[PATCH] D27825: [AVX-512][InstCombine] Teach InstCombine to converted masked vpermv intrinsics into shufflevector instructions

[Craig Topper via Phabricator via llvm-commits]

craig.topper added a comment.

I think in the worst case we may end up selecting a shuffle with large element size and end up being unable to combine the mask into the instruction. But we've likely removed a constant pool load so that may still be a win.


https://reviews.llvm.org/D27825