On Tue, 25 Feb 2025 02:45:46 GMT, Chen Liang <li...@openjdk.org> wrote:
> Remove the intrinsicData field. We can obtain this from the customized MH > when we spin ultra-customized lambda forms. In the long run, we aim to remove > this intrinsic if there is no regression for call site sharing. > > The existing tableSwitch combinator's LF is unnecessarily complex. This patch > also simplifies the tableSwitch combinator. > > Note that I was forced to add `@ForceInline` on immutable lists due to > regressions in `MethodHandlesTableSwitchRandom` with `sorted == true`, which > eliminates the regression. Otherwise, all benchmark results are the same. > Tested java/lang/invoke. src/java.base/share/classes/java/lang/invoke/InvokerBytecodeGenerator.java line 628: > 626: continue; > 627: } > 628: break; // Only inline target MHs if > this is customized I think this could be problematic, as we typically only customize the root method handle in a chain. So, if a table switch handle is used with another combinator, we will never benefit from the intrinsic. src/java.base/share/classes/java/lang/invoke/MethodHandles.java line 7664: > 7662: if (mh.type() != expectedType) > 7663: throw new IllegalArgumentException( > 7664: "Some targets do not have the expected type " + > expectedType + ": " + caseActions); I think we should avoid changing the exception messages here, since it's observable from the outside. FWIW, one of the downsides of only printing out a single method handle, is that a user can't tell which method handle in the list was problematic. ------------- PR Review Comment: https://git.openjdk.org/jdk/pull/23763#discussion_r1979460526 PR Review Comment: https://git.openjdk.org/jdk/pull/23763#discussion_r1979537776
