[Bug rtl-optimization/17236] inefficient code for long long multiply on x86

[bonzini at gnu dot org]

------- Comment #17 from bonzini at gnu dot org  2007-12-19 09:49 -------
With this patch, GCC gets the preferences right, but it does not affect code
generation.

Index: regclass.c
===================================================================
--- regclass.c  (revision 130928)
+++ regclass.c  (working copy)
@@ -1651,9 +1651,15 @@ record_reg_classes (int n_alts, int n_op
                          [(unsigned char) reg_pref[REGNO (op)].prefclass]
                          [(int) classes[i]]);

-                 if (REGNO (ops[i]) != REGNO (ops[j])
-                     && ! find_reg_note (insn, REG_DEAD, op))
-                   alt_cost += 2;
+                 if (REGNO (ops[i]) != REGNO (ops[j]))
+                   {
+                     /* If the pseudo dies, tying it to the duplicate
+                        operand can be advantageous.  */
+                     if (find_reg_note (insn, REG_DEAD, op))
+                       pp->cost[classes[j]]--;
+                     else
+                       alt_cost += 2;
+                   }

                  /* This is in place of ordinary cost computation
                     for this operand, so skip to the end of the


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17236