[PATCH] D42031: [X86] Legalize v32i1 without BWI via splitting to v16i1 rather than the default of promoting to v32i8.

[Simon Pilgrim via Phabricator via llvm-commits]

RKSimon added a comment.

> There are still some regressions in here. I definitely saw some around shuffles. I think we probably should move vXi1 shuffle from lowering to a DAG combine where I think the extend and truncate we have to emit would be better combined.

I've been putting off looking at this but I've been thinking about ways to improve shuffle combining - 2 things come to mind: some form of DemandedVectorElts (preferably in DAGCombine but just in x86 if necessary) and the ability for the x86 shuffle combines to combine through subvector insertion/extraction/concat (and avx512 compress/expand?). Supporting extend/trucate could fit in with the second.


Repository:
  rL LLVM

https://reviews.llvm.org/D42031