forge-indirect-drawing

GPU compute shaders perform frustum culling by writing visibility decisions to an indirect command buffer, enabling indirect drawing without CPU-side visibility tests. Compute shaders evaluate frustum planes and output only visible instances.

GPU-driven rendering offloads visibility decisions from the CPU to the GPU using compute shaders. It is needed for scenes with hundreds or thousands of mesh instances to minimize CPU overhead and reduce draw calls.

SDL_GPU supports indirect drawing via SDL_DrawGPUIndexedPrimitivesIndirect. This function enables dynamic visibility by reading indirect draw commands generated directly by a compute shader.

Portable per-object storage buffers are configured for the pipeline to read and write data. Both compute and vertex stages access these buffers to evaluate frustum planes and render visible instances.

CPU frustum culling creates bottlenecks by serializing visibility tests for many instances. GPU-driven indirect drawing solves this by offloading visibility decisions to compute shaders, bypassing CPU routing entirely.

The first_instance pattern is a portable technique used in GPU-driven indirect rendering to manage per-object data access. It allows the pipeline to correctly index storage buffers when reading indirect draw commands.