Guray Ozen   Compiler Research at NVIDIA Contact  Full Resume

Current Projects at NVIDIA (2025)

Guray currently works on two programming models based on MLIR compiler:

• Cutlass Python DSL: Cutlass 4.0 introduced CuTe Python DSL. We designed this for writing high-performance GPU kernels, integrated with CUTLASS and designed for composable kernel development.
• [See more CuTe DSL]
• TileIR (coming soon): A next-generation MLIR-based IR for portable GPU programming with a focus on kernel composition and performance tuning across ML and HPC workloads.

Research

Guray Ozen is a compiler engineer in the Machine Learning Compiler team at NVIDIA. His current focus centers on optimizing compilers and programming languages for GPU utilization in machine learning (ML) and high-performance computing (HPC).

He has made key contributions to several production-grade compilers, including Clang, Flang, MLIR, IREE, and NVIDIA HPC (formerly PGI).

Previously, he was actively involved in language design for parallel programming models, such as OpenMP and OpenACC. He served as a voting member of the OpenMP Language Committee for NVIDIA and contributed extensively to the OpenACC language specification.

Work

• NVIDIA, Compiler Engineer, ML Compilers, May '24
• Google Research (DeepMind), Compiler Research Engineer, ML Compiler Systems Research, August '22 - '24
• Implemented NVDSL, a Python DSL built on MLIR, simplifying programming for NVIDIA Hopper GPUs while achieving cuBLAS-level performance
• Worked in IREE team
• NVIDIA, Compiler Engineer, NVIDIA HPC Compilers, GPU Group, '18 - '22
  • Worked on C, C++, and Fortran to PTX code generation compiler that accelerated programs on top500 supercomputers
  • Lead engineer of OpenMP GPU Offload Project. Fastest compiler for SPEC ACCEL® benchmarks
  • Implemented Standard Parallelism in Fortran for GPUs
• NVIDIA, Compiler Research Intern, NVIDIA HPC Compilers, GPU Group, '17
• IBM T.J. Watson Research Center, Short-term Researcher, Advanced Compilers Group, Yorktown Heights, New York, USA, '16
• Barcelona Supercomputing Center, Researcher Assistant, Barcelona, Spain, '13 - '17
• Veripark, Akbank Software Engineer, Istanbul, Turkey, '10 - '12

Education

• PhD in Computer Architecture, '18 (Excellent Cum Laude)  - BarcelonaTech, Universitat Politècnica de Catalunya (UPC), Spain.
• Compiler and runtime based parallelization and optimization for GPUs, Advisors: Jesus Labarta, Eduard Ayguade
• [PDF]
• MSc High-Performance Computing, '14  - BarcelonaTech, Universitat Politècnica de Catalunya (UPC), Spain
• BSc in Computer Science Engineering, '10  - Dokuz Eylul University, Turkey

Publications

• NVDSL: Simplifying Tensor Cores with Python-Driven MLIR Metaprogramming
  • ESFOMO@ICML'24, Guray Ozen
  • [PDF]
• Performant Portable OpenMP
  • CC22, Guray Ozen, Michael Wolfe
  • [PDF] [VIDEO]  [DOI]
• Non-Recurring Engineering (NRE) Best Practices: A Case Study with the NERSC/NVIDIA OpenMP Contract
  • SC'21, Christopher Daley, Annemarie Southwell, Rahulkumar Gayatri, Scott Biersdorff, Craig Toepfer, Guray Ozen, Nicholas Wright
  • [PDF]
• OpenMP GPU Offload in Flang and LLVM
  • LLVM-HPC@SC'18, Guray Ozen, Simone Atzeni, Michael Wolfe, Annemarie Southwell, Gary Klimowicz
  • [PDF]
• Collective Dynamic Parallelism for Directive Based GPU Programming Languages and Compilers
  • PACT'16. Guray Ozen, Eduard Ayguade, Jesus Labarta
  • [PDF] [DOI]
• Offloading Support for OpenMP in Clang and LLVM
  • LLVM-HPC@SC'16. Samuel Antao, Alexey Bataev, Arpith Jacob, Alexandre Eichenberger, Georgios Rokos, Matt Martineau, Tian Jin, Guray Ozen, Zehra Sura, Tong Chen, Hyojin Sung, Carlo Bertolli and Kevin O'Brien 
  • [PDF] [SLIDES] [DOI]
• Multiple Target Task Sharing Support for the OpenMP Accelerator Model
  • IWOMP'16. Guray Ozen, Sergi Mateo, Eduard Ayguade, Jesus Labarta and James Beyer
  • [PDF] [SLIDES] [DOI]
• Exploring Dynamic Parallelism in OpenMP 
  • WACCPD@SC'15. Guray Ozen, Eduard Ayguade, Jesus Labarta
  • [PDF] [SLIDES] [DOI]
• On the Roles of the Programmer, the Compiler and the Runtime System When Programming Accelerators in OpenMP
  • IWOMP'14. Guray Ozen, Eduard Ayguade, Jesus Labarta
  • [PDF] [SLIDES] [DOI]

Talks

• Bringing NVIDIA Blackwell support to LLVM and MLIR
  • EuroLLVM'25
  • [VIDEO] [Summary] [SLIDES]
• Zero to Hero: Programming Nvidia Hopper Tensor Core with MLIR's NVGPU Dialect
  • EuroLLVM'24
  • [VIDEO] [Summary] [SLIDES]
• Targeting NVIDIA Hopper using MLIR
  • LLVM-PERF @ CGO'24
  • [Summary] [SLIDES]
• Thinking OpenMP with NVIDIA HPC Compilers
  • NASA Advanced Supercomputing (NAS) Division, '21
  • [Summary] [PDF]
• Accelerating Applications for NERSC's Perlmutter Supercomputer Using OpenMP and NVIDIA's HPC SDK
  • GPU Technology Conference, GTC'21, Silicon Valley Christopher Daley, Guray Ozen
  • [Summary]
• Compiler and Runtime Based Parallelization and Optimization for GPUs
  • SC'18 Doctoral Showcase, Guray Ozen, Eduard Ayguade, Jesus Labarta
  • [Summary] [Poster] [PDF]
• Multi-target Task-Based Programming Model Exploiting OpenACC GPU Kernel
  • GPU Technology Conference, GTC'17, Silicon Valley
  • [SLIDES]

Services

• PC Member at Supercomputing Conference (SC'24)
• PC Member at International Conference on Parallel Processing (ICPP'25)
• PC Member at EuroLLVM (EuroLLVM'24, EuroLLVM'25)
• PC Member at ISC High Performance (ISC'19, ISC'24', ISC'25')
• PC Member at Accelerators and Hybrid Emerging Systems Workshop @ IPDPS ('19, '20, '21, '22, '23)
• PC Member at Extreme Heterogeneity Solutions @ PPoPP ('22, '23, '24)
• Represented NVIDIA at OpenMP Language Committee ('19 - '22)
• Active contributor OpenACC Language Committee ('17 - '22)

Patent

• Program code versions, G. Ozen, M. Wolfe, filed by NVIDIA 2022