[PATCH] D30651: [InlineCost, -Oz] Don't take into account the penalty of a fast call of frequently used functions

[Evgeny Astigeevich via Phabricator via llvm-commits]

eastig added a comment.

In https://reviews.llvm.org/D30651#696071, @MatzeB wrote:

> > "Fast calls are usually optimized to minimize the call penalty, e.g. parameters are passed through registers."
>
> BTW: In terms of code size this statement isn't convincing to me as calls still impose constraints on the register allocator so we end up producing moves: Parameters need to go in specific registers, results come out in specific registers and live-through values have to be moved to callee-saved registers.


If this was an often case we would see code size increases in LNT.


https://reviews.llvm.org/D30651