Re: [Gems-users] How to set up the network file for MESI_SCMP_tiledL2bank_directory protocol

[Calvin <calvincv@xxxxxxxxx>]

Hi Mike,

Thanks for your fast reply. I will try to apply the fix mentioned by
Niket to see if that works. However, I just tried to use the PT_TO_PT
topology but it doesn't work either. Here is my spec in
rubyconfig.defaults:
---------------------------------------------------------------
g_NUM_PROCESSORS: 2
g_NUM_L2_BANKS: 1
g_NUM_MEMORIES: 1
g_PROCS_PER_CHIP: 2
...
g_NETWORK_TOPOLOGY: PT_TO_PT
g_CACHE_DESIGN: NUCA
g_endpoint_bandwidth: 10000
g_adaptive_routing: true
NUMBER_OF_VIRTUAL_NETWORKS: 5
FAN_OUT_DEGREE: 4
g_PRINT_TOPOLOGY: false
---------------------------------------------------------------

The error msg is:
---------------------------------------------------------------
[cpu0] v:0x0000000000012f5c p:0x0000000039f22f5c  sethi %hi(0x13c00), %g2
failed assertion 'msg_destinations.count() == 0' at fn virtual void
PerfectSwitch::wakeup() in
Gems/Install_MESI_SCMP/ruby/network/simple/PerfectSwitch.C:264
failed assertion 'msg_destinations.count() == 0' at fn virtual void
PerfectSwitch::wakeup() in
Gems/Install_MESI_SCMP/ruby/network/simple/PerfectSwitch.C:264
---------------------------------------------------------------

I also tried to use HIERARCHICAL_SWITCH, but got the same error. Could
you help clarify what is the correct set up for using other topology
for this protocol?

--Calvin

On 3/8/07, Mike Marty <mikem@xxxxxxxxxxx> wrote:
> Calvin,
>
> Your network file works fine on my end.  Read my previous message about
> the debugging output however.
>
> You might want to connect the Directory controllers (actually memory
> controllers for this protocol) to seperate switches so that you can
> properly constrain memory bandwidth if you want that.
>
> --Mike