Fixing tests ... (7dc4b071) · Commits · ORNL Quantum Computing Institute / exatn

src/exatn/exatn_numerics.hpp

+20 −5

Original line number	Diff line number	Diff line
		/** ExaTN::Numerics: General client header (free function API)
		REVISION: 2022/01/17
		REVISION: 2022/01/25

		Copyright (C) 2018-2022 Dmitry I. Lyakh (Liakh)
		Copyright (C) 2018-2022 Oak Ridge National Laboratory (UT-Battelle) **/
		@@ -417,6 +417,13 @@ inline bool destroyTensors(TensorNetwork & tensor_network) //inout: tensor n
		inline bool destroyTensorsSync(TensorNetwork & tensor_network) //inout: tensor network
		{return numericalServer->destroyTensorsSync(tensor_network);}

		/ Destroys all currently allocated tensors. /
		inline bool destroyTensors()
		{return numericalServer->destroyTensors();}

		inline bool destroyTensorsSync()
		{return numericalServer->destroyTensorsSync();}


		/ Initializes a tensor to some scalar value. /
		template<typename NumericType>
		@@ -901,12 +908,20 @@ inline bool sync(const ProcessGroup & process_group, //in: chosen group of MPI p

		/** Synchronizes all outstanding tensor operations in the current scope (barrier).
		If ProcessGroup is not provided, defaults to the local process. **/
		inline bool sync(bool wait = true) //in: wait versus test for completion
		{return numericalServer->sync(wait);}
		inline bool sync(bool wait = true, //in: wait versus test for completion
		bool clean_garbage = false) //in: activates garbage collection
		{return numericalServer->sync(wait,clean_garbage);}

		inline bool sync(const ProcessGroup & process_group, //in: chosen group of MPI processes
		bool wait = true) //in: wait versus test for completion
		{return numericalServer->sync(process_group,wait);}
		bool wait = true, //in: wait versus test for completion
		bool clean_garbage = false) //in: activates garbage collection
		{return numericalServer->sync(process_group,wait,clean_garbage);}

		inline bool syncClean()
		{return numericalServer->sync(true,true);}

		inline bool syncClean(const ProcessGroup & process_group)
		{return numericalServer->sync(process_group,true,true);}


		/ Normalizes a tensor to a given 2-norm. /

src/exatn/num_server.cpp

+43 −9

Original line number	Diff line number	Diff line
		/** ExaTN::Numerics: Numerical server
		REVISION: 2022/01/17
		REVISION: 2022/01/25

		Copyright (C) 2018-2022 Dmitry I. Lyakh (Liakh)
		Copyright (C) 2018-2022 Oak Ridge National Laboratory (UT-Battelle) **/
		@@ -1186,19 +1186,19 @@ bool NumServer::sync(const ProcessGroup & process_group, TensorNetwork & network
		return sync(process_group,*(network.getTensor(0)),wait); //synchronization on the output tensor of the tensor network
		}

		bool NumServer::sync(bool wait)
		bool NumServer::sync(bool wait, bool clean_garbage)
		{
		bool success = sync(getCurrentProcessGroup(),wait);
		bool success = sync(getCurrentProcessGroup(),wait,clean_garbage);
		#ifdef CUQUANTUM
		if(comp_backend_ == "cuquantum" && success) tn_exec_handles_.clear();
		#endif
		return success;
		}

		bool NumServer::sync(const ProcessGroup & process_group, bool wait)
		bool NumServer::sync(const ProcessGroup & process_group, bool wait, bool clean_garbage)
		{
		if(!process_group.rankIsIn(process_rank_)) return true; //process is not in the group: Do nothing
		destroyOrphanedTensors(); //garbage collection
		destroyOrphanedTensors(clean_garbage); //garbage collection
		auto success = tensor_rt_->sync(wait);
		if(success){
		if(logging_ > 0) logfile_ << "[" << std::fixed << std::setprecision(6) << exatn::Timer::timeInSecHR(getTimeStampStart())
		@@ -1628,6 +1628,26 @@ bool NumServer::destroyTensorsSync(TensorNetwork & tensor_network)
		return success;
		}

		bool NumServer::destroyTensors()
		{
		bool success = true;
		while(!tensors_.empty()){
		success = destroyTensor(tensors_.begin()->first);
		if(!success) break;
		}
		return success;
		}

		bool NumServer::destroyTensorsSync()
		{
		bool success = true;
		while(!tensors_.empty()){
		success = destroyTensorSync(tensors_.begin()->first);
		if(!success) break;
		}
		return success;
		}

		bool NumServer::initTensorFile(const std::string & name,
		const std::string & filename)
		{
		@@ -3502,17 +3522,21 @@ std::shared_ptr<talsh::Tensor> NumServer::getLocalTensor(const std::string & nam
		return getLocalTensor(iter->second);
		}

		void NumServer::destroyOrphanedTensors()
		void NumServer::destroyOrphanedTensors(bool force)
		{
		//std::cout << "#DEBUG(exatn::NumServer): Destroying orphaned tensors ... "; //debug
		//std::cout << "#DEBUG(exatn::NumServer): Destroying orphaned tensors:\n" << std::flush; //debug
		auto iter = implicit_tensors_.begin();
		while(iter != implicit_tensors_.end()){
		int ref_count = 1;
		auto tens = tensors_.find(iter->first);
		if(tens != tensors_.end()) ++ref_count;
		if(iter->second.use_count() <= ref_count){
		auto sh_use_count = iter->second.use_count();
		//std::cout << "#DEBUG(exatn::NumServer::destroyOrphanedTensors): Orphan candidate found: ExaTN ref count "
		// << ref_count << " VS shared_ptr use count " << sh_use_count << ": "
		// << iter->first << std::endl << std::flush; //debug
		if(force \|\| sh_use_count <= ref_count){
		//std::cout << "#DEBUG(exatn::NumServer::destroyOrphanedTensors): Orphan found with ref count "
		// << ref_count << ": " << iter->first << std::endl; //debug
		// << ref_count << ": " << iter->first << std::endl << std::flush; //debug
		auto tensor_mapper = getTensorMapper(getTensorProcessGroup(iter->first));
		std::shared_ptr<TensorOperation> destroy_op = tensor_op_factory_->createTensorOp(TensorOpCode::DESTROY);
		destroy_op->setTensorOperand(iter->second);
		@@ -3527,6 +3551,16 @@ void NumServer::destroyOrphanedTensors()
		return;
		}

		void NumServer::printAllocatedTensors() const
		{
		std::cout << "#DEBUG(exatn::NumServer::printAllocatedTensors):" << std::endl;
		for(const auto & tens: tensors_){
		std::cout << tens.first << ": Reference count = " << tens.second.use_count() << std::endl;
		}
		std::cout << "#END" << std::endl << std::flush;
		return;
		}

		void NumServer::printImplicitTensors() const
		{
		std::cout << "#DEBUG(exatn::NumServer::printImplicitTensors):" << std::endl;

src/exatn/num_server.hpp

+17 −6

Original line number	Diff line number	Diff line
		/** ExaTN::Numerics: Numerical server
		REVISION: 2022/01/17
		REVISION: 2022/01/25

		Copyright (C) 2018-2022 Dmitry I. Lyakh (Liakh)
		Copyright (C) 2018-2022 Oak Ridge National Laboratory (UT-Battelle) **/
		@@ -446,9 +446,11 @@ public:
		/** Synchronizes execution of all outstanding tensor operations.
		Changing wait to FALSE, only tests for completion.
		If ProcessGroup is not provided, defaults to the local process. **/
		bool sync(bool wait = true);
		bool sync(bool wait = true,
		bool clean_garbage = false);
		bool sync(const ProcessGroup & process_group,
		bool wait = true);
		bool wait = true,
		bool clean_garbage = false);

		/ HIGHER-LEVEL WRAPPERS /

		@@ -628,6 +630,11 @@ public:

		bool destroyTensorsSync(TensorNetwork & tensor_network); //inout: tensor network

		/ Destroys all currently allocated tensors. /
		bool destroyTensors();

		bool destroyTensorsSync();

		/ Initializes a tensor to some scalar value. /
		template<typename NumericType>
		bool initTensor(const std::string & name, //in: tensor name
		@@ -1020,7 +1027,10 @@ public:

		inline double getTimeStampStart() const {return time_start_;}

		/ DEBUG: Prints all currently existing tensors created implicitly. /
		/ DEBUG: Prints all currently existing (allocated) tensors. /
		void printAllocatedTensors() const;

		/ DEBUG: Prints all currently existing (allocated) tensors created implicitly. /
		void printImplicitTensors() const;

		protected:
		@@ -1034,8 +1044,9 @@ protected:
		bool sync(TensorOperation & operation, //in: previously submitted tensor operation
		bool wait = true);

		/ Destroys orphaned tensors (garbage collection). /
		void destroyOrphanedTensors();
		/** Destroys orphaned tensors (garbage collection). Setting <force> to TRUE
		will force destruction regardless of the use count. **/
		void destroyOrphanedTensors(bool force = false);

		private:

src/exatn/tests/NumServerTester.cpp

+282 −47

File changed.

Preview size limit exceeded, changes collapsed.