Functionality overview:
- project structure, components, library organization
- targeted/achieved functionalities
Implementation overview:
- components internals, basic data structures host/device
- random number handling and reproducibility
- workflow(s), scheduling and event loop implementations
- scoring and data transfers
Performance overview
- Current performance...
Code overview:
- background and motivation
- codebase overview, including dependencies
- components overview
Performance:
- Metrics
- OLCF Summit results
- Profiling
VecGeom overview and experience
Current VecGeom GPU support
- data model on host/device and dependency on CUDA
- model data transfer and initialization on GPU
- geometry optimizations (navigation states, single-precision, BVH)
- navigation and current fundamental problems on GPU
- evolution path for VecGeom GPU support: surface models
Magnetic field handling
Celeritas/VecGeom integration
ORANGE: Oak Ridge geometry engine
- background
- features, status, goals
- code examples
Magnetic field in Celeritas
- background and methods
- standalone field performance
- general design and interface with transport
- available processes
- physics validation
- GPU vs CPU performance and issues
Physics interface
- transport loop and control flow
- physics kernels
Physics "perspectives"
- available processes
- data management and components
Physics verification/validation
- geant4 helper app
- current test problems
- key integration goals
- constraints from external frameworks
- Geant4 interfacing approach
- current workflow limitatations
Experiment integration
- potential workflows
- challenges
Acceleritas: Celeritas/Geant4 integration
- offloading strategy and integration details
- theoretical and actual performance
- future work
- summary of performance assessment and main bottlenecks
- further development ideas
- strategies, follow-ups and collaboration
- summary of performance assessment and main bottlenecks
- further development ideas
- strategies, follow-ups and collaboration
