Design meeting: SIMD multiversioning #309
Description
Summary
Multiple designs have been proposed for achieving the functionality of implementing function multiversioning based on SIMD instruction sets.
The main question I'd like answered in this meeting is what is the best way forward -- and if any have major blockers -- between
- ad-hoc, caller-based multiversioning ("target_features(caller)")
- generics based on witness types ("struct-target-features")
- const-generics based ("target_features(generic)")
If time permits, it also would be good to figure out more implementation details for the proposal(s).
More details can be found in the background reading document.
Background reading
https://hackmd.io/@veluca93/simd-multiversioning (and linked documents)
About this issue
This issue corresponds to a lang-team design meeting proposal. It corresponds to a possible topic of discussion that may be scheduled for deeper discussion during one of our design meetings.