Patchwork RFA: Fix debug/44832

login
register
mail settings
Submitter Joern Rennecke
Date July 10, 2010, 4:43 a.m.
Message ID <20100710004303.ez0r2n9bcgsk4w40-nzlynne@webmail.spamcop.net>
Download mbox | patch
Permalink /patch/58460/
State New
Headers show

Comments

Joern Rennecke - July 10, 2010, 4:43 a.m.
Bootstrapped & regression tested on i686-pc-linux-gnu on top of the patch
for PR other/44874 in trunk revision 161952; as expected, the pr44832 test
is fixed for gcc and g++; there are no new regressions.

It also fixes the -fcompare-debug failures on i686-pc-linux-gnu for the
various hand-run test cases we have for this PR.
2010-07-10  Richard Guenther  <rguenther@suse.de>
	    Joern Rennecke  <joern.rennecke@embecosm.com>

	PR debug/44832
	* tree-ssa-live.c (mark_all_vars_used_1): Set TREE_USED for
	LABEL_DECL.
	(remove_unused_scope_block_p): Don't drop TREE_USED LABEL_DECLs
	unless they have DECL_IGNORED_P set.
Richard Guenther - July 10, 2010, 8:14 a.m.
On Sat, Jul 10, 2010 at 6:43 AM, Joern Rennecke <amylaar@spamcop.net> wrote:
> Bootstrapped & regression tested on i686-pc-linux-gnu on top of the patch
> for PR other/44874 in trunk revision 161952; as expected, the pr44832 test
> is fixed for gcc and g++; there are no new regressions.
>
> It also fixes the -fcompare-debug failures on i686-pc-linux-gnu for the
> various hand-run test cases we have for this PR.
>
> 2010-07-10  Richard Guenther  <rguenther@suse.de>
>            Joern Rennecke  <joern.rennecke@embecosm.com>
>
>        PR debug/44832
>        * tree-ssa-live.c (mark_all_vars_used_1): Set TREE_USED for
>        LABEL_DECL.
>        (remove_unused_scope_block_p): Don't drop TREE_USED LABEL_DECLs
>        unless they have DECL_IGNORED_P set.
>
> diff -pu ../../../pr44874/gcc/gcc/tree-ssa-live.c ./tree-ssa-live.c
> --- ../../../pr44874/gcc/gcc/tree-ssa-live.c    2010-07-09
> 19:46:22.862606781 +0100
> +++ ./tree-ssa-live.c   2010-07-09 19:49:13.544981717 +0100
> @@ -384,6 +384,19 @@ mark_all_vars_used_1 (tree *tp, int *wal
>        }
>       set_is_used (t);
>     }
> +  /* remove_unused_scope_block_p requires information about labels
> +     which are not DECL_IGNORED_P to tell if they might be used in the IL.
>  */
> +  if (TREE_CODE (t) == LABEL_DECL)
> +#if 0
> +    /* A label that has not been artifically created by the compiler should
> +       only be used in the IL if it has been explicitly used by the
> +       programmer.  */
> +    gcc_assert (TREE_USED (t) || DECL_IGNORED_P (t));
> +#else
> +    /* Actually, init_vars_expansion clears TREE_USED for LABEL_DECLs too,
> +       so we must compute it here.  */
> +    TREE_USED (t) = 1;
> +#endif

No #if 0 code please, just remove it.

Ok with that change.

Thanks,
Richard.

>   if (IS_TYPE_OR_DECL_P (t))
>     *walk_subtrees = 0;
> @@ -463,6 +476,19 @@ remove_unused_scope_block_p (tree scope)
>       else if ((ann = var_ann (*t)) != NULL
>                && ann->used)
>        unused = false;
> +      else if (TREE_CODE (*t) == LABEL_DECL && TREE_USED (*t))
> +       /* For labels that are still used in the IL, the decision to
> +          preserve them must not depend DEBUG_INFO_LEVEL, otherwise we
> +          risk having different ordering in debug vs.  non-debug builds
> +          during inlining or versioning.
> +          A label appearing here (we have already checked DECL_IGNORED_P)
> +          should not be used in the IL unless it has been explicitly used
> +          before, so we use TREE_USED as an approximation.  */
> +       /* In principle, we should do the same here as for the debug case
> +          below, however, when debugging, there might be additional nested
> +          levels that keep an upper level with a label live, so we have to
> +          force this block to be considered used, too.  */
> +       unused = false;
>
>       /* When we are not doing full debug info, we however can keep around
>         only the used variables for cfgexpand's memory packing saving quite
>
>

Patch

diff -pu ../../../pr44874/gcc/gcc/tree-ssa-live.c ./tree-ssa-live.c
--- ../../../pr44874/gcc/gcc/tree-ssa-live.c	2010-07-09 19:46:22.862606781 +0100
+++ ./tree-ssa-live.c	2010-07-09 19:49:13.544981717 +0100
@@ -384,6 +384,19 @@  mark_all_vars_used_1 (tree *tp, int *wal
 	}
       set_is_used (t);
     }
+  /* remove_unused_scope_block_p requires information about labels
+     which are not DECL_IGNORED_P to tell if they might be used in the IL.  */
+  if (TREE_CODE (t) == LABEL_DECL)
+#if 0
+    /* A label that has not been artifically created by the compiler should
+       only be used in the IL if it has been explicitly used by the
+       programmer.  */
+    gcc_assert (TREE_USED (t) || DECL_IGNORED_P (t));
+#else
+    /* Actually, init_vars_expansion clears TREE_USED for LABEL_DECLs too,
+       so we must compute it here.  */
+    TREE_USED (t) = 1;
+#endif
 
   if (IS_TYPE_OR_DECL_P (t))
     *walk_subtrees = 0;
@@ -463,6 +476,19 @@  remove_unused_scope_block_p (tree scope)
       else if ((ann = var_ann (*t)) != NULL
 		&& ann->used)
 	unused = false;
+      else if (TREE_CODE (*t) == LABEL_DECL && TREE_USED (*t))
+	/* For labels that are still used in the IL, the decision to
+	   preserve them must not depend DEBUG_INFO_LEVEL, otherwise we
+	   risk having different ordering in debug vs.  non-debug builds
+	   during inlining or versioning.
+	   A label appearing here (we have already checked DECL_IGNORED_P)
+	   should not be used in the IL unless it has been explicitly used
+	   before, so we use TREE_USED as an approximation.  */
+	/* In principle, we should do the same here as for the debug case
+	   below, however, when debugging, there might be additional nested
+	   levels that keep an upper level with a label live, so we have to
+	   force this block to be considered used, too.  */
+	unused = false;
 
       /* When we are not doing full debug info, we however can keep around
 	 only the used variables for cfgexpand's memory packing saving quite