[PATCH] D94708: [IR] Introduce llvm.experimental.vector.splice intrinsic

[Paul Walker via Phabricator via llvm-commits]

paulwalker-arm added inline comments.


================
Comment at: llvm/docs/LangRef.rst:16243
+The first two operands are vectors with the same type. The third argument
+``trailing.elts`` is an integer constant that specifies the number of trailing
+elements to extract from the first vector, if this exceeds the known minimum
----------------
vkmr wrote:
> Why is it required for the `trainling.elts` to be an immediate? I imagine there could be a scenario where this value is unknown at compile-time, for instance a recurrence where the order of recurrence is determined at runtime.
At this stage we don't want to muddy the waters by introducing new shuffle behaviours to LLVM, but rather just extend the existing shuffle requirements to cover scalable vectors.  Today `shufflevector` only supports a literal mask and thus at this stage we are enforcing this same requirement to the intrinsic variants.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D94708/new/

https://reviews.llvm.org/D94708