2 replies to this topic

[Starfox]

I'm a bit puzzled by this: I have a ground plane polygon that I draw as such:

const float Extent = 128;
    glBegin(GL_QUADS);
    glVertexAttrib4f(0, -Extent, 0, Extent, 1);
    glVertexAttrib4f(0, Extent, 0, Extent, 1);
    glVertexAttrib4f(0, Extent, 0, -Extent, 1);
    glVertexAttrib4f(0, -Extent, 0, -Extent, 1);
    glEnd();

With the following vertex shader aiming to generate screen aligned texture coordinates in the range [0, 1]:

float2 pack(float2 In)
{
    return (In + 1.0) / 2.0;
}

float4 main(in float4 Position : ATTR0, 
            uniform float4x4 WorldMatrix, 
            uniform float4x4 WorldViewProjectionMatrix, 
            out float2 oScreenSpaceTexCoord : TEXCOORD3) : HPOS
{
    float4 ProjectedCoord = mul(WorldViewProjectionMatrix, Position);
    oScreenSpaceTexCoord = pack(ProjectedCoord.xy / ProjectedCoord.ww);
	return ProjectedCoord;
}

and using the following visualization fragment shader:

float4 main(in float2 ScreenSpaceTexCoord : TEXCOORD3) : COLOR
{
    return float4(ScreenSpaceTexCoord, 0, 1);
}

The results tend to be in the range [0, 1] but they're messed up and they change depending on the camera - ie pixel at coords (x,y) can have a different texcoord based on the camera position, which is wrong - a pixel's texcoord should only depend on its Normalized Device Coordinates and by connection, screen coordinates. Any ideas?

[Hodgman]

IIRC - You've got to perform the perspective divide after interpolation (i.e. in the fragment shader), rather than doing it immediately in the vertex shader.

[Starfox]

Aha, that was it, thanks a million.