GCC 4.2.0 release.

1 files changed, 1453 insertions, 1133 deletions

diff --git a/contrib/gcc/cgraphunit.c b/contrib/gcc/cgraphunit.c
index 75ad3c8..0d3ed47 100644
--- a/contrib/gcc/cgraphunit.c
+++ b/contrib/gcc/cgraphunit.c
@@ -1,5 +1,5 @@
-/* Callgraph based intraprocedural optimizations.
-   Copyright (C) 2003, 2004 Free Software Foundation, Inc.
+/* Callgraph based interprocedural optimizations.
+   Copyright (C) 2003, 2004, 2005 Free Software Foundation, Inc.
    Contributed by Jan Hubicka
 
 This file is part of GCC.
@@ -16,17 +16,139 @@ for more details.
 
 You should have received a copy of the GNU General Public License
 along with GCC; see the file COPYING.  If not, write to the Free
-Software Foundation, 59 Temple Place - Suite 330, Boston, MA
-02111-1307, USA.  */
+Software Foundation, 51 Franklin Street, Fifth Floor, Boston, MA
+02110-1301, USA.  */
+
+/* This module implements main driver of compilation process as well as
+   few basic interprocedural optimizers.
+
+   The main scope of this file is to act as an interface in between
+   tree based frontends and the backend (and middle end)
+
+   The front-end is supposed to use following functionality:
+
+    - cgraph_finalize_function
+
+      This function is called once front-end has parsed whole body of function
+      and it is certain that the function body nor the declaration will change.
+
+      (There is one exception needed for implementing GCC extern inline function.)
+
+    - cgraph_varpool_finalize_variable
+
+      This function has same behavior as the above but is used for static
+      variables.
+
+    - cgraph_finalize_compilation_unit
+
+      This function is called once compilation unit is finalized and it will
+      no longer change.
+
+      In the unit-at-a-time the call-graph construction and local function
+      analysis takes place here.  Bodies of unreachable functions are released
+      to conserve memory usage.
+
+      ???  The compilation unit in this point of view should be compilation
+      unit as defined by the language - for instance C frontend allows multiple
+      compilation units to be parsed at once and it should call function each
+      time parsing is done so we save memory.
+
+    - cgraph_optimize
+
+      In this unit-at-a-time compilation the intra procedural analysis takes
+      place here.  In particular the static functions whose address is never
+      taken are marked as local.  Backend can then use this information to
+      modify calling conventions, do better inlining or similar optimizations.
+
+    - cgraph_assemble_pending_functions
+    - cgraph_varpool_assemble_pending_variables
+
+      In non-unit-at-a-time mode these functions can be used to force compilation
+      of functions or variables that are known to be needed at given stage
+      of compilation
+
+    - cgraph_mark_needed_node
+    - cgraph_varpool_mark_needed_node
+
+      When function or variable is referenced by some hidden way (for instance
+      via assembly code and marked by attribute "used"), the call-graph data structure
+      must be updated accordingly by this function.
+
+    - analyze_expr callback
+
+      This function is responsible for lowering tree nodes not understood by
+      generic code into understandable ones or alternatively marking
+      callgraph and varpool nodes referenced by the as needed.
+
+      ??? On the tree-ssa genericizing should take place here and we will avoid
+      need for these hooks (replacing them by genericizing hook)
+
+    - expand_function callback
+
+      This function is used to expand function and pass it into RTL back-end.
+      Front-end should not make any assumptions about when this function can be
+      called.  In particular cgraph_assemble_pending_functions,
+      cgraph_varpool_assemble_pending_variables, cgraph_finalize_function,
+      cgraph_varpool_finalize_function, cgraph_optimize can cause arbitrarily
+      previously finalized functions to be expanded.
+
+    We implement two compilation modes.
+
+      - unit-at-a-time:  In this mode analyzing of all functions is deferred
+	to cgraph_finalize_compilation_unit and expansion into cgraph_optimize.
+
+	In cgraph_finalize_compilation_unit the reachable functions are
+	analyzed.  During analysis the call-graph edges from reachable
+	functions are constructed and their destinations are marked as
+	reachable.  References to functions and variables are discovered too
+	and variables found to be needed output to the assembly file.  Via
+	mark_referenced call in assemble_variable functions referenced by
+	static variables are noticed too.
+
+	The intra-procedural information is produced and its existence
+	indicated by global_info_ready.  Once this flag is set it is impossible
+	to change function from !reachable to reachable and thus
+	assemble_variable no longer call mark_referenced.
+
+	Finally the call-graph is topologically sorted and all reachable functions
+	that has not been completely inlined or are not external are output.
+
+	??? It is possible that reference to function or variable is optimized
+	out.  We can not deal with this nicely because topological order is not
+	suitable for it.  For tree-ssa we may consider another pass doing
+	optimization and re-discovering reachable functions.
+
+	??? Reorganize code so variables are output very last and only if they
+	really has been referenced by produced code, so we catch more cases
+	where reference has been optimized out.
+
+      - non-unit-at-a-time
+
+	All functions are variables are output as early as possible to conserve
+	memory consumption.  This may or may not result in less memory used but
+	it is still needed for some legacy code that rely on particular ordering
+	of things output from the compiler.
+
+	Varpool data structures are not used and variables are output directly.
+
+	Functions are output early using call of
+	cgraph_assemble_pending_function from cgraph_finalize_function.  The
+	decision on whether function is needed is made more conservative so
+	uninlininable static functions are needed too.  During the call-graph
+	construction the edge destinations are not marked as reachable and it
+	is completely relied upn assemble_variable to mark them.  */
+
 
 #include "config.h"
 #include "system.h"
 #include "coretypes.h"
 #include "tm.h"
 #include "tree.h"
+#include "rtl.h"
+#include "tree-flow.h"
 #include "tree-inline.h"
 #include "langhooks.h"
-#include "hashtab.h"
+#include "pointer-set.h"
 #include "toplev.h"
 #include "flags.h"
 #include "ggc.h"
@@ -40,30 +162,28 @@ Software Foundation, 59 Temple Place - Suite 330, Boston, MA
 #include "c-common.h"
 #include "intl.h"
 #include "function.h"
-
-#define INSNS_PER_CALL 10
+#include "ipa-prop.h"
+#include "tree-gimple.h"
+#include "tree-pass.h"
+#include "output.h"
 
 static void cgraph_expand_all_functions (void);
 static void cgraph_mark_functions_to_output (void);
 static void cgraph_expand_function (struct cgraph_node *);
-static tree record_call_1 (tree *, int *, void *);
-static void cgraph_mark_local_functions (void);
-static void cgraph_optimize_function (struct cgraph_node *);
-static bool cgraph_default_inline_p (struct cgraph_node *n);
-static void cgraph_analyze_function (struct cgraph_node *node);
-static void cgraph_decide_inlining_incrementally (struct cgraph_node *);
-
-/* Statistics we collect about inlining algorithm.  */
-static int ncalls_inlined;
-static int nfunctions_inlined;
-static int initial_insns;
-static int overall_insns;
-
-/* Records tree nodes seen in cgraph_create_edges.  Simply using
+static tree record_reference (tree *, int *, void *);
+static void cgraph_output_pending_asms (void);
+static void cgraph_increase_alignment (void);
+
+/* Lists all assembled variables to be sent to debugger output later on.  */
+static GTY(()) struct cgraph_varpool_node *cgraph_varpool_assembled_nodes_queue;
+
+/* Records tree nodes seen in record_reference.  Simply using
    walk_tree_without_duplicates doesn't guarantee each node is visited
    once because it gets a new htab upon each recursive call from
-   record_calls_1.  */
-static htab_t visited_nodes;
+   record_reference itself.  */
+static struct pointer_set_t *visited_nodes;
+
+static FILE *cgraph_dump_file;
 
 /* Determine if function DECL is needed.  That is, visible to something
    either outside this translation unit, something magic in the system
@@ -73,6 +193,28 @@ static htab_t visited_nodes;
 static bool
 decide_is_function_needed (struct cgraph_node *node, tree decl)
 {
+  tree origin;
+  if (MAIN_NAME_P (DECL_NAME (decl))
+      && TREE_PUBLIC (decl))
+    {
+      node->local.externally_visible = true;
+      return true;
+    }
+
+  /* If the user told us it is used, then it must be so.  */
+  if (node->local.externally_visible)
+    return true;
+
+  if (!flag_unit_at_a_time && lookup_attribute ("used", DECL_ATTRIBUTES (decl)))
+    return true;
+
+  /* ??? If the assembler name is set by hand, it is possible to assemble
+     the name later after finalizing the function and the fact is noticed
+     in assemble_name then.  This is arguably a bug.  */
+  if (DECL_ASSEMBLER_NAME_SET_P (decl)
+      && TREE_SYMBOL_REFERENCED (DECL_ASSEMBLER_NAME (decl)))
+    return true;
+
   /* If we decided it was needed before, but at the time we didn't have
      the body of the function available, then it's still needed.  We have
      to go back and re-check its dependencies now.  */
@@ -80,8 +222,19 @@ decide_is_function_needed (struct cgraph_node *node, tree decl)
     return true;
 
   /* Externally visible functions must be output.  The exception is
-     COMDAT functions that must be output only when they are needed.  */
-  if (TREE_PUBLIC (decl) && !DECL_COMDAT (decl) && !DECL_EXTERNAL (decl))
+     COMDAT functions that must be output only when they are needed.
+
+     When not optimizing, also output the static functions. (see
+     PR24561), but don't do so for always_inline functions, functions
+     declared inline and nested functions.  These was optimized out
+     in the original implementation and it is unclear whether we want
+     to change the behavior here.  */
+  if (((TREE_PUBLIC (decl)
+	|| (!optimize && !node->local.disregard_inline_limits
+	    && !DECL_DECLARED_INLINE_P (decl)
+	    && !node->origin))
+      && !flag_whole_program)
+      && !DECL_COMDAT (decl) && !DECL_EXTERNAL (decl))
     return true;
 
   /* Constructors and destructors are reachable from the runtime by
@@ -89,17 +242,6 @@ decide_is_function_needed (struct cgraph_node *node, tree decl)
   if (DECL_STATIC_CONSTRUCTOR (decl) || DECL_STATIC_DESTRUCTOR (decl))
     return true;
 
-  /* If the user told us it is used, then it must be so.  */
-  if (lookup_attribute ("used", DECL_ATTRIBUTES (decl)))
-    return true;
-
-  /* ??? If the assembler name is set by hand, it is possible to assemble
-     the name later after finalizing the function and the fact is noticed
-     in assemble_name then.  This is arguably a bug.  */
-  if (DECL_ASSEMBLER_NAME_SET_P (decl)
-      && TREE_SYMBOL_REFERENCED (DECL_ASSEMBLER_NAME (decl)))
-    return true;
-
   if (flag_unit_at_a_time)
     return false;
 
@@ -109,6 +251,12 @@ decide_is_function_needed (struct cgraph_node *node, tree decl)
   /* "extern inline" functions are never output locally.  */
   if (DECL_EXTERNAL (decl))
     return false;
+  /* Nested functions of extern inline function shall not be emit unless
+     we inlined the origin.  */
+  for (origin = decl_function_context (decl); origin;
+       origin = decl_function_context (origin))
+    if (DECL_EXTERNAL (origin))
+      return false;
   /* We want to emit COMDAT functions only when absolutely necessary.  */
   if (DECL_COMDAT (decl))
     return false;
@@ -116,13 +264,90 @@ decide_is_function_needed (struct cgraph_node *node, tree decl)
       || (!node->local.disregard_inline_limits
 	  /* When declared inline, defer even the uninlinable functions.
 	     This allows them to be eliminated when unused.  */
-	  && !DECL_DECLARED_INLINE_P (decl) 
-	  && (!node->local.inlinable || !cgraph_default_inline_p (node))))
+	  && !DECL_DECLARED_INLINE_P (decl)
+	  && (!node->local.inlinable || !cgraph_default_inline_p (node, NULL))))
     return true;
 
   return false;
 }
 
+/* Walk the decls we marked as necessary and see if they reference new
+   variables or functions and add them into the worklists.  */
+static bool
+cgraph_varpool_analyze_pending_decls (void)
+{
+  bool changed = false;
+  timevar_push (TV_CGRAPH);
+
+  while (cgraph_varpool_first_unanalyzed_node)
+    {
+      tree decl = cgraph_varpool_first_unanalyzed_node->decl;
+
+      cgraph_varpool_first_unanalyzed_node->analyzed = true;
+
+      cgraph_varpool_first_unanalyzed_node = cgraph_varpool_first_unanalyzed_node->next_needed;
+
+      /* Compute the alignment early so function body expanders are
+	 already informed about increased alignment.  */
+      align_variable (decl, 0);
+
+      if (DECL_INITIAL (decl))
+	{
+	  visited_nodes = pointer_set_create ();
+	  walk_tree (&DECL_INITIAL (decl), record_reference, NULL, visited_nodes);
+	  pointer_set_destroy (visited_nodes);
+	  visited_nodes = NULL;
+	}
+      changed = true;
+    }
+  timevar_pop (TV_CGRAPH);
+  return changed;
+}
+
+/* Optimization of function bodies might've rendered some variables as
+   unnecessary so we want to avoid these from being compiled.
+
+   This is done by pruning the queue and keeping only the variables that
+   really appear needed (ie they are either externally visible or referenced
+   by compiled function). Re-doing the reachability analysis on variables
+   brings back the remaining variables referenced by these.  */
+static void
+cgraph_varpool_remove_unreferenced_decls (void)
+{
+  struct cgraph_varpool_node *next, *node = cgraph_varpool_nodes_queue;
+
+  cgraph_varpool_reset_queue ();
+
+  if (errorcount || sorrycount)
+    return;
+
+  while (node)
+    {
+      tree decl = node->decl;
+      next = node->next_needed;
+      node->needed = 0;
+
+      if (node->finalized
+	  && ((DECL_ASSEMBLER_NAME_SET_P (decl)
+	       && TREE_SYMBOL_REFERENCED (DECL_ASSEMBLER_NAME (decl)))
+	      || node->force_output
+	      || decide_is_variable_needed (node, decl)
+	      /* ??? Cgraph does not yet rule the world with an iron hand,
+		 and does not control the emission of debug information.
+		 After a variable has its DECL_RTL set, we must assume that
+		 it may be referenced by the debug information, and we can
+		 no longer elide it.  */
+	      || DECL_RTL_SET_P (decl)))
+	cgraph_varpool_mark_needed_node (node);
+
+      node = next;
+    }
+  /* Make sure we mark alias targets as used targets.  */
+  finish_aliases_1 ();
+  cgraph_varpool_analyze_pending_decls ();
+}
+
+
 /* When not doing unit-at-a-time, output all functions enqueued.
    Return true when such a functions were found.  */
 
@@ -134,21 +359,105 @@ cgraph_assemble_pending_functions (void)
   if (flag_unit_at_a_time)
     return false;
 
+  cgraph_output_pending_asms ();
+
   while (cgraph_nodes_queue)
     {
       struct cgraph_node *n = cgraph_nodes_queue;
 
       cgraph_nodes_queue = cgraph_nodes_queue->next_needed;
-      if (!n->origin && !DECL_EXTERNAL (n->decl))
+      n->next_needed = NULL;
+      if (!n->global.inlined_to
+	  && !n->alias
+	  && !DECL_EXTERNAL (n->decl))
 	{
 	  cgraph_expand_function (n);
 	  output = true;
 	}
     }
 
+  /* Process CGRAPH_EXPAND_QUEUE, these are functions created during
+     the expansion process.  Note that this queue may grow as its
+     being processed, as the new functions may generate new ones.  */
+  while (cgraph_expand_queue)
+    {
+      struct cgraph_node *n = cgraph_expand_queue;
+      cgraph_expand_queue = cgraph_expand_queue->next_needed;
+      n->next_needed = NULL;
+      cgraph_finalize_function (n->decl, false);
+      output = true;
+    }
+
   return output;
 }
 
+
+/* As an GCC extension we allow redefinition of the function.  The
+   semantics when both copies of bodies differ is not well defined.
+   We replace the old body with new body so in unit at a time mode
+   we always use new body, while in normal mode we may end up with
+   old body inlined into some functions and new body expanded and
+   inlined in others.
+
+   ??? It may make more sense to use one body for inlining and other
+   body for expanding the function but this is difficult to do.  */
+
+static void
+cgraph_reset_node (struct cgraph_node *node)
+{
+  /* If node->output is set, then this is a unit-at-a-time compilation
+     and we have already begun whole-unit analysis.  This is *not*
+     testing for whether we've already emitted the function.  That
+     case can be sort-of legitimately seen with real function
+     redefinition errors.  I would argue that the front end should
+     never present us with such a case, but don't enforce that for now.  */
+  gcc_assert (!node->output);
+
+  /* Reset our data structures so we can analyze the function again.  */
+  memset (&node->local, 0, sizeof (node->local));
+  memset (&node->global, 0, sizeof (node->global));
+  memset (&node->rtl, 0, sizeof (node->rtl));
+  node->analyzed = false;
+  node->local.redefined_extern_inline = true;
+  node->local.finalized = false;
+
+  if (!flag_unit_at_a_time)
+    {
+      struct cgraph_node *n, *next;
+
+      for (n = cgraph_nodes; n; n = next)
+	{
+	  next = n->next;
+	  if (n->global.inlined_to == node)
+	    cgraph_remove_node (n);
+	}
+    }
+
+  cgraph_node_remove_callees (node);
+
+  /* We may need to re-queue the node for assembling in case
+     we already proceeded it and ignored as not needed.  */
+  if (node->reachable && !flag_unit_at_a_time)
+    {
+      struct cgraph_node *n;
+
+      for (n = cgraph_nodes_queue; n; n = n->next_needed)
+	if (n == node)
+	  break;
+      if (!n)
+	node->reachable = 0;
+    }
+}
+
+static void
+cgraph_lower_function (struct cgraph_node *node)
+{
+  if (node->lowered)
+    return;
+  tree_lowering_passes (node->decl);
+  node->lowered = true;
+}
+
 /* DECL has been parsed.  Take it, queue it, compile it at the whim of the
    logic in effect.  If NESTED is true, then our caller cannot stand to have
    the garbage collector run at the moment.  We would need to either create
@@ -160,64 +469,33 @@ cgraph_finalize_function (tree decl, bool nested)
   struct cgraph_node *node = cgraph_node (decl);
 
   if (node->local.finalized)
-    {
-      /* As an GCC extension we allow redefinition of the function.  The
-	 semantics when both copies of bodies differ is not well defined.
-	 We replace the old body with new body so in unit at a time mode
-	 we always use new body, while in normal mode we may end up with
-	 old body inlined into some functions and new body expanded and
-	 inlined in others.
-	 
-	 ??? It may make more sense to use one body for inlining and other
-	 body for expanding the function but this is difficult to do.  */
-
-      /* If node->output is set, then this is a unit-at-a-time compilation
-	 and we have already begun whole-unit analysis.  This is *not*
-	 testing for whether we've already emitted the function.  That
-	 case can be sort-of legitimately seen with real function 
-	 redefinition errors.  I would argue that the front end should
-	 never present us with such a case, but don't enforce that for now.  */
-      if (node->output)
-	abort ();
-
-      /* Reset our datastructures so we can analyze the function again.  */
-      memset (&node->local, 0, sizeof (node->local));
-      memset (&node->global, 0, sizeof (node->global));
-      memset (&node->rtl, 0, sizeof (node->rtl));
-      node->analyzed = false;
-      node->local.redefined_extern_inline = true;
-      while (node->callees)
-	cgraph_remove_edge (node, node->callees->callee);
-
-      /* We may need to re-queue the node for assembling in case
-         we already proceeded it and ignored as not needed.  */
-      if (node->reachable && !flag_unit_at_a_time)
-	{
-	  struct cgraph_node *n;
-
-	  for (n = cgraph_nodes_queue; n; n = n->next_needed)
-	    if (n == node)
-	      break;
-	  if (!n)
-	    node->reachable = 0;
-	}
-    }
+    cgraph_reset_node (node);
 
   notice_global_symbol (decl);
   node->decl = decl;
   node->local.finalized = true;
+  node->lowered = DECL_STRUCT_FUNCTION (decl)->cfg != NULL;
+  if (node->nested)
+    lower_nested_functions (decl);
+  gcc_assert (!node->nested);
 
   /* If not unit at a time, then we need to create the call graph
      now, so that called functions can be queued and emitted now.  */
   if (!flag_unit_at_a_time)
     {
       cgraph_analyze_function (node);
-      cgraph_decide_inlining_incrementally (node);
+      cgraph_decide_inlining_incrementally (node, false);
     }
 
   if (decide_is_function_needed (node, decl))
     cgraph_mark_needed_node (node);
 
+  /* Since we reclaim unreachable nodes at the end of every language
+     level unit, we need to be conservative about possible entry points
+     there.  */
+  if ((TREE_PUBLIC (decl) && !DECL_COMDAT (decl) && !DECL_EXTERNAL (decl)))
+    cgraph_mark_reachable_node (node);
+
   /* If not unit at a time, go ahead and emit everything we've found
      to be reachable at this time.  */
   if (!nested)
@@ -230,15 +508,14 @@ cgraph_finalize_function (tree decl, bool nested)
   if (!TREE_ASM_WRITTEN (decl))
     (*debug_hooks->deferred_inline_function) (decl);
 
-  /* We will never really output the function body, clear the SAVED_INSNS array
-     early then.  */
-  if (DECL_EXTERNAL (decl))
-    DECL_SAVED_INSNS (decl) = NULL;
+  /* Possibly warn about unused parameters.  */
+  if (warn_unused_parameter)
+    do_warn_unused_parameter (decl);
 }
 
 /* Walk tree and record all calls.  Called via walk_tree.  */
 static tree
-record_call_1 (tree *tp, int *walk_subtrees, void *data)
+record_reference (tree *tp, int *walk_subtrees, void *data)
 {
   tree t = *tp;
 
@@ -248,10 +525,16 @@ record_call_1 (tree *tp, int *walk_subtrees, void *data)
       /* ??? Really, we should mark this decl as *potentially* referenced
 	 by this function and re-examine whether the decl is actually used
 	 after rtl has been generated.  */
-      if (TREE_STATIC (t))
-        cgraph_varpool_mark_needed_node (cgraph_varpool_node (t));
+      if (TREE_STATIC (t) || DECL_EXTERNAL (t))
+	{
+	  cgraph_varpool_mark_needed_node (cgraph_varpool_node (t));
+	  if (lang_hooks.callgraph.analyze_expr)
+	    return lang_hooks.callgraph.analyze_expr (tp, walk_subtrees,
+						      data);
+	}
       break;
 
+    case FDESC_EXPR:
     case ADDR_EXPR:
       if (flag_unit_at_a_time)
 	{
@@ -263,104 +546,507 @@ record_call_1 (tree *tp, int *walk_subtrees, void *data)
 	}
       break;
 
-    case CALL_EXPR:
-      {
-	tree decl = get_callee_fndecl (*tp);
-	if (decl && TREE_CODE (decl) == FUNCTION_DECL)
-	  {
-	    cgraph_record_call (data, decl);
-
-	    /* When we see a function call, we don't want to look at the
-	       function reference in the ADDR_EXPR that is hanging from
-	       the CALL_EXPR we're examining here, because we would
-	       conclude incorrectly that the function's address could be
-	       taken by something that is not a function call.  So only
-	       walk the function parameter list, skip the other subtrees.  */
-
-	    walk_tree (&TREE_OPERAND (*tp, 1), record_call_1, data,
-		       visited_nodes);
-	    *walk_subtrees = 0;
-	  }
-	break;
-      }
-
     default:
       /* Save some cycles by not walking types and declaration as we
 	 won't find anything useful there anyway.  */
-      if (DECL_P (*tp) || TYPE_P (*tp))
+      if (IS_TYPE_OR_DECL_P (*tp))
 	{
 	  *walk_subtrees = 0;
 	  break;
 	}
 
       if ((unsigned int) TREE_CODE (t) >= LAST_AND_UNUSED_TREE_CODE)
-	return (*lang_hooks.callgraph.analyze_expr) (tp, walk_subtrees, data);
+	return lang_hooks.callgraph.analyze_expr (tp, walk_subtrees, data);
       break;
     }
 
   return NULL;
 }
 
-/* Create cgraph edges for function calls inside BODY from DECL.  */
+/* Create cgraph edges for function calls inside BODY from NODE.  */
 
-void
-cgraph_create_edges (tree decl, tree body)
+static void
+cgraph_create_edges (struct cgraph_node *node, tree body)
 {
-  /* The nodes we're interested in are never shared, so walk
-     the tree ignoring duplicates.  */
-  visited_nodes = htab_create (37, htab_hash_pointer,
-				    htab_eq_pointer, NULL);
-  walk_tree (&body, record_call_1, decl, visited_nodes);
-  htab_delete (visited_nodes);
+  basic_block bb;
+
+  struct function *this_cfun = DECL_STRUCT_FUNCTION (body);
+  block_stmt_iterator bsi;
+  tree step;
+  visited_nodes = pointer_set_create ();
+
+  /* Reach the trees by walking over the CFG, and note the
+     enclosing basic-blocks in the call edges.  */
+  FOR_EACH_BB_FN (bb, this_cfun)
+    for (bsi = bsi_start (bb); !bsi_end_p (bsi); bsi_next (&bsi))
+      {
+	tree stmt = bsi_stmt (bsi);
+	tree call = get_call_expr_in (stmt);
+	tree decl;
+
+	if (call && (decl = get_callee_fndecl (call)))
+	  {
+	    cgraph_create_edge (node, cgraph_node (decl), stmt,
+				bb->count,
+				bb->loop_depth);
+	    walk_tree (&TREE_OPERAND (call, 1),
+		       record_reference, node, visited_nodes);
+	    if (TREE_CODE (stmt) == MODIFY_EXPR)
+	      walk_tree (&TREE_OPERAND (stmt, 0),
+			 record_reference, node, visited_nodes);
+	  }
+	else
+	  walk_tree (bsi_stmt_ptr (bsi), record_reference, node, visited_nodes);
+      }
+
+  /* Look for initializers of constant variables and private statics.  */
+  for (step = DECL_STRUCT_FUNCTION (body)->unexpanded_var_list;
+       step;
+       step = TREE_CHAIN (step))
+    {
+      tree decl = TREE_VALUE (step);
+      if (TREE_CODE (decl) == VAR_DECL
+	  && (TREE_STATIC (decl) && !DECL_EXTERNAL (decl))
+	  && flag_unit_at_a_time)
+	cgraph_varpool_finalize_decl (decl);
+      else if (TREE_CODE (decl) == VAR_DECL && DECL_INITIAL (decl))
+	walk_tree (&DECL_INITIAL (decl), record_reference, node, visited_nodes);
+    }
+
+  pointer_set_destroy (visited_nodes);
   visited_nodes = NULL;
 }
 
-/* Analyze the function scheduled to be output.  */
+/* Give initial reasons why inlining would fail.  Those gets
+   either NULLified or usually overwritten by more precise reason
+   later.  */
+static void
+initialize_inline_failed (struct cgraph_node *node)
+{
+  struct cgraph_edge *e;
+
+  for (e = node->callers; e; e = e->next_caller)
+    {
+      gcc_assert (!e->callee->global.inlined_to);
+      gcc_assert (e->inline_failed);
+      if (node->local.redefined_extern_inline)
+	e->inline_failed = N_("redefined extern inline functions are not "
+			   "considered for inlining");
+      else if (!node->local.inlinable)
+	e->inline_failed = N_("function not inlinable");
+      else
+	e->inline_failed = N_("function not considered for inlining");
+    }
+}
+
+/* Rebuild call edges from current function after a passes not aware
+   of cgraph updating.  */
+static unsigned int
+rebuild_cgraph_edges (void)
+{
+  basic_block bb;
+  struct cgraph_node *node = cgraph_node (current_function_decl);
+  block_stmt_iterator bsi;
+
+  cgraph_node_remove_callees (node);
+
+  node->count = ENTRY_BLOCK_PTR->count;
+
+  FOR_EACH_BB (bb)
+    for (bsi = bsi_start (bb); !bsi_end_p (bsi); bsi_next (&bsi))
+      {
+	tree stmt = bsi_stmt (bsi);
+	tree call = get_call_expr_in (stmt);
+	tree decl;
+
+	if (call && (decl = get_callee_fndecl (call)))
+	  cgraph_create_edge (node, cgraph_node (decl), stmt,
+			      bb->count,
+			      bb->loop_depth);
+      }
+  initialize_inline_failed (node);
+  gcc_assert (!node->global.inlined_to);
+  return 0;
+}
+
+struct tree_opt_pass pass_rebuild_cgraph_edges =
+{
+  NULL,					/* name */
+  NULL,					/* gate */
+  rebuild_cgraph_edges,			/* execute */
+  NULL,					/* sub */
+  NULL,					/* next */
+  0,					/* static_pass_number */
+  0,					/* tv_id */
+  PROP_cfg,				/* properties_required */
+  0,					/* properties_provided */
+  0,					/* properties_destroyed */
+  0,					/* todo_flags_start */
+  0,					/* todo_flags_finish */
+  0					/* letter */
+};
+
+/* Verify cgraph nodes of given cgraph node.  */
+void
+verify_cgraph_node (struct cgraph_node *node)
+{
+  struct cgraph_edge *e;
+  struct cgraph_node *main_clone;
+  struct function *this_cfun = DECL_STRUCT_FUNCTION (node->decl);
+  basic_block this_block;
+  block_stmt_iterator bsi;
+  bool error_found = false;
+
+  if (errorcount || sorrycount)
+    return;
+
+  timevar_push (TV_CGRAPH_VERIFY);
+  for (e = node->callees; e; e = e->next_callee)
+    if (e->aux)
+      {
+	error ("aux field set for edge %s->%s",
+	       cgraph_node_name (e->caller), cgraph_node_name (e->callee));
+	error_found = true;
+      }
+  if (node->count < 0)
+    {
+      error ("Execution count is negative");
+      error_found = true;
+    }
+  for (e = node->callers; e; e = e->next_caller)
+    {
+      if (e->count < 0)
+	{
+	  error ("caller edge count is negative");
+	  error_found = true;
+	}
+      if (!e->inline_failed)
+	{
+	  if (node->global.inlined_to
+	      != (e->caller->global.inlined_to
+		  ? e->caller->global.inlined_to : e->caller))
+	    {
+	      error ("inlined_to pointer is wrong");
+	      error_found = true;
+	    }
+	  if (node->callers->next_caller)
+	    {
+	      error ("multiple inline callers");
+	      error_found = true;
+	    }
+	}
+      else
+	if (node->global.inlined_to)
+	  {
+	    error ("inlined_to pointer set for noninline callers");
+	    error_found = true;
+	  }
+    }
+  if (!node->callers && node->global.inlined_to)
+    {
+      error ("inlined_to pointer is set but no predecessors found");
+      error_found = true;
+    }
+  if (node->global.inlined_to == node)
+    {
+      error ("inlined_to pointer refers to itself");
+      error_found = true;
+    }
+
+  for (main_clone = cgraph_node (node->decl); main_clone;
+       main_clone = main_clone->next_clone)
+    if (main_clone == node)
+      break;
+  if (!cgraph_node (node->decl))
+    {
+      error ("node not found in cgraph_hash");
+      error_found = true;
+    }
+
+  if (node->analyzed
+      && DECL_SAVED_TREE (node->decl) && !TREE_ASM_WRITTEN (node->decl)
+      && (!DECL_EXTERNAL (node->decl) || node->global.inlined_to))
+    {
+      if (this_cfun->cfg)
+	{
+	  /* The nodes we're interested in are never shared, so walk
+	     the tree ignoring duplicates.  */
+	  visited_nodes = pointer_set_create ();
+	  /* Reach the trees by walking over the CFG, and note the
+	     enclosing basic-blocks in the call edges.  */
+	  FOR_EACH_BB_FN (this_block, this_cfun)
+	    for (bsi = bsi_start (this_block); !bsi_end_p (bsi); bsi_next (&bsi))
+	      {
+		tree stmt = bsi_stmt (bsi);
+		tree call = get_call_expr_in (stmt);
+		tree decl;
+		if (call && (decl = get_callee_fndecl (call)))
+		  {
+		    struct cgraph_edge *e = cgraph_edge (node, stmt);
+		    if (e)
+		      {
+			if (e->aux)
+			  {
+			    error ("shared call_stmt:");
+			    debug_generic_stmt (stmt);
+			    error_found = true;
+			  }
+			if (e->callee->decl != cgraph_node (decl)->decl
+			    && e->inline_failed)
+			  {
+			    error ("edge points to wrong declaration:");
+			    debug_tree (e->callee->decl);
+			    fprintf (stderr," Instead of:");
+			    debug_tree (decl);
+			  }
+			e->aux = (void *)1;
+		      }
+		    else
+		      {
+			error ("missing callgraph edge for call stmt:");
+			debug_generic_stmt (stmt);
+			error_found = true;
+		      }
+		  }
+	      }
+	  pointer_set_destroy (visited_nodes);
+	  visited_nodes = NULL;
+	}
+      else
+	/* No CFG available?!  */
+	gcc_unreachable ();
+
+      for (e = node->callees; e; e = e->next_callee)
+	{
+	  if (!e->aux)
+	    {
+	      error ("edge %s->%s has no corresponding call_stmt",
+		     cgraph_node_name (e->caller),
+		     cgraph_node_name (e->callee));
+	      debug_generic_stmt (e->call_stmt);
+	      error_found = true;
+	    }
+	  e->aux = 0;
+	}
+    }
+  if (error_found)
+    {
+      dump_cgraph_node (stderr, node);
+      internal_error ("verify_cgraph_node failed");
+    }
+  timevar_pop (TV_CGRAPH_VERIFY);
+}
+
+/* Verify whole cgraph structure.  */
+void
+verify_cgraph (void)
+{
+  struct cgraph_node *node;
+
+  if (sorrycount || errorcount)
+    return;
+
+  for (node = cgraph_nodes; node; node = node->next)
+    verify_cgraph_node (node);
+}
+
+/* Output one variable, if necessary.  Return whether we output it.  */
+static bool
+cgraph_varpool_assemble_decl (struct cgraph_varpool_node *node)
+{
+  tree decl = node->decl;
+
+  if (!TREE_ASM_WRITTEN (decl)
+      && !node->alias
+      && !DECL_EXTERNAL (decl)
+      && (TREE_CODE (decl) != VAR_DECL || !DECL_HAS_VALUE_EXPR_P (decl)))
+    {
+      assemble_variable (decl, 0, 1, 0);
+      return TREE_ASM_WRITTEN (decl);
+    }
+
+  return false;
+}
+
+/* Output all variables enqueued to be assembled.  */
+bool
+cgraph_varpool_assemble_pending_decls (void)
+{
+  bool changed = false;
+
+  if (errorcount || sorrycount)
+    return false;
+
+  /* EH might mark decls as needed during expansion.  This should be safe since
+     we don't create references to new function, but it should not be used
+     elsewhere.  */
+  cgraph_varpool_analyze_pending_decls ();
+
+  while (cgraph_varpool_nodes_queue)
+    {
+      struct cgraph_varpool_node *node = cgraph_varpool_nodes_queue;
+
+      cgraph_varpool_nodes_queue = cgraph_varpool_nodes_queue->next_needed;
+      if (cgraph_varpool_assemble_decl (node))
+	{
+	  changed = true;
+	  node->next_needed = cgraph_varpool_assembled_nodes_queue;
+	  cgraph_varpool_assembled_nodes_queue = node;
+	  node->finalized = 1;
+	}
+      else
+        node->next_needed = NULL;
+    }
+  /* cgraph_varpool_nodes_queue is now empty, clear the pointer to the last
+     element in the queue.  */
+  cgraph_varpool_last_needed_node = NULL;
+  return changed;
+}
+/* Output all variables enqueued to be assembled.  */
 static void
+cgraph_varpool_output_debug_info (void)
+{
+  timevar_push (TV_SYMOUT);
+  if (errorcount == 0 && sorrycount == 0)
+    while (cgraph_varpool_assembled_nodes_queue)
+      {
+	struct cgraph_varpool_node *node = cgraph_varpool_assembled_nodes_queue;
+
+	/* Local static variables are never seen by check_global_declarations
+	   so we need to output debug info by hand.  */
+	if (DECL_CONTEXT (node->decl)
+	    && (TREE_CODE (DECL_CONTEXT (node->decl)) == BLOCK
+		|| TREE_CODE (DECL_CONTEXT (node->decl)) == FUNCTION_DECL)
+	    && errorcount == 0 && sorrycount == 0)
+	     (*debug_hooks->global_decl) (node->decl);
+	cgraph_varpool_assembled_nodes_queue = node->next_needed;
+	node->next_needed = 0;
+      }
+  timevar_pop (TV_SYMOUT);
+}
+
+/* Output all asm statements we have stored up to be output.  */
+
+static void
+cgraph_output_pending_asms (void)
+{
+  struct cgraph_asm_node *can;
+
+  if (errorcount || sorrycount)
+    return;
+
+  for (can = cgraph_asm_nodes; can; can = can->next)
+    assemble_asm (can->asm_str);
+  cgraph_asm_nodes = NULL;
+}
+
+/* Analyze the function scheduled to be output.  */
+void
 cgraph_analyze_function (struct cgraph_node *node)
 {
   tree decl = node->decl;
-  struct cgraph_edge *e;
 
   current_function_decl = decl;
+  push_cfun (DECL_STRUCT_FUNCTION (decl));
+  cgraph_lower_function (node);
 
   /* First kill forward declaration so reverse inlining works properly.  */
-  cgraph_create_edges (decl, DECL_SAVED_TREE (decl));
+  cgraph_create_edges (node, decl);
 
   node->local.inlinable = tree_inlinable_function_p (decl);
-  if (!node->local.self_insns)
-    node->local.self_insns
-      = (*lang_hooks.tree_inlining.estimate_num_insns) (decl);
+  if (!flag_unit_at_a_time)
+    node->local.self_insns = estimate_num_insns (decl);
   if (node->local.inlinable)
     node->local.disregard_inline_limits
-      = (*lang_hooks.tree_inlining.disregard_inline_limits) (decl);
-  for (e = node->callers; e; e = e->next_caller)
-    if (e->inline_failed)
-      {
-	if (node->local.redefined_extern_inline)
-	  e->inline_failed = N_("redefined extern inline functions are not "
-				"considered for inlining");
-	else if (!node->local.inlinable)
-	  e->inline_failed = N_("function not inlinable");
-	else
-	  e->inline_failed = N_("function not considered for inlining");
-      }
+      = lang_hooks.tree_inlining.disregard_inline_limits (decl);
+  initialize_inline_failed (node);
   if (flag_really_no_inline && !node->local.disregard_inline_limits)
     node->local.inlinable = 0;
   /* Inlining characteristics are maintained by the cgraph_mark_inline.  */
   node->global.insns = node->local.self_insns;
-  if (!DECL_EXTERNAL (decl))
-    {
-      node->global.cloned_times = 1;
-      node->global.will_be_output = true;
-    }
 
   node->analyzed = true;
+  pop_cfun ();
   current_function_decl = NULL;
+}
 
-  /* Possibly warn about unused parameters.  */
-  if (warn_unused_parameter)
-    do_warn_unused_parameter (decl);
+/* Look for externally_visible and used attributes and mark cgraph nodes
+   accordingly.
+
+   We cannot mark the nodes at the point the attributes are processed (in
+   handle_*_attribute) because the copy of the declarations available at that
+   point may not be canonical.  For example, in:
+
+    void f();
+    void f() __attribute__((used));
+
+   the declaration we see in handle_used_attribute will be the second
+   declaration -- but the front end will subsequently merge that declaration
+   with the original declaration and discard the second declaration.
+
+   Furthermore, we can't mark these nodes in cgraph_finalize_function because:
+
+    void f() {}
+    void f() __attribute__((externally_visible));
+
+   is valid.
+
+   So, we walk the nodes at the end of the translation unit, applying the
+   attributes at that point.  */
+
+static void
+process_function_and_variable_attributes (struct cgraph_node *first,
+                                          struct cgraph_varpool_node *first_var)
+{
+  struct cgraph_node *node;
+  struct cgraph_varpool_node *vnode;
+
+  for (node = cgraph_nodes; node != first; node = node->next)
+    {
+      tree decl = node->decl;
+      if (lookup_attribute ("used", DECL_ATTRIBUTES (decl)))
+	{
+	  mark_decl_referenced (decl);
+	  if (node->local.finalized)
+	     cgraph_mark_needed_node (node);
+	}
+      if (lookup_attribute ("externally_visible", DECL_ATTRIBUTES (decl)))
+	{
+	  if (! TREE_PUBLIC (node->decl))
+	    warning (OPT_Wattributes,
+		     "%J%<externally_visible%> attribute have effect only on public objects",
+		     node->decl);
+	  else
+	    {
+	      if (node->local.finalized)
+		cgraph_mark_needed_node (node);
+	      node->local.externally_visible = true;
+	    }
+	}
+    }
+  for (vnode = cgraph_varpool_nodes; vnode != first_var; vnode = vnode->next)
+    {
+      tree decl = vnode->decl;
+      if (lookup_attribute ("used", DECL_ATTRIBUTES (decl)))
+	{
+	  mark_decl_referenced (decl);
+	  if (vnode->finalized)
+	    cgraph_varpool_mark_needed_node (vnode);
+	}
+      if (lookup_attribute ("externally_visible", DECL_ATTRIBUTES (decl)))
+	{
+	  if (! TREE_PUBLIC (vnode->decl))
+	    warning (OPT_Wattributes,
+		     "%J%<externally_visible%> attribute have effect only on public objects",
+		     vnode->decl);
+	  else
+	    {
+	      if (vnode->finalized)
+		cgraph_varpool_mark_needed_node (vnode);
+	      vnode->externally_visible = true;
+	    }
+	}
+    }
 }
 
 /* Analyze the whole compilation unit once it is parsed completely.  */
@@ -368,23 +1054,42 @@ cgraph_analyze_function (struct cgraph_node *node)
 void
 cgraph_finalize_compilation_unit (void)
 {
-  struct cgraph_node *node;
+  struct cgraph_node *node, *next;
+  /* Keep track of already processed nodes when called multiple times for
+     intermodule optimization.  */
+  static struct cgraph_node *first_analyzed;
+  struct cgraph_node *first_processed = first_analyzed;
+  static struct cgraph_varpool_node *first_analyzed_var;
+
+  if (errorcount || sorrycount)
+    return;
+
+  finish_aliases_1 ();
 
   if (!flag_unit_at_a_time)
     {
+      cgraph_output_pending_asms ();
       cgraph_assemble_pending_functions ();
+      cgraph_varpool_output_debug_info ();
       return;
     }
 
-  cgraph_varpool_assemble_pending_decls ();
   if (!quiet_flag)
-    fprintf (stderr, "\nAnalyzing compilation unit\n");
+    {
+      fprintf (stderr, "\nAnalyzing compilation unit");
+      fflush (stderr);
+    }
 
   timevar_push (TV_CGRAPH);
+  process_function_and_variable_attributes (first_processed,
+					    first_analyzed_var);
+  first_processed = cgraph_nodes;
+  first_analyzed_var = cgraph_varpool_nodes;
+  cgraph_varpool_analyze_pending_decls ();
   if (cgraph_dump_file)
     {
       fprintf (cgraph_dump_file, "Initial entry points:");
-      for (node = cgraph_nodes; node; node = node->next)
+      for (node = cgraph_nodes; node != first_analyzed; node = node->next)
 	if (node->needed && DECL_SAVED_TREE (node->decl))
 	  fprintf (cgraph_dump_file, " %s", cgraph_node_name (node));
       fprintf (cgraph_dump_file, "\n");
@@ -401,15 +1106,19 @@ cgraph_finalize_compilation_unit (void)
 
       node = cgraph_nodes_queue;
       cgraph_nodes_queue = cgraph_nodes_queue->next_needed;
+      node->next_needed = NULL;
 
       /* ??? It is possible to create extern inline function and later using
-	 weak alas attribute to kill it's body. See
+	 weak alias attribute to kill its body. See
 	 gcc.c-torture/compile/20011119-1.c  */
       if (!DECL_SAVED_TREE (decl))
-	continue;
+	{
+	  cgraph_reset_node (node);
+	  continue;
+	}
 
-      if (node->analyzed || !node->reachable || !DECL_SAVED_TREE (decl))
-	abort ();
+      gcc_assert (!node->analyzed && node->reachable);
+      gcc_assert (DECL_SAVED_TREE (decl));
 
       cgraph_analyze_function (node);
 
@@ -417,15 +1126,20 @@ cgraph_finalize_compilation_unit (void)
 	if (!edge->callee->reachable)
 	  cgraph_mark_reachable_node (edge->callee);
 
-      cgraph_varpool_assemble_pending_decls ();
+      /* We finalize local static variables during constructing callgraph
+         edges.  Process their attributes too.  */
+      process_function_and_variable_attributes (first_processed,
+						first_analyzed_var);
+      first_processed = cgraph_nodes;
+      first_analyzed_var = cgraph_varpool_nodes;
+      cgraph_varpool_analyze_pending_decls ();
     }
 
   /* Collect entry points to the unit.  */
-
   if (cgraph_dump_file)
     {
       fprintf (cgraph_dump_file, "Unit entry points:");
-      for (node = cgraph_nodes; node; node = node->next)
+      for (node = cgraph_nodes; node != first_analyzed; node = node->next)
 	if (node->needed && DECL_SAVED_TREE (node->decl))
 	  fprintf (cgraph_dump_file, " %s", cgraph_node_name (node));
       fprintf (cgraph_dump_file, "\n\nInitial ");
@@ -435,28 +1149,35 @@ cgraph_finalize_compilation_unit (void)
   if (cgraph_dump_file)
     fprintf (cgraph_dump_file, "\nReclaiming functions:");
 
-  for (node = cgraph_nodes; node; node = node->next)
+  for (node = cgraph_nodes; node != first_analyzed; node = next)
     {
       tree decl = node->decl;
+      next = node->next;
+
+      if (node->local.finalized && !DECL_SAVED_TREE (decl))
+	cgraph_reset_node (node);
 
       if (!node->reachable && DECL_SAVED_TREE (decl))
 	{
-	  cgraph_remove_node (node);
 	  if (cgraph_dump_file)
 	    fprintf (cgraph_dump_file, " %s", cgraph_node_name (node));
+	  cgraph_remove_node (node);
+	  continue;
 	}
       else
 	node->next_needed = NULL;
+      gcc_assert (!node->local.finalized || DECL_SAVED_TREE (decl));
+      gcc_assert (node->analyzed == node->local.finalized);
     }
   if (cgraph_dump_file)
     {
       fprintf (cgraph_dump_file, "\n\nReclaimed ");
       dump_cgraph (cgraph_dump_file);
     }
+  first_analyzed = cgraph_nodes;
   ggc_collect ();
   timevar_pop (TV_CGRAPH);
 }
-
 /* Figure out what functions we want to assemble.  */
 
 static void
@@ -469,8 +1190,7 @@ cgraph_mark_functions_to_output (void)
       tree decl = node->decl;
       struct cgraph_edge *e;
 
-      if (node->output)
-	abort ();
+      gcc_assert (!node->output);
 
       for (e = node->callers; e; e = e->next_caller)
 	if (e->inline_failed)
@@ -480,45 +1200,29 @@ cgraph_mark_functions_to_output (void)
 	 always inlined, as well as those that are reachable from
 	 outside the current compilation unit.  */
       if (DECL_SAVED_TREE (decl)
+	  && !node->global.inlined_to
 	  && (node->needed
 	      || (e && node->reachable))
-	  && !TREE_ASM_WRITTEN (decl) && !node->origin
+	  && !TREE_ASM_WRITTEN (decl)
 	  && !DECL_EXTERNAL (decl))
 	node->output = 1;
       else
-        DECL_SAVED_INSNS (decl) = NULL;
-    }
-}
-
-/* Optimize the function before expansion.  */
-
-static void
-cgraph_optimize_function (struct cgraph_node *node)
-{
-  tree decl = node->decl;
+	{
+	  /* We should've reclaimed all functions that are not needed.  */
+#ifdef ENABLE_CHECKING
+	  if (!node->global.inlined_to && DECL_SAVED_TREE (decl)
+	      && !DECL_EXTERNAL (decl))
+	    {
+	      dump_cgraph_node (stderr, node);
+	      internal_error ("failed to reclaim unneeded function");
+	    }
+#endif
+	  gcc_assert (node->global.inlined_to || !DECL_SAVED_TREE (decl)
+		      || DECL_EXTERNAL (decl));
 
-  timevar_push (TV_INTEGRATION);
-  /* optimize_inline_calls avoids inlining of current_function_decl.  */
-  current_function_decl = decl;
-  if (flag_inline_trees)
-    {
-      struct cgraph_edge *e;
+	}
 
-      for (e = node->callees; e; e = e->next_callee)
-	if (!e->inline_failed || warn_inline
-	    || (DECL_DECLARED_INLINE_P (e->callee->decl)
-		&& lookup_attribute ("always_inline",
-				     DECL_ATTRIBUTES (e->callee->decl))))
-	  break;
-      if (e)
-        optimize_inline_calls (decl);
     }
-  if (node->nested)
-    {
-      for (node = node->nested; node; node = node->next_nested)
-	cgraph_optimize_function (node);
-    }
-  timevar_pop (TV_INTEGRATION);
 }
 
 /* Expand function specified by NODE.  */
@@ -528,1086 +1232,702 @@ cgraph_expand_function (struct cgraph_node *node)
 {
   tree decl = node->decl;
 
+  /* We ought to not compile any inline clones.  */
+  gcc_assert (!node->global.inlined_to);
+
   if (flag_unit_at_a_time)
     announce_function (decl);
 
-  cgraph_optimize_function (node);
+  cgraph_lower_function (node);
+
+  /* Generate RTL for the body of DECL.  */
+  lang_hooks.callgraph.expand_function (decl);
 
-  /* Generate RTL for the body of DECL.  Nested functions are expanded
-     via lang_expand_decl_stmt.  */
-  (*lang_hooks.callgraph.expand_function) (decl);
-  if (DECL_DEFER_OUTPUT (decl))
-    abort ();
+  /* Make sure that BE didn't give up on compiling.  */
+  /* ??? Can happen with nested function of extern inline.  */
+  gcc_assert (TREE_ASM_WRITTEN (node->decl));
 
   current_function_decl = NULL;
+  if (!cgraph_preserve_function_body_p (node->decl))
+    {
+      DECL_SAVED_TREE (node->decl) = NULL;
+      DECL_STRUCT_FUNCTION (node->decl) = NULL;
+      DECL_INITIAL (node->decl) = error_mark_node;
+      /* Eliminate all call edges.  This is important so the call_expr no longer
+	 points to the dead function body.  */
+      cgraph_node_remove_callees (node);
+    }
+
+  cgraph_function_flags_ready = true;
 }
 
-/* Fill array order with all nodes with output flag set in the reverse
-   topological order.  */
+/* Return true when CALLER_DECL should be inlined into CALLEE_DECL.  */
 
-static int
-cgraph_postorder (struct cgraph_node **order)
+bool
+cgraph_inline_p (struct cgraph_edge *e, const char **reason)
 {
-  struct cgraph_node *node, *node2;
-  int stack_size = 0;
-  int order_pos = 0;
-  struct cgraph_edge *edge, last;
-
-  struct cgraph_node **stack =
-    xcalloc (cgraph_n_nodes, sizeof (struct cgraph_node *));
-
-  /* We have to deal with cycles nicely, so use a depth first traversal
-     output algorithm.  Ignore the fact that some functions won't need
-     to be output and put them into order as well, so we get dependencies
-     right through intline functions.  */
-  for (node = cgraph_nodes; node; node = node->next)
-    node->aux = NULL;
-  for (node = cgraph_nodes; node; node = node->next)
-    if (!node->aux)
-      {
-	node2 = node;
-	if (!node->callers)
-	  node->aux = &last;
-	else
-	  node->aux = node->callers;
-	while (node2)
-	  {
-	    while (node2->aux != &last)
-	      {
-		edge = node2->aux;
-		if (edge->next_caller)
-		  node2->aux = edge->next_caller;
-		else
-		  node2->aux = &last;
-		if (!edge->caller->aux)
-		  {
-		    if (!edge->caller->callers)
-		      edge->caller->aux = &last;
-		    else
-		      edge->caller->aux = edge->caller->callers;
-		    stack[stack_size++] = node2;
-		    node2 = edge->caller;
-		    break;
-		  }
-	      }
-	    if (node2->aux == &last)
-	      {
-		order[order_pos++] = node2;
-		if (stack_size)
-		  node2 = stack[--stack_size];
-		else
-		  node2 = NULL;
-	      }
-	  }
-      }
-  free (stack);
-  return order_pos;
+  *reason = e->inline_failed;
+  return !e->inline_failed;
 }
 
-#define INLINED_TIMES(node) ((size_t)(node)->aux)
-#define SET_INLINED_TIMES(node,times) ((node)->aux = (void *)(times))
 
-/* Return list of nodes we decided to inline NODE into, set their output
-   flag and compute INLINED_TIMES.
 
-   We do simple backtracing to get INLINED_TIMES right.  This should not be
-   expensive as we limit the amount of inlining.  Alternatively we may first
-   discover set of nodes, topologically sort these and propagate
-   INLINED_TIMES  */
+/* Expand all functions that must be output.
+
+   Attempt to topologically sort the nodes so function is output when
+   all called functions are already assembled to allow data to be
+   propagated across the callgraph.  Use a stack to get smaller distance
+   between a function and its callees (later we may choose to use a more
+   sophisticated algorithm for function reordering; we will likely want
+   to use subsections to make the output functions appear in top-down
+   order).  */
 
-static int
-cgraph_inlined_into (struct cgraph_node *node, struct cgraph_node **array)
+static void
+cgraph_expand_all_functions (void)
 {
-  int nfound = 0;
-  struct cgraph_edge **stack;
-  struct cgraph_edge *e, *e1;
-  int sp;
+  struct cgraph_node *node;
+  struct cgraph_node **order = XCNEWVEC (struct cgraph_node *, cgraph_n_nodes);
+  int order_pos = 0, new_order_pos = 0;
   int i;
 
-  /* Fast path: since we traverse in mostly topological order, we will likely
-     find no edges.  */
-  for (e = node->callers; e; e = e->next_caller)
-    if (!e->inline_failed)
-      break;
-
-  if (!e)
-    return 0;
-
-  /* Allocate stack for back-tracking up callgraph.  */
-  stack = xmalloc ((cgraph_n_nodes + 1) * sizeof (struct cgraph_edge));
-  sp = 0;
+  order_pos = cgraph_postorder (order);
+  gcc_assert (order_pos == cgraph_n_nodes);
 
-  /* Push the first edge on to the stack.  */
-  stack[sp++] = e;
+  /* Garbage collector may remove inline clones we eliminate during
+     optimization.  So we must be sure to not reference them.  */
+  for (i = 0; i < order_pos; i++)
+    if (order[i]->output)
+      order[new_order_pos++] = order[i];
 
-  while (sp)
+  for (i = new_order_pos - 1; i >= 0; i--)
     {
-      struct cgraph_node *caller;
-
-      /* Look at the edge on the top of the stack.  */
-      e = stack[sp - 1];
-      caller = e->caller;
-
-      /* Check if the caller destination has been visited yet.  */
-      if (!caller->output)
-	{
-	  array[nfound++] = e->caller;
-	  /* Mark that we have visited the destination.  */
-	  caller->output = true;
-	  SET_INLINED_TIMES (caller, 0);
-	}
-      SET_INLINED_TIMES (caller, INLINED_TIMES (caller) + 1);
-
-      for (e1 = caller->callers; e1; e1 = e1->next_caller)
-	if (!e1->inline_failed)
-	  break;
-
-      if (e1)
-	stack[sp++] = e1;
-      else
+      node = order[i];
+      if (node->output)
 	{
-	  while (true)
-	    {
-	      for (e1 = e->next_caller; e1; e1 = e1->next_caller)
-		if (!e1->inline_failed)
-		  break;
-
-	      if (e1)
-		{
-		  stack[sp - 1] = e1;
-		  break;
-		}
-	      else
-		{
-		  sp--;
-		  if (!sp)
-		    break;
-		  e = stack[sp - 1];
-		}
-	    }
+	  gcc_assert (node->reachable);
+	  node->output = 0;
+	  cgraph_expand_function (node);
 	}
     }
 
-  free (stack);
-
+  free (order);
 
-  if (cgraph_dump_file)
+  /* Process CGRAPH_EXPAND_QUEUE, these are functions created during
+     the expansion process.  Note that this queue may grow as its
+     being processed, as the new functions may generate new ones.  */
+  while (cgraph_expand_queue)
     {
-      fprintf (cgraph_dump_file, " Found inline predecesors of %s:",
-	       cgraph_node_name (node));
-      for (i = 0; i < nfound; i++)
-	{
-	  fprintf (cgraph_dump_file, " %s", cgraph_node_name (array[i]));
-	  if (INLINED_TIMES (array[i]) != 1)
-	    fprintf (cgraph_dump_file, " (%i times)",
-		     (int)INLINED_TIMES (array[i]));
-	}
-      fprintf (cgraph_dump_file, "\n");
+      node = cgraph_expand_queue;
+      cgraph_expand_queue = cgraph_expand_queue->next_needed;
+      node->next_needed = NULL;
+      node->output = 0;
+      node->lowered = DECL_STRUCT_FUNCTION (node->decl)->cfg != NULL;
+      cgraph_expand_function (node);
     }
-
-  return nfound;
 }
 
-/* Return list of nodes we decided to inline into NODE, set their output
-   flag and compute INLINED_TIMES.
+/* This is used to sort the node types by the cgraph order number.  */
 
-   This function is identical to cgraph_inlined_into with callers and callees
-   nodes swapped.  */
+struct cgraph_order_sort
+{
+  enum { ORDER_UNDEFINED = 0, ORDER_FUNCTION, ORDER_VAR, ORDER_ASM } kind;
+  union
+  {
+    struct cgraph_node *f;
+    struct cgraph_varpool_node *v;
+    struct cgraph_asm_node *a;
+  } u;
+};
+
+/* Output all functions, variables, and asm statements in the order
+   according to their order fields, which is the order in which they
+   appeared in the file.  This implements -fno-toplevel-reorder.  In
+   this mode we may output functions and variables which don't really
+   need to be output.  */
 
-static int
-cgraph_inlined_callees (struct cgraph_node *node, struct cgraph_node **array)
+static void
+cgraph_output_in_order (void)
 {
-  int nfound = 0;
-  struct cgraph_edge **stack;
-  struct cgraph_edge *e, *e1;
-  int sp;
+  int max;
+  size_t size;
+  struct cgraph_order_sort *nodes;
   int i;
+  struct cgraph_node *pf;
+  struct cgraph_varpool_node *pv;
+  struct cgraph_asm_node *pa;
 
-  /* Fast path: since we traverse in mostly topological order, we will likely
-     find no edges.  */
-  for (e = node->callees; e; e = e->next_callee)
-    if (!e->inline_failed)
-      break;
-
-  if (!e)
-    return 0;
-
-  /* Allocate stack for back-tracking up callgraph.  */
-  stack = xmalloc ((cgraph_n_nodes + 1) * sizeof (struct cgraph_edge));
-  sp = 0;
+  max = cgraph_order;
+  size = max * sizeof (struct cgraph_order_sort);
+  nodes = (struct cgraph_order_sort *) alloca (size);
+  memset (nodes, 0, size);
 
-  /* Push the first edge on to the stack.  */
-  stack[sp++] = e;
+  cgraph_varpool_analyze_pending_decls ();
 
-  while (sp)
+  for (pf = cgraph_nodes; pf; pf = pf->next)
     {
-      struct cgraph_node *callee;
-
-      /* Look at the edge on the top of the stack.  */
-      e = stack[sp - 1];
-      callee = e->callee;
-
-      /* Check if the callee destination has been visited yet.  */
-      if (!callee->output)
-	{
-	  array[nfound++] = e->callee;
-	  /* Mark that we have visited the destination.  */
-	  callee->output = true;
-	  SET_INLINED_TIMES (callee, 0);
-	}
-      SET_INLINED_TIMES (callee, INLINED_TIMES (callee) + 1);
-
-      for (e1 = callee->callees; e1; e1 = e1->next_callee)
-	if (!e1->inline_failed)
-	  break;
-      if (e1)
-	stack[sp++] = e1;
-      else
+      if (pf->output)
 	{
-	  while (true)
-	    {
-	      for (e1 = e->next_callee; e1; e1 = e1->next_callee)
-		if (!e1->inline_failed)
-		  break;
-
-	      if (e1)
-		{
-		  stack[sp - 1] = e1;
-		  break;
-		}
-	      else
-		{
-		  sp--;
-		  if (!sp)
-		    break;
-		  e = stack[sp - 1];
-		}
-	    }
+	  i = pf->order;
+	  gcc_assert (nodes[i].kind == ORDER_UNDEFINED);
+	  nodes[i].kind = ORDER_FUNCTION;
+	  nodes[i].u.f = pf;
 	}
     }
 
-  free (stack);
-
-  if (cgraph_dump_file)
+  for (pv = cgraph_varpool_nodes_queue; pv; pv = pv->next_needed)
     {
-      fprintf (cgraph_dump_file, " Found inline successors of %s:",
-	       cgraph_node_name (node));
-      for (i = 0; i < nfound; i++)
-	{
-	  fprintf (cgraph_dump_file, " %s", cgraph_node_name (array[i]));
-	  if (INLINED_TIMES (array[i]) != 1)
-	    fprintf (cgraph_dump_file, " (%i times)",
-		     (int)INLINED_TIMES (array[i]));
-	}
-      fprintf (cgraph_dump_file, "\n");
+      i = pv->order;
+      gcc_assert (nodes[i].kind == ORDER_UNDEFINED);
+      nodes[i].kind = ORDER_VAR;
+      nodes[i].u.v = pv;
     }
 
-  return nfound;
-}
-
-/* Perform reachability analysis and reclaim all unreachable nodes.
-   This function also remove unneeded bodies of extern inline functions
-   and thus needs to be done only after inlining decisions has been made.  */
-static bool
-cgraph_remove_unreachable_nodes (void)
-{
-  struct cgraph_node *first = (void *) 1;
-  struct cgraph_node *node;
-  bool changed = false;
-  int insns = 0;
-
-  if (cgraph_dump_file)
-    fprintf (cgraph_dump_file, "\nReclaiming functions:");
-#ifdef ENABLE_CHECKING
-  for (node = cgraph_nodes; node; node = node->next)
-    if (node->aux)
-      abort ();
-#endif
-  for (node = cgraph_nodes; node; node = node->next)
-    if (node->needed && (!DECL_EXTERNAL (node->decl) || !node->analyzed))
-      {
-	node->aux = first;
-	first = node;
-      }
-    else if (node->aux)
-      abort ();
-
-  /* Perform reachability analysis.  As a special case do not consider
-     extern inline functions not inlined as live because we won't output
-     them at all.  */
-  while (first != (void *) 1)
+  for (pa = cgraph_asm_nodes; pa; pa = pa->next)
     {
-      struct cgraph_edge *e;
-      node = first;
-      first = first->aux;
-
-      for (e = node->callees; e; e = e->next_callee)
-	if (!e->callee->aux
-	    && node->analyzed
-	    && (!e->inline_failed || !e->callee->analyzed
-		|| !DECL_EXTERNAL (e->callee->decl)))
-	  {
-	    e->callee->aux = first;
-	    first = e->callee;
-	  }
+      i = pa->order;
+      gcc_assert (nodes[i].kind == ORDER_UNDEFINED);
+      nodes[i].kind = ORDER_ASM;
+      nodes[i].u.a = pa;
     }
 
-  /* Remove unreachable nodes.  Extern inline functions need special care;
-     Unreachable extern inline functions shall be removed.
-     Reachable extern inline functions we never inlined shall get their bodies
-     elliminated
-     Reachable extern inline functions we sometimes inlined will be turned into
-     unanalyzed nodes so they look like for true extern functions to the rest
-     of code.  */
-  for (node = cgraph_nodes; node; node = node->next)
+  for (i = 0; i < max; ++i)
     {
-      if (!node->aux)
+      switch (nodes[i].kind)
 	{
-	  int local_insns;
-	  tree decl = node->decl;
-
-	  if (DECL_SAVED_INSNS (decl))
-	    local_insns = node->local.self_insns;
-	  else
-	    local_insns = 0;
-	  if (cgraph_dump_file)
-	    fprintf (cgraph_dump_file, " %s", cgraph_node_name (node));
-	  if (!node->analyzed || !DECL_EXTERNAL (node->decl))
-	    cgraph_remove_node (node);
-	  else
-	    {
-	      struct cgraph_edge *e;
-
-	      for (e = node->callers; e; e = e->next_caller)
-		if (e->caller->aux)
-		  break;
-	      if (e || node->needed)
-		{
-		  DECL_SAVED_TREE (node->decl) = NULL_TREE;
-		  while (node->callees)
-		    cgraph_remove_edge (node, node->callees->callee);
-		  node->analyzed = false;
-		}
-	      else
-		cgraph_remove_node (node);
-	    }
-	  if (!DECL_SAVED_TREE (decl))
-	    insns += local_insns;
-	  changed = true;
-	}
-    }
-  for (node = cgraph_nodes; node; node = node->next)
-    node->aux = NULL;
-  if (cgraph_dump_file)
-    fprintf (cgraph_dump_file, "\nReclaimed %i insns", insns);
-  return changed;
-}
-
-
-/* Estimate size of the function after inlining WHAT into TO.  */
+	case ORDER_FUNCTION:
+	  nodes[i].u.f->output = 0;
+	  cgraph_expand_function (nodes[i].u.f);
+	  break;
 
-static int
-cgraph_estimate_size_after_inlining (int times, struct cgraph_node *to,
-				     struct cgraph_node *what)
-{
-  return (what->global.insns - INSNS_PER_CALL) * times + to->global.insns;
-}
+	case ORDER_VAR:
+	  cgraph_varpool_assemble_decl (nodes[i].u.v);
+	  break;
 
-/* Estimate the growth caused by inlining NODE into all callees.  */
+	case ORDER_ASM:
+	  assemble_asm (nodes[i].u.a->asm_str);
+	  break;
 
-static int
-cgraph_estimate_growth (struct cgraph_node *node)
-{
-  int growth = 0;
-  int calls_saved = 0;
-  int clones_added = 0;
-  struct cgraph_edge *e;
+	case ORDER_UNDEFINED:
+	  break;
 
-  for (e = node->callers; e; e = e->next_caller)
-    if (e->inline_failed)
-      {
-	growth += ((cgraph_estimate_size_after_inlining (1, e->caller, node)
-		    -
-		    e->caller->global.insns) *e->caller->global.cloned_times);
-	calls_saved += e->caller->global.cloned_times;
-	clones_added += e->caller->global.cloned_times;
-      }
+	default:
+	  gcc_unreachable ();
+	}
+    }
 
-  /* ??? Wrong for self recursive functions or cases where we decide to not
-     inline for different reasons, but it is not big deal as in that case
-     we will keep the body around, but we will also avoid some inlining.  */
-  if (!node->needed && !node->origin && !DECL_EXTERNAL (node->decl))
-    growth -= node->global.insns, clones_added--;
+  cgraph_asm_nodes = NULL;
+}
 
-  if (!calls_saved)
-    calls_saved = 1;
+/* Mark visibility of all functions.
 
-  return growth;
-}
+   A local function is one whose calls can occur only in the current
+   compilation unit and all its calls are explicit, so we can change
+   its calling convention.  We simply mark all static functions whose
+   address is not taken as local.
 
-/* Update insn sizes after inlining WHAT into TO that is already inlined into
-   all nodes in INLINED array.  */
+   We also change the TREE_PUBLIC flag of all declarations that are public
+   in language point of view but we want to overwrite this default
+   via visibilities for the backend point of view.  */
 
 static void
-cgraph_mark_inline (struct cgraph_node *to, struct cgraph_node *what,
-		    struct cgraph_node **inlined, int ninlined,
-		    struct cgraph_node **inlined_callees,
-		    int ninlined_callees)
+cgraph_function_and_variable_visibility (void)
 {
-  int i;
-  int times = 0;
-  int clones = 0;
-  struct cgraph_edge *e;
-  bool called = false;
-  int new_insns;
+  struct cgraph_node *node;
+  struct cgraph_varpool_node *vnode;
 
-  what->global.inlined = 1;
-  for (e = what->callers; e; e = e->next_caller)
+  for (node = cgraph_nodes; node; node = node->next)
     {
-      if (e->caller == to)
+      if (node->reachable
+	  && (DECL_COMDAT (node->decl)
+	      || (!flag_whole_program
+		  && TREE_PUBLIC (node->decl) && !DECL_EXTERNAL (node->decl))))
+	node->local.externally_visible = true;
+      if (!node->local.externally_visible && node->analyzed
+	  && !DECL_EXTERNAL (node->decl))
 	{
-	  if (!e->inline_failed)
-	    continue;
-	  e->inline_failed = NULL;
-	  times++;
-	  clones += e->caller->global.cloned_times;
+	  gcc_assert (flag_whole_program || !TREE_PUBLIC (node->decl));
+	  TREE_PUBLIC (node->decl) = 0;
 	}
-      else if (e->inline_failed)
-	called = true;
-    }
-  if (!times)
-    abort ();
-  ncalls_inlined += times;
-
-  new_insns = cgraph_estimate_size_after_inlining (times, to, what);
-  if (to->global.will_be_output)
-    overall_insns += new_insns - to->global.insns;
-  to->global.insns = new_insns;
-
-  if (!called && !what->needed && !what->origin
-      && flag_unit_at_a_time
-      && !DECL_EXTERNAL (what->decl))
-    {
-      if (!what->global.will_be_output)
-	abort ();
-      clones--;
-      nfunctions_inlined++;
-      what->global.will_be_output = 0;
-      overall_insns -= what->global.insns;
-    }
-  what->global.cloned_times += clones;
-  for (i = 0; i < ninlined; i++)
-    {
-      new_insns =
-	cgraph_estimate_size_after_inlining (INLINED_TIMES (inlined[i]) *
-					     times, inlined[i], what);
-      if (inlined[i]->global.will_be_output)
-	overall_insns += new_insns - inlined[i]->global.insns;
-      inlined[i]->global.insns = new_insns;
+      node->local.local = (!node->needed
+			   && node->analyzed
+			   && !DECL_EXTERNAL (node->decl)
+			   && !node->local.externally_visible);
     }
-  for (i = 0; i < ninlined_callees; i++)
+  for (vnode = cgraph_varpool_nodes_queue; vnode; vnode = vnode->next_needed)
     {
-      inlined_callees[i]->global.cloned_times +=
-	INLINED_TIMES (inlined_callees[i]) * clones;
+      if (vnode->needed
+	  && !flag_whole_program
+	  && (DECL_COMDAT (vnode->decl) || TREE_PUBLIC (vnode->decl)))
+	vnode->externally_visible = 1;
+      if (!vnode->externally_visible)
+	{
+	  gcc_assert (flag_whole_program || !TREE_PUBLIC (vnode->decl));
+	  TREE_PUBLIC (vnode->decl) = 0;
+	}
+     gcc_assert (TREE_STATIC (vnode->decl));
     }
-}
-
-/* Return false when inlining WHAT into TO is not good idea as it would cause
-   too large growth of function bodies.  */
-
-static bool
-cgraph_check_inline_limits (struct cgraph_node *to, struct cgraph_node *what,
-			    struct cgraph_node **inlined, int ninlined,
-			    const char **reason)
-{
-  int i;
-  int times = 0;
-  struct cgraph_edge *e;
-  int newsize;
-  int limit;
-
-  for (e = to->callees; e; e = e->next_callee)
-    if (e->callee == what)
-      times++;
-
-  /* When inlining large function body called once into small function,
-     take the inlined function as base for limiting the growth.  */
-  if (to->local.self_insns > what->local.self_insns)
-    limit = to->local.self_insns;
-  else
-    limit = what->local.self_insns;
 
-  limit += limit * PARAM_VALUE (PARAM_LARGE_FUNCTION_GROWTH) / 100;
+  /* Because we have to be conservative on the boundaries of source
+     level units, it is possible that we marked some functions in
+     reachable just because they might be used later via external
+     linkage, but after making them local they are really unreachable
+     now.  */
+  cgraph_remove_unreachable_nodes (true, cgraph_dump_file);
 
-  newsize = cgraph_estimate_size_after_inlining (times, to, what);
-  if (newsize > PARAM_VALUE (PARAM_LARGE_FUNCTION_INSNS)
-      && newsize > limit)
-    {
-      *reason = N_("--param large-function-growth limit reached");
-      return false;
-    }
-  for (i = 0; i < ninlined; i++)
+  if (cgraph_dump_file)
     {
-      newsize =
-	cgraph_estimate_size_after_inlining (INLINED_TIMES (inlined[i]) *
-					     times, inlined[i], what);
-      if (newsize > PARAM_VALUE (PARAM_LARGE_FUNCTION_INSNS)
-	  && newsize >
-	  inlined[i]->local.self_insns *
-	  (100 + PARAM_VALUE (PARAM_LARGE_FUNCTION_GROWTH)) / 100)
-	{
-	  *reason = N_("--param large-function-growth limit reached while inlining the caller");
-	  return false;
-	}
+      fprintf (cgraph_dump_file, "\nMarking local functions:");
+      for (node = cgraph_nodes; node; node = node->next)
+	if (node->local.local)
+	  fprintf (cgraph_dump_file, " %s", cgraph_node_name (node));
+      fprintf (cgraph_dump_file, "\n\n");
+      fprintf (cgraph_dump_file, "\nMarking externally visible functions:");
+      for (node = cgraph_nodes; node; node = node->next)
+	if (node->local.externally_visible)
+	  fprintf (cgraph_dump_file, " %s", cgraph_node_name (node));
+      fprintf (cgraph_dump_file, "\n\n");
     }
-  return true;
+  cgraph_function_flags_ready = true;
 }
 
-/* Return true when function N is small enough to be inlined.  */
-
-static bool
-cgraph_default_inline_p (struct cgraph_node *n)
+/* Return true when function body of DECL still needs to be kept around
+   for later re-use.  */
+bool
+cgraph_preserve_function_body_p (tree decl)
 {
-  if (!DECL_INLINE (n->decl) || !DECL_SAVED_TREE (n->decl))
-    return false;
-  if (DECL_DECLARED_INLINE_P (n->decl))
-    return n->global.insns < MAX_INLINE_INSNS_SINGLE;
-  else
-    return n->global.insns < MAX_INLINE_INSNS_AUTO;
+  struct cgraph_node *node;
+  if (!cgraph_global_info_ready)
+    return (flag_really_no_inline
+	    ? lang_hooks.tree_inlining.disregard_inline_limits (decl)
+	    : DECL_INLINE (decl));
+  /* Look if there is any clone around.  */
+  for (node = cgraph_node (decl); node; node = node->next_clone)
+    if (node->global.inlined_to)
+      return true;
+  return false;
 }
 
-/* Set inline_failed for all callers of given function to REASON.  */
-
 static void
-cgraph_set_inline_failed (struct cgraph_node *node, const char *reason)
+ipa_passes (void)
 {
-  struct cgraph_edge *e;
-
-  if (cgraph_dump_file)
-    fprintf (cgraph_dump_file, "Inlining failed: %s\n", reason);
-  for (e = node->callers; e; e = e->next_caller)
-    if (e->inline_failed)
-      e->inline_failed = reason;
+  cfun = NULL;
+  tree_register_cfg_hooks ();
+  bitmap_obstack_initialize (NULL);
+  execute_ipa_pass_list (all_ipa_passes);
+  bitmap_obstack_release (NULL);
 }
 
-/* We use greedy algorithm for inlining of small functions:
-   All inline candidates are put into prioritized heap based on estimated
-   growth of the overall number of instructions and then update the estimates.
-
-   INLINED and INLINED_CALEES are just pointers to arrays large enough
-   to be passed to cgraph_inlined_into and cgraph_inlined_callees.  */
+/* Perform simple optimizations based on callgraph.  */
 
-static void
-cgraph_decide_inlining_of_small_functions (struct cgraph_node **inlined,
-					   struct cgraph_node **inlined_callees)
+void
+cgraph_optimize (void)
 {
-  int i;
-  struct cgraph_node *node;
-  fibheap_t heap = fibheap_new ();
-  struct fibnode **heap_node =
-    xcalloc (cgraph_max_uid, sizeof (struct fibnode *));
-  int ninlined, ninlined_callees;
-  int max_insns = ((HOST_WIDEST_INT) initial_insns
-		   * (100 + PARAM_VALUE (PARAM_INLINE_UNIT_GROWTH)) / 100);
-
-  /* Put all inline candidates into the heap.  */
+  if (errorcount || sorrycount)
+    return;
 
-  for (node = cgraph_nodes; node; node = node->next)
+#ifdef ENABLE_CHECKING
+  verify_cgraph ();
+#endif
+  if (!flag_unit_at_a_time)
     {
-      if (!node->local.inlinable || !node->callers
-	  || node->local.disregard_inline_limits)
-	continue;
-
-      if (!cgraph_default_inline_p (node))
-	{
-	  cgraph_set_inline_failed (node,
-	    N_("--param max-inline-insns-single limit reached"));
-	  continue;
-	}
-      heap_node[node->uid] =
-	fibheap_insert (heap, cgraph_estimate_growth (node), node);
+      cgraph_output_pending_asms ();
+      cgraph_varpool_assemble_pending_decls ();
+      cgraph_varpool_output_debug_info ();
+      return;
     }
 
-  if (cgraph_dump_file)
-    fprintf (cgraph_dump_file, "\nDeciding on smaller functions:\n");
-  while (overall_insns <= max_insns && (node = fibheap_extract_min (heap)))
-    {
-      struct cgraph_edge *e;
-      int old_insns = overall_insns;
-
-      heap_node[node->uid] = NULL;
-      if (cgraph_dump_file)
-	fprintf (cgraph_dump_file, 
-		 "\nConsidering %s with %i insns\n"
-		 " Estimated growth is %+i insns.\n",
-		 cgraph_node_name (node), node->global.insns,
-		 cgraph_estimate_growth (node));
-      if (!cgraph_default_inline_p (node))
-	{
-	  cgraph_set_inline_failed (node,
-	    N_("--param max-inline-insns-single limit reached after inlining into the callee"));
-	  continue;
-	}
-      ninlined_callees = cgraph_inlined_callees (node, inlined_callees);
-      for (e = node->callers; e; e = e->next_caller)
-	if (e->inline_failed)
-	  {
-	    /* Marking recursive function inlinine has sane semantic and
-	       thus we should not warn on it.  */
- 	    if (e->caller == node)
- 	      {
- 	        e->inline_failed = "";
- 		continue;
- 	      }
-	    ninlined = cgraph_inlined_into (e->caller, inlined);
-	    if (e->callee->output)
-	      e->inline_failed = "";
-	    if (e->callee->output
-		|| !cgraph_check_inline_limits (e->caller, node, inlined,
-						ninlined, &e->inline_failed))
-	      {
-		for (i = 0; i < ninlined; i++)
-		  inlined[i]->output = 0, inlined[i]->aux = 0;
-		if (cgraph_dump_file)
-		  fprintf (cgraph_dump_file, " Not inlining into %s.\n",
-			   cgraph_node_name (e->caller));
-		continue;
-	      }
-	    cgraph_mark_inline (e->caller, node, inlined, ninlined,
-				inlined_callees, ninlined_callees);
-	    if (heap_node[e->caller->uid])
-	      fibheap_replace_key (heap, heap_node[e->caller->uid],
-				   cgraph_estimate_growth (e->caller));
-
-	    /* Size of the functions we updated into has changed, so update
-	       the keys.  */
-	    for (i = 0; i < ninlined; i++)
-	      {
-		inlined[i]->output = 0, inlined[i]->aux = 0;
-		if (heap_node[inlined[i]->uid])
-		  fibheap_replace_key (heap, heap_node[inlined[i]->uid],
-				       cgraph_estimate_growth (inlined[i]));
-	      }
-	    if (cgraph_dump_file)
-	      fprintf (cgraph_dump_file, 
-		       " Inlined into %s which now has %i insns.\n",
-		       cgraph_node_name (e->caller),
-		       e->caller->global.insns);
-	  }
-
-      /* Similarly all functions called by the function we just inlined
-         are now called more times; update keys.  */
-
-      for (e = node->callees; e; e = e->next_callee)
-	if (e->inline_failed && heap_node[e->callee->uid])
-	  fibheap_replace_key (heap, heap_node[e->callee->uid],
-			       cgraph_estimate_growth (e->callee));
+  process_pending_assemble_externals ();
 
-      for (i = 0; i < ninlined_callees; i++)
-	{
-	  struct cgraph_edge *e;
+  /* Frontend may output common variables after the unit has been finalized.
+     It is safe to deal with them here as they are always zero initialized.  */
+  cgraph_varpool_analyze_pending_decls ();
 
-	  for (e = inlined_callees[i]->callees; e; e = e->next_callee)
-	    if (e->inline_failed && heap_node[e->callee->uid])
-	      fibheap_replace_key (heap, heap_node[e->callee->uid],
-				   cgraph_estimate_growth (e->callee));
+  timevar_push (TV_CGRAPHOPT);
+  if (!quiet_flag)
+    fprintf (stderr, "Performing interprocedural optimizations\n");
 
-	  inlined_callees[i]->output = 0;
-	  inlined_callees[i]->aux = 0;
-	}
-      if (cgraph_dump_file)
-	fprintf (cgraph_dump_file, 
-		 " Inlined %i times for a net change of %+i insns.\n",
-		 node->global.cloned_times, overall_insns - old_insns);
+  cgraph_function_and_variable_visibility ();
+  if (cgraph_dump_file)
+    {
+      fprintf (cgraph_dump_file, "Marked ");
+      dump_cgraph (cgraph_dump_file);
     }
-  while ((node = fibheap_extract_min (heap)) != NULL)
-    if (!node->local.disregard_inline_limits)
-      cgraph_set_inline_failed (node, N_("--param inline-unit-growth limit reached"));
-  fibheap_delete (heap);
-  free (heap_node);
-}
-
-/* Decide on the inlining.  We do so in the topological order to avoid
-   expenses on updating datastructures.  */
-
-static void
-cgraph_decide_inlining (void)
-{
-  struct cgraph_node *node;
-  int nnodes;
-  struct cgraph_node **order =
-    xcalloc (cgraph_n_nodes, sizeof (struct cgraph_node *));
-  struct cgraph_node **inlined =
-    xcalloc (cgraph_n_nodes, sizeof (struct cgraph_node *));
-  struct cgraph_node **inlined_callees =
-    xcalloc (cgraph_n_nodes, sizeof (struct cgraph_node *));
-  int ninlined;
-  int ninlined_callees;
-  int old_insns = 0;
-  int i, y;
-
-  for (node = cgraph_nodes; node; node = node->next)
-    initial_insns += node->local.self_insns;
-  overall_insns = initial_insns;
-
-  nnodes = cgraph_postorder (order);
-
+    
+  /* Don't run the IPA passes if there was any error or sorry messages.  */
+  if (errorcount == 0 && sorrycount == 0)
+    ipa_passes ();
+
+  /* This pass remove bodies of extern inline functions we never inlined.
+     Do this later so other IPA passes see what is really going on.  */
+  cgraph_remove_unreachable_nodes (false, dump_file);
+  cgraph_increase_alignment ();
+  cgraph_global_info_ready = true;
   if (cgraph_dump_file)
-    fprintf (cgraph_dump_file,
-	     "\nDeciding on inlining.  Starting with %i insns.\n",
-	     initial_insns);
-
-  for (node = cgraph_nodes; node; node = node->next)
-    node->aux = 0;
+    {
+      fprintf (cgraph_dump_file, "Optimized ");
+      dump_cgraph (cgraph_dump_file);
+      dump_varpool (cgraph_dump_file);
+    }
+  timevar_pop (TV_CGRAPHOPT);
 
-  if (cgraph_dump_file)
-    fprintf (cgraph_dump_file, "\nInlining always_inline functions:\n");
+  /* Output everything.  */
+  if (!quiet_flag)
+    fprintf (stderr, "Assembling functions:\n");
 #ifdef ENABLE_CHECKING
-  for (node = cgraph_nodes; node; node = node->next)
-    if (node->aux || node->output)
-      abort ();
+  verify_cgraph ();
 #endif
 
-  /* In the first pass mark all always_inline edges.  Do this with a priority
-     so none of our later choices will make this impossible.  */
-  for (i = nnodes - 1; i >= 0; i--)
+  cgraph_mark_functions_to_output ();
+
+  if (!flag_toplevel_reorder)
+    cgraph_output_in_order ();
+  else
     {
-      struct cgraph_edge *e;
+      cgraph_output_pending_asms ();
 
-      node = order[i];
+      cgraph_expand_all_functions ();
+      cgraph_varpool_remove_unreferenced_decls ();
 
-      for (e = node->callees; e; e = e->next_callee)
-	if (e->callee->local.disregard_inline_limits)
-	  break;
-      if (!e)
-	continue;
-      if (cgraph_dump_file)
-	fprintf (cgraph_dump_file,
-		 "\nConsidering %s %i insns (always inline)\n",
-		 cgraph_node_name (e->callee), e->callee->global.insns);
-      ninlined = cgraph_inlined_into (order[i], inlined);
-      for (; e; e = e->next_callee)
-	{
-	  old_insns = overall_insns;
-	  if (!e->inline_failed || !e->callee->local.inlinable
-	      || !e->callee->local.disregard_inline_limits)
-  	    continue;
-  	  if (e->callee->output || e->callee == node)
-	    {
-	      e->inline_failed = N_("recursive inlining");
-	      continue;
-	    }
-	  ninlined_callees =
-	    cgraph_inlined_callees (e->callee, inlined_callees);
-	  cgraph_mark_inline (node, e->callee, inlined, ninlined,
-			      inlined_callees, ninlined_callees);
-	  for (y = 0; y < ninlined_callees; y++)
-	    inlined_callees[y]->output = 0, inlined_callees[y]->aux = 0;
-	  if (cgraph_dump_file)
-	    fprintf (cgraph_dump_file, 
-		     " Inlined into %s which now has %i insns.\n",
-		     cgraph_node_name (node->callees->caller),
-	             node->callees->caller->global.insns);
-	}
-      if (cgraph_dump_file && node->global.cloned_times > 0)
-	fprintf (cgraph_dump_file, 
-		 " Inlined %i times for a net change of %+i insns.\n",
-		 node->global.cloned_times, overall_insns - old_insns);
-      for (y = 0; y < ninlined; y++)
-	inlined[y]->output = 0, inlined[y]->aux = 0;
+      cgraph_varpool_assemble_pending_decls ();
+      cgraph_varpool_output_debug_info ();
     }
-#ifdef ENABLE_CHECKING
-  for (node = cgraph_nodes; node; node = node->next)
-    if (node->aux || node->output)
-      abort ();
-#endif
 
-  if (!flag_really_no_inline)
+  if (cgraph_dump_file)
     {
-      cgraph_decide_inlining_of_small_functions (inlined, inlined_callees);
+      fprintf (cgraph_dump_file, "\nFinal ");
+      dump_cgraph (cgraph_dump_file);
+    }
 #ifdef ENABLE_CHECKING
+  verify_cgraph ();
+  /* Double check that all inline clones are gone and that all
+     function bodies have been released from memory.  */
+  if (flag_unit_at_a_time
+      && !(sorrycount || errorcount))
+    {
+      struct cgraph_node *node;
+      bool error_found = false;
+
       for (node = cgraph_nodes; node; node = node->next)
-	if (node->aux || node->output)
-	  abort ();
+	if (node->analyzed
+	    && (node->global.inlined_to
+		|| DECL_SAVED_TREE (node->decl)))
+	  {
+	    error_found = true;
+	    dump_cgraph_node (stderr, node);
+	  }
+      if (error_found)
+	internal_error ("nodes with no released memory found");
+    }
 #endif
+}
 
-      if (cgraph_dump_file)
-	fprintf (cgraph_dump_file, "\nDeciding on functions called once:\n");
-
-      /* And finally decide what functions are called once.  */
+/* Increase alignment of global arrays to improve vectorization potential.
+   TODO:
+   - Consider also structs that have an array field.
+   - Use ipa analysis to prune arrays that can't be vectorized?
+     This should involve global alignment analysis and in the future also
+     array padding.  */
 
-      for (i = nnodes - 1; i >= 0; i--)
-	{
-	  node = order[i];
-
-	  if (node->callers && !node->callers->next_caller && !node->needed
-	      && node->local.inlinable && node->callers->inline_failed
-	      && !DECL_EXTERNAL (node->decl) && !DECL_COMDAT (node->decl))
-	    {
-	      bool ok = true;
-	      struct cgraph_node *node1;
-
-	      /* Verify that we won't duplicate the caller.  */
-	      for (node1 = node->callers->caller;
-		   node1->callers && !node1->callers->inline_failed
-		   && ok; node1 = node1->callers->caller)
-		if (node1->callers->next_caller || node1->needed)
-		  ok = false;
-	      if (ok)
-		{
-		  const char *dummy_reason;
-		  if (cgraph_dump_file)
-		    fprintf (cgraph_dump_file,
-			     "\nConsidering %s %i insns.\n"
-			     " Called once from %s %i insns.\n",
-			     cgraph_node_name (node), node->global.insns,
-			     cgraph_node_name (node->callers->caller),
-			     node->callers->caller->global.insns);
-		  ninlined = cgraph_inlined_into (node->callers->caller,
-		      				  inlined);
-		  old_insns = overall_insns;
-
-		  /* Inlining functions once would never cause inlining warnings.  */
-		  if (cgraph_check_inline_limits
-		      (node->callers->caller, node, inlined, ninlined,
-		       &dummy_reason))
-		    {
-		      ninlined_callees =
-			cgraph_inlined_callees (node, inlined_callees);
-		      cgraph_mark_inline (node->callers->caller, node, inlined,
-					  ninlined, inlined_callees,
-					  ninlined_callees);
-		      for (y = 0; y < ninlined_callees; y++)
-			inlined_callees[y]->output = 0, inlined_callees[y]->aux = 0;
-		      if (cgraph_dump_file)
-			fprintf (cgraph_dump_file,
-				 " Inlined into %s which now has %i insns"
-				 " for a net change of %+i insns.\n",
-				 cgraph_node_name (node->callers->caller),
-				 node->callers->caller->global.insns,
-				 overall_insns - old_insns);
-		    }
-		  else
-		    {
-		      if (cgraph_dump_file)
-			fprintf (cgraph_dump_file,
-				 " Inline limit reached, not inlined.\n");
-		    }
-		  for (y = 0; y < ninlined; y++)
-		    inlined[y]->output = 0, inlined[y]->aux = 0;
-		}
-	    }
-	}
+static void
+cgraph_increase_alignment (void)
+{
+  if (flag_section_anchors && flag_tree_vectorize)
+    {
+      struct cgraph_varpool_node *vnode;
+
+      /* Increase the alignment of all global arrays for vectorization.  */
+      for (vnode = cgraph_varpool_nodes_queue;
+           vnode;
+           vnode = vnode->next_needed)
+        {
+          tree vectype, decl = vnode->decl;
+          unsigned int alignment;
+
+          if (TREE_CODE (TREE_TYPE (decl)) != ARRAY_TYPE)
+            continue;
+          vectype = get_vectype_for_scalar_type (TREE_TYPE (TREE_TYPE (decl)));
+          if (!vectype)
+            continue;
+          alignment = TYPE_ALIGN (vectype);
+          if (DECL_ALIGN (decl) >= alignment)
+            continue;
+
+          if (vect_can_force_dr_alignment_p (decl, alignment))
+            { 
+              DECL_ALIGN (decl) = TYPE_ALIGN (vectype);
+              DECL_USER_ALIGN (decl) = 1;
+              if (cgraph_dump_file)
+                { 
+                  fprintf (cgraph_dump_file, "Increasing alignment of decl: ");
+                  print_generic_expr (cgraph_dump_file, decl, TDF_SLIM);
+                }
+            }
+        }
     }
-  cgraph_remove_unreachable_nodes ();
-
-  if (cgraph_dump_file)
-    fprintf (cgraph_dump_file,
-	     "\nInlined %i calls, eliminated %i functions, "
-	     "%i insns turned to %i insns.\n\n",
-	     ncalls_inlined, nfunctions_inlined, initial_insns,
-	     overall_insns);
-  free (order);
-  free (inlined);
-  free (inlined_callees);
 }
 
-/* Decide on the inlining.  We do so in the topological order to avoid
-   expenses on updating datastructures.  */
+/* Generate and emit a static constructor or destructor.  WHICH must be
+   one of 'I' or 'D'.  BODY should be a STATEMENT_LIST containing
+   GENERIC statements.  */
 
-static void
-cgraph_decide_inlining_incrementally (struct cgraph_node *node)
+void
+cgraph_build_static_cdtor (char which, tree body, int priority)
 {
-  struct cgraph_edge *e;
-  struct cgraph_node **inlined =
-    xmalloc (sizeof (struct cgraph_node *) * cgraph_n_nodes);
-  struct cgraph_node **inlined_callees =
-    xmalloc (sizeof (struct cgraph_node *) * cgraph_n_nodes);
-  int ninlined;
-  int ninlined_callees;
-  int y;
+  static int counter = 0;
+  char which_buf[16];
+  tree decl, name, resdecl;
 
-  ninlined = cgraph_inlined_into (node, inlined);
+  sprintf (which_buf, "%c_%d", which, counter++);
+  name = get_file_function_name_long (which_buf);
 
-  /* First of all look for always inline functions.  */
-  for (e = node->callees; e; e = e->next_callee)
-    if (e->callee->local.disregard_inline_limits && e->inline_failed
-	/* ??? It is possible that renaming variable removed the function body
-	   in duplicate_decls. See gcc.c-torture/compile/20011119-2.c  */
-	&& DECL_SAVED_TREE (e->callee->decl))
-      {
-	if (e->callee->output || e->callee == node)
-	  {
- 	    e->inline_failed = N_("recursive inlining");
-	    continue;
-	  }
-	ninlined_callees = cgraph_inlined_callees (e->callee, inlined_callees);
-	cgraph_mark_inline (node, e->callee, inlined, ninlined,
-			    inlined_callees, ninlined_callees);
-	for (y = 0; y < ninlined_callees; y++)
-	  inlined_callees[y]->output = 0, inlined_callees[y]->aux = 0;
-      }
+  decl = build_decl (FUNCTION_DECL, name,
+		     build_function_type (void_type_node, void_list_node));
+  current_function_decl = decl;
+
+  resdecl = build_decl (RESULT_DECL, NULL_TREE, void_type_node);
+  DECL_ARTIFICIAL (resdecl) = 1;
+  DECL_IGNORED_P (resdecl) = 1;
+  DECL_RESULT (decl) = resdecl;
+
+  allocate_struct_function (decl);
+
+  TREE_STATIC (decl) = 1;
+  TREE_USED (decl) = 1;
+  DECL_ARTIFICIAL (decl) = 1;
+  DECL_IGNORED_P (decl) = 1;
+  DECL_NO_INSTRUMENT_FUNCTION_ENTRY_EXIT (decl) = 1;
+  DECL_SAVED_TREE (decl) = body;
+  TREE_PUBLIC (decl) = ! targetm.have_ctors_dtors;
+  DECL_UNINLINABLE (decl) = 1;
+
+  DECL_INITIAL (decl) = make_node (BLOCK);
+  TREE_USED (DECL_INITIAL (decl)) = 1;
 
-  if (!flag_really_no_inline)
+  DECL_SOURCE_LOCATION (decl) = input_location;
+  cfun->function_end_locus = input_location;
+
+  switch (which)
     {
-      /* Now do the automatic inlining.  */
-      for (e = node->callees; e; e = e->next_callee)
-	if (e->callee->local.inlinable && e->inline_failed
-	    && cgraph_default_inline_p (e->callee)
-	    && cgraph_check_inline_limits (node, e->callee, inlined,
-					   ninlined, &e->inline_failed)
-	    && DECL_SAVED_TREE (e->callee->decl))
-	  {
-	    /* Marking recursive function inlinine has sane semantic and thus
-	       we should not warn on it.  */
-	    if (e->callee->output || e->callee == node)
-	      {
-		e->inline_failed = "";
-		continue;
-	      }
-	    ninlined_callees = cgraph_inlined_callees (e->callee,
-						       inlined_callees);
-	    cgraph_mark_inline (node, e->callee, inlined, ninlined,
-				inlined_callees, ninlined_callees);
-	    for (y = 0; y < ninlined_callees; y++)
-	      inlined_callees[y]->output = 0, inlined_callees[y]->aux = 0;
-	  }
+    case 'I':
+      DECL_STATIC_CONSTRUCTOR (decl) = 1;
+      break;
+    case 'D':
+      DECL_STATIC_DESTRUCTOR (decl) = 1;
+      break;
+    default:
+      gcc_unreachable ();
     }
 
-  /* Clear the flags set by cgraph_inlined_into.  */
-  for (y = 0; y < ninlined; y++)
-    inlined[y]->output = 0, inlined[y]->aux = 0;
+  gimplify_function_tree (decl);
 
-  free (inlined);
-  free (inlined_callees);
-}
+  /* ??? We will get called LATE in the compilation process.  */
+  if (cgraph_global_info_ready)
+    {
+      tree_lowering_passes (decl);
+      tree_rest_of_compilation (decl);
+    }
+  else
+    cgraph_finalize_function (decl, 0);
 
+  if (targetm.have_ctors_dtors)
+    {
+      void (*fn) (rtx, int);
 
-/* Return true when CALLER_DECL should be inlined into CALLEE_DECL.
-   When returned false and reason is non-NULL, set it to the reason
-   why the call was not inlined.  */
+      if (which == 'I')
+	fn = targetm.asm_out.constructor;
+      else
+	fn = targetm.asm_out.destructor;
+      fn (XEXP (DECL_RTL (decl), 0), priority);
+    }
+}
 
-bool
-cgraph_inline_p (tree caller_decl, tree callee_decl, const char **reason)
+void
+init_cgraph (void)
 {
-  struct cgraph_node *caller = cgraph_node (caller_decl);
-  struct cgraph_node *callee = cgraph_node (callee_decl);
-  struct cgraph_edge *e;
-
-  for (e = caller->callees; e; e = e->next_callee)
-    if (e->callee == callee)
-      {
-	if (e->inline_failed && reason)
-	  *reason = e->inline_failed;
-        return !e->inline_failed;
-      }
-  /* We do not record builtins in the callgraph.  Perhaps it would make more
-     sense to do so and then prune out those not overwritten by explicit
-     function body.  */
-  if (reason)
-    *reason = "originally indirect function calls never inlined";
-  return false;
+  cgraph_dump_file = dump_begin (TDI_cgraph, NULL);
 }
-/* Expand all functions that must be output.
 
-   Attempt to topologically sort the nodes so function is output when
-   all called functions are already assembled to allow data to be
-   propagated across the callgraph.  Use a stack to get smaller distance
-   between a function and it's callees (later we may choose to use a more
-   sophisticated algorithm for function reordering; we will likely want
-   to use subsections to make the output functions appear in top-down
-   order).  */
+/* The edges representing the callers of the NEW_VERSION node were
+   fixed by cgraph_function_versioning (), now the call_expr in their
+   respective tree code should be updated to call the NEW_VERSION.  */
 
 static void
-cgraph_expand_all_functions (void)
+update_call_expr (struct cgraph_node *new_version)
 {
-  struct cgraph_node *node;
-  struct cgraph_node **order =
-    xcalloc (cgraph_n_nodes, sizeof (struct cgraph_node *));
-  int order_pos = 0;
-  int i;
-
-  cgraph_mark_functions_to_output ();
-
-  order_pos = cgraph_postorder (order);
+  struct cgraph_edge *e;
 
-  for (i = order_pos - 1; i >= 0; i--)
-    {
-      node = order[i];
-      if (node->output)
-	{
-	  if (!node->reachable)
-	    abort ();
-	  node->output = 0;
-	  cgraph_expand_function (node);
-	}
-    }
-  free (order);
+  gcc_assert (new_version);
+  for (e = new_version->callers; e; e = e->next_caller)
+    /* Update the call expr on the edges
+       to call the new version.  */
+    TREE_OPERAND (TREE_OPERAND (get_call_expr_in (e->call_stmt), 0), 0) = new_version->decl;
 }
 
-/* Mark all local functions.
 
-   A local function is one whose calls can occur only in the
-   current compilation unit and all it's calls are explicit,
-   so we can change its calling convention.
-   We simply mark all static functions whose address is not taken
-   as local.  */
+/* Create a new cgraph node which is the new version of
+   OLD_VERSION node.  REDIRECT_CALLERS holds the callers
+   edges which should be redirected to point to
+   NEW_VERSION.  ALL the callees edges of OLD_VERSION
+   are cloned to the new version node.  Return the new
+   version node.  */
+
+static struct cgraph_node *
+cgraph_copy_node_for_versioning (struct cgraph_node *old_version,
+				 tree new_decl,
+				 VEC(cgraph_edge_p,heap) *redirect_callers)
+ {
+   struct cgraph_node *new_version;
+   struct cgraph_edge *e, *new_e;
+   struct cgraph_edge *next_callee;
+   unsigned i;
+
+   gcc_assert (old_version);
+
+   new_version = cgraph_node (new_decl);
+
+   new_version->analyzed = true;
+   new_version->local = old_version->local;
+   new_version->global = old_version->global;
+   new_version->rtl = new_version->rtl;
+   new_version->reachable = true;
+   new_version->count = old_version->count;
+
+   /* Clone the old node callees.  Recursive calls are
+      also cloned.  */
+   for (e = old_version->callees;e; e=e->next_callee)
+     {
+       new_e = cgraph_clone_edge (e, new_version, e->call_stmt, 0, e->loop_nest, true);
+       new_e->count = e->count;
+     }
+   /* Fix recursive calls.
+      If OLD_VERSION has a recursive call after the
+      previous edge cloning, the new version will have an edge
+      pointing to the old version, which is wrong;
+      Redirect it to point to the new version. */
+   for (e = new_version->callees ; e; e = next_callee)
+     {
+       next_callee = e->next_callee;
+       if (e->callee == old_version)
+	 cgraph_redirect_edge_callee (e, new_version);
+
+       if (!next_callee)
+	 break;
+     }
+   for (i = 0; VEC_iterate (cgraph_edge_p, redirect_callers, i, e); i++)
+     {
+       /* Redirect calls to the old version node to point to its new
+	  version.  */
+       cgraph_redirect_edge_callee (e, new_version);
+     }
+
+   return new_version;
+ }
+
+ /* Perform function versioning.
+    Function versioning includes copying of the tree and
+    a callgraph update (creating a new cgraph node and updating
+    its callees and callers).
+
+    REDIRECT_CALLERS varray includes the edges to be redirected
+    to the new version.
+
+    TREE_MAP is a mapping of tree nodes we want to replace with
+    new ones (according to results of prior analysis).
+    OLD_VERSION_NODE is the node that is versioned.
+    It returns the new version's cgraph node.  */
+
+struct cgraph_node *
+cgraph_function_versioning (struct cgraph_node *old_version_node,
+			    VEC(cgraph_edge_p,heap) *redirect_callers,
+			    varray_type tree_map)
+{
+  tree old_decl = old_version_node->decl;
+  struct cgraph_node *new_version_node = NULL;
+  tree new_decl;
+
+  if (!tree_versionable_function_p (old_decl))
+    return NULL;
+
+  /* Make a new FUNCTION_DECL tree node for the
+     new version. */
+  new_decl = copy_node (old_decl);
+
+  /* Create the new version's call-graph node.
+     and update the edges of the new node. */
+  new_version_node =
+    cgraph_copy_node_for_versioning (old_version_node, new_decl,
+				     redirect_callers);
+
+  /* Copy the OLD_VERSION_NODE function tree to the new version.  */
+  tree_function_versioning (old_decl, new_decl, tree_map, false);
+  /* Update the call_expr on the edges to call the new version node. */
+  update_call_expr (new_version_node);
+
+  /* Update the new version's properties.
+     Make The new version visible only within this translation unit.
+     ??? We cannot use COMDAT linkage because there is no
+     ABI support for this.  */
+  DECL_EXTERNAL (new_version_node->decl) = 0;
+  DECL_ONE_ONLY (new_version_node->decl) = 0;
+  TREE_PUBLIC (new_version_node->decl) = 0;
+  DECL_COMDAT (new_version_node->decl) = 0;
+  new_version_node->local.externally_visible = 0;
+  new_version_node->local.local = 1;
+  new_version_node->lowered = true;
+  return new_version_node;
+}
 
-static void
-cgraph_mark_local_functions (void)
+/* Produce separate function body for inline clones so the offline copy can be
+   modified without affecting them.  */
+struct cgraph_node *
+save_inline_function_body (struct cgraph_node *node)
 {
-  struct cgraph_node *node;
+  struct cgraph_node *first_clone;
 
-  if (cgraph_dump_file)
-    fprintf (cgraph_dump_file, "\nMarking local functions:");
+  gcc_assert (node == cgraph_node (node->decl));
 
-  /* Figure out functions we want to assemble.  */
-  for (node = cgraph_nodes; node; node = node->next)
+  cgraph_lower_function (node);
+
+  /* In non-unit-at-a-time we construct full fledged clone we never output to
+     assembly file.  This clone is pointed out by inline_decl of original function
+     and inlining infrastructure knows how to deal with this.  */
+  if (!flag_unit_at_a_time)
     {
-      node->local.local = (!node->needed
-		           && DECL_SAVED_TREE (node->decl)
-		           && !TREE_PUBLIC (node->decl));
-      if (cgraph_dump_file && node->local.local)
-	fprintf (cgraph_dump_file, " %s", cgraph_node_name (node));
-    }
-  if (cgraph_dump_file)
-    fprintf (cgraph_dump_file, "\n\n");
-}
+      struct cgraph_edge *e;
 
-/* Perform simple optimizations based on callgraph.  */
+      first_clone = cgraph_clone_node (node, node->count, 0, false);
+      first_clone->needed = 0;
+      first_clone->reachable = 1;
+      /* Recursively clone all bodies.  */
+      for (e = first_clone->callees; e; e = e->next_callee)
+	if (!e->inline_failed)
+	  cgraph_clone_inlined_nodes (e, true, false);
+    }
+  else
+    first_clone = node->next_clone;
 
-void
-cgraph_optimize (void)
-{
+  first_clone->decl = copy_node (node->decl);
+  node->next_clone = NULL;
   if (!flag_unit_at_a_time)
-    return;
-  timevar_push (TV_CGRAPHOPT);
-  if (!quiet_flag)
-    fprintf (stderr, "Performing intraprocedural optimizations\n");
+    node->inline_decl = first_clone->decl;
+  first_clone->prev_clone = NULL;
+  cgraph_insert_node_to_hashtable (first_clone);
+  gcc_assert (first_clone == cgraph_node (first_clone->decl));
 
-  cgraph_mark_local_functions ();
-  if (cgraph_dump_file)
-    {
-      fprintf (cgraph_dump_file, "Marked ");
-      dump_cgraph (cgraph_dump_file);
-    }
+  /* Copy the OLD_VERSION_NODE function tree to the new version.  */
+  tree_function_versioning (node->decl, first_clone->decl, NULL, true);
 
-  cgraph_decide_inlining ();
-  cgraph_global_info_ready = true;
-  if (cgraph_dump_file)
-    {
-      fprintf (cgraph_dump_file, "Optimized ");
-      dump_cgraph (cgraph_dump_file);
-    }
-  timevar_pop (TV_CGRAPHOPT);
+  DECL_EXTERNAL (first_clone->decl) = 0;
+  DECL_ONE_ONLY (first_clone->decl) = 0;
+  TREE_PUBLIC (first_clone->decl) = 0;
+  DECL_COMDAT (first_clone->decl) = 0;
 
-  /* Output everything.  */
-  if (!quiet_flag)
-    fprintf (stderr, "Assembling functions:\n");
-  cgraph_expand_all_functions ();
-  if (cgraph_dump_file)
-    {
-      fprintf (cgraph_dump_file, "\nFinal ");
-      dump_cgraph (cgraph_dump_file);
-    }
+  for (node = first_clone->next_clone; node; node = node->next_clone)
+    node->decl = first_clone->decl;
+#ifdef ENABLE_CHECKING
+  verify_cgraph_node (first_clone);
+#endif
+  return first_clone;
 }
+
+#include "gt-cgraphunit.h"