Problems with vdefault/vwithfield

Frederic Parain frederic.parain at oracle.com
Mon Feb 6 15:09:35 UTC 2017


Handling of vnew and vwithfield are indeed incorrect.
Handling of vdefault of vdefault is correct.

You can push your code if you want, I'll fix vnew and vwithfield
later this week.

Thanks,

Fred

On 02/06/2017 09:35 AM, Frederic Parain wrote:
> I'll look at the issue today.
>
> Fred
>
> On 02/06/2017 08:00 AM, Tobias Hartmann wrote:
>> Hi,
>>
>> Mr. Simms pointed out that in GenerateOopMap::interp1() me may have to
>> pop arguments for vwithdefault. But then we would need to do the same
>> with vnew. I'm not too familiar with that code, Fred, could you have a
>> look?
>>
>> Thanks,
>> Tobias
>>
>> On 06.02.2017 13:39, Tobias Hartmann wrote:
>>> Hi,
>>>
>>> please review the following fix:
>>> http://cr.openjdk.java.net/~thartmann/valhalla/vt_prototype/webrev.06/
>>>
>>> We hit a ShouldNotReachHere in GenerateOopMap::interp1() with
>>> "unexpected opcode: 215" because the _vwithfield bytecode is not
>>> known. I fixed the same problem in
>>> MethodLiveness::BasicBlock::compute_gen_kill_single() and
>>> BCEscapeAnalyzer::iterate_one_block() and modified the tests to
>>> trigger this.
>>>
>>> Thanks,
>>> Tobias
>>>


More information about the valhalla-dev mailing list