Merge ipa-profile pass from pretty-ipa

[Jan Hubicka <hubicka at ucw dot cz>]

Hi,
this patch adds simple pass propagating node frequencies across callgraph.  The
pass is able to determine that functions called only from called once that are
not called inside loop are called once and similar simple analysis.  The
savings are pretty small (about 1-4% on usual spec2000 benchmark, we get lucky
in some cases but those tends to be the simple benchmarks, like gzip), but the
pass is easy too and excercise the profile hot/cold infrastructure more
On SPEC2000 GCC in whole program mode this affects about 300 functions.

Bootstrapped/regtested x86_64-linux, will commit it later today if there
are no complains.

2010-04-18 Jan Hubicka  <jh@suse.cz> 

	* doc/invoke.texi (-fipa-profile): Document.
	* opts.c (decode_options): Enable ipa-profile at -O1.
	* timevar.def (TV_IPA_PROFILE): Define.
	* common.opt (fipa-profile): Add.
	* cgraph.c (cgraph_clone_node): Set local flag and clear vtable method flag
	for clones.
	(cgraph_propagate_frequency): Handle only local ones.
	* tree-pass.h (pass_ipa_profile): Declare.
	* ipa-profile.c (gate_profile): Use flag_ipa_profile.
	(pass_ipa_profile): Use TV_IPA_PROFILE.
	* ipa.c (ipa_profile): New function.
	(gate_ipa_profile): Likewise.
	(pass_ipa_profile): New global variable.
	* passes.c (pass_ipa_profile): New.
Index: doc/invoke.texi
===================================================================
--- doc/invoke.texi	(revision 158772)
+++ doc/invoke.texi	(working copy)
@@ -344,7 +344,7 @@ Objective-C and Objective-C++ Dialects}.
 -fgcse-sm -fif-conversion -fif-conversion2 -findirect-inlining @gol
 -finline-functions -finline-functions-called-once -finline-limit=@var{n} @gol
 -finline-small-functions -fipa-cp -fipa-cp-clone -fipa-matrix-reorg -fipa-pta @gol
--fipa-pure-const -fipa-reference -fipa-struct-reorg @gol
+-fipa-profile -fipa-pure-const -fipa-reference -fipa-struct-reorg @gol
 -fipa-type-escape -fira-algorithm=@var{algorithm} @gol
 -fira-region=@var{region} -fira-coalesce @gol
 -fira-loop-pressure -fno-ira-share-save-slots @gol
@@ -6601,6 +6601,17 @@ Perform interprocedural pointer analysis
 and reference analysis.  This option can cause excessive memory and
 compile-time usage on large compilation units.  It is not enabled by
 default at any optimization level.
+Enabled by default at @option{-O} and higher.
+
+@item -fipa-profile
+@opindex fipa-profile
+Perform interprocedural profile propagation.  The functions called only from
+cold functions are marked as cold. Also functions executed once (such as
+@samp{cold}, @samp{noreturn}, static constructors or destructors) are identified. Cold
+functions and loop less parts of functions executed once are then optimized for
+size.
+
+This flag is enabled by default at @option{-O1}.
 
 @item -fipa-cp
 @opindex fipa-cp
Index: cgraph.c
===================================================================
--- cgraph.c	(revision 158775)
+++ cgraph.c	(working copy)
@@ -1897,6 +1897,8 @@ cgraph_clone_node (struct cgraph_node *n
   new_node->analyzed = n->analyzed;
   new_node->local = n->local;
   new_node->local.externally_visible = false;
+  new_node->local.local = true;
+  new_node->local.vtable_method = false;
   new_node->global = n->global;
   new_node->rtl = n->rtl;
   new_node->count = count;
@@ -2314,7 +2316,7 @@ cgraph_propagate_frequency (struct cgrap
 {
   bool maybe_unlikely_executed = true, maybe_executed_once = true;
   struct cgraph_edge *edge;
-  if (node->needed || node->local.externally_visible)
+  if (!node->local.local)
     return false;
   gcc_assert (node->analyzed);
   if (node->frequency == NODE_FREQUENCY_HOT)
Index: tree-pass.h
===================================================================
--- tree-pass.h	(revision 158772)
+++ tree-pass.h	(working copy)
@@ -459,6 +459,7 @@ extern struct simple_ipa_opt_pass pass_i
 extern struct simple_ipa_opt_pass pass_ipa_struct_reorg;
 extern struct ipa_opt_pass_d pass_ipa_lto_wpa_fixup;
 extern struct ipa_opt_pass_d pass_ipa_lto_finish_out;
+extern struct ipa_opt_pass_d pass_ipa_profile;
 
 extern struct gimple_opt_pass pass_all_optimizations;
 extern struct gimple_opt_pass pass_cleanup_cfg_post_optimizing;
Index: opts.c
===================================================================
--- opts.c	(revision 158772)
+++ opts.c	(working copy)
@@ -858,6 +858,7 @@ decode_options (unsigned int argc, const
   flag_if_conversion2 = opt1;
   flag_ipa_pure_const = opt1;
   flag_ipa_reference = opt1;
+  flag_ipa_profile = opt1;
   flag_merge_constants = opt1;
   flag_split_wide_types = opt1;
   flag_tree_ccp = opt1;
Index: timevar.def
===================================================================
--- timevar.def	(revision 158772)
+++ timevar.def	(working copy)
@@ -59,6 +59,7 @@ DEFTIMEVAR (TV_WHOPR_WPA_IO          , "
 DEFTIMEVAR (TV_WHOPR_LTRANS          , "whopr ltrans")
 DEFTIMEVAR (TV_WHOPR_WPA_LTRANS_EXEC , "whopr wpa->ltrans")
 DEFTIMEVAR (TV_IPA_REFERENCE         , "ipa reference")
+DEFTIMEVAR (TV_IPA_PROFILE           , "ipa profile")
 DEFTIMEVAR (TV_IPA_PURE_CONST        , "ipa pure const")
 DEFTIMEVAR (TV_IPA_TYPE_ESCAPE       , "ipa type escape")
 DEFTIMEVAR (TV_IPA_PTA               , "ipa points-to")
Index: ipa.c
===================================================================
--- ipa.c	(revision 158772)
+++ ipa.c	(working copy)
@@ -741,3 +741,84 @@ debug_cgraph_node_set (cgraph_node_set s
   dump_cgraph_node_set (stderr, set);
 }
 
+/* Simple ipa profile pass propagating frequencies across the callgraph.  */
+
+static unsigned int
+ipa_profile (void)
+{
+  struct cgraph_node **order = XCNEWVEC (struct cgraph_node *, cgraph_n_nodes);
+  struct cgraph_edge *e;
+  int order_pos;
+  bool something_changed = false;
+  int i;
+
+  order_pos = cgraph_postorder (order);
+  for (i = order_pos - 1; i >= 0; i--)
+    {
+      if (order[i]->local.local && cgraph_propagate_frequency (order[i]))
+	{
+	  for (e = order[i]->callees; e; e = e->next_callee)
+	    if (e->callee->local.local && !e->callee->aux)
+	      {
+	        something_changed = true;
+	        e->callee->aux = (void *)1;
+	      }
+	}
+      order[i]->aux = NULL;
+    }
+
+  while (something_changed)
+    {
+      something_changed = false;
+      for (i = order_pos - 1; i >= 0; i--)
+	{
+	  if (order[i]->aux && cgraph_propagate_frequency (order[i]))
+	    {
+	      for (e = order[i]->callees; e; e = e->next_callee)
+		if (e->callee->local.local && !e->callee->aux)
+		  {
+		    something_changed = true;
+		    e->callee->aux = (void *)1;
+		  }
+	    }
+	  order[i]->aux = NULL;
+	}
+    }
+  free (order);
+  return 0;
+}
+
+static bool
+gate_ipa_profile (void)
+{
+  return flag_ipa_profile;
+}
+
+struct ipa_opt_pass_d pass_ipa_profile =
+{
+ {
+  IPA_PASS,
+  "ipa-profile",			/* name */
+  gate_ipa_profile,			/* gate */
+  ipa_profile,			        /* execute */
+  NULL,					/* sub */
+  NULL,					/* next */
+  0,					/* static_pass_number */
+  TV_IPA_PROFILE,		        /* tv_id */
+  0,	                                /* properties_required */
+  0,					/* properties_provided */
+  0,					/* properties_destroyed */
+  0,					/* todo_flags_start */
+  0                                     /* todo_flags_finish */
+ },
+ NULL,				        /* generate_summary */
+ NULL,					/* write_summary */
+ NULL,					/* read_summary */
+ NULL,					/* write_optimization_summary */
+ NULL,					/* read_optimization_summary */
+ NULL,					/* stmt_fixup */
+ 0,					/* TODOs */
+ NULL,			                /* function_transform */
+ NULL					/* variable_transform */
+};
Index: common.opt
===================================================================
--- common.opt	(revision 158772)
+++ common.opt	(working copy)
@@ -693,14 +693,18 @@ fipa-cp-clone
 Common Report Var(flag_ipa_cp_clone) Optimization
 Perform cloning to make Interprocedural constant propagation stronger
 
-fipa-pure-const
-Common Report Var(flag_ipa_pure_const) Init(0) Optimization
-Discover pure and const functions
+fipa-profile
+Common Report Var(flag_ipa_profile) Init(0) Optimization
+Perform interprocedural profile propagation
 
 fipa-pta
 Common Report Var(flag_ipa_pta) Init(0) Optimization
 Perform interprocedural points-to analysis
 
+fipa-pure-const
+Common Report Var(flag_ipa_pure_const) Init(0) Optimization
+Discover pure and const functions
+
 fipa-reference
 Common Report Var(flag_ipa_reference) Init(0) Optimization
 Discover readonly and non addressable static variables
Index: passes.c
===================================================================
--- passes.c	(revision 158772)
+++ passes.c	(working copy)
@@ -803,6 +803,7 @@ init_optimization_passes (void)
 
   p = &all_regular_ipa_passes;
   NEXT_PASS (pass_ipa_whole_program_visibility);
+  NEXT_PASS (pass_ipa_profile);
   NEXT_PASS (pass_ipa_cp);
   NEXT_PASS (pass_ipa_inline);
   NEXT_PASS (pass_ipa_reference);