Volumetric Shader Help

[oyedoyin.dada]

Hello everyone,

I'm currently working on a fluid visualization project and I need to move our working GLSL implementation to Unigine. To get my bearings, I'm trying to setup a very simple test case where;

1: The shader takes in the vector minimum and maximum positions of the fluid volume

2: I check for collisions for rays starting at each pixel, looking in the camera direction with the volume. If there is a collision, I check the distance against the depth buffer at this point (to make sure the fluid isn't appearing over other scene objects)

3: The pixel colour is then specified based on a value read from the fluid result sent as a 3d texture to the shader.

I've gotten this test working with an HLSL shader in Unity as I hope to use that as a conversion guide(I have attached the script, shader and project files), but my current Cpp implementation is Unigine (script is also attached) isn't working. I can send the camera matrices and bind the shader to render but i'm unable to get the texture communication working.

I would really appreciate any help with this problem.

VolumeEffect.cs Volume.shader AppWorldLogic.cpp Shader_Playground.zip

[vovan675]

Hello! It’s not entirely accurate to write shaders directly in plain HLSL. We recommend using our shader API, and ideally, you should work with basemat and its passes. To bind a texture in your shader, use RenderState::setTexture (and be sure to use an INIT_TEXTURE declaration in the shader as well). For defining uniforms, you’ll need to use the CBUFFER declaration. Additionally, there are already pre-bound uniforms available, such as camera position and other useful data. You can refer to shaders\api\uniforms\common.h to see what’s already accessible.

Here’s an example of how to bind and use a color texture and depth buffer with shaders using our shader API

AppWorldLogic.cpp CustomVolume.basemat

[oyedoyin.dada]

Thank you very much! I'll check it out

[oyedoyin.dada]

Finally got it all to work! I had to convert the vertex position to clip space so the volume renders properly in the scene now. The depth buffer logic also works properly which fixes our initial problem :)

I've attached the files with the sample 3d fluid data in case someone else finds it useful

AppWorldLogic.cpp CustomVolume.basemat

[oyedoyin.dada]

Here's the raw render with the actual fluid data from the CUDA-based solver: 

 @vovan675 Do you have a recommendation for an efficient transfer of the data since both the compute and rendering are happening on the GPU? My current implementation requires extracting the CUDA buffer to a float array and then generating a 3D image from that.

[vovan675]

Hello. You should attempt to retrieve the external memory resource of a 3D texture using texture->getResourceExternalMemory() with the FORMAT_USAGE_SHARED flag. Then, use this resource to create a cudaExternalMemoryHandleDesc object. Next, generate a surface from it by calling cudaCreateSurfaceObject. After that, invoke a CUDA kernel function, passing the texture surface as a cudaSurfaceObject_t parameter. Finally, process the surface within the kernel using surf3Dwrite, and this approach should work effectively without any GPU->CPU transfers, writing directly to texture that used for rendering.