speed up sortEventListenersOfSceneGraphPriority by reverse visit

[huangwei1024]

speed up sortEventListenersOfSceneGraphPriority by reverse visit
its more efficient when node tree was big

[crazyhappygame]

You can not setUserData.
It can be used by user.

[huangwei1024]

yes, I known its not a right way which use UserData to storage some visit information for engine. and its simple to add an "SystemData" in Node for optimize

[crazyhappygame]

Could you explain new algorithm and provide profile results. What is the speedup.

[huangwei1024]

for (; i < childrenCount; i++)
{
	child = children.at(i);
	if (child && child->getLocalZOrder() < 0)
	{
		if (child->getUserData() == 0) // here!!!
			continue;
		visitTree(child);
	}
	else
		break;
}

I think its easy idea, if the tree had N nodes, and the sceneGraphListeners had M listeners, you could visit from M listener node to root, and color the path.
second, visit from root to child by the colored path. it could be avoid to visit useless path.

in my test proj, I had tree which contain 6K nodes, but listeners only about 600. I thought the case which listnener much less than nodes was in general. O(N) -> O(cM), c is the coefficient about the avg children and node in path. but in the worst cM = N

[huangwei1024]

sorry, I could not provide the test case, it contain too many complex widgets.
but I test in win32, the old method cost 8ms, and cost 3ms when use the new one.

[huangwei1024]

		if ((nodeData & 1) == 1)
			_globalZOrderNodeMap[node->getGlobalZOrder()].push_back(node);

if (_nodeListenersMap.find(node) != _nodeListenersMap.end())
        {
            _globalZOrderNodeMap[node->getGlobalZOrder()].push_back(node);
        }

and stl unordered_map also more expensive than bit mask.
when the N nodes is big, the cost need be consider