diff --git a/paddle/phi/kernels/cpu/embedding_kernel.cc b/paddle/phi/kernels/cpu/embedding_kernel.cc
index cfba1787a1576e361809ab0d49273f5f70766947..0d937e6364eacc22863ac34ec7a2afe65fee317a 100644
--- a/paddle/phi/kernels/cpu/embedding_kernel.cc
+++ b/paddle/phi/kernels/cpu/embedding_kernel.cc
@@ -48,14 +48,8 @@ struct EmbeddingCPUFunctor {
     dev_ctx_.template Alloc<T>(out_);
     auto* output = out_->data<T>();
 
-#if defined(_OPENMP) && !defined(PADDLE_WITH_CUDA)
-#pragma omp parallel for
-#endif
-
     for (int64_t i = 0; i < ids_numel; ++i) {
-      if (padding_idx_ != kNoPadding && ids[i] == padding_idx_) {
-        memset(output + i * row_width, 0, row_width * sizeof(T));
-      } else {
+      if (padding_idx_ == kNoPadding && ids[i] != padding_idx_) {
         PADDLE_ENFORCE_LT(
             ids[i],
             row_number,
@@ -74,6 +68,17 @@ struct EmbeddingCPUFunctor {
                 "value.",
                 row_number,
                 ids[i]));
+      }
+    }
+
+#if defined(_OPENMP) && !defined(PADDLE_WITH_CUDA)
+#pragma omp parallel for
+#endif
+
+    for (int64_t i = 0; i < ids_numel; ++i) {
+      if (padding_idx_ != kNoPadding && ids[i] == padding_idx_) {
+        memset(output + i * row_width, 0, row_width * sizeof(T));
+      } else {
         memcpy(output + i * row_width,
                table + ids[i] * row_width,
                row_width * sizeof(T));