-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://m5sim.org/cgi-bin/mailman/private/m5-users/attachments/20080918/c7a931f3/attachment-0001.htm
I think you have a misconception about the events. Events can be
reused, so as long as you don't have more than one event outstanding
for a specific task (and you really shouldn't ever do this), you
shouldn't need to dynamically create them. Of the top of my head I
can' think of what the copy constructor looks like for an Event, but
it doesn't seems like it would do what you appear to want it to do. I
would imagine that the event copying is the root of the problem, that
or something subtle about using the comInstEventQueue.
Ali
hi Ali,
I know my reply is late, very sorry. Anyways, I've fixed the thread-ID
error, it turns out I haven't initialized it properly in the constructor
list. So now I can schedule the event to be process()'d at the correct
instruction count in each thread, however I still get a segmentation fault
error when I call suspendContext(tid) or even deactivateThread(tid). Any
tips on how to debug this? And also my reason that I dynamically create
SuspendThreadEvent is because I can only find out the number of
threads(which would be important when Queue-ing the event in the
comInstEventQueue array),instruction count depending on the configuration of
the CPU. I'm fairly new to C++ and after reading about copy constructors I
understand the fundamental problem of my previous code in
scheduleSuspendThreadEvent() method,
so my new scheduleSuspendThreadEvent() is now:
void scheduleSuspendThreadEvent(int _tid,Tick when){
190 // Schedule thread to suspend, regardless of its current state.
191 DPRINTF(Event,"suspend thread:%d \n",_tid);
192
193 new SuspendThreadEvent(comInstEventQueue[_tid],1000,_tid);
//hard-coded to suspend in 1000 instructions each thread; will be changed
later
202 }
I removed the "SuspendThreadEvent suspendThread[Impl::Maxthreads]"(see code
below from previous email) line to prevent issues in copy constructors and
just dyamically created the events(the downside is that I cannot access the
created events after, which I guess is fine as of now). Also, I know that
this is not an M5-related question but a C++ one: Is it possible create an
array of Events in M5 that wouldn't use the standard constructors? The
relevance of my question is that I wanted to pass the thread_id of each
event so that it would be queued in its own comInstEventQueue[].
Thanks in advance!
On Sep 18, 2008, at 4:20 AM, Dean Michael Ancajas wrote:
> hi everybody,
> I've create d an Event called SuspendThreadEvent(same properties- Hide
quoted text -
> with ActivateThreadEvent in src/o3/cpu.cc) which will call
> FullO3CPU<Impl>::suspendContext(int tid), to be able to initialize a
> SuspendThreadEvent instance, I created
> scheduleSuspendThreadEvent(tid,cpu), the problem is when I passed
> the correct parameter to scheduleSuspendThreadEvent but according to
> the generated traces(see below) the value used in the initialization
> is different.
>
> commandline:
> [EMAIL PROTECTED]:~/m5/repos/smt$ ./m5.debug --trace-file=event2.out --
> trace-flags=Event tests/run.py quick/01.hello-2T-smt/alpha/tru64/o3-
> timing
>
>
> in cpu.cc
> if( params->max_insts_pause_each_thread == 0 ){ //number of
> instructiosn has been coded in initialization at cpu.hh
> 279
> 280 for(int tid=0;tid < numThreads; tid++){
> 281
> 282 DPRINTF(Event,"Num of threads: %d\n",numThreads); //
> print number of threads
> 283 DPRINTF(Event,"1. scheduling suspend thread: %d
> \n",tid); //print out thread # to be suspended
> 284 scheduleSuspendThreadEvent(tid,params-
> >max_insts_pause_each_thread);
> 285 DPRINTF(Event,"2. scheduling suspend thread: %d
> \n",tid); //double check if value of tid has changed
> 286 }
> 287 }
>
>
> in cpu.hh
> /** added by dean michael ancajas, Schedule thread to suspend **/
> 189 void scheduleSuspendThreadEvent(int _tid,Tick when){
> 190 // Schedule thread to suspend, regardless of its current
> state.
> 191 DPRINTF(Event,"suspend thread:%d \n",_tid);
> 192
> 193 suspendThreadEvent[_tid] = *(new
> SuspendThreadEvent(comInstEventQueue[_tid],100)); //for now this is
> hard-coded to suspend in 100 instructions each thread
> 194 suspendThreadEvent[_tid].init(_tid,this);
> 201 }
>
>
> Generated traces from event2.out:
>
> 2 0: system.cpu: Num of threads: 2
> 3 0: system.cpu: 1. scheduling suspend thread: 0
> 4 0: system.cpu: suspend thread:0
> 5 0: Event_15: FullO3CPU "Suspend Thread" event scheduled @
> 100
> 6 0: Event_15: thread suspend: 0
> 7 0: system.cpu: 2. scheduling suspend thread: 0
> 8 0: system.cpu: Num of threads: 2
> 9 0: system.cpu: 1. scheduling suspend thread: 1
> 10 0: system.cpu: suspend thread:1
> 11 0: Event_16: FullO3CPU "Suspend Thread" event scheduled @
> 100
> 12 0: Event_16: thread suspend: 9
> 13 0: system.cpu: 2. scheduling suspend thread: 1
> 14 0: Event_2: FullO3CPU tick event scheduled @ 0
>
>
> The first SuspendThreadEvent was initialized properly(line 3's
> thread # and line 6's thread # are equal, but the 2nd is not, line 9
> says it must suspend thread #1 but in line 12 it passed thread #9.
> I'm confused as to where the value started to change. As we can also
> see in line 13, the value of tid in cpu.cc hasn't change after
> calling the scheduleSuspendThreadEvent. So I guess the culprit must
> be in the scheduleSuspendThreadEvent.. Any help/ideas would be
> greatly appreciated.
>
> thanks in advance,
> Dean Michael Ancajas
>
> _______________________________________________
> 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
End of m5-users Digest, Vol 26, Issue 10
****************************************
--
Dean Michael B. Ancajas
_______________________________________________
m5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
