Merge pull request #10941 from mozga-intel/mozga-intel/Top-K

Top K algorithm: parallel version

paddle/fluid/operators/top_k_op.h

@@ -55,6 +55,9 @@ class TopkKernel : public framework::OpKernel<T> {
     // NOTE: eigen shape doesn't affect paddle tensor.
     eg_input.reshape(flat2dims);
 
+#ifdef PADDLE_WITH_MKLML
+#pragma omp parallel for
+#endif
     for (size_t i = 0; i < row; i++) {
       std::vector<std::pair<T, size_t>> vec;
       for (size_t j = 0; j < col; j++) {