On Sat, 18 Feb 2023 09:21:25 GMT, Eirik Bjorsnos <[email protected]> wrote:
> This PR suggests we can speed up `StringLatin1.regionMatchesCI` by applying
> 'the oldest ASCII trick in the book'.
>
> The new static method `CharacterDataLatin1.equalsIgnoreCase` compares two
> latin1 bytes for equality ignoring case. `StringLatin1.regionMatchesCI` is
> updated to use `equalsIgnoreCase`
>
> To verify the correctness of `equalsIgnoreCase`, a new test is added to
> `EqualsIgnoreCase` with an exhaustive verification that all 256x256 latin1
> code point pairs have an `equalsIgnoreCase` consistent with
> Character.toUpperCase, Character.toLowerCase.
>
> Performance is tested for matching and mismatching cases of code point pairs
> picked from the ASCII letter, ASCII number and latin1 letter ranges. Results
> in the first comment below.
src/java.base/share/classes/java/lang/CharacterDataLatin1.java.template line
181:
> 179: return ( U <= 'Z' // In range A-Z
> 180: || (U >= 0xC0 && U <= 0XDE && U != 0xD7)) // ..or
> A-grave-Thorn, excl. multiplication
> 181: && U == (b2 & 0xDF); // b2 has same uppercase
I'm curious if the order of comparisons could alter performance to a small
degree. For example, it might be interesting to compare various permutations
like below to short circuit reject unequal uppercased b2
Suggestion:
// uppercase b1 using 'the oldest ASCII trick in the book'
int U = b1 & 0xDF;
return (U == (b2 & 0xDF))
&& ((U >= 'A' && U <= 'Z') // In range A-Z
|| (U >= 0xC0 && U <= 0XDE && U != 0xD7)) // ..or
A-grave-Thorn, excl. multiplication
-------------
PR: https://git.openjdk.org/jdk/pull/12632
