refactor PMacc event system

src/libPMacc/examples/gameOfLife2D/include/Simulation.hpp

@@ -145,10 +145,10 @@ class Simulation
         // in units of SuperCells to be used by the kernel to identify itself.
         evo.init(
             MappingDesc(layout.getDataSpace(), 1, 1));
-            
+
         buff1.reset(new PMacc::GridBuffer<std::uint8_t, DIM2>(layout, false));
         buff2.reset(new PMacc::GridBuffer<std::uint8_t, DIM2>(layout, false));
-            
+
         PMacc::DataSpace<DIM2> guardingCells(1, 1);
         for (uint32_t i(1); i < PMacc::traits::NumberOfExchanges<DIM2>::value; ++i)
         {
@@ -161,7 +161,7 @@ class Simulation
             buff1->addExchange(PMacc::GUARD, PMacc::Mask(i), guardingCells, BUFF1);
             buff2->addExchange(PMacc::GUARD, PMacc::Mask(i), guardingCells, BUFF2);
         }
-            
+
         // Both next lines are defined in GatherSlice.hpp:
         // -gather saves the MessageHeader object
         // -Then do an Allgather for the gloabalRanks from GC, sort out
@@ -236,7 +236,7 @@ class Simulation
         evo.run<PMacc::BORDER>(
             read.getDeviceBuffer().getDataBox(),
             write.getDeviceBuffer().getDataBox());
-            
+
         // Copy from device to host for saving. All threads and not only the master have to do this.
         write.deviceToHost();
 
@@ -264,7 +264,7 @@ class Simulation
             writeFullImage(write, sFileNameWithoutExt);
         }
     }
-        
+
     //-----------------------------------------------------------------------------
     //!
     //-----------------------------------------------------------------------------

src/libPMacc/include/eventSystem/events/CudaEvent.hpp

@@ -41,25 +41,12 @@ class CudaEvent
      * no data is allocated @see create()
      */
     CudaEvent() :
-        m_event(),
-        m_pStream(),
-        isRecorded(false)
+        m_event(NULL),
+        m_pStream(NULL),
+        isRecorded(false),
+        isValid(false)
     {}
 
-    /**
-     * Copy constructor
-     */
-    CudaEvent(CudaEvent const & other) :
-        m_event(
-            new alpaka::event::Event<AlpakaAccStream>(*other.m_event.get())),
-        m_pStream(other.m_pStream),
-        isRecorded(other.isRecorded)
-    {}
-
-    /**
-     * Move constructor
-     */
-    CudaEvent(CudaEvent && other) = default;
 
     /**
      * Destructor
@@ -80,7 +67,8 @@ class CudaEvent
     static CudaEvent create(AlpakaAccDev const & dev)
     {
         CudaEvent ev;
-        ev.m_event.reset(new alpaka::event::Event<AlpakaAccStream>(dev));
+        ev.m_event = new alpaka::event::Event<AlpakaAccStream>(dev);
+        ev.isValid = true;
         return ev;
     }
 
@@ -89,19 +77,21 @@ class CudaEvent
      */
     static void destroy(CudaEvent& ev)
     {
-        alpaka::wait::wait(*ev.m_event.get());
-        ev.m_event.reset();
+        alpaka::wait::wait(*ev.m_event);
+        ev.isValid = false;
+        delete ev.m_event;
+        ev.m_event = NULL;
     }
 
     /**
      * get native cuda event
      *
      * @return native cuda event
      */
-    alpaka::event::Event<AlpakaAccStream> & operator*() const
+    alpaka::event::Event<AlpakaAccStream> operator*() const
     {
-        assert(m_event);
-        return *m_event.get();
+        assert(isValid);
+        return *m_event;
     }
 
     /**
@@ -111,16 +101,16 @@ class CudaEvent
      */
     bool isFinished() const
     {
-        assert(m_event);
-        return alpaka::event::test(*m_event.get());
+        assert(isValid);
+        return alpaka::event::test(*m_event);
     }
 
     /**
      * get stream in which this event is recorded
      *
      * @return native cuda stream
      */
-    AlpakaAccStream & getCudaStream() const
+    AlpakaAccStream& getCudaStream() const
     {
         assert(isRecorded);
         assert(m_pStream);
@@ -132,19 +122,20 @@ class CudaEvent
      *
      * @param stream native cuda stream
      */
-    void recordEvent(AlpakaAccStream & stream)
+    void recordEvent( AlpakaAccStream* stream)
     {
         /* disallow double recording */
         assert(isRecorded==false);
         isRecorded = true;
-        m_pStream = &stream;
-        alpaka::stream::enqueue(*m_pStream, *m_event.get());
+        m_pStream = stream;
+        alpaka::stream::enqueue(*m_pStream, *m_event);
     }
 
 private:
-    std::unique_ptr<alpaka::event::Event<AlpakaAccStream>> m_event;
-    AlpakaAccStream * m_pStream;
+    alpaka::event::Event<AlpakaAccStream>* m_event;
+    AlpakaAccStream* m_pStream;
     /* state if event is recorded */
     bool isRecorded;
+    bool isValid;
 };
 }

src/libPMacc/include/eventSystem/streams/EventStream.hpp

@@ -43,7 +43,7 @@ class EventStream
      * Creates the cudaStream_t object.
      */
     EventStream(alpaka::dev::Dev<AlpakaAccDev> dev) :
-        stream(dev)
+        stream(new AlpakaAccStream(dev))
     {}
 
     /**
@@ -53,28 +53,28 @@ class EventStream
     virtual ~EventStream()
     {
         //wait for all kernels in stream to finish
-        alpaka::wait::wait(stream);
+        alpaka::wait::wait(*stream);
     }
 
     /**
      * Returns the cudaStream_t object associated with this EventStream.
      * @return the internal cuda stream object
      */
-    AlpakaAccStream & getCudaStream()
+    AlpakaAccStream& getCudaStream()
     {
-        return stream;
+        return *stream;
     }
 
     void waitOn(const CudaEvent& ev)
     {
         if(getCudaStream() != ev.getCudaStream())
         {
-            alpaka::wait::wait(stream, *ev);
+            alpaka::wait::wait(*stream, *ev);
         }
     }
 
 private:
-    AlpakaAccStream stream;
+    AlpakaAccStream* stream;
 };
 
 }

src/libPMacc/include/eventSystem/tasks/StreamTask.hpp

@@ -98,8 +98,9 @@ namespace PMacc
 
     private:
         mutable EventStream* stream;
-        std::unique_ptr<CudaEvent> cudaEvent;
+        CudaEvent cudaEvent;
         bool alwaysFinished;
+        bool hasCudaEvent;
     };
 
 } //namespace PMacc

src/libPMacc/include/eventSystem/tasks/StreamTask.tpp

@@ -34,30 +34,31 @@ namespace PMacc
 inline StreamTask::StreamTask( ) :
 ITask( ),
 stream( NULL ),
-cudaEvent( ),
-alwaysFinished( false )
+alwaysFinished( false ),
+hasCudaEvent(false)
 {
     this->setTaskType( TASK_CUDA );
 }
 
 inline CudaEvent StreamTask::getCudaEvent( ) const
 {
-    assert(cudaEvent);
-    return *cudaEvent.get();
+    assert(hasCudaEvent);
+    return cudaEvent;
 }
 
 inline void StreamTask::setCudaEvent(const CudaEvent& cudaEvent )
 {
-    this->cudaEvent.reset(new CudaEvent(cudaEvent));
+    this->hasCudaEvent = true;
+    this->cudaEvent = cudaEvent;
 }
 
 inline bool StreamTask::isFinished( )
 {
     if ( alwaysFinished )
         return true;
-    if(cudaEvent)
+    if(hasCudaEvent)
     {
-        if ( cudaEvent->isFinished( ) )
+        if ( cudaEvent.isFinished( ) )
         {
             alwaysFinished = true;
             return true;
@@ -83,8 +84,9 @@ inline void StreamTask::setEventStream( EventStream* newStream )
 
 inline void StreamTask::activate( )
 {
-    cudaEvent.reset(new CudaEvent(Environment<>::get().Manager().getEventPool().getNextEvent()));
-    cudaEvent->recordEvent(this->stream->getCudaStream());
+    cudaEvent = Environment<>::get().Manager().getEventPool().getNextEvent();
+    cudaEvent.recordEvent(&(getEventStream()->getCudaStream()));
+    hasCudaEvent = true;
 }
 
 } //namespace PMacc

src/libPMacc/include/particles/memory/boxes/ParticlesBox.hpp

@@ -80,14 +80,15 @@ class ParticlesBox : protected DataBox<PitchedBox<SuperCell<FRAME>, DIM> >
      *
      * @return an empty frame
      */
+    PMACC_NO_NVCC_HDWARNING
     DINLINE FRAME &getEmptyFrame() const
     {
 
         FrameType* tmp = NULL;
         const int maxTries = 13; //magic number is not performance critical
         for (int numTries = 0; numTries < maxTries; ++numTries)
         {
-            tmp = (FrameType*) mallocMC::malloc(sizeof (FrameType));
+            tmp = (FrameType*) ::mallocMC::malloc(sizeof (FrameType));
             if (tmp != NULL)
             {
                 /* disable all particles since we can not assume that newly allocated memory contains zeros */
@@ -119,7 +120,7 @@ class ParticlesBox : protected DataBox<PitchedBox<SuperCell<FRAME>, DIM> >
      */
     DINLINE void removeFrame(FRAME &frame) const
     {
-        mallocMC::free((void*) &frame);
+        ::mallocMC::free((void*) &frame);
     }
 
     HDINLINE

src/libPMacc/include/types.h

@@ -51,11 +51,12 @@ namespace PMacc
     using AlpakaHostDev = alpaka::dev::DevCpu;
 #ifdef PMACC_ACC_CPU
     using AlpakaAccDev = alpaka::dev::DevCpu;
-    //using AlpakaAccStream = alpaka::stream::StreamCpuAsync;
-    using AlpakaAccStream = alpaka::stream::StreamCpuSync;
+    using AlpakaAccStream = alpaka::stream::StreamCpuAsync;
+    //using AlpakaAccStream = alpaka::stream::StreamCpuSync;
     template<
         typename TDim>
     using AlpakaAcc = alpaka::acc::AccCpuOmp2Threads<TDim, AlpakaIdxSize>;
+    //using AlpakaAcc = alpaka::acc::AccCpuOmp2Blocks<TDim, AlpakaIdxSize>;
 #else
     using AlpakaAccDev = alpaka::dev::DevCudaRt;
     using AlpakaAccStream = alpaka::stream::StreamCudaRtAsync;

src/picongpu/include/particles/manipulators/RandomPositionImpl.hpp

@@ -50,6 +50,7 @@ struct RandomPositionImpl
         localCells = subGrid.getLocalDomain().size;
     }
 
+    PMACC_NO_NVCC_HDWARNING
     template<
         typename T_Acc,
         typename T_Particle1,

src/picongpu/include/particles/manipulators/TemperatureImpl.hpp

@@ -58,6 +58,7 @@ struct TemperatureImpl : private T_ValueFunctor
         localCells = subGrid.getLocalDomain().size;
     }
 
+    PMACC_NO_NVCC_HDWARNING
     template<
         typename T_Acc,
         typename T_Particle1,

src/picongpu/include/particles/startPosition/RandomImpl.hpp

@@ -77,6 +77,7 @@ struct RandomImpl
      * @param curParticle the number of this particle: [0, totalNumParsPerCell-1]
      * @return float3_X with components between [0.0, 1.0)
      */
+    PMACC_NO_NVCC_HDWARNING
     DINLINE floatD_X operator()(const uint32_t)
     {
         floatD_X result;

thirdParty/mallocMC/src/include/mallocMC/DistributionPolicies.hpp

@@ -34,7 +34,6 @@
 
 #if defined(MAMC_CUDA_ENABLED) && defined(__CUDACC__)
 
-
 #include "distributionPolicies/XMallocSIMD.hpp"
 #include "distributionPolicies/XMallocSIMD_impl.hpp"
 

thirdParty/mallocMC/src/include/mallocMC/ReservePoolPolicies.hpp

@@ -34,7 +34,6 @@
 
 #if defined(MAMC_CUDA_ENABLED) && defined(__CUDACC__)
 
-
 #include "reservePoolPolicies/SimpleCudaMalloc.hpp"
 #include "reservePoolPolicies/SimpleCudaMalloc_impl.hpp"
 

thirdParty/mallocMC/src/include/mallocMC/alignmentPolicies/Noop_impl.hpp

@@ -41,15 +41,17 @@ namespace AlignmentPolicies{
 
     public:
 
+    MAMC_HOST
     static boost::tuple<void*,size_t> alignPool(void* memory, size_t memsize){
       return boost::make_tuple(memory,memsize);
     }
 
-    MAMC_HOST MAMC_ACCELERATOR
+    MAMC_ACC
     static uint32 applyPadding(uint32 bytes){
       return bytes;
     }
 
+    MAMC_HOST
     static std::string classname(){
       return "Noop";
     }

thirdParty/mallocMC/src/include/mallocMC/alignmentPolicies/Shrink_impl.hpp

@@ -79,6 +79,7 @@ namespace Shrink2NS{
     BOOST_STATIC_ASSERT(dataAlignment && !(dataAlignment & (dataAlignment-1)) ); 
 
     public:
+    MAMC_HOST
     static boost::tuple<void*,size_t> alignPool(void* memory, size_t memsize){
       PointerEquivalent alignmentstatus = ((PointerEquivalent)memory) & (dataAlignment -1);
       if(alignmentstatus != 0)
@@ -102,8 +103,7 @@ namespace Shrink2NS{
       return boost::make_tuple(memory,memsize);
     }
 
-    MAMC_HOST
-    MAMC_ACCELERATOR
+    MAMC_ACC
     static uint32 applyPadding(uint32 bytes){
       return (bytes + dataAlignment - 1) & ~(dataAlignment-1);
     }