Tracking Development Branch Rebased on new Transfomration Maps - DO NOT MERGE

GPU/Common/GPUCommonMath.h

@@ -106,6 +106,7 @@ class GPUCommonMath
   GPUd() constexpr static double Copysignd(double x, double y);
   GPUd() constexpr static float TwoPi() { return 6.2831853f; }
   GPUd() constexpr static float Pi() { return 3.1415927f; }
+  GPUd() constexpr static float Deg2Rad() { return Pi() / 180.f; }
   GPUd() constexpr static float Round(float x);
   GPUd() constexpr static float Floor(float x) { return GPUCA_CHOICE(floorf(x), floorf(x), floor(x)); }
   GPUd() static uint32_t Float2UIntReint(const float& x);

GPU/GPUTracking/Base/GPUParam.h

@@ -80,8 +80,8 @@ struct GPUParam : public internal::GPUParam_t<GPUSettingsRec, GPUSettingsParam>
   void UpdateRun3ClusterErrors(const float* yErrorParam, const float* zErrorParam);
 #endif
 
-  GPUd() constexpr uint32_t tpcMinHitsB5(float qPtB5) const { return CAMath::Abs(qPtB5) > 10 ? 10 : (CAMath::Abs(qPtB5) > 5 ? 15 : 29); } // Minimum hits should depend on Pt, low Pt tracks can have few hits. 29 Hits default, 15 for < 200 mev, 10 for < 100 mev
-  GPUd() constexpr float Alpha(int32_t iSector) const                                                                                     // TODO: Check if this is faster, or the lookup from GPUTPCGeometry
+  GPUd() constexpr uint32_t tpcMinHitsB5Abs(float qPtB5) const { return qPtB5 > 10.f ? 10 : (qPtB5 > 5.f ? (uint32_t)(20 - qPtB5) : (qPtB5 > 3.6f ? (uint32_t)(65 - 10.f * qPtB5) : 29)); }
+  GPUd() constexpr float Alpha(int32_t iSector) const // TODO: Check if this is faster, or the lookup from GPUTPCGeometry
   {
     if (iSector >= (int32_t)o2::tpc::constants::MAXSECTOR / 2) {
       iSector -= o2::tpc::constants::MAXSECTOR / 2;

GPU/GPUTracking/Base/GPUReconstruction.cxx

@@ -295,7 +295,7 @@ int32_t GPUReconstruction::InitPhaseBeforeDevice()
   if (GetProcessingSettings().deterministicGPUReconstruction && GetProcessingSettings().debugLevel >= 6) {
     mProcessingSettings->nTPCClustererLanes = 1;
   }
-  if (GetProcessingSettings().createO2Output > 1 && GetProcessingSettings().runQA && GetProcessingSettings().qcRunFraction == 100.f) {
+  if (GetProcessingSettings().createO2Output > 1 && ((GetProcessingSettings().runQA && GetProcessingSettings().qcRunFraction == 100.f) || GetProcessingSettings().debug.mergerDumpTrackNCl)) {
     mProcessingSettings->createO2Output = 1;
   }
   if (!GetProcessingSettings().createO2Output || !IsGPU()) {

GPU/GPUTracking/Base/GPUReconstructionCPU.cxx

@@ -216,10 +216,10 @@ int32_t GPUReconstructionCPU::ExitDevice()
 int32_t GPUReconstructionCPU::RunChains()
 {
   mMemoryScalers->temporaryFactor = 1.;
-  if (GetProcessingSettings().memoryScalingFuzz) {
+  if (GetProcessingSettings().debug.memoryScalingFuzz) {
     static std::mt19937 rng;
     static std::uniform_int_distribution<uint64_t> dist(0, 1000000);
-    uint64_t fuzzFactor = GetProcessingSettings().memoryScalingFuzz == 1 ? dist(rng) : GetProcessingSettings().memoryScalingFuzz;
+    uint64_t fuzzFactor = GetProcessingSettings().debug.memoryScalingFuzz == 1 ? dist(rng) : GetProcessingSettings().debug.memoryScalingFuzz;
     GPUInfo("Fuzzing memory scaling factor with %lu", fuzzFactor);
     mMemoryScalers->fuzzScalingFactor(fuzzFactor);
   }

GPU/GPUTracking/DataTypes/GPUTPCGeometry.h

@@ -16,7 +16,11 @@
 #define GPUTPCGEOMETRY_H
 
 #include "GPUCommonDef.h"
+#include "GPUCommonMath.h"
 #include "DataFormatsTPC/Constants.h"
+#ifndef GPUCA_GPUCODE_DEVICE
+#include <cmath>
+#endif
 
 namespace o2::gpu
 {
@@ -174,6 +178,8 @@ class GPUTPCGeometry
     const float v = (sector >= NSECTORS / 2) ? -z : z;
     return (250.f - v) * FACTOR_Z2T; // Used in compression, must remain constant at 250cm
   }
+
+  GPUd() static constexpr float kSectAngle() { return 2 * CAMath::Pi() / 18.f; }
 };
 
 } // namespace o2::gpu

GPU/GPUTracking/Definitions/GPUDef.h

@@ -70,16 +70,30 @@
   #ifdef CADEBUG
     #undef CADEBUG
   #endif
-  #if GPUCA_CADEBUG == 1 && !defined(GPUCA_GPUCODE)
-    #define CADEBUG(...) __VA_ARGS__
-    #define CADEBUG2(cmd, ...) {__VA_ARGS__; cmd;}
-    #define GPUCA_CADEBUG_ENABLED
+  #ifdef CADEBUG2
+    #undef CADEBUG2
+  #endif
+  #ifdef CADEBUG3
+    #undef CADEBUG3
+  #endif
+  #if GPUCA_CADEBUG >= 1 && !defined(GPUCA_GPUCODE)
+    #if GPUCA_CADEBUG >= 2
+      #define CADEBUG(...) if (CADEBUG_CONDITION) {__VA_ARGS__;}
+      #define CADEBUG2(cmd, ...) if (CADEBUG_CONDITION) {__VA_ARGS__; cmd;}
+      #define GPUCA_CADEBUG_ENABLED 2
+    #else
+      #define CADEBUG(...) __VA_ARGS__
+      #define CADEBUG2(cmd, ...) {__VA_ARGS__; cmd;}
+      #define GPUCA_CADEBUG_ENABLED 1
+    #endif
+    #define CADEBUG3(...) __VA_ARGS__
   #endif
   #undef GPUCA_CADEBUG
 #endif
 
 #ifndef CADEBUG
   #define CADEBUG(...)
   #define CADEBUG2(cmd, ...) {cmd;}
+  #define CADEBUG3(...)
 #endif
 // clang-format on

GPU/GPUTracking/Definitions/GPUSettingsList.h

@@ -32,7 +32,7 @@ BeginNamespace(o2::gpu)
 
 // Reconstruction parameters for TPC, no bool in here !!!
 BeginSubConfig(GPUSettingsRecTPC, tpc, configStandalone.rec, "RECTPC", 0, "Reconstruction settings", rec_tpc)
-AddOptionRTC(rejectQPtB5, float, 1.f / 0.050f, "", 0, "QPt threshold to reject clusters of TPC tracks (Inverse Pt, scaled to B=0.5T!!!)")
+AddOptionRTC(rejectQPtB5, float, 1.f / 0.050f, "", 0, "QPt threshold to reject clusters of TPC tracks (Inverse Pt, scaled to B=0.5T!!!)") // TODO: Sort these options automatically for parameter size
 AddOptionRTC(hitPickUpFactor, float, 1.f, "", 0, "multiplier for the combined cluster+track error during track following")
 AddOptionRTC(hitSearchArea2, float, 2.f, "", 0, "square of maximum search road of hits during seeding")
 AddOptionRTC(neighboursSearchArea, float, 3.f, "", 0, "area in cm for the search of neighbours, for z only used if searchWindowDZDR = 0")
@@ -102,6 +102,20 @@ AddOptionRTC(rejectEdgeClustersMargin, float, 0.f, "", 0, "Margin in cm of Y pos
 AddOptionRTC(rejectEdgeClustersSigmaMargin, float, 0.f, "", 0, "Margin factor for trackSigmaY when rejecting edge clusters based on uncorrected track Y")
 AddOptionRTC(trackletMaxSharedFraction, float, 0.1f, "", 0, "Max fraction of shared clusters for tracklet")
 AddOptionRTC(trackletMinSharedNormFactor, float, 0.f, "", 0, "Max shared defined as trackletMinSharedNormFactor*max(current_nhits,trackletMinSharedNormFactor*minHits,1)")
+AddOptionRTC(trackFitMinHitsMaxQPtIncrease, float, 0.2f, "", 0, "For the QPt-dependent minimum number of hits of the track fit, abs(QPt) is increased by 1 sigma, but at most by this fraction (0 = use fitted QPt)")
+AddOptionRTC(rebuildTrackMaxSharedFraction, float, 0.1f, "", 0, "Max fraction of shared clusters for rebuilt tracks")
+AddOptionRTC(rebuildTrackMaxNonIntCov, float, 4.f, "", 0, "Max Err2 allowed for non-interpolated cluster attachment during rebuild")
+AddOptionRTC(rebuildErrorWeightFactor, float, 5.f, "", 0, "Factor for the error (distance^2) of a cluster candidate in its hit weight 1 / (1 + factor * error) during rebuild")
+AddOptionRTC(maxSinPhiForFitUpdate, float, 0.9396926f, "", 0, "Max abs(sinPhi) of the track to update it with a cluster in the track fit)") // default sin(70 deg = 0.9396926
+AddOptionRTC(rebuildTrackMaxQPtDrift, float, 0.5f, "", 0, "Stop the track extrapolation during rebuild when QPt changed by more than this fraction of max(1, abs(QPt)) compared to before the extrapolation (0 = disabled)")
+AddOptionRTC(rebuildTrackExtrMaxAvgChi2, float, 4.f, "", 0, "Max average chi2 per cluster to accept an extrapolated segment (0 = disabled)")
+AddOptionRTC(rebuildTrackFallbackErr, float, 1.f, "", 0, "Error (distance^2) assigned to the original clusters of a track used as fallback candidates in rows that could not be searched during rebuild")
+AddOptionRTC(rebuildTrackFallbackTrackChi2, float, 4.f, "", 0, "Chi2 per cluster assumed for the hit weight of tracks without successful fit (NDF <= 0) during rebuild")
+AddOptionRTC(rebuildTrackAltCandidateMaxErrFactor, float, 2.f, "", 0, "Alternative cluster candidates of a row are used only if their distance^2 is below this factor times the distance^2 of the best candidate")
+AddOptionRTC(rebuildTrackAltCandidateGoodErr, float, 0.25f, "", 0, "Alternative cluster candidates of a row with distance^2 below this value are always used")
+AddOptionRTC(rebuildTrackErrFactorOneSided, float, 1.5f, "", 0, "Factor applied to the error (distance^2) of cluster candidates found with the track position from only one side during rebuild")
+AddOptionRTC(rebuildTrackErrFactorExtrapolation, float, 1.f, "", 0, "Factor applied to the error (distance^2) of cluster candidates found by track extrapolation during rebuild")
+AddOptionRTC(rebuildTrackMaxInterpolationDistY, float, 4.f, "", 0, "Max distance in Y between the track positions from both sides to use their interpolation for the cluster search in rows without cluster during rebuild")
 AddOptionRTC(maxTimeBinAboveThresholdIn1000Bin, uint16_t, 500, "", 0, "Except pad from cluster finding if total number of charges in a fragment is above this baseline (disable = 0)")
 AddOptionRTC(maxConsecTimeBinAboveThreshold, uint16_t, 200, "", 0, "Except pad from cluster finding if number of consecutive charges in a fragment is above this baseline (disable = 0)")
 AddOptionRTC(noisyPadSaturationThreshold, uint16_t, 700, "", 0, "Threshold where a timebin is considered saturated, disabling the noisy pad check for that pad")
@@ -135,12 +149,16 @@ AddOptionRTC(cfNoiseSuppressionEpsilon, uint8_t, 10, "", 0, "Cluster Finder: Dif
 AddOptionRTC(cfNoiseSuppressionEpsilonRelative, uint8_t, 76, "", 0, "Cluster Finder: Difference between peak and charge for the charge to count as a minima during noise suppression, relative as fraction of 255")
 AddOptionRTC(cfEdgeTwoPads, uint8_t, 0, "", 0, "Flag clusters with peak on the 2 pads closes to the sector edge as edge cluster")
 AddOptionRTC(nWays, uint8_t, 3, "", 0, "Do N fit passes in final fit of merger (must be odd to end with inward fit)")
-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")
+AddOptionRTC(rebuildTrackInFit, uint8_t, 1, "", 0, "Rebuild track completely during fit based on clusters closed to interpolated track positions")
+AddOptionRTC(rebuildTrackInFitClusterCandidates, uint8_t, 3, "", 0, "Number of cluster candidates per row for rebuilt track")
+AddOptionRTC(disableRebuildAttachment, uint8_t, 0, "", 0, "Bitmask to disable certain attachment steps during track rebuid (1: current row, 2: interpolate missing row, 4: one sided interpolation, 8: original hit, 16: extrapolation)")
+AddOptionRTC(disableMarkAdjacent, uint8_t, 0, "", 0, "Bitmask to disable certain steps during refit to mark adjacenrt clusters (1: current row, 2: extrapolate missing rows, 4: loop following)") // TODO: Add history
+AddOptionRTC(trackFitRejectMode, int8_t, 5, "", 0, "0: no limit on rejection or missed hits, >0: break after n rejected hits")
 AddOptionRTC(rejectIFCLowRadiusCluster, uint8_t, 1, "", 0, "Reject clusters that get the IFC mask error during refit")
 AddOptionRTC(dEdxTruncLow, uint8_t, 2, "", 0, "Low truncation threshold, fraction of 128")
 AddOptionRTC(dEdxTruncHigh, uint8_t, 77, "", 0, "High truncation threshold, fraction of 128")
 AddOptionRTC(extrapolationTracking, int8_t, 1, "", 0, "Enable Extrapolation Tracking (prolong tracks to adjacent sectors to find short segments)")
-AddOptionRTC(disableRefitAttachment, uint8_t, 0, "", 0, "Bitmask to disable certain attachment steps during refit (1: attachment, 2: propagation, 4: loop following, 8: mirroring)")
+AddOptionRTC(looperFollowMode, uint8_t, 1, "", 0, "0 = simple, 1 = advances, for disabling use disableMarkAdjacent = 4")
 AddOptionRTC(rejectionStrategy, uint8_t, o2::gpu::GPUSettings::RejectionStrategyA, "", 0, "Enable rejection of TPC clusters for compression (0 = no, 1 = strategy A, 2 = strategy B)")
 AddOptionRTC(mergeLoopersAfterburner, uint8_t, 1, "", 0, "Run afterburner for additional looper merging")
 AddOptionRTC(compressionTypeMask, uint8_t, o2::gpu::GPUSettings::CompressionFull, "", 0, "TPC Compression mode bits (1=truncate charge/width LSB, 2=differences, 4=track-model)")
@@ -152,7 +170,7 @@ AddOptionRTC(mergerInterpolateErrors, uint8_t, 1, "", 0, "Use interpolation inst
 AddOptionRTC(mergerInterpolateRejectAlsoOnCurrentPosition, uint8_t, 1, "", 0, "When using mergerInterpolateErrors, reject based on chi2 twice computed with interpolated and current track position starting from NDF > mergerNonInterpolateRejectMinNDF")
 AddOptionRTC(mergerNonInterpolateRejectMinNDF, uint8_t, 5, "", 0, "Minimum NDF of track for non-interpolated reject (both for chi2 and absolute distance)")
 AddOptionRTC(mergeCE, uint8_t, 1, "", 0, "Merge tracks accross the central electrode")
-AddOptionRTC(retryRefit, int8_t, 1, "", 0, "Retry refit with seeding errors and without cluster rejection when fit fails (=2 means retry in same kernel, =1 for separate kernel")
+AddOptionRTC(retryRefit, int8_t, 1, "", 0, "Retry refit with seeding errors and without cluster rejection when fit fails")
 AddOptionRTC(enablePID, int8_t, 1, "", 0, "Enable PID response")
 AddOptionRTC(PID_useNsigma, int8_t, 1, "", 0, "Use nSigma instead of absolute distance in PID response")
 AddOptionRTC(adddEdxSubThresholdClusters, int8_t, 1, "", 0, "Add sub threshold clusters in TPC dEdx computation")
@@ -162,8 +180,16 @@ AddOptionRTC(rejectEdgeClustersInSeeding, int8_t, 0, "", 0, "Reject edge cluster
 AddOptionRTC(rejectEdgeClustersInTrackFit, int8_t, 0, "", 0, "Reject edge clusters based on uncorrected track Y during track fit")
 AddOptionRTC(tubeExtraProtectMinRow, uint8_t, 20, "", 0, "Increase Protection, decrease removal by factor 2, when below this row")
 AddOptionRTC(tubeExtraProtectEdgePads, uint8_t, 2, "", 0, "Increase Protection, decrease removal by factor 2, when on this number of pads from the edge")
-
-AddOptionArray(PID_remap, int8_t, 9, (0, 1, 2, 3, 4, 5, 6, 7, 8), "", 0, "Remap Ipid to PID_reamp[Ipid] (no remap if<0)") // BUG: CUDA cannot yet hand AddOptionArrayRTC
+AddOptionRTC(rebuildTrackExtrMaxMissingRows, uint8_t, 20, "", 0, "Maximum total number of rows allowed to be missing during track extrapolation")
+AddOptionRTC(rebuildTrackExtrMinConsecGoodRows, uint8_t, 8, "", 0, "Minimum number of consecutive rows required to accept extrapolated segment")
+AddOptionRTC(rebuildTrackExtrMinConsecGoodRowsEdge, uint8_t, 3, "", 0, "Minimum number of consecutive rows required to accept extrapolated segment")
+AddOptionRTC(rebuildTrackExtrMaxRowsBack, uint8_t, 3, "", 0, "After the track extrapolation, propagate the track back to the last accepted cluster if it is at most this number of rows away, otherwise revert to the track before the extrapolation")
+AddOptionRTC(rebuildTrackExtrConsecGoodRowsMaxGap, uint8_t, 1, "", 0, "Max row gap size to consider the segment as consecutive rows")
+AddOptionRTC(rebuildTrackCandidateErrChi2, uint8_t, 0, "", 0, "Rank cluster candidates during rebuild by chi2 instead of distance^2 (rebuildTrackAltCandidateGoodErr is then in units of chi2)")
+AddOptionRTC(rebuildTrackMaxRowsFromEnd, uint8_t, 10, "", 0, "If the last fitted hit of the outward way of the rebuild is more than this number of rows away from the end of the track, the final fit starts from the original track parameters")
+AddOptionRTC(rebuildTrackFallbackMaxNCl, uint8_t, 20, "", 0, "For tracks with at most this number of clusters, keep the original cluster of a row as candidate during rebuild if no candidate was found but it passes the interpolated chi2 cut (0 = disabled)")
+AddOptionRTC(rebuildTrackNonIntMinNDF, uint8_t, 10, "", 0, "NDF the track must exceed for non-interpolated cluster attachment (one-sided search and extrapolation) during rebuild")
+AddOptionArray(PID_remap, int8_t, 9, (0, 1, 2, 3, 4, 5, 6, 7, 8), "", 0, "Remap Ipid to PID_reamp[Ipid] (no remap if<0)") // BUG: CUDA cannot yet handle AddOptionArrayRTC
 AddHelp("help", 'h')
 EndConfig()
 
@@ -322,6 +348,14 @@ AddOption(tpcDecodingSafetyBuffer, uint16_t, 1000, "", 0, "Scaling Factor: safet
 AddHelp("help", 'h')
 EndConfig()
 
+// Debug Settings
+BeginSubConfig(GPUSettingsProcessingDebug, debug, configStandalone.proc, "DEBUG", 0, "Debugging Settings", proc_debug)
+AddOption(memoryScalingFuzz, uint64_t, 0, "", 0, "Fuzz the memoryScalingFactor (0 disable, 1 enable, >1 set seed", def(1))
+AddOption(mergerMCLabels, bool, false, "", 0, "Create MC labels for merged tracks before refit for debugging")
+AddOption(mergerDumpTrackNCl, bool, false, "", 0, "Dump file with nCl per merged track fit")
+AddHelp("help", 'h')
+EndConfig()
+
 // Settings steering the processing once the device was selected, only available on the host
 BeginSubConfig(GPUSettingsProcessing, proc, configStandalone, "PROC", 0, "Processing settings", proc)
 AddOption(deviceNum, int32_t, -1, "gpuDevice", 0, "Set GPU device to use (-1: automatic, -2: for round-robin usage in timeslice-pipeline)")
@@ -351,7 +385,6 @@ AddOption(memoryAllocationStrategy, int8_t, 0, "", 0, "Memory Allocation Strateg
 AddOption(forceMemoryPoolSize, uint64_t, 1, "memSize", 0, "Force size of allocated GPU / page locked host memory (0 = disable, 1 / 2 = auto, other values set the size of the memory pool)", min(0ul))
 AddOption(forceHostMemoryPoolSize, uint64_t, 0, "hostMemSize", 0, "Force size of allocated host page locked host memory (overriding memSize)", min(0ul))
 AddOption(memoryScalingFactor, float, 1.f, "", 0, "Factor to apply to all memory scalers")
-AddOption(memoryScalingFuzz, uint64_t, 0, "", 0, "Fuzz the memoryScalingFactor (0 disable, 1 enable, >1 set seed", def(1))
 AddOption(tpcInputWithClusterRejection, uint8_t, 0, "", 0, "Indicate whether the TPC input is CTF data with cluster rejection, to tune buffer estimations")
 AddOption(forceMaxMemScalers, uint64_t, 0, "", 0, "Force using the maximum values for all buffers, Set a value n > 1 to rescale all maximums to a memory size of n")
 AddOption(registerStandaloneInputMemory, bool, false, "registerInputMemory", 0, "Automatically register input memory buffers for the GPU")
@@ -398,7 +431,7 @@ AddOption(tpcUseOldCPUDecoding, bool, false, "", 0, "Enable old CPU-based TPC de
 AddOption(tpcApplyCFCutsAtDecoding, bool, false, "", 0, "Apply cluster cuts from clusterization during decoding of compressed clusters")
 AddOption(tpcApplyClusterFilterOnCPU, uint8_t, 0, "", 0, "Apply custom cluster filter of GPUTPCClusterFilter class, 0: off, 1: debug, 2: PbPb23")
 AddOption(tpcWriteClustersAfterRejection, bool, false, "", 0, "Apply TPC rejection strategy before writing clusters")
-AddOption(oclPlatformNum, int32_t, -1, "", 0, "Platform to use, in case the backend provides multiple platforms (OpenCL only, -1 = auto-select, -2 query all platforms (also incompatible))")
+AddOption(oclPlatformNum, int32_t, -1, "", 0, "Platform to use, in case the backend provides multiple platforms (OpenCL only, -1 = auto-select, -2 query all platforms (also incompatible))") // TODO: Create some backend-specific options
 AddOption(oclCompileFromSources, bool, false, "", 0, "Compile OpenCL binary from included source code instead of using included spirv code")
 AddOption(oclOverrideSourceBuildFlags, std::string, "", "", 0, "Override OCL build flags for compilation from source, put a space for empty options")
 AddOption(metalOverrideSourceBuildFlags, std::string, "", "", 0, "Override Metal build flags for compilation from source, put a space for empty options")
@@ -422,6 +455,7 @@ AddSubConfig(GPUSettingsProcessingRTCtechnical, rtctech)
 AddSubConfig(GPUSettingsProcessingParam, param)
 AddSubConfig(GPUSettingsProcessingNNclusterizer, nn)
 AddSubConfig(GPUSettingsProcessingScaling, scaling)
+AddSubConfig(GPUSettingsProcessingDebug, debug)
 AddHelp("help", 'h')
 EndConfig()
 #endif // !__OPENCL__ && !__METAL__

GPU/GPUTracking/Definitions/Parameters/GPUParameters.csv

@@ -52,6 +52,11 @@ GPUTPCGMMergerPrepareForFit_step2,256,,256,256,,,,,,256,256,256,256,,256,,256,25
 GPUTPCGMMergerFinalize_step0,256,,,256,,,,,,256,,256,256,,256,,256,256,
 GPUTPCGMMergerFinalize_step1,256,,,256,,,,,,256,,256,256,,256,,256,256,
 GPUTPCGMMergerFinalize_step2,256,,,256,,,,,,256,,256,256,,256,,256,256,
+GPUTPCGMMergerHitWeights_prepare,256,,,,,,,,,,,,,,,,,,
+GPUTPCGMMergerHitWeights_compute,256,,,,,,,,,,,,,,,,,,
+GPUTPCGMMergerHitWeights_resolve1,256,,,,,,,,,,,,,,,,,,
+GPUTPCGMMergerHitWeights_resolve2,256,,,,,,,,,,,,,,,,,,
+GPUTPCGMMergerHitWeights_resolveShared,256,,,,,,,,,,,,,,,,,,
 GPUTPCGMMergerMergeLoopers_step0,256,,,,,,,,,256,,256,256,,256,,256,256,
 GPUTPCGMMergerMergeLoopers_step1,256,,,,,,,,,256,,256,256,,256,,256,256,
 GPUTPCGMMergerMergeLoopers_step2,256,,,,,,,,,256,,256,256,,256,,256,256,