On Sun, May 8, 2011 at 9:09 AM, Hiuen Tsang <[email protected]> wrote:
> Hi,
>
> Scenario:
> I am trying to generate a trace of all the load/store virtual addresses
> generated by a multithreaded application.
>
> Issues:
> 1. As per one of the threads in this mailing list, this can be achived by
> trapping lsq->virtaddr just before it is reset in the function
> ReorderBufferEntry::commit() in the file ooopipe.cpp.** The addresses
> corresponding to Load instructions are being printed properly. However the
> ones associated with Store are being printed as 0x0. Can you please throw
> some light on as to why this anamoly is being observed?
>
> Command used from simulator: ./start_sim; ls ;./stop_sim
> Modification to ooopipe.cpp: just before the statement lsq->reset() in line
> 2285...
> if( lsq->virtaddr < 0x7f000000){
> if (st)
> cout << hexstring(lsq->virtaddr,64), " S threadid:
> ", threadid, " coreid: ", coreid, endl;
> else if(ld)
> cout << hexstring(lsq->virtaddr,64), " L threadid:
> ", threadid, " coreid: ", coreid, endl;
> }
>
> All the stores that has virtaddr set to 0 are memory fence opcodes. To
correctly trace stores, check if uop.opcode == OP_st.
> 2. Will modifying the different eventhandler in mesiCache.cpp allow me to
> the to trap different cache events ( e.g, the addresses recieved from a
> snoop signal etc.)?
>
Yes.
> 3. This is a naive question. How can I force the emulator to run only the
> instructions associated to the benchmark and stop any kernel code from
> execution?
>
> Currently there is no way to do this. But the stats are collected
separately for use and kernel mode if you are interested in stats only.
- Avadh
> I would very much appreciate any kind of help on this front.
>
> ---
> Thanks & Regards
> Hiuen
>
>
> _______________________________________________
> http://www.marss86.org
> Marss86-Devel mailing list
> [email protected]
> https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel
>
>
_______________________________________________
http://www.marss86.org
Marss86-Devel mailing list
[email protected]
https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel
