Have you tried the benchmark, make benchmark ITEM=string_casecmp_p.yml?

It doesn't seem significant differences.

compare-ruby: ruby 3.5.0dev (2025-06-24T10:29:39Z master 45a2c95) +PRISM [arm64-darwin24]
built-ruby: ruby 3.5.0dev (2025-06-25T13:41:55Z master 077dbb8) +PRISM [arm64-darwin24]

Are you sure you ran the benchmark against my branch? It looks like compare-ruby and build-ruby are both pointing to different commits on the master branch.

Here are my results when comparing my branch against Ruby 3.5.0preview1:

compare-ruby: ruby 3.5.0preview1 (2025-04-18 master d06ec25be4) +PRISM [arm64-darwin24]
built-ruby: ruby 3.5.0dev (2025-06-26T03:14:35Z optimize-string-ca.. 5bc4a83d30) +PRISM [arm64-darwin25]

My code is around 1.7× faster for short strings (36 characters). My version is less efficient for longer strings (1.23× slower for 360-character strings, 1.28× slower for 3,600-character strings, and 1.42× slower for 36,000-character strings). In my testing, the break-even point is around 180 characters. My intuition is that most string comparisons are for strings shorter than 180 characters, especially if those strings are all ASCII characters. Also, if the strings are not equivalent, the comparison will be faster, since it will short-circuit as soon as two characters are not the same.

I think this patch will make most programs run faster. That said, it is not a pure win, so I will understand if you don't accept it. Thank you for taking the time to consider it.