Re: [Gems-users] SMT in GEMS2.1

[Xuan Qi <xqi@xxxxxxxxxxx>]

Thank you very much for the explanation Luke,

However

(1)
>     These asserts are confusing but not incorrect.  Ruby only
> knows about SMT processor numbers, so the cpuNumber passed in should be
> less than numProcs.  In your example cpuNumber and seq_index should only
  be 0,1 if
> the SMT system is set up properly.

If this is the case, the first assert should be:

ASSERT( (cpuNumber >= 0) && (cpuNumber < system_t::inst->m_numSMTProcs) );

instead of:

ASSERT( (cpuNumber >= 0) && (cpuNumber < system_t::inst->m_numProcs) );

Is it?

(2) in system.C

first, the numProcs is defined as:

// establish how many processors there are in the system
  int numProcs = SIM_number_processors();   //for SMT, this is the total
number of LOGICAL processors in the system

then,

m_numSMTProcs = numProcs / CONFIG_LOGICAL_PER_PHY_PROC;

I am confused here because numProcs is first defined as the total number
of LOGICAL processors in the system. Is this the number of SMT processors?
If so, m_numSMTProcs = numProcs / CONFIG_LOGICAL_PER_PHY_PROC doesn't make
sense. It looks to me that numProcs should be set to the number
of PHYSICAL processors initiated by simics, is it? Maybe I misunderstand
the definition of LOGICAL processor?

Thanks again for your help...

Best regards,
Xuan Qi


On Tue, 25 Mar 2008, Luke Yen wrote:

> On Mon, 24 Mar 2008, Xuan Qi wrote:
>
> > Hi everyone,
> >
> > I am trying to configure SMT in GEMS2.1. I checked mfacet.py and
> > GEMS/opal/system/system.C but the code for SMT looks kind of strange
> > to me. Please help to correct my understanding. Thanks a lot.
> >
> > in system.C:
> >  "m_numSMTProcs = numProcs / CONFIG_LOGICAL_PER_PHY_PROC;"
> >
> >  so the number of SMT processor is equal the number of processors
> >  initiated by simics over the number of config_logical on each SMT
> >  processor. And CONFIG_LOGICAL_PER_PHY_PROC should be at least 2 in
> >  order to make use of SMT (2 threads can issue
> >  instructions simutaneously), am I right?
>
>      Yes, you are correct.
> >
> > Next in system.C
> >
> >  "ASSERT( (cpuNumber >= 0) && (cpuNumber < system_t::inst->m_numProcs) );
> >
> >  int32 seq_index = cpuNumber;
> >  ASSERT(seq_index >= 0 && seq_index < system_t::inst->m_numSMTProcs);"
> >
> >  As seq_index is equal to cpuNumber, if numProcs = 4, then seq_index can
> >  be 0, 1, 2, 3. But if i set CONFIG_LOGICAL_PER_PHY_PROC = 2, the
> >  seq_index can only be 0, 1 to satisfy the second assert. It is kind of
> >  conflict, isn't it?
> >
>
>     These asserts are confusing but not incorrect.  Ruby only
> knows about SMT processor numbers, so the cpuNumber passed in should be
> less than numProcs.  In your example cpuNumber and seq_index should only be 0,1 if
> the SMT system is set up properly.
>
>     Luke
> _______________________________________________
> Gems-users mailing list
> Gems-users@xxxxxxxxxxx
> https://lists.cs.wisc.edu/mailman/listinfo/gems-users
> Use Google to search the GEMS Users mailing list by adding "site:https://lists.cs.wisc.edu/archive/gems-users/"; to your search.
>
>