RFR: 8326306: RISC-V: Re-structure MASM calls and jumps [v5]

[Fei Yang]

On Tue, 7 May 2024 07:10:21 GMT, Robbin Ehn <rehn at openjdk.org> wrote:

>> It's a bit annoying, since if we are about performance in the interpreter we should implement the math stubs, e.g. StubRoutines::dcos.
>
> I don't find a nice way to get li for them.
> https://github.com/openjdk/jdk/blob/3b8227ba24c7bc05a8ea23801e3816e8fc80de4e/src/hotspot/cpu/riscv/templateInterpreterGenerator_riscv.cpp#L272
> 
> Any suggestions?
> 
> As I said, if we want to be performant here we should implement the "StubRoutines::dpow()".
> 
> What you think?

Hmm .. It does look bad to me if we use mv instead of movptr like you do in your previous commit [1]. 
Seems that a mv is more reasonable than movptr in this case when addr is out of code cache, isn't it? 
And it's not solely about the currently call sites like in the interpreter, we should also consider future possible uses of `call`. 

[1] https://github.com/openjdk/jdk/pull/18942/commits/d8fbb00b51afbf6a3b7dcd828f623b4add949faf

-------------

PR Review Comment: https://git.openjdk.org/jdk/pull/18942#discussion_r1591939129