Re: [Gems-users] How to collect multicore coherence traffic

[jianhual@xxxxxxxxxxxxxxxx]

Hi John,

Sorry for trouble! I checked your reply information and it seems that it's not the thing I want. Thank you, Let me

clarify my question:

For a multicore platform running multithreaded programs, I want to collect the traffic trace 

for cores which means to export the core's traffic trace into a trace file. The format 

of the trace file is something like follows.

The format of the trace file for a 16-core CMP system

----------------------------------------------------------------------------------

1 U rd 0 1 16 0xa01280

1 U rd 1 15 16 0xb6fb80

1 U rd 2 1 16 0xa01280

1 U rd 3 1 16 0xa01280

1 U rd 4 1 16 0xa01280

1 U rd 5 1 16 0xa01280

1 U rd 6 1 16 0xa01280

1 U rd 7 1 16 0xa01280

1 U rd 8 1 16 0xa01280

1 U rd 9 1 16 0xa01280

1 U rd 10 1 16 0xa01280

1 U rd 11 1 16 0xa01280

1 U rd 12 1 16 0xa01280

1 U rd 13 1 16 0xa01280

1 U rd 14 1 16 0xa01280

1 U rd 15 1 16 0xa01280

--The first field is the injection time stamp.
--The second field is the message type – this is either a U for unicast and an M for multicast (multicast means that write-caused 

   write-invalidate in directory based cache coherence protocols)

--The third field relates to the type of coherence message such as read, write
--The fourth field is the source
--The fifth field is the destination (when the message type is M, this is a bit vector represented as a hex number; 

   for unicast it will only be destinations from 0 to 15) 
--The sixth field is the message size in bytes
--The seventh field is the memory address

--------------------------------------------------------------------------------------

Wish someone can help me on this issue. I just want to know how and where to implement such 

function in GEMS framework. Thanks in advance!

Regards,

Jianhua