Shading Languages & CubeCL: GPU Programming in Rust (Part 3)

This video picks up from where we left off in the last stream, providing a recap of various approaches to GPU programming, including: - Shading languages—WGSL (WebGPU), GLSL (OpenGL), HLSL (Direct3D), and MSL (Metal) - Single-source methods like CUDA, ROCm, and OpenCL We then shift our focus to CubeCL, a Rust-based GPU runtime, and explore how it enables writing GPU kernels directly in Rust. In this video, we'll cover: - Kernel Expansion and Definition: Expanding Rust functions into CubeCL IR - IR Compilation: Translating CubeCL IR to WGSL If you're interested in GPU programming, Rust, or learning more about CubeCL, this video is for you! GitHub: https://github.com/tracel-ai/cubecl Git Gist: https://gist.github.com/nihalpasham/570d4fe01b403985e1eaf620b6613774 #gpu #programming #rust #compiler #heterogeneous #coding #cuda #webgpu Estimated Timecode Breakdown: 00:00:00 – Recap: GPU Programming, Shaders, and Single-Source Concepts 00:02:16 – Recap: High-Level Overview of CubeCL, WebGPU, and CUDA Backends 00:03:03 – Recap: Writing Kernels in Rust – What Does It Look Like? 00:08:50 – Overview: CubeCL Kernel Compilation 00:11:28 – Quick Look: WGSL (WebGPU Shading Language) 00:15:16 – Kernel Expansion 00:39:05 – Kernel Definition 00:48:04 – Final Step: Translating Kernel Definition to WGSL 00:57:40 – Demo (Failed due to requiring Apple's SDK License terms acceptance post-update) 01:03:00 – Cube IR to WGSL: 1-to-1 Mapping Comparison 01:09:04 – Summary and Wrap-up