This repo works to launch a CUDA kernel without using the CUDA driver runtime.
- Install the Python requirements
- Install clang, and update
pstruct.py
with location of libclang.
git clone https://github.com/NVIDIA/open-gpu-kernel-modules.git
./cudaless.sh
gpu_driver.cc
-- The thing that pushes the command queues and stuff
tc_context.cc
-- GPU init, replacement for cuInit and cuCtxCreate