gl_FragCoord has different orientation while rendering into viewport and texture buffer

[Vladimir Sabantsev]

Outputting vec4(gl_FragCoord.xy / DrawViewport.zw, 0.f, 1.f) as outColor[0] for all the scene geometry fragments into viewport directly:


Outputting vec4(gl_FragCoord.xy / DrawViewport.zw, 0.f, 1.f) as outColor[0] for all the scene geometry fragments into texture buffer
and then assigning this texture values to the surface.basecolor:

I had to reorient gl_FragCoord to (gl_FragCoord.x, DrawViewport.w-gl_FragCoord.y) for one of the cameras in order to achieve the pixel-perfect dithering effect between two cameras.

[Josh]

You might want to try getting the camera projection matrix instead. I believe that will account for the flipping of the Y coordinate that happens, because OpenGL considers +1 to face up in screen coordinates.

If it's an ortho camera, just use the CameraProjectionViewMatrix variable.

If it's a perspective camera, I believe you can still get the ortho projection with this function: ExtractCameraOrthoMatrix(in uint cameraID)

[Josh]

Also see Shaders/Utilities/ReconstructPosition.glsl, as you might find some of those functions useful.

[Vladimir Sabantsev]

6 hours ago, Josh said:

You might want to try getting the camera projection matrix instead. I believe that will account for the flipping of the Y coordinate that happens, because OpenGL considers +1 to face up in screen coordinates.
If it's an ortho camera, just use the CameraProjectionViewMatrix variable.
If it's a perspective camera, I believe you can still get the ortho projection with this function: ExtractCameraOrthoMatrix(in uint cameraID)

Ok, not sure how it should help, but I will see if I can utilize projection view matrix for it. Currently I fix this problem for myself with the following trick:

if (CameraRange.x == 0.125f)
{
  // NOTE: gl_FragCoord has different orientation when drawn into texture buffer
  ivec2    FragCoord_Reoriented = ivec2(gl_FragCoord.x, DrawViewport.w - gl_FragCoord.y);
  ...
}

 

2 hours ago, Josh said:

Also see Shaders/Utilities/ReconstructPosition.glsl, as you might find some of those functions useful.

I saw these utility methods while thinking on how to reconstruct position from a depth cubemap. They work as transform from the screen to the world, but in the case stated what I need is just a consistency of the screen coordinates between drawing to the viewport and texture buffer render target.

I also have some guts feeling that depth pre-pass doesn't work while rendering to a texture buffer. Maybe I will make some minimal samples to reproduce both issues, but as for now it doesn't stop me moving towards more difficult scenarios for my visibility filter.

[Josh]

Okay, I think this problem is solved then. Thank you for reporting.