Hi Guys,

I am writhing a game where I have all of the game assets in an std::vector.

But, I am wondering if there is a faster way to do what I am doing.

Essentially, when ever I deal with a particular game asset I get the data from it similar to below

std::vector<ObjectInstance>::iterator it;
for(it=vObject.begin();it<vObject.end();it++)
{
	if(szName==it->szName.c_str())
		return it->x;
}

So, everything is handled this way, collisions, rendering, positioning, etc...

I am finding (in particular collisions) that this seems to be fairly expensive on the CPU as I am calling the above code (or similar to) possibly several hundred (maybe thousand) times per loop. Whilst it works perfectly in its current form, I am worried that when I get a lot of game assets in the scene (currently only four) that this will be too slow.

For example, I have a function that sorts out collisions and if the object is intersecting with another object the function calls a while loop to place the object back where it should have been before the collision.

So, I did a check. If the object is 256 pixels deep into the other object it takes 0.5 seconds for the loop to be exited.

Normally you wouldn't let an object collide that far into another object. But, the recovery time is quite long.

Some testing points the finger at the Vector routine (as it loops over and over).

Is there a faster way to achieve what I am after (while making the code as flexible as it is)?

Thanks in advance

+1 for the std::map<>

With regards to iterating over all objects in a given container, there's really nothing that beats vector. It's unlikely that changing the container will have any significant impact on your collision detection code. Should be perfectly fine for entity logic and rendering though, and there are options to eliminate the necessity of actually deleting elements in the vector to avoid that problem.

For collision detection though, If your collision detection algorithm is O(N^2), there's no container that's going to allow that sort of brute force method to work even remotely efficiently for large N. You need to focus more on the collision detection algorithm itself, partition your objects to minimize the total number of collision tests that need to be made. A simple grid, a BSP tree, etc. That's where you'll see real gains.