[Bug tree-optimization/64910] tree reassociation results in poor code

[law at redhat dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64910

--- Comment #1 from Jeffrey A. Law <law at redhat dot com> ---
Here's the hack that I was playing with which shows the better code sequence:
diff --git a/gcc/tree-ssa-reassoc.c b/gcc/tree-ssa-reassoc.c
index 9952222..4515a4d 100644
--- a/gcc/tree-ssa-reassoc.c
+++ b/gcc/tree-ssa-reassoc.c
@@ -3476,6 +3476,18 @@ swap_ops_for_binary_stmt (vec<operand_entry_t> ops,
       oe1->op = temp.op;
       oe1->rank = temp.rank;
     }
+  else if (TREE_CODE (oe1->op) != INTEGER_CST
+          && TREE_CODE (oe2->op) != INTEGER_CST
+          && TREE_CODE (oe3->op) == INTEGER_CST)
+    {
+      struct operand_entry temp = *oe3;
+      oe3->op = oe2->op;
+      oe3->rank = oe2->rank;
+      oe2->op = oe1->op;
+      oe2->rank = oe1->rank;
+      oe1->op = temp.op;
+      oe1->rank = temp.rank;
+    }
 }

 /* If definition of RHS1 or RHS2 dominates STMT, return the later of those