sarah-ek
Refactor to add more type support
I needed support for 8/16 bit types and min/max for integers, but the changes quickly became unmanageable. So instead I tried reining in the code duplication a bit using some paste macros so it's easier to add more types to existing operations, or add more simple ops. I managed to add a lot of new type support while shrinking code size, and I used cargo-semver-checks to ensure no breaking changes were added.
One thing I'm unsure about are the 8 and 16 bit masked loads - there are several approaches to solving this without AVX-512, but they each have performance tradeoffs. So for the time being, I have them fall back to the scalar implementation on all targets except V4.
sorry for not responding earlier. this looks pretty interesting and i wanna take some time to get a closer look at it
sorry for not responding earlier. this looks pretty interesting and i wanna take some time to get a closer look at it
Of course, it is a pretty massive change. So take your time, I can work off a branch for now. I do have a ton of other ops I also need added on another branch and the macros made that so much less painful than having to do it all manually. So while this one only adds a couple of new functions, I had enough time using the macros to verify it eases maintenance burden a lot in the long run.