WARDen: Specializing Cache Coherence for High-Level Parallel Languages
High-level parallel languages (HLPLs) make it easier to write correct parallel programs.
Disciplined memory usage in these languages enables new optimizations for hardware bottlenecks, such as cache coherence.
In this work, we show how to reduce the costs of cache coherence by integrating the hardware coherence protocol directly with the programming language; no programmer effort or static analysis is required.
We identify a new low-level memory property, WARD (WAW Apathy and RAW Dependence-freedom), \emph{by construction} in HLPL programs.
We design a new coherence protocol, WARDen, to selectively disable coherence using WARD.
We evaluate WARDen with a widely-used HLPL benchmark suite on both current and future x64 machine structures.
WARDen both accelerates the benchmarks (by an average of 1.46x) and reduces energy (by 23%) by eliminating unnecessary data movement and coherency messages.
Mon 27 FebDisplayed time zone: Eastern Time (US & Canada) change
15:40 - 17:00 | Session 3 -- PotpourriMain Conference at Montreal 1-2-3 Chair(s): Bernhard Egger Seoul National University | ||
15:40 26mTalk | Fast Polynomial Evaluation for Correctly Rounded Elementary Functions using the RLIBM Approach Main Conference DOI | ||
16:06 26mTalk | A Game-Based Framework to Compare Program Classifiers and Evaders Main Conference Thaís Regina Damásio Federal University of Minas Gerais, Michael Canesche Federal University of Minas Gerais, Vinícius Pacheco Federal University of Minas Gerais, Marcus Botacin Texas A&M University, Anderson Faustino da Silva State University of Maringá, Fernando Magno Quintão Pereira Federal University of Minas Gerais DOI | ||
16:33 26mTalk | WARDen: Specializing Cache Coherence for High-Level Parallel Languages Main Conference Michael Wilkins Northwestern University, Sam Westrick Carnegie Mellon University, Vijay Kandiah Northwestern University, Alex Bernat Northwestern University, Brian Suchy Northwestern University, Enrico Armenio Deiana Northwestern University, Simone Campanoni Northwestern University, Umut A. Acar Carnegie Mellon University, Peter Dinda Northwestern University, Nikos Hardavellas Northwestern University DOI | ||