Fast way to render lots of spheres

fast_spheres.txt

Setup:
1. Index buffer containing N quads (each 2 triangles), where N is the max amount of spheres. Repeating pattern of {0,1,2,1,3,2} + K*4.
2. No vertex buffer.

Render N*2 triangles, where N is the number of spheres you have.

Vertex shader:
1. Sphere index = N/4 (N = SV_VertexId)
2. Quad coord: Q = float2(N%2, (N%4)/2) * 2.0 - 1.0
3. Transform sphere center -> pos
4. Calculate the optimal oriented bounding box of the sphere. See: http://www.iquilezles.org/www/articles/sphereproj/sphereproj.htm
5. Output bounding box vertex (xy): pos + bbox_xaxis * Q.x + bbox_yaxis * Q.y
6. Bias depth output by sphere radius. (we want front of the sphere)

Pixel shader:
1. Calculate ray-sphere intersection: https://en.wikipedia.org/wiki/Line%E2%80%93sphere_intersection
2. If ray misses sphere, discard the pixel.
3. Calculate surface position and normal using the ray-sphere intersection result.
4. Calculate all g-buffer outputs, shading, etc that you need.
5. Output depth calculated with ray-sphere intersection. 
6. Use SV_DepthGreaterEqual conservative depth output. This way you get HiZ culling (measured up to 6x speedup vs SV_Depth)

Preferably sort spheres front to back, if you have lots of overdraw.

If you have HUGE amount of overdraw, use the two-phase occlusion cull described in page 53 of this SIGGRAPH 2015 presentation:
http://advances.realtimerendering.com/s2015/aaltonenhaar_siggraph2015_combined_final_footer_220dpi.pdf

hello! so where do you get the sphere centers from?

In my case I have a raw buffer written by my GPGPU simulators. I use this technique for debug views of SPH fluids and PBD physics simulation.

In DirectX, you can use one of the following:

Buffer<float4> spheres;
ByteAddressBuffer spheres;
StructuredBuffer<float4> spheres;

Store sphere position to xyz and radius to w. SV_VertexId % 4 gives you the index to address the sphere buffer.

Hello. Could you elaborate on the way you calculate the rays and possibly the ray origin? I get the vertex shader to work to the point where I can render oriented quads, but I'm struggling with implementing the pixel shader. Thank you for compiling this guide btw.