RFR(L) 8205824: Update Graal

Vladimir Kozlov vladimir.kozlov at oracle.com
Wed Jun 27 23:50:50 UTC 2018


On 6/27/18 4:48 PM, Tom Rodriguez wrote:
> 
> 
> Vladimir Kozlov wrote on 6/27/18 2:10 PM:
>> Tom, can you tell about removal of 'trivial' code? Why it was removed?
> 
> It was somewhat ugly hack when originally done and it was supplanted by 
> the much nicer adjustCompilationLevel callback so it hasn't been in use 
> by Graal for almost two years.  Since we were breaking binary 
> compatibility it seemed like a good time to remove it.

Good.

Thanks,
Vladimir

> 
>>
>> Thank you for fixing our jtreg jvmci tests.
>>
>> New file should have Copyright year updated:
>> jdk.vm.ci.common/src/jdk/vm/ci/common/NativeImageReinitialize.java
> 
> I'll update it.
> 
>>
>> Otherwise looks good to me.
> 
> Thanks!
> 
> tom
> 
>>
>> Thanks,
>> Vladimir
>>
>> On 6/27/18 1:45 PM, Tom Rodriguez wrote:
>>> https://bugs.openjdk.java.net/browse/JDK-8205824
>>> http://cr.openjdk.java.net/~never/8205824/webrev/index.html
>>>
>>> This combines a general Graal update with the fix for 
>>> https://bugs.openjdk.java.net/browse/JDK-8204855.  This general Graal 
>>> update also fixes JDK-8204914 and includes a few other fixes from 
>>> Graal.   Unfortunately the webrev is fairly noisy because of blank 
>>> lines inserted because of copyright adjustment and the addition of 
>>> @Override annotations in JVMCI.
>>>
>>> 8204855 is a set of changes to the JVMCI API which are required to 
>>> support the libgraal effort.  Primarily this is about making a strong 
>>> distinction between runtime and compiler objects and state and hiding 
>>> anything which lets you bridge that boundary without be careful. 
>>> Mostly it doesn't change the way JVMCI works apart from some changes 
>>> to the HotSpotSpeculationLog implementation and the associated 
>>> changes in Graal.  These changes are binary incompatible with 
>>> previous versions of JVMCI.  Given the limitations of webrev I can't 
>>> easily provide a webrev which shows just this change.
>>>
>>> The bits were tested with mach5 
>>> builds-tier1,jdk-tier1,hs-tier1,hs-tier2,hs-tier3,hs-tier4-graal,hs-tier5-graal 
>>> and hs-tier6-graal.  The details are in the bug report.
>>>
>>> I'm happy to explain anything about changes the JVMCI and why they 
>>> were required.  The JVMCI API updates have been tested with JDK8 
>>> Graal and will land in the next few days.
>>>
>>> tom


More information about the hotspot-compiler-dev mailing list