 |
Attention | Topic was automatically imported from the old Question2Answer platform. |
 |
Asked By | osimmac |
I’ve created a Naive approach Nbody simulation just to test performance in godot, with gdscript it can run with about 200 rigid bodies, cool, i recreated the nbody part of the code in a GDnative C++ library, and it can run about 600 rigid bodies, better, but with a few optimizations and basic multithreading within the C++ code, i was able to get it running smoothly with 1200 bodies.
However, I would like to find the best method of rendering all these bodies, currently im passing in a reference to each rigidbody to the gdnative library where it keeps a list of all them.
I noticed that with no gravity, just rigidbodies sitting there doing nothing, godot will start lagging once i spawn around 5000 of them, each of these instances have 2 nodes to them, so thats 10000 nodes.
I’ve looked into the “optimization using servers” and have attempted to recreate first part of rendering a sprite in C++ to no avail, it works fine it gdscript. I’ve researched as much as i could to find out what im doing wrong but idk, it will just crash every time i try running my game, this is what i have so far:
void RbSpawner::_init()
{
Godot::print(String("initilizing!"));
ReLo->get_singleton();
//create canvas item, child of node
RID canvasitemrid = Visual->canvas_item_create();
RID thisID = get_canvas_item();
//made this node parent
Visual->canvas_item_set_parent(canvasitemrid, thisID);
img.instance();
img = ReLo->load("res://red.bmp","res//red.bmp");
img->lock();
sprite->set_texture(img);
img->unlock();
//draw sprite on it
//this is a reference!
//add it centered
Visual->canvas_item_add_texture_rect(canvasitemrid, Rect2(sprite->get_texture()->get_size(), sprite->get_texture()->get_size()),sprite);
//add the item 45 degress rotated
Transform2D xform;
xform.rotated(.785398);
//translated (should be down right)
xform.translated(Vector2(20,30));
Visual->canvas_item_set_transform(canvasitemrid,xform);
}
And in my header file here are the variable for this class:
//member variables
private:
ResourceLoader* ReLo;
VisualServer* Visual;
//physics server is an interface for low level 2d physics access.
Physics2DServer* Physics;
//handle for resources unique ID (doesn't grant access to resource by itself, need a server to do stuff.)
RID body;
Ref<Image> img;
Ref<World2D> world;
Ref<RectangleShape2D> shape;
Sprite* sprite;
I’ve tried many variations of that, i hope its a simple fix, Is it even recommended to do this? how much faster would this be than using gdscript to run it? i guess even just running this in gdscript would be much faster than instancing all my rigidbodies in gdscript, but i would like it to run as fast as possible, but still be easy to develop. from what i understand i’d essentially have to replace the scene system for the objects i don’t in it.
btw i was unable to use Ref for many things because i got the error
Ref.hpp:180:31: error: 'class godot::ResourceLoader' has no member named 'unreference'
thank you for reading
i guess i should mention im on 3.2.1 and downloaded the headers about a week ago
osimmac | 2020-04-01 08:10
I’d try asking on IRC at this point (#godotengine-gdnative on irc.freenode.net), since this Q&A platform doesn’t have a lot of GDNative users.
Also, keep in mind rendering might be the bottleneck at this point due to the sheer amount of draw calls.
Calinou | 2020-04-01 08:21
each body just has 1 texture and its only 1 pixel big, i would hope godot rendering/ my computer could handle more, CPU particles can spawn quite a bit, at least 30,000. but yeah its a lot to ask for, im simply stress testing for now, in a game i would optimize the heck out of this and remove objects far away from the player give them a more abstract representation.
im am curious what the most optimal path is though, it would be awesome to have a very fast gravity simulator working in godot, even just 1200 particles it looks pretty cool, 10,000 would be pretty(though unrealistic)
osimmac | 2020-04-01 09:01