For the record, you were using two different process objects in the
first place; the parentheses on the end of "Benchmarks.SPECGCCEIO()"
indicate that it's creating an instance (just standard python class
instantiation syntax). In order to assign the same instance to both
CPUs (which is what you'd want to do for a multithreaded program)
you'd need to assign an instance to a variable and then use that
variable inside the loop.
That said, it's not clear to me what's happening. The fact that cpu1
has 400 squashed instructions means that something is going on there
though. I'd guess that cpu1 is misspeculating so badly that cpu0 gets
to 1000 committed instructions (thus ending the simulation) before
cpu1 gets any. I'm not sure why the misspeculation rates would be so
different though. I'd run with --trace-flags=Exec and see what the
cpus are really doing. If you want you can also try raising the
max_insts to 10000 and see if cpu1 gets any instructions committed.
Steve
On Sun, Jun 22, 2008 at 3:18 PM, prannav shrestha <[EMAIL PROTECTED]> wrote:
>
> HI meng-ju!!
> Evein if I used two different workloads for tiw CPUs the result is still
> the same. Ok Now I have changed the code as:
> .............
> ............
> system.cpu[0].workload = [Benchmarks.SPECGCCEIO()]
> system.cpu[1].workload = [Benchmarks.SPECGAPEIO()]
> for i in xrange(np):
> system.cpu[i].addPrivateSplitL1Caches(L1Cache(size = '16kB'),
> L1Cache(size = '8kB'))
> system.cpu[i].connectMemPorts(system.tol2Bus)
> system.cpu[i].max_insts_all_threads = 1000
> ## system.cpu[1].workload = [Benchmarks.SPECGCCEIO()]
> root = Root(system = system)
> still the result is same. So what mmay be the reason, why M5 2.0b5 is not
> able to run simulation for multiple core?
>
> Regards
> Prannav
>
> Hi all!!
> i tried to simulate multi-core system in M5.0b5 with cache upto level three.
> I am trying to run O3 cpu in SE. Part of corresponding configuration file is
> as below:
>
> ................
> ..............
> CPUClass.clock = '2GHz'
>
> np = 2
>
> system = System(cpu = [DerivO3CPU(cpu_id=i) for i in xrange(np)],
> physmem = PhysicalMemory(range=AddrRange("1024MB")),
> membus = Bus(), mem_mode = test_mem_mode)
>
> system.physmem.port = system.membus.port
>
>
> system.l3 = L3Cache(size='2MB')
> system.tol3Bus = Bus()
> system.l3.cpu_side = system.tol3Bus.port
> system.l3.mem_side = system.membus.port
>
>
> system.l2 = L2Cache(size='1MB')
> system.tol2Bus = Bus()
> system.l2.cpu_side = system.tol2Bus.port
> system.l2.mem_side = system.tol3Bus.port
>
>
> for i in xrange(np):
> system.cpu[i].addPrivateSplitL1Caches(L1Cache(size = '16kB'),
> L1Cache(size = '8kB'))
> system.cpu[i].connectMemPorts(system.tol2Bus)
> system.cpu[i].max_insts_all_threads = 1000
> system.cpu[i].workload = [Benchmarks.SPECGCCEIO()]
> ## system.cpu[1].workload = [Benchmarks.SPECGCCEIO()]
> root = Root(system = system)
>
> Simulation.run(options, root, system, FutureClass)
>
>
> .............
> I have used 2 core in this example, which runs the same program. The problem
> is that, i found that only one of the CPU reaches its maximum instruction
> count to 1000 while the instruction count at another CPU remains zero. The
> m5stat.txt files shows following :
> ......
> sim_insts 1000
> # Number of instructions simulated
> sim_seconds 0.000004
> # Number of seconds simulated
> .........
> ......
> system.cpu0.commit.COM:count 1067
> # Number of instructions committed
> system.cpu0.commit.COM:loads 265
> # Number of loads committed
> system.cpu0.commit.COM:membars 0
> # Number of memory barriers committed
> system.cpu0.commit.COM:refs 388
> # Number of memory references committed
> system.cpu0.commit.COM:swp_count 0
> # Number of s/w prefetches committed
> system.cpu0.commit.branchMispredicts 61
> # The number of times a branch was mispredicted
> system.cpu0.commit.commitCommittedInsts 1067
> # The number of committed instructions
> system.cpu0.commit.commitSquashedInsts 294
> # The number of squashed insts skipped by commit
> system.cpu0.committedInsts 1000
> # Number of Instructions Simulated
> system.cpu0.committedInsts_total 1000
> # Number of Instructions Simulated
>
> .....
> ...
> ...
> system.cpu1.commit.COM:count 0
> # Number of instructions committed
> system.cpu1.commit.COM:loads 0
> # Number of loads committed
> system.cpu1.commit.COM:membars 0
> # Number of memory barriers committed
> system.cpu1.commit.COM:refs 0
> # Number of memory references committed
> system.cpu1.commit.COM:swp_count 0
> # Number of s/w prefetches committed
> system.cpu1.commit.commitSquashedInsts 400
> # The number of squashed insts skipped by commit
> system.cpu1.committedInsts 0
> # Number of Instructions Simulated
> system.cpu1.committedInsts_total 0
> # Number of Instructions Simulated
>
> ..
> why is the instruction count in cpu1 zero.?As I am new in m5, i am not able
> to figure out the problem. Suggestion from anyone would be appreciated..
>
> regards
> prannav
>
>
> _______________________________________________
> m5-users mailing list
> [email protected]
> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
>
_______________________________________________
m5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
