glCopyTexImage

[powerpad]

I render a scene to my viewport and create a rectangular texture. I want to render the viewport to that rectangular texture which works just fine if i define the texture's format to be rgba, however, if the format is rgba32f_ so a 32bit floating point texture the rendering speed drops amazingly.

can someone tell me please what i did wrong or how to do it better (with respect to glCopyTexImage)

here is some code


//set up texture
Gl.glGenTextures(1, out this.iTexture);

Gl.glBindTexture(Gl.GL_TEXTURE_RECTANGLE_ARB, this.iTexture);

Gl.glTexEnvi(Gl.GL_TEXTURE_ENV, Gl.GL_TEXTURE_ENV_MODE, Gl.GL_MODULATE);
Gl.glTexParameteri(Gl.GL_TEXTURE_RECTANGLE_ARB, Gl.GL_TEXTURE_MIN_FILTER, Gl.GL_LINEAR);
Gl.glTexParameteri(Gl.GL_TEXTURE_RECTANGLE_ARB, Gl.GL_TEXTURE_MAG_FILTER, Gl.GL_LINEAR);
Gl.glTexParameteri(Gl.GL_TEXTURE_RECTANGLE_ARB, Gl.GL_TEXTURE_WRAP_S, Gl.GL_CLAMP_TO_EDGE);
Gl.glTexParameteri(Gl.GL_TEXTURE_RECTANGLE_ARB, Gl.GL_TEXTURE_WRAP_T, Gl.GL_CLAMP_TO_EDGE);

Gl.glEnable(Gl.GL_TEXTURE_RECTANGLE_ARB);
//rendering stuff Gl.glBindTexture(Gl.GL_TEXTURE_RECTANGLE_ARB, this.iTexture);
//render scene Gl.glCopyTexImage2D(Gl.GL_TEXTURE_RECTANGLE_ARB, 0, Gl.GL_RGB32F_ARB, 0, 0, 640, 480, 0);

[jide]

Just use RGB and not RGBA. There's absolutely no need for the alpha channel that consumes memory but is not used at all.

Hope that helps.

[powerpad]

it still runs slow (if one could call that running)

[carl_lewis]

try using glCopyTexSubImage(...);
it's much much quicker!



the RGBA32 format uses 32 bits per channel, so that's 4x the memory access of a standard RGBA format. it should take longer.

[Jan]

Is your framebuffer RGBA8 or RGBA32 ?

[powerpad]

I setup the framebuffer with the normal pixel format like PFD_TYPE_RGBA (I hope that this is what you mean) or so ... should I use something different?

[Omaha]

I think they are referring to the format and internal_format parameters of all of the glTex*Image* calls.

[powerpad]

well the glcopyteximage() call is visible in my first post

[Overmind]

No. The question was what pixel format does your framebuffer have?

Because if the framebuffer has RGB(A)8 fixed point format, the glCopyTex(Sub)Image call with internal format RGBA32F has to do a format conversion (in the worst case this has to be done on the CPU), while with internal format RGBA8 it can simply copy the data...

[powerpad]

How I can I set this up or query for the framebuffer format?

[Jan]

The question is: Why do you want to copy RGBA8 data from your framebuffer into a RGBA32 texture? You don't gain anything from that.

However, if you want to render 32 Bit data to your texture, you should check out the "Framebuffer Object" (FBO) extension.

Jan.

[powerpad]

I want to implement some algorithm that is described in a paper. Therein geometry is processed by a pair of shaders (that forwards the x,y,z coordinate of a vertex to the fragment shader - the fragment shader writes x,y,z to r,g,b color values - this I figured I need a floating point texture).