[PATCH] D35320: [X86][SSE] Add support for extending bool vectors bitcasted from scalars.

[Simon Pilgrim via Phabricator via llvm-commits]

RKSimon added a comment.

In https://reviews.llvm.org/D35320#810639, @delena wrote:

> I'm thinking about another algorithm:
>  let's assume that we have sext (bitcast i8 to 8 x i1) to <8 x i16>
>
> I can do the following:
>
> 1. broadcast i8 over the xmm
> 2. "pand" with vector-mask (1, 2, 4, 8, 16 ..)
> 3. compare with zeroinitializer
>
>   If we have "zext" instead of "sext", we can put an additional shift-right at the end.


That makes sense, there may have to be special cases for older CPUs with weaker shift instructions but I'll see what I can do.


Repository:
  rL LLVM

https://reviews.llvm.org/D35320