Mars Saxman
Principal Compiler Engineer at IR Labs
Seattle, Washington, United States
Summary
🤩
RockstarMars Saxman is a Principal Compiler Engineer based in Seattle with nine years of focused experience in compilers and deep learning systems. He has progressed through roles at startups and industry leaders—most recently at IR Labs after compiler and deep learning engineering stints at RISC Zero, Intel, and Gensyn.ai—bringing production-grade performance and concurrency fixes to compute backends. His open-source work on PlaidML demonstrates practical backend refactoring and performance engineering, including replacing function-pointer complexity and adding a Boost thread pool to optimize llvm_cpu parallelism. Mars combines low-level systems expertise with applied ML infrastructure, making him comfortable tuning concurrency to physical cores and eliminating race conditions. Colleagues rely on him to simplify complex codepaths while squeezing measurable runtime improvements out of compiler backends. He often surfaces non-obvious wins by prioritizing maintainability alongside raw performance.
8 years of coding experience
20 years of employment as a software developer
Github Skills (10)
c-language10
backend10
cprogramming-language10
performance-optimization10
back-end-development10
multithreading10
llvm10
boost9
opencl8
metal8
Programming languages (2)
Github contributions (5)
plaidml/plaidml
Dec 2017 - Dec 2020
PlaidML is a framework for making deep learning work everywhere.
Role in this project:
Back-end Developer
Back-end DeveloperContributions:11 reviews, 182 commits, 53 PRs in 3 years
Contributions summary:Mars focused on refactoring and improving the backend of the PlaidML framework. This involved simplifying the backends by removing support for function pointers, which streamlined the code. The user also implemented performance optimizations by integrating a Boost thread pool for parallel execution on the llvm_cpu backend. Further work included fixing a race condition and ensuring that the llvm_cpu concurrency limited to the physical cores, making it more efficient.
pytorchtvmdeep-learningmachine-learningcompiler
mars1026/plaidbench
Feb 2018 - Feb 2018
Benchmarking Keras application network performance
Contributions:5 pushes, 1 branch in 1 day
stress-testingbenchmarkingnetwork-performancebenchmarkkeras-application
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.