Skip to content

WIP Optimized UT bank decoding

Sebastien Ponce requested to merge cherry-pick-cfe62fde into master

cheery-picked form TDR as it can be applied to master

Main changes :

  • dropped error management that seems useless in our case (MC data) and was extremely costful (25% of the total). This needs to be reviewed for real data
  • dropped most of the counters until they are more efficient in Gaudi. Updating them was taking 17% of total time
  • dropped the final sorting of the clusters. They simply do not need to be sorted. Theyare only consumed by 2 algorithms (PrGECFilter and PrStoreUTHit) that do not rely on them to be sorted. This was taking 25% of the total time
  • optimized research of the board the bank comes from, by adding a map rather than a linear search in a vector. To be reviewed, as a flat map could improve further. This was taking 9% of the total time All in all, the speedup achieved by this commit on the decoding of UT is 5.7 !

(cherry picked from commit cfe62fde)

Edited by Marco Cattaneo

Merge request reports