[PATCH] D48789: [X86] Replace (32/64 - n) shift amounts with (neg n) since the shift amount is masked in hardware

[Nirav Dave via Phabricator via llvm-commits]

niravd added a comment.

I think you should just return immediately and let it get replaced the next time.

Alternatively since we can't find an example,should we just do the obviously safe fix and just fail out if the UpdateNodeOperands returns an existing node? UpdateNodeOperands doesn't modify anything when it finds an existing node so this is easy.

In https://reviews.llvm.org/D48789#1208150, @craig.topper wrote:

> So if we call ReplaceNode, do I still continue on to selecting that node? Which I think means we select that node earlier than its other users and prevents the users from seeing it as an 'and' that they can match in their patterns?





Repository:
  rL LLVM

https://reviews.llvm.org/D48789