Resource-Aware GPU Programming
Overview
General-purpose GPU (GPGPU) programming is becoming increasingly common in applications such as data science and machine learning. The memory and performance models of GPUs, however, can be counterintuitive for most programmers, leading to inefficiencies and potentially even performance-related security vulnerabilities. This project aims to build tools to analyze, and automatically optimize, GPGPU programs, as well as languages and frameworks to help programmers to develop more efficient GPGPU programs.
People
- Mark Lou
- Deepika Padmanabhan
Collaborators
Funding
NSF CCF-2007784: Automatic Qualitative and Quantitative Verification of CUDA Code