[PATCH] D158195: [RISCV] Combine (vrot{l,r} vxi16, 8) -> vrev8

[Philip Reames via Phabricator via llvm-commits]

reames added a comment.

In D158195#4595909 <https://reviews.llvm.org/D158195#4595909>, @luke wrote:

> In D158195#4595901 <https://reviews.llvm.org/D158195#4595901>, @craig.topper wrote:
>
>> Do we have reason to believe vrev8 is better than vror?
>
> @reames alluded to it here: https://reviews.llvm.org/D157417#inline-1529308
> Tangentially, DAGCombiner canonicalises rotr/rotl to bswap anyway so this brings fixed-length vector behaviour more in line with scalable

For context, I do *not* think there's a performance difference.  This was mostly a canonicalization thing.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D158195/new/

https://reviews.llvm.org/D158195