is now up on:

http://www.artima.com/cppsource/backyard.html

Walter Bright wrote:
> is now up on:
> 
> http://www.artima.com/cppsource/backyard.html

Great stuff!  I've only read the first page so far, but one thing caught my eye.  Change:

"We can then define a dehydrated pointer as being odd, and a hydrated pointer as being 1."

to:

"We can then define a dehydrated pointer as being odd, and a hydrated pointer as being even."


Sean

Sean Kelly wrote:
> Walter Bright wrote:
>> is now up on:
>>
>> http://www.artima.com/cppsource/backyard.html
> 
> Great stuff!  I've only read the first page so far, but one thing caught my eye.  Change:
> 
> "We can then define a dehydrated pointer as being odd, and a hydrated pointer as being 1."
> 
> to:
> 
> "We can then define a dehydrated pointer as being odd, and a hydrated pointer as being even."

Only about two dozen people reviewed that article, and all missed it <g>. Thanks.

>http://www.artima.com/cppsource/backyard.html

Hmm.  vptr is typically not at offset 0 if the base class is a POD.  It might be, depending on whether the compiler decides to change 'this' on the cast to base instead of just using an offset, but YMMV.

I can't help thinking that if you replace your malloc with nedmalloc or PTMalloc3 then most people will get a bigger speed boost than they will by trying to play dirty at such a low level.

James

James Mansion wrote:
>> http://www.artima.com/cppsource/backyard.html
> 
> Hmm.  vptr is typically not at offset 0 if the base class is a POD.  It might
> be, depending on whether the compiler decides to change 'this' on the cast to
> base instead of just using an offset, but YMMV.
> 
> I can't help thinking that if you replace your malloc with nedmalloc or
> PTMalloc3 then most people will get a bigger speed boost than they will by
> trying to play dirty at such a low level.

The tricks are useful in other ways though.  For example, I work with a piece of software whose "database" is a memory mapped file containing objects with modified vtbls.  The result is screaming fast and isn't something that could be accomplished by other means.  So I view this article more as a window to some of the evil (and sometimes useful) games one can play with C++ than as a template one should follow to increase performance in a typical application.


Sean

James Mansion wrote:
>> http://www.artima.com/cppsource/backyard.html
> 
> Hmm.  vptr is typically not at offset 0 if the base class is a POD.  It might
> be, depending on whether the compiler decides to change 'this' on the cast to
> base instead of just using an offset, but YMMV.

You do have to check that the vptr is in the right place, and adjust accordingly.


> I can't help thinking that if you replace your malloc with nedmalloc or
> PTMalloc3 then most people will get a bigger speed boost than they will by
> trying to play dirty at such a low level.

Why not do both and get double the speedup?