MKLDNN elementwise_mul: Parallelize mul

d14858e4 · Michal Gallus · ed31936b · d14858e4
隐藏空白更改
内联并排

Showing with 1 addition and 0 deletion

paddle/fluid/operators/elementwise_mul_mkldnn_op.cc paddle/fluid/operators/elementwise_mul_mkldnn_op.cc +1 -0

未找到文件。
--- 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 =