Hello,
Is there some way to access and use the rasterization hardware (i.e. the ROPs) on NVIDIA GPUs using CUDA? I was wondering if there is an API for this or even a set of assembly/PTX instructions that can enable this.
No API exists to allow a compute shader (CUDA, DX, or VK) to access ROP.
Is there any plan to expose it in future updates?
NVIDIA representatives posting in these forums have repeatedly stated that they will not comment on future products or product features.
Based on historical observation across the 15+ years of CUDA’s existence, the answer is NO.
You probably have to combine CUDA programs with a separate pixel shader.
Pixel shaders are also translated into SASS, so there exist assembly instructions for ROP access, but they are not allowed in/working for compute shaders.