/cpp-ch/local-engine/Functions/SparkFunctionRoundHalfUp.h
+++ b/cpp-ch/local-engine/Functions/SparkFunctionRoundHalfUp.h
@@ -19,50 +19,28 @@
 #include <Functions/FunctionsRound.h>
 
 
-namespace  local_engine
+namespace local_engine
 {
 using namespace DB;
 
-
-/// Implementation for round half up. Not vectorized.
-
-inline float roundHalfUp(float x)
-{
-    return roundf(x);
-
-    UNREACHABLE();
-}
-
-inline double roundHalfUp(double x)
-{
-    return round(x);
-
-    UNREACHABLE();
-}
-
 template <typename T>
 class BaseFloatRoundingHalfUpComputation
 {
 public:
     using ScalarType = T;
-    using VectorType = T;
+    using VectorType = Float64;
     static const size_t data_count = 1;
 
-    static VectorType load(const ScalarType * in) { return *in; }
-    static VectorType load1(const ScalarType in) { return in; }
-    static VectorType store(ScalarType * out, ScalarType val) { return *out = val;}
+    static VectorType load(const ScalarType * in) { return static_cast<VectorType>(*in); }
+    static VectorType load1(ScalarType in) { return in; }
+    static ScalarType store(ScalarType * out, VectorType val) { return *out = static_cast<ScalarType>(val); }
     static VectorType multiply(VectorType val, VectorType scale) { return val * scale; }
     static VectorType divide(VectorType val, VectorType scale) { return val / scale; }
-    static VectorType apply(VectorType val){return roundHalfUp(val);}
-
-    static VectorType prepare(size_t scale)
-    {
-        return load1(scale);
-    }
+    static VectorType apply(VectorType val) { return round(val); }
+    static VectorType prepare(size_t scale) { return load1(scale); }
 };
 
 
-
 /** Implementation of low-level round-off functions for floating-point values.
   */
 template <typename T, ScaleMode scale_mode>
@@ -140,9 +118,6 @@ struct FloatRoundingHalfUpImpl
 };
 
 
-
-
-
 /** Select the appropriate processing algorithm depending on the scale.
   */
 template <typename T, RoundingMode rounding_mode, TieBreakingMode tie_breaking_mode>