Skip to content

Speedup of the MM electronics simulation

Patrick Scholer requested to merge pscholer/athena:vmmSpeedRound2 into 21.9

This MR addresses the CPU performance issues in the MM digitization discussed in ATLASSIM-4923. It mainly addresses technical improvements in the electronics response function namely:

  • Move as many calculations as possible out of the VMM response loop
  • Only start peak search close to peak (80% of the peak time after the first charge)
  • Search for the peak in time step of 0.5ns (instead of .1ns) and only scan with 0.1 ns precisson if a peak was detected
  • Search in 1 ns steps for a threshold crossing until a first possible peak and search in .1 ns steps once the crossing was identified
  • Avoid copying heavy vectors at one point and return by reference instead (MM_ElectronicsToolInput)

It also introduces functions in the strips response simulation where random numbers are generated to identify cpu bottlenecks. The overall time spent in the elx response simulation got down from 290 s to about 20 s per event using a <mu>=200 event. The main bottleneck left in the MM digi is now the generation of random numbers for the transverse diffusion.

Edited by Patrick Scholer

Merge request reports