Random cleanups [2/4]: canonicalize ctor values

Random cleanups [2/4]: canonicalize ctor values

[Michael Matz]

Hi,

this came up when looking into why the static ctors contain useless trees 
(like casts).  We can simply canonicalize them while varpool analyzes 
pending decls.  It'll look at initialzers once, where we can "gimplify" 
them.  This requires making canonicalize_constructor_val be able to be 
called outside of functions.  And it requires the java frontend not 
leaving a dangling function decl as current (for the static ctor function 
it generates).

Regstrapped on x86_64-linux with the other three cleanups.  Okay for 
trunk?


Ciao,
Michael.
-- 
	* cgraphbuild.c (record_reference): Canonicalize constructor
	values.
	* gimple-fold.c (canonicalize_constructor_val): Accept being called
	without function context.

java/
	* jcf-parse.c (java_emit_static_constructor): Clear cfun and
	current_function_decl.

Index: cgraphbuild.c
===================================================================
--- cgraphbuild.c	(revision 171537)
+++ cgraphbuild.c	(working copy)
@@ -53,6 +53,8 @@ record_reference (tree *tp, int *walk_su
   tree decl;
   struct record_reference_ctx *ctx = (struct record_reference_ctx *)data;
 
+restart:
+
   switch (TREE_CODE (t))
     {
     case VAR_DECL:
@@ -98,6 +100,15 @@ record_reference (tree *tp, int *walk_su
 	  break;
 	}
 
+      t = canonicalize_constructor_val (t);
+      if (t && t != *tp)
+	{
+	  *tp = t;
+	  goto restart;
+	}
+      else
+	t = *tp;
+
       if ((unsigned int) TREE_CODE (t) >= LAST_AND_UNUSED_TREE_CODE)
 	return lang_hooks.callgraph.analyze_expr (tp, walk_subtrees);
       break;
Index: gimple-fold.c
===================================================================
--- gimple-fold.c	(revision 171537)
+++ gimple-fold.c	(working copy)
@@ -106,7 +106,7 @@ can_refer_decl_in_current_unit_p (tree d
   return true;
 }
 
-/* CVAL is value taken from DECL_INITIAL of variable.  Try to transorm it into
+/* CVAL is value taken from DECL_INITIAL of variable.  Try to transform it into
    acceptable form for is_gimple_min_invariant.   */
 
 tree
@@ -131,7 +131,7 @@ canonicalize_constructor_val (tree cval)
 	      || TREE_CODE (base) == FUNCTION_DECL)
 	  && !can_refer_decl_in_current_unit_p (base))
 	return NULL_TREE;
-      if (base && TREE_CODE (base) == VAR_DECL)
+      if (cfun && base && TREE_CODE (base) == VAR_DECL)
 	add_referenced_var (base);
       /* We never have the chance to fixup types in global initializers
          during gimplification.  Do so here.  */
Index: java/jcf-parse.c
===================================================================
--- java/jcf-parse.c.orig	2011-03-26 02:19:03.000000000 +0100
+++ java/jcf-parse.c	2011-03-28 06:16:43.000000000 +0200
@@ -1725,6 +1725,8 @@ java_emit_static_constructor (void)
       DECL_STATIC_CONSTRUCTOR (decl) = 1;
       java_genericize (decl);
       cgraph_finalize_function (decl, false);
+      current_function_decl = NULL;
+      set_cfun (NULL);
     }
 }
 

Re: Random cleanups [2/4]: canonicalize ctor values

[Paolo Bonzini]

On 03/31/2011 03:22 AM, Michael Matz wrote:
> this came up when looking into why the static ctors contain useless trees
> (like casts).  We can simply canonicalize them while varpool analyzes
> pending decls.  It'll look at initialzers once, where we can "gimplify"
> them.  This requires making canonicalize_constructor_val be able to be
> called outside of functions.  And it requires the java frontend not
> leaving a dangling function decl as current (for the static ctor function
> it generates).

canonicalize_constructor_val may be doing useful things on ADDR_EXPR 
too, but you don't call it in that case because you only added your code 
in the default case.  You only reach it when the ADDR_EXPR is wrapped in 
a cast.

Any reason why you didn't add it (possibly wrapped in a loop) _before_ 
the switch statement?

Paolo

Re: Random cleanups [2/4]: canonicalize ctor values

[Richard Guenther]

On Thu, Mar 31, 2011 at 9:14 AM, Paolo Bonzini <bonzini@gnu.org> wrote:
> On 03/31/2011 03:22 AM, Michael Matz wrote:
>>
>> this came up when looking into why the static ctors contain useless trees
>> (like casts).  We can simply canonicalize them while varpool analyzes
>> pending decls.  It'll look at initialzers once, where we can "gimplify"
>> them.  This requires making canonicalize_constructor_val be able to be
>> called outside of functions.  And it requires the java frontend not
>> leaving a dangling function decl as current (for the static ctor function
>> it generates).
>
> canonicalize_constructor_val may be doing useful things on ADDR_EXPR too,
> but you don't call it in that case because you only added your code in the
> default case.  You only reach it when the ADDR_EXPR is wrapped in a cast.
>
> Any reason why you didn't add it (possibly wrapped in a loop) _before_ the
> switch statement?

Indeed.  This comment needs adjustment, too:

      /* We never have the chance to fixup types in global initializers
         during gimplification.  Do so here.  */
      if (TREE_TYPE (TREE_TYPE (cval)) != TREE_TYPE (TREE_OPERAND (cval, 0)))
        cval = build_fold_addr_expr (TREE_OPERAND (cval, 0));

as we now sort-of do this.  In fact it looks like all existing
canonicalize_constructor_val calls are subsumed by the lowering, so,
eventually remove those and move this function to a local place next
to its sole caller.

Richard.

> Paolo
>

Re: Random cleanups [2/4]: canonicalize ctor values

[Michael Matz]

[-- Attachment #1: Type: TEXT/PLAIN, Size: 4919 bytes --]

On Thu, 31 Mar 2011, Richard Guenther wrote:

> > canonicalize_constructor_val may be doing useful things on ADDR_EXPR 
> > too, but you don't call it in that case because you only added your 
> > code in the default case.  You only reach it when the ADDR_EXPR is 
> > wrapped in a cast.
> >
> > Any reason why you didn't add it (possibly wrapped in a loop) _before_ 
> > the switch statement?

Nope, probably I didn't want to pay the cost each time, but you're right.

> Indeed.  This comment needs adjustment, too:
> 
>       /* We never have the chance to fixup types in global initializers
>          during gimplification.  Do so here.  */

Done in new patch.

> as we now sort-of do this.  In fact it looks like all existing 
> canonicalize_constructor_val calls are subsumed by the lowering, so, 
> eventually remove those and move this function to a local place next to 
> its sole caller.

Yeah.  I couldn't immediately convince myself that all simplifications 
will also be done on local vars (for which record_reference won't run), so 
I'll leave this for later (I'm for instance doubtful about the NULL return 
for when the current unit "can't" refer to a symbol).  FWIW I'm currently 
testing with these asserts which should tell me:

-----------------------------------------------------
@@ -152,7 +151,8 @@ get_symbol_constant_value (tree sym)
       tree val = DECL_INITIAL (sym);
       if (val)
        {
-         val = canonicalize_constructor_val (val);
+         tree newval = canonicalize_constructor_val (val);
+         gcc_assert (newval == val);
          if (val && is_gimple_min_invariant (val))
            return val;
          else
@@ -3164,7 +3164,11 @@ fold_ctor_reference (tree type, tree cto
   /* We found the field with exact match.  */
   if (useless_type_conversion_p (type, TREE_TYPE (ctor))
       && !offset)
-    return canonicalize_constructor_val (ctor);
+    {
+      tree newctor = canonicalize_constructor_val (ctor);
+      gcc_assert (newctor == ctor);
+      return newctor;
+    }

   /* We are at the end of walk, see if we can view convert the
      result.  */
@@ -3174,6 +3178,7 @@ fold_ctor_reference (tree type, tree cto
                          TYPE_SIZE (TREE_TYPE (ctor)), 0))
     {
       ret = canonicalize_constructor_val (ctor);
+      gcc_assert (ret == ctor);
       ret = fold_unary (VIEW_CONVERT_EXPR, type, ret);
       if (ret)
        STRIP_NOPS (ret);
-----------------------------------------------------

In the meanwhile, is the below version okay?


Ciao,
Michael.
-- 
	* cgraphbuild.c (record_reference): Canonicalize constructor
	values.
	* gimple-fold.c (canonicalize_constructor_val): Accept being called
	without function context.

java/
	* jcf-parse.c (java_emit_static_constructor): Clear cfun and
	current_function_decl.

Index: cgraphbuild.c
===================================================================
--- cgraphbuild.c.orig	2011-03-29 17:08:15.000000000 +0200
+++ cgraphbuild.c	2011-03-31 17:43:29.000000000 +0200
@@ -53,6 +53,12 @@ record_reference (tree *tp, int *walk_su
   tree decl;
   struct record_reference_ctx *ctx = (struct record_reference_ctx *)data;
 
+  t = canonicalize_constructor_val (t);
+  if (!t)
+    t = *tp;
+  else if (t != *tp)
+    *tp = t;
+
   switch (TREE_CODE (t))
     {
     case VAR_DECL:
Index: gimple-fold.c
===================================================================
--- gimple-fold.c.orig	2011-03-29 17:08:15.000000000 +0200
+++ gimple-fold.c	2011-03-31 17:42:45.000000000 +0200
@@ -106,7 +106,7 @@ can_refer_decl_in_current_unit_p (tree d
   return true;
 }
 
-/* CVAL is value taken from DECL_INITIAL of variable.  Try to transorm it into
+/* CVAL is value taken from DECL_INITIAL of variable.  Try to transform it into
    acceptable form for is_gimple_min_invariant.   */
 
 tree
@@ -131,10 +131,9 @@ canonicalize_constructor_val (tree cval)
 	      || TREE_CODE (base) == FUNCTION_DECL)
 	  && !can_refer_decl_in_current_unit_p (base))
 	return NULL_TREE;
-      if (base && TREE_CODE (base) == VAR_DECL)
+      if (cfun && base && TREE_CODE (base) == VAR_DECL)
 	add_referenced_var (base);
-      /* We never have the chance to fixup types in global initializers
-         during gimplification.  Do so here.  */
+      /* Fixup types in global initializers.  */
       if (TREE_TYPE (TREE_TYPE (cval)) != TREE_TYPE (TREE_OPERAND (cval, 0)))
 	cval = build_fold_addr_expr (TREE_OPERAND (cval, 0));
     }
Index: java/jcf-parse.c
===================================================================
--- java/jcf-parse.c.orig	2011-03-29 17:08:15.000000000 +0200
+++ java/jcf-parse.c	2011-03-29 17:12:05.000000000 +0200
@@ -1725,6 +1725,8 @@ java_emit_static_constructor (void)
       DECL_STATIC_CONSTRUCTOR (decl) = 1;
       java_genericize (decl);
       cgraph_finalize_function (decl, false);
+      current_function_decl = NULL;
+      set_cfun (NULL);
     }
 }
 

Re: Random cleanups [2/4]: canonicalize ctor values

[Richard Guenther]

On Thu, Mar 31, 2011 at 5:57 PM, Michael Matz <matz@suse.de> wrote:
> On Thu, 31 Mar 2011, Richard Guenther wrote:
>
>> > canonicalize_constructor_val may be doing useful things on ADDR_EXPR
>> > too, but you don't call it in that case because you only added your
>> > code in the default case.  You only reach it when the ADDR_EXPR is
>> > wrapped in a cast.
>> >
>> > Any reason why you didn't add it (possibly wrapped in a loop) _before_
>> > the switch statement?
>
> Nope, probably I didn't want to pay the cost each time, but you're right.
>
>> Indeed.  This comment needs adjustment, too:
>>
>>       /* We never have the chance to fixup types in global initializers
>>          during gimplification.  Do so here.  */
>
> Done in new patch.
>
>> as we now sort-of do this.  In fact it looks like all existing
>> canonicalize_constructor_val calls are subsumed by the lowering, so,
>> eventually remove those and move this function to a local place next to
>> its sole caller.
>
> Yeah.  I couldn't immediately convince myself that all simplifications
> will also be done on local vars (for which record_reference won't run), so
> I'll leave this for later (I'm for instance doubtful about the NULL return
> for when the current unit "can't" refer to a symbol).  FWIW I'm currently
> testing with these asserts which should tell me:
>
> -----------------------------------------------------
> @@ -152,7 +151,8 @@ get_symbol_constant_value (tree sym)
>       tree val = DECL_INITIAL (sym);
>       if (val)
>        {
> -         val = canonicalize_constructor_val (val);
> +         tree newval = canonicalize_constructor_val (val);
> +         gcc_assert (newval == val);
>          if (val && is_gimple_min_invariant (val))
>            return val;
>          else
> @@ -3164,7 +3164,11 @@ fold_ctor_reference (tree type, tree cto
>   /* We found the field with exact match.  */
>   if (useless_type_conversion_p (type, TREE_TYPE (ctor))
>       && !offset)
> -    return canonicalize_constructor_val (ctor);
> +    {
> +      tree newctor = canonicalize_constructor_val (ctor);
> +      gcc_assert (newctor == ctor);
> +      return newctor;
> +    }
>
>   /* We are at the end of walk, see if we can view convert the
>      result.  */
> @@ -3174,6 +3178,7 @@ fold_ctor_reference (tree type, tree cto
>                          TYPE_SIZE (TREE_TYPE (ctor)), 0))
>     {
>       ret = canonicalize_constructor_val (ctor);
> +      gcc_assert (ret == ctor);
>       ret = fold_unary (VIEW_CONVERT_EXPR, type, ret);
>       if (ret)
>        STRIP_NOPS (ret);
> -----------------------------------------------------
>
> In the meanwhile, is the below version okay?

If it bootstraps & tests ok then yes.  The java parts look obvious.

Thanks,
Richard.

>
> Ciao,
> Michael.
> --
>        * cgraphbuild.c (record_reference): Canonicalize constructor
>        values.
>        * gimple-fold.c (canonicalize_constructor_val): Accept being called
>        without function context.
>
> java/
>        * jcf-parse.c (java_emit_static_constructor): Clear cfun and
>        current_function_decl.
>
> Index: cgraphbuild.c
> ===================================================================
> --- cgraphbuild.c.orig  2011-03-29 17:08:15.000000000 +0200
> +++ cgraphbuild.c       2011-03-31 17:43:29.000000000 +0200
> @@ -53,6 +53,12 @@ record_reference (tree *tp, int *walk_su
>   tree decl;
>   struct record_reference_ctx *ctx = (struct record_reference_ctx *)data;
>
> +  t = canonicalize_constructor_val (t);
> +  if (!t)
> +    t = *tp;
> +  else if (t != *tp)
> +    *tp = t;
> +
>   switch (TREE_CODE (t))
>     {
>     case VAR_DECL:
> Index: gimple-fold.c
> ===================================================================
> --- gimple-fold.c.orig  2011-03-29 17:08:15.000000000 +0200
> +++ gimple-fold.c       2011-03-31 17:42:45.000000000 +0200
> @@ -106,7 +106,7 @@ can_refer_decl_in_current_unit_p (tree d
>   return true;
>  }
>
> -/* CVAL is value taken from DECL_INITIAL of variable.  Try to transorm it into
> +/* CVAL is value taken from DECL_INITIAL of variable.  Try to transform it into
>    acceptable form for is_gimple_min_invariant.   */
>
>  tree
> @@ -131,10 +131,9 @@ canonicalize_constructor_val (tree cval)
>              || TREE_CODE (base) == FUNCTION_DECL)
>          && !can_refer_decl_in_current_unit_p (base))
>        return NULL_TREE;
> -      if (base && TREE_CODE (base) == VAR_DECL)
> +      if (cfun && base && TREE_CODE (base) == VAR_DECL)
>        add_referenced_var (base);
> -      /* We never have the chance to fixup types in global initializers
> -         during gimplification.  Do so here.  */
> +      /* Fixup types in global initializers.  */
>       if (TREE_TYPE (TREE_TYPE (cval)) != TREE_TYPE (TREE_OPERAND (cval, 0)))
>        cval = build_fold_addr_expr (TREE_OPERAND (cval, 0));
>     }
> Index: java/jcf-parse.c
> ===================================================================
> --- java/jcf-parse.c.orig       2011-03-29 17:08:15.000000000 +0200
> +++ java/jcf-parse.c    2011-03-29 17:12:05.000000000 +0200
> @@ -1725,6 +1725,8 @@ java_emit_static_constructor (void)
>       DECL_STATIC_CONSTRUCTOR (decl) = 1;
>       java_genericize (decl);
>       cgraph_finalize_function (decl, false);
> +      current_function_decl = NULL;
> +      set_cfun (NULL);
>     }
>  }
>

Re: Random cleanups [2/4]: canonicalize ctor values

[Michael Matz]

Hi,

On Thu, 31 Mar 2011, Richard Guenther wrote:

> > In the meanwhile, is the below version okay?
> 
> If it bootstraps & tests ok then yes.  The java parts look obvious.

So, we indeed can't remove the other calls to 
canonicalize_constructor_val, because of local ctors.  And fortran has a 
similar problem with java.  Instead of fixing up all these places of 
resetting cfun (where otherwise the frontends don't deal at all with it, 
it's mostly just set from the various cgraph routines), I decided to 
simply clear this at the appropriate place in 
cgraph_finalize_compilation_unit.

Regstrapping in progress again.  Still okay if that works?


Ciao,
Michael.
-- 
	* cgraphbuild.c (record_reference): Canonicalize constructor
	values.
	* gimple-fold.c (canonicalize_constructor_val): Accept being called
	without function context.
	* cgraphunit.c (cgraph_finalize_compilation_unit): Clear
	current_function_decl and cfun.

Index: cgraphbuild.c
===================================================================
--- cgraphbuild.c.orig	2011-04-03 11:28:45.000000000 +0200
+++ cgraphbuild.c	2011-04-03 11:31:21.000000000 +0200
@@ -53,6 +53,12 @@ record_reference (tree *tp, int *walk_su
   tree decl;
   struct record_reference_ctx *ctx = (struct record_reference_ctx *)data;
 
+  t = canonicalize_constructor_val (t);
+  if (!t)
+    t = *tp;
+  else if (t != *tp)
+    *tp = t;
+
   switch (TREE_CODE (t))
     {
     case VAR_DECL:
Index: gimple-fold.c
===================================================================
--- gimple-fold.c.orig	2011-04-03 11:28:45.000000000 +0200
+++ gimple-fold.c	2011-04-03 11:31:21.000000000 +0200
@@ -106,7 +106,7 @@ can_refer_decl_in_current_unit_p (tree d
   return true;
 }
 
-/* CVAL is value taken from DECL_INITIAL of variable.  Try to transorm it into
+/* CVAL is value taken from DECL_INITIAL of variable.  Try to transform it into
    acceptable form for is_gimple_min_invariant.   */
 
 tree
@@ -131,10 +131,9 @@ canonicalize_constructor_val (tree cval)
 	      || TREE_CODE (base) == FUNCTION_DECL)
 	  && !can_refer_decl_in_current_unit_p (base))
 	return NULL_TREE;
-      if (base && TREE_CODE (base) == VAR_DECL)
+      if (cfun && base && TREE_CODE (base) == VAR_DECL)
 	add_referenced_var (base);
-      /* We never have the chance to fixup types in global initializers
-         during gimplification.  Do so here.  */
+      /* Fixup types in global initializers.  */
       if (TREE_TYPE (TREE_TYPE (cval)) != TREE_TYPE (TREE_OPERAND (cval, 0)))
 	cval = build_fold_addr_expr (TREE_OPERAND (cval, 0));
     }
Index: cgraphunit.c
===================================================================
--- cgraphunit.c.orig	2011-04-03 11:41:46.000000000 +0200
+++ cgraphunit.c	2011-04-03 11:50:01.000000000 +0200
@@ -1066,6 +1066,11 @@ cgraph_finalize_compilation_unit (void)
 {
   timevar_push (TV_CGRAPH);
 
+  /* If we're here there's no current function anymore.  Some frontends
+     are lazy in clearing these.  */
+  current_function_decl = NULL;
+  set_cfun (NULL);
+
   /* Do not skip analyzing the functions if there were errors, we
      miss diagnostics for following functions otherwise.  */
 

Re: Random cleanups [2/4]: canonicalize ctor values

[Richard Guenther]

On Sun, Apr 3, 2011 at 12:14 PM, Michael Matz <matz@suse.de> wrote:
> Hi,
>
> On Thu, 31 Mar 2011, Richard Guenther wrote:
>
>> > In the meanwhile, is the below version okay?
>>
>> If it bootstraps & tests ok then yes.  The java parts look obvious.
>
> So, we indeed can't remove the other calls to
> canonicalize_constructor_val, because of local ctors.  And fortran has a
> similar problem with java.  Instead of fixing up all these places of
> resetting cfun (where otherwise the frontends don't deal at all with it,
> it's mostly just set from the various cgraph routines), I decided to
> simply clear this at the appropriate place in
> cgraph_finalize_compilation_unit.
>
> Regstrapping in progress again.  Still okay if that works?

Ok.

Thanks,
Richard.

>
> Ciao,
> Michael.
> --
>        * cgraphbuild.c (record_reference): Canonicalize constructor
>        values.
>        * gimple-fold.c (canonicalize_constructor_val): Accept being called
>        without function context.
>        * cgraphunit.c (cgraph_finalize_compilation_unit): Clear
>        current_function_decl and cfun.
>
> Index: cgraphbuild.c
> ===================================================================
> --- cgraphbuild.c.orig  2011-04-03 11:28:45.000000000 +0200
> +++ cgraphbuild.c       2011-04-03 11:31:21.000000000 +0200
> @@ -53,6 +53,12 @@ record_reference (tree *tp, int *walk_su
>   tree decl;
>   struct record_reference_ctx *ctx = (struct record_reference_ctx *)data;
>
> +  t = canonicalize_constructor_val (t);
> +  if (!t)
> +    t = *tp;
> +  else if (t != *tp)
> +    *tp = t;
> +
>   switch (TREE_CODE (t))
>     {
>     case VAR_DECL:
> Index: gimple-fold.c
> ===================================================================
> --- gimple-fold.c.orig  2011-04-03 11:28:45.000000000 +0200
> +++ gimple-fold.c       2011-04-03 11:31:21.000000000 +0200
> @@ -106,7 +106,7 @@ can_refer_decl_in_current_unit_p (tree d
>   return true;
>  }
>
> -/* CVAL is value taken from DECL_INITIAL of variable.  Try to transorm it into
> +/* CVAL is value taken from DECL_INITIAL of variable.  Try to transform it into
>    acceptable form for is_gimple_min_invariant.   */
>
>  tree
> @@ -131,10 +131,9 @@ canonicalize_constructor_val (tree cval)
>              || TREE_CODE (base) == FUNCTION_DECL)
>          && !can_refer_decl_in_current_unit_p (base))
>        return NULL_TREE;
> -      if (base && TREE_CODE (base) == VAR_DECL)
> +      if (cfun && base && TREE_CODE (base) == VAR_DECL)
>        add_referenced_var (base);
> -      /* We never have the chance to fixup types in global initializers
> -         during gimplification.  Do so here.  */
> +      /* Fixup types in global initializers.  */
>       if (TREE_TYPE (TREE_TYPE (cval)) != TREE_TYPE (TREE_OPERAND (cval, 0)))
>        cval = build_fold_addr_expr (TREE_OPERAND (cval, 0));
>     }
> Index: cgraphunit.c
> ===================================================================
> --- cgraphunit.c.orig   2011-04-03 11:41:46.000000000 +0200
> +++ cgraphunit.c        2011-04-03 11:50:01.000000000 +0200
> @@ -1066,6 +1066,11 @@ cgraph_finalize_compilation_unit (void)
>  {
>   timevar_push (TV_CGRAPH);
>
> +  /* If we're here there's no current function anymore.  Some frontends
> +     are lazy in clearing these.  */
> +  current_function_decl = NULL;
> +  set_cfun (NULL);
> +
>   /* Do not skip analyzing the functions if there were errors, we
>      miss diagnostics for following functions otherwise.  */
>
>

Re: Random cleanups [2/4]: canonicalize ctor values

[H.J. Lu]

On Sun, Apr 3, 2011 at 3:14 AM, Michael Matz <matz@suse.de> wrote:
> Hi,
>
> On Thu, 31 Mar 2011, Richard Guenther wrote:
>
>> > In the meanwhile, is the below version okay?
>>
>> If it bootstraps & tests ok then yes.  The java parts look obvious.
>
> So, we indeed can't remove the other calls to
> canonicalize_constructor_val, because of local ctors.  And fortran has a
> similar problem with java.  Instead of fixing up all these places of
> resetting cfun (where otherwise the frontends don't deal at all with it,
> it's mostly just set from the various cgraph routines), I decided to
> simply clear this at the appropriate place in
> cgraph_finalize_compilation_unit.
>
> Regstrapping in progress again.  Still okay if that works?
>
>
> Ciao,
> Michael.
> --
>        * cgraphbuild.c (record_reference): Canonicalize constructor
>        values.

>
> Index: cgraphbuild.c
> ===================================================================
> --- cgraphbuild.c.orig  2011-04-03 11:28:45.000000000 +0200
> +++ cgraphbuild.c       2011-04-03 11:31:21.000000000 +0200
> @@ -53,6 +53,12 @@ record_reference (tree *tp, int *walk_su
>   tree decl;
>   struct record_reference_ctx *ctx = (struct record_reference_ctx *)data;
>
> +  t = canonicalize_constructor_val (t);
> +  if (!t)
> +    t = *tp;
> +  else if (t != *tp)
> +    *tp = t;
> +
>   switch (TREE_CODE (t))
>     {
>     case VAR_DECL:

This change caused:

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48440

-- 
H.J.