Alphatensor matrix multiplication algorithms

[jamen]

You've probably seen Alphatensor discovering new matrix multiplication algorithms. Is there interest in porting these to glam?

[bitshifter]

I hadn't seen that, very interesting. I would potentially be interested in seeing the scalar implementations. They may or may not translate well to SIMD. I only skimmed the paper but I wonder if they took floating point accuracy into account when coming up with new algorithms.

[jamen]

Forgot to link https://github.com/drbh/simd-alphatensor-rs. Here is mat4x4 * mat4x4

[bitshifter]

Hmm, they're only benchmarking with i32, it would be interesting to see f32.