next »

[Jay Jeon]

http://www.youtube.com/v/0DrcxkcNeVk
this is my pysics engine demo.
but fps is very low.(about 30? or 20?)
I want to make faster. and I really wonder about aquaria's fps.
How could you keep 60 fps? besides, render and compute many things same time.
how? using separated space? or what?

[dustin]

um... well what language is it written in?  Also are you using hardware acceleration for the graphics?  If not it could easily be your rendering method slowing things down.  Even so if your using a fast language (like c/c++ etc.) it does seem slow for just simple no torque physics like that.  I know in my java game I don't have a physics detection going on exactly like your but it's pretty simlar in terms of speed usage I would imagine (100 or so things colliding with 10 or so other objects which would be roughly equivalent to 32 things colliding with 32 other things) and I seem to be able to pretty easily get constant 60 fps and java is slower then c/c++.  I know your video is in flash so if this is programmed in flash then I would imagine that is the main reason why your fps is low...

If you want general tips in how to handle collision detection faster then I would suggest splitting the screen up into areas.  For example divide the screen into 4 quadrants, everything in the upper left only needs to worry about collision with other things in the upper left etc. That kinda thing helps a lot because then instead of 32 objects colliding with 32 other objects (1024 collisions possible) you only have to check

1.  each object with the 4 quadrants worst case (32*4=128)
2.  each object with the others in it's quadrant assuming equal distribution (32/4*32/4*4=256) the 32/4 are the number of objects in each quadrant and there are 4 quadrants in total.

Those add up to 128+256=384 as apposed to 1024.  Again though I would make sure nothings wrong before making your code more complicated even if it's not that much.  You could be not adressing the real problem.

[BenH]

Why not use an existing physics engine like Box2D or ODE?

[Jay Jeon]

Why not use an existing physics engine like Box2D or ODE?

I think Box2D isn't good to compute a lots of thing at the same time.
(Box2D is too sensitive )
That's why I use my simple pysics engine.

and dustin thank you.now I got it. :D
(Iam using C/C++ with directX.)

[dustin]

Why not use an existing physics engine like Box2D or ODE?

I can see not using one of thse just because it's fun coding your own and less complicated if you want a simpler physics engine...

and dustin thank you.now I got it. Cheesy
(Iam using C/C++ with directX.)

you are welcome however if your using c++ it seems weird that you are getting slow down with only 50 by 50 objects.  Just as a test I tried it with my own code which like I said is not exactly like yours but still.  I managed to do about 500*500 rectangular collision detections (plus displaying, and stuff) at 30 fps in java.  It seems like there might be something wrong with your code that's slowing it down a ton.

[Jay Jeon]

you are welcome however if your using c++ it seems weird that you are getting slow down with only 50 by 50 objects.

Oh. Actually, now I use 450 boxes.

[dustin]

oh ok well sweet then!

[Alex May]

If you want general tips in how to handle collision detection faster then I would suggest splitting the screen up into areas.  For example divide the screen into 4 quadrants, everything in the upper left only needs to worry about collision with other things in the upper left etc. That kinda thing helps a lot because then instead of 32 objects colliding with 32 other objects (1024 collisions possible) you only have to check

1.  each object with the 4 quadrants worst case (32*4=128)
2.  each object with the others in it's quadrant assuming equal distribution (32/4*32/4*4=256) the 32/4 are the number of objects in each quadrant and there are 4 quadrants in total.

Those add up to 128+256=384 as apposed to 1024.  Again though I would make sure nothings wrong before making your code more complicated even if it's not that much.  You could be not adressing the real problem.

This would cause problems for objects along the boundaries between the quadrants intersecting each other.

A better way would be to divide the space up into smaller partitions (let's call them buckets) and the have each object do collisions with things in adjacent buckets as well as the one it is in.

Also see Quadtrees.

[dustin]

yeah I know it's probably not the absolute best but it is the simplest I know of and it sure cuts down on the needed calculations.  Also you don't get the problems along the  boundaries as long as you specify that each object can be in multiple quadrants.  At least I don't see how...

Splitting the screen up into more then 4 areas is also what I was thinking of (I just suggested splitting it up into areas) but I didn't want to make my calculations over complicated

[Alex May]

Ah, please forgive a tired programmer for being pedantic on a monday morning

[Jay Jeon]

yeah, I've heard about Quad Tree on my class.
but I don't know how it works exactly. and It's so complicated to me.
I think the best way is simple thing.

[Alex May]

I've never found a straight-up bucket sort to be particularly lacking, personally. I've never implemented a quadtree!

[DrDerekDoctors]

Another vote for bucket sort here! I use 'em in my engine and I can have craploads of stuff colliding at once without slowdown.