Request for reviews (S): 6877254: Server vm crashes with "no branches off of store slice" when run with CMS and UseSuperWord (default)

[Changpeng Fang]

http://cr.openjdk.java.net/~cfang/6877254/webrev.01/

Updated with the test removed!

* if (mem != my_store) {*


Thanks,

Changpeng


On 09/08/09 21:34, Tom Rodriguez wrote:
>
> On Sep 8, 2009, at 6:30 PM, Vladimir Kozlov wrote:
>
>> I have concerns about StoreCMNode::Ideal(aidx) implementation.
>>
>> The result of memory_at() is never the same mergemem node.
>> I either mergemem->base_memory() or mergemem->in(aidx)
>> so your check will be always true.
>
> You're right but i think that just means we can drop that test.
>
>>
>> Also mergemem nodes could be chained when the result of memory_at()
>> is also mergemem.
>
> I figured Ideal would be called again if there was more work to do and 
> it should be a rare case anyway.  The mem of a StoreCM should only be 
> something besides a Store if the StoreCM is dead or if something weird 
> has gone on with cloning.  These changes are pretty conservative since 
> we don't have a clear set of rules for when a StoreCM is dead though I 
> think we could if we wanted to.
>
> tom
>
>>
>> You can look on methods in memnode.cpp which search through
>> mergemem for stores: step_through_mergemem(), find_previous_store().
>>
>> Thanks,
>> Vladimir
>>
>> Changpeng Fang wrote:
>>> Need additional reviews (Tom (never) proposed the implementation, so 
>>> I assume he is oked )!
>>> Comments are in the webrev.
>>> On 09/01/09 11:07, Changpeng Fang wrote:
>>>> http://cr.openjdk.java.net/~cfang/6877254/webrev.00/
>>>>
>>>> Thanks,
>>>>
>>>> Changpeng
>>> Thanks,
>>> Changpeng
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/attachments/20090910/4f9d9da9/attachment.html