Have you ever been "nerd sniped"? When I realized that exceptions are capable of reducing binaries sizes, it felt like the only thing standing in the way of its adoption was its performance concerns. I couldn't let it go. I wanted to see how far the algorithm could be pushed. In 2024, I set out to just that: making C++ exceptions blazingly fast for embedded platforms. A year later, I've achieved a +90% performance improvement for ARM microcontrollers.
This talk takes you behind the scenes of this journey. I'll share the complete roadmap—from initial benchmarking to optimization—revealing techniques applicable to any performance-critical C++ system. But this isn't just a technical story; it's a testament to the power of community collaboration and how the C++ community summoned this talk. Even if you've sworn off exceptions forever, this talk still delivers value. It's a course in performance optimization which applies across the entire C++ ecosystem.
Join me for a deep dive into what happens when obsession meets optimization—and discover just how fast C++ can really be. Whether you're writing embedded firmware, high-frequency trading systems, a game, or a web server in C++ you'll walk away with techniques to dramatically improve your code.
You'll discover:
- Building precise measurement frameworks for cpu cycle improvements
- Navigating and modernizing a 25-year-old codebase
- Demystifying and eliminating the actual sources of non-determinism
- Leveraging community expertise to shape product requirements
- Using data-oriented design to convert O(log(n)) operations to O(1)
- Working within fixed ABI constraints without compromising performance
- Exploring the future landscape of C++ exception handling
