ITS GPU: Make threads and blocks configurable from CLI (#13596)

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

@@ -184,7 +184,6 @@ class TimeFrameGPU : public TimeFrame
   void registerHostMemory(const int);
   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(IndexTableUtils*, const TrackingParameters& trkParam, const TimeFrameGPUParameters&, const int, const int);
   void initDeviceSAFitting();
   void loadTrackingFrameInfoDevice(const int);

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

@@ -59,6 +59,8 @@ void trackSeedHandler(CellSeed* trackSeeds,
                       float maxChi2ClusterAttachment,
                       float maxChi2NDF,
                       const o2::base::Propagator* propagator,
-                      const o2::base::PropagatorF::MatCorrType matCorrType);
+                      const o2::base::PropagatorF::MatCorrType matCorrType,
+                      const int nBlocks,
+                      const int nThreads);
 } // namespace o2::its
 #endif // ITSTRACKINGGPU_TRACKINGKERNELS_H_

Detectors/ITSMFT/ITS/tracking/GPU/cuda/TimeFrameGPU.cu

@@ -362,27 +362,6 @@ void TimeFrameGPU<nLayers>::initialise(const int iteration,
                                        const int maxLayers,
                                        IndexTableUtils* utils,
                                        const TimeFrameGPUParameters* gpuParam)
-{
-  mGpuStreams.resize(mGpuParams.nTimeFrameChunks);
-  mHostNTracklets.resize((nLayers - 1) * mGpuParams.nTimeFrameChunks, 0);
-  mHostNCells.resize((nLayers - 2) * mGpuParams.nTimeFrameChunks, 0);
-
-  auto init = [&]() -> void {
-    this->initDevice(utils, trkParam, *gpuParam, maxLayers, iteration);
-  };
-  std::thread t1{init};
-  RANGE("tf_cpu_initialisation", 1);
-  o2::its::TimeFrame::initialise(iteration, trkParam, maxLayers);
-  // registerHostMemory(maxLayers);
-  t1.join();
-}
-
-template <int nLayers>
-void TimeFrameGPU<nLayers>::initialiseHybrid(const int iteration,
-                                             const TrackingParameters& trkParam,
-                                             const int maxLayers,
-                                             IndexTableUtils* utils,
-                                             const TimeFrameGPUParameters* gpuParam)
 {
   mGpuStreams.resize(mGpuParams.nTimeFrameChunks);
   o2::its::TimeFrame::initialise(iteration, trkParam, maxLayers);

Detectors/ITSMFT/ITS/tracking/GPU/cuda/TrackerTraitsGPU.cxx

@@ -20,6 +20,7 @@
 
 #include "ITStrackingGPU/TrackerTraitsGPU.h"
 #include "ITStrackingGPU/TrackingKernels.h"
+#include "ITStracking/TrackingConfigParam.h"
 
 namespace o2::its
 {
@@ -28,7 +29,7 @@ constexpr int UnusedIndex{-1};
 template <int nLayers>
 void TrackerTraitsGPU<nLayers>::initialiseTimeFrame(const int iteration)
 {
-  mTimeFrameGPU->initialiseHybrid(iteration, mTrkParams[iteration], nLayers);
+  mTimeFrameGPU->initialise(iteration, mTrkParams[iteration], nLayers);
   mTimeFrameGPU->loadTrackingFrameInfoDevice(iteration);
 }
 
@@ -397,7 +398,7 @@ void TrackerTraitsGPU<nLayers>::findRoads(const int iteration)
     }
     mTimeFrameGPU->createTrackITSExtDevice(trackSeeds);
     mTimeFrameGPU->loadTrackSeedsDevice(trackSeeds);
-
+    auto& conf = o2::its::ITSGpuTrackingParamConfig::Instance();
     trackSeedHandler(
       mTimeFrameGPU->getDeviceTrackSeeds(),             // CellSeed* trackSeeds,
       mTimeFrameGPU->getDeviceArrayTrackingFrameInfo(), // TrackingFrameInfo** foundTrackingFrameInfo,
@@ -408,7 +409,9 @@ void TrackerTraitsGPU<nLayers>::findRoads(const int iteration)
       mTrkParams[0].MaxChi2ClusterAttachment,           // float maxChi2ClusterAttachment,
       mTrkParams[0].MaxChi2NDF,                         // float maxChi2NDF,
       mTimeFrameGPU->getDevicePropagator(),             // const o2::base::Propagator* propagator
-      mCorrType);                                       // o2::base::PropagatorImpl<float>::MatCorrType
+      mCorrType,                                        // o2::base::PropagatorImpl<float>::MatCorrType
+      conf.nBlocks,
+      conf.nThreads);
 
     mTimeFrameGPU->downloadTrackITSExtDevice(trackSeeds);
 

Detectors/ITSMFT/ITS/tracking/GPU/cuda/TrackingKernels.cu

@@ -717,9 +717,11 @@ void trackSeedHandler(CellSeed* trackSeeds,
                       float maxChi2ClusterAttachment,
                       float maxChi2NDF,
                       const o2::base::Propagator* propagator,
-                      const o2::base::PropagatorF::MatCorrType matCorrType)
+                      const o2::base::PropagatorF::MatCorrType matCorrType,
+                      const int nBlocks,
+                      const int nThreads)
 {
-  gpu::fitTrackSeedsKernel<<<20, 256>>>(
+  gpu::fitTrackSeedsKernel<<<nBlocks, nThreads>>>(
     trackSeeds,               // CellSeed* trackSeeds,
     foundTrackingFrameInfo,   // TrackingFrameInfo** foundTrackingFrameInfo,
     tracks,                   // o2::its::TrackITSExt* tracks,
@@ -734,4 +736,4 @@ void trackSeedHandler(CellSeed* trackSeeds,
   gpuCheckError(cudaPeekAtLastError());
   gpuCheckError(cudaDeviceSynchronize());
 }
-} // namespace o2::its
+} // namespace o2::its

Detectors/ITSMFT/ITS/tracking/GPU/cuda/VertexerTraitsGPU.cxx

@@ -37,6 +37,7 @@ void VertexerTraitsGPU::initialise(const TrackingParameters& trackingParams, con
 {
   mTimeFrameGPU->initialise(0, trackingParams, 3, &mIndexTableUtils, &mTfGPUParams);
 }
+
 void VertexerTraitsGPU::updateVertexingParameters(const std::vector<VertexingParameters>& vrtPar, const TimeFrameGPUParameters& tfPar)
 {
   mVrtParams = vrtPar;

Detectors/ITSMFT/ITS/tracking/GPU/cuda/VertexingKernels.cu

@@ -39,8 +39,23 @@ void trackletFinderHandler(const Cluster* clustersNextLayer,    // 0 2
                            const unsigned int startRofId,
                            const unsigned int rofSize,
                            const float phiCut,
-                           const size_t maxTrackletsPerCluster)
+                           const unsigned int maxTrackletsPerCluster,
+                           const int nBlocks,
+                           const int nThreads)
 {
+  gpu::trackleterKernelMultipleRof<Mode><<<nBlocks, nThreads>>>(
+    clustersNextLayer,       // const Cluster* clustersNextLayer,    // 0 2
+    clustersCurrentLayer,    // const Cluster* clustersCurrentLayer, // 1 1
+    sizeNextLClusters,       // const int* sizeNextLClusters,
+    sizeCurrentLClusters,    // const int* sizeCurrentLClusters,
+    nextIndexTables,         // const int* nextIndexTables,
+    Tracklets,               // Tracklet* Tracklets,
+    foundTracklets,          // int* foundTracklets,
+    utils,                   // const IndexTableUtils* utils,
+    startRofId,              // const unsigned int startRofId,
+    rofSize,                 // const unsigned int rofSize,
+    phiCut,                  // const float phiCut,
+    maxTrackletsPerCluster); // const unsigned int maxTrackletsPerCluster = 1e2
 }
 /*
 GPUd() float smallestAngleDifference(float a, float b)
@@ -96,7 +111,7 @@ GPUd() void printOnBlock(const unsigned int bId, const char* str, Args... args)
   }
 }
 
-GPUg() void printBufferOnThread(const int* v, size_t size, const int len = 150, const unsigned int tId = 0)
+GPUg() void printBufferOnThread(const int* v, unsigned int size, const int len = 150, const unsigned int tId = 0)
 {
   if (blockIdx.x * blockDim.x + threadIdx.x == tId) {
     for (int i{0}; i < size; ++i) {
@@ -109,7 +124,7 @@ GPUg() void printBufferOnThread(const int* v, size_t size, const int len = 150,
   }
 }
 
-GPUg() void printBufferOnThreadF(const float* v, size_t size, const unsigned int tId = 0)
+GPUg() void printBufferOnThreadF(const float* v, unsigned int size, const unsigned int tId = 0)
 {
   if (blockIdx.x * blockDim.x + threadIdx.x == tId) {
     printf("vector :");
@@ -127,7 +142,7 @@ GPUg() void resetTrackletsKernel(Tracklet* tracklets, const int nTracklets)
   }
 }
 
-GPUg() void dumpFoundTrackletsKernel(const Tracklet* tracklets, const int* nTracklet, const size_t nClustersMiddleLayer, const int maxTrackletsPerCluster)
+GPUg() void dumpFoundTrackletsKernel(const Tracklet* tracklets, const int* nTracklet, const unsigned int nClustersMiddleLayer, const int maxTrackletsPerCluster)
 {
   for (int iCurrentLayerClusterIndex = blockIdx.x * blockDim.x + threadIdx.x; iCurrentLayerClusterIndex < nClustersMiddleLayer; iCurrentLayerClusterIndex += blockDim.x * gridDim.x) {
     const int stride{iCurrentLayerClusterIndex * maxTrackletsPerCluster};
@@ -160,15 +175,15 @@ GPUg() void trackleterKernelSingleRof(
   int* foundTracklets,
   const IndexTableUtils* utils,
   const short rofId,
-  const size_t maxTrackletsPerCluster = 1e2)
+  const unsigned int maxTrackletsPerCluster = 1e2)
 {
   const int phiBins{utils->getNphiBins()};
   const int zBins{utils->getNzBins()};
   // loop on layer1 clusters
   for (int iCurrentLayerClusterIndex = blockIdx.x * blockDim.x + threadIdx.x; iCurrentLayerClusterIndex < sizeCurrentLClusters; iCurrentLayerClusterIndex += blockDim.x * gridDim.x) {
     if (iCurrentLayerClusterIndex < sizeCurrentLClusters) {
       unsigned int storedTracklets{0};
-      const size_t stride{iCurrentLayerClusterIndex * maxTrackletsPerCluster};
+      const unsigned int stride{iCurrentLayerClusterIndex * maxTrackletsPerCluster};
       const Cluster& currentCluster = clustersCurrentLayer[iCurrentLayerClusterIndex];
       const int4 selectedBinsRect{VertexerTraits::getBinsRect(currentCluster, (int)Mode, 0.f, 50.f, phiCut / 2, *utils)};
       if (selectedBinsRect.x != 0 || selectedBinsRect.y != 0 || selectedBinsRect.z != 0 || selectedBinsRect.w != 0) {
@@ -218,7 +233,7 @@ GPUg() void trackleterKernelMultipleRof(
   const short startRofId,
   const short rofSize,
   const float phiCut,
-  const size_t maxTrackletsPerCluster = 1e2)
+  const unsigned int maxTrackletsPerCluster = 1e2)
 {
   const int phiBins{utils->getNphiBins()};
   const int zBins{utils->getNzBins()};
@@ -235,7 +250,7 @@ GPUg() void trackleterKernelMultipleRof(
     // single rof loop on layer1 clusters
     for (int iCurrentLayerClusterIndex = threadIdx.x; iCurrentLayerClusterIndex < nClustersCurrentLayerRof; iCurrentLayerClusterIndex += blockDim.x) {
       unsigned int storedTracklets{0};
-      const size_t stride{iCurrentLayerClusterIndex * maxTrackletsPerCluster};
+      const unsigned int stride{iCurrentLayerClusterIndex * maxTrackletsPerCluster};
       const Cluster& currentCluster = clustersCurrentLayerRof[iCurrentLayerClusterIndex];
       const int4 selectedBinsRect{VertexerTraits::getBinsRect(currentCluster, (int)Mode, 0.f, 50.f, phiCut / 2, *utils)};
       if (selectedBinsRect.x != 0 || selectedBinsRect.y != 0 || selectedBinsRect.z != 0 || selectedBinsRect.w != 0) {
@@ -276,7 +291,7 @@ template <bool initRun>
 GPUg() void trackletSelectionKernelSingleRof(
   const Cluster* clusters0,
   const Cluster* clusters1,
-  const size_t nClustersMiddleLayer,
+  const unsigned int nClustersMiddleLayer,
   Tracklet* tracklets01,
   Tracklet* tracklets12,
   const int* nFoundTracklet01,
@@ -436,7 +451,7 @@ GPUg() void computeCentroidsKernel(
   Line* lines,
   int* nFoundLines,
   int* nExclusiveFoundLines,
-  const size_t nClustersMiddleLayer,
+  const unsigned int nClustersMiddleLayer,
   float* centroids,
   const float lowHistX,
   const float highHistX,
@@ -446,7 +461,7 @@ GPUg() void computeCentroidsKernel(
 {
   const int nLines = nExclusiveFoundLines[nClustersMiddleLayer - 1] + nFoundLines[nClustersMiddleLayer - 1];
   const int maxIterations{nLines * (nLines - 1) / 2};
-  for (size_t currentThreadIndex = blockIdx.x * blockDim.x + threadIdx.x; currentThreadIndex < maxIterations; currentThreadIndex += blockDim.x * gridDim.x) {
+  for (unsigned int currentThreadIndex = blockIdx.x * blockDim.x + threadIdx.x; currentThreadIndex < maxIterations; currentThreadIndex += blockDim.x * gridDim.x) {
     int iFirstLine = currentThreadIndex / nLines;
     int iSecondLine = currentThreadIndex % nLines;
     // All unique pairs
@@ -496,7 +511,7 @@ GPUg() void computeZCentroidsKernel(
   const int binOpeningX,
   const int binOpeningY)
 {
-  for (size_t currentThreadIndex = blockIdx.x * blockDim.x + threadIdx.x; currentThreadIndex < nLines; currentThreadIndex += blockDim.x * gridDim.x) {
+  for (unsigned int currentThreadIndex = blockIdx.x * blockDim.x + threadIdx.x; currentThreadIndex < nLines; currentThreadIndex += blockDim.x * gridDim.x) {
     if (tmpVtX[0].value || tmpVtX[1].value) {
       float tmpX{lowHistX + tmpVtX[0].key * binSizeHistX + binSizeHistX / 2};
       int sumWX{tmpVtX[0].value};
@@ -543,7 +558,7 @@ GPUg() void computeVertexKernel(
   const int minContributors,
   const int binOpeningZ)
 {
-  for (size_t currentThreadIndex = blockIdx.x * blockDim.x + threadIdx.x; currentThreadIndex < binOpeningZ; currentThreadIndex += blockDim.x * gridDim.x) {
+  for (unsigned int currentThreadIndex = blockIdx.x * blockDim.x + threadIdx.x; currentThreadIndex < binOpeningZ; currentThreadIndex += blockDim.x * gridDim.x) {
     if (currentThreadIndex == 0) {
       if (tmpVertexBins[2].value > 1 && (tmpVertexBins[0].value || tmpVertexBins[1].value)) {
         float z{lowHistZ + tmpVertexBins[2].key * binSizeHistZ + binSizeHistZ / 2};

Detectors/ITSMFT/ITS/tracking/include/ITStracking/TrackingConfigParam.h

@@ -55,7 +55,6 @@ struct VertexerParamConfig : public o2::conf::ConfigurableParamHelper<VertexerPa
 };
 
 struct TrackerParamConfig : public o2::conf::ConfigurableParamHelper<TrackerParamConfig> {
-
   // Use TGeo for mat. budget
   bool useMatCorrTGeo = false;
   bool useFastMaterial = false;
@@ -89,24 +88,13 @@ struct TrackerParamConfig : public o2::conf::ConfigurableParamHelper<TrackerPara
   O2ParamDef(TrackerParamConfig, "ITSCATrackerParam");
 };
 
-struct GpuRecoParamConfig : public o2::conf::ConfigurableParamHelper<GpuRecoParamConfig> {
+struct ITSGpuTrackingParamConfig : public o2::conf::ConfigurableParamHelper<ITSGpuTrackingParamConfig> {
   // GPU-specific parameters
-  size_t tmpCUBBufferSize = 1e5; // In average in pp events there are required 4096 bytes
-  size_t maxTrackletsPerCluster = 1e2;
-  size_t clustersPerLayerCapacity = 2.5e5;
-  size_t clustersPerROfCapacity = 1.5e3;
-  // size_t trackletsCapacity = maxTrackletsPerCluster * clustersPerLayerCapacity;
-  size_t validatedTrackletsCapacity = 1e5;
-  size_t cellsLUTsize = validatedTrackletsCapacity;
-  size_t maxNeighboursSize = 1e4;
-  size_t neighboursLUTsize = maxNeighboursSize;
-  size_t maxRoadPerRofSize = 5e2; // pp!
-  size_t maxLinesCapacity = 1e2;
-  size_t maxVerticesCapacity = 5e4;
-  size_t nTimeFramePartitions = 3;
-  int maxGPUMemoryGB = -1;
+  unsigned int tmpCUBBufferSize = 1e5; // In average in pp events there are required 4096 bytes
+  int nBlocks = 20;
+  int nThreads = 256;
 
-  O2ParamDef(GpuRecoParamConfig, "ITSGpuRecoParam");
+  O2ParamDef(ITSGpuTrackingParamConfig, "ITSGpuTrackingParam");
 };
 
 } // namespace its

Detectors/ITSMFT/ITS/tracking/src/TrackingConfigParam.cxx

@@ -18,10 +18,10 @@ namespace its
 {
 static auto& sVertexerParamITS = o2::its::VertexerParamConfig::Instance();
 static auto& sCATrackerParamITS = o2::its::TrackerParamConfig::Instance();
-static auto& sGpuRecoParamITS = o2::its::GpuRecoParamConfig::Instance();
+static auto& sGpuRecoParamITS = o2::its::ITSGpuTrackingParamConfig::Instance();
 
 O2ParamImpl(o2::its::VertexerParamConfig);
 O2ParamImpl(o2::its::TrackerParamConfig);
-O2ParamImpl(o2::its::GpuRecoParamConfig);
+O2ParamImpl(o2::its::ITSGpuTrackingParamConfig);
 } // namespace its
 } // namespace o2

Detectors/ITSMFT/ITS/tracking/src/TrackingLinkDef.h

@@ -30,7 +30,7 @@
 #pragma link C++ class o2::its::TrackerParamConfig + ;
 #pragma link C++ class o2::conf::ConfigurableParamHelper < o2::its::TrackerParamConfig> + ;
 
-#pragma link C++ class o2::its::GpuRecoParamConfig + ;
-#pragma link C++ class o2::conf::ConfigurableParamHelper < o2::its::GpuRecoParamConfig> + ;
+#pragma link C++ class o2::its::ITSGpuTrackingParamConfig + ;
+#pragma link C++ class o2::conf::ConfigurableParamHelper < o2::its::ITSGpuTrackingParamConfig> + ;
 
 #endif

Detectors/ITSMFT/ITS/tracking/src/Vertexer.cxx

@@ -104,7 +104,7 @@ void Vertexer::getGlobalConfiguration()
 {
   auto& vc = o2::its::VertexerParamConfig::Instance();
   vc.printKeyValues(true, true);
-  auto& grc = o2::its::GpuRecoParamConfig::Instance();
+  auto& grc = o2::its::ITSGpuTrackingParamConfig::Instance();
 
   // This is odd: we override only the parameters for the first iteration.
   // Variations for the next iterations are set in the trackingInterfrace.