Many members of the LLVM community from both industry and academia are working towards addressing an important problem: shipping software as LLVM IR for more flexible analysis and transformation. Examples of these efforts include technologies such as `-fembed-bitcode`, ThinLTO, and WLLVM.
We propose a BoF for these parties and all interested to meet and discuss the benefits and technical challenges involved, learn about each others' goals and use-cases, and to identify collaboration opportunities across these overlapping projects.
Our interest: We at UIUC are developing a system called "ALLVM" in which all components are represented as LLVM IR first and foremost. Our goal is to explore the potential benefits of the approach for improving performance, strengthening security, and simplifying failure diagnosis for production code. A second goal is to make ALLVM available widely as a platform for research. As part of this ongoing project we are developing and automating the construction of complete LLVM-based representations of real-world software, as well as building an ecosystem of supporting tools.
