On Sun, 12 Aug 2012, jinsong wrote:
Hi All, These two days I try to run parsec in Alpha FS mode with ruby (use MOESI_hammer protocol). After attempted with several combinations in the simulation, I found such situation: Without ruby attached (i.e., fs.py is used), I can successfully run parsec with and without the checkpoint. However, with ruby attached (i.e., ruby_fs.py is used), different combinations used in the simulation got different results as below: 1. cpu-type=timing, checkpoint disabled, simulation result: succeed; 2. cpu-type=timing, checkpoint enabled, simulation result: failed; 3. cpu-type=detailed, checkpoint enabled or disabled, simulation result: failed. In these simulations, only situation 1 succeed on the condition that "swithcpu" should be disabled. Each time I switched cpu type from timing to detailed, the program aborts. As for situations 2 and 3, they always abort with certain errors. I search the gem5 mail-list archive and found that Nilay and Yuan have pointed out "In essence, the checkpointing support was added to Ruby with the assumption that simulation would stop after the checkpoint is created." However, the problem remains unsolved. In summary, has anyone successfully run parsec in Alpha FS mode with ruby and checkpoint? Would you please share your experience with me? Thank you very much!
You should report the errors that you got as output for situations 2 and 3. Also I have posted a patch on the reviewboard ( http://reviews.gem5.org) that alters the behavior of the Alpha architecture when it starts up. You might want to try running your simulations with this patch applied.
-- Nilay _______________________________________________ gem5-users mailing list [email protected] http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
