Exploiting GPU On-chip Shared Memory for Accelerating Schedulability Analysis