Bridging the gap between system-level and chip-level performance optimization