RICH Photon Reco - More efficient mirror data cache update implementation.

Implements improvements to the method SIMDQuarticPhotonReco::MirrorData::update() which caches SIMD information when associated mirror pointers change. Main changes are :

• Avoid pointer dereferences, which are potentially slow due to cache issues, when they are not strictly required (i.e. pointer has not changed).
• Use private minimal implementations of Point, Vector and Plane that allow direct access to data members, that in turn allows MirrorData::update() to be a lot more efficient.

Relies on LHCb!1108 (merged)

• [2018-01-31 19:17] Validation started with lhcb-gaudi-merge#402
• [2018-02-02 15:53] Validation started with lhcb-upgrade-hackathon#3
• [2018-02-02 20:51] Validation started with lhcb-gaudi-merge#406
• [2018-02-03 00:09] Validation started with lhcb-clang-test#850
• [2018-02-03 00:11] Validation started with lhcb-head#1743
• [2018-02-03 00:11] Validation started with lhcb-gaudi-head-py3#53
• [2018-02-03 00:14] Validation started with lhcb-tdr-test#47
• [2018-02-03 00:16] Validation started with lhcb-gaudi-head#1745

marked as a Work In Progress

changed title from RICH Photon Reco - Mirror Update to WIP : RICH Photon Reco - Mirror Update

changed the description

added 2 commits

• edc738f6 - Allow more flexibility in template types in QuarticSolverNewton.h
• a7e881ed - Use private Point, Vector and Plane types in Quartic reconstruction to allow for…

Compare with previous version

changed title from WIP : RICH Photon Reco - Mirror Update to WIP : RICH Photon Reco - More efficient mirror data update implementation.

changed the description

unmarked as a Work In Progress

changed title from WIP : RICH Photon Reco - More efficient mirror data update implementation. to RICH Photon Reco - More efficient mirror data update implementation.

changed the description

changed title from RICH Photon Reco - More efficient mirror data update implementation. to RICH Photon Reco - More efficient mirror data cache update implementation.

mentioned in merge request !886 (merged)

changed the description

Just for the record, this introduces tiny numerical differences in some of the Brunel tests - look at diffs between https://lhcb-nightlies.cern.ch/logs/tests/nightly/lhcb-upgrade-hackathon/3/x86_64-centos7-gcc62-opt/Brunel/ and https://lhcb-nightlies.cern.ch/logs/tests/nightly/lhcb-upgrade-hackathon/2/x86_64-centos7-gcc62-opt/Brunel/, where the only difference is this MR

merged

mentioned in commit a6620e61

Tiny diffs like these are not impossible. One thing the MR removes is an un-neccessary (re)-normalisation of the (A,B,C,D) parameters used to describe a Plane. The normalisation is not required, as the parameters are already normalised, by removing this step could introduce a some minor diffs, in the lowest bits.