diff --git a/paddle/pten/CMakeLists.txt b/paddle/pten/CMakeLists.txt
index 8990def122601e99678b213aa16f3f69bf8b0ffc..1888d0b52809fbf5e50ee4c7b29b3f678590f06f 100644
--- a/paddle/pten/CMakeLists.txt
+++ b/paddle/pten/CMakeLists.txt
@@ -24,7 +24,7 @@ add_subdirectory(tests)
 
 # make an unity target for compile deps
 set(PTEN_DEPS convert_utils dense_tensor pten_context kernel_factory kernel_context)
-set(PTEN_DEPS scale_kernel_eigen)
+set(PTEN_DEPS ${PTEN_DEPS} scale_kernel_eigen)
 set(PTEN_DEPS ${PTEN_DEPS} math_cpu linalg_cpu creation_cpu manipulation_cpu)
 set(PTEN_DEPS ${PTEN_DEPS} nary unary binary)
 if(WITH_GPU OR WITH_ROCM)
diff --git a/paddle/pten/kernels/cuda/math.h b/paddle/pten/kernels/cuda/math.h
index c9f2f04494f84f24ff62d537b550a3d32f87922a..b2bca5e41185e09f2621a549b2f5da47e25f44f8 100644
--- a/paddle/pten/kernels/cuda/math.h
+++ b/paddle/pten/kernels/cuda/math.h
@@ -36,14 +36,6 @@ void Mean(const CUDAContext& dev_ctx,
           DataType out_dtype,
           DenseTensor* out);
 
-template <typename T>
-void Scale(const CUDAContext& dev_ctx,
-           const DenseTensor& x,
-           const Scalar& scale,
-           float bias,
-           bool bias_after_scale,
-           DenseTensor* out);
-
 template <typename T>
 void Add(const CUDAContext& dev_ctx,
          const DenseTensor& x,