Commit 0256df3a authored by Dmitry I. Lyakh's avatar Dmitry I. Lyakh
Browse files

Fixed Host buffer leak problem 



Signed-off-by: default avatarDmitry I. Lyakh <quant4me@gmail.com>
parent 7dc4b071

monitor.sh

0 → 100755
+3 −0
Original line number Diff line number Diff line 
#!/bin/bash

#Will monitor the total Flop count and current memory usage:

watch --interval 1 "tail --lines=1024 ./exatn_exec_thread.0.log | grep usage | tail --lines=1"

src/exatn/num_server.cpp

+11 −3
Original line number Diff line number Diff line 
/** ExaTN::Numerics: Numerical server

REVISION: 2022/01/25

REVISION: 2022/01/26



Copyright (C) 2018-2022 Dmitry I. Lyakh (Liakh)

Copyright (C) 2018-2022 Oak Ridge National Laboratory (UT-Battelle) **/
@@ -518,6 +518,9 @@ bool NumServer::submitOp(std::shared_ptr<TensorOperation> operation) 
    std::cout << "#ERROR(exatn::NumServer::submitOp): Attempt to CREATE an already existing tensor "

              << tensor->getName() << std::endl << std::flush;

    submitted = false;

   //}else{ //debug

    //const auto & tens_name = tensor->getName();

    //std::cout << "#DEBUG(exatn::NumServer::submitOp): Created tensor " << tens_name << std::endl << std::flush;

   }

  }else if(operation->getOpcode() == TensorOpCode::DESTROY){

   auto tensor = operation->getTensorOperand(0);
@@ -526,6 +529,9 @@ bool NumServer::submitOp(std::shared_ptr<TensorOperation> operation) 
    std::cout << "#ERROR(exatn::NumServer::submitOp): Attempt to DESTROY a non-existing tensor "

              << tensor->getName() << std::endl << std::flush;

    submitted = false;

   //}else{ //debug

    //const auto & tens_name = tensor->getName();

    //std::cout << "#DEBUG(exatn::NumServer::submitOp): Destroyed tensor " << tens_name << std::endl << std::flush;

   }

  }

  //Submit tensor operation to tensor runtime:
@@ -1632,7 +1638,8 @@ bool NumServer::destroyTensors() 
{

 bool success = true;

 while(!tensors_.empty()){

  success = destroyTensor(tensors_.begin()->first);

  const auto tens_name = tensors_.begin()->first;

  success = destroyTensor(tens_name);

  if(!success) break;

 }

 return success;
@@ -1642,7 +1649,8 @@ bool NumServer::destroyTensorsSync() 
{

 bool success = true;

 while(!tensors_.empty()){

  success = destroyTensorSync(tensors_.begin()->first);

  const auto tens_name = tensors_.begin()->first;

  success = destroyTensorSync(tens_name);

  if(!success) break;

 }

 return success;

src/exatn/tests/NumServerTester.cpp

+52 −2
Original line number Diff line number Diff line
@@ -21,7 +21,7 @@ 


#define EXATN_TEST0

#define EXATN_TEST1

/*#define EXATN_TEST2

#define EXATN_TEST2

#define EXATN_TEST3

#define EXATN_TEST4

#define EXATN_TEST5
@@ -48,11 +48,12 @@ 
#define EXATN_TEST26

#define EXATN_TEST27 //requires input file from source

#define EXATN_TEST28 //requires input file from source

#define EXATN_TEST29*/

#define EXATN_TEST29

#define EXATN_TEST30

#define EXATN_TEST31 //requires input file from source

#define EXATN_TEST32

#define EXATN_TEST33

#define EXATN_TEST34





#ifdef EXATN_TEST0
@@ -4127,6 +4128,55 @@ TEST(NumServerTester, CuTensorNet) { 
#endif



#ifdef EXATN_TEST33

TEST(NumServerTester, IsometricAIEM) {

 using exatn::TensorShape;

 using exatn::TensorSignature;

 using exatn::Tensor;

 using exatn::TensorNetwork;

 using exatn::TensorExpansion;

 using exatn::TensorOperator;

 using exatn::TensorElementType;

 using exatn::TensorRange;



 const auto TENS_ELEM_TYPE = TensorElementType::COMPLEX64;



 //exatn::resetLoggingLevel(1,2); //debug



 std::size_t free_mem = 0;

 auto used_mem = exatn::getMemoryUsage(&free_mem);

 std::cout << "#MSG(exatn): Backend tensor memory usage on entrance = "

           << used_mem << std::endl << std::flush;

 assert(used_mem == 0);



 bool success = true;



 //Create tensors:

 success = exatn::createTensor("A",TENS_ELEM_TYPE,TensorShape{}); assert(success);

 success = exatn::createTensor("B",TENS_ELEM_TYPE,TensorShape{4,4}); assert(success);

 success = exatn::createTensor("C",TENS_ELEM_TYPE,TensorShape{4,4}); assert(success);



 //Init tensors:

 success = exatn::initTensorRnd("A"); assert(success);

 success = exatn::initTensorRnd("B"); assert(success);

 success = exatn::initTensor("C",0.0); assert(success);



 //Contract tensors:

 success = exatn::contractTensors("C(u1,u0)+=A()*B(u0,u1)",1.0); assert(success);



 //Destroy tensors:

 success = exatn::sync(); assert(success);

 success = exatn::destroyTensor("C"); assert(success);

 success = exatn::destroyTensor("B"); assert(success);

 success = exatn::destroyTensor("A"); assert(success);



 //Synchronize:

 success = exatn::syncClean(); assert(success);

 //exatn::resetLoggingLevel(0,0);

 //Grab a beer!

}

#endif



#ifdef EXATN_TEST34

TEST(NumServerTester, TensorComposite) {

 using exatn::TensorShape;

 using exatn::TensorSignature;

src/runtime/executor/node_executors/talsh/node_executor_talsh.cpp

+6 −0
Original line number Diff line number Diff line
@@ -673,6 +673,7 @@ int TalshNodeExecutor::execute(numerics::TensorOpContract & op, 
 }



 //std::cout << "#DEBUG(exatn::runtime::node_executor_talsh): Tensor contraction " << op.getIndexPattern() << std::endl; //debug

 //const auto host_buf_free_mem = talshDeviceBufferFreeSize(0,DEV_HOST); //debug

 auto error_code = tens0.contractAccumulate((task_res.first)->second.get(),

                                            op.getIndexPatternReduced(),

                                            tens1,tens2,
@@ -720,6 +721,11 @@ int TalshNodeExecutor::execute(numerics::TensorOpContract & op, 
  double flop_count = talsh_submitted_flops_.load() + op.getFlopEstimate() * tensorElementTypeOpFactor(tensor1.getElementType());

  talsh_submitted_flops_.store(flop_count);

 }

 /*if(talshDeviceBufferFreeSize(0,DEV_HOST) < host_buf_free_mem){ //debug

  std::cout << "#FATAL(exatn::runtime::TalshNodeExecutor): Host buffer leak detected for tensor contraction:\n";

  op.printIt();

  std::abort();

 }*/

 return error_code;

}

src/utils/errors.hpp

+27 −3
Original line number Diff line number Diff line 
/** ExaTN: Error handling

REVISION: 2021/09/27

REVISION: 2022/01/26



Copyright (C) 2018-2021 Dmitry I. Lyakh (Liakh)

Copyright (C) 2018-2021 Oak Ridge National Laboratory (UT-Battelle) **/

Copyright (C) 2018-2022 Dmitry I. Lyakh (Liakh)

Copyright (C) 2018-2022 Oak Ridge National Laboratory (UT-Battelle) **/



#ifndef EXATN_ERRORS_HPP_

#define EXATN_ERRORS_HPP_



#ifndef NO_LINUX

#include <execinfo.h> //Linux only

#include <stdio.h>

#endif



#include <iostream>

#include <string>
@@ -57,6 +62,25 @@ inline void print_variadic_pack(Arg&& arg, Args&&... args) 
 return print_variadic_pack(std::forward<Args>(args)...);

}



#ifndef NO_LINUX

inline void print_backtrace() //Linux only

{

 int MAX_CALLSTACK_DEPTH = 256;

 int callstack_depth = 0;

 void * addresses[MAX_CALLSTACK_DEPTH];

 char ** funcs;

 callstack_depth = backtrace(addresses,MAX_CALLSTACK_DEPTH);

 funcs = backtrace_symbols(addresses,callstack_depth);

 if(funcs != nullptr){

  for(int i = 0; i < callstack_depth; ++i){

   printf("%s\n",funcs[i]);

  }

  free(funcs);

 }

 return;

}

#endif



} //namespace exatn



#endif //EXATN_ERRORS_HPP_