Open Source Summit North America 2023

Innovation is the fuel and power of open-source projects. To propel that innovation, every year the toolchain community focuses on creating new tools for open-source developers. This presentation aims to show some of the new GNU toolchains features for 2023. This year the Glibc project has added the capability to return high-quality randomness from the kernel through three new functions. At the same time, it has added a new function that allows a caller to release the memory of a dying process for better memory management of applications. The GCC compiler now has optimizations for five new x86 platforms, these optimizations enable ISA extensions for video codec, integer FMA, and Vector Neural Network Instructions based on INT8 among others. This presentation will show some examples in C code for developers to take advantage of these new ISA. GCC 13 will also have Address Sanitizer and Link-time optimization improvements. Having a better understanding of these innovations in toolchains allows developers to showcase the best of new platform architecture technology for users’ applications as well as boost the innovation and security of incoming projects.

Opportunities for parallel computing are everywhere, from laptops with multicore processors and GPUs to commodity clusters, the cloud, and supercomputers. Yet even as parallelism becomes more ubiquitous, performance-oriented parallel programming remains challenging, and over time has arguably become even more so. This talk will introduce Chapel, a programming language that supports code that is clear and concise, yet whose performance and scalability competes with or beats traditional approaches like Fortran/C/C++, POSIX threads, MPI, OpenMP, CUDA, etc. Recent applications of Chapel in Aerodynamics, Data Science, and Atmospheric Research will be presented. Attendees can expect to leave the talk with an understanding of Chapel’s key features and their uses for students, hobbyists, and scientists who would benefit from scalable, parallel computations.

The SFrame format is a new stack trace format, the support for which we recently introduced into the GNU Toolchain. It represents the information needed for generating simple stack traces. The SFrame format keeps track of the minimal necessary information needed for stack tracing: Canonical Frame Address (CFA), Frame Pointer (FP), Return Address (RA). This format is designed for those applications which need fast, online stack traces, while relieving the need for them to preserve the frame pointer. This can be useful, for instance, in production-like environments, in which many constraints may exist. Support for the SFrame format is available in Binutils 2.40 (Spec: https://sourceware.org/binutils/docs/sframe-spec.html). This talk will introduce the format, its support in Binutils and showcase how simple it is to generate stack traces using SFrame. It will also discuss the next steps for the format and its adoption in the GNU/Linux ecosystem. Since this work was last presented, several new ideas and changes (including a new name) were implemented. This talk will cover the progress made since then, and discuss ideas for possible use cases. The talk will also showcase some libsframe APIs and show how to easily generate stack traces using SFrame.