As discussed in [1] [2], in java spec, there is no restrictions on how an NaN output should be based on the input NaN. That means an implementation can return any NaN and does not need to preserve payloads in the input NaN.
Till now, riscv intrinsics treat NaN specially and process them in a branch, which impacts the performance. It's worth to re-visit all the NaN related intrinsics on riscv, to check if it's necessary and maybe re-implement intrinsics and related tests.

[1] https://github.com/openjdk/jdk/pull/17450#pullrequestreview-1828999418
[2] https://mail.openjdk.org/pipermail/riscv-port-dev/2022-December/000706.html