Skip to content
GitLab
Projects Groups Snippets
  • /
  • Help
    • Help
    • Support
    • Community forum
    • Submit feedback
  • Sign in
  • R Rec
  • Project information
    • Project information
    • Activity
    • Labels
    • Members
  • Repository
    • Repository
    • Files
    • Commits
    • Branches
    • Tags
    • Contributors
    • Graph
    • Compare
    • Locked Files
  • Issues 289
    • Issues 289
    • List
    • Boards
    • Service Desk
    • Milestones
    • Iterations
    • Requirements
  • Jira
    • Jira
  • Merge requests 61
    • Merge requests 61
  • CI/CD
    • CI/CD
    • Pipelines
    • Jobs
    • Schedules
    • Test Cases
  • Deployments
    • Deployments
    • Environments
    • Releases
  • Monitor
    • Monitor
    • Incidents
  • Analytics
    • Analytics
    • Value stream
    • CI/CD
    • Code review
    • Insights
    • Issue
    • Repository
  • Activity
  • Graph
  • Create a new issue
  • Jobs
  • Commits
  • Issue Boards
Collapse sidebar
  • LHCbLHCb
  • Rec
  • Merge requests
  • !193

Calocounterswitch

  • Review changes

  • Download
  • Email patches
  • Plain diff
Merged Olivier Deschamps requested to merge CALOCOUNTERSWITCH into master Aug 17, 2016
  • Overview 0
  • Commits 2
  • Pipelines 0
  • Changes 92

From @deschamp : Implement counter switch in calo reco packages

From @jonrob (via !122 (merged) ) : Some proposed changes to CaloRec to speed up the running in HLT. This update does two main things

  • Speed up CaloCorrectionBase::getParams, which is a significant contribution to the total time taken by several Calo algorithms, by removing the map look up for parameters. Map look ups are not super fast at the best of times, and this was made worse as the key was a string. In this case a map is not required, as the keys are all known before hand and in fact just a series of enums, numerating from 0. So instead I use a simple pre-filled vector (with a status flag to say which are 'active').
  • Another large contribution was from math functions, so exp, log, atan etc. This update just changes these to use the fast vectorised VDT versions. The precision of these methods is almost as good as the IEEE standards, differing in a few of the least significant bits. Should be more than enough for the precision of the numbers being processed, but this should be confirmed by a CALO expert. Together, these updates should save O(a few %) on the total HLT2 time. Full profiling results will follow when I have them.
Assignee
Assign to
Reviewers
Request review from
Time tracking
Source branch: CALOCOUNTERSWITCH