On Fri, 8 Apr 2022 12:20:32 GMT, Claes Redestad <redes...@openjdk.org> wrote:
>> A few additional enhancements aiming to improve VH performance in the >> interpreter: >> >> - Flatten `TypeAndInvokers`: adds a pointer to `VarHandle` (a small increase >> 40->48) but removes an object and an indirection on any instance actually >> used - and might avoid allocating the `MethodHandle[]` unnecessarily on some >> instances >> - Have `checkExactAccessMode` return the directness of the `VarHandle` so >> that we can avoid some `isDirect` method calls. >> >> Baseline, `-Xint` >> >> Benchmark Mode Cnt Score Error Units >> VarHandleExact.exact_exactInvocation avgt 30 478.324 ? 5.762 ns/op >> VarHandleExact.generic_exactInvocation avgt 30 392.114 ? 1.644 ns/op >> VarHandleExact.generic_genericInvocation avgt 30 822.484 ? 1.865 ns/op >> >> >> Patched, `-Xint` >> >> Benchmark Mode Cnt Score Error Units >> VarHandleExact.exact_exactInvocation avgt 30 437.704 ? 5.320 ns/op >> VarHandleExact.generic_exactInvocation avgt 30 374.512 ? 3.154 ns/op >> VarHandleExact.generic_genericInvocation avgt 30 757.054 ? 1.237 ns/op >> >> >> No significant performance difference in normal mode. > > Claes Redestad has updated the pull request incrementally with one additional > commit since the last revision: > > Simplified as suggested by @ExE-Boss The changes look good. You might want to run few of the memory segment JMH benchmarks - especially the ones in "LoopOverNonConstant" and "TestAdaptVarHandles", to make sure they don't regress. ------------- Marked as reviewed by mcimadamore (Reviewer). PR: https://git.openjdk.java.net/jdk/pull/8160