[PATCH] D55898: [x86] add movddup specialization for build vector lowering (PR37502)

[Sanjay Patel via Phabricator via llvm-commits]

spatel added a comment.

Proposal for load folding improvement is here:  
D55936 <https://reviews.llvm.org/D55936>

Also, I reviewed the diffs from:
https://bugs.llvm.org/show_bug.cgi?id=32558
...which was a perf reason for reverting the earlier and more general fix. This patch won't hit that same problem because we are checking that the 2 operands are different (if all 4 operands are the same, then we should defer that to more general splat lowering). AFAICT, the code for that benchmark is going to improve via this fix + the load folding...at least it looks better on paper, no telling what will happen in practice. :)


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D55898/new/

https://reviews.llvm.org/D55898