cgraph.c (cgraph_turn_edge_to_speculative): Return newly introduced edge; fix typo in sanity check.

	* cgraph.c (cgraph_turn_edge_to_speculative): Return newly
	introduced edge; fix typo in sanity check.
	(cgraph_resolve_speculation): Export; improve diagnostic.
	(cgraph_redirect_edge_call_stmt_to_callee): Better diagnostic; cancel
	speculation at type mismatch.
	* cgraph.h (cgraph_turn_edge_to_speculative): Update.
	(cgraph_resolve_speculation): Declare.
	(symtab_can_be_discarded): New function.
	* value-prof.c (gimple_ic_transform): Remove actual transform code.
	* ipa-inline-transform.c (speculation_removed): New global var.
	(clone_inlined_nodes): See if speculation can be removed.
	(inline_call): If speculations was removed, we growths may not match.
	* ipa-inline.c (can_inline_edge_p): Add DISREGARD_LIMITS parameter.
	(speculation_useful_p): New function.
	(resolve_noninline_speculation): New function.
	(inline_small_functions): Resolve useless speculations.
	* ipa-inline.h (speculation_useful_p): Declare
	* ipa.c (can_replace_by_local_alias): Simplify.
	(ipa_profile): Produce speculative calls in non-lto, too;
	add simple cost model; produce local aliases.

From-SVN: r201683

gcc/ChangeLog

+2013-08-13  Jan Hubicka  <jh@suse.cz>
+
+	* cgraph.c (cgraph_turn_edge_to_speculative): Return newly
+	introduced edge; fix typo in sanity check.
+	(cgraph_resolve_speculation): Export; improve diagnostic.
+	(cgraph_redirect_edge_call_stmt_to_callee): Better diagnostic; cancel
+	speculation at type mismatch.
+	* cgraph.h (cgraph_turn_edge_to_speculative): Update.
+	(cgraph_resolve_speculation): Declare.
+	(symtab_can_be_discarded): New function.
+	* value-prof.c (gimple_ic_transform): Remove actual transform code.
+	* ipa-inline-transform.c (speculation_removed): New global var.
+	(clone_inlined_nodes): See if speculation can be removed.
+	(inline_call): If speculations was removed, we growths may not match.
+	* ipa-inline.c (can_inline_edge_p): Add DISREGARD_LIMITS parameter.
+	(speculation_useful_p): New function.
+	(resolve_noninline_speculation): New function.
+	(inline_small_functions): Resolve useless speculations.
+	* ipa-inline.h (speculation_useful_p): Declare
+	* ipa.c (can_replace_by_local_alias): Simplify.
+	(ipa_profile): Produce speculative calls in non-lto, too;
+	add simple cost model; produce local aliases.
+
 2013-08-13  David Malcolm  <dmalcolm@redhat.com>
 
 	* config/i386/t-i386 (i386.o): Rename stray PIPELINE_H to

gcc/cgraph.c

@@ -1040,9 +1040,11 @@ cgraph_set_edge_callee (struct cgraph_edge *e, struct cgraph_node *n)
 
    At this time the function just creates the direct call,
    the referencd representing the if conditional and attaches
-   them all to the orginal indirect call statement.  */
+   them all to the orginal indirect call statement.  
 
-void
+   Return direct edge created.  */
+
+struct cgraph_edge *
 cgraph_turn_edge_to_speculative (struct cgraph_edge *e,
 				 struct cgraph_node *n2,
 				 gcov_type direct_count,
@@ -1073,6 +1075,7 @@ cgraph_turn_edge_to_speculative (struct cgraph_edge *e,
 			      IPA_REF_ADDR, e->call_stmt);
   ref->lto_stmt_uid = e->lto_stmt_uid;
   ref->speculative = e->speculative;
+  return e2;
 }
 
 /* Speculative call consist of three components:
@@ -1107,7 +1110,7 @@ cgraph_speculative_call_info (struct cgraph_edge *e,
       if (e2->call_stmt)
 	{
 	  e = cgraph_edge (e->caller, e2->call_stmt);
-	  gcc_assert (!e->speculative && !e->indirect_unknown_callee);
+	  gcc_assert (e->speculative && !e->indirect_unknown_callee);
 	}
       else
 	for (e = e->caller->callees; 
@@ -1147,7 +1150,7 @@ cgraph_redirect_edge_callee (struct cgraph_edge *e, struct cgraph_node *n)
    Remove the speculative call sequence and return edge representing the call.
    It is up to caller to redirect the call as appropriate. */
 
-static struct cgraph_edge *
+struct cgraph_edge *
 cgraph_resolve_speculation (struct cgraph_edge *edge, tree callee_decl)
 {
   struct cgraph_edge *e2;
@@ -1159,12 +1162,21 @@ cgraph_resolve_speculation (struct cgraph_edge *edge, tree callee_decl)
     {
       if (dump_file)
 	{
-	  fprintf (dump_file, "Speculative indirect call %s/%i => %s/%i has "
-		   "turned out to have contradicitng known target ",
-		   xstrdup (cgraph_node_name (edge->caller)), edge->caller->symbol.order,
-		   xstrdup (cgraph_node_name (e2->callee)), e2->callee->symbol.order);
-	  print_generic_expr (dump_file, callee_decl, 0);
-          fprintf (dump_file, "\n");
+	  if (callee_decl)
+	    {
+	      fprintf (dump_file, "Speculative indirect call %s/%i => %s/%i has "
+		       "turned out to have contradicting known target ",
+		       xstrdup (cgraph_node_name (edge->caller)), edge->caller->symbol.order,
+		       xstrdup (cgraph_node_name (e2->callee)), e2->callee->symbol.order);
+	      print_generic_expr (dump_file, callee_decl, 0);
+	      fprintf (dump_file, "\n");
+	    }
+	  else
+	    {
+	      fprintf (dump_file, "Removing speculative call %s/%i => %s/%i\n",
+		       xstrdup (cgraph_node_name (edge->caller)), edge->caller->symbol.order,
+		       xstrdup (cgraph_node_name (e2->callee)), e2->callee->symbol.order);
+	    }
 	}
     }
   else
@@ -1264,12 +1276,24 @@ cgraph_redirect_edge_call_stmt_to_callee (struct cgraph_edge *e)
       cgraph_speculative_call_info (e, e, e2, ref);
       if (gimple_call_fndecl (e->call_stmt))
 	e = cgraph_resolve_speculation (e, gimple_call_fndecl (e->call_stmt));
-      else
+      if (!gimple_check_call_matching_types (e->call_stmt, e->callee->symbol.decl,
+					     true))
 	{
+	  e = cgraph_resolve_speculation (e, NULL);
 	  if (dump_file)
-	    fprintf (dump_file, "Expanding speculative call of %s/%i -> %s/%i\n",
+	    fprintf (dump_file, "Not expanding speculative call of %s/%i -> %s/%i\n"
+		     "Type mismatch.\n",
 		     xstrdup (cgraph_node_name (e->caller)), e->caller->symbol.order,
 		     xstrdup (cgraph_node_name (e->callee)), e->callee->symbol.order);
+	}
+      else
+	{
+	  if (dump_file)
+	    fprintf (dump_file, "Expanding speculative call of %s/%i -> %s/%i count:"
+		     HOST_WIDEST_INT_PRINT_DEC"\n",
+		     xstrdup (cgraph_node_name (e->caller)), e->caller->symbol.order,
+		     xstrdup (cgraph_node_name (e->callee)), e->callee->symbol.order,
+		     (HOST_WIDEST_INT)e->count);
 	  gcc_assert (e2->speculative);
 	  push_cfun (DECL_STRUCT_FUNCTION (e->caller->symbol.decl));
 	  new_stmt = gimple_ic (e->call_stmt, cgraph (ref->referred),

gcc/cgraph.h

@@ -726,7 +726,7 @@ bool cgraph_propagate_frequency (struct cgraph_node *node);
 struct cgraph_node * cgraph_function_node (struct cgraph_node *,
 					   enum availability *avail = NULL);
 bool cgraph_get_body (struct cgraph_node *node);
-void
+struct cgraph_edge *
 cgraph_turn_edge_to_speculative (struct cgraph_edge *,
 				 struct cgraph_node *,
 				 gcov_type, int);
@@ -783,6 +783,7 @@ struct cgraph_node *cgraph_function_versioning (struct cgraph_node *,
 						basic_block, const char *);
 void tree_function_versioning (tree, tree, vec<ipa_replace_map_p, va_gc> *,
 			       bool, bitmap, bool, bitmap, basic_block);
+struct cgraph_edge *cgraph_resolve_speculation (struct cgraph_edge *, tree);
 
 /* In cgraphbuild.c  */
 unsigned int rebuild_cgraph_edges (void);
@@ -1398,4 +1399,16 @@ symtab_real_symbol_p (symtab_node node)
     return false;
   return true;
 }
+
+/* Return true if NODE can be discarded by linker from the binary.  */
+
+static inline bool
+symtab_can_be_discarded (symtab_node node)
+{
+  return (DECL_EXTERNAL (node->symbol.decl)
+	  || (DECL_ONE_ONLY (node->symbol.decl)
+	      && node->symbol.resolution != LDPR_PREVAILING_DEF
+	      && node->symbol.resolution != LDPR_PREVAILING_DEF_IRONLY
+	      && node->symbol.resolution != LDPR_PREVAILING_DEF_IRONLY_EXP));
+}
 #endif  /* GCC_CGRAPH_H  */

gcc/ipa-inline-transform.c

@@ -46,6 +46,7 @@ along with GCC; see the file COPYING3.  If not see
 
 int ncalls_inlined;
 int nfunctions_inlined;
+bool speculation_removed;
 
 /* Scale frequency of NODE edges by FREQ_SCALE.  */
 
@@ -134,6 +135,7 @@ clone_inlined_nodes (struct cgraph_edge *e, bool duplicate,
 		     bool update_original, int *overall_size)
 {
   struct cgraph_node *inlining_into;
+  struct cgraph_edge *next;
 
   if (e->caller->global.inlined_to)
     inlining_into = e->caller->global.inlined_to;
@@ -186,9 +188,17 @@ clone_inlined_nodes (struct cgraph_edge *e, bool duplicate,
   e->callee->global.inlined_to = inlining_into;
 
   /* Recursively clone all bodies.  */
-  for (e = e->callee->callees; e; e = e->next_callee)
-    if (!e->inline_failed)
-      clone_inlined_nodes (e, duplicate, update_original, overall_size);
+  for (e = e->callee->callees; e; e = next)
+    {
+      next = e->next_callee;
+      if (!e->inline_failed)
+        clone_inlined_nodes (e, duplicate, update_original, overall_size);
+      if (e->speculative && !speculation_useful_p (e, true))
+	{
+	  cgraph_resolve_speculation (e, NULL);
+	  speculation_removed = true;
+	}
+    }
 }
 
 
@@ -218,6 +228,7 @@ inline_call (struct cgraph_edge *e, bool update_original,
   bool predicated = inline_edge_summary (e)->predicate != NULL;
 #endif
 
+  speculation_removed = false;
   /* Don't inline inlined edges.  */
   gcc_assert (e->inline_failed);
   /* Don't even think of inlining inline clone.  */
@@ -267,6 +278,7 @@ inline_call (struct cgraph_edge *e, bool update_original,
      error due to INLINE_SIZE_SCALE roudoff errors.  */
   gcc_assert (!update_overall_summary || !overall_size || new_edges_found
 	      || abs (estimated_growth - (new_size - old_size)) <= 1
+	      || speculation_removed
 	      /* FIXME: a hack.  Edges with false predicate are accounted
 		 wrong, we should remove them from callgraph.  */
 	      || predicated);

gcc/ipa-inline.h

@@ -226,6 +226,7 @@ inline_hints do_estimate_edge_hints (struct cgraph_edge *edge);
 void initialize_growth_caches (void);
 void free_growth_caches (void);
 void compute_inline_parameters (struct cgraph_node *, bool);
+bool speculation_useful_p (struct cgraph_edge *e, bool anticipate_inlining);
 
 /* In ipa-inline-transform.c  */
 bool inline_call (struct cgraph_edge *, bool, vec<cgraph_edge_p> *, int *, bool);

gcc/ipa.c

@@ -768,11 +768,7 @@ bool
 can_replace_by_local_alias (symtab_node node)
 {
   return (symtab_node_availability (node) > AVAIL_OVERWRITABLE
-	  && !DECL_EXTERNAL (node->symbol.decl)
-	  && (!DECL_ONE_ONLY (node->symbol.decl)
-	      || node->symbol.resolution == LDPR_PREVAILING_DEF
-	      || node->symbol.resolution == LDPR_PREVAILING_DEF_IRONLY
-	      || node->symbol.resolution == LDPR_PREVAILING_DEF_IRONLY_EXP));
+	  && !symtab_can_be_discarded (node));
 }
 
 /* Mark visibility of all functions.
@@ -1407,53 +1403,9 @@ ipa_profile (void)
   bool something_changed = false;
   int i;
   gcov_type overall_time = 0, cutoff = 0, cumulated = 0, overall_size = 0;
-
-  /* Produce speculative calls: we saved common traget from porfiling into
-     e->common_target_id.  Now, at link time, we can look up corresponding
-     function node and produce speculative call.  */
-  if (in_lto_p)
-    {
-      struct cgraph_edge *e;
-      struct cgraph_node *n,*n2;
-
-      init_node_map (false);
-      FOR_EACH_DEFINED_FUNCTION (n)
-	{
-	  bool update = false;
-
-	  for (e = n->indirect_calls; e; e = e->next_callee)
-	    if (e->indirect_info->common_target_id)
-	      {
-		n2 = find_func_by_profile_id (e->indirect_info->common_target_id);
-		if (n2)
-		  {
-		    if (dump_file)
-		      {
-			fprintf (dump_file, "Indirect call -> direct call from"
-				 " other module %s/%i => %s/%i, prob %3.2f\n",
-				 xstrdup (cgraph_node_name (n)), n->symbol.order,
-				 xstrdup (cgraph_node_name (n2)), n2->symbol.order,
-				 e->indirect_info->common_target_probability
-				 / (float)REG_BR_PROB_BASE);
-		      }
-		    cgraph_turn_edge_to_speculative
-		      (e, n2,
-		       apply_scale (e->count,
-				    e->indirect_info->common_target_probability),
-		       apply_scale (e->frequency,
-				    e->indirect_info->common_target_probability));
-		    update = true;
-		  }
-		else
-		  if (dump_file)
-		    fprintf (dump_file, "Function with profile-id %i not found.\n",
-			     e->indirect_info->common_target_id);
-	       }
-	     if (update)
-	       inline_update_overall_summary (n);
-	   }
-	del_node_map ();
-    }
+  struct cgraph_node *n,*n2;
+  int nindirect = 0, ncommon = 0, nunknown = 0, nuseless = 0, nconverted = 0;
+  bool node_map_initialized = false;
 
   if (dump_file)
     dump_histogram (dump_file, histogram);
@@ -1523,6 +1475,106 @@ ipa_profile (void)
   histogram.release();
   free_alloc_pool (histogram_pool);
 
+  /* Produce speculative calls: we saved common traget from porfiling into
+     e->common_target_id.  Now, at link time, we can look up corresponding
+     function node and produce speculative call.  */
+
+  FOR_EACH_DEFINED_FUNCTION (n)
+    {
+      bool update = false;
+
+      for (e = n->indirect_calls; e; e = e->next_callee)
+	{
+	  if (n->count)
+	    nindirect++;
+	  if (e->indirect_info->common_target_id)
+	    {
+	      if (!node_map_initialized)
+	        init_node_map (false);
+	      node_map_initialized = true;
+	      ncommon++;
+	      n2 = find_func_by_profile_id (e->indirect_info->common_target_id);
+	      if (n2)
+		{
+		  if (dump_file)
+		    {
+		      fprintf (dump_file, "Indirect call -> direct call from"
+			       " other module %s/%i => %s/%i, prob %3.2f\n",
+			       xstrdup (cgraph_node_name (n)), n->symbol.order,
+			       xstrdup (cgraph_node_name (n2)), n2->symbol.order,
+			       e->indirect_info->common_target_probability
+			       / (float)REG_BR_PROB_BASE);
+		    }
+		  if (e->indirect_info->common_target_probability
+		      < REG_BR_PROB_BASE / 2)
+		    {
+		      nuseless++;
+		      if (dump_file)
+			fprintf (dump_file,
+				 "Not speculating: probability is too low.\n");
+		    }
+		  else if (!cgraph_maybe_hot_edge_p (e))
+		    {
+		      nuseless++;
+		      if (dump_file)
+			fprintf (dump_file,
+				 "Not speculating: call is cold.\n");
+		    }
+		  else if (cgraph_function_body_availability (n2)
+			   <= AVAIL_OVERWRITABLE
+			   && symtab_can_be_discarded ((symtab_node) n2))
+		    {
+		      nuseless++;
+		      if (dump_file)
+			fprintf (dump_file,
+				 "Not speculating: target is overwritable "
+				 "and can be discarded.\n");
+		    }
+		  else
+		    {
+		      /* Target may be overwritable, but profile says that
+			 control flow goes to this particular implementation
+			 of N2.  Speculate on the local alias to allow inlining.
+		       */
+		      if (!symtab_can_be_discarded ((symtab_node) n2))
+			n2 = cgraph (symtab_nonoverwritable_alias ((symtab_node)n2));
+		      nconverted++;
+		      cgraph_turn_edge_to_speculative
+			(e, n2,
+			 apply_scale (e->count,
+				      e->indirect_info->common_target_probability),
+			 apply_scale (e->frequency,
+				      e->indirect_info->common_target_probability));
+		      update = true;
+		    }
+		}
+	      else
+		{
+		  if (dump_file)
+		    fprintf (dump_file, "Function with profile-id %i not found.\n",
+			     e->indirect_info->common_target_id);
+		  nunknown++;
+		}
+	    }
+	 }
+       if (update)
+	 inline_update_overall_summary (n);
+     }
+  if (node_map_initialized)
+    del_node_map ();
+  if (dump_file && nindirect)
+    fprintf (dump_file,
+	     "%i indirect calls trained.\n"
+	     "%i (%3.2f%%) have common target.\n"
+	     "%i (%3.2f%%) targets was not found.\n"
+	     "%i (%3.2f%%) speculations seems useless.\n"
+	     "%i (%3.2f%%) speculations produced.\n",
+	     nindirect,
+	     ncommon, ncommon * 100.0 / nindirect,
+	     nunknown, nunknown * 100.0 / nindirect,
+	     nuseless, nuseless * 100.0 / nindirect,
+	     nconverted, nconverted * 100.0 / nindirect);
+
   order_pos = ipa_reverse_postorder (order);
   for (i = order_pos - 1; i >= 0; i--)
     {

gcc/value-prof.c

@@ -1431,8 +1431,6 @@ gimple_ic_transform (gimple_stmt_iterator *gsi)
   gimple stmt = gsi_stmt (*gsi);
   histogram_value histogram;
   gcov_type val, count, all, bb_all;
-  gcov_type prob;
-  gimple modify;
   struct cgraph_node *direct_call;
 
   if (gimple_code (stmt) != GIMPLE_CALL)
@@ -1452,12 +1450,6 @@ gimple_ic_transform (gimple_stmt_iterator *gsi)
   count = histogram->hvalue.counters [1];
   all = histogram->hvalue.counters [2];
 
-  if (4 * count <= 3 * all)
-    {
-      gimple_remove_histogram_value (cfun, stmt, histogram);
-      return false;
-    }
-
   bb_all = gimple_bb (stmt)->count;
   /* The order of CHECK_COUNTER calls is important -
      since check_counter can correct the third parameter
@@ -1469,10 +1461,9 @@ gimple_ic_transform (gimple_stmt_iterator *gsi)
       return false;
     }
 
-  if (all > 0)
-    prob = GCOV_COMPUTE_SCALE (count, all);
-  else
-    prob = 0;
+  if (4 * count <= 3 * all)
+    return false;
+
   direct_call = find_func_by_profile_id ((int)val);
 
   if (direct_call == NULL)
@@ -1488,12 +1479,21 @@ gimple_ic_transform (gimple_stmt_iterator *gsi)
 	}
       return false;
     }
-  gimple_remove_histogram_value (cfun, stmt, histogram);
 
   if (!check_ic_target (stmt, direct_call))
-    return false;
-
-  modify = gimple_ic (stmt, direct_call, prob, count, all);
+    {
+      if (dump_file)
+	{
+	  fprintf (dump_file, "Indirect call -> direct call ");
+	  print_generic_expr (dump_file, gimple_call_fn (stmt), TDF_SLIM);
+	  fprintf (dump_file, "=> ");
+	  print_generic_expr (dump_file, direct_call->symbol.decl, TDF_SLIM);
+	  fprintf (dump_file, " transformation skipped because of type mismatch");
+	  print_gimple_stmt (dump_file, stmt, 0, TDF_SLIM);
+	}
+      gimple_remove_histogram_value (cfun, stmt, histogram);
+      return false;
+    }
 
   if (dump_file)
     {
@@ -1501,10 +1501,8 @@ gimple_ic_transform (gimple_stmt_iterator *gsi)
       print_generic_expr (dump_file, gimple_call_fn (stmt), TDF_SLIM);
       fprintf (dump_file, "=> ");
       print_generic_expr (dump_file, direct_call->symbol.decl, TDF_SLIM);
-      fprintf (dump_file, " transformation on insn ");
+      fprintf (dump_file, " transformation on insn postponned to ipa-profile");
       print_gimple_stmt (dump_file, stmt, 0, TDF_SLIM);
-      fprintf (dump_file, " to ");
-      print_gimple_stmt (dump_file, modify, 0, TDF_SLIM);
       fprintf (dump_file, "hist->count "HOST_WIDEST_INT_PRINT_DEC
 	       " hist->all "HOST_WIDEST_INT_PRINT_DEC"\n", count, all);
     }