How to install redesign-prototye

Build Nektar++ e.g. on current master

• See README.md under Nektar++ repository
• For development, the Debug option is recommended
• Note that the SIMD options have to be turned off, if you want to use CUDA support (Should be by default).

Build Nektar++ redesign-prototype

In main directory of the redesign, create a build directory cd into it, configure and build via:

mkdir build
cd build
cmake ../ -DNektar++_DIR=~/code/nektarMaster/build/dist/lib64/nektar++/cmake -DCMAKE_BUILD_TYPE=Debug
make 

If boost is not available on system, build boost (e.g. v1.71.1)

• One option is via the Ubuntu package manager sudo apt install libboost-all-dev
• Another option is to download boost and install it manually

After installing boost, run again:

cmake ../ -DNektar++_DIR=~/code/nektarMaster/build/dist/lib64/nektar++/cmake -DCMAKE_BUILD_TYPE=Debug
make

When the installation was successful, check it via tests:

ctest --output-on-failure

Build Nektar++ redesign-prototype with CUDA

To build the redesign project with CUDA:

• The CUDA option must be turned on
• The SIMD option must be turned off

This can be obtained by using the following command

cmake ../ -DNektar++_DIR=~/code/nektarMaster/build/dist/lib64/nektar++/cmake -DCMAKE_BUILD_TYPE=Debug -DNEKTAR_USE_CUDA=ON -DNEKTAR_USE_SIMD=OFF -DCMAKE_CUDA_ARCHITECTURES=86
make

NOTE: The CMAKE_CUDA_ARCHITECTURES parameter must be set to the appropriate compute capability of the GPU. On Linux, the compute capbability can be obtained by using the following command

nvidia-smi --query-gpu=compute_cap --format=csv

Here is a list of CUDA compute capability:

• 6.0 -> -DCMAKE_CUDA_ARCHITECTURES=60
• 6.1 -> -DCMAKE_CUDA_ARCHITECTURES=61
• 6.2 -> -DCMAKE_CUDA_ARCHITECTURES=62
• 7.0 -> -DCMAKE_CUDA_ARCHITECTURES=70
• 7.2 -> -DCMAKE_CUDA_ARCHITECTURES=72
• 7.5 -> -DCMAKE_CUDA_ARCHITECTURES=75
• 8.0 -> -DCMAKE_CUDA_ARCHITECTURES=80
• 8.6 -> -DCMAKE_CUDA_ARCHITECTURES=86
• 8.7 -> -DCMAKE_CUDA_ARCHITECTURES=87
• 8.9 -> -DCMAKE_CUDA_ARCHITECTURES=89
• 9.0 -> -DCMAKE_CUDA_ARCHITECTURES=90