Request for reviews (S): 6823453: DeoptimizeALot causes fastdebug server jvm to fail with assert(false,"unscheduable graph")

[Vladimir Kozlov]

Thank you, Tom

It passed CTW and JPRT, so I am going to push it.

Thanks,
Vladimir

Tom Rodriguez wrote:
> That looks much nicer.
> 
> tom
> 
> On Apr 2, 2009, at 7:32 PM, Vladimir Kozlov wrote:
> 
>> How about this
>>
>> http://cr.openjdk.java.net/~kvn/6823453/webrev.04
>>
>> I also replaced the call remove_globally_dead_node()
>> with igvn->replace_node(catchproj, phase->C->top()).
>> It can't be used for nodes with out edges (kills regions otherwise).
>>
>> Thanks,
>> Vladimir
>>
>> Tom Rodriguez wrote:
>>> I think this:
>>> +       Node* catchproj = 
>>> proj_out(0)->unique_out()->as_Multi()->proj_out(0);
>>> has to be coded more carefully.  That's the shape of a well-formed 
>>> fall through edge but there's no guarantee it's well formed.  It's 
>>> too bad the extract_call_projections logic isn't packaged up a little 
>>> differently.
>>> tom
>>> On Apr 2, 2009, at 3:39 PM, Vladimir Kozlov wrote:
>>>> New (fourth?) version of the fix suggested by Tom.
>>>>
>>>> I added the test case when array length is not defined
>>>> until loop optimizations. Only this version can compile it.
>>>>
>>>> http://cr.openjdk.java.net/~kvn/6823453/webrev.03
>>>>
>>>> Thanks,
>>>> Vladimir
>