Hyperchaining Optimizations for an LLVM-Based Binary Translator on x86-64 and RISC-V Platforms

Jyun Kai Lai, Wuu Yang

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

1 Scopus citations

Abstract

Rabbit is an LLVM-based hybrid binary translator with many diverse optimizations to improve the performance. In addition to platform-independent hyperchaining (indep), Rabbit also includes platform-dependent hyperchaining (dep) on both x86-64 and RISC-V architectures for both direct and indirect branches. The dep optimizations leverage architecture-specific instructions and patches to achieve the same effect as the indep optimiztion but gains more performance improvement. The experimental results show that the platform-dependent hyperchaining can achieve 1.08x and 1.05x speedup in comparison with platform-independent hyperchaining for direct and indirect branches, respectively. The experimental results also show that platform-dependent hyperchaining incurs little memory space overhead in comparison with platform-independent hyperchaining.

Original languageEnglish
Title of host publication50th International Conference on Parallel Processing Workshop, ICPP 2021 - Proceedings
PublisherAssociation for Computing Machinery
ISBN (Electronic)9781450384414
DOIs
StatePublished - 9 Aug 2021
Event50th International Conference on Parallel Processing Workshop, ICPP 2021 - Virtual, Online, United States
Duration: 9 Aug 202112 Aug 2021

Publication series

NameACM International Conference Proceeding Series

Conference

Conference50th International Conference on Parallel Processing Workshop, ICPP 2021
Country/TerritoryUnited States
CityVirtual, Online
Period9/08/2112/08/21

Keywords

  • Binary translation
  • Block chaining
  • Hyperchaining
  • Llvm
  • Optimization
  • Rabbit
  • Risc-v
  • X86-64

Fingerprint

Dive into the research topics of 'Hyperchaining Optimizations for an LLVM-Based Binary Translator on x86-64 and RISC-V Platforms'. Together they form a unique fingerprint.

Cite this