Re: [PATCH] Re-work get_object_alignment (again)

[Richard Guenther <rguenther@suse.de>]

On Thu, 19 Jul 2012, Markus Trippelsdorf wrote:

> On 2012.07.17 at 15:10 +0200, Richard Guenther wrote:
> > Comments welcome, of course.
> 
> This patch apparently miscompiles the Linux kernel, which just
> hangs during early boot:
> 
> ...
> SLUB: Genslabs=15, HWalign=64, Order=0-3, MinObjects=0, CPUs=4, Nodes=1
> Hierarchical RCU implementation.
> NR_IRQS:4352 nr_irqs:712 16
> Extended CMOS year: 2000
> Console: colour VGA+ 80x25
> console [tty0] enabled
> (hang)
> 
> See: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54031

The following fixes one issue (but reportedly does not fix the above
issue).

Bootstrapped and tested on x86_64-unknown-linux-gnu, applied to trunk.

Richard.

2012-07-20  Richard Guenther  <rguenther@suse.de>

	* builtins.c (get_object_alignment_2): Correct offset handling
	when using type alignment of a MEM_REF kind base.

Index: gcc/builtins.c
===================================================================
--- gcc/builtins.c	(revision 189656)
+++ gcc/builtins.c	(working copy)
@@ -346,12 +346,10 @@ get_object_alignment_2 (tree exp, unsign
 
       known_alignment
 	= get_pointer_alignment_1 (addr, &ptr_align, &ptr_bitpos);
-      bitpos += ptr_bitpos;
       align = MAX (ptr_align, align);
 
-      if (TREE_CODE (exp) == MEM_REF
-	  || TREE_CODE (exp) == TARGET_MEM_REF)
-	bitpos += mem_ref_offset (exp).low * BITS_PER_UNIT;
+      /* The alignment of the pointer operand in a TARGET_MEM_REF
+	 has to take the variable offset parts into account.  */
       if (TREE_CODE (exp) == TARGET_MEM_REF)
 	{
 	  if (TMR_INDEX (exp))
@@ -369,9 +367,19 @@ get_object_alignment_2 (tree exp, unsign
       /* When EXP is an actual memory reference then we can use
 	 TYPE_ALIGN of a pointer indirection to derive alignment.
 	 Do so only if get_pointer_alignment_1 did not reveal absolute
-	 alignment knowledge.  */
-      if (!addr_p && !known_alignment)
-	align = MAX (TYPE_ALIGN (TREE_TYPE (exp)), align);
+	 alignment knowledge and if using that alignment would
+	 improve the situation.  */
+      if (!addr_p && !known_alignment
+	  && TYPE_ALIGN (TREE_TYPE (exp)) > align)
+	align = TYPE_ALIGN (TREE_TYPE (exp));
+      else
+	{
+	  /* Else adjust bitpos accordingly.  */
+	  bitpos += ptr_bitpos;
+	  if (TREE_CODE (exp) == MEM_REF
+	      || TREE_CODE (exp) == TARGET_MEM_REF)
+	    bitpos += mem_ref_offset (exp).low * BITS_PER_UNIT;
+	}
     }
   else if (TREE_CODE (exp) == STRING_CST)
     {