The increasing parallelism brought by CUDA and GPU libraries benefits PCL as well as other open-source communities greatly. However, the test for these GPU-accelerated code is a bit awkward on those common CI without free GPU. Therefore, building a generic tool to allow CPU to run CUDA without a GPU is inspiring and valuable. And integrating this tool into the CI of PCL will bring more convenience and confidence to CI build.
The toolkits used in this project will also provide an option for AMD users to build and run CUDA codes in AMD GPUs.
This proposal aims to build a generic tool to allow CPU to run CUDA code, and introduce this feature to PCL as necessary.