2. Compile imate from Source with CUDA#
2.1. Install C++ Compiler and OpenMP#
Compile imate with either of GCC, Clang/LLVM, or Intel C++ compiler.
Install GNU GCC Compiler
sudo apt install build-essential libomp-dev
sudo yum group install "Development Tools"
sudo dnf group install "Development Tools"
Then, export C and CXX variables by
export CC=/usr/local/bin/gcc
export CXX=/usr/local/bin/g++
Install Clang/LLVN Compiler
sudo apt install clang libomp-dev
sudo yum install yum-utils
sudo yum-config-manager --enable extras
sudo yum makecache
sudo yum install clang
sudo dnf install yum-utils
sudo dnf config-manager --enable extras
sudo dnf makecache
sudo dnf install clang
Then, export C and CXX variables by
export CC=/usr/local/bin/clang
export CXX=/usr/local/bin/clang++
Install Intel oneAPI Compiler
To install Intel Compiler see Intel oneAPI Base Toolkit.
2.2. Install CUDA Compiler and Development Libraries#
Attention
The minimum version of CUDA to compile imate is CUDA 10.0.
To download and install the CUDA Toolkit on both Linux and Windows, refer to the NVIDIA Developer website. It’s important to note that NVIDIA’s installation instructions on their website include the entire CUDA Toolkit, which is typically quite large (over 6 GB in size).
However, for compiling imate, you don’t need to install the entire CUDA Toolkit. Instead, only the CUDA compiler and a few specific development libraries, such as cuBLAS and cuSparse, are required. Below are simplified installation instructions for Linux, allowing you to perform a minimal CUDA installation with only the necessary libraries. Note that in the following, you may change CUDA_VERSION to the CUDA version that you wish to install.
# Set to the desired cuda version
CUDA_VERSION="12-3"
# Machine architecture
ARCH=$(uname -m | grep -q -e 'x86_64' && echo 'x86_64' || echo 'sbsa')
# OS Version
UBUNTU_VERSION=$(awk -F= '/^VERSION_ID/{gsub(/"/, "", $2); print $2}' /etc/os-release)
OS_VERSION=$(dpkg --compare-versions "$UBUNTU_VERSION" "ge" "22.04" && echo "2204" || echo "2004")
# Add CUDA Repository
sudo apt update
sudo apt install wget -y
wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu${OS_VERSION}/${ARCH}/cuda-keyring_1.1-1_all.deb -P /tmp
sudo dpkg -i /tmp/cuda-keyring_1.1-1_all.deb
rm /tmp/cuda-keyring_1.1-1_all.deb
# Install required CUDA libraries
sudo apt-get update
sudo apt install -y \
cuda-nvcc-${CUDA_VERSION} \
libcublas-${CUDA_VERSION} \
libcublas-dev-${CUDA_VERSION} \
libcusparse-${CUDA_VERSION} \
libcusparse-dev-${CUDA_VERSION}
# Set to the desired cuda version
CUDA_VERSION="12-3"
# Machine architecture
ARCH=$(uname -m | grep -q -e 'x86_64' && echo 'x86_64' || echo 'sbsa')
# OS Version
OS_VERSION=$(awk -F= '/^VERSION_ID/{gsub(/"/, "", $2); print $2}' /etc/os-release)
# Add CUDA Repository
sudo yum install -y yum-utils
sudo yum-config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/rhel${OS_VERSION}/${ARCH}/cuda-rhel${OS_VERSION}.repo
# Install required CUDA libraries
sudo yum install --setopt=obsoletes=0 -y \
cuda-nvcc-${CUDA_VERSION} \
cuda-cudart-devel-${CUDA_VERSION} \
libcublas-${CUDA_VERSION} \
libcublas-devel-${CUDA_VERSION} \
libcusparse-${CUDA_VERSION} \
libcusparse-devel-${CUDA_VERSION}
# Set to the desired cuda version
CUDA_VERSION="12-3"
# Machine architecture
ARCH=$(uname -m | grep -q -e 'x86_64' && echo 'x86_64' || echo 'sbsa')
# OS Version
OS_VERSION=$(awk -F= '/^VERSION_ID/{gsub(/"/, "", $2); print $2}' /etc/os-release)
# Add CUDA Repository
sudo dnf install -y dnf-utils
sudo dnf config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/rhel${OS_VERSION}/${ARCH}/cuda-rhel${OS_VERSION}.repo
# Install required CUDA libraries
sudo dnf install --setopt=obsoletes=0 -y \
cuda-nvcc-${CUDA_VERSION} \
cuda-cudart-devel-${CUDA_VERSION} \
libcublas-${CUDA_VERSION} \
libcublas-devel-${CUDA_VERSION} \
libcusparse-${CUDA_VERSION} \
libcusparse-devel-${CUDA_VERSION}
Update PATH with the CUDA installation location by
echo 'export PATH=/usr/local/cuda/bin${PATH:+:${PATH}}' >> ~/.bashrc
source ~/.bashrc
Check if the CUDA compiler is available with which nvcc.
2.3. Load CUDA Compiler on GPU Cluster#
If you are compiling imate on a GPU cluster, chances are the CUDA Toolkit is already installed. If the cluster uses module interface, load CUDA as follows.
First, check if a CUDA module is available by
module avail
Load both CUDA and GCC by
module load cuda gcc
You may specify CUDA version if multiple CUDA versions are available, such as by
module load cuda/11.7 gcc/6.3
You may check if CUDA Compiler is available with which nvcc.
2.4. Configure Compile-Time Environment Variables#
Specify the home directory of CUDA Toolkit by setting either of the variables CUDA_HOME, CUDA_ROOT, or CUDA_PATH. The home directory should be a path containing the executable /bin/nvcc. For instance, if /usr/local/cuda/bin/nvcc exists, export the following:
export CUDA_HOME=/usr/local/cuda
To permanently set this variable, place the above line in a profile file, such as in ~/.bashrc, or ~/.profile, and source this file, for instance by
echo 'export CUDA_HOME=/usr/local/cuda${CUDA_HOME:+:${CUDA_HOME}}' >> ~/.bashrc
source ~/.bashrc
To compile imate with CUDA, export the following flag variable
export USE_CUDA=1
2.5. Enable Dynamic Loading (optional)#
When imate is complied, the CUDA libraries bundle with the final installation of imate package, making it over 700MB. While this is generally not an issue for most users, often a small package is preferable if the installed package has to be distributed to other machines. To this end, enable the dynamic loading feature of imate. In this case, the CUDA libraries do not bundle with the imate installation, rather, imate loads the existing CUDA libraries of the host machine at runtime. To enable dynamic loading, simply set:
export CUDA_DYNAMIC_LOADING=1
2.6. Compile and Install#
Get the source code of imate with
git clone https://github.com/ameli/imate.git
Compile and install by
cd imate
python -m pip install .