Merge branch 'AliceO2Group:dev' into new-detector4

Author: pkurash

CCDB/include/CCDB/BasicCCDBManager.h

@@ -330,7 +330,7 @@ T* CCDBManagerInstance::getForRun(std::string const& path, int runNumber, bool s
 template <typename T>
 T* CCDBManagerInstance::getSpecificForRun(std::string const& path, int runNumber, MD metaData)
 {
-  auto [start, stop] = getRunDuration(runNumber);
+  auto [start, stop] = getRunDuration(runNumber, mFatalWhenNull);
   if (start < 0 || stop < 0) {
     if (mFatalWhenNull) {
       reportFatal(std::string("Failed to get run duration for run ") + std::to_string(runNumber) + std::string(" from CCDB"));

CMakeLists.txt

@@ -41,8 +41,6 @@ o2_build_sanity_checks()
 set(CMAKE_CXX_STANDARD 20)
 set(CMAKE_CXX_STANDARD_REQUIRED TRUE)
 
-include(dependencies/FindONNXRuntime.cmake)
-
 include(O2CheckCXXFeatures)
 o2_check_cxx_features()
 

Common/Constants/include/CommonConstants/PhysicsConstants.h

@@ -34,6 +34,7 @@ enum Pdg {
   kB0 = 511,
   kB0Bar = -511,
   kBPlus = 521,
+  kBCPlus = 541,
   kBS = 531,
   kBSBar = -531,
   kD0 = 421,
@@ -63,6 +64,8 @@ enum Pdg {
   kPhi = 333,
   kSigmaC0 = 4112,
   kSigmaCPlusPlus = 4222,
+  kSigmaCStar0 = 4114,
+  kSigmaCStarPlusPlus = 4224,
   kX3872 = 9920443,
   kXi0 = 3322,
   kXiB0 = 5232,
@@ -84,6 +87,7 @@ enum Pdg {
 constexpr double MassB0 = 5.27966;
 constexpr double MassB0Bar = 5.27966;
 constexpr double MassBPlus = 5.27934;
+constexpr double MassBCPlus = 6.27447;
 constexpr double MassBS = 5.36692;
 constexpr double MassBSBar = 5.36692;
 constexpr double MassD0 = 1.86484;
@@ -113,6 +117,8 @@ constexpr double MassKPlusStar892 = 0.89167;
 constexpr double MassPhi = 1.019461;
 constexpr double MassSigmaC0 = 2.45375;
 constexpr double MassSigmaCPlusPlus = 2.45397;
+constexpr double MassSigmaCStar0 = 2.51848;
+constexpr double MassSigmaCStarPlusPlus = 2.51841;
 constexpr double MassX3872 = 3.87165;
 constexpr double MassXi0 = 1.31486;
 constexpr double MassXiB0 = 5.7919;

Common/Constants/include/CommonConstants/make_pdg_header.py

@@ -89,6 +89,7 @@ class Pdg(Enum):
     kB0 = 511
     kB0Bar = -511
     kBPlus = 521
+    kBCPlus = 541
     kBS = 531
     kBSBar = -531
     kD0 = 421
@@ -118,6 +119,8 @@ class Pdg(Enum):
     kPhi = 333
     kSigmaC0 = 4112
     kSigmaCPlusPlus = 4222
+    kSigmaCStar0 = 4114
+    kSigmaCStarPlusPlus = 4224
     kX3872 = 9920443
     kXi0 = 3322
     kXiB0 = 5232

Common/DCAFitter/README.md

@@ -93,3 +93,41 @@ In this case the relevant correlation coefficient of the cov.matrix is redefined
 
 `DCAFitterN::setBadCovPolicy(DCAFitterN::OverrideAnFlag);` continue fit with overridden cov.matrix but set the propagation failure flag (can be checked using the same `isPropagationFailure(int cand = 0)` method).
 
+## Fit status
+The fitter provides a fit status for each candidate, which can be retrieved using:
+```
+FitStatus status = ft.getFitStatus(int cand = 0);
+```
+The possible values are:
+```
+enum FitStatus : uint8_t {    // part of the DCAFitterN class
+     None,                    // no status set (should not be possible!)
+
+     /* Good Conditions */
+     Converged, // fit converged
+     MaxIter,   // max iterations reached before fit convergence (can still be a good vertex)
+
+     /* Error Conditions */
+     NoCrossing,      // no reasonable crossing was found
+     RejRadius,       // radius of crossing was not acceptable
+     RejTrackX,       // one candidate track x was below the minimum required radius
+     RejTrackRoughZ,  // rejected by rough cut on tracks Z difference
+     RejChi2Max,      // rejected by maximum chi2 cut
+     FailProp,        // propagation of at least prong to PCA failed
+     FailInvCov,      // inversion of cov.-matrix failed
+     FailInvWeight,   // inversion of Ti weight matrix failed
+     FailInv2ndDeriv, // inversion of 2nd derivatives failed
+     FailCorrTracks,  // correction of tracks to updated x failed
+     FailCloserAlt,   // alternative PCA is closer
+};
+```
+This is allows to track where candiate fit was abondended.
+```
+int nc = ft.process(tr0,tr1,tr2);
+auto status = ft.getFitStatus();
+if (nc) {
+     // status can either be FitStatus::Converged or FitStatus::MaxIter
+}
+// status can be on of the error conditions
+```
+A more thorough example is given in `testDCAFitterN.cxx`.

Common/ML/CMakeLists.txt

@@ -9,21 +9,14 @@
 # granted to it by virtue of its status as an Intergovernmental Organization
 # or submit itself to any jurisdiction.
 
-# Pass ORT variables as a preprocessor definition
-if(ORT_ROCM_BUILD)
-    add_compile_definitions(ORT_ROCM_BUILD=1)
-endif()
-if(ORT_CUDA_BUILD)
-    add_compile_definitions(ORT_CUDA_BUILD=1)
-endif()
-if(ORT_MIGRAPHX_BUILD)
-    add_compile_definitions(ORT_MIGRAPHX_BUILD=1)
-endif()
-if(ORT_TENSORRT_BUILD)
-    add_compile_definitions(ORT_TENSORRT_BUILD=1)
-endif()
-
 o2_add_library(ML
                SOURCES src/OrtInterface.cxx
                TARGETVARNAME targetName
-               PRIVATE_LINK_LIBRARIES O2::Framework ONNXRuntime::ONNXRuntime)
+               PRIVATE_LINK_LIBRARIES O2::GPUCommon onnxruntime::onnxruntime)
+
+# Pass ORT variables as a preprocessor definition
+target_compile_definitions(${targetName} PRIVATE
+    $<$<BOOL:${ORT_ROCM_BUILD}>:ORT_ROCM_BUILD>
+    $<$<BOOL:${ORT_CUDA_BUILD}>:ORT_CUDA_BUILD>
+    $<$<BOOL:${ORT_MIGRAPHX_BUILD}>:ORT_MIGRAPHX_BUILD>
+    $<$<BOOL:${ORT_TENSORRT_BUILD}>:ORT_TENSORRT_BUILD>)

Common/ML/include/ML/3rdparty/GPUORTFloat16.h

@@ -535,22 +535,22 @@ GPUdi() uint16_t BFloat16Impl<Derived>::ToUint16Impl(float v) noexcept
     result = kPositiveQNaNBits;
   } else {
     auto get_msb_half = [](float fl) {
-      uint16_t result;
+      uint16_t res;
 #ifdef GPUCA_GPUCODE
-      o2::gpu::CAMath::memcpy(&result, reinterpret_cast<char*>(&fl) + sizeof(uint16_t), sizeof(uint16_t));
+      o2::gpu::CAMath::memcpy(&res, reinterpret_cast<char*>(&fl) + sizeof(uint16_t), sizeof(uint16_t));
 #else
 #ifdef __cpp_if_constexpr
       if constexpr (detail::endian::native == detail::endian::little)
 #else
       if (detail::endian::native == detail::endian::little)
 #endif
       {
-        std::memcpy(&result, reinterpret_cast<char*>(&fl) + sizeof(uint16_t), sizeof(uint16_t));
+        std::memcpy(&res, reinterpret_cast<char*>(&fl) + sizeof(uint16_t), sizeof(uint16_t));
       } else {
-        std::memcpy(&result, &fl, sizeof(uint16_t));
+        std::memcpy(&res, &fl, sizeof(uint16_t));
       }
 #endif
-      return result;
+      return res;
     };
 
     uint16_t upper_bits = get_msb_half(v);
@@ -882,4 +882,4 @@ static_assert(sizeof(BFloat16_t) == sizeof(uint16_t), "Sizes must match");
 } // namespace OrtDataType
 
 } // namespace o2
-#endif
+#endif

Common/ML/include/ML/OrtInterface.h

@@ -22,9 +22,17 @@
 #include <memory>
 #include <map>
 #include <thread>
+#include <unordered_map>
 
 // O2 includes
-#include "Framework/Logger.h"
+#include "GPUCommonLogger.h"
+
+namespace Ort
+{
+struct SessionOptions;
+struct MemoryInfo;
+struct Env;
+} // namespace Ort
 
 namespace o2
 {
@@ -36,14 +44,52 @@ class OrtModel
 {
 
  public:
-  // Constructor
+  // Constructors & destructors
   OrtModel() = default;
-  OrtModel(std::unordered_map<std::string, std::string> optionsMap) { reset(optionsMap); }
-  void init(std::unordered_map<std::string, std::string> optionsMap) { reset(optionsMap); }
-  void reset(std::unordered_map<std::string, std::string>);
+  OrtModel(std::unordered_map<std::string, std::string> optionsMap) { init(optionsMap); }
+  void init(std::unordered_map<std::string, std::string> optionsMap)
+  {
+    initOptions(optionsMap);
+    initEnvironment();
+  }
+  virtual ~OrtModel() = default;
+
+  // General purpose
+  void initOptions(std::unordered_map<std::string, std::string> optionsMap);
+  void initEnvironment();
+  void initSession();
+  void memoryOnDevice(int32_t = 0);
   bool isInitialized() { return mInitialized; }
+  void resetSession();
 
-  virtual ~OrtModel() = default;
+  // Getters
+  std::vector<std::vector<int64_t>> getNumInputNodes() const { return mInputShapes; }
+  std::vector<std::vector<int64_t>> getNumOutputNodes() const { return mOutputShapes; }
+  std::vector<std::string> getInputNames() const { return mInputNames; }
+  std::vector<std::string> getOutputNames() const { return mOutputNames; }
+  Ort::SessionOptions* getSessionOptions();
+  Ort::MemoryInfo* getMemoryInfo();
+  Ort::Env* getEnv();
+  int32_t getIntraOpNumThreads() const { return intraOpNumThreads; }
+  int32_t getInterOpNumThreads() const { return interOpNumThreads; }
+
+  // Setters
+  void setDeviceId(int32_t id) { deviceId = id; }
+  void setIO();
+  void setActiveThreads(int threads) { intraOpNumThreads = threads; }
+  void setIntraOpNumThreads(int threads)
+  {
+    if (deviceType == "CPU") {
+      intraOpNumThreads = threads;
+    }
+  }
+  void setInterOpNumThreads(int threads)
+  {
+    if (deviceType == "CPU") {
+      interOpNumThreads = threads;
+    }
+  }
+  void setEnv(Ort::Env*);
 
   // Conversion
   template <class I, class O>
@@ -53,41 +99,36 @@ class OrtModel
   template <class I, class O> // class I is the input data type, e.g. float, class O is the output data type, e.g. OrtDataType::Float16_t from O2/Common/ML/include/ML/GPUORTFloat16.h
   std::vector<O> inference(std::vector<I>&);
 
-  template <class I, class O> // class I is the input data type, e.g. float, class O is the output data type, e.g. O2::gpu::OrtDataType::Float16_t from O2/GPU/GPUTracking/ML/convert_float16.h
+  template <class I, class O>
   std::vector<O> inference(std::vector<std::vector<I>>&);
 
-  template <class I, class O> // class I is the input data type, e.g. float, class O is the output data type, e.g. OrtDataType::Float16_t from O2/Common/ML/include/ML/GPUORTFloat16.h
-  void inference(I*, size_t, O*);
-
-  // template<class I, class T, class O> // class I is the input data type, e.g. float, class T the throughput data type and class O is the output data type
-  // std::vector<O> inference(std::vector<I>&);
-
-  // Reset session
-  void resetSession();
+  template <class I, class O>
+  void inference(I*, int64_t, O*);
 
-  std::vector<std::vector<int64_t>> getNumInputNodes() const { return mInputShapes; }
-  std::vector<std::vector<int64_t>> getNumOutputNodes() const { return mOutputShapes; }
-  std::vector<std::string> getInputNames() const { return mInputNames; }
-  std::vector<std::string> getOutputNames() const { return mOutputNames; }
+  template <class I, class O>
+  void inference(I**, int64_t, O*);
 
-  void setActiveThreads(int threads) { intraOpNumThreads = threads; }
+  void release(bool = false);
 
  private:
-  // ORT variables -> need to be hidden as Pimpl
+  // ORT variables -> need to be hidden as pImpl
   struct OrtVariables;
   OrtVariables* pImplOrt;
 
   // Input & Output specifications of the loaded network
   std::vector<const char*> inputNamesChar, outputNamesChar;
   std::vector<std::string> mInputNames, mOutputNames;
-  std::vector<std::vector<int64_t>> mInputShapes, mOutputShapes;
+  std::vector<std::vector<int64_t>> mInputShapes, mOutputShapes, inputShapesCopy, outputShapesCopy; // Input shapes
+  std::vector<int64_t> inputSizePerNode, outputSizePerNode;                                         // Output shapes
+  int32_t mInputsTotal = 0, mOutputsTotal = 0;                                                      // Total number of inputs and outputs
 
   // Environment settings
   bool mInitialized = false;
-  std::string modelPath, device = "cpu", dtype = "float", thread_affinity = ""; // device options should be cpu, rocm, migraphx, cuda
-  int intraOpNumThreads = 1, interOpNumThreads = 1, deviceId = 0, enableProfiling = 0, loggingLevel = 0, allocateDeviceMemory = 0, enableOptimizations = 0;
+  std::string modelPath, envName = "", deviceType = "CPU", thread_affinity = ""; // device options should be cpu, rocm, migraphx, cuda
+  int32_t intraOpNumThreads = 1, interOpNumThreads = 1, deviceId = -1, enableProfiling = 0, loggingLevel = 0, allocateDeviceMemory = 0, enableOptimizations = 0;
 
   std::string printShape(const std::vector<int64_t>&);
+  std::string printShape(const std::vector<std::vector<int64_t>>&, std::vector<std::string>&);
 };
 
 } // namespace ml