[PATCH] D114071: [X86][Costmodel] `getReplicationShuffleCost()`: promote 1 bit-wide elements to 8 bit when have AVX512BW+AVX512VBMI

[Roman Lebedev via Phabricator via llvm-commits]

lebedev.ri added a comment.

In D114071#3142360 <https://reviews.llvm.org/D114071#3142360>, @RKSimon wrote:

> LGTM (VBMI)

Thank you for the review!



================
Comment at: llvm/lib/Target/X86/X86TargetTransformInfo.cpp:3677
     }
     return bailout();
   default:
----------------
RKSimon wrote:
> lebedev.ri wrote:
> > RKSimon wrote:
> > > AVX512F will use (pretty awful) vXi32 shuffles: https://simd.godbolt.org/z/YYzjaf7Wh
> > Yes, those are pretty awful, i'm not sure if there's much hope for plain AVX512F,
> > we need AVX512BW or AVX512DQ for this i'd say.
> Its probably worth adding them instead of scalarization bailout though.
I mean, yes, it is just not obvious to me how to do that without hardcoding them.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D114071/new/

https://reviews.llvm.org/D114071