Previous topic

Groupvel Kludge Unintended Consequences

Next topic

This Page

geant4_opticks_integration¶

tconcentric
- setup
- remnant pflags discrep ?
- RESOLVED seqmat truncation difference, c2 down to 0.91
- seqmat still some truncation difference ?
- seqmat mismatch mostly resolved
- psel selection of truncated
- opticks seqmat
- ISSUE: seqmat mismatches, zeros
- how should truncation be handled ?
- ISSUE : negative times on exceeding time domain
- RESOLVED: bad REjoin onto scatter, 3 in 1M
  - elaborated fix
  - initial fix
  - issue rare SC/AB at truncation
- MOSTLY RESOLVED: pflags inconsistency : due to incomplete REjoins, AB not scrubbed and replaced with RE
  - Mostly Resolved
- Nov 4, 2016 : seqhis chi2 now less than 1
- RESOLVED : scattered counts and distrib mismatch
- MATCH : no selection CFG4/OK rayleigh scattering in isolation
- RESOLVED : scatter discrep
- FIXED : wrong wavelength compression/decompression in CFG4
- ASIS : polarization de-normalizing
- viz
- truncation control
- FIXED Longstanding pflags issue
- FIXED : Opticks not doing “TO BT BT BT BR ..” by polz correction
- FIXED : chi2 large contribs from SC lines
- dbgzero lines
  - seqmat truncation discrep, lots of zeros in tail
  - pflags
  - FIXED seqhis
- remaining discreps by cmx selection
- truncation avoidance trick
- zeros in long tail issue, from truncation ?
- with flags fixed
- FIXED : funny badflag 00
- avoiding truncation by pushing bouncemax and recordmax
- switch to unified model fixes incorrectly specular DR
- post DR discrep
- seqhis after optical surfaces hookup with test geometry
- seqhis
- lack DR due to lack of complete Optical Surface ? with test geometry
- topline 8ccccd 4% Opticks deficit
Profiling Duplication
- TODO
- OpticksProfile
- Related : npy-/Timer
Boolean CSG on GPU
- TODO: numerical/chi2 history comparison with CFG4 booleans
- TODO : python trees into opticks
- TODO : randbox photon source
- TODO : more natural test tree construction
- TODO: support transforms in op nodes
- TODO: support empties, ie non-perfect trees
- TODO : algorithm characteristic comparisons
- Boolean Debugging
- DONE: boolean csg tree implementation
- FIXED Issue : ray trace “near/tmin” clipping fails to see inside booleans
  - scene_epsilon
- FIXED Issue : boolean insides invisible from outside
- FIXED : Issue : cannot see booleans from inside
- FIXED : Issue : ray trace of box shows slab intersects extending behind the box
  - CUDA fminf/fmaxf/max infinity/nan handling ?
- FIXED Issue : boolean intersection “lens” : boundary disappears from inside
- FIXED Issue : lens not bending light
- approach
SurLib with Test Geometry
tpmt
Review Analytic Geometry Workflow
- Changes needed
- Test Geometry
Sliced Composited Raytraced Rasterized
- Test geometries
- Oglrap
Tracer Crash
- Tidy Actions
OKG4 G4GUN Shakedown
- Red Gensteps everywhere issue
- Tao commit changing Scintillaton and Cerenkov
- Debug running
- Peculiarities
- Known sources of difference
JUNO JPMT Shakedown
- Issue : default torch source goes off somewhere in TT
- Issue : segv due to no input gensteps
Material Code Mapping Generalization
- env/chroma/G4DAEChroma
- Mapping Codes ?
- Where is genstep path configured ?
- Review Genstep
- Issue : Sep 2017 : Still getting lookup fails
- Issue : Dec 2016 : Lookup fails with live g4gun
- Legacy Approach
  - Translate on load
  - Lookups
- Changes
Geant4 Optical Code History
- Geant4 On Github
- Scattering Bug
- ProposeVelocity does nothing as CalculateVelocityForOpticalPhoton is repeatedly called
- GROUPVEL From Wrong Material Issue ?
GROUPVEL
- Approach
- Issue 1
- Issue 2
- See Also
- Transport gets velocity from preStep
- GROUPVEL Injection at tail of PSDIP has no effect
- Can I replace G4Transportation with a debug version ?
  - AddTransportation
- DsG4OpBoundaryProcess dumping : looks like getting groupvel from Ac instead of LS and MO
- Back to basics after moving to fine domain (1nm)
- tconcentric check
- Review
- Now get G4 warnings when run without groupvel option
- Opticks GROUPVEL
- G4 GROUPVEL
Reemission Review
- recsel segv : last one issue
- fixing REJOIN recording
- checking matching with extremes of material properties
- switch off reemission with nore option
- tlaser : sizable differences in many categories : how to proceed ?
- push stats to 1M have 35 CRecorder/Rec discrepant seqhis/seqmat
- seqhis machinery inconsistency between CRecorder and Rec
- How to proceed ?
- Adopting DYBOp into CFG4
- REEMISSIONPROB is not a standard G4 property
Multi Event
- Overview
- FIXED : Issue
- FIX
- Workaround
- Things to try
  - DONE : bufferTest.cu bufferTest.cc
  - Trivial Program works
tconcentric_distrib
- Look for presentational hists
- Comparing interpolation blib
- After move to c2shape comparisons : listing seq points with sum of distrib chi2 > 20
- DONE machinery shakedown
- multiplot slice(0,10) quick look
- tconcentric agreement sufficient to move on to distribs
tconcentric pflags mismatch from_truncation handling
- Status
- Issue
- FIXED BY ENFORCING HARD_TRUNCATE IN CANNED RUNNING
- Internal CFG4 Opticks event inconsistency
- CRecorder::RecordPhoton
- tconcentric.py
- Selecting mismatch using PFLAGS_DEBUG
- Dump first mismatch photon
- Dump 2nd mismatch
- Truncation handling
post recording has seqmat zeros
- Overview
- After Fix seqmat chi2 is 0.93
- FIX REQUIRED matSwap BASED ON NEXT BOUNDARY
- Issue
- Cause
- Opticks code check shows no kludging for DR SURFACE_DREFLECT
- CFG4 Dumping of mismatched photons
- Flags : NA NAN_ABORT actually means boundary status StepTooSmall
- Differences between live and canned modes of CRecorder
Review Geant4 Stepping Time Velocity
- where does G4 set times anyhow ?
- G4TrackingManager::ProcessOneTrack
- G4Step::InitializeStep Track->PreStepPoint
- G4StepPoint::AddGlobalTime
- G4Step::UpdateTrack PostStepPoint->Track
Interpolation Mismatch Continued
- Issue
- TODO: check time distrib diff matches expectation from groupvel interpol mismatch
- Testing finebndtex
- How to up samples ?
- Possible Approaches to reduce interpolation mismatch
- Dumping Interpol deviations
SURFACE_DREFLECT (DR) diffuse reflection
- issue : now resolved
- CFG4 SURFACE_DREFLECT from LobeReflection or LambertianReflection
- Cause of discrepant reflection, different imps
- Reimplement Opticks diffuse reflection
- SURFACE_DREFLECT discrepant polarization b,c
- NumPy DoReflection calc
- chi2 after fix, notably the pflags decreased from 1.21 to 1.07
Test Fails
SURFACE_ABSORB
- NEXT
- dbgseqhis
- Boundary 37 (with os:RSOilSurface) : wrong way around ? inner not outer surf ?
- Concentric test
- Compare with tdefault where photons distributed randomly from point source
- Complement checks...
- Visual inspection of photon termination points
- Surface info comparison with CInterpolationTest and OInterpolationTest
- Commenting out ESRAir reflectivity diddle doesnt fix the 20%
- 1M 2016 Oct 28 seqhis
- SA Opticks
- G4 SA
tlaser
- 1M 2016 Oct 28 seqhis
- DONE: badflags fixed by categorizing SAM:SameMaterial as BT in cfg4-/OpStatus
- DONE: property comparison using G4 interpolation and Opticks tex interpolation
- Using constant material prop values gives much better agreement
- With fixpol doesnt change much
- Progressive masking for following discreps step by step
- After REJOIN fix still large discreps, eg top line SA
- Optical Surface Trace
- Prior to fixing aim
- Laser aim issue
- after fix aiming, restricted to top seq
- After fix CG4 skin surfaces
- Termination boundaries
- probable cause CG4 logical skin surfaces lacking lv
- After fix CG4 logical skin surfaces
- full seq following fixed aim
- initial ana
Badflags
Interpol Mismatch
tboxlaser
- check fx machinery
tbox
OKG4 tpmt revisited
- FIXED : CTestDetector::kludgePhotoCathode was incorrectly using dielectric_dielectric
- Perhaps surface conversion not yet implemented for CTestDetector ?
- G4 EFFICIENCY
- Divergence following many changes : g4 not stopping at cathode ?
- FIXED : tpmt takes exception to duplicated OpaqueVacuum material name
Missing CFG4 Surface Detect
- Issue
- Observations
- Fixed ? N
- CGDMLDetector vs CTestDetector
- Writing G4DAE Optical surfaces
- Reading G4DAE/ggeo Optical Surfaces
- Ancient GDML Export has no surfaces OR optical props (is also has no material properties)
- CGDMLDetector::addMPT
- CGDMLDetector::addSurfaces ?
- NEXT
- Code Trace photon SD flags
OKG4 Torch Shakedown
- Issues Overview
- What would be useful for debugging ?
  - tlaser testing
  - seqvol : volume sequence indexing
- Simulation
- Analysis
- Viz
- Initially not in same ballpark, after an afternoon get into same ballpark
  - zero flags, SR?
- dbgseqhis c0cac0cccd
- FIXED : Positional bug
- bouncemax zero check
- FIXED : Material reporting with CG4
seqvol : volume sequence indexing
- tlaser node index dumping
- dbgseqhis
- Where did the surface handling come from in Opticks...
- op –bnd : shows no isur/osur duping but there are more skin surfs that border surfs, SO THIS IS A BUG
- optical_buffer
- closest_hit_propagate
- instance_identity comes from the intersects
- mesh_intersect
- identityBuffer
- FaceRepeatedIdentityBuffer
OKG4 vizg4 noshow
- Issue
- tpmt.py
- tokg4.py
G4/Op Integration Overview
- Objectives
- See Also
- DONE : High Level Cleanup/Refactor
- WIP : Multi event handling
  - TODO : G4 driven multievent
- DONE : init and per-event distinction
- DONE : Optical Step Collection
- DONE : Material Code Mapping Generalization
- DONE : Integrated G4GUN running
- DONE : genstep handling rationalize
- DONE : avoid duplication between OKMgr and OKG4Mgr
- NEXT : OKMgr/OKG4Mgr propagation multi-event test
- NEXT : loaded propagation viz lacks index
- NEXT : OKG4 genstep shakedown
- WIP : Integrated Torch running debug
- FIXED : OKG4 : Material Map chicken/egg problem
  - Mapping A
  - Mapping B
  - Fix
- NEXT : G4/Op Comparison of generation distribs
- NEXT : event handling in integrated mode
- NEXT : move CG4 event creation later for multi-event
- Approach
  - ggeoview-/App
  - OpticksEngine base class
  - OpticksApp
  - How does Opticks and G4 need to interface ?
- DONE
- DEBUGGING
- TODO
Interop Failures
- FIXED : Interop mode with seed buffer : get hard crash even with trivial
- Trying to do raytrace when loaded, fails for lack of record buffer
- Multi event testing CUDA memory error
Multi Event : Seed Stuck at Zero for 2nd Event
- Interop Mode
  - WITH_SEED_BUFFER
- Compute Mode
- DONE: Getting OptiX to notice CUDA/Thrust change to Photon buffer
  - Dirty CUDA Interop Buffers
- FIX: Using BUFFER_COPY_ON_DIRTY for photon buffer and manually marking dirty
- Maybe Use Separate INPUT_ONLY SEED BUFFER ?
- TODO: Measure multievent compute speed using INPUT_OUTPUT seeded photon buffer vs INPUT only seed buffer
OptiX Changing Event Buffer
- Objective
- Observation : trying to prelaunch in init fails
  - Context validation fails without evt buffers
- Changing Event Buffers ?
- Changing OpenGL Buffers
  - glBufferData
- oglrap- Scene::uploadEvent
NEXT : G4 Optical Step Collection
- Objective
- Stages
- Review old collection with G4DAEChroma
- Approach to reviving optical step collection
- DONE : Near Standard G4 Scintillation/Cerenkov Step Collection
- TODO: details of scintillation generation, compare Opticks vs G4 generated distrib
Opticks Engineering
Material Properties
- [WORKED AROUND] Issue : old GDML export omits material properties
Broken Indexing
Broken PMT Test
- FIXED
- Issue : photon record positional qwns all zero
G4GUN Debug
- Integration starting point : pure g4gun
- Evt debug with npy-/g4gun.py
  - domains
- Opticks Space Domain
- G4/GDML Space Domain
- Geometry Selection
  - shape
G4GUN Material Reporting
- Approach
- Fishy flags in history
Taming Geant4 Verbosity
- FIXED Issue : g4 geometry cleanup WARNINGs
- FIXED Issue : g4 process verbosity control
- Issue : g4 couples table noise, comment call to DumpCutValuesTable()
- Issue : g4 init is slow, how to cache the physics tables ?
- Issue : g4 Em noise control
- Issue : g4 noise control
[FIXED] GUI Photon Flag Names All NULL
- progress
- initial issue
[FIXED] GUI Photon Selection Failure
- FIXED
- Background
- Approach to solving
- Not Working
- Working
High Level Refactor
- Objective
- Approach
- DONE
- Tests Commands To Run Whilst Refactoring
- [FIXED] OP record vis broken by OpticksEvent simplifications
- [FIXED] Left field flakiness
- FIXED: Photon record coloring M key seems wrong
- [FIXED] CFG4 load count mismatch assert
Time Goes Backward
- [FIXED] Issue : with GDML geometry opticalphoton time going backwards