readme_gpu.md

Software Requirements:

CMake cmake.org:

• minimum version 3.13

CUDA developer.nvidia.com/cuda-zone:

• Minimum CUDA Version 9.0
• Minimum Compute Capability 3.0, because of maximal number of Blocks in x direction
• Recommended Compute Capability 6.0, because of atomics for double precision floating point data (GKS only)

Paraview www.paraview.org:

• any version, for example the most recent

C++ Compiler:

• with C++11 support, for example gcc6.3 or Visual C++ 14.0

How to get VirtualFluidsGPU:

Option 1: use git

1. checkout out https://git.irmb.bau.tu-bs.de/VirtualFluids/VirtualFluidsGPU.git with your credentials

Option 2: without git

1. go to git.irmb.tu-bs.de
2. Log in with your credentials
3. click on VirtualFluids/VirtualFluidsGPU
4. click on the download symbol on the top right and download zip/tar.gz file

How to build VirtualFluidsGPU:

1. CMake the project
2. set the output path in targets/apps/LidDrivenCavity/LidDrivenCavity.cpp
3. build the project ("compile")
4. run the generated executable (usually in /bin/)

Known Issues:

If CMake does not find CUDA_CUT_INCLUDE_DIR use and set the correct CUDA Pathes in CMakeLists.txt in the base directory in lines 35, 36.

VirtualFluidsGPU results files:

VirtualFluidsGPU generates a the time series of output files directly in the output path. In Paraview these time series can be read directly.

VirtualFluidsGPU change between Double and Single Precision:

Option 1:

1. go to CMakeLists.txt in the base directory
2. go to line 83 to switch ON/OFF VF_DOUBLE_ACCURACY

Option 2: Check/uncheck the item VF_DOUBLE_ACCURACY in CMake.

Documentation:

The doxygen generated documentation can be found here.