🔬 CUDA Project Template

A simple and extensible template for CUDA projects.

🌟 Features

• Easy CMake integration — build separate CPU and GPU static libraries.
• Smart GPU memory management via memory_cuda.cuh:
  • cuda::unique_ptr<T>
    A GPU-aware smart pointer that owns device memory allocated with cudaMalloc().
    Automatically releases it using cudaFree() when it goes out of scope.
  • cuda::shared_ptr<T>
    A reference-counted smart pointer that provides shared ownership of GPU-allocated memory. Multiple instances can safely share the same device pointer.
    This implementation is thread-safe, thanks to atomic reference counting.
  • Safe-by-default — dereferencing on the host side is explicitly disabled (operator*, ->, and [] are deleted).

---

📘 Example

#include "memory_cuda.cuh"

__global__ void kernel(float* data) {
    int idx = threadIdx.x;
    data[idx] *= 2.0f;
}

int main() {
    // Allocate GPU memory and wrap it in a smart pointer
    auto d_ptr = cuda::make_unique<float>(10);

    // Launch a kernel that uses the allocated device memory
    kernel<<<1, 10>>>(d_ptr.get());
    cudaDeviceSynchronize();

    return 0;
}

📑 License

This project is licensed under GPL-3.0 license. See the LICENSE file for the full license text.