tree-ssa-loop-im.c: New file.

* tree-ssa-loop-im.c: New file.
	* Makefile.in (tree-ssa-loop-im.o): Add.
	* cfgloop.c (superloop_at_depth): New function.
	* cfgloop.h (superloop_at_depth): Declare.
	* common.opt (ftree-lim): New flag.
	* expr.c (array_ref_up_bound): New function.
	* params.def (PARAM_LIM_EXPENSIVE): New parameter.
	* timevar.def (TV_LIM): New timevar.
	* tree-dfa.c (compute_immediate_uses): Respect TDFA_USE flags when
	computing immediate uses of a phi node.
	* tree-flow.h (struct tree_ann_common_d): Add aux field.
	(loop_commit_inserts, for_each_index, tree_ssa_lim): Declare.
	* tree-optimize.c (init_tree_optimization_passes): Add pass_lim.
	* tree-pass.h (pass_lim): Declare.
	* tree-ssa-loop.c (tree_ssa_loop_im, gate_tree_ssa_loop_im): New
	functions.
	(pass_lim): New pass structure.
	* tree-eh.c (tree_could_trap_p): Handle ARRAY_REFs correctly.
	* tree.c (in_array_bounds_p): New function.
	* tree.h (TREE_THIS_NOTRAP): Define also for ARRAY_REFs.
	(in_array_bounds_p, array_ref_up_bound): Declare.
	* doc/invoke.texi (-ftree-lim, --param lim-expensive): Document.
	* doc/passes.texi (tree-ssa-loop-im.c): Document.

From-SVN: r84441

gcc/ChangeLog

@@ -1,3 +1,29 @@
+2004-07-09  Zdenek Dvorak  <rakdver@atrey.karlin.mff.cuni.cz>
+
+	* tree-ssa-loop-im.c: New file.
+	* Makefile.in (tree-ssa-loop-im.o): Add.
+	* cfgloop.c (superloop_at_depth): New function.
+	* cfgloop.h (superloop_at_depth): Declare.
+	* common.opt (ftree-lim): New flag.
+	* expr.c (array_ref_up_bound): New function.
+	* params.def (PARAM_LIM_EXPENSIVE): New parameter.
+	* timevar.def (TV_LIM): New timevar.
+	* tree-dfa.c (compute_immediate_uses): Respect TDFA_USE flags when
+	computing immediate uses of a phi node.
+	* tree-flow.h (struct tree_ann_common_d): Add aux field.
+	(loop_commit_inserts, for_each_index, tree_ssa_lim): Declare.
+	* tree-optimize.c (init_tree_optimization_passes): Add pass_lim.
+	* tree-pass.h (pass_lim): Declare.
+	* tree-ssa-loop.c (tree_ssa_loop_im, gate_tree_ssa_loop_im): New
+	functions.
+	(pass_lim): New pass structure.
+	* tree-eh.c (tree_could_trap_p): Handle ARRAY_REFs correctly.
+	* tree.c (in_array_bounds_p): New function.
+	* tree.h (TREE_THIS_NOTRAP): Define also for ARRAY_REFs.
+	(in_array_bounds_p, array_ref_up_bound): Declare.
+	* doc/invoke.texi (-ftree-lim, --param lim-expensive): Document.
+	* doc/passes.texi (tree-ssa-loop-im.c): Document.
+
 2004-07-09  Richard Henderson  <rth@redhat.com>
 
 	* builtins.c (expand_builtin_stpcpy): Don't modify len.

gcc/Makefile.in

@@ -901,7 +901,7 @@ OBJS-common = \
  cfg.o cfganal.o cfgbuild.o cfgcleanup.o cfglayout.o cfgloop.o		   \
  cfgloopanal.o cfgloopmanip.o loop-init.o loop-unswitch.o loop-unroll.o	   \
  cfgrtl.o combine.o conflict.o convert.o coverage.o cse.o cselib.o 	   \
- dbxout.o ddg.o tree-ssa-loop-ch.o loop-invariant.o			   \
+ dbxout.o ddg.o tree-ssa-loop-ch.o loop-invariant.o tree-ssa-loop-im.o	   \
  debug.o df.o diagnostic.o dojump.o dominance.o loop-doloop.o		   \
  dwarf2asm.o dwarf2out.o emit-rtl.o except.o explow.o loop-iv.o		   \
  expmed.o expr.o final.o flow.o fold-const.o function.o gcse.o		   \
@@ -1686,6 +1686,10 @@ tree-ssa-loop-ch.o : tree-ssa-loop-ch.c $(TREE_FLOW_H) $(CONFIG_H) \
    $(SYSTEM_H) $(RTL_H) $(TREE_H) $(TM_P_H) $(CFGLOOP_H) tree-inline.h \
    output.h diagnostic.h $(TIMEVAR_H) $(TM_H) coretypes.h $(TREE_DUMP_H) \
    tree-pass.h flags.h
+tree-ssa-loop-im.o : tree-ssa-loop-im.c $(TREE_FLOW_H) $(CONFIG_H) \
+   $(SYSTEM_H) $(RTL_H) $(TREE_H) $(TM_P_H) $(CFGLOOP_H) domwalk.h $(PARAMS_H)\
+   output.h diagnostic.h $(TIMEVAR_H) $(TM_H) coretypes.h $(TREE_DUMP_H) \
+   tree-pass.h flags.h
 tree-ssa-alias.o : tree-ssa-alias.c $(TREE_FLOW_H) $(CONFIG_H) $(SYSTEM_H) \
    $(RTL_H) $(TREE_H) $(TM_P_H) $(EXPR_H) $(GGC_H) tree-inline.h $(FLAGS_H) \
    function.h $(TIMEVAR_H) tree-alias-common.h convert.h $(TM_H) coretypes.h \

gcc/cfgloop.c

@@ -101,6 +101,20 @@ flow_loop_nested_p (const struct loop *outer, const struct loop *loop)
 	 && loop->pred[outer->depth] == outer;
 }
 
+/* Returns superloop of LOOP at given DEPTH.  */
+
+struct loop *
+superloop_at_depth (struct loop *loop, unsigned depth)
+{
+  if (depth > (unsigned) loop->depth)
+    abort ();
+
+  if (depth == (unsigned) loop->depth)
+    return loop;
+
+  return loop->pred[depth];
+}
+
 /* Dump the loop information specified by LOOP to the stream FILE
    using auxiliary dump callback function LOOP_DUMP_AUX if non null.  */
 

gcc/cfgloop.h

@@ -256,6 +256,7 @@ extern bool flow_loop_outside_edge_p (const struct loop *, edge);
 extern bool flow_loop_nested_p	(const struct loop *, const struct loop *);
 extern bool flow_bb_inside_loop_p (const struct loop *, const basic_block);
 extern struct loop * find_common_loop (struct loop *, struct loop *);
+struct loop *superloop_at_depth (struct loop *, unsigned);
 extern int num_loop_insns (struct loop *);
 extern int average_num_loop_insns (struct loop *);
 extern unsigned get_loop_level (const struct loop *);

gcc/common.opt

@@ -844,6 +844,10 @@ ftree-fre
 Common Report Var(flag_tree_fre)
 Enable Full Redundancy Elimination (FRE) on trees
 
+ftree-lim
+Common Report Var(flag_tree_lim) Init(1)
+Enable loop invariant motion on trees
+
 ftree-loop-optimize
 Common Report Var(flag_tree_loop_optimize) Init(1)
 Enable loop optimizations on tree level

gcc/doc/invoke.texi

@@ -313,6 +313,7 @@ in the following sections.
 -funroll-all-loops  -funroll-loops  -fpeel-loops @gol
 -funswitch-loops  -fold-unroll-loops  -fold-unroll-all-loops @gol
 -ftree-pre  -ftree-ccp  -ftree-dce -ftree-loop-optimize @gol
+-ftree-lim @gol
 -ftree-dominator-opts -ftree-dse -ftree-copyrename @gol
 -ftree-ch -ftree-sra -ftree-ter -ftree-lrs -ftree-fre @gol
 --param @var{name}=@var{value}
@@ -4401,6 +4402,14 @@ usually increases code size.
 Perform loop optimizations on trees.  This flag is enabled by default at -O
 and higher.
 
+@item -ftree-lim
+Perform loop invariant motion on trees.  This pass moves only invartiants that
+would be hard to handle on rtl level (function calls, operations that expand to
+nontrivial sequences of insns).  With @option{-funswitch-loops} it also moves
+operands of conditions that are invariant out of the loop, so that we can use
+just trivial invariantness analysis in loop unswitching.  The pass also includes
+store motion.
+
 @item -ftree-sra
 Perform scalar replacement of aggregates.  This pass replaces structure
 references with scalars to prevent committing structures to memory too
@@ -5174,6 +5183,9 @@ The maximum number of insns of an unswitched loop.
 @item max-unswitch-level
 The maximum number of branches unswitched in a single loop.
 
+@item lim-expensive
+The minimum cost of an expensive expression in the loop invariant motion.
+
 @item max-iterations-to-track
 
 The maximum number of iterations of a loop the brute force algorithm

gcc/doc/passes.texi

@@ -363,9 +363,20 @@ and is described by @code{pass_pre}.
 
 @item Loop optimization
 
-TODO: Presumably we're going to do something with loops here.  At
-present we don't, and this is a placeholder.  The pass is located
-in @file{tree-ssa-loop.c} and is described by @code{pass_loop}.
+The main driver of the pass is placed in @file{tree-ssa-loop.c}
+and described by @code{pass_loop}.
+
+The optimizations performed by this pass are:
+
+Loop invariant motion.  This pass moves only invariants that
+would be hard to handle on rtl level (function calls, operations that expand to
+nontrivial sequences of insns).  With @option{-funswitch-loops} it also moves
+operands of conditions that are invariant out of the loop, so that we can use
+just trivial invariantness analysis in loop unswitching.  The pass also includes
+store motion.  The pass is implemented in @file{tree-ssa-loop-im.c}.
+
+The optimizations also use various utility functions contained in
+@file{cfgloop.c}, @file{cfgloopanal.c} and @file{cfgloopmanip.c}.
 
 @item Conditional constant propagation
 

gcc/expr.c

@@ -5706,6 +5706,23 @@ array_ref_low_bound (tree exp)
   return fold_convert (TREE_TYPE (TREE_OPERAND (exp, 1)), integer_zero_node);
 }
 
+/* Return a tree representing the upper bound of the array mentioned in
+   EXP, an ARRAY_REF.  */
+
+tree
+array_ref_up_bound (tree exp)
+{
+  tree domain_type = TYPE_DOMAIN (TREE_TYPE (TREE_OPERAND (exp, 0)));
+
+  /* If there is a domain type and it has an upper bound, use it, substituting
+     for a PLACEHOLDER_EXPR as needed.  */
+  if (domain_type && TYPE_MAX_VALUE (domain_type))
+    return SUBSTITUTE_PLACEHOLDER_IN_EXPR (TYPE_MAX_VALUE (domain_type), exp);
+
+  /* Otherwise fail.  */
+  return NULL_TREE;
+}
+
 /* Return a tree representing the offset, in bytes, of the field referenced
    by EXP.  This does not include any offset in DECL_FIELD_BIT_OFFSET.  */
 

gcc/params.def

@@ -312,6 +312,13 @@ DEFPARAM(PARAM_MAX_CSE_PATH_LENGTH,
 	 "The maximum length of path considered in cse",
 	 10)
 
+/* The cost of expression in loop invariant motion that is considered
+   expensive.  */
+DEFPARAM(PARAM_LIM_EXPENSIVE,
+	 "lim-expensive",
+	 "The minimum cost of an expensive expression in the loop invariant motion",
+	 20)
+
 /* The product of the next two is used to decide whether or not to
    use .GLOBAL_VAR.  See tree-dfa.c.  */
 DEFPARAM(PARAM_GLOBAL_VAR_THRESHOLD,

gcc/timevar.def

@@ -82,6 +82,7 @@ DEFTIMEVAR (TV_TREE_DCE		     , "tree conservative DCE")
 DEFTIMEVAR (TV_TREE_CD_DCE	     , "tree aggressive DCE")
 DEFTIMEVAR (TV_TREE_DSE		     , "tree DSE")
 DEFTIMEVAR (TV_TREE_LOOP	     , "tree loop optimization")
+DEFTIMEVAR (TV_LIM                   , "loop invariant motion")
 DEFTIMEVAR (TV_TREE_CH		     , "tree copy headers")
 DEFTIMEVAR (TV_TREE_SSA_TO_NORMAL    , "tree SSA to normal")
 DEFTIMEVAR (TV_TREE_NRV		     , "tree NRV optimization")

gcc/tree-dfa.c

@@ -178,7 +178,20 @@ compute_immediate_uses (int flags, bool (*calc_for)(tree))
       tree phi;
 
       for (phi = phi_nodes (bb); phi; phi = PHI_CHAIN (phi))
-	compute_immediate_uses_for_phi (phi, calc_for);
+	{
+	  if (is_gimple_reg (PHI_RESULT (phi)))
+	    {
+	      if (!(flags & TDFA_USE_OPS))
+		continue;
+	    }
+	  else
+	    {
+	      if (!(flags & TDFA_USE_VOPS))
+		continue;
+	    }
+
+	  compute_immediate_uses_for_phi (phi, calc_for);
+	}
 
       for (si = bsi_start (bb); !bsi_end_p (si); bsi_next (&si))
         {

gcc/tree-eh.c

@@ -1705,7 +1705,7 @@ tree_could_trap_p (tree expr)
   bool honor_nans = false;
   bool honor_snans = false;
   bool fp_operation = false;
-  tree t;
+  tree t, base, idx;
 
   if (TREE_CODE_CLASS (code) == '<'
       || TREE_CODE_CLASS (code) == '1'
@@ -1722,14 +1722,32 @@ tree_could_trap_p (tree expr)
 
   switch (code)
     {
-    case ARRAY_REF:
-    case ARRAY_RANGE_REF:
     case COMPONENT_REF:
     case REALPART_EXPR:
     case IMAGPART_EXPR:
     case BIT_FIELD_REF:
-      t = get_base_address (expr);
-      return !t || tree_could_trap_p (t);
+      t = TREE_OPERAND (expr, 0);
+      return tree_could_trap_p (t);
+
+    case ARRAY_RANGE_REF:
+      /* Let us be conservative here for now.  We might be checking bounds of
+	 the access similarly to the case below.  */
+      if (!TREE_THIS_NOTRAP (expr))
+	return true;
+
+      base = TREE_OPERAND (expr, 0);
+      return tree_could_trap_p (base);
+
+    case ARRAY_REF:
+      base = TREE_OPERAND (expr, 0);
+      idx = TREE_OPERAND (expr, 1);
+      if (tree_could_trap_p (base))
+	return true;
+
+      if (TREE_THIS_NOTRAP (expr))
+	return false;
+
+      return !in_array_bounds_p (expr);
 
     case INDIRECT_REF:
       return !TREE_THIS_NOTRAP (expr);

gcc/tree-flow.h

@@ -83,6 +83,10 @@ struct tree_ann_common_d GTY(())
   /* Annotation type.  */
   enum tree_ann_type type;
 
+ /* Auxiliary info specific to a pass.  At all times, this
+    should either point to valid data or be NULL.  */
+  PTR GTY ((skip (""))) aux;
+
   /* The value handle for this expression.  Used by GVN-PRE.  */
   tree GTY((skip)) value_handle;
 };
@@ -636,6 +640,8 @@ struct tree_niter_desc
 
 /* In tree-ssa-loop*.c  */
 
+void tree_ssa_lim (struct loops *);
+
 void number_of_iterations_cond (tree, tree, tree, enum tree_code, tree, tree,
 				struct tree_niter_desc *);
 bool number_of_iterations_exit (struct loop *, edge,
@@ -645,6 +651,8 @@ tree find_loop_niter_by_eval (struct loop *, edge *);
 void estimate_numbers_of_iterations (struct loops *);
 tree can_count_iv_in_wider_type (struct loop *, tree, tree, tree, tree);
 void free_numbers_of_iterations_estimates (struct loops *);
+void loop_commit_inserts (void);
+bool for_each_index (tree *, bool (*) (tree, tree *, void *), void *);
 
 /* In tree-flow-inline.h  */
 static inline int phi_arg_from_edge (tree, edge);

gcc/tree-optimize.c

@@ -328,6 +328,7 @@ init_tree_optimization_passes (void)
 
   p = &pass_loop.sub;
   NEXT_PASS (pass_loop_init);
+  NEXT_PASS (pass_lim);
   NEXT_PASS (pass_loop_done);
   *p = NULL;
 

gcc/tree-pass.h

@@ -107,6 +107,7 @@ extern struct tree_opt_pass pass_tail_recursion;
 extern struct tree_opt_pass pass_tail_calls;
 extern struct tree_opt_pass pass_loop;
 extern struct tree_opt_pass pass_loop_init;
+extern struct tree_opt_pass pass_lim;
 extern struct tree_opt_pass pass_loop_done;
 extern struct tree_opt_pass pass_ch;
 extern struct tree_opt_pass pass_ccp;

gcc/tree-ssa-loop.c

@@ -111,6 +111,39 @@ struct tree_opt_pass pass_loop_init =
   0					/* todo_flags_finish */
 };
 
+/* Loop invariant motion pass.  */
+
+static void
+tree_ssa_loop_im (void)
+{
+  if (!current_loops)
+    return;
+
+  tree_ssa_lim (current_loops);
+}
+
+static bool
+gate_tree_ssa_loop_im (void)
+{
+  return flag_tree_lim != 0;
+}
+
+struct tree_opt_pass pass_lim = 
+{
+  "lim",				/* name */
+  gate_tree_ssa_loop_im,		/* gate */
+  tree_ssa_loop_im,			/* execute */
+  NULL,					/* sub */
+  NULL,					/* next */
+  0,					/* static_pass_number */
+  TV_LIM,				/* tv_id */
+  PROP_cfg,				/* properties_required */
+  0,					/* properties_provided */
+  0,					/* properties_destroyed */
+  0,					/* todo_flags_start */
+  TODO_dump_func                	/* todo_flags_finish */
+};
+
 /* Loop optimizer finalization.  */
 
 static void

gcc/tree.c

@@ -5650,6 +5650,34 @@ build_empty_stmt (void)
 }
 
 
+/* Returns true if it is possible to prove that the index of
+   an array access REF (an ARRAY_REF expression) falls into the
+   array bounds.  */
+
+bool
+in_array_bounds_p (tree ref)
+{
+  tree idx = TREE_OPERAND (ref, 1);
+  tree min, max;
+
+  if (TREE_CODE (idx) != INTEGER_CST)
+    return false;
+	    
+  min = array_ref_low_bound (ref);
+  max = array_ref_up_bound (ref);
+  if (!min
+      || !max
+      || TREE_CODE (min) != INTEGER_CST
+      || TREE_CODE (max) != INTEGER_CST)
+    return false;
+
+  if (tree_int_cst_lt (idx, min)
+      || tree_int_cst_lt (max, idx))
+    return false;
+
+  return true;
+}
+
 /* Return true if T (assumed to be a DECL) must be assigned a memory
    location.  */
 

gcc/tree.h

@@ -303,7 +303,7 @@ struct tree_common GTY(())
 	   ..._TYPE
 
        TREE_THIS_NOTRAP in
-          INDIRECT_REF
+          INDIRECT_REF, ARRAY_REF, ARRAY_RANGE_REF
 
    deprecated_flag:
 
@@ -798,7 +798,12 @@ extern void tree_operand_check_failed (int, enum tree_code,
 /* Nonzero means this node will not trap.  In an INDIRECT_REF, means
    accessing the memory pointed to won't generate a trap.  However,
    this only applies to an object when used appropriately: it doesn't
-   mean that writing a READONLY mem won't trap.  */
+   mean that writing a READONLY mem won't trap.
+   
+   In ARRAY_REF and ARRAY_RANGE_REF means that we know that the index
+   (or slice of the array) always belongs to the range of the array.
+   I.e. that the access will not trap, provided that the access to
+   the base to the array will not trap.  */
 #define TREE_THIS_NOTRAP(NODE) ((NODE)->common.nothrow_flag)
 
 /* In a VAR_DECL, PARM_DECL or FIELD_DECL, or any kind of ..._REF node,
@@ -2722,6 +2727,7 @@ extern tree build_method_type (tree, tree);
 extern tree build_offset_type (tree, tree);
 extern tree build_complex_type (tree);
 extern tree array_type_nelts (tree);
+extern bool in_array_bounds_p (tree);
 
 extern tree value_member (tree, tree);
 extern tree purpose_member (tree, tree);
@@ -3256,6 +3262,11 @@ extern tree array_ref_element_size (tree);
 
 extern tree array_ref_low_bound (tree);
 
+/* Return a tree representing the upper bound of the array mentioned in
+   EXP, an ARRAY_REF.  */
+
+extern tree array_ref_up_bound (tree);
+
 /* Return a tree representing the offset, in bytes, of the field referenced
    by EXP.  This does not include any offset in DECL_FIELD_BIT_OFFSET.  */