Page Comparison

2017-09-04 Meeting notes

• Anthony: git-svn bridge
• Irana: interactions: list vs vector
• Anthony: setIndSpecies

See log below:

• Thomas: IntersectionOfWalls can now handle fine details (previously could cause nan values); nan values now get treated
• Irana: Bug fix cgHandler at boundaries
• Irana: Simplification of Kernel/CMakeLists.txt
• Irana Commented CG, sped up neighborhood detection in CG<CGCoordinates::XZ, CGFunctions::Lucy, CGFields::StandardFields>::evaluateContact
• Irana: vector instead of list in BaseInteractable.

Because of templates, not all files with the implementations of functions can be compiled separately, see e.g. this cppfaq for an explanation of templates and compiling.

Currently, there are some .cc files included instead of compiled, see e.g. CGPoint. Can we rename this to CGPoint.hcc (or something else) to make a distinction between implementation files that are included vs compiled?

untitled.m

Here the results of computation time from a quasi 2D silo simulation (Trunk, r2478) with periodic particles. Particle number is increasing, so I can measure speed as function on N. Speed is at 50'000 particles*timesteps per real second. So a silo simulation (timeMax=3) takes 3 weeks.

In 2012, the chute simulations, we could simulate 8e3 particles and 2e6 timesteps in 10 days (18'500 particles*timesteps per real second), so it got quicker.

But: Under idealised conditions (small number of particles, no boundaries/walls), Mercury was at ~500'000 particles*timesteps per real second (measured in Santorini). My guess: the ghost particles are very costly.