[PATCH] D24338: [InlineCost] Remove CallPenalty and change MinSizeThreshold to 5

[James Molloy via llvm-commits]

jmolloy added a comment.

Hi,

I agree completely, and this is actually already the case. The argument setup is accounted for; currently *on top of* the call penalty.

I see the inline cost metric for performance being highly correlated to IR code size, as the big gains from inlining come from code simplification and constant propagation, which is reflected in number of resulting IR instructions.


Repository:
  rL LLVM

https://reviews.llvm.org/D24338