1 reply to this topic

[Hornsj3]

Hi,

I am implementing GPU ray-cast terrain in Directx11, rendered based on heightmap data.

In order to make it more efficient I am attempting to implement the algorithm outlined in the following paper. http://www.tevs.eu/project_i3d08.html


My experience with DirectX is limited. I was wondering what the best approach would be to populate the mip levels of the texture pyramid outlined in the paper. Is this something to be done on the CPU side, or the GPU side? Can anyone think of a specific implementation for this?


As far as the paper I linked goes I'm just looking for the way to implement the mip-map data structure.


******************************************************Process Description **********************************************
1. Create a Texture of NxN.
2. Directx generates the mipmap levels.
3. I overwrite the auto generated mip map values with custom values according to the algorithm.
***************************************************************************************************************************

Are these steps best done in a compute shader, rendered as individual render targets per mip level, or on the CPU? An implementation hint would also help.



Thanks,

Josh

[Hornsj3]

Maybe I should rephrase this.

I'm wondering what the recommended way to populate mip-map data is.

Do I load an image file as a resource, calculate the subresource data, set the descriptor to generate mip levels, and assign the resource data to the texture object? This is a CPU solution.

Is there a better GPU solution?