> (As a side note, I would expect this conditional branch to slow down more code than it speeds up. But remember that compiler writers measure an "optimization"

> (As a side note, I would expect this conditional branch to slow down more code than it speeds up. But remember that compiler writers measure an "optimization" as successful if they can find any example where the "optimization" saves time.) Wildly false, and I have no idea where the author is getting this idea from. If you regress people's code in LLVM, your patch gets reverted.