Liouville Solver

In this repository we have a CUDA solution for the computation of the solution to the Liouville Equation in general dimensions. We will explain the steps so you can familiarize with the code and test it for yourself! (Any optimizations/errors, please let me know)

REQUIREMENTS

• General:

  • At least 2 cores
  • An NVIDIA graphics card (compute capability >= 6.0)
  • Windows or any Linux distro (not tested in all)
  • Python (>3.0) installed

• Linux/WSL - specific:

  • g++ compiler (g++ >=11.3.0).
  • nvcc compiler (nvcc >= 11.5).
  • GNU Make >= 4.3.
  • CMake >= 3.22 (required).
  • Boost Library (>= 1.81.0).
  • save it in the folder: /usr/include/ or in the same directory as the project, under the names boost, or boost_dir.

• Windows - specific:

  • C++ compiler: MSVC (>= 19.36).
  • CUDA compiler: NVIDIA >= 12.2.91.
  • CMake >= 3.22 (required).
  • Boost Library (>= 1.81.0) .
  • save its include folder in the include folder of MSVC or in the same directory as the project, under the names boost, or boost_dir.

Create CUDA experiment from .json file and python function (easier)

This method is the easiest and assumes that you have Python installed (used Python 3.12, but works with previous versions).

• Create the .json file:
  • There is an example in Definition_examples/vdp_tests.json.
  • In the saving key, we introduce the kind of saving we want. Y means that we save all the frames, and thus the two following keys (first_frame and last_frame) can be left at 0, or any other value.
  • All other parameters are self-explanatory (will be detailed).
• To launch the simulation:
  • First verify that the output folder exists. If not, create it.
  • Then, open the terminal at the directory level.
  • Type: python setSimulator.py "<your_json_filename>.json"
  • If your json example definition contains multiple experiments, it will compile, build and execute them sequentially.

Create CUDA experiment '.cuh' file and launch from terminal (more complicated)

• Set the build system in the CMakeLists.txt file (Linux or Windows)

• To initialize the project:

  • LINUX Type make with the specific option you want:
  • make compile_X: erase the previous build and output folders and compiles the solution.
  • make total_X: Same as before, but runs the executable afterwards.
  • make update_X: Updates files in the build directory and executes the executable.
    • Use compile and total on the first time it is executed in a computer, and use update afterwards.
    • The executable is generated in ./build/X.
    • Options for X are debug and release.
  • WINDOWS Type cmake -S . -B ./build After it's done, type cmake --build ./build --config Y (Y can be either Debug or Release, without the quotes)

• To run the simulation (without the make commands), go to ./build/app/Y and then call the executable:

  • ./Simulation for Linux/WSL
  • start Simulation.exe for Windows

• There are several examples in 1, 2 and even 3 dimensions with several combinations of parameters, impulse terms, etc. I believe it is well-explained, but feel free to experiment with different combinations. Note that different combinations may imply that time-stepping, number of parameter samples or even domain size must be changed.

• DO NOT CHANGE NAMES OF VARIABLES in the Case_definition.cuh document.

• The simulator will adapt to your GPU memory availability as long as you have the amount of memory necessary to hold the corresponding number of variables for 1 parameter sample (that is, if there is memory for 1 parameter sample, then you can run for 400, for example).

• Note that the variable used for storage will grow very fast when performing simulations in >2D for a large amount of timesteps.

• Added the possibility of storing full, or partial information in separate .bin (binary) files, where each file can be 4 GB at most. This can also be changed in the Case_definition.cuh file. Files are stored in parallel, but the naming makes sure that the simulation is stored correctly.

• There are Matlab scripts to visualize the information obtained from the simulation (currently it is separated for 1, 2 and 3D cases). Feel free to experiment with the visualization tools.

For any other doubts, feel free to ask for help at: vizenzo97@gmail.com or vibees@upv.es