Re: [Gems-users] Simulation is hanging

[Jayaram Bobba <bobba@xxxxxxxxxxx>]

Are you seeing consistency check failures due to overlap between escape
actions and transactions?

xactIsolationCheck in TransactionSimicsProcessor.C checks that escape
accesses do not overlap transactional accesses. In our experience, overlap
happens due to false sharing. In such cases, we deal with it by padding
out data structures accessed within escape actions (e.g., see transaction.h)
If overlap is not due to false sharing, then you have trickier semantic
issues to handle (see nested Logtm paper in ASPLOS06).

If you are getting the alternate error report, then you might be running 
into an
implementation bug.

Jayaram

Ricardo Quislant del Barrio wrote:
> Hi all,
> I'm simulating STAMP large transactional benchmarks with Ruby.
> I've been able to simulate "kmeans" benchmark without incidences but 
> when trying
> with other benchmarks  (labyrinth or yada) the simulation hangs and it 
> doesn't end.
> With "genome" and "vacation" benchmarks the simulation ends some times 
> and others
> (the more frequently), it hangs.
> Sometimes are shown XACT_CONSISTENCY_CHECKER conflicts.
> I build the checkpoints inside Simics 3.0.31: Solaris 9, gcc 3.4.6.
> GEMS version: 2.0
> Perhaps, i need to do some complex modifications to benchmarks. I've 
> just changed the
> macros BEGIN_XACT, COMMIT_XACT from STAMP to point the ones in 
> transaction.h from
> GEMS, and I've also inserted tm_bind_to_cabinet, Barrier_breaking, 
> set_transaction_registers and BEGIN_WORKLOAD_TRANSACTION before the 
> first transaction begins.
> I've dissasemble the code of the compiled benchmark and the offset in 
> set_transaction_registers
> is ok.
> If XACT_CONSISTENCY_CHECKER conflicts could be the problem, ¿How can I 
> fix them?
> Thanks,
> Ricardo
>
> _______________________________________________
> Gems-users mailing list
> Gems-users@xxxxxxxxxxx
> https://lists.cs.wisc.edu/mailman/listinfo/gems-users
> Use Google to search the GEMS Users mailing list by adding "site:https://lists.cs.wisc.edu/archive/gems-users/"; to your search.
>
>