RFR(XS): 8250233: -XX:+CITime triggers guarantee(events != NULL) in jvmci.cpp

[Vladimir Kozlov]

We simply missing EnableJVMCI flag check! That is why JVMCI is not initialized. EnableJVMCI should be true in hosted mode.

I looked and I see few problematic places in compileBroker.cpp  statistic code guarded by #if INCLUDE_JVMCI but which 
does not check EnableJVMCI or comp->is_jvmci(). I think it should be fixed.

Also I think JVMCI_event_1 is useless now in print_compilation_timers() because its output does not go into tty.

Thanks,
Vladimir

On 7/27/20 12:51 PM, Tom Rodriguez wrote:
> Doug is away, but it seems like we could just remove the JVMCI_event_1 call in print_compilation_timers.  It's not a 
> particularly worthwhile notification and all the other events should be able to safely assume that JVMCI has actually 
> been initialized which I think was probably the point of the guarantee.  If that's not sufficient then we need to 
> convert that guarantee in a check for NULL and return.
> 
> tom
> 
> Vladimir Kozlov wrote on 7/27/20 12:35 PM:
>> Nope, the check is correct. In hosted mode Graal is used as Java application and not as JIT compiler so that 
>> UseJVMCICompiler flag is false.
>>
>> The problem is really caused by recent 8248321 changes - it is regression. Doug, please advice how to fix it.
>> I think the new JVMCI events code should be adjusted for hosted mode.
>>
>> Thanks,
>> Vladimir
>>
>> On 7/24/20 5:53 AM, Schmidt, Lutz wrote:
>>> Resending after updating subject line with bug id.
>>> Sorry for the spam.
>>> Lutz
>>>
>>> On 24.07.20, 14:51, "Schmidt, Lutz" <lutz.schmidt at sap.com> wrote:
>>>
>>>      Dear all,
>>>
>>>      may I please request reviews for this small fix? I would even say it is a trivial fix. It inverts an if 
>>> condition such that JVMCI specific code is called only when JVMCI compilation is enabled via UseJVMCICompiler.
>>>
>>>      Bug:    https://bugs.openjdk.java.net/browse/JDK-8250233
>>>      Webrev: http://cr.openjdk.java.net/~lucy/webrevs/8250233.00/
>>>
>>>      Local testing looks good. jdk/submit tests pending.
>>>
>>>      Thank you!
>>>      Lutz
>>>
>>>
>>>
>>>