[Bug bootstrap/41626] New: GCC does not bootstrap on powerpc64

[Bug bootstrap/41626] New: GCC does not bootstrap on powerpc64

[meissner at gcc dot gnu dot org]

GCC does not bootstrap on powerpc64.  It fails in the 2nd stage, when it links
build/gensupport.o.  On the powerpc, the statements:

static struct obstack obstack;
struct obstack *rtl_obstack = &obstack;

On the powerpc generate a reference to an unknown external 'obstack', while on
the x86_64, it generates 'obstack' as a bss item.  The option
-fkeep-static-consts does not help the issue.


-- 
           Summary: GCC does not bootstrap on powerpc64
           Product: gcc
           Version: 4.5.0
            Status: UNCONFIRMED
          Severity: blocker
          Priority: P3
         Component: bootstrap
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: meissner at gcc dot gnu dot org
 GCC build triplet: powerpc64-unknown-linux-gnu
  GCC host triplet: powerpc64-unknown-linux-gnu
GCC target triplet: powerpc64-unknown-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41626

[Bug bootstrap/41626] GCC does not bootstrap on powerpc64

[meissner at gcc dot gnu dot org]

------- Comment #1 from meissner at gcc dot gnu dot org  2009-10-07 21:14 -------
Created an attachment (id=18746)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=18746&action=view)
Simplified test case that fails on the powerpc.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41626

[Bug bootstrap/41626] [4.5 Regression] GCC does not bootstrap on powerpc64

[rguenth at gcc dot gnu dot org]

-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |hubicka at gcc dot gnu dot
                   |                            |org
            Summary|GCC does not bootstrap on   |[4.5 Regression] GCC does
                   |powerpc64                   |not bootstrap on powerpc64
   Target Milestone|---                         |4.5.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41626

[Bug bootstrap/41626] [4.5 Regression] GCC does not bootstrap on powerpc64

[meissner at gcc dot gnu dot org]

------- Comment #2 from meissner at gcc dot gnu dot org  2009-10-07 22:00 -------
Subversion revision 152520 is the revision that breaks this.

The change in question is:
009-10-07  Jan Hubicka  <jh@suse.cz>

        * lto-symtab.c (lto_cgraph_replace_node): Assert that inline clones has
        no address taken.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41626

[Bug middle-end/41626] [4.5 Regression] GCC does not bootstrap on powerpc64

[pinskia at gcc dot gnu dot org]

------- Comment #3 from pinskia at gcc dot gnu dot org  2009-10-07 22:25 -------
This testcase works at -O0 but fails at -O1.


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2009-10-07 22:25:08
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41626

[Bug middle-end/41626] [4.5 Regression] GCC does not bootstrap on powerpc64

[pinskia at gcc dot gnu dot org]

------- Comment #4 from pinskia at gcc dot gnu dot org  2009-10-07 22:34 -------
Reverting just the varpool.c file back to what it was, allows this testcase to
work.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41626

[Bug middle-end/41626] [4.5 Regression] GCC does not bootstrap on powerpc64

[pinskia at gcc dot gnu dot org]

------- Comment #5 from pinskia at gcc dot gnu dot org  2009-10-07 22:37 -------
This part is what is causing it:
!       /* When reading back varpool at LTO time, we re-construct the queue in
order
!          to have "needed" list right by inserting all needed nodes into
varpool.
!        We however don't want to re-analyze already analyzed nodes.  */
!       if (!analyzed)
!       {
!         gcc_assert (!in_lto_p);
!           /* Compute the alignment early so function body expanders are
!            already informed about increased alignment.  */
!           align_variable (decl, 0);

!           if (DECL_INITIAL (decl))
!           record_references_in_initializer (decl);
!       }

Moving the record_references_in_initializer outside of the !analyzed condition
allows this testcase to work.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41626

[Bug middle-end/41626] [4.5 Regression] GCC does not bootstrap on powerpc64

[danglin at gcc dot gnu dot org]

------- Comment #6 from danglin at gcc dot gnu dot org  2009-10-08 13:10 -------
Bootstrap also fails on armv5tejl-unknown-linux-gnueabi:

build/gensupport.o:(.data+0x20): undefined reference to `obstack'


-- 

danglin at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |danglin at gcc dot gnu dot
                   |                            |org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41626

[Bug middle-end/41626] [4.5 Regression] GCC does not bootstrap on powerpc64

[hubicka at ucw dot cz]

------- Comment #7 from hubicka at ucw dot cz  2009-10-08 14:18 -------
Subject: Re:  [4.5 Regression] GCC does not bootstrap on powerpc64

Hi,
the following patch should fix the problem.  I am
bootstrapping/regtesting it at x86_64-linux, but it would be nice to
know if it actually fixes the bootstrap. It does work on reduced
testcase for me.

Honza

        * cgraphbuild.c (record_reference): When parameter DATA is NULL,
        do not mark cgraph nodes as needed.
        (record_references_in_initializer): Add new only_vars parameter.
        * cgraph.h (record_references_in_initializer): New parameter.
        * varasm.c (assemble_variable): Update call.
        * varpool.c (varpool_analyze_pending_decls): Always look for
        referenced vars.
Index: cgraphbuild.c
===================================================================
--- cgraphbuild.c       (revision 152557)
+++ cgraphbuild.c       (working copy)
@@ -33,13 +33,16 @@ along with GCC; see the file COPYING3.  
 #include "tree-pass.h"

 /* Walk tree and record all calls and references to functions/variables.
-   Called via walk_tree: TP is pointer to tree to be examined.  */
+   Called via walk_tree: TP is pointer to tree to be examined.  
+   When DATA is non-null, record references to callgraph.
+   */

 static tree
-record_reference (tree *tp, int *walk_subtrees, void *data ATTRIBUTE_UNUSED)
+record_reference (tree *tp, int *walk_subtrees, void *data)
 {
   tree t = *tp;
   tree decl;
+  bool do_callgraph = data != NULL;

   switch (TREE_CODE (t))
     {
@@ -57,7 +60,7 @@ record_reference (tree *tp, int *walk_su
       /* Record dereferences to the functions.  This makes the
         functions reachable unconditionally.  */
       decl = TREE_OPERAND (*tp, 0);
-      if (TREE_CODE (decl) == FUNCTION_DECL)
+      if (TREE_CODE (decl) == FUNCTION_DECL && do_callgraph)
        cgraph_mark_address_taken_node (cgraph_node (decl));
       break;

@@ -218,13 +221,15 @@ struct gimple_opt_pass pass_build_cgraph
 };

 /* Record references to functions and other variables present in the
-   initial value of DECL, a variable.  */
+   initial value of DECL, a variable.  
+   When ONLY_VARS is true, we mark needed only variables, not functions.  */

 void
-record_references_in_initializer (tree decl)
+record_references_in_initializer (tree decl, bool only_vars)
 {
   struct pointer_set_t *visited_nodes = pointer_set_create ();
-  walk_tree (&DECL_INITIAL (decl), record_reference, NULL, visited_nodes);
+  walk_tree (&DECL_INITIAL (decl), record_reference, 
+            only_vars ? NULL : decl, visited_nodes);
   pointer_set_destroy (visited_nodes);
 }

Index: cgraph.h
===================================================================
--- cgraph.h    (revision 152557)
+++ cgraph.h    (working copy)
@@ -459,7 +459,7 @@ struct cgraph_node *cgraph_function_vers
                                                bitmap);
 void tree_function_versioning (tree, tree, VEC (ipa_replace_map_p,gc)*, bool,
bitmap);
 struct cgraph_node *save_inline_function_body (struct cgraph_node *);
-void record_references_in_initializer (tree);
+void record_references_in_initializer (tree, bool);
 bool cgraph_process_new_functions (void);

 bool cgraph_decide_is_function_needed (struct cgraph_node *, tree);
Index: varasm.c
===================================================================
--- varasm.c    (revision 152557)
+++ varasm.c    (working copy)
@@ -2082,7 +2082,7 @@ assemble_variable (tree decl, int top_le
             Without this, if the variable is placed in a
             section-anchored block, the template will only be marked
             when it's too late.  */
-         record_references_in_initializer (to);
+         record_references_in_initializer (to, false);
        }

       decl = to;
Index: varpool.c
===================================================================
--- varpool.c   (revision 152557)
+++ varpool.c   (working copy)
@@ -365,9 +365,9 @@ varpool_analyze_pending_decls (void)
             already informed about increased alignment.  */
           align_variable (decl, 0);

-          if (DECL_INITIAL (decl))
-           record_references_in_initializer (decl);
        }
+      if (DECL_INITIAL (decl))
+       record_references_in_initializer (decl, analyzed);
       changed = true;
     }
   timevar_pop (TV_CGRAPH);


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41626

[Bug middle-end/41626] [4.5 Regression] GCC does not bootstrap on powerpc64

[meissner at gcc dot gnu dot org]

------- Comment #8 from meissner at gcc dot gnu dot org  2009-10-08 18:59 -------
Created an attachment (id=18754)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=18754&action=view)
Patch that fixes the powerpc bootstrap problem.

This is the patch that I checked in that came from Jan Hubicka.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41626

[Bug middle-end/41626] [4.5 Regression] GCC does not bootstrap on powerpc64

[meissner at gcc dot gnu dot org]

------- Comment #9 from meissner at gcc dot gnu dot org  2009-10-08 19:00 -------
Patch checked into mainline, subversion id 152569.


-- 

meissner at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41626