Enable event-local memory pool in PV reconstruction and selection algorithms

Depends on LHCb!2343 (merged), makes sense to test with Phys!650 (merged).

Changes the transient type for PVs to be LHCb::Event::v2::RecVertices.

Also:

• Propagates allocators through PrFilterIPSoA.cpp
• Enables the event-local memory pool for vertex relations.
• Enables the event-local memory pool for CombineTracks and adds a missing reserve().
• Enables the event-local memory pool for both TrackBeamLineVertexFinderSoA both for its output and intermediate objects.

Note that CombineTracks will not compile without gaudi/Gaudi!1017 (closed) or gaudi/Gaudi!1031 (merged).

cc: @nnolte @wouter @ahennequ

cc @adudziak

mentioned in merge request !1845 (closed)

@sstahl , @pseyfert @olupton Can somebody explain to me how this is related to Paul's zips? !1831 (closed) and related? Is it temporary until Paul's stuff is merged? or is it to suppress it?

added 2 commits

• ffd8fdc3 - Migrate high level selection, PV etc. algorithms to use the event-local memory pool.
• 7a7a09ab - Migrate PrFilterIPSoA to propagate allocators.

Compare with previous version

mentioned in merge request Phys!650 (merged)

mentioned in merge request LHCb!2343 (merged)

changed the description

I'd be more radical here: We are never going to change the number of bins anymore, so why not use an std::array?

mentioned in issue LHCb#73 (closed)

@wouter @adudziak please hit the "Approve" button once you've checked and agree with the changes

assigned to @wouter and @adudziak

Let me raise my concern. What is timescale of this being merge? vs what is timescale for the evaluation of PV performance between GPU and CPU? Do we want to really change the PV class right now? Allen already made transformer to std::vectorLHCb::Event::v2::RecVertex and is able to run performance checks, which @dovombru (!1897 (merged) and related Moore) has made last week. Also the PV checker which is basically there (waiting for the minor changes) is using for now std::vectorLHCb::Event::v2::RecVertex, changing this to LHCb::Event::v2::RecVertices should be indeed trivial, but it will introduce the chain of changes also Allen code. Could somebody responsible for the timescale comment here? @gligorov @sstahl

added PV finding enhancement labels

added lhcb-head label

approved this merge request

added needs Gaudi release label

removed lhcb-head label

added 1 commit

• d1e1cd83 - Avoid explicit EventContext in CombineTracks.

Compare with previous version