Rahul Joshi
Principal Compiler Engineer at NVIDIA
California, United States
Summary
🤩
Rockstar🎓
Top SchoolRahul Joshi is a Principal Compiler Engineer based in California with over two decades of deep compiler and GPU systems experience and six years focused on ML compiler toolchains. He has driven backend compiler work at NVIDIA and scalable XLA/GPU efforts at Google, and now shapes compiler/runtime strategy at NVIDIA for modern ML workloads. Rahul’s open-source contributions to flagship projects like TensorFlow, OpenXLA and IREE show specialization in MLIR, HLO/XLA consistency checks, BEF conversions and GPU kernel argument handling that improve both correctness and performance. He has a track record of eliminating legacy cruft and modernizing IR/codegen (e.g., removing old HLO reliance and adopting variadic patterns) to reduce complexity and compile time. Comfortable across research and production—having published and implemented compiler optimizations since his UIUC research days—he blends low-level GPU ABI knowledge with practical ML runtime engineering. Colleagues rely on him for subtle correctness fixes in sharding and control-flow that prevent brittle distributed-execution bugs.
6 years of coding experience
22 years of employment as a software developer
BE Computer Engineering, BE Computer Engineering at COEP Technological UniversityMS Computer Science, MS Computer Science at University of Illinois Urbana-Champaign
Github Skills (24)
c-language10
python10
operation10
testing10
mlr10
distributed-computing10
compiler-compiler10
tensorflow10
parallel-computing10
hla10
xla10
compiler10
jax10
cprogramming-language10
test-automation10
Programming languages (6)
Github contributions (5)
openxla/xla
Jun 2020 - Jan 2023
A machine learning compiler for GPUs, CPUs, and ML accelerators
Role in this project:
Back-end Developer
Back-end DeveloperContributions:62 reviews, 236 commits, 81 comments in 2 years 7 months
Contributions summary:Rahul's contributions primarily involved enhancing the XLA compiler, with a focus on improving code generation and efficiency. Their work included eliminating unnecessary code segments, adopting variadic programming techniques, and fixing issues in slice operations. Furthermore, the user was responsible for updating the system to handle more modern MLIR concepts, and they improved the handling of data types by improving the HLO code. Additionally, the user removed reliance on legacy HLO features, such as those found in the custom call and tuple representation.
compilercommunity-drivenmachine-learningmodular
tensorflow/tensorflow
May 2020 - Jan 2023
An Open Source Machine Learning Framework for Everyone
Role in this project:Back-end Developer
Back-end DeveloperContributions:48 reviews, 375 commits, 1 PR in 2 years 8 months
Contributions summary:Rahul's contributions focused on improving sharding consistency within the XLA compiler, specifically for control flow instructions like while loops and conditionals. They implemented verification checks to ensure data sharding is maintained across different operations within these control flow constructs. The changes primarily involved modifying the HLO verifier to enforce consistency across instructions in the XLA compiler which likely increases code stability. They also made additional changes to handle all-reduce operations and the dataflow analysis code.
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.