Home

Giorgis Georgakoudis

Principal Computer Scientist, LLNL

Compiler engineering, JIT compilation, Python for HPC, and parallel programming systems.

I build compiler and runtime systems for high-performance and heterogeneous computing, with emphasis on Clang, LLVM, MLIR, OpenMP, and Numba. My work spans from JIT compilation and Python compiler stacks to runtime specialization and optimization for CPU, GPU, and accelerator platforms, with the goal of improving performance portability and developer productivity for large-scale parallel applications.

Focus Areas

• Compiler infrastructure and optimization with LLVM, Clang, and MLIR
• JIT compilation and runtime specialization for heterogeneous systems
• Python and parallel programming systems with OpenMP and Numba

Projects

Proteus

Proteus is a programmable JIT compiler for C and C++ applications that enables embedded JIT compilation and runtime specialization across CPU, CUDA, and HIP execution paths. Through annotations and programmatic APIs, it folds runtime values into generated code to unlock optimizations such as loop unrolling, constant propagation, and control-flow simplification beyond what static compilation can achieve.

• Main developer and Principal Investigator, directing a team of 6 researchers and software developers.

PyOMP

PyOMP extends Numba with OpenMP parallel programming abstractions for Python, including directives, runtime functions, and GPU offloading through target constructs. It provides a familiar OpenMP programming model for Python users while enabling JIT-compiled execution across CPU and GPU systems.

• Main developer and technical lead

Mneme

Mneme is a framework for recording and replaying GPU kernel executions as standalone, reproducible executables. Built around an LLVM instrumentation pass, it captures the execution context, LLVM IR, and required device memory state for each kernel, enabling isolated replay, compiler experimentation, and autotuning of launch parameters without modifying the original application workflow.

• Co-developer of Mneme, a GPU kernel recording and replay system for compiler-driven analysis and autotuning

OpenMP

• Chair of the OpenMP Python Language Subcommittee, announced by the OpenMP ARB in February 2026
• Contributor to LLVM and OpenMP compiler/runtime technology

Publications & Service

• Publications: Google Scholar
• Open-source: GitHub
• Professional: LinkedIn
• Curriculum Vitae:
  • 1-page (PDF)
  • Short (PDF)
  • Full (PDF)

Contact

Work

georgakoudis1@llnl.gov

This page reflects personal work and views, not official positions of LLNL.