On Wed, 15 Jun 2022 22:00:22 GMT, Paul Sandoz <psan...@openjdk.org> wrote:
> I would expect passing an non-constant exception mapper should work. The C2 > intrinsic never accesses the exception mapper argument, and a range check > failure results in a deoptimzation back to the interpreter from which the > exception mapper is then operated on. However, i think its worth just double > checking generated code. Both @iwanowww and @rwestrel have confirmed that that should be indeed the case. I did some test with PrintInlining with seemed to show the method as correctly inlined. For extra safety I will run all memory access benchmarks, and compare before/after. ------------- PR: https://git.openjdk.org/jdk19/pull/24
