🔒 Locked

Smart Pointer performance overhead

Started by cypherx Feb 23, 2005 at 6:40 PM 3 replies 1.1k views

Original Post

Feb 23, 2005 at 6:40 PM

I'm currently using smart pointers in my game engine, but I'm a little nervous about the performance cost of dereferencing. Here's a really simplified version of my render loop:


class Drawable
{
    virtual void Render()=0; 
};

struct Object
{
    ptr<Drawable> drawable; 
    /* also contains other object properties */ 
}; 

for (list< ptr<Object> >::iterator iter = objectList.begin(); iter != objectList.end(); iter++)
{
    (*iter)->drawable->Render(); 
}

So to render any object the engine has to: 1) Dereference list< ptr

SumDude

Feb 23, 2005 at 6:43 PM

i don't think it would become much of a problem unless your recreating Doom 3 or Half life 2.

me22

Feb 23, 2005 at 9:33 PM

You can make a local copy of the location in a plain-old-pointer if you need to access stuff through it multipule times.

I think that the overhead of the smart pointer itself ought not to be significant, since the alternative is quite messy.

Sneftel

Feb 23, 2005 at 9:38 PM

Is it really that difficult to *put* 1000 objects in your scene, and *see* if it's too slow? Performance metrics like this are usually more or less useless; the only valid metric is real-world performance in your actual situation.

Promit

Feb 23, 2005 at 11:06 PM

It's probably (almost definitely) of no significant cost.

But if you're really worried about it, you could steal boost::shared_ptr (loki might have a smart pointer too, not sure). In this case, the code is probably metaprogrammed to resolve the dereference at compile time in most, if not all cases.

[EDIT] Read this!

Topic Locked

This topic has been locked by a moderator. New replies are not allowed.