GPU QA: Use single GPU QA instance for double-pipeline

Author: davidrohr

GPU/GPUTracking/Benchmark/standalone.cxx

@@ -501,6 +501,7 @@ int ReadEvent(int n)
   if ((configStandalone.proc.runQA || configStandalone.eventDisplay) && !configStandalone.QA.noMC) {
     chainTracking->ForceInitQA();
     snprintf(filename, 256, "events/%s/mc.%d.dump", configStandalone.eventsDir, n);
+    chainTracking->GetQA()->UpdateChain(chainTracking);
     if (chainTracking->GetQA()->ReadO2MCData(filename)) {
       snprintf(filename, 256, "events/%s/mc.%d.dump", configStandalone.eventsDir, 0);
       if (chainTracking->GetQA()->ReadO2MCData(filename) && configStandalone.proc.runQA) {
@@ -734,12 +735,14 @@ int main(int argc, char** argv)
       recAsync->SetDebugLevelTmp(configStandalone.proc.debugLevel);
     }
     chainTrackingAsync = recAsync->AddChain<GPUChainTracking>();
+    chainTrackingAsync->SetQAFromForeignChain(chainTracking);
   }
   if (configStandalone.proc.doublePipeline) {
     if (configStandalone.proc.debugLevel >= 3) {
       recPipeline->SetDebugLevelTmp(configStandalone.proc.debugLevel);
     }
     chainTrackingPipeline = recPipeline->AddChain<GPUChainTracking>();
+    chainTrackingPipeline->SetQAFromForeignChain(chainTracking);
   }
 #ifdef GPUCA_HAVE_O2HEADERS
   if (!configStandalone.proc.doublePipeline) {

GPU/GPUTracking/Global/GPUChainTracking.cxx

@@ -373,11 +373,14 @@ int GPUChainTracking::Init()
   }
 
   if (GPUQA::QAAvailable() && (GetProcessingSettings().runQA || GetProcessingSettings().eventDisplay)) {
-    mQA.reset(new GPUQA(this));
+    auto& qa = mQAFromForeignChain ? mQAFromForeignChain->mQA : mQA;
+    if (!qa) {
+      qa.reset(new GPUQA(this));
+    }
   }
   if (GetProcessingSettings().eventDisplay) {
 #ifndef GPUCA_ALIROOT_LIB
-    mEventDisplay.reset(GPUDisplayInterface::getDisplay(GetProcessingSettings().eventDisplay, this, mQA.get()));
+    mEventDisplay.reset(GPUDisplayInterface::getDisplay(GetProcessingSettings().eventDisplay, this, GetQA()));
 #endif
     if (mEventDisplay == nullptr) {
       throw std::runtime_error("Error loading event display");
@@ -479,19 +482,21 @@ int GPUChainTracking::PrepareEvent()
 
 int GPUChainTracking::ForceInitQA()
 {
-  if (!mQA) {
-    mQA.reset(new GPUQA(this));
+  auto& qa = mQAFromForeignChain ? mQAFromForeignChain->mQA : mQA;
+  if (!qa) {
+    qa.reset(new GPUQA(this));
   }
-  if (!mQA->IsInitialized()) {
-    return mQA->InitQA();
+  if (!GetQA()->IsInitialized()) {
+    return GetQA()->InitQA();
   }
   return 0;
 }
 
 int GPUChainTracking::Finalize()
 {
-  if (GetProcessingSettings().runQA && mQA->IsInitialized() && !(mConfigQA && mConfigQA->shipToQC)) {
-    mQA->DrawQAHistograms();
+  if (GetProcessingSettings().runQA && GetQA()->IsInitialized() && !(mConfigQA && mConfigQA->shipToQC) && !mQAFromForeignChain) {
+    GetQA()->UpdateChain(this);
+    GetQA()->DrawQAHistograms();
   }
   if (GetProcessingSettings().debugLevel >= 6) {
     mDebugFile->close();
@@ -670,8 +675,8 @@ int GPUChainTracking::RunChain()
     mCompressionStatistics.reset(new GPUTPCClusterStatistics);
   }
   const bool needQA = GPUQA::QAAvailable() && (GetProcessingSettings().runQA || (GetProcessingSettings().eventDisplay && (mIOPtrs.nMCInfosTPC || GetProcessingSettings().runMC)));
-  if (needQA && mQA->IsInitialized() == false) {
-    if (mQA->InitQA(GetProcessingSettings().runQA ? -GetProcessingSettings().runQA : -1)) {
+  if (needQA && GetQA()->IsInitialized() == false) {
+    if (GetQA()->InitQA(GetProcessingSettings().runQA ? -GetProcessingSettings().runQA : -1)) {
       return 1;
     }
   }
@@ -796,7 +801,8 @@ int GPUChainTracking::RunChainFinalize()
   const bool needQA = GPUQA::QAAvailable() && (GetProcessingSettings().runQA || (GetProcessingSettings().eventDisplay && mIOPtrs.nMCInfosTPC));
   if (needQA && mFractionalQAEnabled) {
     mRec->getGeneralStepTimer(GeneralStep::QA).Start();
-    mQA->RunQA(!GetProcessingSettings().runQA);
+    GetQA()->UpdateChain(this);
+    GetQA()->RunQA(!GetProcessingSettings().runQA);
     mRec->getGeneralStepTimer(GeneralStep::QA).Stop();
     if (GetProcessingSettings().debugLevel == 0) {
       GPUInfo("Total QA runtime: %d us", (int)(mRec->getGeneralStepTimer(GeneralStep::QA).GetElapsedTime() * 1000000));

GPU/GPUTracking/Global/GPUChainTracking.h

@@ -156,9 +156,10 @@ class GPUChainTracking : public GPUChain, GPUReconstructionHelpers::helperDelega
   const GPUTPCGMMerger& GetTPCMerger() const { return processors()->tpcMerger; }
   GPUTPCGMMerger& GetTPCMerger() { return processors()->tpcMerger; }
   GPUDisplayInterface* GetEventDisplay() { return mEventDisplay.get(); }
-  const GPUQA* GetQA() const { return mQA.get(); }
-  GPUQA* GetQA() { return mQA.get(); }
+  const GPUQA* GetQA() const { return mQAFromForeignChain ? mQAFromForeignChain->mQA.get() : mQA.get(); }
+  GPUQA* GetQA() { return mQAFromForeignChain ? mQAFromForeignChain->mQA.get() : mQA.get(); }
   int ForceInitQA();
+  void SetQAFromForeignChain(GPUChainTracking* chain) { mQAFromForeignChain = chain; }
 
   // Processing functions
   int RunTPCClusterizer(bool synchronizeOutput = true);
@@ -254,6 +255,7 @@ class GPUChainTracking : public GPUChain, GPUReconstructionHelpers::helperDelega
   // Display / QA
   bool mDisplayRunning = false;
   std::unique_ptr<GPUDisplayInterface> mEventDisplay;
+  GPUChainTracking* mQAFromForeignChain = nullptr;
   std::unique_ptr<GPUQA> mQA;
   std::unique_ptr<GPUTPCClusterStatistics> mCompressionStatistics;
 

GPU/GPUTracking/Interface/GPUO2Interface.cxx

@@ -80,6 +80,9 @@ int GPUO2Interface::Initialize(const GPUO2InterfaceConfiguration& config)
   }
   for (unsigned int i = 0; i < mNContexts; i++) {
     mCtx[i].mChain = mCtx[i].mRec->AddChain<GPUChainTracking>(mConfig->configInterface.maxTPCHits, mConfig->configInterface.maxTRDTracklets);
+    if (i) {
+      mCtx[i].mChain->SetQAFromForeignChain(mCtx[0].mChain);
+    }
     mCtx[i].mChain->mConfigDisplay = &mConfig->configDisplay;
     mCtx[i].mChain->mConfigQA = &mConfig->configQA;
     mCtx[i].mRec->SetSettings(&mConfig->configGRP, &mConfig->configReconstruction, &mConfig->configProcessing, &mConfig->configWorkflow);
@@ -145,6 +148,7 @@ void GPUO2Interface::DumpEvent(int nEvent, GPUTrackingInOutPointers* data)
     if (mConfig->configProcessing.runMC) {
       mCtx[0].mChain->ForceInitQA();
       snprintf(fname, 1024, "mc.%d.dump", nEvent);
+      mCtx[0].mChain->GetQA()->UpdateChain(mCtx[0].mChain);
       mCtx[0].mChain->GetQA()->DumpO2MCData(fname);
     }
 #endif

GPU/GPUTracking/qa/GPUQA.h

@@ -113,6 +113,7 @@ class GPUQA
   int ReadO2MCData(const char* filename);
   static bool QAAvailable() { return true; }
   bool IsInitialized() { return mQAInitialized; }
+  void UpdateChain(GPUChainTracking* chain) { mTracking = chain; }
 
   const std::vector<TH1F>& getHistograms1D() const { return *mHist1D; }
   const std::vector<TH2F>& getHistograms2D() const { return *mHist2D; }