Stephan Herhut
Distinguished Compiler Engineer at NVIDIA
Paderborn, North Rhine-Westphalia, Germany
Summary
🤩
Rockstar🎓
Top SchoolStephan Herhut is a Distinguished Compiler Engineer with 12 years of experience building production-focused compilers, runtimes, and developer tools for machine learning, web, mobile and embedded systems. Now at NVIDIA after leading XLA CPU/GPU and MLIR GPU code generation efforts at Google, he combines deep expertise in compiler architecture, MLIR, LLVM integration and structured code generation with hands-on backend implementation. He has driven performance and size optimization across V8, Android R8, TensorFlow and IREE, and contributed key lowering and backend pipelines for GPU execution in open-source ML compilers. A pragmatic technical lead and mentor, he is known for establishing new teams and sites, improving debugging and JIT workflows, and pushing LLVM updates across large runtimes—bringing a research-trained perspective (PhD) to production compiler engineering.
12 years of coding experience
13 years of employment as a software developer
PhD Computer Science, PhD Computer Science at University of HertfordshireDiplom Informatiker Computer Science, Diplom Informatiker Computer Science at Kiel University
German, English
Github Skills (23)
c-language10
compiler-development10
llvm10
machine-learning10
mlr10
compiler-compiler10
deep-learning10
tensorflow10
gpu10
compiler10
xla10
cprogramming-language10
python9
bazel9
nvidia9
Programming languages (10)
Github contributions (5)
openxla/xla
Sep 2019 - Jan 2023
A machine learning compiler for GPUs, CPUs, and ML accelerators
Role in this project:
Back-end Developer & ML Engineer
Back-end Developer & ML EngineerContributions:12 reviews, 50 commits, 2 PRs in 3 years 5 months
Contributions summary:Stephan contributed to the development of the XLA compiler, specifically focusing on the MLIR GPU backend. Their work involved creating and integrating lowering pipelines to transform LHLO code into LLVM/NVVM dialects for GPU execution. The user added support for constant values, reduce operations, and fusion regions within the HLO and LHLO dialect emitters. They also implemented code to handle scalar arguments and rewrite kernel signatures for optimized execution.
compilercommunity-drivenmachine-learningmodular
tensorflow/tensorflow
Apr 2019 - Jan 2023
An Open Source Machine Learning Framework for Everyone
Role in this project:Back-end Developer
Back-end DeveloperContributions:80 reviews, 335 commits, 6 PRs in 3 years 9 months
Contributions summary:Stephan primarily contributed to the TensorFlow project by implementing and modifying code related to the GPU backend and MLIR (Multi-Level Intermediate Representation) compilation pipeline. Their work included adding functionality to dump intermediate MLIR passes for debugging, registering a diagnostic handler for error reporting during lowering, and optimizing the buffer packing transformation. Additionally, the user refactored kernel generation code and modified the JIT compilation process. This indicates a focus on improving the performance and debugging capabilities of the TensorFlow framework, specifically on GPU hardware.
pythondata-sciencedeep-learningmlmachine-learning
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.