Hi guys,

I'm trying to approach the method of Dual Paraboloid Mapping, the method itself is integrated, now I'm thinking of optimizations, like can it be integrated in one render pass. Recently I have been informed that you can use multiple camera views in one pass with using the geometry shader, and then exporting to different render targets, whether this would be a performance improvement, I can't tell as I have never done it.

Question 1. How can the geometry shader be used to export multiple SV_Positions according to the render targets?

When reading a theoretical and practical paper on Dual Paraboloid Mapping by Jazon Zink, he explained the artifacts that can appear, which I receive myself.

Question 2. Is there a way to fake away these artifacts, or a proper method to?

Thank You, as always.

I just realized something, when coming close to the vertices, they somehow jump:

So when any vertex come to the corner of the screen, they jump away.

The implementation for rendering a single pass dual paraboloid map can be found in Hieroglyph 3 (see the MirrorMirror sample). In there, I use the geometry shader to render both paraboloid maps at once. To be honest, I never tried to do it in a two pass method, so I can't really give any real information about comparing the speeds. Some people stay away from the geometry shader at all costs, but I haven't really had any performance problems while using it. Just as FYI, there is a chapter in Practical Rendering and Computation with Direct3D 11 that deals with setting up the single pass rendering of dual paraboloid maps.

What is probably happening with your vertices is that once they go behind the paraboloid's center point then the method used to find the vertex position puts it on the opposite side of the paraboloid map. You should detect when this is the case and either clip the pixels affected or clamp the vertex position to the plane where z = 0 in paraboloid space.

Fixing the issues with primitive size is not really possible unless you tessellate things a bit better. You could use a stronger LOD system, or possibly use the tessellation stages to improve the tessellation based on paraboloid map size. But again, you may run into performance problems with the additional stages being active...

About the issue with the vertices, it still appears, here's what I do:

VS: // After applying world*view*proj
output.position = output.position / output.position.w;

// The decider between the front and back side
output.position.z *= _dpmSetting;

float L = length( output.position.xyz ); 
output.position = output.position / L;

output.z_value = output.position.z;

output.position.z = output.position.z + 1; 
output.position.x = output.position.x / output.position.z; 
output.position.y = output.position.y / output.position.z;

output.position.z = L / 500;							// set a depth value for correct z-buffering
output.position.w = 1;									// set w to 1 so there is no w divide

PS:
clip( input.z_value );

But no luck.

Do you have a culling mode enabled? Either back face or front face culling?

This is how I initialize my D3D11_RASTERIZER_DESC stage:

D3D11_RASTERIZER_DESC rasterizerState;
rasterizerState.FillMode = D3D11_FILL_SOLID;
rasterizerState.CullMode = D3D11_CULL_BACK;
rasterizerState.FrontCounterClockwise = false;
rasterizerState.DepthBias = false;
rasterizerState.DepthBiasClamp = 0;
rasterizerState.SlopeScaledDepthBias = 0;
rasterizerState.DepthClipEnable = true;
rasterizerState.ScissorEnable = true;
rasterizerState.MultisampleEnable = false;
rasterizerState.AntialiasedLineEnable = false;

But nothing changed, though I may have done it wrong.

Well after countless tries, I can't seem to solve the jumping of vertices.

output.position = output.position / output.position.w;

When this line is removed the problem no longer persists, but again, that creates another issue as the map doesn't become correct.

Creation of D3D11_RASTERIZER_DESC :
D3D11_RASTERIZER_DESC rasterizerState;
rasterizerState.FillMode = D3D11_FILL_SOLID;
rasterizerState.CullMode = D3D11_CULL_BACK;
rasterizerState.FrontCounterClockwise = false;
rasterizerState.DepthBias = false;
rasterizerState.DepthBiasClamp = 0;
rasterizerState.SlopeScaledDepthBias = 0;
rasterizerState.DepthClipEnable = true;
rasterizerState.ScissorEnable = true;
rasterizerState.MultisampleEnable = false;
rasterizerState.AntialiasedLineEnable = false;

dev->CreateRasterizerState(&rasterizerState, &RsRMAP);

Shader work:

VS
output.position = mul(position, worldMatrix);
output.position = mul(output.position, viewMatrix);
output.position = mul(output.position, projectionMatrix);

output.position = output.position / output.position.w;

// The decider between the front and back side
output.position.z *= _dpmSetting; // 1 front || -1 back

// Determine the distance between the paraboloid origin and the vertex.
float L = length( output.position.xyz );

// Normalize the vector to the vertex
output.position = output.position / L;

// Save the z-component of the normalized vector
output.z_value = output.position.z;

output.position.z = output.position.z + 1;
output.position.x = output.position.x / output.position.z;
output.position.y = output.position.y / output.position.z;
output.position.z = (L - 0.1)/(500.0-0.1);

output.position.w = 1;

PS
clip( input.z_value );

Now what on the earth have I done wrong... again...

As a test, I would say to change the color of the vertices based on their output.z_value attribute. If the value is greater than 0, then use one color, and if it is less than 0 use another color. This will help you identify which vertices you should be displaying and which ones should not be. I am willing to bet that the ones that you shouldn't see are the ones jumping to the other side. This geometry should be captured in the other paraboloid map, and hence be clipped from the current one. In fact, that is why it jumps - because the paraboloid transformation is only correct when z > 0!

Result:

HLSL:

if ( input.z_value > 0)
{
	output.Diffuse = float4(0, 1, 0, 1);
}
else
{
	output.Diffuse = float4(1, 0, 0, 1);
}

clip( input.z_value );

output.Lighting = float4(1, 1, 1, 1);

return output;

I must be doing something really stupid right now.