[Gems-users] About data prefetch

[lori zhuang <lorizhuang@xxxxxxxxx>]

List,

     I wonder if Ruby simulates the behavior of data prefetch. It seems like it does according to this link https://lists.cs.wisc.edu/archive/gems-users/2007-November/msg00102.shtml. But it seems like it doesn't according to the code in SimicsDriver.C:

int SimicsDriver::makeRequest(memory_transaction_t *mem_trans) {

    ...

    ...

    // skip unsupported types

    if (isUnhandledTransaction(mem_trans)) {

    ret = Unhandled;

    } else {

     // perform makeRequest()

      int proc = SIMICS_get_proc_no(mem_trans->s.ini_ptr);

     ret = m_processors[proc]->makeRequest(mem_trans);

   }

   ...

  ...

}

bool SimicsDriver::isUnhandledTransaction(memory_transaction_t* mem_trans) {

  // only handle user data?

  if (USER_MODE_DATA_ONLY) {

    if(PRIV_MODE) {

      return true;

    }

    if(mem_trans->s.type == Sim_Trans_Instr_Fetch) {

      return true;

    }

  }

  // no prefetches

  if (mem_trans->s.type == Sim_Trans_Prefetch) {

    return true;

  }

  // no cache flush

  if (mem_trans->s.type == Sim_Trans_Cache) {

    return true;

  }

  // no DMA & IO

  if (IS_DEV_MEM_OP(mem_trans->s.ini_type) ||

      IS_OTH_MEM_OP(mem_trans->s.ini_type) ||

      mem_trans->s.physical_address > uinteger_t(RubyConfig::memorySizeBytes())

     ) {

    return true;

  }

  return false;

}

Ruby skips supported types memory request and it threats Sim_Trans_Prefetch as unsupported types. Hence I think Ruby doesn't simulates data prefetching. Am I right?

Since I'm going to add a prefetch instruction of the SPARC v9 instruction set to a benchmark and see if any speedups. That's why I care so much about prefetching. Any suggestion is highly appreciated. 

Thanks,

lorizhuang