RFR: JDK-8262994: Refactor String.split to help method inlining
Sergey Tsypanov
stsypanov at openjdk.org
Wed Dec 28 20:44:48 UTC 2022
On Tue, 27 Dec 2022 20:12:51 GMT, Christian Wimmer <cwimmer at openjdk.org> wrote:
> The method `String.split` contains a fast-path when the regular expression parameter is not really a regular expression, but just a single split character.
> This fast path vs. slow path check can be constant folded when the regular expression parameter is a literal constant - a quite frequent pattern (for example, all JDK usages of `String.split` have a constant expression parameter). But method inlining in JIT and AOT compilers can usually not inline `String.split` because the method body is too large. Factoring out the actual fast-path splitting logic into a separate method solves this problem: the JIT or AOT compiler can inline `String.split`, constant-fold the fast/slow path check, and then only the invoke of either the fast path or the slow path remains.
Is there any benchmark proving the benefit?
-------------
PR: https://git.openjdk.org/jdk/pull/11791
More information about the core-libs-dev
mailing list