Skip to content

Latest commit

 

History

History
100 lines (83 loc) · 8.14 KB

processors.md

File metadata and controls

100 lines (83 loc) · 8.14 KB

Legacy ICARUS processor inventory

This inventory describes the validated ICARUS overlay configuration, not every possible LArSoft Supera configuration. Runtime order is not the proposed porting order.

Authoritative baseline

There are two scheduling layers. art reconstructs hits and space points, runs mcreco, and invokes LArSoftSuperaDriver through the superaMC analyzer. The driver then runs the following LArCV ProcessDriver sequence. mcreco and cluster3D are prerequisites, not entries in this processor list.

Processors in execution order

There are 16 top-level instances, with 13 distinct processor types.

Order Instance (type when different) Configured behavior Proposed responsibility
1 MultiPartVrtx (SuperaMCTruth) Read generator and optional-in-practice cosmgen MCTruth; emit interaction/neutrino and generator-particle records under mpv. Adapter extraction; shared representation and explicit selection policy.
2 SuperaBBoxInteraction Establish the common fixed 3D grid and assign metadata to configured products. Detector/configuration input in the adapter; grid semantics in shared code.
3 SuperaMCParticleCluster Read simplemerge MCParticles and mcreco tracks/showers; build particle groups, energy clusters, semantic labels, and diagnostic products. Uses nested true-to-reco matching. Split LArSoft extraction from shared grouping and labeling. Hardest migration stage.
4 SuperaSimEnergyDeposit Voxelize filtersed SimEnergyDepositLite into sed, using the already-produced particle groups and mcreco shower daughters. Adapter reads deposits; shared code owns grouping and aggregation.
5 SuperaSpacePointCryoE (SuperaSpacePoint) Read east cluster3D points and hit associations; produce reconstructed charge and auxiliary tensors. Adapter extracts points/hits; shared code handles voxel collisions and selection.
6 SuperaSpacePointCryoW (SuperaSpacePoint) The corresponding west-cryostat conversion. Same implementation, configured for another input collection.
7 Tensor3DFromCluster3D Convert pcluster and sed cluster collections into sparse tensors, retaining input voxel values (PITypeList: [1,1]). Shared tensor transformation.
8 CombineTensor3DGhost (CombineTensor3D) Combine east/west reconstructed tensors into pcluster_semantics_ghost, with min pooling (PoolType: 0). Shared tensor transformation.
9 ThresholdTensor3D Paint the reconstructed voxel support with semantic value 5, the configured ghost value. Shared labeling policy.
10 CombineTensor3D Combine the painted tensor with pcluster_semantics, again with min pooling, to form the final semantic/ghost tensor. Shared labeling policy; explicitly test overlap behavior.
11 ParticleCorrector Derive corrected particle records from pcluster particles and pcluster_highE clusters; includes cluster-based endpoint correction. Exact changed fields/defaults need review. Shared post-processing.
12 EmptyTensorFilter Require at least one voxel in pcluster_semantics_ghost. Filtering is enabled; rejection can prevent subsequent processors from running. Pipeline selection policy.
13 RescaleChargeTensor3DCryoE (RescaleChargeTensor3D) Use three planes of hit keys/charges and the pcluster reference to produce reco_cryoE_rescaled. Review shared-hit charge apportionment. Shared post-processing using adapter-provided associations.
14 RescaleChargeTensor3DCryoW (RescaleChargeTensor3D) The corresponding west-cryostat charge rescaling. Same shared implementation, separate inputs.
15 SuperaOptical Translate opflashCryoE and opflashCryoW into flash records: times, photoelectrons, positions/widths, flags, wire information, and collection-local IDs. Adapter extraction; independent output serialization.
16 SuperaCRT Translate overlayCRTHit into CRT records, including timestamps, photoelectrons, detector identifiers, positions, and uncertainties. Experiment-specific extraction behind a shared record contract.

The LArCV2 installed headers identify PoolType: 0 as min pooling and PIType: 1 as input-voxel values. Full collision, in-place update, correction, and charge rescaling behavior still needs implementation-level review and small tests. Configuration names alone are not sufficient specifications.

Nested processor: SuperaTrue2RecoVoxel3D

This runs inside SuperaMCParticleCluster; it is not a seventeenth top-level entry. It connects merge SimChannels with both cluster3D point collections and their hit associations. The baseline enables twofold matching, true positions, and post-averaging. It uses a 100-electron threshold, a 15-tick window, a 3-sigma window setting, a 0.425 cm post-averaging threshold, and a voxel-distance setting of 3.0. Units and interactions between cuts must be checked against the code.

The adapter must expose the relevant evidence and detector context; the matching policy needs a separately reviewed home. Do not hide this algorithm in a reader.

Details that matter for the port

  • The grid has bottom corner [-412.89, -181.86, -894.951] cm, side lengths [1843.2, 1843.2, 1843.2] cm, and [0.3, 0.3, 0.3] cm voxels.
  • SuperaMCParticleCluster sets both UseSimEnergyDeposit and UseSimEnergyDepositLite to false. Its main filling path uses SimChannels. The separate SuperaSimEnergyDeposit processor uses SEDLite. These are different inputs and cannot be substituted silently.
  • The particle semantic priority is [1,2,0,3,4]; the configuration identifies labels 0 through 4 as shower, track, Michel, delta, and low-energy scattering.
  • Space-point charge selection is configured as [-1000,50000]. Wire information is enabled. Hit amplitude/time/RMS/multiplicity, hit count, and occupancy outputs are dropped; hit keys and charges remain for rescaling.
  • StoreAbsTime: true is configured for SEDLite, but the selected source helper fills only energy clusters. See F-002.
  • This input has generator MCTruth and no cosmgen product. The MCTruth processor reports a missing handle and continues. Missing and empty products must remain distinguishable in the new contract.
  • Real-data activity in overlay has no complete simulation ancestry. A ghost label describes this pipeline's matching result; it is not proof that a reconstructed point is physically spurious.

Source entry points

The pinned Supera tree contains SuperaMCTruth.cxx, SuperaBBoxInteraction.cxx, SuperaMCParticleCluster.cxx, SuperaTrue2RecoVoxel3D.cxx, SuperaSimEnergyDeposit.cxx, SuperaSpacePoint.cxx, SuperaOptical.cxx, and SuperaCRT.cxx. The remaining processor implementations belong to LArCV2, not this Supera tree. For review, use the installed v2_2_6 interfaces and the corresponding source revision; do not silently inspect a newer LArCV implementation.

These ownership assignments are proposals. The current Atomic event model contains particle inputs and unassociated deposits; it does not provide a complete schema for reconstructed hits, flashes, and CRT records. Establish that interface explicitly before implementing conversions.