[clang] [Clang][Lexer][Performance] Optimize Lexer whitespace skipping logic (PR #180819)

[Oliver Hunt via cfe-commits]

ojhunt wrote:

> > We should also check whether a loop that simply checks for all whitespaces is faster. All the non-' ' whitespaces are consecutive numbers together, so it boils down to a `(c >= '\t' && c <= '\r') || c == ' '` (though `\n` has special handling, so it might not be as perfect).
> 
> `isHorizontalWhitespace` only matches `' '`, `'\t'`, `'\f'`, `'\v'`. I tried replacing the conditions with
> 
> ```c++
> *CurPtr == ' ' || *CurPtr == '\t' || *CurPtr == '\f' || *CurPtr == '\v'
> ```
> 
> but it is slower, which is understandable seeing [what it optimizes to](https://godbolt.org/z/P6coT8nn5).

That code looks perfectly reasonable - you just need to look at it when its part of a loop post-inlining: https://godbolt.org/z/sTz88Y9f9


https://github.com/llvm/llvm-project/pull/180819