This repo is for developing integration of GPU acceleration through CUDA for McVine.
The following dependencies are required to run the code in this repository:
- CMake 3.8 or newer
- CUDA 8.0 or newer
- This project supports separable compilation of device code. As a result, this should only be run on a Nvidia GPU with a Compute Capability of 2.0 or higher. You can find a list of GPUs and their Compute Capabilities here. Note: The hyperlinked website does not list all CUDA capable Nvidia GPUs. Some GPUs that aren't listed (i.e. GeForce GTX 1050 Mobile) can still be used.
- A C++ compiler that supports C++11
The code in this repository is organized as follows:
bash: This directory stores bash scripts that can be used to easily compile and run the code.cuda: This directory stores the main CUDA code for the repository, such as- CUDA Kernels
- CUDA Device Code
- Host (CPU) Code that calls CUDA Kernels
src: This directory stores any plain C++ (no CUDA) code that is needed. Currently, it only contains the main function for the non-test code.test: This directory stores the unit tests for the code base. All these tests are made with googletest.test_separate_compilation: This directory is not part of the main repository. It is an example directory that shows how to manually compile a CUDA project with separable compilation for device code.