Hi Rick,
Regarding the O3 CPU tracing misspeculated instructions, it only
prints out valid, committed instructions. The key function for
determining what instructions get printed out is when dump() gets
called on the traceData. You'll notice that it only gets called in
commit_impl.hh on instructions that have committed. Right now there's
actually no option for printing out misspeculated instructions, though
in the future it can be added in.
Regarding the unexpected output that you're getting, the problem is
probably that the retireHead() function gets called on both squashed
and committed instructions. It may be a bit of a misnomer in that
case, as it really removes the head instruction (it depends on your
definition of retire). It is definitely a misnomer that the
instruction in there is set as committed, regardless of it being
squashed or committed.
I would suggest you look for the places where dump() gets called and
put your call to setCommitted there instead. If you still have issues
with garbage output let me know.
Kevin
Quoting Rick Strong <[EMAIL PROTECTED]>:
> I modified the InstRecord class to keep track of commit time of an
> instruction. I update commit time
> when src/cpu/o3/rob_impl.hh:retireHead calls head_inst->setCommitted.
> This goes to
> src/cpu/base_dyn_inst.hh:setCommitted(). I added to this code:
>
> setCommitted(){
> status.set(Committed)
> #if TRACING_ON
> if (traceData != NULL){
> traceData->setCommitted();
> }
> #endif
> }
>
> I also modify src/cpu/exetrace.cc to print out the committed time at the
> beginning of each line. The weird thing is some of instructions have
> times that are negative, zero, or other garbadge. Either the
> ExeTracerRecord is called before commit on some instructions or some of
> the instructions are misspeculated and commit times is not being set.
>
> Any ideas?
>
> Best,
> -Rick
>
> Steve Reinhardt wrote:
>> Are you getting misspeculated instructions even when you set the flag
>> to turn them off?
>>
>> On Thu, Oct 30, 2008 at 4:54 PM, Rick Strong <[EMAIL PROTECTED]> wrote:
>>
>>> Hi,
>>>
>>> For an exec trace, it seems that both misspeculated and
>>> non-misspeculated instructions are being printed out for the o3cpu as
>>> the src/cpu/o3/thread_context.hh returns false no mater what. Is this
>>> the case or am I missing a level of indirection?
>>>
>>> Best,
>>> -Rick
>>>
>>> _______________________________________________
>>> 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
>
>
>
_______________________________________________
m5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
