Reese Levine

Santa Cruz, CA

I am a sixth (and final) year PhD candidate and an NDSEG Fellow at UC Santa Cruz, where I work with Tyler Sorensen and am generally affiliated with the LSD Lab (I also co-organize the LSD seminar). My research interests are in parallel and concurrent programming. My work has focused on GPU programming models, including designing and evaluating techniques to test the conformance of compilers and hardware to memory model specifications [1, 2], and testing and improving the safety properties of GPU programming languages in the face of data races [3].

I am currently collaborating on projects related to GPU security and compiler correctness. I am also actively contributing to llama.cpp in order to bring GPU-accelerated inference to the browser using WebGPU, opening up research opportunities in portable performance, efficient LLM quantization, and improving GPU programming models.

I am currently exploring full-time opportunities! I am interested in working on GPU (or other accelerator) programming languages and architecture, AI infrastructure (especially but not limited to local LLM inference), or other roles related to parallel and concurrent programming. Please reach out if you have an opening or want to learn more about my research.

news

Aug 15, 2025	Our work on assessing and addressing WebGPU memory safety in the presence of data races, SafeRace, was accepted to OOPSLA 2025.
Jun 06, 2025	Our SIGARCH blog on finding memory model errors in a recent GPU synchronization paper was released, highlighting opportunties for future research and collaboration.
Oct 31, 2024	I presented our work on testing memory consistency at Stanford University.
Feb 12, 2024	I presented our work on testing the Vulkan memory model at Vulkanised 2024.
Jan 15, 2024	I gave a talk on evolving weak memory models for evolving architectures at the Future of Weak Memory workshop at POPL 2024.