XFLUIDS is a parallelized SYCL C++ solver for large-scale high-resolution simulations of compressible multi-component reacting flows. It is developed by Prof. Shucheng Pan's group at the School of Aeronautics, Northwestern Polytechincal University.

main developers:

• Jinlong Li ([email protected])
• Shucheng Pan ([email protected])

other contributors:

• Yixuan Lian, Renfei Zhang

If you use XFLUIDS for academic aplications, please cite our paper:

Jinlong Li, Shucheng Pan (2024). XFLUIDS: A SYCL-based unified cross-architecture heterogeneous simulation solver for compressible reacting flows. arXiv:2403.05910. (https://arxiv.org/abs/2403.05910)

• Support CPU, GPU (integrated & discrete), and FPGA without porting the code
• General for multi-vendor devices (Intel/NVIDIA/AMD/Hygon ... )
• High portability, productivity, and performace
• GPU-aware MPI
• Highly optimized kernels & device functions for multicomponent flows and chemical reaction
• ongoing work: sharp-interface method, curvilinear mesh, turbulence models ...

The following gpus have been tested:

• NVIDIA
  • Data center GPU: A100, P100
  • Gaming GPS: RTX 4090, RTX 3090/3080/3070/3060TI, T600, RTX 1080
• AMD
  • Data center GPU: MI50
  • Gaming GPS: RX 7900XTX, RX 6800XT, Pro VII
• Intel
  • Gaming GPS: ARC A770/A380
  • Integrated GPUs: UHD P630, UHD 750

1. libboost_filesystem as external lib while gcc internal filesystem is missing
2. one of two SYCL implentation:
   • AdaptiveCpp
   • Intel oneAPI as compiler
3. XFLUIDS automatically resolve boost and AdaptiveCpp dependencies, but oneAPI need to be installed manually if you use Intel oneAPI.

• 1.1.1. install boost-version-1.83(needed by AdaptiveCpp)

• 1.1.2. install AdaptiveCpp

• 1.1.3. add bin, libs and includes of AdaptiveCpp and dependencies to ENV PATHs

• 1.1.4. XFLUIDS use find_package(AdaptiveCpp) targetting AdaptiveCpp compile system, set cmake option AdaptiveCpp_DIR

  cmake -DAdaptiveCpp_DIR=/path/to/AdaptiveCpp/lib/cmake/AdaptiveCpp ..

• 1.1.1.intel oneapi version >= 2023.0.0 as compiler

• 1.1.2.codeplay Solutions for NVIDIA and AMD backends if GPU targets are needed

• 1.1.3.1.activate environment for oneAPI appended codeplay sultion libs

  source /opt/intel/oneapi/setvars.sh  --force --include-intel-llvm

• 1.2.3.2.or you can use the script files(only basic environments are included)

  source ./scripts/oneAPI/oneapi_base.sh

$ acpp-info
=================Backend information===================
Loaded backend 0: OpenMP
  Found device: hipSYCL OpenMP host device
Loaded backend 1: CUDA
  Found device: NVIDIA GeForce RTX 3070
=================Device information===================
***************** Devices for backend OpenMP *****************
Device 0:
General device information:
  Name: hipSYCL OpenMP host device
  Backend: OpenMP
  Vendor: the hipSYCL project
  Arch: <native-cpu>
  Driver version: 1.2
  Is CPU: 1
  Is GPU: 0
***************** Devices for backend CUDA *****************
Device 0:
General device information:
  Name: NVIDIA GeForce RTX 3070
  Backend: CUDA
  Vendor: NVIDIA
  Arch: sm_86
  Driver version: 12000
  Is CPU: 0
  Is GPU: 1

$ sycl-ls
[opencl:acc:0] Intel(R) FPGA Emulation Platform for OpenCL(TM), Intel(R) FPGA Emulation Device 1.2
[opencl:cpu:1] Intel(R) OpenCL, AMD Ryzen 7 5800X 8-Core Processor 3.0
[ext_oneapi_cuda:gpu:0] NVIDIA CUDA BACKEND, NVIDIA T600 0.0 [CUDA 11.5]

• set integer platform_id and device_id for targetting different backends("DeviceSelect" in json file or option: -dev)

  auto device = sycl::platform::get_platforms()[platform_id].get_devices()[device_id];
  sycl::queue q(device);

• CMAKE_BUILD_TYPE is set to "Release" by default, SYCL code would target to host while ${CMAKE_BUILD_TYPE}==Debug
• set INIT_SAMPLE as the problem being tested, path to "species_list.dat" and "reaction_list.dat" should be given to MIXTURE_MODEL
• MPI and AWARE-MPI support added in project, AWARE_MPI need specific GPU-ENABLED mpi version, details referenced in [4-mpi-libs]("4. MPI libs")
• VENDOR_SUBMIT allows throwing some parallism tuning cuda/hip model to their GPU, only supportted by AdaptiveCpp compile environment

• 3.2.1.Build with cmake

• build with cmake

  cd ./XFLUIDS
  mkdir build && cd ./build && cmake .. && make -j

• 3.2.2.Local machine running

• XFLUIDS automatically read <XFLUIDS/settings/*.json> file depending on INIT_SAMPLE setting

  ./XFLUIDS

• Append options to XFLUIDS in cmd for another settings, all options are optional, all options are listed in [6. executable file options]("6. Executable file options")

  ./XFLUIDS -dev=1,1,0
  mpirun -n mx*my*mz ./XFLUIDS -mpi=mx,my,mz -dev=1,0,0

• 3.2.2.Slurm sbatch running on Hygon(KunShan) supercompute center

  cd ./XFLUIDS/scripts/KS-DCU
  sbatch ./1node.slurm
  sbatch ./2node.slurm

• cmake system of this project browse libmpi.so automatically in path of ${MPI_PATH}/lib, please export MPI_PATH to the mpi you want:

  export MPI_PATH=/home/ompi

  -- MPI settings:
  --   MPI_HOME:/home/ompi
  --   MPI_INC: /home/ompi/include added
  --   MPI_CXX lib located: /home/ompi/lib/libmpi.so found

• reading commits in src file: <${workspaceFolder}/src/read_ini/settings/read_json.cpp>

• Set "OutDAT", "OutVTI" as 1 in .json file

• import .dat files of all ranks of one Step for visualization, points overlapped between boundarys of ranks(3D parallel tecplot format file visualization is not supportted, using tecplot for 1D visualization is recommended)

• use paraview to open *.pvti files for MPI visualization(1D visualization is not allowed, using paraview for 2/3D visualization is recommended);

@misc{li2024xfluids,
      title={XFLUIDS: A SYCL-based unified cross-architecture heterogeneous simulation solver for compressible reacting flows}, 
      author={Jinlong Li and Shucheng Pan},
      year={2024},
      eprint={2403.05910},
      archivePrefix={arXiv}
}

XFLUIDS has received financial support from the following fundings:

• The Guanghe foundation (Grant No. ghfund202302016412)
• The National Natural Science Foundation of China (Grant No. 11902271)