DeNovo: Rethinking the Memory Hierarchy for Disciplined Parallelism | IEEE Conference Publication | IEEE Xplore