getAndAddInt() breaks AOT compilation for AArch64

[stewartd.qdt]

I just uploaded a patch for this fix to https://github.com/oracle/graal/pull/446. 

It is essentially the same as Zhongwei proposed and does not contain any advanced handling for immediates. I think at this point simpler is better. It also appeared to me that Hotspot does not have special handling for the immediate cases either, so I chose the simple route. 

Please have a look and my apologies for the breakage.

Daniel

-----Original Message-----
From: Zhongwei Yao [mailto:zhongwei.yao at linaro.org] 
Sent: Wednesday, May 30, 2018 5:43 AM
To: dean.long at oracle.com
Cc: Andrew Haley <aph at redhat.com>; Daniel Stewart <stewartd at qti.qualcomm.com>; stewartd.qdt <stewartd.qdt at qualcommdatacenter.com>; graal-dev at openjdk.java.net
Subject: Re: getAndAddInt() breaks AOT compilation for AArch64

Hi, all,

Since Daniel is on holiday, here are my 2 methods to get it pass:

  - one is a quick fix for this error (the fix doesn't include optimization for special-case integer values as Andrew mentioned).

  - and another is a workaround which disables "registerUnsafeReadAndAddPlugins" completely.

Here are the patches: http://cr.openjdk.java.net/~zyao/getAndAddInt/

Both of them are OK when running "jaotc --output libjava.base.so --module java.base" and pass Graal unit tests.

Could you take a review?

And what shall I do if the patch is OK? Shall I create a pull request in Graal for it? Or should we just wait for Daniel's full fix for it?


On 30 May 2018 at 06:51,  <dean.long at oracle.com> wrote:
> On 5/29/18 9:57 AM, Andrew Haley wrote:
>>
>> On 05/29/2018 05:45 PM, Daniel Stewart wrote:
>>>
>>> Unfortunately I'm on holiday until Friday and have no access to be 
>>> able to patch this. I'll work on it once I'm back if someone doesn't 
>>> get to it first.
>>>
>>> Sorry for the breakage.
>>
>> It's OK, we can work around it for now.  Better that you get to fix 
>> it yourself.  And if someone reading this is thinking about importing 
>> Graal to OpenJDK, this would not be a good time.
>> :-)
>>
>
> Can we include a work-around in the Graal import rather than delaying it?
>
> dl



--
Best regards,
Zhongwei