Something new: CubeCL, Writing Pure Rust GPU Kernels.

This stream is mostly me exploring a new Rust crate I came across. Since the project is quite interesting, I recorded it so I can revisit it in the future. P.S.: The audio quality isn't the best, so you might want to turn up the volume. 00:00:00 Intro to CubeCL 00:05:45 CubeCL uses proc-macros to annotate kernels 00:15:18 Auto-vectorization, Auto-tuning, Comptime 00:20:15 What is CubeCL at its core - a JIT compiler 00:26:09 CubeCL basically has a rust front-end for its Compiler 00:27:00 Wrap up Apple Metal Matrix Multiply (tiling) Kernels: https://github.com/nihalpasham/mtl #compiler #gpu #rust #heterogeneous #performance #machinelearning