Zetian Qin
Compiler Engineering and Developer Productivity Engineer, ByteDance
Zetian Qin is a core R&D engineer for ByteDance’s C++ server-side compiler toolchain and a contributor to the LLVM community. He has long focused on building ByteDance’s internal C++ compilation and build ecosystem, advancing compiler toolchains, and optimizing compilation efficiency for large-scale engineering projects. He has deep expertise in build systems, toolchain infrastructure, and compilation performance optimization for ultra-large-scale C++ codebases. He is responsible for driving the development of ByteDance’s C++ compilation and build ecosystem and the evolution of its compiler toolchain. His work focuses on continuously improving compiler toolchains, build infrastructure, and developer productivity through systematic analysis of compilation bottlenecks and ongoing optimization of the build pipeline, helping reduce compilation costs for large-scale C++ projects while improving build efficiency and the overall developer experience.
Topic
Deploying libc++ in ByteDance Data Centers: Implementation and Challenges(Joint Presentation)
ByteDance operates C++ services across tens of millions of CPU cores, which have long relied on libstdc++. Migrating these services to libc++ has delivered performance improvements of 5–10% across nearly all business workloads. This talk will introduce the large-scale migration process and the multidimensional challenges encountered along the way, including prebuilt libraries, differences between standard library implementations, toolchains, and compilation and build systems. It will also share practical experience with AI-assisted migration. Through a range of both performance improvement and regression cases, the session will further analyze performance characteristics and optimization opportunities in C++ standard libraries. Outline 1. The Current State of STL Usage in ByteDance Data Centers Prebuilt library ecosystem Compilation and build infrastructure 2. libc++ vs. libstdc++ 3. libc++ Migration: Implementation and Challenges Challenges introduced by prebuilt libraries Compatibility and stability challenges arising from differences between STL implementations Challenges for the compilation and build infrastructure In-depth analysis of performance improvements and regressions during the migration from libstdc++ to libc++ Practical experience with AI-assisted migration 4. The Internal Evolution and Development of libc++ at ByteDance Audience Takeaways Large-scale migrations from libstdc++ to libc++ remain relatively uncommon among major Internet companies in China. ByteDance completed the migration of its core services to LLVM in 2021, while continuing to use the GNU-based libstdc++ standard library. In 2026, to achieve greater performance gains and compatibility, ByteDance migrated from libstdc++ to libc++ based on LLVM 20. Through this process, the team accumulated extensive experience in deploying libc++ across large-scale production C++ workloads. Attendees will gain practical insights into how to successfully deploy libc++ across large-scale C++ services, including a comparison of libstdc++ and libc++, performance analysis, and an examination of issues arising from ABI differences. The session will also share advanced practices and experience from making in-depth modifications and optimizations to libc++.
