Fallback to CPU tile light assign

[shrekshao]

Finally it turns out that even with gather approach, we still cannot use shader to assign lights to tile in parallel, since one single thread cannot write multiple pixels to texture (assign multiple lights to buffer). MRT won't work cuz the number is too limited compared to num of lights.

I'm finally falling back to use cpu to assign lights to each tile. Reading from multiple textures (functioning as tile-lightIdx buffer) is still doable (like Sijie Tian's Tile-based deferred one).

Let me know if there's anything I'm missing.

[shrekshao]

Do you recommend to try WebCL for this project?

[likangning93]

I think Kai said WebCL is kind of dead. CPU-based light tiling seems like it'll be the best way to go :(

I'm pretty out of the loop here, but we definitely can't do something like, run a fragment shader over all the pixels in the tile gbuffer? Maybe each tile is 128x128 pixels, and we set a cap of 128 * 32 lights (pack each light's attributes into 6 floats or something). Each light could get a unique ID in the light gbuffer. That way, I think each thread would just have to compute which light/tile it is in charge of, compute if it intersects the tile, and write just that value as true/false.

[shrekshao]

Sounds like something worth a try. Actually we only need to store the light idx, no need to pack all attributes. The process as I understand goes like this:

• light culling step
  • render a full quad
  • tile size 16 * 16, use a RGBA float texture, so each tile has 1024 entries, our scene can have at most 1024 lights.
  • each thread in this tile handle overlapping test for one light. Write 1.0 for true, 0.0 for false.
• light accumulation step
  • each thread/pixel find its tile, then loop through the 16 * 16 RGBA buffer, if there's an 1.0, retrieve the light attributes with the index, and shade this pixel

Potential Problems/Limitations:

• Total light number depends on the tile size. 16x16 -> 1024 lights, 8x8 -> 256 lights. (not a big problem I think)
• Need one more pass to get min and max depth for each tile frustum planes for light culling (can be ommited if only use left, right, up, front planes for culling, in this case even depth prepass is not needed)
• At the final shading stage, will the if statement judging if a light is overlapping lead to serious branching issue which results in each thread still parsing through every light in the scene? (One good news is this light overlapping list is same for threads in one tile. But is this tile aligned perfectly with that launched by the shader?)

Overall I think it is worth a try. Thanks Gary. Let me know if there's any misunderstanding