[PATCH] D19435: [LowerExpectIntrinsic] pin default likely/unlikely weights to min/max values

[Amaury SECHET via llvm-commits]

deadalnix added a comment.

This basically forces the branch to be as likely/unlikely as possible. I think this makes sense. One should not expect the compiler to do a good job when provided inaccurate infos.

As long as weight can be specified explicitly, legacy code that relied on the 64/4 behavior can continue to made to behave the way it is intended to. I'd give some time for @davidxl to comment, but as far as I'm concerned, this is for the best.


http://reviews.llvm.org/D19435