JOGL TextureRenderer slow?

OpenGL Development JOGL Development
#1

Hi,

I’m still fighting to get some Graphics2D onto my JOGL scene.
I’m using TextureRenderer to create a small 100*100 texture. But then TextureRenderer.getTexture() is damn slow!
If I call it only 10 times in a row, the CPU usage jumps to over 60%.
I used “-Dsun.java2d.opengl=True”

What am I doing wrong? Isn’t there a simple and efficient way to draw some 2D graphics?

Bye
Chris

#2

maybe if you can post some of your code, we can look into this.

as for drawing 2D graphics, maybe the Overlay-class which uses the TextureRenderer is more useful.

#3

Hi emzic,

Here is my quick test-routine which I call every frame (Imagine I’d like to create animated 2D graphics).
I guess it does the same as the Overlay class (I took a look into the Overlay and TextureRenderer source to see how it works)

As I noted, when I call my “render2D” 10 times in a loop the CPU usage rises to over 60%.


  TextureRenderer renderer = null;

  int render2D(GL gl, GLDrawable gld)
  {

    if (renderer == null)
    {        	
   	renderer = new TextureRenderer(100,100,true);
    	renderer.setSmoothing(false);
    }
    
   
    Graphics2D g2 = renderer.createGraphics();

    g2.setBackground(new Color(0,0,0,0)); //transparent background
    g2.clearRect(0,0,100,100); //clear all
    
    g2.fillRect(0,0,10,10);
    g2.setPaint(new Color(99,0,255));
    g2.drawRect(0,0,22,22);

    g2.dispose();
    
    
    
    renderer.markDirty(0,0,500,500);     
    Texture tex = renderer.getTexture();
    tex.bind();
   
    
    //draw orthogonal quad. 
    int x = 45;
    int y = 45;
    int w = 100;
    int h = 100;
    
    y = gld.getHeight()-y;
    
    gl.glBegin(GL.GL_QUADS);
        
    gl.glTexCoord2i(0,1);
    gl.glVertex3d(x,y-h,0f);
    
    gl.glTexCoord2i(1,1);
    gl.glVertex3d(x+w,y-h,0f);
    
    gl.glTexCoord2i(1,0);
    gl.glVertex3d(x+w,y,0f);
    
    gl.glTexCoord2i(0,0);
    gl.glVertex3d(x,y,0f);
    
    
    gl.glEnd();
 

  	return 0;
  }
#4

Just something that seemed odd to me but why are you marking the TextureRenderer dirty from (0,0) to (500,500) since the width and height of the renderer are 100 each.

#5

That’s because it’s messy test code. :wink: . But it doesn’t solve the actual problem…

#6

Try not to create Graphics2D with each render cycle, but instead reuse it!

Not:


Graphics2D g2 = renderer.createGraphics();
...
g2.dispose();

But:


Graphics2D g2; // somewhere in field declaration

// and within render2D()
if (g2 == null) {
    g2 = renderer.createGraphics();
}
#7

Hi Martin,

sorry, but that doesn’t help :frowning:
I think renderer.getTexture() is the bottleneck…

#8

Could you try profiling it just to make sure?

#9

[quote]Could you try profiling it just to make sure?
[/quote]
I’ve never used a java-profiler before…
I can only say:
If I remove the call to getTexture the CPU usage is 7 times smaller.

#10

well getTexture() does update all of the texture that you have markedDirty before. so it does quite a bit of work.

to profile it, just put a System.nanoTime() before and after and measure the difference.

also try making the texturenderer resolution a power of two. 128x128 and not 100x100.
also, never mark something dirty that is outside this resolution since it will result in an openGL error which has a big impact on performance.

#11

Hi Zoomby,

I mean, what do you expect?

10 times

creating graphics content
Drawing a 100x100 picture
uploading that picture as texture
displaying texture

PER Frame.

I assume you are running with at least 30fps.