[boost] Re: new approach to smart ptrs -- no new()s is good news

From: Greg Colvin (gcolvin_at_[hidden])
Date: 2000-01-28 23:45:39

From: <jsiek_at_[hidden]>
> I'd be willing to bet this new linked implementation is faster in most
> situations. Going to memory is getting slower and slower relative to
> clock speeds, and the extra copying just gives the cpu something to do
> instead of stalling :)

The copy code for the current implementation expands to this:

      if (pn != rpn) {
         if (--*pn == 0) { delete px; delete pn; }
         px = rpx;
         ++*(pn = rpn);
      }

The copy code for Mark's implementation expands to this:

      lastref = !left && !right;
      if (left || right)
      {
         if (left)
            left->right = right;
         if (right)
            right->left = left;
         right = left = 0;
      }
      if (lastref && ptr)
         delete ptr;
      ptr = pNewLeft->ptr;
      left = pNewLeft;
      right = pNewLeft->right;
      pNewLeft->right = this;
      if (right)
         right->left = this;

> It would be interesting to see the performance difference in the
> pathological case of doing nothing but a bunch of ptr assignments in a
> tight loop.

Yes, it would.  Mark?

> Greg Colvin writes:
>  > A very interesting implementation.  Its existence may be a good reason to
>  > deprecate the use_count() member of shared_ptr.  I'm not too sure that
>  > you would see a 4X improvement in typical applications though, since your
>  > test does a lot of allocation and very little copying.  I'd be interested
>  > in seeing the results of more thorough tests.
> 
> ------------------------------------------------------------------------
> Want to send money instantly to anyone, anywhere, anytime?
> You can today at X.com - and we'll give you $20 to try it!  Sign 
> up today at X.com.  It's quick, free, & there's no obligation!
> http://click.egroups.com/1/332/1/_/9351/_/949120014/
> 
> -- 20 megs of disk space in your group's Document Vault
> -- http://www.egroups.com/docvault/boost/?m=1
> 
> 
>