https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127559
--- Comment #4 from Tobias Burnus <burnus at gcc dot gnu.org> ---
In cp/semantics.cc:
finish_omp_target_clauses (location_t loc, tree body, tree *clauses_ptr)
{
...
if (TREE_TYPE (t) && LAMBDA_TYPE_P (TREE_TYPE (t)))
{
tree lt = TREE_TYPE (t);
gcc_assert (CLASS_TYPE_P (lt));
if (!data->lambda_objects_accessed.contains (t)
/* Do not prepare to create target maps for locally declared
lambdas or anonymous ones. */
&& !data->local_decls.contains (t)
&& TREE_CODE (t) != TARGET_EXPR)
data->lambda_objects_accessed.add (t);
*walk_subtrees = 0;
return NULL_TREE;
}
}
* * *
For the code in question (bottom of comment 0), there is for the first lambda:
get_omega
which has data->local_decls.contains (t) == true.
And then for the second lambda, there is first
ker
which is also in local_decls – but then additionally an INIT_EXPR:
(ker.lrn_bwd_kernel()::<lambda()>::<get_omega capture> = get_omega)
where 'ker' is of course still in 'local_decls' - and the RHS is a
TARGET_EXPR - which 'get_omega' being also still in local_decls.
* * *
Not adding but continuing processing (i.e. not setting subtree walk)
will still add the LHS - while just checking the RHS will ICE as
if (current_object)
{
tree this_expr = TREE_OPERAND (current_object, 0);
fails.
* * *
The following seems to be a tad cleaner as comment 3 and seems to work ...
--- a/gcc/cp/semantics.cc
+++ b/gcc/cp/semantics.cc
@@ -11283,7 +11283,8 @@ finish_omp_target_clauses_r (tree *tp, int
tree lt = TREE_TYPE (t);
gcc_assert (CLASS_TYPE_P (lt));
- if (!data->lambda_objects_accessed.contains (t)
+ if (TREE_CODE(t) != INIT_EXPR
+ && !data->lambda_objects_accessed.contains (t)
/* Do not prepare to create target maps for locally declared
lambdas or anonymous ones. */
&& !data->local_decls.contains (t)
