Improve performnace with Mesh?

hnim · October 2012

i saw this video

11000 textured quad (rectangle shape/sprite ? ) were update at frame-rate ~30. So i wonder if we can make same thing with Gideros. If yes, maybe we will have a simple batch?

Nascode · October 2012

Ah, dont let @atilim look at this video! He wont take a rest until Gideros surpassed that vid

hnim · October 2012

he must see this. he must see this.

)

techdojo · October 2012

Gotta love Codea

hnim · October 2012

bad luck. i do some try, but it's too slow, slower than using bitmap

techdojo · October 2012

That surprises me - I would have thought a single mesh to be more efficient than rendering lot's of separate bitmaps, fancy sharing your code so we can see if we can identify the bottlenecks?

hnim · October 2012

i made a simple example
in bitmaptest, i add 11000 bitmap (8x8px), each frame i change their position to a random point. I got fps 59 on Desktop player
in meshtest, i set vertex to create 11000 rect, each frame i setVertex to new position. I got fps 40 on Desktop player.

atilim · October 2012

@hnim I think your bottleneck is Lua and not the drawing.

I've done a simple test. Here is my code:

local function add(t, ...)
	local n = select("#", ...)	
	for i=1,n do
		t[#t+1] = select(i, ...)
	end
end
 
application:setBackgroundColor(0x306090)
 
local vertices = {}
local indices = {}
local velocities = {}
 
local nrect = 0
 
for i = -400, 400, 5 do
	for j = -400, 400, 5 do
		local x = i / 275
		local y = -j / 275
 
		-- implicit heart function is from <a href="http://mathworld.wolfram.com/HeartCurve.html" rel="nofollow">http://mathworld.wolfram.com/HeartCurve.html</a>
		local a = x * x + y * y - 1
		local f = a * a * a - x * x * y * y * y
 
		if f < 0 then
			local n = #vertices/2
 
			-- add 4 corners
			add(vertices, i,     j)
			add(vertices, i + 4, j)
			add(vertices, i + 4, j + 4)
			add(vertices, i,     j + 4)
 
			-- add 2 triangles to indices
			add(indices, n + 1, n + 2, n + 3)
			add(indices, n + 1, n + 3, n + 4)
 
			-- and velocities
			add(velocities, x + math.random(-100,100)/200, -y + math.random(-100,100)/200)
 
			nrect = nrect + 1
		end
	end
end
 
print("# of rects = "..nrect)
 
local mesh = Mesh.new()
 
mesh:setVertexArray(vertices)
mesh:setIndexArray(indices)
mesh:setPosition(768/2, 1024/2)
 
stage:addChild(mesh)
 
local function onEnterFrame()
	for i = 1,#vertices,8 do
		local ind = (i - 1) / 4 + 1
		local vx = velocities[ind]
		local vy = velocities[ind+1]
 
		vertices[i+0] = vertices[i+0] + vx
		vertices[i+1] = vertices[i+1] + vy
		vertices[i+2] = vertices[i+2] + vx
		vertices[i+3] = vertices[i+3] + vy
		vertices[i+4] = vertices[i+4] + vx
		vertices[i+5] = vertices[i+5] + vy
		vertices[i+6] = vertices[i+6] + vx
		vertices[i+7] = vertices[i+7] + vy
	end
 
	mesh:setVertexArray(vertices)
end
 
stage:addEventListener(Event.ENTER_FRAME, onEnterFrame)

This code creates and explodes 11073 rects and runs at ~15 fps on iPad 3.

* If I comment out the explosion calculation code, it runs at 60fps.
* If I explode 1000 random rects at each frame, it runs at 60fps.
* Without explosion calculation code, it's possible to display 5 * 11073 = 55365 rects at 60 fps.

Conclusion: if you try to move >10000 rects, Lua can be really a bottleneck.
Guess: If the explosion calculation code is developed in C/C++ as a plugin, it may be possible to explode ~50000 rects at 60fps.

hnim · October 2012

@atitlim: that's really Lua problem. I simple convert your code to object model, frame rate drops significant (from ~60 to 45 on Desktop Player).

i attached my test project here if someone can take a look.

Howdy, Stranger!

Categories

In this Discussion

Top Posters

Improve performnace with Mesh?

Comments

Howdy, Stranger!

Quick Links

Categories

In this Discussion

Top Posters

Improve performnace with Mesh?

Comments