[PATCH] D31509: [InstCombine] Combine vector shuffles if the same operand can be reused

[Sanjay Patel via Phabricator via llvm-commits]

spatel added a comment.

In https://reviews.llvm.org/D31509#723393, @jonpa wrote:

> Comparing to last year, I see that there is a smaller problem with shuffles today. It also seems they may have gotten worse recently. Does anybody know what the causes might be here?


Can you post examples/bug reports of the IR that you are seeing? I have a draft of a patch that appears to solve this case and the case in PR30630. It's an instsimplify patch, so it will avoid controversy about target costs and difficult intermediate shuffle masks...although it may be masking the root cause in the vectorizers if that's where these chains of useless shuffles are originating.


https://reviews.llvm.org/D31509