The LLVM OMPD ompd_get_task_in_parallel() function has a typo in the CUDA device code. It misspells "ompd_nvptx_parallel_info_t" as "ompd_nvptx_paralel_info". Diff to fix it follows:
-------------------------- libompd/src/omp-debug.cpp --------------------------
index fae2b63..8e9b645 100644
@@ -865,7 +865,7 @@ ompd_rc_t ompd_get_task_in_parallel(
if (parallel_handle->ah->kind == OMPD_DEVICE_KIND_CUDA) {
ret = TValue(context, parallel_handle->th)
- .cast("ompd_nvptx_paralel_info", 0,
+ .cast("ompd_nvptx_parallel_info_t", 0,
OMPD_SEGMENT_CUDA_PTX_GLOBAL)
.access("parallel_tasks")
.cast("omptarget_nvptx_TaskDescr", 1,