diff --git a/paddle/fluid/operators/elementwise_mul_mkldnn_op.cc b/paddle/fluid/operators/elementwise_mul_mkldnn_op.cc
index d66c58bd450737fc9bc3ce0788f9c3e3e9c126ba..36e88cd7895c35f7f0d268a78f421f63d0e84ff0 100644
--- a/paddle/fluid/operators/elementwise_mul_mkldnn_op.cc
+++ b/paddle/fluid/operators/elementwise_mul_mkldnn_op.cc
@@ -144,6 +144,7 @@ class ElementwiseMulMKLDNNKernel : public framework::OpKernel<T> {
 
           mul_func_t mul_func = (mul_func_t) mul.getCode();
 
+          #pragma omp parallel for collapse(2)
           for (int ni = 0; ni < n; ni++) {
             for (int ci = 0; ci < C; ci++) {
               auto ptr_x =