The code generated for vector bitcasts is wrong, as the generic should be `vecN<f32>` instead of `f32` There are also missing casts between many other types of numeric types, as well as casts between vec2h and u32, i32, etc...