vizcon2 talk

This is the source (slightly modified to fit in here and with links added) of my talk given (online) at vizcon2 on March 15, 2020

• vizcon2 talks
• this talk

Intro

vizcon2: VTKView.jl – vtk based in-situ visualization in Julia

Jürgen Fuhrmann

Weierstrass Institute (WIAS) Berlin

github.com/j_fu

My secret interest in visualization

Idea

• Duty: Numerical Math, PDE based simulation algorithms, application projects
• Secret interest: in-situ visualization from within 1/2/3D unstructured grid PDE solvers
  • Plot output during simulation run
  • Non-intrusive API so it can be added to any code
  • Utilization of GPU hardware

History

• Did use some GKS in the early times
• Rolled my own: gms
• Switched to OpenGL: gltools
  • Uses fixed function pipeline OpenGL standard
  • gltools2 with FLTK gui included in WIAS simulation code (we think about making this available to Julia)
• vtkfig

Why vtk ?

• Fixed function pipeline OpenGL deprecated
• Don't want to mess with shaders, Vulkan etc.
• Very few intermediate level interfaces for GPU graphics available
  • three.js but this sits in the browser
  • Josef Heinen's gr, but this also uses fixed function pipeline
  • Big chance for Makie also outside of Julia
• VTK provides high level primitives for grid data based visualization So why not... ?

VTK

• Started in 1993
• BSD license
• Well maintained, reasonable evolution with legacy edges
  • Modern OpenGL
  • hopefully Vulkan (fresh discourse thread ) EDIT 2020-02-24: This seems to be on the way
  • smart pointers
  • C++11 features
  • First class citizen of CMake universe (both maintained by Kitware)
• Library behind Paraview, VisIt ...
• used by US labs ==> big interest (IMHO) in keeping this going
• ==> investment in learning appears to be safe.

Some cons

• Lots of examples available, but steep learning curve nevertheless
• Event loop based API hard to incorporate into simulation codes
• Rather ugly plot attributes
• It's YUGE

VTK basic structure + features

• Fill vtkDataSet with grid, point scalars etc. (more or less 1:1 mapping with VTK files)
• Create rendering pipelines by chaining various filters to extract isolines, apply colors ...
  • some similarity with Makie's lift
  • Paraview is the GUI for this

isoline_filter->SetInputConnection(transgeometry->GetOutputPort());

• Create Actors which make things visible on the screen

vtkfig aka fixed rendering pipeline VTK viewer

• Several fixed but configurable pipelines for standard visualization tasks
  • vtkfig::Figure
    • vtkfig::ScalarView
    • vtkfig::VectorView
    • vtkfig::GridView
    • vtkfig::XYPlot
• Figures can be added to a vtkfig::Frame
• Event loop and rendering in separate thread At the cost of having all data copied ...
• Interactivity inherited from vtk
• File output: images, video, vtk
• Duck typing based API for filling data structures ==> easy integration into C++ code
• Memory management via C++11 std::shared_ptr
• Additional C API
• Builds on Linux, MacOS, Unix Subsystem of Win10

VTKView.jl

• Make vtkfig available from Julia
• Integration via C API and ccall
• Attempt on a more "julian" API
• No precompilation time ;-)

Plans

vtkfig

• One thread per frame in multiple frame case
• No time for GUI...
• Try out Paraview catalyst (send vtkDataset to Paraview)
• Try out higher order elements (vtk has some stuff for this)

VTKView.jl

• Map full API of vtkfig (still missing streamlines some state toggles)
• jll
  • builds with vtk-8.2.0 but only without cross compiling jll works for linux-x86_64
  • vtk-9.0.0 seems to support cross compiling need to learn how to export cross compilation tools to client codes ...

Ideas

• Backend for plot recipes ?