Hi!
When a taskloop doesn't have any iterations, GOMP_taskloop* takes an early
return, doesn't create any tasks and more importantly, doesn't create
a taskgroup and doesn't register task reductions. But, the code emitted
in the callers assumes task reductions have been registered and performs
the reduction handling and task reduction unregistration. The pointer
to the task reduction private variables is reused, on input it is the alignment
and only on output it is the pointer, so in the case taskloop with no iterations
the caller attempts to dereference the alignment value as if it was a pointer
and crashes. We could in the early returns register the task reductions
only to have them looped over and unregistered in the caller, but I think
it is better to tell the caller there is nothing to task reduce and bypass
all that.
Bootstrapped/regtested on x86_64-linux and i686-linux, committed to trunk
so far.
2021-05-11 Jakub Jelinek <ja...@redhat.com>
PR middle-end/100471
* omp-low.c (lower_omp_task_reductions): For OMP_TASKLOOP, if data
is 0, bypass the reduction loop including
GOMP_taskgroup_reduction_unregister call.
* taskloop.c (GOMP_taskloop): If GOMP_TASK_FLAG_REDUCTION and not
GOMP_TASK_FLAG_NOGROUP, when doing early return clear the task
reduction pointer.
* testsuite/libgomp.c/task-reduction-4.c: New test.
--- gcc/omp-low.c.jj 2021-05-10 12:22:30.391452318 +0200
+++ gcc/omp-low.c 2021-05-10 13:39:39.366205162 +0200
@@ -8781,7 +8781,7 @@ lower_omp_task_reductions (omp_context *
tree num_thr_sz = create_tmp_var (size_type_node);
tree lab1 = create_artificial_label (UNKNOWN_LOCATION);
tree lab2 = create_artificial_label (UNKNOWN_LOCATION);
- tree lab3 = NULL_TREE;
+ tree lab3 = NULL_TREE, lab7 = NULL_TREE;
gimple *g;
if (code == OMP_FOR || code == OMP_SECTIONS)
{
@@ -8846,6 +8846,14 @@ lower_omp_task_reductions (omp_context *
NULL_TREE, NULL_TREE);
tree data = create_tmp_var (pointer_sized_int_node);
gimple_seq_add_stmt (end, gimple_build_assign (data, t));
+ if (code == OMP_TASKLOOP)
+ {
+ lab7 = create_artificial_label (UNKNOWN_LOCATION);
+ g = gimple_build_cond (NE_EXPR, data,
+ build_zero_cst (pointer_sized_int_node),
+ lab1, lab7);
+ gimple_seq_add_stmt (end, g);
+ }
gimple_seq_add_stmt (end, gimple_build_label (lab1));
tree ptr;
if (TREE_CODE (TYPE_SIZE_UNIT (record_type)) == INTEGER_CST)
@@ -9209,6 +9217,8 @@ lower_omp_task_reductions (omp_context *
g = gimple_build_call (t, 1, build_fold_addr_expr (avar));
}
gimple_seq_add_stmt (end, g);
+ if (lab7)
+ gimple_seq_add_stmt (end, gimple_build_label (lab7));
t = build_constructor (atype, NULL);
TREE_THIS_VOLATILE (t) = 1;
gimple_seq_add_stmt (end, gimple_build_assign (avar, t));
--- libgomp/taskloop.c.jj 2021-01-04 10:25:56.074038599 +0100
+++ libgomp/taskloop.c 2021-05-10 12:32:04.024191809 +0200
@@ -51,20 +51,32 @@ GOMP_taskloop (void (*fn) (void *), void
/* If parallel or taskgroup has been cancelled, don't start new tasks. */
if (team && gomp_team_barrier_cancelled (&team->barrier))
- return;
+ {
+ early_return:
+ if ((flags & (GOMP_TASK_FLAG_NOGROUP | GOMP_TASK_FLAG_REDUCTION))
+ == GOMP_TASK_FLAG_REDUCTION)
+ {
+ struct gomp_data_head { TYPE t1, t2; uintptr_t *ptr; };
+ uintptr_t *ptr = ((struct gomp_data_head *) data)->ptr;
+ /* Tell callers GOMP_taskgroup_reduction_register has not been
+ called. */
+ ptr[2] = 0;
+ }
+ return;
+ }
#ifdef TYPE_is_long
TYPE s = step;
if (step > 0)
{
if (start >= end)
- return;
+ goto early_return;
s--;
}
else
{
if (start <= end)
- return;
+ goto early_return;
s++;
}
UTYPE n = (end - start + s) / step;
@@ -73,13 +85,13 @@ GOMP_taskloop (void (*fn) (void *), void
if (flags & GOMP_TASK_FLAG_UP)
{
if (start >= end)
- return;
+ goto early_return;
n = (end - start + step - 1) / step;
}
else
{
if (start <= end)
- return;
+ goto early_return;
n = (start - end - step - 1) / -step;
}
#endif
--- libgomp/testsuite/libgomp.c/task-reduction-4.c.jj 2021-05-10
12:31:37.628479637 +0200
+++ libgomp/testsuite/libgomp.c/task-reduction-4.c 2021-05-10
12:30:57.966912118 +0200
@@ -0,0 +1,21 @@
+/* PR middle-end/100471 */
+
+extern void abort (void);
+
+int c;
+
+int
+main ()
+{
+#pragma omp parallel
+#pragma omp single
+ {
+ int r = 0, i;
+ #pragma omp taskloop reduction(+:r)
+ for (i = 0; i < c; i++)
+ r++;
+ if (r != 0)
+ abort ();
+ }
+ return 0;
+}
Jakub
