Kristof Beyls

Compilers play a crucial role in hardening software against security attacks. As compiler engineers we experience an increase in demand for security-related features: we simply work on security-related features more often. We find it tough to analyze whether the hardenings we implement are easy or hard to circumvent by attackers. After chatting about this with many compiler developers, our experience is that most feel their work would benefit from a deeper understanding of attacks and hardening techniques. After having looked around, we didn't find much educational material that gives a broad overview, covering all aspects compiler developers ought to know about. Therefore, we recently started an open source book titled "Low Level Software Security for Compiler developers" at https://github.com/llsoftsec/llsoftsecbook/. It aims to improve the industry-wide knowledge about security hardening in compilers and related tools; ultimately leading to more innovation and better implementations of security features. In this presentation, we'll explain the rationale for this new open source project in more detail. We will discuss what content we have so far and what content we plan to add. The project very much welcomes new contributors: we need more new content, more review of content, discussion of ideas for how to make the book better, improvement in the design and layout of the produced HTML and PDF output, etc. We hope this presentation will reach both anyone interested in learning more about low-level software security and anyone interested in helping to grow this project further.

Ever wondered how the LLVM project and community works? Want to get some advice on how to most effectively contribute? Now is your chance at FOSDEM to ask experienced developers directly. This panel will host a number of experienced LLVM developers, answering any relevant questions from the audience.

In case questions from the audience do not fill the entire slot, a set of prepared questions focussed on how to get started working with and contributing to LLVM will be raised for the panelists to answer.

The panel will consist of the following experienced contributors to LLVM: Kristof Beyls, Peter Smith, Jeroen Dobbelaere, Nick Desaulniers

Last year, at the 2016 LLVM dev room, I presented ideas on how to improve LNT to make performance tracking of LLVM generated code easier. We've implemented most ideas since, and we're happy to see that our ideas work out in practice: quite often, understanding the root cause for a performance regression now takes about 5 to 10 minutes of work, whereas before it typically took us half a day to produce all data that is now recorded automatically.

I intend to give a demo of how we use LNT at ARM's LLVM teams. I'll also discuss how LNT is useful for other projects that need to track performance of generated code. LNT is starting to be used by at least also some GCC, java jit and even a Verilog-to-C++ compiler teams.

I'll also give the necessary pointers on what you need to get started on LNT for your own projects.

You can find a preview of the kind of things I plan to demo at http://blog.llvm.org/2016/06/using-lnt-to-track-performance.html

Ensuring that top-of-trunk consistently generates high-quality code remains harder than it should be. Continuous integration (CI) setups that track correctness of top-of-trunk work pretty well today since they automatically report correctness regressions with low false positive rate to committers. In comparison, the output generated by CI setups that track performance require far more human effort to interpret.

In this talk, I’ll describe why I think effective performance tracking is hard and what problems need solving, with a focus on our real world experiences and observations.

As part of the bring-up of one of the public performance tracking bots, I’ve done an in-depth analysis of its performance and noise characteristics. The insights gained from this analysis drove a number of improvements to LNT and the test-suite in the past year. I hope that sharing these insights will help others in setting up low-noise performance-tracking bots.

I’ll conclude by summarizing what seem to be the most important missing pieces of CI functionality to make the performance-tracking infrastructure as effective as the correctness-tracking infrastructure.

This presentation has been given before at the US LLVM dev meeting in San Jose. Given the interest there and the mostly non-overlapping audience between the FOSDEM llvm dev room and the US dev meeting, I think it's worthwhile to repeat this presentation at FOSDEM.

AArch64 is ARM's 64-bit architecture. I'll present a short history of how the AArch64 support in LLVM was implemented, a summary of which general AArch64 features are well supported by LLVM and a summary of which areas need more work.