Gus Smith

I think we could incorporate some quantization-like experiments in Glenside. We could keep sample values in the analysis (like Ruler's cvec). We could then have rewrites that quantize/manipulate the program...

there are some weird ops that will take some time to support

Clippy returns a ton of warnings. Go through and clean them up.

This was an update in egg's API that I missed. We need to explicitly union eclasses in `apply_one` in some cases (specifically the cases where we're returning an explicitly-constructed `vec![..]`.)

Glenside might translate well to some of the dialects of MLIR (e.g. affine).

Just as we implement matmul blocking, we should also implement conv2d channel splitting

This has come up multiple times. You want to avoid layout transformations when going back and forth to accelerators. I just haven't found a good model to do an experiment...