[Ns-developers] A Reason for Slowness of NS3

[Mathieu Lacage]

On Sat, 2009-02-14 at 08:34 -0800, Tom Henderson wrote:

> Do you understand yet why Calendar is not doing better than Map in most 
> (all?) of our scenarios?

No. I looked at the calendar code carefully and I believe that I did not
introduce any performance bug in the calendar code proper. Someone who
knows the original ns-2 code better (hint!) should probably try to
review what I did in search of anything I might have screwed up.

> - TypeId tid = GetTypeId();
> + const TypeId& tid = GetTypeId();
> 
> won't the copy in this case just be optimized away by compilers, making 
> return by reference here unnecessary?  Does gcc optimize this case?

1) The answer to this question does not really matter because return a
const reference is simply not an option from a maintenance point of view
so, arguing about whether or not it is necessary from a performance
point of view is not really useful.

2) but, yes, from a purely academic point of view, you are right, gcc
optimizes this case:

-bash-3.2$ cat test.cc
#include "simple.h"
int main (int argc, char *argv[])
{
  Simple s = Do ();
  s.Foo (2);
  return 0;
}
-bash-3.2$ cat simple.h
class Simple
{
public:
  Simple (const Simple &o);
  Simple ();
  ~ Simple ();
  Simple &operator = (const Simple &o);
  void Foo (int v) ;
private:
  int m_v;
};

Simple Do (void);
-bash-3.2$ 
-bash-3.2$ cat simple.cc
#include "simple.h"
Simple::Simple (const Simple &o) {}
Simple::Simple () {}
Simple::~ Simple () {}
Simple &
Simple::operator = (const Simple &o)
{}

void Simple::Foo (int v) {m_v = v;}
Simple Do (void)
{
  return Simple ();
}
-bash-3.2$ g++ -O3 ./test.cc ./simple.cc
-bash-3.2$ objdump -d ./a.out |c++filt -t |less
00000000004005e0 <main>:
  4005e0:       48 89 5c 24 f0          mov    %rbx,-0x10(%rsp)
  4005e5:       48 89 6c 24 f8          mov    %rbp,-0x8(%rsp)
  4005ea:       48 83 ec 28             sub    $0x28,%rsp
  4005ee:       48 89 e7                mov    %rsp,%rdi
  4005f1:       e8 ba 00 00 00          callq  4006b0 <Do()>
  4005f6:       be 02 00 00 00          mov    $0x2,%esi
  4005fb:       48 89 e7                mov    %rsp,%rdi
  4005fe:       e8 9d 00 00 00          callq  4006a0 <Simple::Foo(int)>

See that there is no call to Simple::Simple(const Simple&);

Mathieu