Choose timezone
Your profile timezone:
BLonD code development meeting
→
Europe/Zurich
864/1-B04 (CERN)
General info:
- reminder of January deadline for publication
- Assembler to be started
Alexander: FCC synchrotron radiation compensation via noise
- results agree with theory for bunch length decrease and noise amplitude
- to be done: add phase loop and synchro loop
- simulate ramp
Simon: first attempts for an assembler
- loop over track routines
- map, turn number, args, kwargs called by function
- map (order of tracks) defined in main file
- occasional interactions (plots, statistics) -> using add_function every x turns
Simon: frequency offset from design frequency
- frequency programme that differs from the design one (omega_rf, and phi_rf slippage calculated from that) -> to be tested by other users after merge
- special case: fixed frequency for PSB injection
- unittests and example cases
- custom exceptions defined that can replace the "RuntimeError"
Kostis: literature review on approximation techniques
- need approximation, quality evaluation, run-time controller
- physics and simulations is all about finding the right approximation
- well-known techniques: loop perforation, tiling, algorithm exchange, numerical approximation, relaxing synchronisation
- hardware approximations: reduced precision units (GPU/CPU/FPGA), parrot transformation -> for reduced code regions (neural processing units NPU)
- quality evaluation: based on user-specified function
- approximation opportunities: programmers-pecified, compiler-based, learning-based (predict error via on/off-line training)
- on-line reconsiderations of approximation techniques: "steepest ascending hill" algorithm, evaluation using canary inputs, within fixed/adaptable re-evaluation intervals
Round table:
- Markus: save points at regular intervals to restart the simulation from the point it was interrupted
- Save and reload from a given point: what input is needed?
- Save and reload defined for every object?
- Objectives: reload crashed simulations, different simulations that change only towards the end
- Alex: mirror to gitlab
- Simon: beam.intensity bug to be fixed in next pull request
There are minutes attached to this event.
Show them.