This fixes the Ada bootstrap failure introduced by alloca folding.
We now fold alloca (0) to &auto-with-size-zero which confuses us.
I didn't exactly investigate but what I think happens is that we
expand that &auto-with-size-zero to NULL instead of
virtual_stack_dynamic_rtx (see zero-size special-case in
allocate_dynamic_stack_space) and Ada ends up dereferencing the
pointer returned from alloca (0) (something to investigate for
the Ada folks I guess), something which "works" if we just
return a random stack address.
The following patch restores previous behavior by simply not
folding alloca (0).
Bootstrapped on x86_64-unknown-linux-gnu, applied to trunk.
Richard.
2011-09-02 Richard Guenther <rguent...@suse.de>
* tree-ssa-ccp.c (fold_builtin_alloca_for_var): Do not
fold alloca (0).
(ccp_fold_stmt): Continue replacing args when folding
alloca fails.
Index: gcc/tree-ssa-ccp.c
===================================================================
--- gcc/tree-ssa-ccp.c (revision 178460)
+++ gcc/tree-ssa-ccp.c (working copy)
@@ -1702,10 +1687,14 @@ fold_builtin_alloca_for_var (gimple stmt
/* Detect constant argument. */
arg = get_constant_value (gimple_call_arg (stmt, 0));
- if (arg == NULL_TREE || TREE_CODE (arg) != INTEGER_CST
+ if (arg == NULL_TREE
+ || TREE_CODE (arg) != INTEGER_CST
|| !host_integerp (arg, 1))
return NULL_TREE;
+
size = TREE_INT_CST_LOW (arg);
+ if (size == 0)
+ return NULL_TREE;
/* Heuristic: don't fold large vlas. */
threshold = (unsigned HOST_WIDE_INT)PARAM_VALUE (PARAM_LARGE_STACK_FRAME);
@@ -1804,12 +1793,12 @@ ccp_fold_stmt (gimple_stmt_iterator *gsi
if (gimple_call_alloca_for_var_p (stmt))
{
tree new_rhs = fold_builtin_alloca_for_var (stmt);
- bool res;
- if (new_rhs == NULL_TREE)
- return false;
- res = update_call_from_tree (gsi, new_rhs);
- gcc_assert (res);
- return true;
+ if (new_rhs)
+ {
+ bool res = update_call_from_tree (gsi, new_rhs);
+ gcc_assert (res);
+ return true;
+ }
}
/* Propagate into the call arguments. Compared to replace_uses_in
