- Update CUDA-Clang container to Clang 14.
- Add CUDA-aware MPICH there.
- Add hipSYCL build with MPI for CUDA.
- Bump hipSYCL version.
- Bump UCX from 1.10.1 to 1.13.0, since the old one had troubles compiling with new clang.
- Add weekly tests with 2 NVIDIA GPUs and GPU-aware MPI.
Pipeline with GPU-direct halo exchange: https://gitlab.com/gromacs/gromacs/-/pipelines/605876672
Pipeline with GPU-direct halo exchange with broken pack/unpack kernels, to make sure we're testing them: https://gitlab.com/gromacs/gromacs/-/pipelines/605878986