FSC CUDA Studies

Prerequisites

• CMake (3.24+)

  This is highly experimental code, so a very late CMake is expected.

  Follow this guide to install the latest CMake from standard repositories.

• ninja-build

  Ninja is known to be faster than Unix make. Install with sudo apt-get install ninja-build

• ccache

  Caches previous compilations to speed up compiling as much as possible

Building

We strongly recommending using the provided CMake preset. If you are using vscode, the CMake extension will choose the correct preset Linux X64 Default (instead of using compiler kits). If you are building from the command line, run

cmake -S . --preset=linux-default # NO -B build option
cmake --build --preset=build-debug # Or build-release

Running

A rich example of integrating an ensemble of dynamical systems is provided with plot output. Run

./build/out/examples/Release/run_simulation

❗ This examples calls up to python to do matplotlib plotting. Ensure you have matplotlib and numpy installed, and you are building and running the code under ths SAME python environment.

Testing

Two rudimentary tests are provided. Reading these tests is educational.

• test_basic_kernel_launch.cu fleshes out the parallel matrix addition example in the CUDA programming model article, adding correct input/output argument setup
• test_function_pointer.cu is inspired by the Pass Function Pointers to Kernels CUDA article but simplified, minimally demonstrating the difficult task of passing functions to kernels.

If you are using vscode, the CMake extension will help you choose the correct preset Run GTest unit tests then Run CTest button will correctly run all tests.

If you are testing from the command line, run

ctest --preset=test-gtest