[PATCH] D105827: [SLP]Workaround for InsertSubVector cost.

[Simon Pilgrim via Phabricator via llvm-commits]

RKSimon added a comment.

In D105827#2875513 <https://reviews.llvm.org/D105827#2875513>, @ABataev wrote:

> In D105827#2875503 <https://reviews.llvm.org/D105827#2875503>, @RKSimon wrote:
>
>> LGTM - I'm wondering if SK_InsertSubvector is actually usable in most cases as a shuffle kind or whether we're better off replacing it with something like SK_WidenSubvector + SK_ConcatSubvectors patterns?
>
> Not sure if I understand it correctly but I think SK_InsertSubvector is better here, just need a proper cost for it.

My main annoyance with SK_InsertSubvector is that it doesn't match an actual pattern we can create with a single shufflevector instruction.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D105827/new/

https://reviews.llvm.org/D105827