[PATCH] D149749: [AArch64][SVE] Custom ISelLowering for 256b `shuffle_vector v, undef, <1, 1, 0, 0>`

[Eli Friedman via Phabricator via llvm-commits]

efriedma added a comment.

For NEON, we would use the PerfectShuffle tables for something like this... should we try to use those tables here?  I mean, I guess it's kind of narrow to implement perfect shuffle tables specifically for `<4 x double>`, but it might make sense...

We should probably also consider implementing a general-purpose fallback for shuffling that doesn't involve the stack.  For a shuffle with one source, we can use tbl; I guess for the general case we'd have to use tbl+tbl+orr.  (Sort of messy, but almost certainly better than the default fallback of storing to the stack element by element.)


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D149749/new/

https://reviews.llvm.org/D149749