Skip to content

Commit 0982c20

Browse files
davidrohrdavidrohr
committed
GPU TPC: Implement ambiguity solving for rebuild tracks cluster association
1 parent 0cf2585 commit 0982c20

File tree

14 files changed

+258
-77
lines changed

14 files changed

+258
-77
lines changed

‎GPU/GPUTracking/Definitions/GPUSettingsList.h‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -136,6 +136,7 @@ AddOptionRTC(cfEdgeTwoPads, uint8_t, 0, "", 0, "Flag clusters with peak on the 2
136136
AddOptionRTC(nWays, uint8_t, 3, "", 0, "Do N fit passes in final fit of merger (must be odd to end with inward fit)")
137137
AddOptionRTC(rebuildTrackInFit, uint8_t, 1, "", 0, "Rebuild track completely during fit based on clusters closed to interpolated track positions")
138138
AddOptionRTC(rebuildTrackInFitClusterCandidates, uint8_t, 3, "", 0, "Number of cluster candidates per row for rebuilt track")
139+
AddOptionRTC(rebuildTrackMaxSharedFraction, float, 0.1f, "", 0, "Max fraction of shared clusters for rebuilt tracks")
139140
AddOptionRTC(trackFitRejectMode, int8_t, 5, "", 0, "0: no limit on rejection or missed hits, >0: break after n rejected hits, <0: reject at max -n hits")
140141
AddOptionRTC(rejectIFCLowRadiusCluster, uint8_t, 1, "", 0, "Reject clusters that get the IFC mask error during refit")
141142
AddOptionRTC(dEdxTruncLow, uint8_t, 2, "", 0, "Low truncation threshold, fraction of 128")

‎GPU/GPUTracking/Global/GPUChainTrackingDebugAndProfiling.cxx‎

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -180,8 +180,8 @@ void GPUChainTracking::PrintMemoryRelations()
180180
GPUInfo("MEMREL SectorTracks NCl %d NTrk %d", processors()->tpcTrackers[i].NHitsTotal(), *processors()->tpcTrackers[i].NTracks());
181181
GPUInfo("MEMREL SectorTrackHits NCl %d NTrkH %d", processors()->tpcTrackers[i].NHitsTotal(), *processors()->tpcTrackers[i].NTrackHits());
182182
}
183-
GPUInfo("MEMREL Tracks NCl %d NTrk %d", processors()->tpcMerger.NMaxClusters(), processors()->tpcMerger.NMergedTracks());
184-
GPUInfo("MEMREL TrackHitss NCl %d NTrkH %d", processors()->tpcMerger.NMaxClusters(), processors()->tpcMerger.NMergedTrackClusters());
183+
GPUInfo("MEMREL Tracks NCl %d NTrk %d", processors()->tpcMerger.NClusters(), processors()->tpcMerger.NMergedTracks());
184+
GPUInfo("MEMREL TrackHitss NCl %d NTrkH %d", processors()->tpcMerger.NClusters(), processors()->tpcMerger.NMergedTrackClusters());
185185
}
186186

187187
void GPUChainTracking::PrepareKernelDebugOutput()
@@ -209,7 +209,7 @@ void GPUChainTracking::PrintKernelDebugOutput()
209209
void GPUChainTracking::PrintOutputStat()
210210
{
211211
int32_t nTracks = 0, nAttachedClusters = 0, nAttachedClustersFitted = 0, nAdjacentClusters = 0;
212-
uint32_t nCls = GetProcessingSettings().doublePipeline ? mIOPtrs.clustersNative->nClustersTotal : processors()->tpcMerger.NMaxClusters();
212+
uint32_t nCls = GetProcessingSettings().doublePipeline ? mIOPtrs.clustersNative->nClustersTotal : processors()->tpcMerger.NClusters();
213213
if (GetProcessingSettings().createO2Output > 1) {
214214
nTracks = mIOPtrs.nOutputTracksTPCO2;
215215
nAttachedClusters = mIOPtrs.nMergedTrackHits;

‎GPU/GPUTracking/Global/GPUChainTrackingMerger.cxx‎

Lines changed: 10 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -195,12 +195,8 @@ int32_t GPUChainTracking::RunTPCTrackingMerger(bool synchronizeOutput)
195195
Merger.CheckCollectedTracks();
196196
}
197197

198-
uint32_t maxId = Merger.NMaxClusters();
199-
if (maxId > Merger.NMaxClusters()) {
200-
throw std::runtime_error("mNMaxClusters too small");
201-
}
202-
runKernel<GPUMemClean16>({{numBlocks, -ThreadCount(), 0, deviceType, RecoStep::TPCMerging}}, MergerShadow.SharedCount(), maxId * sizeof(*MergerShadow.SharedCount()));
203-
runKernel<GPUMemClean16>({{numBlocks, -ThreadCount(), 0, deviceType, RecoStep::TPCMerging}}, MergerShadow.ClusterAttachment(), maxId * sizeof(*MergerShadow.ClusterAttachment()));
198+
runKernel<GPUMemClean16>({{numBlocks, -ThreadCount(), 0, deviceType, RecoStep::TPCMerging}}, MergerShadow.SharedCount(), Merger.NClusters() * sizeof(*MergerShadow.SharedCount()));
199+
runKernel<GPUMemClean16>({{numBlocks, -ThreadCount(), 0, deviceType, RecoStep::TPCMerging}}, MergerShadow.ClusterAttachment(), Merger.NClusters() * sizeof(*MergerShadow.ClusterAttachment()));
204200
runKernel<GPUTPCGMMergerPrepareForFit, 0>(GetGridAuto(0, deviceType));
205201
CondWaitEvent(waitForTransfer, &mEvents->single);
206202
runKernel<GPUTPCGMMergerSortTracksQPt>(GetGridAuto(0, deviceType));
@@ -230,6 +226,12 @@ int32_t GPUChainTracking::RunTPCTrackingMerger(bool synchronizeOutput)
230226
}
231227
runKernel<GPUTPCGMMergerTrackFit>(doGPU ? GetGrid(Merger.NMergedTracks(), 0) : GetGridAuto(0), mergerSortTracks ? 1 : 0, 0);
232228
if (param().rec.tpc.rebuildTrackInFit) {
229+
runKernel<GPUMemClean16>({{numBlocks, -ThreadCount(), 0, deviceType, RecoStep::TPCMerging}}, MergerShadow.HitWeights(), Merger.NClusters() * sizeof(*MergerShadow.HitWeights()));
230+
for (int32_t i = 0; i < param().rec.tpc.rebuildTrackInFitClusterCandidates; i++) {
231+
runKernel<GPUTPCGMMergerHitWeights, GPUTPCGMMergerHitWeights::compute>(doGPU ? GetGrid(Merger.NMergedTracks(), 0) : GetGridAuto(0), i);
232+
}
233+
runKernel<GPUTPCGMMergerHitWeights, GPUTPCGMMergerHitWeights::resolve>(doGPU ? GetGrid(Merger.NMergedTracks(), 0) : GetGridAuto(0), 0);
234+
runKernel<GPUTPCGMMergerHitWeights, GPUTPCGMMergerHitWeights::resolveShared>(doGPU ? GetGrid(Merger.NMergedTracks(), 0) : GetGridAuto(0), 0);
233235
runKernel<GPUTPCGMMergerTrackFit>(doGPU ? GetGrid(Merger.NMergedTracks(), 0) : GetGridAuto(0), mergerSortTracks ? 1 : 0, 1);
234236
}
235237
runKernel<GPUTPCGMMergerFollowLoopers>(GetGridAuto(0));
@@ -268,7 +270,7 @@ int32_t GPUChainTracking::RunTPCTrackingMerger(bool synchronizeOutput)
268270
GPUMemCpy(RecoStep::TPCMerging, Merger.MergedTracksdEdx(), MergerShadow.MergedTracksdEdx(), Merger.NMergedTracks() * sizeof(*Merger.MergedTracksdEdx()), outputStream, 0);
269271
}
270272
GPUMemCpy(RecoStep::TPCMerging, Merger.Clusters(), MergerShadow.Clusters(), Merger.NMergedTrackClusters() * sizeof(*Merger.Clusters()), outputStream, 0);
271-
GPUMemCpy(RecoStep::TPCMerging, Merger.ClusterAttachment(), MergerShadow.ClusterAttachment(), Merger.NMaxClusters() * sizeof(*Merger.ClusterAttachment()), outputStream, 0);
273+
GPUMemCpy(RecoStep::TPCMerging, Merger.ClusterAttachment(), MergerShadow.ClusterAttachment(), Merger.NClusters() * sizeof(*Merger.ClusterAttachment()), outputStream, 0);
272274
}
273275
if (GetProcessingSettings().outputSharedClusterMap) {
274276
TransferMemoryResourceLinkToHost(RecoStep::TPCMerging, Merger.MemoryResOutputState(), outputStream, nullptr, waitEvent);
@@ -358,7 +360,7 @@ int32_t GPUChainTracking::RunTPCTrackingMerger(bool synchronizeOutput)
358360
}
359361

360362
if (GetProcessingSettings().debugLevel >= 2) {
361-
GPUInfo("TPC Merger Finished (output clusters %d / input clusters %d)", Merger.NMergedTrackClusters(), Merger.NClusters());
363+
GPUInfo("TPC Merger Finished (output clusters %d / input clusters %d)", Merger.NMergedTrackClusters(), Merger.NSectorHits());
362364
}
363365
return 0;
364366
}

‎GPU/GPUTracking/Global/GPUErrorCodes.h‎

Lines changed: 15 additions & 14 deletions
Original file line numberDiff line numberDiff line change
@@ -33,20 +33,21 @@ GPUCA_ERROR_CODE(13, ERROR_SECTORDATA_HITINROW_OVERFLOW, SectorRow, Value, Max)
3333
GPUCA_ERROR_CODE(14, ERROR_SECTORDATA_BIN_OVERFLOW, SectorRow, Value, Max)
3434
GPUCA_ERROR_CODE(15, ERROR_SECTORDATA_Z_OVERFLOW, Sector, Value)
3535
GPUCA_ERROR_CODE(16, ERROR_MERGER_HIT_OVERFLOW, Value, Max)
36-
GPUCA_ERROR_CODE(17, ERROR_MERGER_TRACK_OVERFLOW, Value, Max)
37-
GPUCA_ERROR_CODE(18, ERROR_COMPRESSION_ROW_HIT_OVERFLOW, SectorRow, Value, Max)
38-
GPUCA_ERROR_CODE(19, ERROR_LOOPER_MATCH_OVERFLOW, Value, Max)
39-
GPUCA_ERROR_CODE(20, ERROR_CF_PEAK_OVERFLOW, Sector, Value, Max)
40-
GPUCA_ERROR_CODE(21, ERROR_CF_CLUSTER_OVERFLOW, Sector, Value, Max)
41-
GPUCA_ERROR_CODE(22, ERROR_CF_ROW_CLUSTER_OVERFLOW, SectorRow, Value, Max)
42-
GPUCA_ERROR_CODE(23, ERROR_CF_GLOBAL_CLUSTER_OVERFLOW, SectorRow, Value, Max)
43-
GPUCA_ERROR_CODE(24, ERROR_DECOMPRESSION_ATTACHED_CLUSTER_OVERFLOW, SectorRow, Value, Max)
44-
GPUCA_ERROR_CODE(25, MAX_OVERFLOW_ERROR_NUMBER) // Overflow errors are detected as errno <= MAX_OVERFLOW_ERROR_NUMBER
36+
GPUCA_ERROR_CODE(17, ERROR_MERGER_REBUILD_HIT_OVERFLOW, Value, Max)
37+
GPUCA_ERROR_CODE(18, ERROR_MERGER_TRACK_OVERFLOW, Value, Max)
38+
GPUCA_ERROR_CODE(19, ERROR_COMPRESSION_ROW_HIT_OVERFLOW, SectorRow, Value, Max)
39+
GPUCA_ERROR_CODE(20, ERROR_LOOPER_MATCH_OVERFLOW, Value, Max)
40+
GPUCA_ERROR_CODE(21, ERROR_CF_PEAK_OVERFLOW, Sector, Value, Max)
41+
GPUCA_ERROR_CODE(22, ERROR_CF_CLUSTER_OVERFLOW, Sector, Value, Max)
42+
GPUCA_ERROR_CODE(23, ERROR_CF_ROW_CLUSTER_OVERFLOW, SectorRow, Value, Max)
43+
GPUCA_ERROR_CODE(24, ERROR_CF_GLOBAL_CLUSTER_OVERFLOW, SectorRow, Value, Max)
44+
GPUCA_ERROR_CODE(25, ERROR_DECOMPRESSION_ATTACHED_CLUSTER_OVERFLOW, SectorRow, Value, Max)
45+
GPUCA_ERROR_CODE(26, MAX_OVERFLOW_ERROR_NUMBER) // Overflow errors are detected as errno <= MAX_OVERFLOW_ERROR_NUMBER
4546

46-
GPUCA_ERROR_CODE(26, ERROR_TPCZS_INVALID_ROW, SectorRow) // Data from invalid row is skipped
47-
GPUCA_ERROR_CODE(27, ERROR_TPCZS_INVALID_NADC, SectorCRU, SamplesInPage, SamplesWritten) // Invalid number of ADC samples in header, existing samples were decoded
48-
GPUCA_ERROR_CODE(28, ERROR_TPCZS_INCOMPLETE_HBF, SectorCRU, PacketCount, NextPacketCount) // Part of HBF is missing, decoding incomplete
49-
GPUCA_ERROR_CODE(29, ERROR_TPCZS_INVALID_OFFSET, SectorEndpoint, Value, Expected) // Raw page is skipped since it contains invalid payload offset
50-
GPUCA_ERROR_CODE(29, MAX_GPUCA_ERROR_NUMBER)
47+
GPUCA_ERROR_CODE(27, ERROR_TPCZS_INVALID_ROW, SectorRow) // Data from invalid row is skipped
48+
GPUCA_ERROR_CODE(28, ERROR_TPCZS_INVALID_NADC, SectorCRU, SamplesInPage, SamplesWritten) // Invalid number of ADC samples in header, existing samples were decoded
49+
GPUCA_ERROR_CODE(29, ERROR_TPCZS_INCOMPLETE_HBF, SectorCRU, PacketCount, NextPacketCount) // Part of HBF is missing, decoding incomplete
50+
GPUCA_ERROR_CODE(30, ERROR_TPCZS_INVALID_OFFSET, SectorEndpoint, Value, Expected) // Raw page is skipped since it contains invalid payload offset
51+
GPUCA_ERROR_CODE(30, MAX_GPUCA_ERROR_NUMBER)
5152

5253
// #define GPUCA_CHECK_TPCZS_CORRUPTION

0 commit comments

Comments
 (0)