GPU/TPC: extend noisy pad filter with Highly Ionising Particle filter (#15424)

* GPU/TPC: extend noisy pad filter with Highly Ionising Particle filter

* Fix race condition when filter is disabled

* Add fallback for deterministic mode

* format

* OpenCL: Enable 64bit integer atomics

* Revert "OpenCL: Enable 64bit integer atomics"

This reverts commit a53a04b.

* Avoid writing to nCluster in HIPClusterizer kernel

* Fix MC labels  with HIP clusters

* Fix integer cast warning

Author: fweig

GPU/Common/GPUCommonAlgorithm.h

@@ -354,6 +354,11 @@ GPUdi() uint8_t warp_broadcast_FUNC<uint8_t>(uint8_t v, int32_t i)
 #define warp_scan_inclusive_add(v) warp_scan_inclusive_add_FUNC(v)
 #define warp_broadcast(v, i) warp_broadcast_FUNC(v, i)
 
+[[nodiscard]] GPUdi() int32_t work_group_count(bool pred)
+{
+  return work_group_reduce_add((int32_t)pred);
+}
+
 #elif (defined(__CUDACC__) || defined(__HIPCC__))
 // CUDA and HIP work the same way using cub, need just different header
 
@@ -416,6 +421,16 @@ GPUdi() T warp_broadcast_FUNC(T v, int32_t i)
 #endif
 }
 
+[[nodiscard]] GPUdi() bool work_group_any(bool pred)
+{
+  return __syncthreads_or(pred);
+}
+
+[[nodiscard]] GPUdi() uint32_t work_group_count(bool pred)
+{
+  return __syncthreads_count(pred);
+}
+
 #else
 // Trivial implementation for the CPU
 
@@ -449,6 +464,16 @@ GPUdi() T warp_broadcast(T v, int32_t i)
   return v;
 }
 
+[[nodiscard]] GPUdi() bool work_group_any(bool pred)
+{
+  return pred;
+}
+
+[[nodiscard]] GPUdi() uint32_t work_group_count(bool pred)
+{
+  return pred;
+}
+
 #endif
 
 #ifdef GPUCA_ALGORITHM_STD

GPU/GPUTracking/DataTypes/GPUTPCExtraADC.h

@@ -0,0 +1,25 @@
+// Copyright 2019-2026 CERN and copyright holders of ALICE O2.
+// See https://alice-o2.web.cern.ch/copyright for details of the copyright holders.
+// All rights not expressly granted are reserved.
+//
+// This software is distributed under the terms of the GNU General Public
+// License v3 (GPL Version 3), copied verbatim in the file "COPYING".
+//
+// In applying this license CERN does not waive the privileges and immunities
+// granted to it by virtue of its status as an Intergovernmental Organization
+// or submit itself to any jurisdiction.
+
+/// \file GPUTPCExtraADC.h
+/// \author Felix Weiglhofer
+
+#include "GPUDefConstantsAndSettings.h"
+#include "DataFormatsTPC/Digit.h"
+#include <array>
+#include <vector>
+
+namespace o2::gpu
+{
+struct GPUTPCExtraADC {
+  std::array<std::vector<tpc::Digit>, tpc::constants::MAXSECTOR> digitsBySector;
+};
+} // namespace o2::gpu

GPU/GPUTracking/Definitions/GPUSettingsList.h

@@ -112,6 +112,9 @@ AddOptionRTC(trackletMinSharedNormFactor, float, 0.f, "", 0, "Max shared defined
 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")
+AddOptionRTC(hipTailFilter, uint8_t, 0, "", 0, "Enable Highly Ionising Particle tail filter in CheckPadBaseline (0 = disable, 1 = filter tails)")
+AddOptionRTC(hipTailFilterThreshold, uint16_t, 100, "", 0, "Threshold that must be exceeded for a timebin to be counted towards Highly Ionising Particle tail")
+AddOptionRTC(hipTailFilterAlpha, float, 0.5f, "", 0, "Smoothing factor for the exponential Highly Ionising Particle tail filter")
 AddOptionRTC(occupancyMapTimeBins, uint16_t, 16, "", 0, "Number of timebins per histogram bin of occupancy map (0 = disable occupancy map)")
 AddOptionRTC(occupancyMapTimeBinsAverage, uint16_t, 0, "", 0, "Number of timebins +/- to use for the averaging")
 AddOptionRTC(trackFitCovLimit, uint16_t, 1000, "", 0, "Abort fit when y/z cov exceed the limit")

GPU/GPUTracking/Definitions/Parameters/GPUParameters.csv

@@ -60,6 +60,8 @@ GPUTPCGMO2Output_output,256,,,,,,,,,,,,,,,256
 GPUTPCStartHitsFinder,256,,"[1024, 2]","[1024, 7]",256,256,256,256,256,512,512,512,,,,608
 GPUTPCStartHitsSorter,256,,"[1024, 5]","[512, 7]",256,256,256,256,256,"[512, 1]","[512, 1]","[512, 1]",,,,608
 GPUTPCCFCheckPadBaseline,576,,"[576, 2]","[576, 2]",,,,,,"[576, 2]",,,,,,"[576, 2]"
+GPUTPCCFHIPTailConnector,256,,256,256,,,,,,256,
+GPUTPCCFHIPClusterizer,256,,256,256,,,,,,256,
 GPUTPCCFChargeMapFiller_fillIndexMap,512,,512,512,,,,,,448,,,,,,448
 GPUTPCCFChargeMapFiller_fillFromDigits,512,,512,512,,,,,,448,,,,,,448
 GPUTPCCFChargeMapFiller_findFragmentStart,512,,512,512,,,,,,448,,,,,,448

GPU/GPUTracking/Global/GPUChainTracking.h

@@ -71,6 +71,7 @@ struct CfFragment;
 class GPUTPCClusterFinder;
 struct GPUSettingsProcessing;
 struct GPUSettingsRec;
+struct GPUTPCExtraADC;
 
 class GPUChainTracking : public GPUChain
 {
@@ -298,13 +299,15 @@ class GPUChainTracking : public GPUChain
   int32_t RunChainFinalize();
   void OutputSanityCheck();
   int32_t RunTPCTrackingSectors_internal();
-  int32_t RunTPCClusterizer_prepare(bool restorePointers);
+  int32_t RunTPCClusterizer_prepare(bool restorePointers, const GPUTPCExtraADC& extraADCs);
 #ifndef GPUCA_RUN2
-  std::pair<uint32_t, uint32_t> RunTPCClusterizer_transferZS(int32_t iSector, const CfFragment& fragment, int32_t lane);
+  std::pair<uint32_t, uint32_t> RunTPCClusterizer_transferZS(int32_t iSector, const CfFragment& fragment, int32_t lane, const GPUTPCExtraADC& extraADCs);
   void RunTPCClusterizer_compactPeaks(GPUTPCClusterFinder& clusterer, GPUTPCClusterFinder& clustererShadow, int32_t stage, bool doGPU, int32_t lane);
   std::pair<uint32_t, uint32_t> TPCClusterizerDecodeZSCount(uint32_t iSector, const CfFragment& fragment);
   std::pair<uint32_t, uint32_t> TPCClusterizerDecodeZSCountUpdate(uint32_t iSector, const CfFragment& fragment);
   void TPCClusterizerEnsureZSOffsets(uint32_t iSector, const CfFragment& fragment);
+  void TPCClusterizerTransferExtraADC(GPUTPCClusterFinder& clusterer, GPUTPCClusterFinder& clustererShadow, int lane, const GPUTPCExtraADC& extraADCs);
+  void TPCClusterizerCheckExtraADCZeros(GPUTPCClusterFinder& clusterer, GPUTPCClusterFinder& clustererShadow, int lane, const GPUTPCExtraADC& extraADCs);
 #endif
   void RunTPCTrackingMerger_MergeBorderTracks(uint8_t mergeMode, GPUReconstruction::krnlDeviceType deviceType);
   void RunTPCTrackingMerger_Resolve(int8_t useOrigTrackParam, int8_t mergeAll, GPUReconstruction::krnlDeviceType deviceType);