Work Plan

• Test Documentation (all; currently Sourav)

• Test runs on HPC (Eric)

• Make a suite of tests which new commits should pass before going to stable branch (Everyone write one)

  • Basic tests (Saleh)

    • CPU vs GPU (integrator * Nbodies, short duration) [done]

    • Energy conservation (GPU integrators * Nbodies, long duration) [done]

    • GPU (Nbodies * monitors, 1 integrator) [should be easy to add to test_integrators.cmake]

  • For scattering & hermite_adap (Aaron)

  • Write monitor/stopper that checks for collisions (Sourav)

• Test attributes for body/system (Sourav while writting collision monitor)

• Check benchmarks (fix scripts; Saleh)

  • Integration Algorithm * Chuncks per block

  • For now, test chunck_size manually, since it is set as cmake parameter & changing requires recompiling. chunck_size is a property of the ensemble and needs to be known at compile time. Nbodies (Nb) is template parameter for integrator.

  • See if there’s a simple rule for choosing efficient chunck_size and chunks_per_block

  • max[ Nb*(Nb-1)/2,3*Nb]*chunck_size*chuncks_per_block ⇐ 512 = max threads per block (so currently 3⇐nbod⇐32?)

  • In principle, someone could make an integrator that didn’t cache accelerations and used only 3*Nb threads per system, so as to allow 3⇐nbod⇐170 (not for next release)

• Fix MVS integrator/propagator (Saleh? or Eric if not easy)

• Release

• Write paper

• Make collisions monitor parallel (Saleh?)

• Make useful error messages (as they come up)