WMM : a resilient Weak Memory Model
Author(s)
Zhang, Sizhuo![Thumbnail](/bitstream/handle/1721.1/103667/953456743-MIT.pdf.jpg?sequence=3&isAllowed=y)
DownloadFull printable version (612.9Kb)
Alternative title
Weak Memory Model : a resilient Weak Memory Model
Resilient Weak Memory Model
Other Contributors
Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science.
Advisor
Arvind.
Terms of use
Metadata
Show full item recordAbstract
A good memory model should have a precise definition that can be understood by any computer architect readily. It should also be resilient in the sense that it should not break when new microarchitecture optimizations are introduced to improve single-threaded performance. We introduce WMM, a new weak memory model, which meets these criteria. WMM permits all load-store reorderings except a store is not allowed to overtake a load. WMM also permits both memory dependency speculation and load-value prediction. We define the operational semantics of WMM using a novel conceptual device called invalidation buffer, which achieves the effect of out-of-order instruction execution even when instructions are executed in-order and one-at-a-time. We show via examples where memory fences need to be inserted for different programming paradigms. We highlight the differences between WMM and other weak memory models including Release Consistency and Power. Our preliminary performance evaluation using the SPLASH benchmarks shows that WMM implementation performs significantly better than the aggressive implementations of SC. WMM holds the promise to be a vendor-independent stable memory model which will not stifle microarchitectural innovations.
Description
Thesis: S.M., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, 2016. This electronic version was submitted by the student author. The certified thesis is available in the Institute Archives and Special Collections. Cataloged from student-submitted PDF version of thesis. Includes bibliographical references (pages 61-65).
Date issued
2016Department
Massachusetts Institute of Technology. Department of Electrical Engineering and Computer SciencePublisher
Massachusetts Institute of Technology
Keywords
Electrical Engineering and Computer Science.