Conditional compilation of ZGC

Stefan Karlsson stefan.karlsson at
Tue May 15 09:07:52 UTC 2018

Hi again,

To mimic the unconditional usages of names and causes by the other GCs, 
I removed the changes to gcCause and gcName, and left the collectedHeap 
enum. I ran this through tier1,tier2, and hs-tier3 testing.


On 2018-05-09 18:58, Stefan Karlsson wrote:
> Hi Stuart,
> On 2018-05-09 18:16, Stuart Monteith wrote:
>> Hi,
>>     I'm not a reviewer. I only have a couple of observations to make,
>> it otherwise looks OK, AFAICT.
> Thanks for taking a look at this!
> Hopefully it's straight forward to enable compilation of ZGC on AArch64 
> by adding a line to the section that sets up JVM_FEATURES_zgc.
>> src/hotspot/cpu/x86/
>>    Did you intend on changing all of the occurrences of "
>> predicate(UseAVX == 3); " into "predicate((UseAVX == 3));" in this
>> patch?
> IIRC, this was necessary to get the code to compile on OSX. I probably 
> should have extracted it as a separate patch.
>> src/hotspot/share/gc/shared/collectedHeap.hpp:
>>   You excluded "Z" from the enum here;
>> src/hotspot/share/gc/shared/gcName.hpp, should you do the same for
>> line 201 in src/hotspot/share/gc/shared/collectedHeap.hpp ?
> There is a hard-coded dependency between the C++ enum values in 
> CollectedHeap and the Java enum values in 
> so I've left Z in CollectedHeap, at least for now.
> Regarding the gcName.hpp. I think I might have to change that to always 
> include Z. There might be some interaction with JFR that requires all 
> values to be present. It's at least a bit odd that the other GC names 
> are not guarded by their respective INCLUDE_<GC> defines. I'll take a 
> look at this when I get back to the office next week.
> Cheers,
> StefanK
>> BR,
>>     Stuart
>> On 9 May 2018 at 14:15, Stefan Karlsson <stefan.karlsson at> 
>> wrote:
>>> Hi all,
>>> This is the first version of the patch to allow the compilation of 
>>> ZGC to be
>>> turned on and off. This patch passes tier1, tier2, and hs-tier3 
>>> (except for
>>> the test in gcbarriers that need to be updated). This mostly shows 
>>> that we
>>> can build ZGC on Linux x86_64 and Solaris Sparc, and that the other
>>> platforms still build without ZGC. It also shows that we don't break the
>>> other GCs.
>>> There are a few changes that we want to do for next version:
>>> 1) .ad file replace the fatal("ZGC specific code") with
>>> ShouldNotReachHere(), and maybe assert that we are running with ZGC.
>>> 2) Remove the INCLUDE_ZGC from gcCause and gcName, or add INCLUDE_<GC>
>>> guards for the other GCs.
>>> 3) c2_globals.hpp further cleanup. Probably move these flags into a 
>>> gc/z/c2
>>> directory.
>>> Cheers,
>>> StefanK

More information about the zgc-dev mailing list