Hi Pritha, I am aware of that specific problem. The current, checked-in version of Ruby does not correctly handle load-locked and store-conditional instructions. I have a patch that fixes that specific issue, as well as many other patches that fix other issues, however I haven't found the time to clean them up and check them in. Unfortunately, I'll have to ask you to wait a bit for those patches. Like I said in my initial email, the protocols in M5 aren't quite ready for "primetime". I apologize for not being able to immediately help you.
Brad From: [email protected] [mailto:[email protected]] On Behalf Of Pritha Ghoshal Sent: Sunday, July 18, 2010 5:54 PM To: M5 Developer List Subject: [m5-dev] Parsec on Ruby Hi, I am trying to run Parsec on a Ruby based machine. What is the way that has been used till now? I have tried to boot a system with Ruby, am facing some problems in that regarding the boot up of cpu-s. So I tried to create a checkpoint and run it using ruby based machine. I am facing some problem in that too. Even though the checkpoint can be run completely using a nonruby based machine, a ruby based machine hangs after sometime. I tried to compare a ruby based machine memory trace with a non ruby based memory trace. I found that a specific set of addresses are getting looped continuously. While looking thoroughly, I found the data address for the normal ALPHA mode is different from the Ruby based ALPHA. The content in the register seems to be different. Normal Alpha 2265201113000: system.cpu3 T0 : @_spin_unlock_bh+152 : ldl r1,0(r16) : MemRead : D=0x0000000000000001 A=0xfffffc001f4fdf68 Ruby based Alpha 2265521254000: system.cpu3 T0 : @_spin_unlock_bh+152 : ldl r1,0(r16) : MemRead : D=0x0000000000000001 A=0xfffffc001f3f04e8 I am not sure, why this is happening, because both the executions are using the same checkpoint. I guess due to the change in data, there is a loop getting created which should not have been for correct execution. Could anyone help me regarding this? Also, is there a way to run Parsec benchmark in any other way? Thanks Pritha
_______________________________________________ m5-dev mailing list [email protected] http://m5sim.org/mailman/listinfo/m5-dev
