[solved] SpriteBatch vs multiple love.graphics.draw(). Why I got identical FPS?

[opex]

Please explain why fps is the same?
I wrote a small test, and I don't understand what happens...
I always thought that rendering through SpriteBatch should work faster than multiples love.graphics.draw().
Where is my mistake?

Code: Select all

local a,b,c
local q1,q2
local batch1,batch2,batch3
local mode=1
local yPos,direction=100,1
local t1={}
local t2={}

local function getModeName(m)
  if m==1 then
    return 'multiple draws'
  elseif m==2 then
    return 'stream batch'
  elseif m==3 then
    return 'dynamic batch'
  elseif m==4 then
    return 'static batch'
  end
end

function love.load()
  a=love.graphics.newImage('1.png')
  b=love.graphics.newImage('2.png')
  c=love.graphics.newImage('3.png')

  q1=love.graphics.newQuad(0,0,160,160,c:getWidth(),c:getHeight())
  q2=love.graphics.newQuad(160,0,160,160,c:getWidth(),c:getHeight())

  batch1=love.graphics.newSpriteBatch(c,2000,'stream')
  batch2=love.graphics.newSpriteBatch(c,2000,'dynamic')
  for i=1,1000 do
    t1[#t1+1] = batch2:add(q1,i,yPos)
    t2[#t2+1] = batch2:add(q2,i,yPos+200)             
  end
  batch3=love.graphics.newSpriteBatch(c,2000,'static')
  for i=1,1000 do
    batch3:add(q1,i,yPos)
    batch3:add(q2,i,yPos+200)             
  end
end

function love.draw()
  if mode==1 then --multiple draws
    for i=1,1000 do
      love.graphics.draw(a,i,yPos)
      love.graphics.draw(b,i,yPos+200)
    end    
  elseif mode==2 then --stream batch
    love.graphics.draw(batch1)
  elseif mode==3 then --dynamic batch
    love.graphics.draw(batch2)
  elseif mode==4 then --static batch
    love.graphics.draw(batch3)
  end
  love.graphics.print('click to change draw mode\nfps: '..love.timer.getFPS()..'\nmode: '..getModeName(mode))
end

function love.mousepressed()
  mode = mode+1
  if mode>4 then
    mode=1
  end
end

function love.update(dt)
  --change y position
  yPos=yPos+100*dt*direction
  if yPos<100 then
    direction=1
  elseif yPos>200 then
    direction=-1
  end

  --change stream batch
  batch1:clear()
  for i=1,1000 do
    batch1:add(q1,i,yPos)
    batch1:add(q2,i,yPos+200)             
  end

  --change dynamic batch
  for i=1,1000 do
    batch2:set(t1[i],q1,i,yPos)
    batch2:set(t2[i],q2,i,yPos+200)             
  end

end

[raidho36]

That's because by default FPS is locked with vsync and won't go any higher, and you're not drawing nearly enough sprites to saturate draw call queue for FPS to drop below that.

[opex]

I turned off vsync and nothing changed.
In my test love.graphics.draw() is called 2000 times every frame,
but SpriteBatch draws only one time. Why FPS identical?
Probably LOVE have internal optimisation mechanism in love.graphics.draw()?
What test should I write to see that the SpriteBatch really works faster?

[zorg]

opex: Try your test code with 40 000 images instead of a measly 1000.

[drunken_munki]

I ran your code with a conf file setting vsync off and with the multiple draw I had 100 fps, all other modes had 500 fps. However the last mode stopped the animation from moving around.

[opex]

I wrote another test, where I compare speed 400 different pngs vs 1 sprite atlas.
I saw that SpriteBatch works about 40 percent faster even on 1000 draws.
The question is resolved.

Interesting fact: there is no difference in speed when using 'stream','dynamic' or 'static' SpriteBatch

[zorg]

Interesting fact: there is no difference in speed when using 'stream','dynamic' or 'static' SpriteBatch

Time to write another test where you actually modify sprite properities or the sprites themselves in the batches, then you may see a difference between the SpriteBatch modes. (static should probably perform worse than dynamic, and dynamic worse than stream, if you modify all the sprites each frame)

[slime]

The amount of performance difference between the SpriteBatch and Mesh usage hints for a given use-case are extremely dependent on the graphics driver and hardware you use. So while there might be no difference for you on your computer, there will probably be a difference on a user's computer.