RFR: AArch64: JDK-8241665: Configuring --with-jvm-features=-compiler2 fails to build on AArch64

[Andrew Dinn]

On 28/03/2020 18:39, Andrew Haley wrote:
> On 3/28/20 4:20 PM, Andrew Dinn wrote:
>>
>> On 27/03/2020 20:00, Vladimir Kozlov wrote:
>>> Your fix looks fine to me. It matches guards in sharedRuntime_x86_64.cpp.
>>>
>>> About jvmci tests failures. I built jdk without C2 on x86 and reproduced
>>> failures you saw.
>>> I filed https://bugs.openjdk.java.net/browse/JDK-8241779
>>> It should not block your fix.
>>
>> Thanks very much for the review and for validating my test reports.
>>
>> Another review of the patch itself would probably be a good idea --
>> Andrew Haley, perhaps?
> 
> Looks good.
> 
> Out of curiosity, does anyone build this way?
Good question. Clearly, /I/ only did this to check Vladimir Ivanov's C2
changes. Given this has been broken on AArch64 since the first JVMCI
patch some years ago that's a definitive yes for anyone using AArch64.
howeer small a subset that may be.

Of course, if someone wants to use Graal as the JIT via JVMCI then
building with -compiler2 may be a very reasonable choice, especially if
they are using libgraal. That said, I suppose including compiler2 in
that use case probably only means more code/static data on disk since C2
will never get invoked. So, it's maybe far from crucial for performance.

regards,


Andrew Dinn
-----------
Senior Principal Software Engineer
Red Hat UK Ltd
Registered in England and Wales under Company Registration No. 03798903
Directors: Michael Cunningham, Michael ("Mike") O'Neill