Hi Yu,

We don't use the memory trace writer much... in fact I just found that it never made the transition from v1.1 to v2.0 beta, and this is the first time I know of that anyone has noticed. (Since we rewrote the memory system interface between those versions, the trace writer needs to be rewritten too, and no one has bothered.)

If you're using v1.1, then you need to go into your config file and create either an M5Writer or ITXWriter object (see python/m5/objects/MemoryTrace.py) and assign it to the mem_trace parameter of the cache whose accesses you want to trace. It's not something you can enable purely via the command line.

If you're using v2.0, then my suggestion is just to hack in some code to write out the address at the top of the handleAccess() method in src/mem/cache/cache_impl.hh. It's not that complex, and that way you can output whatever format you'd like. If you don't want to mess around with adding parameters to the cache to enable/disable tracing for particular caches (which is admittedly much more complex than it should be right now... we're working on simplifying that...), then you can just add some per-cache flags in cache.hh and set them based on the value of _name in the constructor (which is at the top of cache_impl.hh).

Hope that helps...

Steve

Yu Zhang wrote:
Hi,

I’d like to get the information of memory traffic for an application. However, when I use command –MemoryTrace.trace=<filename>, there is no trace file dumped. For a single core processor and only one application is running, the default thread is 0, isn’t it? So, I don’t need to specify the parameter “MemoryTrace.thread”, which is 0 by default. I wonder why the MemoryTrace doesn’t work. BTW: I’m using M5_1.1.

Thanks,

Yu


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

_______________________________________________
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

Reply via email to