https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89779
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Blocks| |88945
--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> ---
So we fail to run into the defensive
static inline bool
tree_nop_conversion (const_tree exp)
{
...
inner_type = TREE_TYPE (TREE_OPERAND (exp, 0));
if (!inner_type)
return false;
now because the type is not NULL but error_mark_node. Of course the real
issue is that we are operating on trees with released SSA names at all.
The above is from r158045
2010-04-07 Richard Guenther <rguent...@suse.de>
PR middle-end/42617
* emit-rtl.c (set_mem_attributes_minus_bitpos): Do not
discard plain indirect references.
* fold-const.c (operand_equal_p): Guard against NULL_TREE type.
* tree.c (tree_nop_conversion): Likewise.
where the tree_nop_conversion already has other defenses:
if (TREE_OPERAND (exp, 0) == error_mark_node)
return false;
we can consolidate those checks now.
Index: gcc/tree.c
===================================================================
--- gcc/tree.c (revision 269832)
+++ gcc/tree.c (working copy)
@@ -12812,13 +12812,10 @@ tree_nop_conversion (const_tree exp)
if (!CONVERT_EXPR_P (exp)
&& TREE_CODE (exp) != NON_LVALUE_EXPR)
return false;
- if (TREE_OPERAND (exp, 0) == error_mark_node)
- return false;
outer_type = TREE_TYPE (exp);
inner_type = TREE_TYPE (TREE_OPERAND (exp, 0));
-
- if (!inner_type)
+ if (inner_type == error_mark_node)
return false;
return tree_nop_conversion_p (outer_type, inner_type);
Index: gcc/fold-const.c
===================================================================
--- gcc/fold-const.c (revision 269832)
+++ gcc/fold-const.c (working copy)
@@ -2973,11 +2973,6 @@ operand_equal_p (const_tree arg0, const_
|| TREE_TYPE (arg1) == error_mark_node)
return 0;
- /* Similar, if either does not have a type (like a released SSA name),
- they aren't equal. */
- if (!TREE_TYPE (arg0) || !TREE_TYPE (arg1))
- return 0;
-
/* We cannot consider pointers to different address space equal. */
if (POINTER_TYPE_P (TREE_TYPE (arg0))
&& POINTER_TYPE_P (TREE_TYPE (arg1))
The underlying issue is that IVOPTs removes unused IVs but keeps
loop->control_ivs uncleared (it clears the SCEV cache which also
resets loop->nb_iterations). The best fix for IVOPTs is probably
to simply delay removal of those IVs and also properly free
control-iv information.
Referenced Bugs:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88945
[Bug 88945] ICE in fold_convert_loc in FRE when using -fdump-tree-fre-details
