re PR rtl-optimization/40761 (IRA memory hog for insanely nested loops)

2012-01-19  Vladimir Makarov  <vmakarov@redhat.com>

	PR rtl-optimization/40761
	* ira-int.h (struct ira_loop_tree_node): Add comment for member
	loop.  Add new member loop_num.
	(IRA_LOOP_NODE_BY_INDEX): Modify the check.
	(ira_build): Remove the parameter.

	* ira.c (ira_print_disposition): Use loop_num instead of
	loop->num.
	(ira.c): Do not build CFG loops for one region allocation.  Remove
	argument from ira_build call.

	* ira-build.c (init_loop_tree_node): New function.
	(create_loop_tree_nodes): Use it.  Separate the case when CFG
	loops are not built.
	(more_one_region_p): Check current_loops.
	(finish_loop_tree_nodes): Separate the case when CFG loops are not
	built.
	(add_loop_to_tree): Process loop equal to NULL too.
	(form_loop_tree): Separate the case when CFG loops are not built.
	Use explicitly number for the root.
	(rebuild_regno_allocno_maps, create_loop_tree_node_allocnos): Add
	an assertion.
	(ira_print_expanded_allocno, loop_compare_func): Use loop_num
	instead of loop->num.
	(mark_loops_for_removal): Ditto.  Use loop_num instead of
	loop->num.
	(mark_all_loops_for_removal): Ditto.
	(remove_unnecessary_regions): Separate the case when CFG loops
	are not built.
	(ira_build): Remove the parameter.  Use explicit number of regions
	when CFG loops are not built.

	* ira-color.c (print_loop_title): Separate the case for the root
	node.  Use loop_num instead of loop->num.
	(move_spill_restore): Use loop_num instead of loop->num.

	* ira-emit.c (setup_entered_from_non_parent_p): Add an assertion.
	(change_loop): Ditto.
	(change_loop): Use loop_num instead of loop->num.

	* ira-lives.c (process_bb_node_lives): Ditto.

	* ira-costs.c (print_allocno_costs, find_costs_and_classes):
	Ditto.

	* ira-conflicts.c (print_allocno_conflicts): Ditto.

From-SVN: r183312

gcc/ChangeLog

+2012-01-19  Vladimir Makarov  <vmakarov@redhat.com>
+
+	PR rtl-optimization/40761
+	* ira-int.h (struct ira_loop_tree_node): Add comment for member
+	loop.  Add new member loop_num.
+	(IRA_LOOP_NODE_BY_INDEX): Modify the check.
+	(ira_build): Remove the parameter.
+
+	* ira.c (ira_print_disposition): Use loop_num instead of
+	loop->num.
+	(ira.c): Do not build CFG loops for one region allocation.  Remove
+	argument from ira_build call.
+
+	* ira-build.c (init_loop_tree_node): New function.
+	(create_loop_tree_nodes): Use it.  Separate the case when CFG
+	loops are not built.
+	(more_one_region_p): Check current_loops.
+	(finish_loop_tree_nodes): Separate the case when CFG loops are not
+	built.
+	(add_loop_to_tree): Process loop equal to NULL too.
+	(form_loop_tree): Separate the case when CFG loops are not built.
+	Use explicitly number for the root.
+	(rebuild_regno_allocno_maps, create_loop_tree_node_allocnos): Add
+	an assertion.
+	(ira_print_expanded_allocno, loop_compare_func): Use loop_num
+	instead of loop->num.
+	(mark_loops_for_removal): Ditto.  Use loop_num instead of
+	loop->num.
+	(mark_all_loops_for_removal): Ditto.
+	(remove_unnecessary_regions): Separate the case when CFG loops
+	are not built.
+	(ira_build): Remove the parameter.  Use explicit number of regions
+	when CFG loops are not built.
+
+	* ira-color.c (print_loop_title): Separate the case for the root
+	node.  Use loop_num instead of loop->num.
+	(move_spill_restore): Use loop_num instead of loop->num.
+
+	* ira-emit.c (setup_entered_from_non_parent_p): Add an assertion.
+	(change_loop): Ditto.
+	(change_loop): Use loop_num instead of loop->num.
+
+	* ira-lives.c (process_bb_node_lives): Ditto.
+
+	* ira-costs.c (print_allocno_costs, find_costs_and_classes):
+	Ditto.
+
+	* ira-conflicts.c (print_allocno_conflicts): Ditto.
+
 2012-01-19  Jakub Jelinek  <jakub@redhat.com>
 
 	PR libmudflap/40778

gcc/ira-color.c

@@ -1670,7 +1670,6 @@ assign_hard_reg (ira_allocno_t a, bool retry_p)
       update_conflict_hard_regno_costs (full_costs, aclass, false);
     }
   min_cost = min_full_cost = INT_MAX;
-
   /* We don't care about giving callee saved registers to allocnos no
      living through calls because call clobbered registers are
      allocated first (it is usual practice to put them first in
@@ -2011,7 +2010,7 @@ ira_loop_edge_freq (ira_loop_tree_node_t loop_node, int regno, bool exit_p)
   edge e;
   VEC (edge, heap) *edges;
 
-  ira_assert (loop_node->loop != NULL
+  ira_assert (current_loops != NULL && loop_node->loop != NULL
 	      && (regno < 0 || regno >= FIRST_PSEUDO_REGISTER));
   freq = 0;
   if (! exit_p)
@@ -2662,14 +2661,19 @@ print_loop_title (ira_loop_tree_node_t loop_tree_node)
   edge e;
   edge_iterator ei;
 
-  ira_assert (loop_tree_node->loop != NULL);
+  if (loop_tree_node->parent == NULL)
+    fprintf (ira_dump_file,
+	     "\n  Loop 0 (parent -1, header bb%d, depth 0)\n    bbs:",
+	     NUM_FIXED_BLOCKS);
+  else
+    {
+      ira_assert (current_loops != NULL && loop_tree_node->loop != NULL);
       fprintf (ira_dump_file,
 	       "\n  Loop %d (parent %d, header bb%d, depth %d)\n    bbs:",
-	   loop_tree_node->loop->num,
-	   (loop_tree_node->parent == NULL
-	    ? -1 : loop_tree_node->parent->loop->num),
+	       loop_tree_node->loop_num, loop_tree_node->parent->loop_num,
 	       loop_tree_node->loop->header->index,
 	       loop_depth (loop_tree_node->loop));
+    }
   for (subloop_node = loop_tree_node->children;
        subloop_node != NULL;
        subloop_node = subloop_node->next)
@@ -2681,7 +2685,7 @@ print_loop_title (ira_loop_tree_node_t loop_tree_node)
 	      && ((dest_loop_node = IRA_BB_NODE (e->dest)->parent)
 		  != loop_tree_node))
 	    fprintf (ira_dump_file, "(->%d:l%d)",
-		     e->dest->index, dest_loop_node->loop->num);
+		     e->dest->index, dest_loop_node->loop_num);
       }
   fprintf (ira_dump_file, "\n    all:");
   EXECUTE_IF_SET_IN_BITMAP (loop_tree_node->all_allocnos, 0, j, bi)
@@ -3011,7 +3015,7 @@ move_spill_restore (void)
 		  fprintf
 		    (ira_dump_file,
 		     "      Moving spill/restore for a%dr%d up from loop %d",
-		     ALLOCNO_NUM (a), regno, loop_node->loop->num);
+		     ALLOCNO_NUM (a), regno, loop_node->loop_num);
 		  fprintf (ira_dump_file, " - profit %d\n", -cost);
 		}
 	      changed_p = true;

gcc/ira-conflicts.c

@@ -419,6 +419,7 @@ process_regs_for_copy (rtx reg1, rtx reg2, bool constraint_p,
     {
       ira_allocno_t a1 = ira_curr_regno_allocno_map[REGNO (reg1)];
       ira_allocno_t a2 = ira_curr_regno_allocno_map[REGNO (reg2)];
+
       if (!allocnos_conflict_for_copy_p (a1, a2) && offset1 == offset2)
 	{
 	  cp = ira_add_allocno_copy (a1, a2, freq, constraint_p, insn,
@@ -765,7 +766,7 @@ print_allocno_conflicts (FILE * file, bool reg_p, ira_allocno_t a)
       if ((bb = ALLOCNO_LOOP_TREE_NODE (a)->bb) != NULL)
         fprintf (file, "b%d", bb->index);
       else
-        fprintf (file, "l%d", ALLOCNO_LOOP_TREE_NODE (a)->loop->num);
+        fprintf (file, "l%d", ALLOCNO_LOOP_TREE_NODE (a)->loop_num);
       putc (')', file);
     }
 
@@ -796,7 +797,7 @@ print_allocno_conflicts (FILE * file, bool reg_p, ira_allocno_t a)
 		fprintf (file, ",b%d", bb->index);
 	      else
 		fprintf (file, ",l%d",
-			 ALLOCNO_LOOP_TREE_NODE (conflict_a)->loop->num);
+			 ALLOCNO_LOOP_TREE_NODE (conflict_a)->loop_num);
 	      putc (')', file);
 	    }
 	}

gcc/ira-costs.c

@@ -1391,7 +1391,7 @@ print_allocno_costs (FILE *f)
       if ((bb = ALLOCNO_LOOP_TREE_NODE (a)->bb) != NULL)
 	fprintf (f, "b%d", bb->index);
       else
-	fprintf (f, "l%d", ALLOCNO_LOOP_TREE_NODE (a)->loop->num);
+	fprintf (f, "l%d", ALLOCNO_LOOP_TREE_NODE (a)->loop_num);
       fprintf (f, ") costs:");
       for (k = 0; k < cost_classes_ptr->num; k++)
 	{
@@ -1789,7 +1789,7 @@ find_costs_and_classes (FILE *dump_file)
 		    fprintf (dump_file, "b%d", bb->index);
 		  else
 		    fprintf (dump_file, "l%d",
-			     ALLOCNO_LOOP_TREE_NODE (a)->loop->num);
+			     ALLOCNO_LOOP_TREE_NODE (a)->loop_num);
 		  fprintf (dump_file, ") best %s, allocno %s\n",
 			   reg_class_names[best],
 			   reg_class_names[regno_aclass[i]]);

gcc/ira-emit.c

@@ -438,6 +438,7 @@ setup_entered_from_non_parent_p (void)
   unsigned int i;
   loop_p loop;
 
+  ira_assert (current_loops != NULL);
   FOR_EACH_VEC_ELT (loop_p, ira_loops.larray, i, loop)
     if (ira_loop_nodes[i].regno_allocno_map != NULL)
       ira_loop_nodes[i].entered_from_non_parent_p
@@ -565,6 +566,7 @@ change_loop (ira_loop_tree_node_t node)
 
   if (node != ira_loop_tree_root)
     {
+      ira_assert (current_loops != NULL);
       
       if (node->bb != NULL)
 	{
@@ -580,7 +582,7 @@ change_loop (ira_loop_tree_node_t node)
       if (internal_flag_ira_verbose > 3 && ira_dump_file != NULL)
 	fprintf (ira_dump_file,
 		 "      Changing RTL for loop %d (header bb%d)\n",
-		 node->loop->num, node->loop->header->index);
+		 node->loop_num, node->loop->header->index);
 
       parent = ira_curr_loop_tree_node->parent;
       map = parent->regno_allocno_map;

gcc/ira-int.h

@@ -87,7 +87,8 @@ struct ira_loop_tree_node
 {
   /* The node represents basic block if children == NULL.  */
   basic_block bb;    /* NULL for loop.  */
-  struct loop *loop; /* NULL for BB.  */
+  /* NULL for BB or for loop tree root if we did not build CFG loop tree.  */
+  struct loop *loop;
   /* NEXT/SUBLOOP_NEXT is the next node/loop-node of the same parent.
      SUBLOOP_NEXT is always NULL for BBs.  */
   ira_loop_tree_node_t subloop_next, next;
@@ -103,6 +104,9 @@ struct ira_loop_tree_node
   /* All the following members are defined only for nodes representing
      loops.  */
 
+  /* The loop number from CFG loop tree.  The root number is 0.  */
+  int loop_num;
+
   /* True if the loop was marked for removal from the register
      allocation.  */
   bool to_remove_p;
@@ -176,8 +180,9 @@ extern ira_loop_tree_node_t ira_loop_nodes;
 /* Two access macros to the nodes representing loops.  */
 #if defined ENABLE_IRA_CHECKING && (GCC_VERSION >= 2007)
 #define IRA_LOOP_NODE_BY_INDEX(index) __extension__			\
-(({ ira_loop_tree_node_t const _node = (&ira_loop_nodes[index]);\
-     if (_node->children == NULL || _node->bb != NULL || _node->loop == NULL)\
+(({ ira_loop_tree_node_t const _node = (&ira_loop_nodes[index]);	\
+     if (_node->children == NULL || _node->bb != NULL			\
+         || (_node->loop == NULL && current_loops != NULL))		\
        {								\
          fprintf (stderr,						\
                   "\n%s: %d: error in %s: it is not a loop node\n",	\
@@ -989,7 +994,7 @@ extern int *ira_allocate_cost_vector (reg_class_t);
 extern void ira_free_cost_vector (int *, reg_class_t);
 
 extern void ira_flattening (int, int);
-extern bool ira_build (bool);
+extern bool ira_build (void);
 extern void ira_destroy (void);
 
 /* ira-costs.c */

gcc/ira-lives.c

@@ -1123,7 +1123,7 @@ process_bb_node_lives (ira_loop_tree_node_t loop_tree_node)
 
 	  if (internal_flag_ira_verbose > 2 && ira_dump_file != NULL)
 	    fprintf (ira_dump_file, "   Insn %u(l%d): point = %d\n",
-		     INSN_UID (insn), loop_tree_node->parent->loop->num,
+		     INSN_UID (insn), loop_tree_node->parent->loop_num,
 		     curr_point);
 
 	  /* Mark each defined value as live.  We need to do this for

gcc/ira.c

@@ -718,7 +718,7 @@ ira_print_disposition (FILE *f)
 	if ((bb = ALLOCNO_LOOP_TREE_NODE (a)->bb) != NULL)
 	  fprintf (f, "b%-3d", bb->index);
 	else
-	  fprintf (f, "l%-3d", ALLOCNO_LOOP_TREE_NODE (a)->loop->num);
+	  fprintf (f, "l%-3d", ALLOCNO_LOOP_TREE_NODE (a)->loop_num);
 	if (ALLOCNO_HARD_REGNO (a) >= 0)
 	  fprintf (f, " %3d", ALLOCNO_HARD_REGNO (a));
 	else
@@ -3614,14 +3614,16 @@ ira (FILE *f)
   ira_move_loops_num = ira_additional_jumps_num = 0;
 
   ira_assert (current_loops == NULL);
+  if (flag_ira_region == IRA_REGION_ALL || flag_ira_region == IRA_REGION_MIXED)
+    {
       flow_loops_find (&ira_loops);
       record_loop_exits ();
       current_loops = &ira_loops;
+    }
 
   if (internal_flag_ira_verbose > 0 && ira_dump_file != NULL)
     fprintf (ira_dump_file, "Building IRA IR\n");
-  loops_p = ira_build (flag_ira_region == IRA_REGION_ALL
-		       || flag_ira_region == IRA_REGION_MIXED);
+  loops_p = ira_build ();
 
   ira_assert (ira_conflicts_p || !loops_p);
 
@@ -3745,8 +3747,11 @@ do_reload (void)
 
   flag_ira_share_spill_slots = saved_flag_ira_share_spill_slots;
 
+  if (current_loops != NULL)
+    {
       flow_loops_free (&ira_loops);
       free_dominance_info (CDI_DOMINATORS);
+    }
   FOR_ALL_BB (bb)
     bb->loop_father = NULL;
   current_loops = NULL;