WCHR-regent

Overview

This code implements the WCHR scheme (see docs/ for documentation about the scheme) to solve the Navier-Stokes equations for compressible fluid flow.

Pre-requisites

• The code is written in the Regent programming language based on the legion runtime.

• SuperLU to solve block-tridagonal matrices.

  • If you're unsure about how to compile SuperLU, make a build directory in the SuperLU source directory and from it run cmake .. -DBUILD_SHARED_LIBS=ON -DUSE_XSDK_DEFAULTS=ON -DCMAKE_INSTALL_PREFIX=<path-to-superlu-installation> -DCMAKE_BUILD_TYPE=Release and then do make; make install.

• HDF5 for file I/O. Make sure Regent is also installed with HDF5 support.

Compiling the code

To compile the code, first navigate to the /src/ directory. Ensure that the problem.rg symlink is linked to the right problem file in the src/problems/ directory. In the problem file, you can set the number of grid points in each direction, the time stepping settings and set the initial conditions. Also set the SUPERLU_PATH environment variable to the location of the SuperLU installation. To use the HDF5 I/O features, set USE_IO=1 and set the HDF_ROOT environment variable to the path of your HDF5 library installation to use the file I/O features.

Compile the code to an executable using SAVEOBJ=1 <path/to/regent.py> main.rg. This will generate the executable called wchr. Note that the problem size is a compile time constant and changing the problem size requires recompilation.

Running the code

If the code is compiled to an executable, run the code using ./wchr -ll:cpu <number of CPUs> [OPTIONS]. Else, run the code directly using <path/to/regent.py> main.rg -ll:cpu <number of CPUs> [OPTIONS].

OPTIONS
  -h               : Print the usage and exit.
  -prefix {prefix} : Use {prefix} as prefix for file I/O.
  -stats {nstats}  : Print stats only every {nstats} steps.
  -p {value}       : Set the number of parallel tasks to {value} (default = 1).

Changing the problem

To change the problem you want to run, simply link the problem.rg symlink to a new problem file using ln -s <path/to/new/problem.rg> problem.rg.

TODO

• Add non-periodic boundary conditions (possibly with ghost cells)
• Custom block-tridiagonal solver to remove SuperLU dependency and reduce round-off errors
• Get the SPMD transformation working
• Make the code GPU compatible (at least the interpolation and solve routines)

Authors

• Akshay Subramaniam ([email protected])
• Man Long Wong ([email protected])