[PATCH] D20645: Avoid doing binary search.

[Rui Ueyama via llvm-commits]

ruiu added a comment.

I was surprised you said 5% overall improvement isn't worth it. The number seems really significant to me. It is not a microbenchmark but is a real result of linking clang. Just +41 lines improves the overall link time by 5%? To me, it is between "excellent" and "it's too good to be true so I'll apply this and verify it myself."

You can see the drop of number of branch mispredictions in the perf numbers.


http://reviews.llvm.org/D20645