Implement changes from the CUDA framework review (#429)

Rename the cudautils namespace to cms::cuda or cms::cudatest, and drop the CUDA prefix from the symbols defined there. Always record and query the CUDA event, to minimize need for error checking in CUDAScopedContextProduce destructor. Add comments to highlight the pieces in CachingDeviceAllocator that have been changed wrt. cub. Various other updates and clean up: - enable CUDA for compute capability 3.5. - clean up CUDAService, CUDA tests and plugins. - add CUDA existence protections to BuildFiles. - mark thread-safe static variables with CMS_THREAD_SAFE.
cms-patatrack · Nov 27, 2020 · e0c507e · e0c507e
1 parent e451702
commit e0c507e
Show file tree

Hide file tree

Showing 4 changed files with 21 additions and 21 deletions.
diff --git a/SimTracker/TrackerHitAssociation/plugins/ClusterTPAssociationProducerCUDA.cc b/SimTracker/TrackerHitAssociation/plugins/ClusterTPAssociationProducerCUDA.cc
@@ -4,7 +4,7 @@
 
 #include <cuda_runtime.h>
 
-#include "CUDADataFormats/Common/interface/CUDAProduct.h"
+#include "CUDADataFormats/Common/interface/Product.h"
 #include "CUDADataFormats/SiPixelDigi/interface/SiPixelDigisCUDA.h"
 #include "CUDADataFormats/TrackingRecHit/interface/TrackingRecHit2DCUDA.h"
 #include "DataFormats/Common/interface/DetSetVector.h"
@@ -29,7 +29,7 @@
 #include "FWCore/Utilities/interface/InputTag.h"
 #include "Geometry/Records/interface/TrackerDigiGeometryRecord.h"
 #include "Geometry/TrackerGeometryBuilder/interface/TrackerGeometry.h"
-#include "HeterogeneousCore/CUDACore/interface/CUDAScopedContext.h"
+#include "HeterogeneousCore/CUDACore/interface/ScopedContext.h"
 #include "HeterogeneousCore/CUDAUtilities/interface/cudaCheck.h"
 #include "SimDataFormats/Track/interface/SimTrackContainer.h"
 #include "SimDataFormats/TrackerDigiSimLink/interface/PixelDigiSimLink.h"
@@ -71,10 +71,10 @@ class ClusterTPAssociationProducerCUDA : public edm::global::EDProducer<> {
   edm::EDGetTokenT<edmNew::DetSetVector<Phase2TrackerCluster1D>> phase2OTClustersToken_;
   edm::EDGetTokenT<TrackingParticleCollection> trackingParticleToken_;
 
-  edm::EDGetTokenT<CUDAProduct<SiPixelDigisCUDA>> tGpuDigis;
-  edm::EDGetTokenT<CUDAProduct<TrackingRecHit2DCUDA>> tGpuHits;
+  edm::EDGetTokenT<cms::cuda::Product<SiPixelDigisCUDA>> tGpuDigis;
+  edm::EDGetTokenT<cms::cuda::Product<TrackingRecHit2DCUDA>> tGpuHits;
 
-  edm::EDPutTokenT<CUDAProduct<ProductCUDA>> tokenGPUProd_;
+  edm::EDPutTokenT<cms::cuda::Product<ProductCUDA>> tokenGPUProd_;
 
   clusterSLOnGPU::Kernel m_gpuAlgo;
 };
@@ -94,12 +94,12 @@ ClusterTPAssociationProducerCUDA::ClusterTPAssociationProducerCUDA(const edm::Pa
           cfg.getParameter<edm::InputTag>("phase2OTClusterSrc"))),
       trackingParticleToken_(
           consumes<TrackingParticleCollection>(cfg.getParameter<edm::InputTag>("trackingParticleSrc"))),
-      tGpuDigis(
-          consumes<CUDAProduct<SiPixelDigisCUDA>>(cfg.getParameter<edm::InputTag>("heterogeneousPixelDigiClusterSrc"))),
-      tGpuHits(
-          consumes<CUDAProduct<TrackingRecHit2DCUDA>>(cfg.getParameter<edm::InputTag>("heterogeneousPixelRecHitSrc"))),
+      tGpuDigis(consumes<cms::cuda::Product<SiPixelDigisCUDA>>(
+          cfg.getParameter<edm::InputTag>("heterogeneousPixelDigiClusterSrc"))),
+      tGpuHits(consumes<cms::cuda::Product<TrackingRecHit2DCUDA>>(
+          cfg.getParameter<edm::InputTag>("heterogeneousPixelRecHitSrc"))),
       m_gpuAlgo(cfg.getParameter<bool>("dumpCSV")) {
-  tokenGPUProd_ = produces<CUDAProduct<ProductCUDA>>();
+  tokenGPUProd_ = produces<cms::cuda::Product<ProductCUDA>>();
 }
 
 void ClusterTPAssociationProducerCUDA::fillDescriptions(edm::ConfigurationDescriptions &descriptions) {
@@ -159,12 +159,12 @@ void ClusterTPAssociationProducerCUDA::produce(edm::StreamID streamID,
 
   auto mapping = makeMap(iEvent);
 
-  edm::Handle<CUDAProduct<SiPixelDigisCUDA>> gd;
+  edm::Handle<cms::cuda::Product<SiPixelDigisCUDA>> gd;
   iEvent.getByToken(tGpuDigis, gd);
-  edm::Handle<CUDAProduct<TrackingRecHit2DCUDA>> gh;
+  edm::Handle<cms::cuda::Product<TrackingRecHit2DCUDA>> gh;
   iEvent.getByToken(tGpuHits, gh);
 
-  CUDAScopedContextProduce ctx{*gd};
+  cms::cuda::ScopedContextProduce ctx{*gd};
   auto const &gDigis = ctx.get(*gd);
   auto const &gHits = ctx.get(*gh);
   auto ndigis = gDigis.nDigis();

diff --git a/SimTracker/TrackerHitAssociation/src/classes.h b/SimTracker/TrackerHitAssociation/src/classes.h
@@ -5,7 +5,7 @@
 #include "DataFormats/Common/interface/AssociationMap.h"
 #include "DataFormats/TrackerRecHit2D/interface/OmniClusterRef.h"
 #include "SimTracker/TrackerHitAssociation/interface/ClusterTPAssociation.h"
-#include "CUDADataFormats/Common/interface/CUDAProduct.h"
+#include "CUDADataFormats/Common/interface/Product.h"
 #include "SimTracker/TrackerHitAssociation/interface/trackerHitAssociationHeterogeneous.h"
 
 #include "DataFormats/Common/interface/AssociationMap.h"

diff --git a/SimTracker/TrackerHitAssociation/src/classes_def.xml b/SimTracker/TrackerHitAssociation/src/classes_def.xml
@@ -22,7 +22,7 @@
   <class name="edm::Wrapper<std::map<OmniClusterRef, std::vector<TrackingParticleRef> > >" persistent="false" /> 
 
   <class name="trackerHitAssociationHeterogeneous::ProductCUDA" persistent="false"/>
-  <class name="CUDAProduct<trackerHitAssociationHeterogeneous::ProductCUDA>" persistent="false"/>
-  <class name="edm::Wrapper<CUDAProduct<trackerHitAssociationHeterogeneous::ProductCUDA>>" persistent="false"/>
+  <class name="cms::cuda::Product<trackerHitAssociationHeterogeneous::ProductCUDA>" persistent="false"/>
+  <class name="edm::Wrapper<cms::cuda::Product<trackerHitAssociationHeterogeneous::ProductCUDA>>" persistent="false"/>
 
 </lcgdict>
diff --git a/SimTracker/TrackerHitAssociation/test/ClusterTPCUDAdump.cc b/SimTracker/TrackerHitAssociation/test/ClusterTPCUDAdump.cc
@@ -1,6 +1,6 @@
 #include <cuda_runtime.h>
 
-#include "CUDADataFormats/Common/interface/CUDAProduct.h"
+#include "CUDADataFormats/Common/interface/Product.h"
 #include "DataFormats/Common/interface/Handle.h"
 #include "FWCore/Framework/interface/ConsumesCollector.h"
 #include "FWCore/Framework/interface/ESHandle.h"
@@ -16,7 +16,7 @@
 #include "FWCore/Utilities/interface/EDGetToken.h"
 #include "FWCore/Utilities/interface/InputTag.h"
 #include "FWCore/Utilities/interface/RunningAverage.h"
-#include "HeterogeneousCore/CUDACore/interface/CUDAScopedContext.h"
+#include "HeterogeneousCore/CUDACore/interface/ScopedContext.h"
 #include "HeterogeneousCore/CUDAServices/interface/CUDAService.h"
 #include "SimTracker/TrackerHitAssociation/interface/trackerHitAssociationHeterogeneous.h"
 
@@ -34,20 +34,20 @@ class ClusterTPCUDAdump : public edm::global::EDAnalyzer<> {
 private:
   void analyze(edm::StreamID streamID, edm::Event const& iEvent, const edm::EventSetup& iSetup) const override;
   const bool m_onGPU;
-  edm::EDGetTokenT<CUDAProduct<ProductCUDA>> tokenGPU_;
+  edm::EDGetTokenT<cms::cuda::Product<ProductCUDA>> tokenGPU_;
 };
 
 ClusterTPCUDAdump::ClusterTPCUDAdump(const edm::ParameterSet& iConfig) : m_onGPU(iConfig.getParameter<bool>("onGPU")) {
   if (m_onGPU) {
-    tokenGPU_ = consumes<CUDAProduct<ProductCUDA>>(iConfig.getParameter<edm::InputTag>("clusterTP"));
+    tokenGPU_ = consumes<cms::cuda::Product<ProductCUDA>>(iConfig.getParameter<edm::InputTag>("clusterTP"));
   } else {
   }
 }
 
 void ClusterTPCUDAdump::analyze(edm::StreamID streamID, edm::Event const& iEvent, const edm::EventSetup& iSetup) const {
   if (m_onGPU) {
     auto const& hctp = iEvent.get(tokenGPU_);
-    CUDAScopedContextProduce ctx{hctp};
+    cms::cuda::ScopedContextProduce ctx{hctp};
 
     auto const& ctp = ctx.get(hctp);
     auto const& soa = ctp.view();