Attached is a benchmark which I run on top of a Valhalla branch on which I added support for latest JEP 419 API. The branch can be found here:

https://github.com/mcimadamore/valhalla/tree/valhalla%2Bpanama

The benchmark results are somewhat surprising: using jdk/jdk numbers are much better than when using valhalla (note that the benchmark does not use primitive classes).

Also, using `-prof gc` reveals that there's a lot more allocation in valhalla than in jdk/jdk:

jdk/jdk

Benchmark Mode Cnt Score Error Units
PointerBenchmarkSimple.testPointerPointerSimpleInt avgt 30 1.562 ± 0.025 ms/op
PointerBenchmarkSimple.testPointerPointerSimpleInt:·gc.alloc.rate avgt 30 ≈ 10⁻³ MB/sec
PointerBenchmarkSimple.testPointerPointerSimpleInt:·gc.alloc.rate.norm avgt 30 1.348 ± 0.054 B/op
PointerBenchmarkSimple.testPointerPointerSimpleInt:·gc.count avgt 30 ≈ 0 counts

valhalla

Benchmark Mode Cnt Score Error Units
PointerBenchmarkSimple.testPointerPointerSimpleInt avgt 30 5.149 ± 0.071 ms/op
PointerBenchmarkSimple.testPointerPointerSimpleInt:·gc.alloc.rate avgt 30 3711.721 ± 50.648 MB/sec
PointerBenchmarkSimple.testPointerPointerSimpleInt:·gc.alloc.rate.norm avgt 30 40001540.338 ± 95.530 B/op
PointerBenchmarkSimple.testPointerPointerSimpleInt:·gc.churn.G1_Eden_Space avgt 30 3696.540 ± 213.533 MB/sec
PointerBenchmarkSimple.testPointerPointerSimpleInt:·gc.churn.G1_Eden_Space.norm avgt 30 39838313.042 ± 2247813.921 B/op
PointerBenchmarkSimple.testPointerPointerSimpleInt:·gc.churn.G1_Survivor_Space avgt 30 0.029 ± 0.015 MB/sec
PointerBenchmarkSimple.testPointerPointerSimpleInt:·gc.churn.G1_Survivor_Space.norm avgt 30 310.224 ± 166.228 B/op
PointerBenchmarkSimple.testPointerPointerSimpleInt:·gc.count avgt 30 169.000 counts
PointerBenchmarkSimple.testPointerPointerSimpleInt:·gc.time avgt 30 116.000 ms