Use starpu_execute_on_each_worker_ex() instead of starpu_execute_on_each_worker()
Compare changes
@@ -131,7 +131,7 @@ int RUNTIME_starpu_ws_alloc(CHAMELEON_starpu_ws_t **workspace,
@@ -141,7 +141,7 @@ int RUNTIME_starpu_ws_free(CHAMELEON_starpu_ws_t *workspace)