On Thu, 30 Mar 2023 20:06:02 GMT, Chen Liang <li...@openjdk.org> wrote:
> This patch optimizes ClassDesc.resolveConstantDesc for array classes. > Otherwise, the performance of reference class resolution remain unchanged. > https://jmh.morethan.io/?gist=111095eecf7a735fe8c96e4b952b58c1 It also > includes a benchmark for MethodTypeDesc resolution, which currently does not > yet depend on class desc resolution but can be later optimized if it does > switch. src/java.base/share/classes/java/lang/constant/ConstantUtils.java line 130: > 128: > 129: static boolean isPrimitiveArray(String > referenceTypeDescriptorString) { > 130: return > referenceTypeDescriptorString.charAt(referenceTypeDescriptorString.length() - > 1) != ';'; it also needs to check if it starts with `[` Suggestion: return referenceTypeDescriptorString.startsWith("[") && referenceTypeDescriptorString.charAt(referenceTypeDescriptorString.length() - 1) != ';'; test/micro/org/openjdk/bench/java/lang/constant/MethodTypeDescResolve.java line 53: > 51: @Fork(1) > 52: @State(Scope.Benchmark) > 53: public class MethodTypeDescResolve { Better to separate this benchmark from this PR. Probably the optimization done in [PR 10991](https://github.com/openjdk/jdk/pull/10991) should be a separate patch from JDK-8294961 and this benchmark could be included there. test/micro/org/openjdk/bench/java/lang/constant/ReferenceClassDescResolve.java line 120: > 118: return resolveReference(lookup, cd); > 119: } > 120: }, I think this benchmark doesn't need this case as that's the implementation of `resolveConstantDesc`. Was it there when you developed this fix? Same for MethodTypeDesc benchmark. ------------- PR Review Comment: https://git.openjdk.org/jdk/pull/13252#discussion_r1153874210 PR Review Comment: https://git.openjdk.org/jdk/pull/13252#discussion_r1153877556 PR Review Comment: https://git.openjdk.org/jdk/pull/13252#discussion_r1153876110
