SYCL Containers

Self-contained containers for SYCL developement, originally written for personal usage.

llvm-cuda

Contains:

• llvm 15.0.0
• DPC++
• CUDA 14.4
• oneMKL
• Lapack
• oneTBB
• dpct (c2s)
• boost 1.65

hipSYCL

Contains:

• hipSYCL (latest)
• CUDA 14.4
• llvm 13
• oneMKL
• boost 1.65

oneAPI-cuda

Contains:

• intel oneapi basekit (oneMKL, oneDPL, ...)
• CUDA 14.4
• llvm 15.0.0
• boost 1.65

Getting started

Before building the container, please make sure you have the NVIDIA container toolkit properly set up. If you don't, first setup the package repository and the GPG key:

distribution=$(. /etc/os-release;echo $ID$VERSION_ID) \
      && curl -fsSL https://nvidia.github.io/libnvidia-container/gpgkey | sudo gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg \
      && curl -s -L https://nvidia.github.io/libnvidia-container/$distribution/libnvidia-container.list | \
            sed 's#deb https://#deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://#g' | \
            sudo tee /etc/apt/sources.list.d/nvidia-container-toolkit.list

then just run

sudo apt-get update
sudo apt-get install -y nvidia-docker2

and eventually restart the Docker daemon

sudo systemctl restart docker

Build

Let's assume you want to build llvm-cuda. Then run

docker build -f llvm-cuda/Dockerfile -t mspronesti/sycl-container:llvm-cuda .

The image requires roughly 23 GB, make sure to have enough space. Also beware it takes roughly 2 hours to build.

Run

Once the image is build, just run it as follows

docker run -it --gpus all mspronesti/sycl-container:llvm-cuda bash

Build a SYCL application on NVidia GPU

Once you got a shell, write your code or clone it from wherever it is and compile it making sure to use clang++ as compiler and to specify the following compilation options, among the others, in you Makefile or CMakeLists

-fsycl -fsycl-targets=nvptx64-nvidia-cuda

Build a SYCL application using HipSYCL on NVidia GPU or OMP.

Compile it using syclcc, specifying the extra option --hipsycl-gpu-arch. For instance

syclcc --hipsycl-gpu-arch=sm_75 dummy.cpp -o dummy

If you want instead to run it on CPU, specify --hipsycl-targets.

syclcc --hipsycl-targets=omp dummy.cpp -o dummy

Alternatively, export HIPSYCL_TARGETS before compiling. For instance

export HIPSYCL_TARGETS=cuda:sm_75