🚀 CUDA hello world (#7)

Author: bstaber

.github/workflows/cpp-tests.yml

@@ -55,7 +55,7 @@ jobs:
         shell: spack-bash {0}
         run: |
           spack env activate lobxp
-          cmake -S . -B build -DCMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE}
+          cmake -S . -B build -DCMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE} -DBUILD_CUDA=OFF
 
       - name: Build
         run: cmake --build build -j $(nproc)

README.md

@@ -4,4 +4,62 @@
 
 The purpose of this repository is to provide a space for exploring numerical methods, algorithms, and patterns in C++ in a practical and modular way. It is not intended as a tutorial or comprehensive learning resource, but rather as a set of working examples and references.
 
-Contributions are welcome. If you spot an issue, find an area that could be improved, or want to add your own example, feel free to open an issue or submit a pull request.
+Contributions are welcome. If you spot an issue, find an area that could be improved, or want to add your own example, feel free to open an issue or submit a pull request.
+
+# Developper guide
+
+## Spack
+
+We use spack to manage dependencies and build the projects. To get started, clone the repository and run:
+
+```bash
+git clone https://github.com/bstaber/cppplorers.git
+cd cppXplorers
+spack env activate .
+spack concretize -f
+spack install
+```
+
+This will set up the environment and install the necessary packages. You can then enter the environment with:
+
+```bash
+spack env activate .
+```
+
+Note that we set the CUDA dependency externally, so if you want to use CUDA, make sure to have it installed and configured properly.
+
+## Building
+
+Each crate has its own `CMakeLists.txt` file and can be built independently. However, we provide a top-level `CMakeLists.txt` that can be used to build all crates at once. To build all crates, run:
+
+```bash
+just
+```
+
+This will run the `justfile` which builds all crates and runs all tests.
+
+## Adding a new crate
+
+To add a new crate, create a new directory under `crates/` and add a `CMakeLists.txt` file. You can use one of the existing crates as a template. Make sure to update the top-level `CMakeLists.txt` to include your new crate.
+
+## Documenting a crate
+We use `mdbook` to document each crate. You can find the book source files in the `books/` directory.  To build and serve the book, run:
+
+```bash
+just serve-book
+```
+
+This will build the book and serve it at `http://localhost:3000`. You can then navigate to the book in your web browser.
+
+## Just
+
+We use `just` as a task runner to automate common tasks. You can find the `justfile` in the root directory. Some common tasks include:
+
+- `just build`: Build all crates
+- `just test`: Run all tests
+- `just serve-book`: Build and serve the documentation book
+- `just clean`: Clean all build artifacts
+- `just fmt`: Format all code using `clang-format`
+- `just lint`: Run `clang-tidy` on all code
+- `just configure`: Configure all crates with CMake
+- `just`: Configure, build, and test all crates

crates/CMakeLists.txt

@@ -1,3 +1,6 @@
 add_subdirectory("kf_linear")
 add_subdirectory("simple_optimizers")
-add_subdirectory("ar_models")
+add_subdirectory("ar_models")
+if(BUILD_CUDA)
+  add_subdirectory("cuda_hello")
+endif()

crates/cuda_hello/CMakeLists.txt

@@ -0,0 +1,15 @@
+cmake_minimum_required(VERSION 3.25)
+project(cuda_hello LANGUAGES CXX CUDA)
+
+set(CMAKE_CUDA_STANDARD 17)
+set(CMAKE_CUDA_STANDARD_REQUIRED ON)
+
+set(CMAKE_BUILD_RPATH_USE_ORIGIN ON)
+set(CMAKE_INSTALL_RPATH_USE_LINK_PATH ON)
+
+add_executable(cuda_hello main.cu)
+
+find_package(CUDAToolkit REQUIRED)
+target_link_libraries(cuda_hello PRIVATE CUDA::cudart)
+
+set_property(TARGET cuda_hello PROPERTY CUDA_ARCHITECTURES native)

crates/cuda_hello/main.cu

@@ -0,0 +1,15 @@
+#include <cstdio>
+
+// Kernel function that runs on the GPU
+__global__ void hello_kernel() {
+    printf("Hello from GPU thread %d in block %d!\n", threadIdx.x, blockIdx.x);
+}
+
+int main() {
+    // Launch 2 blocks with 4 threads each
+    hello_kernel<<<2, 4>>>();
+    cudaDeviceSynchronize();  // wait for GPU to finish
+
+    printf("Hello from CPU!\n");
+    return 0;
+}

spack.lock

@@ -6,7 +6,8 @@ spack:
   specs:
     - eigen@3.4
     - catch2@3
-
+    - cuda
+    - cmake
   packages:
     all:
       target: [x86_64]
@@ -15,6 +16,11 @@ spack:
         - spec: cmake@3.22.1
           prefix: /usr
       buildable: false
-
+    cuda:
+      externals:
+        - spec: cuda@12.0
+          prefix: /usr/local/cuda-12.0
+      buildable: false
   config:
-    install_tree: $spack/opt/spack
+    install_tree:
+      root: $spack/opt/spack