RFR(M) 8179100: [AOT] Rewrite redundant initialization elimination

[Igor Veresov]

Tom suggested to use data flow to do this. Otherwise I was missing the case when initialization happen on both paths of a diamond.
So, no scheduling this time, only walking the control flow.

Here’s the new review: http://cr.openjdk.java.net/~iveresov/8179100/webrev.01 <http://cr.openjdk.java.net/~iveresov/8179100/webrev.01>

igor

> On Apr 21, 2017, at 4:34 PM, Vladimir Kozlov <vladimir.kozlov at oracle.com> wrote:
> 
> Okay.
> 
> Thanks,
> Vladimir
> 
> On 4/21/17 3:57 PM, Igor Veresov wrote:
>> 
>>> On Apr 21, 2017, at 3:27 PM, Vladimir Kozlov <vladimir.kozlov at oracle.com> wrote:
>>> 
>>> On 4/21/17 1:55 PM, Igor Veresov wrote:
>>>> Redundant initialization elimination can be rewritten using the dominator tree visitor pattern. This makes code much more readable. It also now handle the case of the dominating initialization of a subclass. And since now we have more time the phase requires a unit test.
>>> 
>>> I don't see unit test.
>> 
>> Yeah, forgot it, webrev updated in-place.
>> 
>> igor
>> 
>>> 
>>>> 
>>>> Webrev: http://cr.openjdk.java.net/~iveresov/8179100/webrev.00
>>> 
>>> Seems good. But I assume Graal guys will review it too.
>>> 
>>>> (It’s a complete rewrite, so don’t look at the diffs, they are messy).
>>> 
>>> Thank you for warning ;)
>>> 
>>> Thanks,
>>> Vladimir
>>> 
>>>> 
>>>> I’ll be pushing this through graal-core and doing a graal update after that.
>>>> 
>>>> igor
>>>> 
>> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/attachments/20170425/6ae3d76d/attachment.html>