[patch, ice, ia64] Patch for PR middle-end/41551

[patch, ice, ia64] Patch for PR middle-end/41551

[Steve Ellcey]

PR 41551 is an ICE on IA64, Jakub Jelinek put a proposed patch for the
defect in the bug report but didn't submit it to gcc-patches (as far as
I could tell) so I have created a patch with his change, added a
testcase and tested the patch on IA64 HP-UX and Linux with no
regressions.

OK to checkin?

Steve Ellcey
sje@cup.hp.com



2009-11-23  Steve Ellcey  <sje@cup.hp.com>
	    Jakub Jelinek <jakub@redhat.com>

	* function.c (instantiate_virtual_regs_in_insn): Copy to new reg
	before forcing mode.


Index: function.c
===================================================================
--- function.c	(revision 154430)
+++ function.c	(working copy)
@@ -1598,7 +1598,10 @@ instantiate_virtual_regs_in_insn (rtx in
       if (!safe_insn_predicate (insn_code, i, x))
 	{
 	  start_sequence ();
-	  x = force_reg (insn_data[insn_code].operand[i].mode, x);
+	  if (REG_P (x))
+	    x = copy_to_reg (x);
+	  else
+	    x = force_reg (insn_data[insn_code].operand[i].mode, x);
 	  seq = get_insns ();
 	  end_sequence ();
 	  if (seq)




2009-11-23  Steve Ellcey  <sje@cup.hp.com>

	* gcc.dg/pr41551.c: New test.


Index: gcc.dg/pr41551.c
===================================================================
--- gcc.dg/pr41551.c	(revision 0)
+++ gcc.dg/pr41551.c	(revision 0)
@@ -0,0 +1,10 @@
+/* { dg-do compile } */
+/* { dg-options "-O2" } */
+
+/* Make sure we do not ICE.  */
+
+int main(void)
+{
+ int var, *p = &var;
+ return (double)(unsigned long)(p);
+}

Re: [patch, ice, ia64] Patch for PR middle-end/41551

[Richard Henderson]

On 11/23/2009 09:34 AM, Steve Ellcey wrote:
> +	  if (REG_P (x))
> +	    x = copy_to_reg (x);
> +	  else

I think I'd prefer we add something like

   gcc_assert (REGNO (x) <= LAST_VIRTUAL_REGISTER);

before that copy_to_reg.  Ok with that change.


r~