RFR: 8239423: [TESTBUG] jdk/jfr/jvm/TestJFRIntrinsic.java failed with -XX:-TieredCompilation
David Holmes
david.holmes at oracle.com
Fri Feb 21 02:25:31 UTC 2020
On 20/02/2020 11:22 pm, Ao Qi wrote:
> On Thu, Feb 20, 2020 at 9:08 PM David Holmes <david.holmes at oracle.com> wrote:
>>
>> On 20/02/2020 8:33 pm, Ao Qi wrote:
>>>> Ah! Now I see the problem. In WhiteBox we have:
>>>>
>>>> if (comp_level > MIN2((CompLevel) TieredStopAtLevel,
>>>> CompLevel_highest_tier)) {
>>>> tty->print_cr("WB error: invalid compilation level %d", comp_level);
>>>> return false;
>>>> }
>>>>
>>>> but TieredStopAtLevel should be ignored when tiered compilation is
>>>> disabled, or not present. I actually think it is a bug that the VM
>>>> allows you to set TieredStopAtLevel even though it is ignored on those
>>>> configurations. The WB code should be changed to something like:
>>>>
>>>> #ifdef TIERED
>>>> CompLevel cl_min = MIN2((CompLevel) TieredStopAtLevel,
>>>> CompLevel_highest_tier);
>>>> #else
>>>> CompLevel cl_min = CompLevel_highest_tier;
>>>> #endif
>>>> if (comp_level > cl_min) {
>>>> tty->print_cr("WB error: invalid compilation level %d", comp_level);
>>>> return false;
>>>> }
>>>>
>>>> There are other places in WhiteBox that need similar fixes.
>>>
>>> Could you give some comments or advice on this one:
>>> http://cr.openjdk.java.net/~aoqi/8239423/webrev.01/ ?
>>>
>>> I think TieredStopAtLevel should be ignored when
>>> -XX:-TieredCompilation is used (even a build with tiered
>>> configration), so I used TieredCompilation to determine the cl_min,
>>> not "#ifdef TIERED". What do you think?
>>
>> Yes you are quite right! That is the correct way to handle this.
>>
>> I assume the test passes in all combinations now?
>
> Yes. I have tested all the combinations. Also tested tier1. No
> regression found so far.
We need to find all tests that use the modified WB functions and see if
they all pass under normal and non-tiered configurations.
Thanks,
David
> Cheers,
> Ao Qi
>
>>
>> Thanks,
>> David
>>
>>> Cheers,
>>> Ao Qi
>>>
>
More information about the hotspot-dev
mailing list