Marcello Maggioni
Software Engineer at Google DeepMind
Cupertino, California, United States
Summary
🤩
Rockstar🎓
Top SchoolMarcello Maggioni is a seasoned software engineer with 26 years of experience specializing in compiler back-ends and ML accelerator toolchains, currently at Google DeepMind in Cupertino. He has driven compiler and codegen efforts across Apple (including Apple Silicon GPU work), Google TPUs, and startups—co-founding Brium before its acquisition by AMD—bridging low-level runtime systems with high-performance hardware. A prolific contributor to open-source ML compiler projects like XLA and TensorFlow, he has implemented complex HLO ops, sharding fixes, and memory/latency optimizations that directly improve distributed and accelerator workloads. Marcello combines deep systems engineering with hands-on ML compiler design, and his career reveals a consistent focus on squeezing performance from emerging GPU/accelerator architectures.
26 years of coding experience
12 years of employment as a software developer
Athens Programme Data Compression, Athens Programme Data Compression at Czech Technical University in PragueMaster of Science (M.Sc.) Computer Engineering, Master of Science (M.Sc.) Computer Engineering at Politecnico di MilanoLiceo Scientifico
Italian, English, Spanish
Github Skills (16)
ml10
xla10
compiler10
machine-learning10
c-language10
cprogramming-language10
operation10
sharding10
hla10
distributed-computing10
linear-algebra9
computation9
numerical9
numeric9
compiler-design9
Programming languages (4)
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 & ML Engineer
Back-end Developer & ML EngineerContributions:127 commits in 2 years 7 months
Contributions summary:Marcello primarily contributed to the XLA (Accelerated Linear Algebra) compiler by adding support for new HLO (High-Level Operations) and expanding existing functionality. Their work involved implementing the kLogistic HLO, incorporating kCbrt as a transcendental operation, and adding an expander pass for the kLogistic function, including both TAHN and EXP-based expansion strategies. They also modified cost analysis and HLO parsing, as well as updating other parts of the project. The user seems to be adding features related to implementing various mathematical functions within the compiler.
compilercommunity-drivenmachine-learningmodular
tensorflow/tensorflow
Jun 2020 - Jan 2023
An Open Source Machine Learning Framework for Everyone
Role in this project:Back-end Developer
Back-end DeveloperContributions:3 reviews, 126 commits, 4 comments in 2 years 7 months
Contributions summary:Marcello primarily contributed to the XLA compiler, focusing on features related to sharding and distributed computation. Their work involved fixing bugs in sharding propagation, specifically addressing issues related to spatial dimensions and manual sharding in operations like GetTupleElement. Additionally, the user implemented and tested code to optimize the memory usage of HLO instructions. The user also improved and optimized the performance of various aspects of the compiler including improvements to the latency hiding scheduler.
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.