RFR(S): 8190432 [MVT] Value Types should have a default pre-allocated

[Frederic Parain]

Thank you Tobias,

Here’s an updated webrev:

http://cr.openjdk.java.net/~fparain/8190432/webrev.01/index.html

It includes all the changes from your review as well as the fix for
the build issue. I’ve tested builds on all platforms with JPRT.

Regards,

Fred


> On Nov 1, 2017, at 09:35, Tobias Hartmann <tobias.hartmann at oracle.com> wrote:
> 
> Hi Fred,
> 
> just hit this build problem on JPRT when trying another patch on top:
> 
> /opt/jprt/T/P1/120645.tobias/s/src/hotspot/share/memory/oopFactory.cpp: In static member function 'static arrayOopDesc* oopFactory::new_valueArray(Klass*, int, Thread*)':
> /opt/jprt/T/P1/120645.tobias/s/src/hotspot/share/memory/oopFactory.cpp:112:45: error: invalid conversion from 'oop' to 'instanceOop' [-fpermissive]
>   instanceOop value = vklass->default_value();
> 
> Thanks,
> Tobias
> 
> On 01.11.2017 09:15, Tobias Hartmann wrote:
>> Hi Fred,
>> this looks good to me! Some style comments:
>> classFileParser.cpp
>> - line 1724: whitespace between "index" and "++"
>> - line 5770: you can merge the else and the if below
>> interpreterRuntime.cpp
>> - line 220 and 359: Wouldn't it make sense to move the asserts into ValueKlass::default_value()? Because right now you are only checking at vwithfield and vdefault.
>> I've filed JDK-8190458 for the C2 changes.
>> Best regards,
>> Tobias
>> On 31.10.2017 18:35, Frederic Parain wrote:
>>> Please review this small changeset to pre-allocate a default value:
>>> 
>>> Bug:
>>> https://bugs.openjdk.java.net/browse/JDK-8190432
>>> 
>>> Webrev:
>>> http://cr.openjdk.java.net/~fparain/8190432/webrev.00/index.html
>>> 
>>> The changeset injects a static field into each Java mirror of a
>>> value type to keep a reference to the pre-allocated value.
>>> 
>>> All hotspot_valhalla tests pass.
>>> 
>>> Thank you,
>>> 
>>> Fred