Change PrUTHitHandler and algorithms for SOACollection

Needs LHCb!3035 (merged) addresses LHCb#128 (closed)

added 1 commit

• 4dec3cde - fix some small bugs

Compare with previous version

mentioned in merge request !2424 (merged)

added 28 commits

• 4dec3cde...510c53b3 - 26 commits from branch master
• 2ec43ea1 - change PrUTHitHandler for SOACollection
• fea74bba - fix some small bugs

Compare with previous version

added 1 commit

• 2017b89d - Port algorithms to PrUTHits with SOACollection

Compare with previous version

/ci-test --merge LHCb!3035 (merged)

added ci-test-triggered label

• [2021-05-07 22:36] Validation started with lhcb-master-mr#2286
• [2021-05-10 10:16] Validation started with lhcb-master-mr#2289
• [2021-05-10 20:41] Validation started with lhcb-master-mr#2293
• [2021-05-10 22:40] Validation started with lhcb-master-mr#2294

mentioned in merge request LHCb!3035 (merged)

changed title from WIP: Change PrUTHitHandler for SOACollection to WIP: Change PrUTHitHandler and algorithms for SOACollection

added 2 commits

• f9066f85 - Adapt PrKalmanFilter for SOA collection in UT hits
• c4d7cc56 - and format

Compare with previous version

added 1 commit

• 80d08486 - remove obsolete PrVeloUTTrack and converter for it

Compare with previous version

/ci-test --merge LHCb!3035 (merged)

added Tracking UT cleanup labels

added hlt1-throughput-decreased label

added 1 commit

• 12d89e54 - Add back padding to avoid FPEs

Compare with previous version

/ci-test --merge LHCb!3035 (merged)

added 1 commit

• b4c64313 - reserve 10k hits

Compare with previous version

/ci-test --merge LHCb!3035 (merged)

This is ready to go from my side @axu

Some comments:

• It leads to a small throughput loss in HLT1 (CPU). This unfortunately happened so far for all algorithms that started using the SOACollection, but at least is compensated by the last UT MR, which increased the throughput.

It needs a ref update. Changes in references are:

• The lead-lead references change. This is expected, as before they actually were overflowing the maximum allowed number of UT hits (which this MR fixes).
• Many tests for the slot with FMA change fail. This is a known side-effect when changing to the SOACollection, already seen here: !2303 (comment 4177379). This are just small numerical changes and has no effect on the performance (all these tests in the non-FMA slots are fine).
• Some tests see a timeout, but I assume this is unrelated.

Note that this needs to go with LHCb!3035 (merged)

marked this merge request as ready