ITS-GPU: various fixes + separate vertexing kernels (AliceO2Group#13473)

* Staging

* Move vertexer kernels away from the traits

* Fix GPU APIs compatibility

* Fix Hybrid interfaces for CPU

* Checkpoint work

* Unregister memory in sync and async when iteration == par.size()-1

* Fix USEMatCorrNONE case in gpu fit

* Update usage

* Make TrackingFrameInfo array const

* Update methods

Author: mconcas

Detectors/ITSMFT/ITS/tracking/GPU/ITStrackingGPU/TimeFrameGPU.h

@@ -174,9 +174,9 @@ class GpuTimeFrameChunk
 template <int nLayers = 7>
 class TimeFrameGPU : public TimeFrame
 {
- public:
   friend class GpuTimeFrameChunk<nLayers>;
 
+ public:
   TimeFrameGPU();
   ~TimeFrameGPU();
 
@@ -185,7 +185,7 @@ class TimeFrameGPU : public TimeFrame
   void unregisterHostMemory(const int);
   void initialise(const int, const TrackingParameters&, const int, IndexTableUtils* utils = nullptr, const TimeFrameGPUParameters* pars = nullptr);
   void initialiseHybrid(const int, const TrackingParameters&, const int, IndexTableUtils* utils = nullptr, const TimeFrameGPUParameters* pars = nullptr);
-  void initDevice(const int, IndexTableUtils*, const TrackingParameters& trkParam, const TimeFrameGPUParameters&, const int, const int);
+  void initDevice(IndexTableUtils*, const TrackingParameters& trkParam, const TimeFrameGPUParameters&, const int, const int);
   void initDeviceSAFitting();
   void loadTrackingFrameInfoDevice(const int);
   void loadUnsortedClustersDevice();
@@ -227,7 +227,8 @@ class TimeFrameGPU : public TimeFrame
   TrackITSExt* getDeviceTrackITSExt() { return mTrackITSExtDevice; }
   gpuPair<int, int>* getDeviceNeighbours(const int layer) { return mNeighboursDevice[layer]; }
   TrackingFrameInfo* getDeviceTrackingFrameInfo(const int);
-  TrackingFrameInfo** getDeviceArrayTrackingFrameInfo() { return mTrackingFrameInfoDeviceArray; }
+  // TrackingFrameInfo** getDeviceArrayTrackingFrameInfo() { return mTrackingFrameInfoDeviceArray; }
+  const TrackingFrameInfo** getDeviceArrayTrackingFrameInfo() const { return mTrackingFrameInfoDeviceArray; }
   Cluster** getDeviceArrayClusters() const { return mClustersDeviceArray; }
   Cluster** getDeviceArrayUnsortedClusters() const { return mUnsortedClustersDeviceArray; }
   Tracklet** getDeviceArrayTracklets() const { return mTrackletsDeviceArray; }
@@ -275,7 +276,7 @@ class TimeFrameGPU : public TimeFrame
   TrackITSExt* mTrackITSExtDevice;
   std::array<gpuPair<int, int>*, nLayers - 2> mNeighboursDevice;
   std::array<TrackingFrameInfo*, nLayers> mTrackingFrameInfoDevice;
-  TrackingFrameInfo** mTrackingFrameInfoDeviceArray;
+  const TrackingFrameInfo** mTrackingFrameInfoDeviceArray;
 
   // State
   std::vector<Stream> mGpuStreams;

Detectors/ITSMFT/ITS/tracking/GPU/ITStrackingGPU/TrackingKernels.h

@@ -16,9 +16,7 @@
 #include "DetectorsBase/Propagator.h"
 #include "GPUCommonDef.h"
 
-namespace o2
-{
-namespace its
+namespace o2::its
 {
 class CellSeed;
 namespace gpu
@@ -40,9 +38,9 @@ GPUd() bool fitTrack(TrackITSExt& track,
 template <int nLayers = 7>
 GPUg() void fitTrackSeedsKernel(
   CellSeed* trackSeeds,
-  TrackingFrameInfo** foundTrackingFrameInfo,
+  const TrackingFrameInfo** foundTrackingFrameInfo,
   o2::its::TrackITSExt* tracks,
-  const size_t nSeeds,
+  const unsigned int nSeeds,
   const float Bz,
   const int startLevel,
   float maxChi2ClusterAttachment,
@@ -53,15 +51,14 @@ GPUg() void fitTrackSeedsKernel(
 } // namespace gpu
 
 void trackSeedHandler(CellSeed* trackSeeds,
-                      TrackingFrameInfo** foundTrackingFrameInfo,
+                      const TrackingFrameInfo** foundTrackingFrameInfo,
                       o2::its::TrackITSExt* tracks,
-                      const size_t nSeeds,
+                      const unsigned int nSeeds,
                       const float Bz,
                       const int startLevel,
                       float maxChi2ClusterAttachment,
                       float maxChi2NDF,
                       const o2::base::Propagator* propagator,
                       const o2::base::PropagatorF::MatCorrType matCorrType);
-} // namespace its
-} // namespace o2
+} // namespace o2::its
 #endif // ITSTRACKINGGPU_TRACKINGKERNELS_H_

Detectors/ITSMFT/ITS/tracking/GPU/ITStrackingGPU/VertexingKernels.h

@@ -0,0 +1,59 @@
+// Copyright 2019-2020 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.
+///
+
+#ifndef ITSTRACKINGGPU_VERTEXINGKERNELS_H_
+#define ITSTRACKINGGPU_VERTEXINGKERNELS_H_
+#include "ITStracking/MathUtils.h"
+#include "ITStracking/Configuration.h"
+#include "ITStracking/ClusterLines.h"
+#include "ITStracking/Tracklet.h"
+
+#include "ITStrackingGPU/Utils.h"
+#include "ITStrackingGPU/ClusterLinesGPU.h"
+#include "ITStrackingGPU/VertexerTraitsGPU.h"
+#include "ITStrackingGPU/TracerGPU.h"
+
+#include "GPUCommonArray.h"
+
+namespace o2::its::gpu
+{
+#ifdef GPUCA_GPUCODE // GPUg() global kernels must only when compiled by GPU compiler
+template <TrackletMode Mode>
+GPUg() void trackleterKernelMultipleRof(
+  const Cluster* clustersNextLayer,    // 0 2
+  const Cluster* clustersCurrentLayer, // 1 1
+  const int* sizeNextLClusters,
+  const int* sizeCurrentLClusters,
+  const int* nextIndexTables,
+  Tracklet* Tracklets,
+  int* foundTracklets,
+  const IndexTableUtils* utils,
+  const unsigned int startRofId,
+  const unsigned int rofSize,
+  const float phiCut,
+  const size_t maxTrackletsPerCluster);
+#endif
+template <TrackletMode Mode>
+void trackletFinderHandler(const Cluster* clustersNextLayer,    // 0 2
+                           const Cluster* clustersCurrentLayer, // 1 1
+                           const int* sizeNextLClusters,
+                           const int* sizeCurrentLClusters,
+                           const int* nextIndexTables,
+                           Tracklet* Tracklets,
+                           int* foundTracklets,
+                           const IndexTableUtils* utils,
+                           const unsigned int startRofId,
+                           const unsigned int rofSize,
+                           const float phiCut,
+                           const size_t maxTrackletsPerCluster = 1e2);
+} // namespace o2::its::gpu
+#endif

Detectors/ITSMFT/ITS/tracking/GPU/cuda/CMakeLists.txt

@@ -22,7 +22,8 @@ o2_add_library(ITStrackingCUDA
                        TimeFrameGPU.cu
                        TracerGPU.cu
                        TrackingKernels.cu
-                       VertexerTraitsGPU.cu
+                       VertexingKernels.cu
+                       VertexerTraitsGPU.cxx
                        Utils.cu
                PUBLIC_INCLUDE_DIRECTORIES ../
                PUBLIC_LINK_LIBRARIES O2::ITStracking