Choose timezone
Your profile timezone:
BLonD code development meeting
→
Europe/Zurich
864/1-B04 (CERN)
Markus: Fitting
- Gaussian.profile -> store data internally or just return: return is OK for everyone
- impedances -> had many internal arrays stored that caused memory problems
- global default: to be presented next time
Kostis: New implementation of drift equation -- runtime evaluation
- Alex: unittests to be shown next time
- New: sqrt (C lib) - > causes a slow-down of about x2 for 1M particles; better for more particles
- VDT fast sqrt -> not really faster, less accurate (polynomial)
- correct: "beam_delta" -> "double beam_delta" in the loop
- like this, every thread has an own copy of the variable
- no race condition (overwriting by threads)
- takes the same time
- Still the drift takes less time than kick & impedance, so we can afford the slow-down
- benchmark was done for alpha=2 order
- repeat for next time with alpha=0 (used more often)
Alex: blond-org/blond-common Fitting package
- profile.py -> all fittings collected in fit(time, data); return is different for each functio
- unittests: separate folder -> using assert equal w/ certain precision
- perftest: performance test; what is the precision and runtime?
- with Kostis: have a look together
- distributions: interfaces/analytic_distribution
- fitting_overhaul branch/test_profile_benchmark.py
- Mihály: try noise injection on data
- Markus implemented f(string) prints; requiring python 3.6 or above - WE APPROVE
Ivan: Vlasov solver
- stability studies in tracking: w/ phase or energy kick, threshold chosen changes result
- Vlasov eq w/ small perturbation
- orthogonal mode expansion (Oide & Yokoya)
- matrix equation (Lebedev)
- Implemented both in "Melody"
- cutoff frequency changes results, but 2 methods agree
- Im(Z/n) = const is an unphysical mode
- code not yet available for users
There are minutes attached to this event.
Show them.