This was a prototype for integrating Celeritas into Geant4 for offloading
support. The key components of Soon's research implementation have been
integrated into Celeritas as part of the accel
library. See the latest
Acceleritas documentation for details on the supported capabilities.
With this incantation we hope to use the magic of GPU accelerators and Geant4's new tasking model to offload EM showers from Geant4 to Celeritas to accelerate HEP simulations.
- CUDA
- Geant4 (Version 11.0) (with GEANT4_USE_GDML=ON)
- VecGeom (Version 1.1.18) (with DBACKEND="Scalar" and CUDA="ON")
- ROOT (Version 6.XX.X)
An example CMake command looks like with the CUDA Capability 7.0 (assuming that celeritas required products are installed at ${INSTALL_DIR}:
cmake ${SRC_DIR}/acceleritas \
-D CELERITAS_USE_CUDA=ON \
-D CELERITAS_USE_Geant4=ON \
-D CELERITAS_USE_VecGeom=ON \
-D CELERITAS_USE_ROOT=ON \
-D CMAKE_CUDA_ARCHITECTURES=70 \
-D CMAKE_BUILD_TYPE="RelWithDebInfo" \
-D CMAKE_CXX_FLAGS="-Wall -Wextra -pedantic -Werror" \
-D CMAKE_PREFIX_PATH="${INSTALL_DIR};${XERCES_C_DIR}" \
-D Geant4_DIR=${GEANT4_BUILD_DIR}
make -j
ln -s ${SRC_DIR}/acceleritas/app/acc01/input input
./app/acc01/acc01 -m input/run.mac -t 4