adding comment (#1193)

src/ngraph/runtime/cpu/pass/cpu_mat_fusion.cpp

@@ -303,6 +303,12 @@ std::shared_ptr<Node> fuse_group_convolution(const std::shared_ptr<Node>& n)
     }
 
     //TF-flavoured group convolution needs channels re-arranged
+    //MKLDNN requires group slicing to be done on OC
+    //MKLDNN [4,2,-]
+    //ordering w00 w01 w10 w11 w20 w21 w30 w31 produces g00 g01 g10 g11
+    //whereas
+    //TF    [2,4,-]
+    //ordering w00 w01 w02 w03 w10 w11 w12 w13 produces g00 g10 g01 g11
     const size_t CONCAT_AXIS_OC = 0;
     if (!slices.empty())
     {