Get the latest tech news

Optimizing a Rust GPU matmul kernel

I read the excellent post [Optimizing a WebGPU Matmul Kernel for 1TFLOP+

I abstracted the CPU-side code that talks to the GPU using generics and traits so I could easily slot in different kernels and their settings while writing the blog post. I abstracted the CPU testing harness code using generics and traits so I could easily slot in different kernels and their settings while writing the blog post. Leveraging standard tools like rustfmt minimizes cognitive overhead and avoids the hassle of configuring third-party formatters of varying quality.

Get the Android app