Runtime scheduling of dynamic parallelism on accelerator-based multi-core systems