MiniTorch

Build

Create a separate build directory and configure with CMake from the project root.

cmake -S . -B build
cmake --build build

By default, the project tries to detect CUDA automatically. If you want to be explicit, use one of the options below.

Prerequisites

To reproduce the CUDA build on this setup, make sure these tools are installed:

sudo apt install cmake g++-12 nvidia-cuda-toolkit

If your CUDA installation lives somewhere else, adjust the compiler and toolkit paths in the commands below.

CPU-only build

cmake -S . -B build -DUSE_CUDA=OFF
cmake --build build

CUDA build with nvcc

Use this when CUDA is installed and nvcc is available. This toolchain also needs a supported host compiler; on this setup that means GCC/G++ 12.

cmake -S . -B build \
  -DUSE_CUDA=ON \
  -DCMAKE_CUDA_HOST_COMPILER=/usr/bin/g++-12
cmake --build build

If nvcc is not on your PATH, or CUDA lives in a non-default location, use:

cmake -S . -B build \
  -DUSE_CUDA=ON \
  -DCMAKE_CUDA_COMPILER=/usr/lib/nvidia-cuda-toolkit/bin/nvcc \
  -DCUDAToolkit_ROOT=/usr/lib/nvidia-cuda-toolkit \
  -DCMAKE_CUDA_HOST_COMPILER=/usr/bin/g++-12
cmake --build build

If you do not have GCC/G++ 12 installed yet, install it first and use that compiler for the CUDA configure step.

Run all tests

cd build
ctest --output-on-failure

ctest runs the full test suite that was generated for the current build. If you configured CUDA support, this includes the CUDA-backed tests as well.