[PATCH] D134438: POC patch to demonstrate how new intrinsics for interleaved load/store could be used in LoopVectorize

[Philip Reames via Phabricator via llvm-commits]

reames added a comment.

Interesting!  This is a different approach than I'd been considering.

I had been picturing adding intrinsics for each of the shuffles used during interleave lowering.  All three have reasonable lowerings on RISCV - not sure about SVE.  Your approach would make matching segmented load/store easier, but is a larger divergence from the way we currently handle fixed length interleaves.

On the other hand, I'd been debating something similar to this approach for strided load/store intrinsics.  Would you have some time to chat on a call and debate approaches here?


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D134438/new/

https://reviews.llvm.org/D134438