Conversation
Extend MatchTPCITS to work with ITS clusters provided either as a single (monolithic) input or per layer, with layer-dependent ROF length and bias (staggered readout): - ITS clusters, sizes, cluster ROFRecords and MC labels are stored per layer, clusters are addressed by the composed ID (layer<<28)+index_in_layer (all in slot 0 for the monolithic input) - all ITS ROF timings (per-layer lengths and biases in BC and mus, the clock layer defining the ITS tracks ROFs granularity) are derived from the DPLAlpideParam object set via setAlpideParam; the setITSROFrameLength... and setITSTimeBiasInBC setters are removed - interaction candidates are related to the cluster ROFs of every AfterBurner layer within the optional abROFMarginMUS margin (allowing up to 2 compatible ROFs per layer) and are cut at the last clock-layer cluster ROF - AfterBurner reworked for CPU efficiency: unused clusters are filtered and (chip,Z)-sorted once per TF into per (layer, ROF) blocks, built only for the ROFs referenced by candidates with seeds; ITSChipClustersRefs is replaced by thread-local per-layer views (compact Y,Z,id,chip cluster info) refreshed only when the processed group of candidates changes ROFs - tpcits-match-workflow can request the per-layer ITS clusters input (through RecoContainer::setITSPerLayer) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…hing MatchTPCITS previously assigned every ITS track the full clock-layer ROF duration as its time bracket. TrackITS now carries a per-track TimeStamp (BC since TF start, symmetric error, ROF bias already applied), which is typically much narrower than the ROF, especially in staggered running. - prepareITSData builds the per-track bracket from getTimeStamp(), widened by the new itsTimeStampMarginBC margin (BC, both edges), falling back to the nominal ROF bracket when the time stamp is invalid (legacy input) - mITSROFTimes is extended to the envelope of the nominal ROF bracket and the actual per-track brackets, keeping the TPC-side and triggered-mode ITS ROF entry caches conservative - mITSMaxROFOverhangMUS tracks how far track brackets extend past their ROF end in the current TF; doMatching's continuous-mode entry lookup is shifted by this amount so no compatible track is skipped - sorting by bracket min time still cannot mix tracks of different ROFs (the tracker guarantees the raw lower edge stays within the assigned ROF, and the margin shifts all tracks alike), so the existing mITSTimeStart assignment and the tBracket-based break/continue gates in doMatching remain valid without a LUT rebuild; the long-dead RejectOnTgl ROF-skip code (whose precondition never held with mixed layers) is removed - refitTrackTPCITS derives the fallback ITS time error from the track's own bracket (delta()/sqrt(12)) instead of the nominal clock-layer ROF resolution Net effect: most TPC x ITS pairs are now rejected by the cheap bracket overlap check in doMatching before the sqrt/kinematic comparisons, with no new containers and a single extra float member. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
With the per-layer (staggered readout) ITS clusters input the cluster index kept by TrackITS is local to its layer, so the layer must be encoded into the stored reference for the consumers to be able to find the cluster. MatchTPCITS already decoded the ITS/TRACKCLSID entries as such composed IDs, but the tracker pushed the bare per-layer index, which was correct only for the layer 0. - ITSTrackingInterface::run composes the stored reference as (layer << ClusLayerShift) + index_in_layer; with the monolithic clusters input the layer slot is 0 and the composed ID stays equal to the flat index, so the non-staggered output is unchanged - the ID composition/decomposition and the max number of separately provided ITS/MFT cluster layers move from MatchTPCITS.h to the new lightweight DataFormatsITSMFT/ClusterID.h, so that both the producer and the (many) consumers can use them without pulling in GlobalTracking; RecoContainer.h includes it and keeps MaxITSLayers/MaxMFTLayers as aliases - the layer field is shifted by 27 rather than 28 bits, so that it can accommodate the MFT layers too: the bit 31 is unusable, since the negative values of the composed ID are reserved for the "no cluster" flags - the unused TrackITSExt::setClusterIndex, carrying its own hardcoded copy of the composition (and writing to the packed slot while getClusterIndex reads the layer slot), is removed Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
d2ba570 to
1472563
Compare
|
@mpuccio @f3sch Contrary to what I thought, the compound indices |
Sorry I may misunderstand but why is <<27 needed for MFT should not <<28 shift give you already 0-15 range? |
@f3sch these IDs are signed ints and the negative values are used internally as flags. But even with 27 we still have more than enough room to accommodate all clusters: at 50kH PbPb we have <5M clusters, we can accommodate 134M? |
Ah ok, thanks. Indeed more than enough headroom. |
This branch has not been deployed
3 participants
No description provided.