The JDK makes use of isnanf for testing float NaN. musl doesn't provide that macro/function, and instead provides a unified macro isnan which handles both float and double arguments.

I updated the JDK code to use isnan instead of isnanf, but that change may be incorrect for other platforms. This needs to be verified, and corrected if wrong.

Note that this needs to be verified across all the relevant platforms, toolchains and toolchain versions in use.