GPU: Some compilation fixes without ROOT headers

Common/MathUtils/include/MathUtils/SMatrixGPU.h

@@ -29,6 +29,7 @@
 #include "GPUCommonMath.h"
 #include "GPUCommonAlgorithm.h"
 #include "GPUCommonLogger.h"
+#include "GPUCommonTypeTraits.h"
 
 namespace o2::math_utils::detail
 {
@@ -468,6 +469,9 @@ class SMatrixGPU
   GPUd() const T& operator()(unsigned int i, unsigned int j) const;
   GPUd() T& operator()(unsigned int i, unsigned int j);
 
+  template <typename Y, typename X>
+  GPUd() friend X& operator<<(Y& y, const SMatrixGPU&);
+
   class SMatrixRowGPU
   {
    public:
@@ -512,6 +516,15 @@ class SMatrixGPU
   R mRep;
 };
 
+#ifndef __OPENCL__ // TODO: current C++ for OpenCL 2021 is at C++17, so no concepts. But we don't need this trick for OpenCL anyway, so we can just hide it.
+template <class T, unsigned int D1, unsigned int D2, class R, typename Y, typename X = Y>
+  requires(sizeof(typename X::traits_type::pos_type) != 0) // do not provide a template to fair::Logger, etc... (pos_type is a member type of all std::ostream classes)
+GPUd() X& operator<<(Y& y, const SMatrixGPU<T, D1, D2, R>&)
+{
+  return y;
+}
+#endif
+
 template <class T, unsigned int D1, unsigned int D2, class R>
 GPUdi() SMatrixGPU<T, D1, D2, R>::SMatrixGPU(SMatrixIdentity)
 {

GPU/Common/GPUCommonConstants.h

@@ -17,7 +17,7 @@
 
 #include "GPUCommonDef.h"
 
-#if !defined(__OPENCL__) || defined(__OPENCLCPP__)
+#if !defined(__OPENCL1__)
 namespace GPUCA_NAMESPACE::gpu::gpu_common_constants
 {
 static CONSTEXPR const float kCLight = 0.000299792458f;

GPU/Common/GPUCommonDef.h

@@ -30,7 +30,7 @@
 //Some GPU configuration settings, must be included first
 #include "GPUCommonDefSettings.h"
 
-#if (!defined(__OPENCL__) || defined(__OPENCLCPP__)) && (!(defined(__CINT__) || defined(__ROOTCINT__)) || defined(__CLING__)) && defined(__cplusplus) && __cplusplus >= 201103L
+#if !defined(__OPENCL1__) && (!(defined(__CINT__) || defined(__ROOTCINT__)) || defined(__CLING__)) && defined(__cplusplus) && __cplusplus >= 201103L
   #define GPUCA_NOCOMPAT // C++11 + No old ROOT5 + No old OpenCL
   #ifndef __OPENCL__
     #define GPUCA_NOCOMPAT_ALLOPENCL // + No OpenCL at all
@@ -82,7 +82,7 @@
   #define GPUCA_NAMESPACE o2
 #endif
 
-#if (defined(__CUDACC__) && defined(GPUCA_CUDA_NO_CONSTANT_MEMORY)) || (defined(__HIPCC__) && defined(GPUCA_HIP_NO_CONSTANT_MEMORY)) || (defined(__OPENCL__) && !defined(__OPENCLCPP__) && defined(GPUCA_OPENCL_NO_CONSTANT_MEMORY)) || (defined(__OPENCLCPP__) && defined(GPUCA_OPENCLCPP_NO_CONSTANT_MEMORY))
+#if (defined(__CUDACC__) && defined(GPUCA_CUDA_NO_CONSTANT_MEMORY)) || (defined(__HIPCC__) && defined(GPUCA_HIP_NO_CONSTANT_MEMORY)) || (defined(__OPENCL1__) && defined(GPUCA_OPENCL_NO_CONSTANT_MEMORY)) || (defined(__OPENCLCPP__) && defined(GPUCA_OPENCLCPP_NO_CONSTANT_MEMORY))
   #define GPUCA_NO_CONSTANT_MEMORY
 #elif defined(__CUDACC__) || defined(__HIPCC__)
   #define GPUCA_HAS_GLOBAL_SYMBOL_CONSTANT_MEM

GPU/Common/GPUCommonMath.h

@@ -31,7 +31,7 @@
 #include <cstdint>
 #endif
 
-#if !defined(__OPENCL__) || defined(__OPENCLCPP__)
+#if !defined(__OPENCL1__)
 namespace GPUCA_NAMESPACE
 {
 namespace gpu
@@ -220,7 +220,7 @@ GPUdi() uint32_t GPUCommonMath::Float2UIntReint(const float& x)
 {
 #if defined(GPUCA_GPUCODE_DEVICE) && (defined(__CUDACC__) || defined(__HIPCC__))
   return __float_as_uint(x);
-#elif defined(GPUCA_GPUCODE_DEVICE) && (defined(__OPENCL__) || defined(__OPENCLCPP__))
+#elif defined(GPUCA_GPUCODE_DEVICE) && defined(__OPENCL__)
   return as_uint(x);
 #else
   return reinterpret_cast<const uint32_t&>(x);
@@ -289,7 +289,7 @@ GPUhdi() void GPUCommonMath::SinCosd(double x, double& s, double& c)
 
 GPUdi() uint32_t GPUCommonMath::Clz(uint32_t x)
 {
-#if (defined(__GNUC__) || defined(__clang__) || defined(__CUDACC__) || defined(__HIPCC__)) && (!defined(__OPENCL__) || defined(__OPENCLCPP__))
+#if (defined(__GNUC__) || defined(__clang__) || defined(__CUDACC__) || defined(__HIPCC__)) && !defined(__OPENCL1__)
   return x == 0 ? 32 : CHOICE(__builtin_clz(x), __clz(x), __builtin_clz(x)); // use builtin if available
 #else
   for (int32_t i = 31; i >= 0; i--) {
@@ -303,7 +303,7 @@ GPUdi() uint32_t GPUCommonMath::Clz(uint32_t x)
 
 GPUdi() uint32_t GPUCommonMath::Popcount(uint32_t x)
 {
-#if (defined(__GNUC__) || defined(__clang__) || defined(__CUDACC__) || defined(__HIPCC__)) && (!defined(__OPENCL__) /*|| defined(__OPENCLCPP__)*/) // TODO: remove OPENCLCPP workaround when reported SPIR-V bug is fixed
+#if (defined(__GNUC__) || defined(__clang__) || defined(__CUDACC__) || defined(__HIPCC__)) && (!defined(__OPENCL__) /* !defined(__OPENCL1__)*/) // TODO: exclude only OPENCLC (not CPP) when reported SPIR-V bug is fixed
   // use builtin if available
   return CHOICE(__builtin_popcount(x), __popc(x), __builtin_popcount(x));
 #else
@@ -563,7 +563,7 @@ GPUdii() void GPUCommonMath::AtomicMinInternal(GPUglobalref() GPUgeneric() GPUAt
 
 #undef CHOICE
 
-#if !defined(__OPENCL__) || defined(__OPENCLCPP__)
+#if !defined(__OPENCL1__)
 }
 }
 #endif

GPU/Common/GPUCommonRtypes.h

@@ -20,14 +20,14 @@
 #if defined(GPUCA_STANDALONE) || (defined(GPUCA_O2_LIB) && !defined(GPUCA_O2_INTERFACE)) || defined(GPUCA_GPUCODE) // clang-format off
   #if !defined(ROOT_Rtypes) && !defined(__CLING__)
     #define GPUCOMMONRTYPES_H_ACTIVE
+    struct MUST_NOT_USE_Rtypes_h {};
+    typedef MUST_NOT_USE_Rtypes_h TClass;
     #define ClassDef(name,id)
     #define ClassDefNV(name, id)
     #define ClassDefOverride(name, id)
     #define ClassImp(name)
     #define templateClassImp(name)
     #ifndef GPUCA_GPUCODE_DEVICE
-//      typedef uint64_t ULong64_t;
-//      typedef uint32_t UInt_t;
       #include <iostream>
     #endif
   #endif

GPU/Common/GPUCommonTypeTraits.h

@@ -21,7 +21,7 @@
 #ifndef GPUCA_GPUCODE_COMPILEKERNELS
 #include <type_traits>
 #endif
-#elif !defined(__OPENCL__) || defined(__OPENCLCPP__)
+#elif !defined(__OPENCL1__)
 // We just reimplement some type traits in std for the GPU
 namespace std
 {

GPU/Common/GPUROOTSMatrixFwd.h

@@ -52,7 +52,7 @@ template <class T, uint32_t D1, uint32_t D2>
 class MatRepStdGPU;
 } // namespace detail
 
-#if !defined(GPUCA_STANDALONE) && !defined(GPUCA_GPUCODE)
+#if !defined(GPUCA_STANDALONE) && !defined(GPUCA_GPUCODE) && !defined(GPUCOMMONRTYPES_H_ACTIVE)
 template <typename T, uint32_t N>
 using SVector = ROOT::Math::SVector<T, N>;
 template <class T, uint32_t D1, uint32_t D2, class R>

GPU/GPUTracking/Base/GPUParam.inc

@@ -17,7 +17,7 @@
 
 #include "GPUParam.h"
 #include "GPUTPCGMMergedTrackHit.h"
-#if !defined(__OPENCL__) || defined(__OPENCLCPP__)
+#if !defined(__OPENCL1__)
 #include "GPUTPCClusterOccupancyMap.h"
 #endif
 
@@ -228,7 +228,7 @@ GPUdi() void MEM_LG(GPUParam)::UpdateClusterError2ByState(int16_t clusterState,
 MEM_CLASS_PRE()
 GPUdi() float MEM_LG(GPUParam)::GetUnscaledMult(float time) const
 {
-#if !defined(__OPENCL__) || defined(__OPENCLCPP__)
+#if !defined(__OPENCL1__)
   if (!occupancyMap) {
     return 0.f;
   }

GPU/GPUTracking/Base/opencl-common/GPUReconstructionOCL.cl

@@ -14,6 +14,11 @@
 
 // clang-format off
 #define __OPENCL__
+#if defined(__cplusplus) && __cplusplus >= 201703L
+  #define __OPENCLCPP__
+#else
+  #define __OPENCL1__
+#endif
 #define GPUCA_GPUTYPE_OPENCL
 
 #ifdef __OPENCLCPP__

GPU/GPUTracking/Base/opencl2/CMakeLists.txt

@@ -32,8 +32,6 @@ set(OCL_DEFINECL "-D$<JOIN:$<TARGET_PROPERTY:O2::GPUTracking,COMPILE_DEFINITIONS
             -I${CMAKE_SOURCE_DIR}/Detectors/TRD/base/src
             -I${CMAKE_SOURCE_DIR}/Detectors/Base/src
             -I${CMAKE_SOURCE_DIR}/DataFormats/Reconstruction/src
-            -I${CMAKE_SOURCE_DIR}/Detectors/ITSMFT/ITS/tracking/cuda/include
-            -D__OPENCLCPP__
 )
 
 set(SRCS GPUReconstructionOCL2.cxx)

GPU/GPUTracking/DataTypes/CalibdEdxContainer.cxx

@@ -12,14 +12,13 @@
 /// \file  CalibdEdxContainer.cxx
 /// \author Matthias Kleiner <[email protected]>
 
-#include "CalibdEdxContainer.h"
-
-#if !defined(GPUCA_GPUCODE) && !defined(GPUCA_STANDALONE)
+#if !defined(GPUCA_STANDALONE)
 #include "TFile.h"
 #include "TPCBase/CalDet.h"
 #include "Framework/Logger.h"
 #include "clusterFinderDefs.h"
 #endif
+#include "CalibdEdxContainer.h"
 
 using namespace GPUCA_NAMESPACE::gpu;
 using namespace o2::tpc;

GPU/GPUTracking/DataTypes/CalibdEdxTrackTopologyPol.cxx

@@ -9,6 +9,7 @@
 // granted to it by virtue of its status as an Intergovernmental Organization
 // or submit itself to any jurisdiction.
 
+#include "Rtypes.h"
 #include "CalibdEdxTrackTopologyPol.h"
 
 #include <cstddef>
@@ -20,6 +21,7 @@
 using namespace o2::tpc;
 
 #if !defined(GPUCA_GPUCODE) && !defined(GPUCA_STANDALONE) // code invisible on GPU and in the standalone compilation
+#include "NDPiecewisePolynomials.inc"
 void CalibdEdxTrackTopologyPol::dumpToTree(const uint32_t nSamplingPoints[/* Dim */], const char* outName) const
 {
   for (uint32_t i = 0; i < FFits; i++) {

GPU/GPUTracking/DataTypes/CalibdEdxTrackTopologySpline.cxx

@@ -14,11 +14,10 @@
 ///
 /// \author  Matthias Kleiner <[email protected]>
 
-#include "CalibdEdxTrackTopologySpline.h"
-
 #if !defined(GPUCA_STANDALONE)
 #include "TFile.h"
 #endif
+#include "CalibdEdxTrackTopologySpline.h"
 
 using namespace GPUCA_NAMESPACE::gpu;
 using namespace o2::tpc;

GPU/GPUTracking/DataTypes/CalibdEdxTrackTopologySpline.h

@@ -19,12 +19,12 @@
 
 #include "FlatObject.h"
 #include "Spline.h"
+#include "GPUCommonRtypes.h"
 #ifdef GPUCA_HAVE_O2HEADERS
 #include "DataFormatsTPC/Defs.h"
 #endif
 
 #if !defined(GPUCA_GPUCODE) && !defined(GPUCA_STANDALONE) // code invisible on GPU and in the standalone compilation
-#include "Rtypes.h"                                       // for ClassDefNV
 #include <fmt/format.h>
 #endif
 

GPU/GPUTracking/DataTypes/GPUDataTypes.h

@@ -125,7 +125,7 @@ namespace gpu
 #define GPUCA_RECO_STEP GPUDataTypes
 #endif
 
-#if defined(__OPENCL__) && !defined(__OPENCLCPP__)
+#if defined(__OPENCL1__)
 MEM_CLASS_PRE() // Macro with some template magic for OpenCL 1.2
 #endif
 class GPUTPCTrack;

GPU/GPUTracking/DataTypes/GPUO2DataTypes.h

@@ -17,7 +17,7 @@
 
 // Pull in several O2 headers with basic data types, or load a header with empty fake classes if O2 headers not available
 
-#if defined(GPUCA_HAVE_O2HEADERS) && (!defined(__OPENCL__) || defined(__OPENCLCPP__))
+#if defined(GPUCA_HAVE_O2HEADERS) && !defined(__OPENCL1__)
 #include "DataFormatsTPC/ClusterNative.h"
 #include "DataFormatsTPC/Digit.h"
 #include "DetectorsBase/MatLayerCylSet.h"
@@ -27,7 +27,7 @@
 #include "GPUO2FakeClasses.h"
 #endif
 
-#if !defined(__OPENCL__) || defined(__OPENCLCPP__)
+#if !defined(__OPENCL1__)
 #include "GPUdEdxInfo.h"
 #endif
 

GPU/GPUTracking/DataTypes/GPUSettings.h

@@ -45,7 +45,7 @@ class GPUSettings
                               RejectionStrategyA = 1,
                               RejectionStrategyB = 2 };
 
-#if !defined(__OPENCL__) || defined(__OPENCLCPP__)
+#if !defined(__OPENCL1__)
   static CONSTEXPR const uint32_t TPC_MAX_TF_TIME_BIN = ((256 * 3564 + 2 * 8 - 2) / 8);
 #endif
 };

GPU/GPUTracking/DataTypes/GPUTPCGMPolynomialField.h

@@ -29,7 +29,7 @@ namespace gpu
 class GPUTPCGMPolynomialField
 {
  public:
-#if !defined(__OPENCL__) || defined(__OPENCLCPP__)
+#if !defined(__OPENCL1__)
   GPUTPCGMPolynomialField() : mNominalBz(0.f)
   {
     Reset();
@@ -94,7 +94,7 @@ class GPUTPCGMPolynomialField
   float mItsBz[NITSM];
 };
 
-#if !defined(__OPENCL__) || defined(__OPENCLCPP__)
+#if !defined(__OPENCL1__)
 
 inline void GPUTPCGMPolynomialField::Reset()
 {

GPU/GPUTracking/DataTypes/GPUTPCGeometry.h

@@ -34,7 +34,7 @@ namespace gpu
 // Should be unified, but cannot take the contants from the official headers for now, since we want it to be constexpr
 class GPUTPCGeometry // TODO: Make values constexpr
 {
-#if defined(__OPENCL__) && !defined(__OPENCLCPP__)
+#if defined(__OPENCL1__)
   GPUTPCGeometry(); // Fake constructor declaration for OpenCL due to static members, does not exist!
 #endif
 #ifdef GPUCA_TPC_GEOMETRY_O2
@@ -63,7 +63,7 @@ class GPUTPCGeometry // TODO: Make values constexpr
   const float mPadHeight[10] GPUCA_CPP11_INIT(= {.75f, .75f, .75f, .75f, 1.f, 1.f, 1.2f, 1.2f, 1.5f, 1.5f});
   const float mPadWidth[10] GPUCA_CPP11_INIT(= {.416f, .420f, .420f, .436f, .6f, .6f, .608f, .588f, .604f, .607f});
 
-#if !defined(__OPENCL__) || defined(__OPENCLCPP__)
+#if !defined(__OPENCL1__)
   static CONSTEXPR float FACTOR_T2Z GPUCA_CPP11_INIT(= 250.f / 512.f); // Used in compression, must remain constant at 250cm, 512 time bins!
 #endif
 
@@ -95,7 +95,7 @@ class GPUTPCGeometry // TODO: Make values constexpr
   const float mPadHeight[3] GPUCA_CPP11_INIT(= {.75f, 1.f, 1.5f});
   const float mPadWidth[3] GPUCA_CPP11_INIT(= {.4f, .6f, .6f});
 
-#if !defined(__OPENCL__) || defined(__OPENCLCPP__)
+#if !defined(__OPENCL1__)
   static CONSTEXPR float FACTOR_T2Z GPUCA_CPP11_INIT(= 250.f / 1024.f); // Used in compression, must remain constant at 250cm, 1024 time bins!
 #endif
 
@@ -109,7 +109,7 @@ class GPUTPCGeometry // TODO: Make values constexpr
   GPUd() int32_t EndOROC2() const { return GPUCA_ROW_COUNT; }
 #endif
  private:
-#if !defined(__OPENCL__) || defined(__OPENCLCPP__)
+#if !defined(__OPENCL1__)
   static CONSTEXPR float FACTOR_Z2T GPUCA_CPP11_INIT(= 1.f / FACTOR_T2Z);
 #endif
  public:
@@ -120,7 +120,7 @@ class GPUTPCGeometry // TODO: Make values constexpr
   GPUd() float PadWidth(int32_t row) const { return (mPadWidth[GetRegion(row)]); }
   GPUd() uint8_t NPads(int32_t row) const { return mNPads[row]; }
 
-#if !defined(__OPENCL__) || defined(__OPENCLCPP__)
+#if !defined(__OPENCL1__)
   GPUd() float LinearPad2Y(int32_t slice, int32_t row, float pad) const
   {
     const float u = (pad - 0.5f * mNPads[row]) * PadWidth(row);

GPU/GPUTracking/Definitions/GPUDefConstantsAndSettings.h

@@ -51,7 +51,7 @@
 #if defined(GPUCA_NSLICES) || defined(GPUCA_ROW_COUNT)
   #error GPUCA_NSLICES or GPUCA_ROW_COUNT already defined, do not include GPUTPCGeometry.h before!
 #endif
-#if defined(GPUCA_HAVE_O2HEADERS) && defined(GPUCA_TPC_GEOMETRY_O2) && (!defined(__OPENCL__) || defined(__OPENCLCPP__)) && !(defined(ROOT_VERSION_CODE) && ROOT_VERSION_CODE < 393216)
+#if defined(GPUCA_HAVE_O2HEADERS) && defined(GPUCA_TPC_GEOMETRY_O2) && !defined(__OPENCL1__) && !(defined(ROOT_VERSION_CODE) && ROOT_VERSION_CODE < 393216)
   //Use definitions from the O2 headers if available for nicer code and type safety
   #include "DataFormatsTPC/Constants.h"
   #define GPUCA_NSLICES o2::tpc::constants::MAXSECTOR

GPU/GPUTracking/Definitions/GPUSettingsList.h

@@ -131,6 +131,7 @@ AddOptionRTC(cfNoiseSuppressionEpsilonRelative, uint8_t, 76, "", 0, "Cluster Fin
 AddOptionRTC(nWays, uint8_t, 3, "", 0, "Do N fit passes in final fit of merger")
 AddOptionRTC(nWaysOuter, int8_t, 0, "", 0, "Store outer param")
 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(rejectIFCLowRadiusCluster, uint8_t, 0, "", 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(globalTracking, int8_t, 1, "", 0, "Enable Global Tracking (prolong tracks to adjacent sectors to find short segments)")
@@ -345,7 +346,8 @@ AddOption(drawTracksAndFilter, bool, false, "", 0, "Use AND filter instead of OR
 AddOption(propagateLoopers, bool, false, "", 0, "Enabale propagation of loopers")
 AddOption(clustersOnly, bool, false, "", 0, "Visualize clusters only")
 AddOption(clustersOnNominalRow, bool, false, "", 0, "Show clusters at nominal x of pad row for early-transformed data")
-AddOption(separateGlobalTracks, bool, false, "", 0, "Separate global tracks")
+AddOption(separateGlobalTracks, bool, false, "", 0, "Draw track segments propagated to adjacent sectors separately")
+AddOption(splitCETracks, int8_t, -1, "", 0, "Split CE tracks when they cross the central electrode (-1 = for triggered data)")
 AddOption(markClusters, int32_t, 0, "", 0, "Mark clusters")
 AddOption(markFakeClusters, int32_t, 0, "", 0, "Mark fake clusters")
 AddOption(markAdjacentClusters, int32_t, 0, "", 0, "Mark adjacent clusters")

GPU/GPUTracking/Interface/GPUO2InterfaceQA.cxx

@@ -12,11 +12,11 @@
 /// \file GPUO2InterfaceQA.cxx
 /// \author David Rohr
 
+#include "TGraphAsymmErrors.h"
 #include "GPUParam.h"
 #include "GPUQA.h"
 #include "GPUO2InterfaceConfiguration.h"
 #include "GPUO2InterfaceQA.h"
-#include "TGraphAsymmErrors.h"
 
 using namespace o2::gpu;
 using namespace o2::tpc;