Skip to content

Improve VMM CPU performance

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

This MR speeds up the simulation of the VMM shaper in the MM digi. Two major points were improved:

  • Break calculation of the response after the first peak was detected
  • Move a calculation of pow out of the shaper response loop and calculate it just once in initialize

These improvements happen in the context of an overall speedup of the MM digi discussed in ATLASSIM-4923

Update on the CPU times obtained with RunUpgradeTest:

  • nightly of 27 of October:
    <cpu>: ( 6513.333 +/- 3365.182 ) ms
    <cpu_user>: ( 6442.889 +/- 3359.384 ) ms
    <cpu_sys>: ( 70.444 +/- 10.169 ) ms
    <real>: ( 6599.111 +/- 3411.147 ) ms

  • with this MR on top:
    <cpu>: ( 4018.222 +/- 1415.981 ) ms
    <cpu_user>: ( 3862.222 +/- 1361.094 ) ms
    <cpu_sys>: ( 156.000 +/- 55.178 ) ms
    <real>: ( 4522.667 +/- 1698.112 ) ms

Edited by Patrick Scholer

Merge request reports

Loading