Descend: A Safe GPU Systems Programming Language

Explore a 20-minute conference talk from PLDI 2024 introducing Descend, a novel safe GPU systems programming language. Discover how Descend addresses the challenges of programming massively parallel GPU hardware by enforcing safe memory management through ownership and lifetime tracking in the type system. Learn about Descend's holistic GPU programming model, which hierarchically schedules computations over grid, blocks, warps, and threads. Understand how extended Borrow checking and the concept of views ensure safe parallel access patterns, preventing data races. Compare Descend's approach to existing GPU programming languages like CUDA and OpenCL, and examine its potential for expressing real-world GPU programs with competitive performance while maintaining strong safety guarantees. Gain insights into the future of safe GPU programming and its implications for developing efficient, bug-free parallel applications.

[PLDI24] Descend: A Safe GPU Systems Programming Language