This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
[PATCH] Fix intrinsic_pack.f90, tree-nested not marking some decls as addressable
- From: Andrew Pinski <pinskia at physics dot uc dot edu>
- To: GCC Patches <gcc-patches at gcc dot gnu dot org>
- Cc: tobias dot schlueter at physik dot uni-muenchen dot de
- Date: Tue, 5 Oct 2004 14:24:05 -0400
- Subject: [PATCH] Fix intrinsic_pack.f90, tree-nested not marking some decls as addressable
The problem here is that we have "&a[4];" which we change to
&FRAME.117.a[4]; in tree-nested but we don't mark the frame as
addressable so we get an ICE. This fixes the problem by instead
of checking if the immediate part of the ADDR_EXPR we should check
that we changed any part of the address expression.
OK? Bootstrapped and tested on powerpc-darwin.
The testcase is already in the testsuite and fails currently.
Thanks,
Andrew Pinski
ChangeLog:
* tree-nested.c (walk_stmt_info): Add changed field.
(convert_nonlocal_reference): Set changed to when we
change a decl to unnested decl.
<case ADDR_EXPR>: Instead of checking if the immediate part
of the ADDR_EXPR changed, check the field changed.
Use recompute_tree_invarant_for_addr_expr instead of unsetting
TREE_INVARIANT.
(convert_local_reference): Set changed to when we
change a decl to unnested decl.
<case ADDR_EXPR>: Instead of checking if the immediate part
of the ADDR_EXPR changed, check the field changed.
Also call recompute_tree_invarant_for_addr_expr on the ADDR_EXPR.
Index: tree-nested.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/tree-nested.c,v
retrieving revision 2.19
diff -u -p -r2.19 tree-nested.c
--- tree-nested.c 20 Sep 2004 20:38:17 -0000 2.19
+++ tree-nested.c 5 Oct 2004 18:08:51 -0000
@@ -518,6 +518,7 @@ struct walk_stmt_info
tree_stmt_iterator tsi;
struct nesting_info *info;
bool val_only;
+ bool changed;
};
/* A subroutine of walk_function. Iterate over all sub-statements of
*TP. */
@@ -732,6 +747,7 @@ convert_nonlocal_reference (tree *tp, in
tree target_context = decl_function_context (t);
struct nesting_info *i;
tree x;
+ wi->changed = true;
for (i = info->outer; i->context != target_context; i = i->outer)
continue;
@@ -770,17 +786,17 @@ convert_nonlocal_reference (tree *tp, in
case ADDR_EXPR:
{
bool save_val_only = wi->val_only;
- tree save_sub = TREE_OPERAND (t, 0);
+ wi->changed = false;
wi->val_only = false;
walk_tree (&TREE_OPERAND (t, 0), convert_nonlocal_reference, wi,
NULL);
wi->val_only = true;
- if (save_sub != TREE_OPERAND (t, 0))
+ if (wi->changed)
{
/* If we changed anything, then TREE_INVARIANT is be wrong,
since we're no longer directly referencing a decl. */
- TREE_INVARIANT (t) = 0;
+ recompute_tree_invarant_for_addr_expr (t);
/* If the callback converted the address argument in a context
where we only accept variables (and min_invariant, presumably),
@@ -874,6 +890,7 @@ convert_local_reference (tree *tp, int *
field = lookup_field_for_decl (info, t, NO_INSERT);
if (!field)
break;
+ wi->changed = true;
x = get_frame_field (info, info->context, field, &wi->tsi);
if (wi->val_only)
@@ -885,17 +902,19 @@ convert_local_reference (tree *tp, int *
case ADDR_EXPR:
{
bool save_val_only = wi->val_only;
- tree save_sub = TREE_OPERAND (t, 0);
+ wi->changed = false;
wi->val_only = false;
walk_tree (&TREE_OPERAND (t, 0), convert_local_reference, wi, NULL);
wi->val_only = save_val_only;
/* If we converted anything ... */
- if (TREE_OPERAND (t, 0) != save_sub)
+ if (wi->changed)
{
/* Then the frame decl is now addressable. */
TREE_ADDRESSABLE (info->frame_decl) = 1;
+
+ recompute_tree_invarant_for_addr_expr (t);
/* If we are in a context where we only accept values, then
compute the address into a temporary. */