Hongtao Yu
Compiler Software Engineer at Facebook
Redmond, Washington, United States
Summary
🤩
Rockstar🎓
Top SchoolHongtao Yu is a compiler software engineer with a decade of experience building and optimizing compilers for large-scale server and ML platforms. Currently on Facebook's LLVM Server Compiler team, he focuses on code generation and profile-guided optimizations for production server workloads. Previously at Microsoft he contributed to VC++ and .NET AOT/JIT toolchains, bringing practical systems experience across both ahead-of-time and just-in-time compilation. An active open-source contributor, Hongtao has improved PyTorch’s inductor and Triton backends—optimizing GEMM kernels, reducing register pressure, and adding support for emerging numeric types like FP8 and BF16. He combines a strong academic foundation (PhD) with hands-on kernel-level performance tuning, and is known for pragmatic refactors that make complex numeric and view-handling logic both faster and easier to maintain. Based in Redmond, he blends research-driven techniques with production-ready engineering to push compiler performance on GPUs and servers.
10 years of coding experience
9 years of employment as a software developer
bachelor, Computer Science and Technology, bachelor, Computer Science and Technology at University of Science and Technology of ChinaDoctor of Philosophy (Ph.D.), Computer Science and Engineering, Doctor of Philosophy (Ph.D.), Computer Science and Engineering at Institute of Computing Technology, Chinese Academy of Sciences
English, Chinese
Github Skills (17)
auto-tuning10
pytorch10
c-language10
python10
llvm10
mlr10
triton10
gpu10
compiler-optimization10
cuda10
tensor10
amdgpu10
cprogramming-language10
compiler-development9
machine-learning9
Programming languages (6)
Github contributions (5)
triton-lang/triton
Sep 2023 - Mar 2025
Development repository for the Triton language and compiler
Role in this project:
Backend Developer
Backend DeveloperContributions:127 reviews, 98 PRs, 48 pushes in 1 year 6 months
Contributions summary:Hongtao focused on backend development, specifically contributing to the Triton compiler. Their work included supporting new data types (Fp8E4M3Nv, Bf16) and related conversions within the compiler's backend code. They also implemented features to enable per-pass IR printing in the `triton-translate` tool, which aids in debugging and understanding the compilation process. Additional contributions included handling `AtomicCASOp` operations in GPU IR conversion and generalizing loop pipelining for improved performance.
compilerprogramming-languagecode-generationtriton
pytorch/pytorch
Feb 2021 - Feb 2021
Tensors and Dynamic neural networks in Python with strong GPU acceleration
Role in this project:ML Engineer & Performance Engineer
ML Engineer & Performance EngineerContributions:35 reviews, 1 commit, 23 PRs in 1 day
Contributions summary:Hongtao primarily focused on optimizing the performance of the PyTorch framework's inductor feature, contributing to the development and enhancement of the compiler. Their work includes refactoring code related to complex number handling and view operations. A significant portion of the commits revolves around optimizing Triton kernels and experimenting with autotuning techniques to improve matrix multiplication (gemm) performance and reduce register pressure for both NVIDIA and AMD GPUs, leading to notable compilation time and overall performance improvements.
pythongpu-accelerationdeep-learninggpunumpy
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.