Overhaul for multi-threading. Needs some testing to validate, but first signs are good.

input.txt

@@ -1,5 +1,6 @@
 ----------Data Information----------
 OutputFile: /media/data/gwm17/mask_tests/10B3Hea_16800keV_5Lia_74B.root
+NumberOfThreads: 6
 ----------Reaction Information----------
 NumberOfSamples: 100000
 begin_chain

src/Kinematics/main.cpp

@@ -24,6 +24,7 @@ int main(int argc, char** argv)
 			return 1;
 		}
 		calculator.Run();
+		//calculator.RunSingleThread();
 	}
 	catch(const std::exception& e)
 	{

src/Mask/CMakeLists.txt

@@ -50,7 +50,13 @@ target_sources(Mask PRIVATE
     ThreeStepSystem.h
     TwoStepSystem.cpp
     TwoStepSystem.h
+    ThreadPool.h
+    FileWriter.h
+    FileWriter.cpp
 )
 
-target_link_libraries(Mask catima MaskDict ${ROOT_LIBRARIES})
+set(THREADS_PREFER_PTHREAD_FLAG On)
+find_package(Threads REQUIRED)
+
+target_link_libraries(Mask catima MaskDict ${ROOT_LIBRARIES} Threads::Threads)
 set_target_properties(Mask PROPERTIES ARCHIVE_OUTPUT_DIRECTORY ${MASK_LIBRARY_DIR})

src/Mask/FileWriter.cpp

@@ -0,0 +1,75 @@
+#include "FileWriter.h"
+
+namespace Mask {
+
+    FileWriter::FileWriter() :
+        m_file(nullptr), m_tree(nullptr), m_queueSize(0)
+    {
+    }
+
+    FileWriter::FileWriter(const std::string& filename, const std::string& treename) :
+        m_file(nullptr), m_tree(nullptr)
+    {
+        m_file = TFile::Open(filename.c_str(), "RECREATE");
+        if(m_file != nullptr && m_file->IsOpen())
+        {
+            m_tree = new TTree(treename.c_str(), treename.c_str());
+            m_tree->Branch("nuclei", &m_dataHandle);
+        }
+    }
+
+    FileWriter::~FileWriter()
+    {
+        Close();
+    }
+
+    void FileWriter::Open(const std::string& filename, const std::string& treename)
+    {
+        if(m_file != nullptr || m_tree != nullptr)
+            Close();
+
+        m_file = TFile::Open(filename.c_str(), "RECREATE");
+        if(m_file != nullptr && m_file->IsOpen())
+        {
+            m_tree = new TTree(treename.c_str(), treename.c_str());
+            m_tree->Branch("nuclei", &m_dataHandle);
+        }
+    }
+
+    void FileWriter::Close()
+    {
+        if(m_file->IsOpen())
+        {
+            if(m_tree != nullptr)
+                m_tree->Write(m_tree->GetName(), TObject::kOverwrite);
+
+            m_file->Close();
+            delete m_file;
+            m_file = nullptr;
+        }
+    }
+
+    void FileWriter::PushData(const std::vector<Nucleus>& data)
+    {
+        std::scoped_lock<std::mutex> guard(m_queueMutex);
+        m_queue.push(data);
+        ++m_queueSize;
+    }
+
+    bool FileWriter::Write()
+    {
+        if(m_queueSize == 0)
+            return false;
+
+        //Aquire lock for as short a time as possible
+        {
+            std::scoped_lock<std::mutex> guard(m_queueMutex);
+            m_dataHandle = m_queue.front();
+            m_queue.pop();
+        }
+        --m_queueSize;
+
+        m_tree->Fill();
+        return true;
+    }
+}

src/Mask/FileWriter.h

@@ -0,0 +1,45 @@
+#ifndef FILE_WRITER_H
+#define FILE_WRITER_H
+
+#include "Nucleus.h"
+
+#include "TFile.h"
+#include "TTree.h"
+
+#include <queue>
+#include <mutex>
+#include <atomic>
+
+namespace Mask {
+
+    class FileWriter
+    {
+    public:
+        FileWriter();
+        FileWriter(const std::string& filename, const std::string& treename);
+        ~FileWriter();
+
+        bool IsOpen() const { return m_file->IsOpen(); }
+        bool IsTree() const { return m_tree == nullptr ? false : true; }
+
+        std::size_t GetQueueSize() const { return m_queueSize; } //Implicitly thread-safe
+
+        void PushData(const std::vector<Nucleus>& data); //Thread-safe
+        bool Write(); //Not thread safe!
+
+        void Open(const std::string& filename, const std::string& treename);
+        void Close(); //Not thread safe!
+
+    private:
+        TFile* m_file;
+        TTree* m_tree;
+
+        std::vector<Nucleus> m_dataHandle;
+
+        std::mutex m_queueMutex;
+        std::atomic<std::size_t> m_queueSize;
+        std::queue<std::vector<Nucleus>> m_queue;
+    };
+}
+
+#endif

src/Mask/MaskApp.cpp

@@ -8,7 +8,7 @@
 namespace Mask {
 
 	MaskApp::MaskApp() :
-		m_system(nullptr)
+		m_system(nullptr), m_resources(nullptr)
 	{
 		std::cout<<"----------Monte Carlo Simulation of Kinematics----------"<<std::endl;
 	}
@@ -16,6 +16,8 @@ namespace Mask {
 	MaskApp::~MaskApp() 
 	{
 		delete m_system;
+		for(std::size_t i=0; i<m_systemList.size(); i++)
+			delete m_systemList[i];
 	}
 	
 	bool MaskApp::LoadConfig(const std::string& filename) 
@@ -31,10 +33,18 @@ namespace Mask {
 		std::string junk;
 		std::getline(input, junk);
 		input>>junk>>m_outputName;
+		input>>junk>>m_nthreads;
 		std::getline(input, junk);
 		std::getline(input, junk);
 		input>>junk>>m_nsamples;
 
+		std::cout<<"Allocating resources... Asking for " << m_nthreads << " threads...";
+		m_resources = std::make_unique<ThreadPool>(m_nthreads);
+		std::cout<<" Complete."<<std::endl;
+
+		std::cout<<"Outputing data to file: " << m_outputName <<std::endl;
+		m_fileWriter.Open(m_outputName, "SimTree");
+
 		std::vector<StepParameters> params;
 		int z, a;
 		while(input>>junk)
@@ -98,10 +108,20 @@ namespace Mask {
 		m_system = CreateSystem(params);
 		if(m_system == nullptr || !m_system->IsValid())
 		{
-			std::cerr<<"Param size: "<<params.size()<<std::endl;
 			std::cerr<<"Failure to parse reaction system... configuration not loaded."<<std::endl;
 			return false;
 		}
+
+		for(uint32_t i=0; i<m_nthreads; i++)
+		{
+			m_systemList.push_back(CreateSystem(params));
+			if(m_systemList.back() == nullptr || !m_systemList.back()->IsValid())
+			{
+				std::cerr<<"Failure to parse reaction system... configuration not loaded."<<std::endl;
+				return false;
+			}
+		}
+
 		std::getline(input, junk);
 		std::getline(input, junk);
 
@@ -132,9 +152,13 @@ namespace Mask {
 		}
 		m_system->SetLayeredTarget(target);
 
-		std::cout<<"Outputing data to file: "<<m_outputName<<std::endl;
-		std::cout<<"Reaction equation: "<<GetSystemName()<<std::endl;
-		std::cout<<"Number of samples: "<<GetNumberOfSamples()<<std::endl;
+		for(auto system : m_systemList)
+		{
+			system->SetLayeredTarget(target);
+		}
+
+		std::cout<<"Reaction equation: "<<m_system->GetSystemEquation()<<std::endl;
+		std::cout<<"Number of samples: "<<m_nsamples<<std::endl;
 	
 		return true;
 	}
@@ -146,6 +170,42 @@ namespace Mask {
 	}
 
 	void MaskApp::Run()
+	{
+		std::cout<<"Running simulation..."<<std::endl;
+		if(m_systemList.size() != m_nthreads)
+		{
+			return;
+		}
+
+		//Give our thread pool some tasks
+		for(auto system : m_systemList)
+			m_resources->PushJob({std::bind(&MaskApp::RunChunk, std::ref(*this), std::placeholders::_1), system});
+
+		uint64_t count = 0;
+		double percent = 0.05;
+		uint64_t flushVal = m_nsamples*percent;
+		uint64_t flushCount = 0;
+		while(true)
+		{
+			if(count == flushVal)
+			{
+				count = 0;
+				++flushCount;
+				std::cout<<"\rPercent of data written to disk: "<<percent*flushCount*100<<"%"<<std::flush;
+			}
+
+			if(m_resources->IsFinished() && m_fileWriter.GetQueueSize() == 0)
+				break;
+			else if(m_fileWriter.Write())
+				++count;
+		}
+
+		std::cout<<std::endl;
+		std::cout<<"Complete."<<std::endl;
+		std::cout<<"---------------------------------------------"<<std::endl;
+	}
+	
+	void MaskApp::RunSingleThread()
 	{
 		std::cout<<"Running simulation..."<<std::endl;
 		if(m_system == nullptr) 
@@ -158,11 +218,11 @@ namespace Mask {
 		tree->Branch("nuclei", m_system->GetNuclei());
 	
 		//For progress tracking
-		uint32_t percent5 = 0.05*m_nsamples;
-		uint32_t count = 0;
-		uint32_t npercent = 0;
+		uint64_t percent5 = 0.05*m_nsamples;
+		uint64_t count = 0;
+		uint64_t npercent = 0;
 	
-		for(uint32_t i=0; i<m_nsamples; i++) 
+		for(uint64_t i=0; i<m_nsamples; i++) 
 		{
 			if(++count == percent5) 
 			{
@@ -183,4 +243,18 @@ namespace Mask {
 		std::cout<<"---------------------------------------------"<<std::endl;
 	}
 
+	void MaskApp::RunChunk(ReactionSystem* system)
+	{
+		if(system == nullptr)
+			return;
+
+		uint64_t samples = m_nsamples / m_nthreads;
+
+		for(uint64_t i=0; i<samples; i++)
+		{
+			system->RunSystem();
+			m_fileWriter.PushData(*(system->GetNuclei()));
+		}
+	}
+
 }

src/Mask/MaskApp.h

@@ -7,6 +7,10 @@
 #include "TwoStepSystem.h"
 #include "ThreeStepSystem.h"
 #include "RxnType.h"
+#include "ThreadPool.h"
+#include "FileWriter.h"
+
+#include <memory>
 
 namespace Mask {
 
@@ -17,18 +21,23 @@ namespace Mask {
 		~MaskApp();
 		bool LoadConfig(const std::string& filename);
 		bool SaveConfig(const std::string& filename);
-		int GetNumberOfSamples() const { return m_nsamples; }
-		const std::string GetSystemName() const { return m_system == nullptr ? "Invalid System" : m_system->GetSystemEquation(); }
-		const std::string GetOutputName() const { return m_outputName; }
-		const RxnType GetReactionType() const { return m_rxnType; }
 
 		void Run();
+		void RunSingleThread();
 
 	private:
+		void RunChunk(ReactionSystem* system);
+
 		ReactionSystem* m_system;
+
 		std::string m_outputName;
 		RxnType m_rxnType;
-		int m_nsamples;
+		uint64_t m_nsamples;
+		uint32_t m_nthreads;
+
+		std::vector<ReactionSystem*> m_systemList; //One system for each thread
+		FileWriter m_fileWriter;
+		std::unique_ptr<ThreadPool> m_resources;
 	
 	};
 

src/Mask/RandomGenerator.cpp

@@ -1,7 +1,6 @@
 #include "RandomGenerator.h"
 
 namespace Mask {
-	RandomGenerator* RandomGenerator::s_instance = new RandomGenerator();
 	
 	RandomGenerator::RandomGenerator()
 	{

src/Mask/RandomGenerator.h

@@ -10,12 +10,15 @@ namespace Mask {
 	public:
 		~RandomGenerator();
 		std::mt19937& GetGenerator() { return rng; }
-		static RandomGenerator& GetInstance() { return *s_instance; }
+		static RandomGenerator& GetInstance()
+		{
+			static thread_local RandomGenerator s_instance;
+			return s_instance;
+		}
 
 	private:
 		RandomGenerator();
 		
-		static RandomGenerator* s_instance;
 		std::mt19937 rng;
 	};
 

src/Mask/ThreadPool.h

@@ -0,0 +1,115 @@
+/*
+    ThreadPool.h
+    Simple thread pool implementation. Jobs are q'd, and as threads are made available, jobs are dispatched to these threads.
+
+    GWM -- Aug 2022
+*/
+#ifndef THREAD_POOL_H
+#define THREAD_POOL_H
+
+#include "ReactionSystem.h"
+
+#include <vector>
+#include <atomic>
+#include <thread>
+#include <mutex>
+#include <functional>
+#include <queue>
+#include <condition_variable>
+
+
+namespace Mask {
+
+	using JobFunction = std::function<void(ReactionSystem*)>;
+	struct Job
+	{
+		JobFunction func;
+		ReactionSystem* argument;
+	};
+
+	class ThreadPool
+	{
+	public:
+		ThreadPool(uint32_t nthreads) :
+			m_isStopped(false), m_numberRunning(0), m_queueSize(0), m_initShutdown(false)
+		{
+			for(uint32_t i=0; i<nthreads; i++)
+			{
+				m_pool.push_back(std::thread(std::bind(&ThreadPool::ExecuteJob, std::ref(*this))));
+			}
+		}
+		~ThreadPool()
+		{
+			if(!m_isStopped)
+				Shutdown();
+		}
+
+		void PushJob(Job job)
+		{
+			{
+				std::unique_lock<std::mutex> guard(m_poolMutex);
+				m_queue.push(job);
+				m_queueSize++;
+			}
+			m_wakeCondition.notify_one();
+		}
+
+		bool IsFinished()
+		{
+			return m_numberRunning == 0 && m_queueSize == 0;
+		}
+
+		void Shutdown()
+		{
+			m_initShutdown = true;
+			m_wakeCondition.notify_all();
+			for(auto& thread : m_pool)
+			{
+				thread.join();
+			}
+
+			m_pool.clear();
+			m_isStopped = true;
+		}
+
+	private:
+		void ExecuteJob()
+		{
+			while(true)
+			{
+				Job job;
+				{
+					std::unique_lock<std::mutex> guard(m_poolMutex);
+					m_wakeCondition.wait(guard, [this](){ 
+						return (!m_queue.empty() || m_initShutdown); 
+					});
+					if(m_initShutdown)
+						return;
+	
+					job = m_queue.front();
+					m_queue.pop();
+				}
+
+				//Change number running first so that no crash 
+				m_numberRunning++;
+				m_queueSize--;
+
+				job.func(job.argument);
+				m_numberRunning--;
+			}
+			
+		}
+
+		std::queue<Job, std::deque<Job>> m_queue;
+		std::vector<std::thread> m_pool;
+		std::mutex m_poolMutex;
+		std::condition_variable m_wakeCondition;
+
+		bool m_isStopped;
+		std::atomic<int> m_numberRunning;
+		std::atomic<int> m_queueSize;
+		std::atomic<bool> m_initShutdown;
+	};
+}
+
+#endif