[PATCH] D46326: ThinLTO+CFI: short-circuit direct calls to jump table entries

[Vlad Tsyrklevich via Phabricator via llvm-commits]

vlad.tsyrklevich added a comment.

Did you forget to update the change? I'll run it against the failing chrome tests after. CFI-icall presents a <=1% performance regression for chrome so it would be hard to pick up a difference. From my past look at it, most of the regression appeared to because of cfi-icall checks at every malloc/free (because we override their implementations and require a function pointer call.)


https://reviews.llvm.org/D46326