GPU: Improve configuration dump

Author: davidrohr

GPU/GPUTracking/Base/GPUReconstruction.cxx

@@ -40,6 +40,8 @@
 #include "GPUReconstruction.h"
 #include "GPUReconstructionIncludes.h"
 #include "GPUROOTDumpCore.h"
+#include "GPUConfigDump.h"
+#include "GPUChainTracking.h"
 
 #include "GPUMemoryResource.h"
 #include "GPUChain.h"
@@ -212,18 +214,30 @@ int GPUReconstruction::Init()
 
 int GPUReconstruction::InitPhaseBeforeDevice()
 {
+  if (mProcessingSettings.printSettings) {
+    if (mSlaves.size() || mMaster) {
+      printf("\nConfig Dump %s\n", mMaster ? "Slave" : "Master");
+    }
+    const GPUChainTracking* chTrk;
+    for (unsigned int i = 0; i < mChains.size(); i++) {
+      if ((chTrk = dynamic_cast<GPUChainTracking*>(mChains[i].get()))) {
+        break;
+      }
+    }
+    GPUConfigDump::dumpConfig(&param().rec, &mProcessingSettings, chTrk ? chTrk->GetQAConfig() : nullptr, chTrk ? chTrk->GetEventDisplayConfig() : nullptr, &mDeviceBackendSettings, &mRecoSteps);
+  }
 #ifndef GPUCA_HAVE_O2HEADERS
-  mRecoSteps.setBits(RecoStep::ITSTracking, false);
-  mRecoSteps.setBits(RecoStep::TRDTracking, false);
-  mRecoSteps.setBits(RecoStep::TPCConversion, false);
-  mRecoSteps.setBits(RecoStep::TPCCompression, false);
-  mRecoSteps.setBits(RecoStep::TPCdEdx, false);
+  mRecoSteps.steps.setBits(RecoStep::ITSTracking, false);
+  mRecoSteps.steps.setBits(RecoStep::TRDTracking, false);
+  mRecoSteps.steps.setBits(RecoStep::TPCConversion, false);
+  mRecoSteps.steps.setBits(RecoStep::TPCCompression, false);
+  mRecoSteps.steps.setBits(RecoStep::TPCdEdx, false);
   mProcessingSettings.createO2Output = false;
 #endif
-  mRecoStepsGPU &= mRecoSteps;
-  mRecoStepsGPU &= AvailableRecoSteps();
+  mRecoSteps.stepsGPUMask &= mRecoSteps.steps;
+  mRecoSteps.stepsGPUMask &= AvailableGPURecoSteps();
   if (!IsGPU()) {
-    mRecoStepsGPU.set((unsigned char)0);
+    mRecoSteps.stepsGPUMask.set((unsigned char)0);
   }
 
   if (mProcessingSettings.forceMemoryPoolSize >= 1024 || mProcessingSettings.forceHostMemoryPoolSize >= 1024) {
@@ -286,7 +300,7 @@ int GPUReconstruction::InitPhaseBeforeDevice()
   if (!mProcessingSettings.createO2Output || !IsGPU()) {
     mProcessingSettings.clearO2OutputFromGPU = false;
   }
-  if (!(mRecoStepsGPU & GPUDataTypes::RecoStep::TPCMerging)) {
+  if (!(mRecoSteps.stepsGPUMask & GPUDataTypes::RecoStep::TPCMerging)) {
     mProcessingSettings.mergerSortTracks = false;
   }
   if (!IsGPU()) {
@@ -296,7 +310,7 @@ int GPUReconstruction::InitPhaseBeforeDevice()
   if (param().rec.nonConsecutiveIDs) {
     param().rec.tpc.disableRefitAttachment = 0xFF;
   }
-  if (!(mRecoStepsGPU & RecoStep::TPCMerging) || !param().rec.tpc.mergerReadFromTrackerDirectly) {
+  if (!(mRecoSteps.stepsGPUMask & RecoStep::TPCMerging) || !param().rec.tpc.mergerReadFromTrackerDirectly) {
     mProcessingSettings.fullMergerOnGPU = false;
   }
   if (mProcessingSettings.debugLevel > 3 || !mProcessingSettings.fullMergerOnGPU || mProcessingSettings.deterministicGPUReconstruction) {
@@ -1113,7 +1127,7 @@ void GPUReconstruction::UpdateSettings(const GPUSettingsGRP* g, const GPUSetting
     mProcessingSettings.debugLevel = p->debugLevel;
     mProcessingSettings.resetTimers = p->resetTimers;
   }
-  GPURecoStepConfiguration w = {mRecoSteps, mRecoStepsGPU, mRecoStepsInputs, mRecoStepsOutputs};
+  GPURecoStepConfiguration w = mRecoSteps;
   param().UpdateSettings(g, p, &w);
   if (mInitialized) {
     WriteConstantParams();
@@ -1161,10 +1175,10 @@ void GPUReconstruction::SetSettings(const GPUSettingsGRP* grp, const GPUSettings
     mProcessingSettings = *proc;
   }
   if (workflow) {
-    mRecoSteps = workflow->steps;
-    mRecoStepsGPU &= workflow->stepsGPUMask;
-    mRecoStepsInputs = workflow->inputs;
-    mRecoStepsOutputs = workflow->outputs;
+    mRecoSteps.steps = workflow->steps;
+    mRecoSteps.stepsGPUMask &= workflow->stepsGPUMask;
+    mRecoSteps.inputs = workflow->inputs;
+    mRecoSteps.outputs = workflow->outputs;
   }
   param().SetDefaults(&mGRPSettings, rec, proc, workflow);
 }

GPU/GPUTracking/Base/GPUReconstruction.h

@@ -210,10 +210,10 @@ class GPUReconstruction
   int NStreams() const { return mNStreams; }
   const void* DeviceMemoryBase() const { return mDeviceMemoryBase; }
 
-  RecoStepField GetRecoSteps() const { return mRecoSteps; }
-  RecoStepField GetRecoStepsGPU() const { return mRecoStepsGPU; }
-  InOutTypeField GetRecoStepsInputs() const { return mRecoStepsInputs; }
-  InOutTypeField GetRecoStepsOutputs() const { return mRecoStepsOutputs; }
+  RecoStepField GetRecoSteps() const { return mRecoSteps.steps; }
+  RecoStepField GetRecoStepsGPU() const { return mRecoSteps.stepsGPUMask; }
+  InOutTypeField GetRecoStepsInputs() const { return mRecoSteps.inputs; }
+  InOutTypeField GetRecoStepsOutputs() const { return mRecoSteps.outputs; }
   int getRecoStepNum(RecoStep step, bool validCheck = true);
   int getGeneralStepNum(GeneralStep step, bool validCheck = true);
 
@@ -286,7 +286,7 @@ class GPUReconstruction
   int ReadStructFromFile(const char* file, T* obj);
 
   // Others
-  virtual RecoStepField AvailableRecoSteps() { return RecoStep::AllRecoSteps; }
+  virtual RecoStepField AvailableGPURecoSteps() { return RecoStep::AllRecoSteps; }
   virtual bool CanQueryMaxMemory() { return false; }
 
   // Pointers to tracker classes
@@ -304,10 +304,7 @@ class GPUReconstruction
   GPUOutputControl mInputControl;                       // Prefefined input memory location for reading standalone dumps
   std::unique_ptr<GPUMemorySizeScalers> mMemoryScalers; // Scalers how much memory will be needed
 
-  RecoStepField mRecoSteps = RecoStep::AllRecoSteps;
-  RecoStepField mRecoStepsGPU = RecoStep::AllRecoSteps;
-  InOutTypeField mRecoStepsInputs = 0;
-  InOutTypeField mRecoStepsOutputs = 0;
+  GPURecoStepConfiguration mRecoSteps;
 
   std::string mDeviceName = "CPU";
 

GPU/GPUTracking/Base/GPUReconstructionCPU.h

@@ -198,7 +198,7 @@ inline int GPUReconstructionCPU::runKernel(krnlSetup&& setup, Args&&... args)
   if (myStep == GPUCA_RECO_STEP::NoRecoStep) {
     throw std::runtime_error("Failure running general kernel without defining RecoStep");
   }
-  int cpuFallback = IsGPU() ? (setup.x.device == krnlDeviceType::CPU ? 2 : (mRecoStepsGPU & myStep) != myStep) : 0;
+  int cpuFallback = IsGPU() ? (setup.x.device == krnlDeviceType::CPU ? 2 : (mRecoSteps.stepsGPUMask & myStep) != myStep) : 0;
   unsigned int& nThreads = setup.x.nThreads;
   unsigned int& nBlocks = setup.x.nBlocks;
   const unsigned int stream = setup.x.stream;

GPU/GPUTracking/Base/opencl/GPUReconstructionOCL1.h

@@ -40,7 +40,7 @@ class GPUReconstructionOCL1Backend : public GPUReconstructionOCL
   template <class S, class T, int I, bool MULTI>
   S& getKernelObject();
 
-  RecoStepField AvailableRecoSteps() override { return (RecoStep::TPCSliceTracking); }
+  RecoStepField AvailableGPURecoSteps() override { return (RecoStep::TPCSliceTracking); }
   bool ContextForAllPlatforms() override { return true; }
   bool CheckPlatform(unsigned int i) override;
   int GetOCLPrograms() override;

GPU/GPUTracking/CMakeLists.txt

@@ -68,7 +68,7 @@ set(SRCS
     TRDTracking/GPUTRDTrackerKernels.cxx
     Base/GPUParam.cxx)
 
-set(SRCS_DATATYPES DataTypes/GPUDataTypes.cxx)
+set(SRCS_DATATYPES DataTypes/GPUDataTypes.cxx DataTypes/GPUConfigDump.cxx)
 set(SRCS_DATATYPE_HEADERS DataTypes/GPUTPCGMPolynomialField.cxx)
 
 set(HDRS_CINT_O2 Merger/GPUTPCGMMergedTrack.h Merger/GPUTPCGMSliceTrack.h Merger/GPUTPCGMBorderTrack.h)

GPU/GPUTracking/DataTypes/GPUConfigDump.cxx

@@ -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.
+
+/// \file GPUConfigDump.cxx
+/// \author David Rohr
+
+#include "GPUConfigDump.h"
+#include "GPUDataTypes.h"
+#include "GPUSettings.h"
+
+#include <functional>
+#include <iostream>
+#include <cstdio>
+
+#include "utils/qconfig_helpers.h"
+
+using namespace GPUCA_NAMESPACE::gpu;
+
+namespace
+{
+GPUSettingsStandalone configStandalone;
+std::vector<std::function<void()>> qprint_global;
+#define QCONFIG_PRINT
+#include "utils/qconfig.h"
+#undef QCONFIG_PRINT
+} // namespace
+
+void GPUConfigDump::dumpConfig(const GPUSettingsRec* rec, const GPUSettingsProcessing* proc, const GPUSettingsQA* qa, const GPUSettingsDisplay* display, const GPUSettingsDeviceBackend* device, const GPURecoStepConfiguration* workflow)
+{
+  if (rec) {
+    qConfigPrint(*rec, "rec.");
+  }
+  if (proc) {
+    qConfigPrint(*proc, "proc.");
+  }
+  if (qa) {
+    qConfigPrint(*qa, "QA.");
+  }
+  if (display) {
+    qConfigPrint(*display, "display.");
+  }
+  if (device) {
+    std::cout << "\n\tGPUSettingsDeviceBackend:\n"
+              << "\tdeviceType = " << (int)device->deviceType << "\n"
+              << "\tforceDeviceType = " << (device->forceDeviceType ? "true" : "false") << "\n"
+              << "\tslave = " << (device->master ? "true" : "false") << "\n";
+  }
+  if (workflow) {
+    printf("\n\tReconstruction steps / inputs / outputs:\n\tReco Steps = 0x%08x\n\tReco Steps GPU = 0x%08x\n\tInputs = 0x%08x\n\tOutputs = 0x%08x\n", (unsigned int)workflow->steps, (unsigned int)workflow->stepsGPUMask, (unsigned int)workflow->inputs, (unsigned int)workflow->outputs);
+  }
+}

GPU/GPUTracking/DataTypes/GPUConfigDump.h

@@ -0,0 +1,37 @@
+// 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.
+
+/// \file GPUConfigDump.h
+/// \author David Rohr
+
+#ifndef GPUCONFIGDUMP_H
+#define GPUCONFIGDUMP_H
+
+#include "GPUCommonDef.h"
+
+namespace GPUCA_NAMESPACE::gpu
+{
+struct GPUSettingsRec;
+struct GPUSettingsProcessing;
+struct GPUSettingsQA;
+struct GPUSettingsDisplay;
+struct GPUSettingsDeviceBackend;
+struct GPURecoStepConfiguration;
+
+class GPUConfigDump
+{
+ public:
+  static void dumpConfig(const GPUSettingsRec* rec, const GPUSettingsProcessing* proc, const GPUSettingsQA* qa, const GPUSettingsDisplay* display, const GPUSettingsDeviceBackend* device, const GPURecoStepConfiguration* workflow);
+};
+
+} // namespace GPUCA_NAMESPACE::gpu
+
+#endif

GPU/GPUTracking/DataTypes/GPUDataTypes.h

@@ -17,7 +17,7 @@
 
 #include "GPUCommonDef.h"
 
-// These are basic and non-comprex data types, which will also be visible on the GPU.
+// These are basic and non-complex data types, which will also be visible on the GPU.
 // Please add complex data types required on the host but not GPU to GPUHostDataTypes.h and forward-declare!
 #ifndef GPUCA_GPUCODE_DEVICE
 #include <cstddef>

GPU/GPUTracking/Definitions/GPUSettingsList.h

@@ -282,6 +282,7 @@ AddOption(tpcDownscaledEdx, unsigned char, 0, "", 0, "If != 0, downscale dEdx pr
 AddOption(tpcMaxAttachedClustersPerSectorRow, unsigned int, 51000, "", 0, "Maximum number of TPC attached clusters which can be decoded per SectorRow")
 AddOption(tpcUseOldCPUDecoding, bool, false, "", 0, "Enable old CPU-based TPC decoding")
 AddOption(RTCcacheFolder, std::string, "./rtccache/", "", 0, "Folder in which the cache file is stored")
+AddOption(printSettings, bool, false, "", 0, "Print all settings when initializing")
 AddVariable(eventDisplay, GPUCA_NAMESPACE::gpu::GPUDisplayFrontendInterface*, nullptr)
 AddSubConfig(GPUSettingsProcessingRTC, rtc)
 AddSubConfig(GPUSettingsProcessingParam, param)
@@ -503,9 +504,9 @@ AddHelp("help", 'h')
 AddHelpAll("helpall", 'H')
 AddSubConfig(GPUSettingsRec, rec)
 AddSubConfig(GPUSettingsProcessing, proc)
-AddSubConfig(GPUSettingsTFSim, TF)
 AddSubConfig(GPUSettingsQA, QA)
 AddSubConfig(GPUSettingsDisplay, display)
+AddSubConfig(GPUSettingsTFSim, TF)
 AddSubConfig(GPUSettingsEG, EG)
 EndConfig()
 #endif // BeginConfig

GPU/GPUTracking/Global/GPUChainTracking.h

@@ -159,6 +159,8 @@ class GPUChainTracking : public GPUChain, GPUReconstructionHelpers::helperDelega
   GPUQA* GetQA() { return mQAFromForeignChain ? mQAFromForeignChain->mQA.get() : mQA.get(); }
   int ForceInitQA();
   void SetQAFromForeignChain(GPUChainTracking* chain) { mQAFromForeignChain = chain; }
+  const GPUSettingsDisplay* GetEventDisplayConfig() const { return mConfigDisplay; }
+  const GPUSettingsQA* GetQAConfig() const { return mConfigQA; }
 
   // Processing functions
   int RunTPCClusterizer(bool synchronizeOutput = true);