[Mlir-commits] [mlir] [mlir][ArmSVE] add zip1 intrinsic (PR #79270)

[Cullen Rhodes]

c-rhodes wrote:

> > IIUC, that's not possible for SVE. From https://llvm.org/docs/LangRef.html#id189:
> 
> For SVE/scalable vectors there is `@llvm.experimental.vector.interleave2` which a slightly higher level abstraction for this: https://llvm.org/docs/LangRef.html#llvm-experimental-vector-interleave2-intrinsic

Oh cool, this is nicer, I'll use this intrinsic instead.

> Though the MLIR `vector.shuffle` operation currently cannot model a scalable zip, so in MLIR this would probably be `vector.scalable.interleave` (following from `vector.scalable.insert/extract`, which are also experimental LLVM intrinsics).

Sounds good 👍 I'll update the patches to use the target-agnostic interleave intrinsic, and we can look to add an op later.

https://github.com/llvm/llvm-project/pull/79270