[tree-ssa] Mainline merge as of 2003-07-06

[tree-ssa] Mainline merge as of 2003-07-06

[Diego Novillo]

The branch is now ready for including g95.  Many thanks to Steven
Bosscher for his help in removing merge conflicts.

The merge brought in quite a number of changes and there are some new
failures that I will be addressing after committing the merge:

	 FAIL: gcc.c-torture/execute/builtins/string-8.c execution
	 FAIL: gcc.c-torture/execute/builtins/string-8.c execution
	 FAIL: gcc.c-torture/execute/builtins/string-8.c execution
	 FAIL: gcc.c-torture/execute/stdio-opt-1.c execution
	 FAIL: gcc.c-torture/execute/stdio-opt-1.c execution
	 FAIL: gcc.c-torture/execute/stdio-opt-1.c execution
	 FAIL: gcc.c-torture/execute/stdio-opt-1.c execution
	 FAIL: gcc.c-torture/execute/string-opt-16.c execution
	 FAIL: gcc.c-torture/execute/string-opt-16.c execution
	 FAIL: gcc.c-torture/execute/string-opt-17.c execution
	 FAIL: gcc.c-torture/execute/string-opt-17.c execution
	 FAIL: gcc.c-torture/execute/string-opt-17.c execution
	 FAIL: gcc.c-torture/execute/string-opt-17.c execution
	 FAIL: gcc.c-torture/execute/string-opt-17.c execution
	 FAIL: gcc.c-torture/execute/string-opt-17.c execution
	 FAIL: gcc.c-torture/execute/string-opt-7.c execution
	 FAIL: gcc.c-torture/execute/string-opt-7.c execution
	 FAIL: gcc.c-torture/execute/string-opt-7.c execution
	 FAIL: gcc.c-torture/execute/string-opt-7.c execution
	 FAIL: gcc.c-torture/execute/string-opt-7.c execution

All these failures are due to the tree-ssa optimizers no expanding
strlen into a builtin.  We leave the function call and cause the tests
to execute the strlen() function included in the tests.



	 FAIL: gcc.dg/compat/vector-1 c_compat_x_tst.o compile
	 FAIL: gcc.dg/compat/vector-1 c_compat_y_tst.o compile
	 FAIL: gcc.dg/compat/vector-2 c_compat_x_tst.o compile
	 FAIL: gcc.dg/compat/vector-2 c_compat_y_tst.o compile

These are failures inherited from mainline.


	 FAIL: gcc.dg/asm-7.c  (test for warnings
	 FAIL: gcc.dg/asm-7.c  (test for warnings
	 FAIL: gcc.dg/asm-7.c  (test for warnings
	 FAIL: gcc.dg/asm-7.c  (test for warnings
	 FAIL: gcc.dg/asm-7.c (test for excess errors)

These are due to a known limitation in the gimplifier.  We don't deal
with invalid lvalues in __asm__ expressions.



	 FAIL: g++.dg/init/array10.C (test for excess errors)

Inherited from mainline.


	 FAIL: g++.old-deja/g++.brendan/crash13.C  (test for errors
	 FAIL: g++.old-deja/g++.brendan/crash13.C (test for excess errors)
	 FAIL: g++.old-deja/g++.bugs/900205_03.C  (test for errors
	 FAIL: g++.old-deja/g++.bugs/900205_03.C  (test for errors
	 FAIL: g++.old-deja/g++.bugs/900205_03.C  (test for errors
	 FAIL: g++.old-deja/g++.bugs/900205_03.C  (test for errors
	 FAIL: g++.old-deja/g++.bugs/900205_03.C (test for excess errors)
	 FAIL: g++.old-deja/g++.law/friend5.C  (test for errors
	 FAIL: g++.old-deja/g++.law/friend5.C  (test for errors
	 FAIL: g++.old-deja/g++.law/friend5.C (test for excess errors)
	 FAIL: g++.old-deja/g++.mike/p646.C  (test for warnings
	 FAIL: g++.old-deja/g++.mike/p646.C  (test for warnings
	 FAIL: g++.old-deja/g++.mike/p646.C  (test for warnings
	 FAIL: g++.old-deja/g++.mike/p646.C  (test for warnings
	 FAIL: g++.old-deja/g++.mike/p646.C  (test for warnings
	 FAIL: g++.old-deja/g++.mike/p646.C (test for excess errors)
	 FAIL: g++.old-deja/g++.robertl/eb132.C (test for excess errors)

Line number mismatch in warnings.  As far as I can tell, the line number
information is messed up in cp/semantics.c:2453

2452      /* Replace AGGR_INIT_EXPRs with appropriate CALL_EXPRs.  */
2453      walk_tree_without_duplicates (&DECL_SAVED_TREE (fn),
2454                                    simplify_aggr_init_exprs_r,
2455                                    NULL);

The line number gets reset from the last line in the function to the
first one.



	 FAIL: libmudflap.c++/fail24-frag.cxx (-O3) output pattern test
	 FAIL: libmudflap.c++/pass28-frag.cxx (-O3) execution test
	 FAIL: libmudflap.c++/pass31-frag.cxx (-O3) execution test

For some reason I still don't understand, mudflap is emitting more error
messages than expected by dejagnu.  The messages didn't look wrong to
me, but I'm not very familiar with this.



	 FAIL: 27_io/basic_ostream/inserters_arithmetic/char/1.cc execution test
	 FAIL: 27_io/basic_ostream/inserters_arithmetic/char/4402.cc execution test
	 FAIL: 27_io/basic_ostream/inserters_arithmetic/wchar_t/1.cc execution test

These expose a bug in the dominator optimizations.  I'm looking at these
today.

Bootstrapped and tested on x86, ia64, alpha and amd64.  My ppc box is
giving me problems that seem unrelated to the merge.  But I believe that
it works on pcc as well, if not please let me know and/or help me fix
it.


Diego.

Re: [tree-ssa] Mainline merge as of 2003-07-06

[Diego Novillo]

On Mon, 2003-07-21 at 10:03, Diego Novillo wrote:
> The branch is now ready for including g95.  Many thanks to Steven
> Bosscher for his help in removing merge conflicts.
> 
Forgot to include the patch to some of the things that needed to be
fixed post-merge.


2003-07-21  Steven Bosscher  <steven@gcc.gnu.org>
            Diego Novillo  <dnovillo@redhat.com>

	* Makefile.in (install-po): Check if $CATALOGS is empty to avoid
	buggy shells.
	* basic-block.h (rtl_verify_flow_info): Remove.
	(tree_verify_flow_info): Remove.
	(verify_flow_info): Declare.
	* builtins.c: Rearrange to simplify merges.  Add #if 0 around
	expand_ functions that are not used in the branch and move new code
	to the end of the file.
	* c-opts.c (c_common_handle_option): Move handling of -fdump- to
	opts.c.
	* cfgloopmanip.c (loop_split_edge_with_NULL): Remove.  Update all
	users.
	* common.opt: Add all the tree-ssa switches.
	* opts.c: Handle them.
	* flags.h (flag_tree_cp): Remove unused variable.
	(enum pta_type): Move from tree-must-alias.h
	(flag_tree_points_to): Likewise.
	* toplev.c (flag_tree_cp): Remove unused variable.
	* tree-cfg.c: Move cfg_hooks structures and functions for
	trees from cfghooks.c.
	* tree-mudflap.c (mudflap_enqueue_decl): Don't use %D to
	avoid warning about format specifiers.

2003-07-21  Diego Novillo  <dnovillo@redhat.com>

	* gimplify.c (gimplify_function_tree): Move gimplification of the
	function body ...
	(gimplify_body): ... here.
	* tree-simple.h (gimplify_body): Declare.
	* tree-inline.c (initialize_inlined_parameters): If the
	emitted assignment is not in GIMPLE form, gimplify the
	body of assignments emitted.

--- Makefile.in	2003/07/16 22:27:12	1.1
+++ Makefile.in	2003/07/21 00:35:31
@@ -4116,18 +4116,20 @@ update-po: $(CATALOGS:.gmo=.pox)
 # with the distribution.
 install-po:
 	$(SHELL) $(srcdir)/mkinstalldirs $(DESTDIR)$(datadir)
-	for cat in $(CATALOGS); do \
-	  lang=`basename $$cat | sed 's/\.gmo$$//'`; \
-	  if [ -f $$cat ]; then :; \
-	  elif [ -f $(srcdir)/$$cat ]; then cat=$(srcdir)/$$cat; \
-	  else continue; \
-	  fi; \
-	  dir=$(localedir)/$$lang/LC_MESSAGES; \
-	  echo $(SHELL) $(srcdir)/mkinstalldirs $(DESTDIR)$$dir; \
-	  $(SHELL) $(srcdir)/mkinstalldirs $(DESTDIR)$$dir || exit 1; \
-	  echo $(INSTALL_DATA) $$cat $(DESTDIR)$$dir/$(PACKAGE).mo; \
-	  $(INSTALL_DATA) $$cat $(DESTDIR)$$dir/$(PACKAGE).mo; \
-	done
+	if [ "${CATALOGS}" != "" ] ; then \
+	    for cat in "$(CATALOGS)"; do \
+		lang=`basename $$cat | sed 's/\.gmo$$//'`; \
+		if [ -f $$cat ]; then :; \
+		elif [ -f $(srcdir)/$$cat ]; then cat=$(srcdir)/$$cat; \
+		else continue; \
+		fi; \
+		dir=$(localedir)/$$lang/LC_MESSAGES; \
+		echo $(SHELL) $(srcdir)/mkinstalldirs $(DESTDIR)$$dir; \
+		$(SHELL) $(srcdir)/mkinstalldirs $(DESTDIR)$$dir || exit 1; \
+		echo $(INSTALL_DATA) $$cat $(DESTDIR)$$dir/$(PACKAGE).mo; \
+		$(INSTALL_DATA) $$cat $(DESTDIR)$$dir/$(PACKAGE).mo; \
+	    done \
+	fi
 
 # Rule for regenerating the message template (gcc.pot).
 # Instead of forcing everyone to edit POTFILES.in, which proved impractical,
--- basic-block.h	2003/07/16 22:27:12	1.1
+++ basic-block.h	2003/07/16 23:57:09
@@ -593,8 +593,7 @@ extern void free_aux_for_edges		PARAMS (
 /* This function is always defined so it can be called from the
    debugger, and it is declared extern so we don't get warnings about
    it being unused.  */
-extern void rtl_verify_flow_info	PARAMS ((void));
-extern void tree_verify_flow_info	PARAMS ((void));
+extern void verify_flow_info		PARAMS ((void));
 
 typedef struct conflict_graph_def *conflict_graph;
 
--- builtins.c	2003/07/16 22:27:12	1.1
+++ builtins.c	2003/07/16 23:17:55
@@ -79,9 +79,9 @@ tree built_in_decls[(int) END_BUILTINS];
    required to implement the function call in all cases).  */
 tree implicit_built_in_decls[(int) END_BUILTINS];
 
-/* FIXME: When merging tree-ssa into mainline, remove all the #if 0 and
-   lines commented with /*@ */.  They represent code that was removed in
-   the branch.  They were only left in to simplify mainline->branch merges.  */
+/* FIXME: When merging tree-ssa into mainline, remove all the #if 0 and lines
+   commented with '@' markers.  They represent code that was removed in the
+   branch.  They were only left in to simplify mainline->branch merges.  */
 
 /* Trigonometric and mathematical constants used in builtin folding.  */
 static bool builtin_dconsts_init = 0;
@@ -125,10 +125,10 @@ static rtx builtin_memcpy_read_str (void
 /*@static rtx expand_builtin_strspn (tree, rtx, enum machine_mode);*/
 /*@static rtx expand_builtin_strcspn (tree, rtx, enum machine_mode);*/
 static rtx expand_builtin_memcpy (tree, rtx, enum machine_mode);
-/*@static rtx expand_builtin_mempcpy (tree, rtx, enum machine_mode, int);*/
+static rtx expand_builtin_mempcpy (tree, rtx, enum machine_mode, int);
 static rtx expand_builtin_memmove (tree, rtx, enum machine_mode);
 static rtx expand_builtin_bcopy (tree);
-static rtx expand_builtin_strcpy (tree, rtx, enum machine_mode);
+/*@static rtx expand_builtin_strcpy (tree, rtx, enum machine_mode);*/
 static rtx expand_builtin_stpcpy (tree, rtx, enum machine_mode);
 static rtx builtin_strncpy_read_str (void *, HOST_WIDE_INT, enum machine_mode);
 static rtx expand_builtin_strncpy (tree, rtx, enum machine_mode);
@@ -2629,7 +2629,6 @@ expand_builtin_memcpy (tree arglist, rtx
     }
 }
 
-#if 0
 /* Expand a call to the mempcpy builtin, with arguments in ARGLIST.
    Return 0 if we failed the caller should emit a normal call,
    otherwise try to get the result in TARGET, if convenient (and in
@@ -2731,7 +2730,6 @@ expand_builtin_mempcpy (tree arglist, rt
       return 0;
     }
 }
-#endif
 
 /* Expand expression EXP, which is a call to the memmove builtin.  Return 0
    if we failed the caller should emit a normal call.  */
@@ -2813,6 +2811,7 @@ expand_builtin_bcopy (tree arglist)
   return expand_builtin_memmove (newarglist, const0_rtx, VOIDmode);
 }
 
+#if 0
 /* Expand expression EXP, which is a call to the strcpy builtin.  Return 0
    if we failed the caller should emit a normal call, otherwise try to get
    the result in TARGET, if convenient (and in mode MODE if that's
@@ -2843,6 +2842,7 @@ expand_builtin_strcpy (tree arglist, rtx
   return expand_expr (build_function_call_expr (fn, arglist),
 		      target, mode, EXPAND_NORMAL);
 }
+#endif
 
 /* Expand a call to the stpcpy builtin, with arguments in ARGLIST.
    Return 0 if we failed the caller should emit a normal call,
@@ -5027,11 +5027,13 @@ expand_builtin (tree exp, rtx target, rt
 	return target;
       break;
 
+#if 0
     case BUILT_IN_STRCPY:
       target = expand_builtin_strcpy (arglist, target, mode);
       if (target)
 	return target;
       break;
+#endif
 
     case BUILT_IN_STRNCPY:
       target = expand_builtin_strncpy (arglist, target, mode);
@@ -5755,7 +5757,7 @@ fold_builtin (tree exp)
     case BUILT_IN_STRLEN:
       if (validate_arglist (arglist, POINTER_TYPE, VOID_TYPE))
 	{
-	  tree len = c_strlen (TREE_VALUE (arglist));
+	  tree len = c_strlen (TREE_VALUE (arglist), 0);
 	  if (len)
 	    {
 	      /* Convert from the internal "sizetype" type to "size_t".  */
@@ -6682,7 +6684,7 @@ simplify_builtin_strcpy (arglist)
   if (!fn)
     return 0;
 
-  len = c_strlen (TREE_VALUE (TREE_CHAIN (arglist)));
+  len = c_strlen (TREE_VALUE (TREE_CHAIN (arglist)), 1);
   if (len == 0)
     return 0;
 
@@ -6717,7 +6719,7 @@ simplify_builtin_strncpy (arglist)
     return 0;
   else
     {
-      tree slen = c_strlen (TREE_VALUE (TREE_CHAIN (arglist)));
+      tree slen = c_strlen (TREE_VALUE (TREE_CHAIN (arglist)), 0);
       tree len = TREE_VALUE (TREE_CHAIN (TREE_CHAIN (arglist)));
       tree fn;
 
@@ -6890,8 +6892,8 @@ simplify_builtin_strcmp (arglist)
       return fold (build (MINUS_EXPR, integer_type_node, ind1, ind2));
     }
 
-  len = c_strlen (arg1);
-  len2 = c_strlen (arg2);
+  len = c_strlen (arg1, 0);
+  len2 = c_strlen (arg2, 0);
 
   if (len)
     len = size_binop (PLUS_EXPR, ssize_int (1), len);
@@ -7016,16 +7018,16 @@ simplify_builtin_strncmp (arglist)
   /* Perhaps one of the strings is really constant, if so prefer
      that constant length over the other string's length.  */
   if (p1)
-    len = c_strlen (arg1);
+    len = c_strlen (arg1, 0);
   else if (p2)
-    len = c_strlen (arg2);
+    len = c_strlen (arg2, 0);
 
   /* If we still don't have a len, try either string arg as long
      as they don't have side effects.  */
   if (!len && !TREE_SIDE_EFFECTS (arg1))
-    len = c_strlen (arg1);
+    len = c_strlen (arg1, 0);
   if (!len && !TREE_SIDE_EFFECTS (arg2))
-    len = c_strlen (arg2);
+    len = c_strlen (arg2, 0);
   /* If we still don't have a length, punt.  */
   if (!len)
     return 0;
@@ -7289,7 +7291,7 @@ simplify_builtin_fputs (arglist, ignore,
 
   /* Get the length of the string passed to fputs.  If the length
      can't be determined, punt.  */
-  if (!(len = c_strlen (TREE_VALUE (arglist)))
+  if (!(len = c_strlen (TREE_VALUE (arglist), 0))
       || TREE_CODE (len) != INTEGER_CST)
     return 0;
 
--- c-opts.c	2003/07/16 22:27:12	1.1
+++ c-opts.c	2003/07/17 20:44:50
@@ -139,7 +139,6 @@ missing_arg (enum opt_code code)
     case OPT_d:
     case OPT_fabi_version_:
     case OPT_fbuiltin_:
-    case OPT_fdump_:
     case OPT_fname_mangling_version_:
     case OPT_ftabstop_:
     case OPT_fexec_charset_:
@@ -738,11 +737,6 @@ c_common_handle_option (size_t scode, co
       cpp_opts->dollars_in_ident = value;
       break;
 
-    case OPT_fdump_:
-      if (!dump_switch_p (arg))
-	result = 0;
-      break;
-
     case OPT_ffreestanding:
       value = !value;
       /* Fall through...  */
--- cfgloopmanip.c	2003/07/16 22:27:12	1.1
+++ cfgloopmanip.c	2003/07/19 00:22:04
@@ -29,8 +29,6 @@ Software Foundation, 59 Temple Place - S
 #include "cfglayout.h"
 #include "output.h"
 
-static basic_block loop_split_edge_with_NULL PARAMS ((edge, struct loops *));
-
 static struct loop * duplicate_loop (struct loops *, struct loop *,
 				     struct loop *);
 static void duplicate_subloops (struct loops *, struct loop *, struct loop *);
@@ -104,7 +102,7 @@ remove_bbs (dominance_info dom, basic_bl
     {
       remove_bb_from_loops (bbs[i]);
       delete_from_dominance_info (dom, bbs[i]);
-      delete_block (bbs[i]);
+      delete_basic_block (bbs[i]);
     }
 }
 
@@ -351,7 +349,7 @@ remove_path (struct loops *loops, edge e
      fix -- when e->dest has exactly one predecessor, this corresponds
      to blocks dominated by e->dest, if not, split the edge.  */
   if (e->dest->pred->pred_next)
-    e = loop_split_edge_with_NULL (e, loops)->pred;
+    e = loop_split_edge_with (e, NULL_RTX, loops)->pred;
 
   /* It may happen that by removing path we remove one or more loops
      we belong to.  In this case first unloop the loops, then proceed
@@ -1200,7 +1198,7 @@ force_single_succ_latches (struct loops 
       for (e = loop->header->pred; e->src != loop->latch; e = e->pred_next)
 	continue;
 
-      loop_split_edge_with_NULL (e, loops);
+      loop_split_edge_with (e, NULL_RTX, loops);
     }
   loops->state |= LOOPS_HAVE_SIMPLE_LATCHES;
 }
@@ -1247,61 +1245,3 @@ loop_split_edge_with (edge e, rtx insns,
 
   return new_bb;
 }
-
-/* Same function as "loop_split_edge_with (e, NULL, loops)", 
-   independent of the tree/rtl representation.  */
-
-static basic_block
-loop_split_edge_with_NULL (e, loops)
-     edge e;
-     struct loops *loops;
-{
-  basic_block src, dest, new_bb;
-  struct loop *loop_c;
-  edge new_e;
-  
-  src = e->src;
-  dest = e->dest;
-
-  loop_c = find_common_loop (src->loop_father, dest->loop_father);
-
-  /* Create basic block for it.  */
-
-  if (cfg_level == AT_RTL_LEVEL)
-    new_bb = create_basic_block (NULL_RTX, NULL_RTX, EXIT_BLOCK_PTR->prev_bb);
-  else
-    new_bb = create_bb ();
-
-  add_to_dominance_info (loops->cfg.dom, new_bb);
-  add_bb_to_loop (new_bb, loop_c);
-  new_bb->flags = 0;
-
-  new_e = make_edge (new_bb, dest, (cfg_level == AT_RTL_LEVEL) ? EDGE_FALLTHRU : 0);
-  new_e->probability = REG_BR_PROB_BASE;
-  new_e->count = e->count;
-  if (e->flags & EDGE_IRREDUCIBLE_LOOP)
-    {
-      new_bb->flags |= BB_IRREDUCIBLE_LOOP;
-      new_e->flags |= EDGE_IRREDUCIBLE_LOOP;
-    }
-
-  new_bb->count = e->count;
-  new_bb->frequency = EDGE_FREQUENCY (e);
-
-  if (cfg_level == AT_RTL_LEVEL)
-    cfg_layout_redirect_edge (e, new_bb);
-  else
-    redirect_edge_succ (e, new_bb);
-  
-  alloc_aux_for_block (new_bb, sizeof (struct reorder_block_def));
-
-  set_immediate_dominator (loops->cfg.dom, new_bb, src);
-  set_immediate_dominator (loops->cfg.dom, dest,
-    recount_dominator (loops->cfg.dom, dest));
-
-  if (dest->loop_father->latch == src)
-    dest->loop_father->latch = new_bb;
-  
-  return new_bb;
-}
-
--- common.opt	2003/07/17 20:26:10	1.1
+++ common.opt	2003/07/20 14:26:10
@@ -235,6 +235,15 @@ Common
 fdiagnostics-show-location=
 Common Joined RejectNegative
 
+fdisable-gimple
+Common RejectNegative
+
+fdisable-tree-ssa
+Common RejectNegative
+
+fdump-
+Common Joined RejectNegative
+
 fdump-unnumbered
 Common
 
@@ -349,6 +358,9 @@ Common RejectNegative Joined UInteger
 fmove-all-movables
 Common
 
+fmudflap
+Common RejectNegative
+
 fnew-ra
 Common
 
@@ -517,6 +529,27 @@ Common
 ftrapv
 Common
 
+ftree-ccp
+Common
+
+ftree-copyprop
+Common
+
+ftree-dce
+Common
+
+ftree-dominator-opts
+Common
+
+ftree-must-alias
+Common
+
+ftree-points-to=
+Common Joined RejectNegative
+
+ftree-pre
+Common
+
 funit-at-a-time
 Common
 
--- flags.h	2003/07/16 22:27:12	1.1
+++ flags.h	2003/07/17 02:12:46
@@ -692,9 +692,6 @@ extern int flag_disable_tree_ssa;
 /* Enable SSA-PRE on trees.  */
 extern int flag_tree_pre;
 
-/* Enable SSA-CP on trees. */
-extern int flag_tree_cp;
-
 /* Enable SSA-CCP on trees.  */
 extern int flag_tree_ccp;
 
@@ -710,6 +707,14 @@ extern int flag_tree_dom;
 /* Enable promotion of virtual to real operands in must-alias situations.  */
 extern int flag_tree_must_alias;
 
+/* Enable points-to analysis on trees.  */
+enum pta_type
+  {
+    PTA_NONE,
+    PTA_ANDERSEN
+  };
+extern enum pta_type flag_tree_points_to;
+
 /* Nonzero means put zero initialized data in the bss section.  */
 extern int flag_zero_initialized_in_bss;
 
--- gimplify.c	2003/07/16 22:27:12	1.1
+++ gimplify.c	2003/07/17 22:55:16
@@ -215,12 +215,9 @@ keep_function_tree_in_gimple_form (tree 
 int
 gimplify_function_tree (tree fndecl)
 {
-  tree fnbody;
   int done;
+  tree fnbody;
   tree oldfn;
-  tree tmp;
-  const char *saved_input_filename = input_filename;
-  int saved_lineno = input_line;
 
   /* Don't bother doing anything if the program has errors.  */
   if (errorcount || sorrycount)
@@ -235,28 +232,46 @@ gimplify_function_tree (tree fndecl)
   if (fnbody == NULL_TREE)
     return 0;
 
-  timevar_push (TV_TREE_GIMPLIFY);
-
   oldfn = current_function_decl;
   current_function_decl = fndecl;
 
+  done = gimplify_body (&fnbody, fndecl);
+
+  DECL_SAVED_TREE (fndecl) = fnbody;
+  current_function_decl = oldfn;
+
+  return done;
+}
+
+
+/* Gimplify the body of statements pointed by BODY_P.  FNDECL is the
+   function decl containing BODY.  */
+
+int
+gimplify_body (tree *body_p, tree fndecl)
+{
+  int done;
+  const char *saved_input_filename = input_filename;
+  int saved_lineno = input_line;
+
+  timevar_push (TV_TREE_GIMPLIFY);
+
   push_gimplify_context ();
 
   /* Unshare most shared trees in the body.  */
-  unshare_all_trees (fnbody);
+  unshare_all_trees (*body_p);
 
   /* Gimplify the function's body.  */
-  done = gimplify_stmt (&fnbody);
+  done = gimplify_stmt (body_p);
 
   /* Unshare again, in case gimplification was sloppy.  */
-  unshare_all_trees (fnbody);
+  unshare_all_trees (*body_p);
 
   /* If there isn't an outer BIND_EXPR, add one.  */
-  tmp = fnbody;
-  if (TREE_CODE (tmp) != BIND_EXPR)
+  if (TREE_CODE (*body_p) != BIND_EXPR)
     {
-      fnbody = build (BIND_EXPR, void_type_node, NULL_TREE, fnbody, NULL_TREE);
-      TREE_SIDE_EFFECTS (fnbody) = 1;
+      *body_p = build (BIND_EXPR, void_type_node, NULL_TREE, *body_p, NULL_TREE);
+      TREE_SIDE_EFFECTS (*body_p) = 1;
       input_line = get_lineno (fndecl);
       input_filename = get_filename (fndecl);
     }
@@ -266,20 +281,17 @@ gimplify_function_tree (tree fndecl)
       input_filename = saved_input_filename;
     }
 
-  DECL_SAVED_TREE (fndecl) = fnbody;
-
   /* Declare the new temporary variables.  */
-  declare_tmp_vars (gimplify_ctxp->temps, fnbody);
+  declare_tmp_vars (gimplify_ctxp->temps, *body_p);
 
   pop_gimplify_context ();
 
-  current_function_decl = oldfn;
-
   timevar_pop (TV_TREE_GIMPLIFY);
 
   return done;
 }
 
+
 /* Gimplification of expression trees.  */
 
 /* Gimplify an expression which appears at statement context; usually, this
--- opts.c	2003/07/16 22:27:12	1.1
+++ opts.c	2003/07/20 14:24:13
@@ -488,6 +488,10 @@ decode_options (unsigned int argc, const
       flag_crossjumping = 1;
       flag_if_conversion = 1;
       flag_if_conversion2 = 1;
+      flag_tree_ccp = 1;
+      flag_tree_dce = 1;
+      flag_tree_copyprop = 1;
+      flag_tree_dom = 1;
     }
 
   if (optimize >= 2)
@@ -512,6 +516,11 @@ decode_options (unsigned int argc, const
       flag_delete_null_pointer_checks = 1;
       flag_reorder_blocks = 1;
       flag_reorder_functions = 1;
+      /* FIXME: Apparently, dominator optimizations get in the way of
+	 SSA-PRE.  For now, if -ftree-pre is given, disable dominator
+	 optimizations at -O2 and higher.  */
+      if (flag_tree_pre)
+	flag_tree_dom = 0;
     }
 
   if (optimize >= 3)
@@ -901,6 +910,19 @@ common_handle_option (size_t scode, cons
 	return 0;
       break;
 
+    case OPT_fdisable_gimple:
+      flag_disable_gimple = value;
+      break;
+
+    case OPT_fdisable_tree_ssa:
+      flag_disable_tree_ssa = value;
+      break;
+
+    case OPT_fdump_:
+      if (!dump_switch_p (arg))
+	return 0;
+      break;
+
     case OPT_fdump_unnumbered:
       flag_dump_unnumbered = value;
       break;
@@ -1060,6 +1082,10 @@ common_handle_option (size_t scode, cons
       flag_move_all_movables = value;
       break;
 
+    case OPT_fmudflap:
+      flag_mudflap = value;
+      break;
+
     case OPT_fnew_ra:
       flag_new_regalloc = value;
       break;
@@ -1309,6 +1335,44 @@ common_handle_option (size_t scode, cons
       flag_trapv = value;
       break;
 
+    case OPT_ftree_ccp:
+      flag_tree_ccp = value;
+      break;
+
+    case OPT_ftree_copyprop:
+      flag_tree_copyprop = value;
+      break;
+
+    case OPT_ftree_dce:
+      flag_tree_dce = value;
+      break;
+
+    case OPT_ftree_dominator_opts:
+      flag_tree_dom = value;
+      break;
+
+    case OPT_ftree_must_alias:
+      flag_tree_must_alias = value;
+      break;
+
+    case OPT_ftree_points_to_:
+      if (!strcmp (arg, "andersen"))
+#ifdef HAVE_BANSHEE
+        flag_tree_points_to = PTA_ANDERSEN;
+#else
+        warning ("Andersen's PTA not available - libbanshee not compiled.");
+#endif
+      else
+	{
+	  warning ("`%s`: unknown points-to analysis algorithm", arg);
+	  return 0;
+	}
+      break;
+
+    case OPT_ftree_pre:
+      flag_tree_pre = value;
+      break;
+
     case OPT_funit_at_a_time:
       flag_unit_at_a_time = value;
       break;
--- toplev.c	2003/07/16 22:27:12	1.1
+++ toplev.c	2003/07/17 01:54:49
@@ -964,9 +964,6 @@ enum pta_type flag_tree_points_to = PTA_
 /* Enable SSA-CCP on trees.  */
 int flag_tree_ccp = 0;
 
-/* Enable SSA-CP on trees.  */
-int flag_tree_cp = 0;
-
 /* Enable SSA-Copy propagation on trees.  */
 int flag_tree_copyprop;
 
--- tree-cfg.c	2003/07/16 22:27:12	1.1
+++ tree-cfg.c	2003/07/18 01:09:12
@@ -107,6 +107,8 @@ static inline bool stmt_starts_bb_p (tre
 static inline bool stmt_ends_bb_p (tree);
 static void find_contained_blocks (tree *, bitmap, tree **);
 static void compute_reachable_eh (tree);
+static int tree_verify_flow_info (void);
+static basic_block tree_make_forwarder_block (basic_block, int, int, edge, int);
 
 /* Flowgraph optimization and cleanup.  */
 static void remove_unreachable_blocks (void);
@@ -190,6 +192,21 @@ static tree_stmt_iterator find_insert_lo
   at line 6, it sets NEXT_BLOCK_LINK (s3) to 's4'.  */
 #define NEXT_BLOCK_LINK(STMT)	TREE_CHAIN (STMT)
 
+/* FIXME These need to be filled in with appropriate pointers.  But this
+   implies an ABI change in some functions.  */
+struct cfg_hooks tree_cfg_hooks = {
+  tree_verify_flow_info,
+  NULL,				/* dump_bb  */
+  NULL,				/* create_basic_block  */
+  NULL,				/* redirect_edge_and_branch  */
+  NULL,				/* redirect_edge_and_branch_force  */
+  NULL,				/* delete_basic_block  */
+  NULL,				/* split_block  */
+  NULL,				/* can_merge_blocks_p  */
+  NULL,				/* merge_blocks  */
+  tree_split_edge,		/* cfgh_split_edge  */
+  tree_make_forwarder_block	/* cfgh_make_forward_block  */
+};
 
 /*---------------------------------------------------------------------------
 			      Create basic blocks
@@ -4752,9 +4776,77 @@ tree_split_edge (edge edge_in)
 }
 
 
-/* Verifies that the flow information is ok.  */
+/* Verifies that the flow information is OK.  */
 
-void 
+static int 
 tree_verify_flow_info (void)
 {
+  return 0;
+}
+
+
+/* Split BB into entry part and rest; if REDIRECT_LATCH, redirect edges
+   marked as latch into entry part, analogically for REDIRECT_NONLATCH.
+   In both of these cases, ignore edge EXCEPT.  If CONN_LATCH, set edge
+   between created entry part and BB as latch one.  Return created entry
+   part.  */
+
+static basic_block
+tree_make_forwarder_block (basic_block bb, int redirect_latch,
+                           int redirect_nonlatch, edge except, int conn_latch)
+{
+  edge e, next_e, fallthru;
+  basic_block dummy;
+
+  /* Create the new basic block.  */
+  dummy = create_bb (); 
+  alloc_aux_for_block (dummy, sizeof (struct bb_ann_d));
+  dummy->count = bb->count;
+  dummy->frequency = bb->frequency;
+  dummy->loop_depth = bb->loop_depth;
+  dummy->head_tree_p = NULL;
+  dummy->end_tree_p = NULL;
+
+  /* Redirect the incoming edges.  */
+  dummy->pred = bb->pred;
+  bb->pred = NULL;
+  for (e = dummy->pred; e; e = e->pred_next)
+    e->dest = dummy;
+  
+  fallthru = make_edge (dummy, bb, 0);
+  
+  HEADER_BLOCK (dummy) = 0;
+  HEADER_BLOCK (bb) = 1;
+  
+  /* Redirect back edges we want to keep.  */
+  for (e = dummy->pred; e; e = next_e)
+    {
+      next_e = e->pred_next;
+      if (e == except
+	  || !((redirect_latch && LATCH_EDGE (e))
+	       || (redirect_nonlatch && !LATCH_EDGE (e))))
+	{
+	  dummy->frequency -= EDGE_FREQUENCY (e);
+	  dummy->count -= e->count;
+	  if (dummy->frequency < 0)
+	    dummy->frequency = 0;
+	  if (dummy->count < 0)
+	    dummy->count = 0;
+	  redirect_edge_succ (e, bb);
+	}
+    }
+  
+  alloc_aux_for_edge (fallthru, sizeof (int));
+  LATCH_EDGE (fallthru) = conn_latch;
+
+  return dummy;
+}
+
+/* Initialization of functions specific to the tree IR.  */
+
+void 
+tree_register_cfg_hooks ()
+{
+  cfg_level = AT_TREE_LEVEL;
+  cfg_hooks = &tree_cfg_hooks;
 }
--- tree-inline.c	2003/07/16 22:27:12	1.1
+++ tree-inline.c	2003/07/17 23:21:30
@@ -537,6 +537,7 @@ initialize_inlined_parameters (id, args,
   tree a;
   tree p;
   tree vars = NULL_TREE;
+  bool gimplify_init_stmts_p = false;
 
   /* Figure out what the parameters are.  */
   parms = DECL_ARGUMENTS (fn);
@@ -634,12 +635,19 @@ initialize_inlined_parameters (id, args,
 	 the argument to the proper type in case it was promoted.  */
       if (value)
 	{
+	  tree rhs = convert (TREE_TYPE (var), value);
+
 	  /* We want to use MODIFY_EXPR, not INIT_EXPR here so that we
 	     keep our trees in gimple form.  */
-	  init_stmt = build (MODIFY_EXPR, TREE_TYPE (var), var,
-			     convert (TREE_TYPE (var), value));
+	  init_stmt = build (MODIFY_EXPR, TREE_TYPE (var), var, rhs);
 	  init_stmts = add_stmt_to_compound (init_stmts, void_type_node,
 					     init_stmt);
+
+	  /* If the conversion needed to assign VALUE to VAR is not a
+	     GIMPLE expression, flag that we will need to gimplify
+	     INIT_STMTS at the end.  */
+	  if (!is_gimple_rhs (rhs))
+	    gimplify_init_stmts_p = true;
 	}
 
 #if 0
@@ -671,6 +679,10 @@ initialize_inlined_parameters (id, args,
 					 value);
     }
 
+  if (gimplify_init_stmts_p
+      && keep_function_tree_in_gimple_form (fn))
+    gimplify_body (&init_stmts, fn);
+
   add_var_to_bind_expr (bind_expr, vars);
   return init_stmts;
 }
--- tree-mudflap.c	2003/07/16 22:27:12	1.1
+++ tree-mudflap.c	2003/07/18 01:09:13
@@ -1303,7 +1304,8 @@ mudflap_enqueue_decl (tree obj, const ch
 	  found_p = 1;
 
       if (found_p)
-	warning ("mudflap cannot track lifetime of `%D'", obj);
+	warning ("mudflap cannot track lifetime of `%s'",
+		 IDENTIFIER_POINTER (DECL_NAME (obj)));
       else
 	{
 	  VARRAY_PUSH_TREE (deferred_static_decls, obj);
--- tree-simple.h	2003/07/16 22:27:12	1.1
+++ tree-simple.h	2003/07/17 22:51:26
@@ -88,6 +88,7 @@ typedef enum fallback_t {
 } fallback_t;
 int gimplify_expr (tree *, tree *, tree *, int (*) (tree), fallback_t);
 int gimplify_stmt (tree *);
+int gimplify_body (tree *, tree);
 
 /* Miscellaneous helpers.  */
 tree get_base_symbol (tree);

Re: [tree-ssa] Mainline merge as of 2003-07-06

[Toon Moene]

Diego Novillo wrote:

> Bootstrapped and tested on x86, ia64, alpha and amd64.  My ppc box is
> giving me problems that seem unrelated to the merge.  But I believe that
> it works on pcc as well, if not please let me know and/or help me fix
> it.

I'll bootstrap tonight on my Titanium PowerBook 
(powerpc-unknown-linux-gnu; Debian testing as of last Sunday).

Cheers and thanks for all this hard work !

-- 
Toon Moene - mailto:toon@moene.indiv.nluug.nl - phoneto: +31 346 214290
Saturnushof 14, 3738 XG  Maartensdijk, The Netherlands
Maintainer, GNU Fortran 77: http://gcc.gnu.org/onlinedocs/g77_news.html
GNU Fortran 95: http://gcc-g95.sourceforge.net/ (under construction)

Re: [tree-ssa] Mainline merge as of 2003-07-06

[Toon Moene]

Toon Moene wrote:

> Diego Novillo wrote:
> 
>> Bootstrapped and tested on x86, ia64, alpha and amd64.  My ppc box is
>> giving me problems that seem unrelated to the merge.  But I believe that
>> it works on pcc as well, if not please let me know and/or help me fix
>> it.
> 
> I'll bootstrap tonight on my Titanium PowerBook 
> (powerpc-unknown-linux-gnu; Debian testing as of last Sunday).

I get the following error on linking libstdc++-v3:

.libs/strstream.o: could not read symbols: Bad value

whereas compilation of strstream.cc seems to have completed normally ...

-- 
Toon Moene - mailto:toon@moene.indiv.nluug.nl - phoneto: +31 346 214290
Saturnushof 14, 3738 XG  Maartensdijk, The Netherlands
Maintainer, GNU Fortran 77: http://gcc.gnu.org/onlinedocs/g77_news.html
GNU Fortran 95: http://gcc-g95.sourceforge.net/ (under construction)

Re: [tree-ssa] Mainline merge as of 2003-07-06

[Diego Novillo]

On Tue, 2003-07-22 at 04:17, Toon Moene wrote:

> I get the following error on linking libstdc++-v3:
> 
> .libs/strstream.o: could not read symbols: Bad value
> 
Thanks Toon.  It's definitely a problem with mainline as of the merge
date.  I did an incremental merge last night and I'm now testing PPC and
PPC64.  If they both work, I will update the branch with the new merge.

> whereas compilation of strstream.cc seems to have completed normally ...
>
This is what I'm seeing here as well.


Thanks.  Diego.

Re: [tree-ssa] Mainline merge as of 2003-07-06

[Diego Novillo]

On Tue, 2003-07-22 at 04:17, Toon Moene wrote:

> I get the following error on linking libstdc++-v3:
> 
> .libs/strstream.o: could not read symbols: Bad value
> 
> whereas compilation of strstream.cc seems to have completed normally ...
>
Toon,

PPC should be working again.  Could you give it another try?


Thanks.  Diego.

Re: [tree-ssa] Mainline merge as of 2003-07-06

[Toon Moene]

Diego Novillo wrote:
> On Tue, 2003-07-22 at 04:17, Toon Moene wrote:
> 
> 
>>I get the following error on linking libstdc++-v3:
>>
>>.libs/strstream.o: could not read symbols: Bad value
>>
>>whereas compilation of strstream.cc seems to have completed normally ...
>>
> 
> Toon,
> 
> PPC should be working again.  Could you give it another try?

I will, tonight.  In the mean time, I got this for c, f77, f95:

http://gcc.gnu.org/ml/gcc-testresults/2003-07/msg01415.html

Cheers,

-- 
Toon Moene - mailto:toon@moene.indiv.nluug.nl - phoneto: +31 346 214290
Saturnushof 14, 3738 XG  Maartensdijk, The Netherlands
Maintainer, GNU Fortran 77: http://gcc.gnu.org/onlinedocs/g77_news.html
GNU Fortran 95: http://gcc-g95.sourceforge.net/ (under construction)

Re: [tree-ssa] Mainline merge as of 2003-07-06

[Toon Moene]

Toon Moene wrote:

> Diego Novillo wrote:

>> PPC should be working again.  Could you give it another try?
> 
> I will, tonight.

This try runs into a comparison error:

Bootstrap comparison failure!
gcc.o differs

(CVS'd at 17:40 UTC on the 23rd).

-- 
Toon Moene - mailto:toon@moene.indiv.nluug.nl - phoneto: +31 346 214290
Saturnushof 14, 3738 XG  Maartensdijk, The Netherlands
Maintainer, GNU Fortran 77: http://gcc.gnu.org/onlinedocs/g77_news.html
GNU Fortran 95: http://gcc-g95.sourceforge.net/ (under construction)

Re: [tree-ssa] Mainline merge as of 2003-07-06

[Toon Moene]

I wrote:

>> Diego Novillo wrote:
> 
> 
>>> PPC should be working again.  Could you give it another try?

> This try runs into a comparison error:
> 
> Bootstrap comparison failure!
> gcc.o differs
> 
> (CVS'd at 17:40 UTC on the 23rd).

Probably an operator error (I forgot to reapply the g95 patch to the 
tree-ssa checkout).

Will try again tonight.

-- 
Toon Moene - mailto:toon@moene.indiv.nluug.nl - phoneto: +31 346 214290
Saturnushof 14, 3738 XG  Maartensdijk, The Netherlands
Maintainer, GNU Fortran 77: http://gcc.gnu.org/onlinedocs/g77_news.html
GNU Fortran 95: http://gcc-g95.sourceforge.net/ (under construction)

Re: [tree-ssa] Mainline merge as of 2003-07-06

[Toon Moene]

I wrote:

> Probably an operator error (I forgot to reapply the g95 patch to the 
> tree-ssa checkout).
> 
> Will try again tonight.

No luck.  I CVS'd the tree-ssa branch at 13:00 UTC on the 24th, applied 
a brand new g95 patch: 24th of July, 8:58 UTC, but I still run into the 
comparison difference:

Bootstrap comparison failure!
gcc.o differs

This shouldn't hold up inclusion of g95 in the branch, though ...

-- 
Toon Moene - mailto:toon@moene.indiv.nluug.nl - phoneto: +31 346 214290
Saturnushof 14, 3738 XG  Maartensdijk, The Netherlands
Maintainer, GNU Fortran 77: http://gcc.gnu.org/onlinedocs/g77_news.html
GNU Fortran 95: http://gcc-g95.sourceforge.net/ (under construction)

Re: [tree-ssa] Mainline merge as of 2003-07-06

[Diego Novillo]

On Fri, 2003-07-25 at 05:13, Toon Moene wrote:

> Bootstrap comparison failure!
> gcc.o differs
> 
How odd, this is the same problem I had a few weeks ago on my ppc box. 
But I've been bootstrapping on ppc and ppc64 without any problems since
the latest merge.  I wonder if this is the same -fno-strict-aliasing
problems that Andrew Pinski has reported.  Unfortunately I have no
access to non-linux PPC.

Perhaps one way for me to debug this would be for you to run the
testsuite with the stage1 compiler and send me the .i files for any test
that fails for you but not for me (my ppc testers send their testresults
every night).
 

Diego.