LLVM: A Compilation Framework for Lifelong Program Analysis & Transformation

  • Authors:

  • Chris Lattner;Vikram Adve

  • Affiliations:

  • -;-

  • Venue:
  • Proceedings of the international symposium on Code generation and optimization: feedback-directed and runtime optimization
  • Year:
  • 2004

Quantified Score

Hi-index 0.03

Visualization

Abstract

This paper describes LLVM (Low Level Virtual Machine),a compiler framework designed to support transparent, lifelongprogram analysis and transformation for arbitrary programs,by providing high-level information to compilertransformations at compile-time, link-time, run-time, and inidle time between runs.LLVM defines a common, low-levelcode representation in Static Single Assignment (SSA) form,with several novel features: a simple, language-independenttype-system that exposes the primitives commonly used toimplement high-level language features; an instruction fortyped address arithmetic; and a simple mechanism that canbe used to implement the exception handling features ofhigh-level languages (and setjmp/longjmp in C) uniformlyand efficiently.The LLVM compiler framework and coderepresentation together provide a combination of key capabilitiesthat are important for practical, lifelong analysis andtransformation of programs.To our knowledge, no existingcompilation approach provides all these capabilities.We describethe design of the LLVM representation and compilerframework, and evaluate the design in three ways: (a) thesize and effectiveness of the representation, including thetype information it provides; (b) compiler performance forseveral interprocedural problems; and (c) illustrative examplesof the benefits LLVM provides for several challengingcompiler problems.