RFR: 8287567: AArch64: Implement post-call NOPs

[Dean Long]

On Wed, 1 Jun 2022 07:32:35 GMT, Dean Long <dlong at openjdk.org> wrote:

>>> Is there any advantage to having the first instruction be "nop" instead of a branch that skips over the movks?
>>> 
>>> ```
>>> b done
>>> movk zr, lo
>>> movk zr, hi
>>> done:
>>> ```
>>> 
>>> I was going to suggest:
>>> 
>>> ```
>>>  b done
>>>  (raw data)
>>>  done:
>>> ```
>>> 
>>> but then it might be hard for NativePostCallNop::check() to prevent false positives.
>> 
>> The NOP is required for deoptimization.
>
>> The NOP is required for deoptimization.
> 
> I could be wrong, but I thought aarch64 allowed undefined instructions to be patched on top of branch instructions too.

> @dean-long , are you happy with that explanation?

Yes, and thanks for the interesting performance data.  I had an ulterior motive for my question.  I was hoping there would be a platform-independent pattern that new ports could use initially, then optimize later. I think the

 b done
 (raw data)
 done:

pattern will work as long as call sites without a post-call NOP disambiguate by using a regular NOP.  Then NativePostCallNop::check() will only see either a NOP or a branch.

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

PR: https://git.openjdk.java.net/jdk/pull/8955