g4daeview

• DAEVIEW FUNCTIONS
  • DEPENDENCIES
    • Delta : OSX 10.9.2 chroma virtualenv python, based off macports py27
    • G4PB : OSX 10.5.8 macports py26
• G4DAE and G4DAEVIEW Development History
• g4daeview.py
  • Usage Tips
  • Usage text
  • Event Loading/slicing tips
  • Receive ChromaPhotonList from Remote Geant4
  • Load/Save ChromaPhotonList
  • Screen Captures and apache publishing
  • Grab Screen Captures including GLUT menus
  • Partial geometry
  • Target Mode
    • interactive target selection
  • near/far clipping controls
    • ISSUES
  • solid picking
    • TODO
  • placemarks
  • markers
  • Viewpoint Bookmarks
    • ISSUES
  • Animation
    • interpolate between bookmarks
    • orbiting mode
    • animation control
    • photon propagation time control : interactive and auto scanning
  • remote control
  • raycast launch control
    • interactive switch to metric pixels presentation
  • screen capture
  • movie capture
  • Parallel/Orthographic projection
  • Presentation
• G4DAEView Chroma
  • Single PMT Test
  • A Few PMTs and radial shield
• g4daeview debugging
  • Tricks to work out whats gone wrong
  • Issue
  • Near behaving like far
  • Removing the scale kludge
  • Targetting outer volumes draw blanks
  • Near
  • Light control
  • Unproject
  • Trackball
  • Remote Control
  • Modes of operation
    • scaled mode
    • target mode
  • Scenarios
    • detail view
    • outside view
  • Lighting
• G4DAEVIEW Lights
  • reference
  • Light Positioning
    • Controlling a Light’s Position and Direction
  • Understanding Lights
  • glumpy lights
  • refs
• Daeview Dev Notes
  • Limitations
    • GPU Out-of-memory during BVH construction with full Juno geometry (50M nodes?)
  • Old Issues To Be Revisited
    • More Bizarrness
    • Animation
  • Reflectors
    • Bizarre
    • Interpolation
    • Interpolation Jumps (FIXED)
• DAEVertexBuffer Dev Notes
  • Interop
    • Too Many Moving parts
    • AoS or SoA
    • Data flow
    • What needs to be shared between CUDA and OpenGL ?
    • Single VBO approach (array-of-structs)
    • NVIDIA Example
    • Targetted googling
      • andyswarm
    • Separate VBO approach (struct-of-arrays)
    • glBindBuffer
  • OpenGL Drawing Techniques
    • glDrawElements
    • glDrawElements offset
    • glPushClientAttrib
    • glMultiDrawElements
  • Read OpenGL buffer back into numpy arrays
• DAEPhotonsShader
• OpenGL Shader Language
  • GLSL 120
  • GLSL 120 Extensions
    • GL_EXT_gpu_shader4
    • GL_EXT_geometry_shader4
    • glVertexAttribIPointer API hunt
    • GL_NV_vertex_program4
  • Fixed Pipeline and Shaders
  • Bitwise operations glsl
  • line strip finnicky
  • shaders and integers dont mix in version 120
• Photon Level Debug
  • Issues
    • spagetti style and surface step selection incompatible
    • Pushing Up the Slots causes BUS error (AVOIDED)
    • Single Photon Debug
    • Photon Picking
    • Make Bialkali absorb rate ?
    • Quite a lot of many step histories
    • Visual distinction for zombie photons
    • Difficult to zoom into closeups
    • Photon Appearance
    • ESR Step start Asymmetry
    • Material Step Starts
    • Discrepancy between presentation modes (FIXED)
    • Material Selection not compatible with Spagetti style
    • Timeconstant for reemitted photons ?
    • Infinite wavelength for reemitted photons (FIXED)
    • Disappearing/Reappearing Photon 3126 : FIXED
    • Missing NO_HIT : FIXED
    • Dropouts : 91 long bouncers out of 4165
  • Fixed Issues
  • Repeatability/Seeding Doubts
  • Techniques
    • daephotonsanalyzer.sh
    • truncation
    • debugphoton
    • history selection
    • Restrict to photons with n-step histories
    • Restrict birth time range, allowing to examine cohorts
    • cohort mode, third value in cohort string
    • photon highlighting
    • style playoff
• Material Pair Debug
  • Thoughts on how to proceed
  • Geant4 step level comparison
  • Machinery
  • DAEPhotonsAnalyzer.get_material_pairs
    • Both end material check
    • Implement step selection based in from and to materials
    • Broken chain
  • Examine the big hitters
    • Observations
  • Long lived
• Propagated Flags Mismatch
• G4DAEView Performance
  • Pushing to max-slots 100 bogs down performance
  • Apple OpenGL Profiler
  • LINE_STRIP slow performance report
• Optical Photon Marshalling
  • Overview/Issues
  • Numpy Serialize into buffer ?
  • Efficient Selection of Numpy array, eg just the hits with PMTID > 0
  • Current lifecycle of optical photon data
  • Photon Data
  • Using C++ placement new for dummy OP ?
  • Alternate Serialization/Compression to ROOT TObject
• Propagate Timeouts
  • Objectives
  • mock001 : normal incidence NAN_ABORT
  • mock001 : axis aligned photons issue
  • Code Improvements
  • Dependencies
  • DONE
  • Issues
    • timeouts
  • mock photons
  • vbo vs non-vbo hit count difference
  • mocknuwa propagation testing over network
  • file based propagation testing
  • visualize initial positions by holding propagation
  • vbo propagation
  • non-vbo propagation
  • threading sensor ids back to caller (vbo)
  • threading sensor ids back to caller (non-vbo)
• BVH Splits
• Datamodel
  • Performance Data Model examples
    • MySQL Performance Schema
  • Instrumenting Python Code
  • Profiling Python
  • Projects for instrumentation
  • Graphite : Graphing infrastructure
    • whisper
• G4DAEView geometry cache
  • Fixed Problem was the time sorting again
• Serialization
  • Current Solution : NPY
  • Alternatives serialization/deserialization to ROOT ?
    • NEXT
    • contenders
    • comparisons
    • capnproto
    • cereal
    • boost serialization
• G4DAEView Dependencies
  • Dependencies Tree
  • Alternatives to GLUT ?
    • Implications of ancient OpenGL 2.1
    • GLFW
    • glfwcy with PyOpenGL and OpenGL 3.2
    • glfw python wrappers
    • migration from glumpy to glfw
• G4DAEView Ideas
  • JSON State vending
  • Tight zooming not easy
  • Attempts to do rotation interpolation with no geometry visible cause NaN fails
  • During VBO structural changes
  • Live Style Change issues FIXED
  • Step Interpolation Visualization IMPLEMENTED
    • What about non-Chroma devices ?
    • Geometry Shaders
  • Node selection only applies to to particular geometry dyb/juno/lxe
  • Live Updating Test
    • worker
    • client
  • Next
    • improved structuring
    • Histogramming
      • issues
    • Photon provenance
    • ChromaOtherList for transport over ZMQ
    • Geant4 Chroma Process
    • Avoid Geometry Duplication
    • interactive target switching NEEDS OVERHAUL
    • smaller things
• G4DAEView Architecture
  • Division of concerns
• G4DAEView Issues
  • Fallovers
  • Photon Presentation
• NAN Abort Photon
• VBO Propagation Call Sequence needs simplification now that using NPL
• G4DAECHROMA transport testing
  • Same Node Success
    • Client sends workload to broker
    • Worker polls the broker for smth to do
  • Normal Topology Success, via tunneling both client and worker
• G4DAEChroma Memory Leak
  • Executive Summary
  • Initial Observation of 30G VSIZE
  • Large VSIZE
  • Not ballooning to 30G
  • vmmap
  • vmmap -resident
• G4DAEChroma.cpp : ZMQResponder, CUDA kernel call
  • Progress
  • Motivation
  • Dependencies
  • Initialization
  • Propagation
  • NPY Geometry Cache
  • Potential Dependencies
    • CuPP
    • boost::program_options
    • boost::numeric::ublas::vector<double>
    • cuda::thrust
    • CUDA Unified Memory
  • Other Libs
• G4DAEChroma CUDA tune
  • Small Problem Size
  • Large problem size
  • Variation of time with workload
  • Larger than 512 ?
  • Straggler Mode
  • Propagate Queues
• Visualizing GenSteps