GPU-projects

GPU computing assignments and projects for graduate students. The repository uses CMake to build CUDA examples, standalone assignments, and a reusable set of CPU, GPU, and OpenGL support modules.

Repository structure

.
├── base/       Shared CPU, GPU, and graphics modules
├── examples/   Standalone CUDA examples
├── projects/   Course assignments and larger projects
└── cmake/      Submission-package helpers

The base/ directory is intentionally not a standalone CMake target. Each project lists the shared files it uses in its own CMakeLists.txt, so dependencies remain explicit and submission packages can include only the required files.

Requirements

• CMake 3.25 or newer
• A C++20 compiler
• CUDA Toolkit with nvcc and the CUDA runtime development files
• A CUDA-capable GPU and a compatible driver for running CUDA targets
• vcpkg, or equivalent pre-installed packages, when building targets that use third-party libraries

Dependency setup with vcpkg

The recommended setup is to clone vcpkg into the repository root:

git clone https://github.com/microsoft/vcpkg.git vcpkg

Bootstrap vcpkg once:

# Windows PowerShell
.\vcpkg\bootstrap-vcpkg.bat

# Linux/macOS
./vcpkg/bootstrap-vcpkg.sh

Then edit CMakeUserPresets.json.example. Replace each $env{VCPKG_ROOT}/scripts/buildsystems/vcpkg.cmake with ${sourceDir}/vcpkg/scripts/buildsystems/vcpkg.cmake, and copy the result to CMakeUserPresets.json:

cp CMakeUserPresets.json.example CMakeUserPresets.json

Choose a *-vcpkg configure/build preset afterwards. CMake uses the repository's vcpkg.json in manifest mode and installs the declared packages automatically for the selected triplet.

Configure and build

The committed CMakePresets.json contains provider-neutral presets for Visual Studio, GCC, and Clang. After following the vcpkg setup above, use one of the *-vcpkg presets.

On Windows, the recommended workflow is to open the repository directory directly in Visual Studio. In the CMake configuration selector, choose either windows-vs-debug-vcpkg or windows-vs-release-vcpkg, then build the desired target from Visual Studio.

From Developer PowerShell for VS in Windows Terminal, run the equivalent Windows commands:

cmake --preset windows-vs-release-vcpkg
cmake --build --preset windows-vs-release-vcpkg

On Linux, use the corresponding preset directly from a terminal:

# GCC release
cmake --preset linux-gcc-release-vcpkg
cmake --build --preset linux-gcc-release-vcpkg

For debug builds, replace release with debug in the preset name.

If dependencies are already installed and visible to CMake, use a provider-neutral preset such as linux-gcc-release instead:

cmake --preset linux-gcc-release
cmake --build --preset linux-gcc-release

Submission

Every project under projects/ provides a submit_<project-name> target. For example, to package project1:

On Windows, run the following command from Developer PowerShell for VS in Windows Terminal:

cmake --build --preset windows-vs-release-vcpkg --target submit_project1

On Linux:

cmake --build --preset linux-gcc-release-vcpkg --target submit_project1

The resulting archive is written below the selected preset's binary directory: out/build/<preset>/projects/<project-name>/submission/<project-name>.zip.

Project 0: CUDA device query

projects/project0 is a CUDA Runtime API device-query program. It reports properties and capabilities for every detected CUDA device, and checks peer access between devices when multiple GPUs are available.

Project 1: Mandelbrot renderer

projects/project1 is a Mandelbrot renderer with interactive OpenGL and one-shot offline PNG modes. It supports high-precision center/zoom values, configurable resolution, and 1x1 or 2x2 sampling. The PROJECT1_OFFLINE_ONLY CMake option builds it without the online OpenGL dependencies. See the project1 README for usage, command-line options, controls, and offline-build details.