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

[Keno Fischer via Phabricator via llvm-commits]

loladiro added a comment.

If a TTI hook is not the right way to go here, maybe we can find a way to actually express what we want here, maybe via metadata or some other specification on shufflevector that would tell the optimizer not to touch the shufflemask unless it's confident it can improve upon it. As it is, I see my CPU with a perfectly good shuffle instruction shuffle the same vector three times, because LLVM refuses to combine these shuffles - pretty embarrassing ;).


https://reviews.llvm.org/D31509