src/hotspot/share/opto/superword.cpp

*** 1443,1453 ****
  
    for (DUIterator_Fast imax, i = s1->fast_outs(imax); i < imax; i++) {
      Node* t1 = s1->fast_out(i);
  
      // Only allow operand swap on commuting operations
!     if (!t1->is_Add() && !t1->is_Mul()) {
        break;
      }
  
      // Now find t1's packset
      Node_List* p2 = NULL;
--- 1443,1453 ----
  
    for (DUIterator_Fast imax, i = s1->fast_outs(imax); i < imax; i++) {
      Node* t1 = s1->fast_out(i);
  
      // Only allow operand swap on commuting operations
!     if (!t1->is_Add() && !t1->is_Mul() && !VectorNode::is_muladds2i(t1)) {
        break;
      }
  
      // Now find t1's packset
      Node_List* p2 = NULL;
*** 1498,1510 ****
--- 1498,1524 ----
      for (i2++; i2 < ct; i2++) if (u2->in(i2) == d2) break;
      if (i1 != i2) {
        if ((i1 == (3-i2)) && (u2->is_Add() || u2->is_Mul())) {
          // Further analysis relies on operands position matching.
          u2->swap_edges(i1, i2);
+       } else if (VectorNode::is_muladds2i(u2) && u1 != u2) {
+         if (i1 == 5 - i2) { // ((i1 == 3 && i2 == 2) || (i1 == 2 && i2 == 3) || (i1 == 1 && i2 == 4) || (i1 == 4 && i2 == 1))
+           u2->swap_edges(1, 2);
+           u2->swap_edges(3, 4);
+         }
+         if (i1 == 3 - i2 || i1 == 7 - i2) { // ((i1 == 1 && i2 == 2) || (i1 == 2 && i2 == 1) || (i1 == 3 && i2 == 4) || (i1 == 4 && i2 == 3))
+           u2->swap_edges(2, 3);
+           u2->swap_edges(1, 4);
+         }
+         return false; // Just swap the edges, the muladds2i nodes get packed in follow_use_defs
        } else {
          return false;
        }
+     } else if (i1 == i2 && VectorNode::is_muladds2i(u2) && u1 != u2) {
+       u2->swap_edges(1, 3);
+       u2->swap_edges(2, 4);
+       return false; // Just swap the edges, the muladds2i nodes get packed in follow_use_defs
      }
    } while (i1 < ct);
    return true;
  }