[PATCH] D37286: [AVX512] Correct isel patterns to support selecting masked vbroadcastf32x2/vbroadcasti32x2

[Ayman Musa via Phabricator via llvm-commits]

aymanmus added a comment.

That's right, but in the merge-mask version it doesn't really improve anything, you must have the mov instruction from xmm1 to xmm0, so folding the mask into the mov or the shuffle/duplicate is equivalent.
Only in the zero-mask version it can save us the last masked mov (if we fold the mask), but still IACA showed no throughput improvement.


Repository:
  rL LLVM

https://reviews.llvm.org/D37286