diff --git a/paddle/phi/kernels/onednn/split_kernel.cc b/paddle/phi/kernels/onednn/split_kernel.cc
index 94c0924e36e00bf60c5e0fbc4287c2321b458b80..057aa7325f165c3251c395dcbb6dcecc4f96b8c8 100644
--- a/paddle/phi/kernels/onednn/split_kernel.cc
+++ b/paddle/phi/kernels/onednn/split_kernel.cc
@@ -69,10 +69,8 @@ void SplitWithNumKernel(const Context& dev_ctx,
                         std::vector<DenseTensor*> outs) {
   int axis_value = axis_scalar.to<int>();
   auto input_axis_dim = x.dims().at(axis_value);
-  std::vector<int64_t> sections_vec;
-  for (int i = 0; i < num; ++i) {
-    sections_vec.push_back(input_axis_dim / num);
-  }
+  const std::vector<int64_t> sections_vec(num, input_axis_dim / num);
+
   IntArray sections(sections_vec);
   SplitKernel<T, Context>(dev_ctx, x, sections, axis_scalar, outs);
 }