RE: JavaMemoryModel: The range of memory models now in play

From: Bill Pugh (pugh@cs.umd.edu)
Date: Thu Aug 08 2002 - 17:29:17 EDT


Yes, the total order in the execution trace has to be consistent with
the order of actions in each thread of the program (i.e., with
program order).

        Bill

At 4:26 PM -0500 8/8/02, Sarita Adve wrote:
>Bill,
>
>> -----Original Message-----
>> From: owner-javamemorymodel@cs.umd.edu
>> [mailto:owner-javamemorymodel@cs.umd.edu] On Behalf Of Bill Pugh
>> Sent: Thursday, August 08, 2002 3:11 PM
>> To: yyang@cs.utah.edu; Sarita Adve
>> Cc: javamemorymodel@cs.umd.edu
>> Subject: JavaMemoryModel: The range of memory models now in play
>>
>
>
>> Jason and Sarita: do you agree that your models fit between A and B?
>
>This is correct for my model except for one technicality below, which
>may or may not matter depending on the direction you want to take this
>discussion:
>
>> Consider two programming models, A and B.
>>
>> In both models, you check the validity of a totally ordered
>> execution trace E.
>
>Saying only that an execution trace is totally ordered is somewhat
>"content-free" unless some properties are assigned to the total
>ordering. I believe for model B, you are implicitly assuming that the
>above total order is consistent with program order. For my model, I can
>also construct a total order where a read returns the value from a
>previous write. But this order isn't consistent with program order, and
>for hardware with caches and write buffers, this order would be on write
>suboperations rather than write operations, but I don't think you were
>considering such orders.
>
>Also, there is more similarity between your and my model, but I assume
>you will get to that in a subsequent message.
>
>Sarita

-------------------------------
JavaMemoryModel mailing list - http://www.cs.umd.edu/~pugh/java/memoryModel



This archive was generated by hypermail 2b29 : Thu Oct 13 2005 - 07:00:41 EDT