# Portable SIMD 2021-06-07 Meeting

###### tags: `Portable SIMD` `Minutes`

[Last Meeting] | [This Meeting's Chat] | [Next Meeting]

[Last Meeting]: https://hackmd.io/tRDC_oOBRCeZrvK7ZMuXNQ
<!-- [This Meeting's Chat]: -->
<!-- [Next Meeting]: -->

## Agenda

- Jacob: what did llvm-dev say?
- whoops! No updates so far.
- Jubilee: stability annotations?
- https://github.com/rust-lang/stdsimd/issues/70
- We probably want to use a subtree: https://github.com/git/git/blob/master/contrib/subtree/git-subtree.txt
- bjorn3: cranelift?
- No updates so far.
- scatter/gather?
- draft on `feat/gather`!
- do we know more about what we want from trig?
- not yet!
- to-nightly PR status?
- RFC draft status?
- is this even required first?
- does documentation for a vector of pointers proposal exist?
- no! :D
- some information was discovered: LLVM supports `getelementptr` (aka `gep`) for vectors-of-pointers, so adding that capability to the compiler as simd intrinsics is possible if we want.
- I wound up having to use casting pointers to ints and then back to pointers to get vectorized ops on pointers. That doesn't use `gep`, so may have weaker optimization.
- Does that impose an implementation burden of any kind on e.g. cranelift?
- Do we want to someday do this?
- do we have an example where rustc SIMD performance regresses relative to clang on x86-64, aarch64, or another architecture?

## Resolutions

0. It is prophesied that Jacob will remember to follow up on llvm-dev this time. :eyes:
1.