[PATCH] D86559: [Sema, CodeGen] Allow [[likely]] and [[unlikely]] on labels

[Mark de Wever via Phabricator via cfe-commits]

Mordante added a comment.

In D86559#2243575 <https://reviews.llvm.org/D86559#2243575>, @staffantj wrote:

> As one of the SG14 industry members driving this, I'm firmly in the camp that this is what we're expecting. In the first case the 1/2 case are "neutral". This is a very explicit, and local, marker. Anything else makes it so vague as to be unusable for fine tuned code.

Thanks for your interest and affirming this looks like the right path to take.

> I should also make the point that we are not talking about a feature that is expected, or indeed should be, used by anyone other than someone with an exceedingly good understanding of what is going on. This is not a "teach everyone about it, it's safe" feature. It's there to produce a very fine-grained control in those cases where it really matters, and where profiling-guided optimizations would produce exactly the wrong result. Using this feature should be an automatic "is this needed" question in a code review. It is needed sometimes, just rarely.

I think it's hard to predict how the feature will be used. For example if a well known C++ gives a presentation at a conference about the attributes it might be more used. Even though as humans we're bad at guessing the performance bottleneck in our code I still think there are cases where the attribute can be used without testing. For example when writing a cache in an application you can mark the cache-hit to be more likely. If that isn't the case there's no reason for adding a cache. (Of course it would still be wise to measure.)


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D86559/new/

https://reviews.llvm.org/D86559