leap-accelerate compilation is compatible with g++ and clang++ on debian or ubuntu. Support for macOS is currently experimental.
- g++ 9.3.0
- cuda 10.1
- boost 1.71.0
- casacore 3.1.2
- g++ 6.3.0
- cuda 9.0
- boost 1.63.0
- cmake 3.15.1
- casacore 3.1.2
- sudo apt-get install gcc g++ gdb doxygen cmake casacore-dev clang-tidy-10 libboost1.71-all-dev libgsl-dev
- https://developer.nvidia.com/cuda-downloads?target_os=Linux&target_arch=x86_64&target_distro=Ubuntu&target_version=2004&target_type=deblocal
- sudo apt-get install gcc g++ gdb doxygen cmake casacore-dev clang-tidy-10 libboost1.65-all-dev libgsl-dev
- https://developer.nvidia.com/cuda-10.1-download-archive-update2?target_os=Linux&target_arch=x86_64&target_distro=Ubuntu&target_version=1804&target_type=deblocal
- https://askubuntu.com/questions/355565/how-do-i-install-the-latest-version-of-cmake-from-the-command-line
- sudo apt-get install gcc-6 g++-6 gdb doxygen casacore-dev libboost1.58-all-dev libgsl-dev
- https://developer.nvidia.com/cuda-92-download-archive?target_os=Linux&target_arch=x86_64&target_distro=Ubuntu&target_version=1604&target_type=deblocal
Use cmake .. -D<OPTION>=<VALUE> ... or ccmake .. to set cmake options.
Setting an environment variable of the same name will also override these cmake options:
CUDA_ENABLED - Enables building with cuda support
CMAKE_CUDA_ARCHITECTURES - Selects the target cuda streaming multiprocessor and compute levels (default is all)
WERROR - Enables warnings as Errors
WCONVERSION - Enables warnings on implicit numeric conversions
TRACE - Traces data to the local directory
CMAKE_RUN_CLANG_TIDY - Enables running clang-tidy with the compiler
USE_PCH - Use pre-compile headers internally, if possible (defaults to ON)
From the repository root folder run:
git submodule update --init --recursive
NOTE: pulling exernal submodules is now automated by CMake. When downloading the source files via tools other than git the folder external/ will need to be copied manually.
cmake -B build/Debug -DCMAKE_BUILD_TYPE=Debug -DCMAKE_CXX_FLAGS_DEBUG="-g -O1" -DCUDA_ENABLED=TRUE
With tracing to file:
cmake -B build/Debug -DCMAKE_BUILD_TYPE=Debug -DCMAKE_CXX_FLAGS_DEBUG="-g -O1" -DCUDA_ENABLED=TRUE -DTRACE=ON
With gcovr analysis:
cmake -B build/Debug -DCMAKE_BUILD_TYPE=Debug -DCMAKE_CXX_FLAGS_DEBUG="-g -O1" -DCMAKE_CXX_FLAGS="-coverage" -DCMAKE_EXE_LINKER_FLAGS="-coverage"
cmake -B build/Release -DCMAKE_BUILD_TYPE=Release -DCUDA_ENABLED=TRUE
If cmake fails to detect CUDA try adding the following hint variables:
export CUDA_HOME=/usr/local/cuda
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:${CUDA_HOME}/lib64:${CUDA_HOME}/extras/CUPTI/lib64
export PATH=$PATH:$CUDA_HOME/binTesting provided via googletest. To test using the google test runner, test binaries can be executed directly using the following commands from the output folder:
./bin/tests/LeapAccelerate.Tests
./bin/tests/LeapAccelerateCLI.Tests
To test using CTest use the following command in build/linux:
make test or ctest
for verbose output use:
ctest --verbose or ctest --output-on-failure
Generated documentation is available locally at the following file location:
docs/sphinx/index.html
Once deployed to a branch the docs will be available here:
https://developer.skao.int/projects/icrar-leap-accelerate/en/latest/
cmake ../.. -DCMAKE_BUILD_TYPE=Debug -DCMAKE_CXX_FLAGS_DEBUG="-g -O1" -DCMAKE_CXX_FLAGS="-coverage" -DCMAKE_EXE_LINKER_FLAGS="-coverage"
make coverage
Doxygen documentation is generated for all C++ and cuda files with the following target:
make doxygen
Sphinx/Breath/Exhale docs is a dependent target generated with the following command:
make sphinx
The CI/CD on gitlab used a pre-built base build image along with a cpp_build_base image to speed this process.