pkestene / ramsesgpu

Astrophysics MHD simulation code optimized for large cluster of GPU

WARNING: This code is mostly deprecated. Have a look at ppkMHD

Magneto Rotational Instability simulation in a shearing box setup (800x1600x800) made in 2013 on TGCC/CURIE using 256 GPUs. Here Reynolds number is 25000 and Prandtl number is 4.

RamsesGPU code

RamsesGPU website

http://www.maisondelasimulation.fr/projects/RAMSES-GPU/html/index.html

• See doxygen-generated documentation in doc sub-directory

• Quickstart for building RAMSES-GPU using CMake (recommended)

Default CUDA compilation flags can be passed to cmake using env variable CUDAFLAGS, or directly set CMAKE_CUDA_FLAGS on the configuration command line (see below).

0. git clone https://github.com/pkestene/ramsesGPU.git
1. cd ramsesGPU; mkdir build
2. cmake -DUSE_CUDA=ON -DUSE_MPI=ON -DCMAKE_CUDA_FLAGS="-arch=sm_50" ..
3. make

You should get executable ramsesGPU_mpi_cuda. Explore other flag using the ccmake user interface.

• Quickstart for building RAMSES-GPU using autotools (deprecated)

0. make sure to have up-to-date autotools on you build system (autoconf, automake, libtool, m4); then run sh autogen.sh
1. configure --with-cuda=
2. make (or make -j N to speed-up compilation time; you might need to execute make several times when using option -j)

Note: make sure to have CUDA toolkit installed, and environment variables PATH and LD_LIBRARY_PATH correctly set.

This will build the monoCPU / monoGPU version of the programme to solve hydro/MHD problems. Executable are located in src subdirectory and named euler_cpu / euler_gpu

Execute a test run:

	cd src
	./euler_gpu --param ../data/jet2d_gpu.ini

This will start a Hydrodynamics 2D jet simulation run. Output files can be in multiple file format (VTK, Hdf5, ...).

To visualize Hdf5 output, just run:

	paraview --data=./jet2d_gpu.xmf

Contact, questions, comments:

pierre.kestener at cea.fr

ramsesgpu's Issues

cmake error upon generate

Cmake reports an error after configuring, during generate stage, but afterward the project builds fine:

-- Configuring done
CMake Error at src/CMakeLists.txt:25 (add_executable):
  Cannot find source file:

    gpu_macros.cpp

  Tried extensions .c .C .c++ .cc .cpp .cxx .m .M .mm .h .hh .h++ .hm .hpp
  .hxx .in .txx


-- Generating done
-- Build files have been written to: /home/pkestene/etudes/GPU/Hydro/git/github/ramsesGPU/build_cmake_cpu

The problem could be related to cmake issue:
https://gitlab.kitware.com/cmake/cmake/issues/17462
To be investigated.

Develop alternative build system using cmake

Currently the system uses autotools; it's more than 10 years old.
It relies on some dirty tricks to handle cpp file as if they were cuda, in order to able either a cpu or gpu build.

Several issues makes the code difficult to build on most current linux platform.

Let's try to replace autotools with cmake.

Autotools broken

Autotools (top-level autoconf.ac and m4 macros) need to be refreshed to build again.
Let's fix that.