1. GeantV fast simulation ideas and perspectives Andrei Gheata for the GeantV collaboration CERN, May 25-26, 2016

2. THE GEANTV PROJECT G.Amadio (UNESP), Ananya (CERN), J.Apostolakis (CERN), A.Arora (CERN), M.Bandieramonte (CERN), A.Bhattacharyya (BARC), C.Bianchini (UNESP), R.Brun (CERN), P.Canal (FNAL), F.Carminati (CERN), L.Durhem (intel), D.Elvira (FNAL), A.Gheata (CERN), M.Gheata (CERN), I.Goulas (CERN), R.Iope (UNESP), S.Jun (FNAL), H.Kumawat (BARC), G.Lima (FNAL), A.Mohanty (BARC), T.Nikitina (CERN), M.Novak (CERN), W.Pokorski (CERN), A.Ribon (CERN), R.Sehgal (BARC), O.Shadura (CERN), S.Vallecorsa (CERN), S.Wenzel (CERN), Y.Zhang (CERN) 2GeantV fast simulation

3. Background In the best case the speedup from GeantV will be O(10) From the estimations collected informally, at least O(100) is needed This is the realm of fast simulation 3GeantV fast simulation

4. Definition 4 “Learning” the response function fastsim GeantV fast simulation

5. History Fastsim has historically been introduced almost as an “afterthought” or a “user hook” in simulation programmes – Geant4 introduced fastsim hooks early, but they were not “abused” – Why? In GeantV we would like to introduce fast simulation in the framework 5GeantV fast simulation

6. What and how? It is possible to activate fast simulation on a user filter – e.g. region / particle /energy basis Provide generic tools to correlate output information (e.g. digits, rec points, or even tracks) with MC truth for selected particles A choice of basic fast simulation algorithms will be proposed for trackers and for calos 6GeantV fast simulation

7. Too detector specific? The same was said for detailed simulation before EGS came out! User code definitely needs to be involved, so the approach must be user code friendly… We do not pretend to cater for all cases, but to offer a reasonable “library” of algorithms on which to build, and a framework to easily embed user code 7GeantV fast simulation

8. Scheduler Basketizer Geometry navigator Geometry algorithms Physics Basket of tracks x-sections Reactions Dispatching MIMD SIMD GeantV: Trying things differently GeantV fast simulation8

9. Basketizer Outputs Geometry Filters Geometry Filters Baskets (Fast)Physics Filters (Fast)Physics Filters Stepper Input queue Vector stepper Vector stepper Step sampling Filter neutrals (Field) Propagator Step limiter reshuffle Step limiter reshuffle VecGeom navigator VecGeom navigator Full geometry Simplified geometry Physics sampler Phys. Process post-step Phys. Process post-step Secondaries TO SCHEDULER Vol 1 Vol 2 Vol 3 Vol n e + /e - γ e + /e - γ Some more detail Coproc. broker 9 High throughput GeantV fast simulation Scheduler MT vector/scalar processing

10. Who controls the game? Fast simulation creates bridges between simulation and tracking/reconstruction How to interconnect in an experiment independent way these components? All is about scheduling the work and data flow, using a common concurrent infrastructure GeantV proposes tools for this kind of approach – Simulation is a natural place to apply these parameterizations and to check their effect – The library to learn the parameterizations and compare with MC truth is independent, but can be delivered with GeantV (as long as it stays generic) GeantV fast simulation10

11. GeantV concurrent tasks GeantV fast simulation11 Data stream Task T*T* T*T* U U Data stream Scheduler Filter queue std::vector (fired on threshold) Basketizer

12. The complex scheduling business GeantV fast simulation12 Provide scalable & balanced workload – Minimize memory footprint – Minimize cool-down phase (tails) Using concurrent event injection and prioritized processing Adaptive behavior to maximize performance – Dynamic switch of vector/scalar processing – Learning dynamically the “important” filters – Adjust dynamically event slots to control memory Accommodate additional concurrent processing in the simulation workflow – Hits/digits/kinematics I/O – Digitization/reconstruction tasks Efficient concurrent basketizers – Filtering tracks by several possible locality criteria Geometry, (fast)physics process – Giving reasonable size vectors all along the simulation

13. GeantV simulation flow GeantV fast simulation13 Data stream Step sampler (xsec mgr.) Step sampler (xsec mgr.) Geometry stepper (navigator) Geometry stepper (navigator) RK propagator VolumeFilter VolumeFilter Physics (msc) Physics (msc) Relocation (geometry) Ionization (eloss) Ionization (eloss) Process mgr. (discrete) Process mgr. (discrete) e+e-e+e- e+e-e+e- n n hadr. Transport cuts mgr. ParticleFilters ParticleFilters Scheduler secondaries User stepping GeantHit Block GeantHit Block Data stream Kinematics mgr. hepmc3/user

14. Embedding fast simulation GeantV fast simulation14 Data stream SimulationFilters SimulationFilters FastSim filters FastSim filters Scheduler particle, region, energy, … Full simulation FastGeom filters FastGeom filters Geometry/mat. parameterizations Frozen showers/ fast tracking Summable digits Parameterized det. response Parameterized det. response Tracks exiting tracker Data stream Digitizers Digitizers Raw event Data stream Trackers Trackers The framework can construct the parameter database learning from full sim + response from user module

15. Combined simulation 15 Unparameterized track Calorimeter parametrization Tracker parametrization Track “regeneration” Calorimeter parametrization Calorimeter leakage Tracker parametrization GeantV fast simulation

16. Framework functions Provide tools to – Build libraries of parametrized quantities from detailed simulation – Compare detailed and parametrized simulation – Allow user-driven switch between different options (full, parametrization, param+regeneration) Apply these tools to simple examples (provided) – Generalizing in experiment-independent manner starting from these 16GeantV fast simulation

17. And… Parametrized simulation is considerably more “vector friendly” then detailed one 17GeantV fast simulation

18. In summary We are not proposing a “one fit all” solution We are proposing a “toolbox” providing scheduling and interconnection hooks, plus simple examples We want to learn from you new use cases, discuss how to accommodate them 18GeantV fast simulation