On Fri, 27 Mar 2026 06:35:40 GMT, Stefan Karlsson <[email protected]> wrote:
>> src/hotspot/share/prims/jvm.cpp line 564:
>>
>>> 562: Klass* klass = o->klass();
>>> 563:
>>> 564: return klass->is_flatArray_klass();
>>
>> Given the definition:
>>
>> #ifdef _LP64
>> bool oopDesc::is_flatArray() const {
>> markWord mrk = mark();
>> return (mrk.is_unlocked()) ? mrk.is_flat_array() :
>> klass()->is_flatArray_klass();
>> }
>>
>> I think this needs to retain the original form and not call
>> `is_flatArray_klass` directly.
>
> I don't think it does. The mark word is just another way to query the
> information that we get from by querying the class.
>
> In fact, I have a prototype that completely removes these `functions` from
> the mark word and let all C++ code go through the Klass pointer. This cuts
> down on the complexity and some of the risks of those bits being unstable
> while inflating object monitors. Left are the JIT compilers using those bits,
> and having to navigate around the locking code.
If the markword logic is going away then sure. I have to query why we have/had
two ways of querying this though? Optimization?
-------------
PR Review Comment:
https://git.openjdk.org/valhalla/pull/2261#discussion_r3000690165
