Commit de680b5d authored by Doak, Peter W.'s avatar Doak, Peter W.
Browse files

fix performance regression caused by singleMeasLogging

parent aa2f5698

include/dca/phys/dca_step/cluster_solver/ctaux/ctaux_cluster_solver.hpp

+2 −2
Original line number Diff line number Diff line
@@ -104,7 +104,7 @@ public: 
  auto local_G_k_w() const;



  void computeG_k_w(const SpGreensFunction& G0, const SpGreensFunction& M_k_w,

                    SpGreensFunction& G_k_w);

                    SpGreensFunction& G_k_w) const;



  void setSampleConfiguration(const io::Buffer&) {}
@@ -538,7 +538,7 @@ void CtauxClusterSolver<device_t, Parameters, Data, DIST>::symmetrize_measuremen 
template <dca::linalg::DeviceType device_t, class Parameters, class Data, DistType DIST>

void CtauxClusterSolver<device_t, Parameters, Data, DIST>::computeG_k_w(const SpGreensFunction& G0,

                                                                        const SpGreensFunction& M_k_w,

                                                                        SpGreensFunction& G_k_w) {

                                                                        SpGreensFunction& G_k_w) const {

  const int matrix_dim = nu::dmn_size();

  dca::linalg::Matrix<std::complex<double>, dca::linalg::CPU> G0_times_M_matrix(

      "GO_M_matrix", matrix_dim, matrix_dim);

include/dca/phys/dca_step/cluster_solver/stdthread_qmci/stdthread_qmci_accumulator.hpp

+2 −0
Original line number Diff line number Diff line
@@ -81,6 +81,8 @@ public: 
    measuring_ = false;

  }



  std::size_t get_meas_id() const { return meas_id_; }



private:

  int thread_id_;

  bool measuring_;

include/dca/phys/dca_step/cluster_solver/stdthread_qmci/stdthread_qmci_cluster_solver.hpp

+31 −25
Original line number Diff line number Diff line
@@ -88,7 +88,8 @@ public: 
  };



  auto transformMFunction(const MFuncAndSign& mfs) const;

  auto computeSingleMeasurement_G_k_w(const SpGreensFunction& M_k_w);

  auto computeSingleMeasurement_G_k_w(const SpGreensFunction& M_k_w) const ;

  void logSingleMeasurement(StdThreadAccumulatorType& accumulator, int stamping_period, bool log_MFunction) const;



private:

  void startWalker(int id);
@@ -415,7 +416,8 @@ void StdThreadQmciClusterSolver<QmciSolver>::iterateOverLocalMeasurements( 
      f(meas_id, n_local_meas, print);

  }

  else {

    throw std::runtime_error("Non fix-meas-per-walker accumulation is suspect and disabled at this time.");

    throw std::runtime_error(

        "Non fix-meas-per-walker accumulation is suspect and disabled at this time.");

    // Perform the total number of loop with a shared atomic counter.

    // for (int meas_id = measurements_done_++; meas_id < n_local_meas; meas_id = measurements_done_++)

    //   f(meas_id, n_local_meas, print);
@@ -434,14 +436,27 @@ auto StdThreadQmciClusterSolver<QmciSolver>::transformMFunction(const MFuncAndSi 
// accumulators causing diversion from the normal algorithm. a way to merge this code should be

// found.

template <class QmciSolver>

auto StdThreadQmciClusterSolver<QmciSolver>::computeSingleMeasurement_G_k_w(const SpGreensFunction& M_k_w)

{

auto StdThreadQmciClusterSolver<QmciSolver>::computeSingleMeasurement_G_k_w(

    const SpGreensFunction& M_k_w) const {

  SpGreensFunction G_k_w("G_k_w");

  QmciSolver::computeG_k_w(data_.G0_k_w_cluster_excluded, M_k_w, G_k_w);



  return G_k_w;

}



template <class QmciSolver>

void StdThreadQmciClusterSolver<QmciSolver>::logSingleMeasurement(StdThreadAccumulatorType& accumulator_obj, int stamping_period, bool log_MFunction) const {

  if (accumulator_obj.get_meas_id() % stamping_period == 0) {

    auto mfs = ThisType::getSingleMFunc(accumulator_obj);

    auto M_k_w = ThisType::transformMFunction(mfs);

    auto single_meas_G_k_w = ThisType::computeSingleMeasurement_G_k_w(M_k_w);

    if (log_MFunction)

      accumulator_obj.logPerConfigurationMFunction(M_k_w, mfs.sign);

    accumulator_obj.logPerConfigurationGreensFunction(single_meas_G_k_w);

    accumulator_obj.clearSingleMeasurement();

  }

}



template <class QmciSolver>

void StdThreadQmciClusterSolver<QmciSolver>::startAccumulator(int id, const Parameters& parameters) {

  Profiler::start_threading(id);
@@ -452,7 +467,8 @@ void StdThreadQmciClusterSolver<QmciSolver>::startAccumulator(int id, const Para 
  accumulator_obj.initialize(dca_iteration_);



  std::unique_ptr<std::exception> exception_ptr;



  bool log_MFunction = parameters.per_measurement_MFunction();

  auto stamping_period = parameters.stamping_period();

  try {

    while (true) {

      {
@@ -471,15 +487,8 @@ void StdThreadQmciClusterSolver<QmciSolver>::startAccumulator(int id, const Para 
      {

        Profiler profiler("accumulating", "stdthread-MC-accumulator", __LINE__, id);

        accumulator_obj.measure();

        if (last_iteration_) {

          auto mfs =  getSingleMFunc(accumulator_obj);

          auto M_k_w = transformMFunction(mfs);

          auto single_meas_G_k_w = computeSingleMeasurement_G_k_w(M_k_w);

          if(parameters.per_measurement_MFunction())

            accumulator_obj.logPerConfigurationMFunction(M_k_w, mfs.sign);

          accumulator_obj.logPerConfigurationGreensFunction(single_meas_G_k_w);

          accumulator_obj.clearSingleMeasurement();

        }

        if (accumulator_log && stamping_period)

          logSingleMeasurement(accumulator_obj, stamping_period, log_MFunction);

        accumulator_obj.finishMeasuring();

      }

    }
@@ -511,7 +520,8 @@ void StdThreadQmciClusterSolver<QmciSolver>::startAccumulator(int id, const Para 
}



template <class QmciSolver>

void StdThreadQmciClusterSolver<QmciSolver>::startWalkerAndAccumulator(int id, const Parameters& parameters) {

void StdThreadQmciClusterSolver<QmciSolver>::startWalkerAndAccumulator(int id,

                                                                       const Parameters& parameters) {

  Profiler::start_threading(id);



  // Create and warm a walker.
@@ -531,6 +541,9 @@ void StdThreadQmciClusterSolver<QmciSolver>::startWalkerAndAccumulator(int id, c 


  std::unique_ptr<std::exception> current_exception;



  bool log_MFunction = parameters.per_measurement_MFunction();

  auto stamping_period = parameters.stamping_period();



  try {

    iterateOverLocalMeasurements(id, [&](const int meas_id, const int n_meas, const bool print) {

      {
@@ -542,15 +555,8 @@ void StdThreadQmciClusterSolver<QmciSolver>::startWalkerAndAccumulator(int id, c 
        accumulator_obj.updateFrom(walker, concurrency_.get_id(), walker.get_thread_id(),

                                   walker.get_meas_id(), last_iteration_);

        accumulator_obj.measure();

        if (last_iteration_) {

          auto mfs = getSingleMFunc(accumulator_obj);

          auto M_k_w = transformMFunction(mfs);

          auto single_meas_G_k_w = computeSingleMeasurement_G_k_w(M_k_w);

          if(parameters.per_measurement_MFunction())

            accumulator_obj.logPerConfigurationMFunction(M_k_w, mfs.sign);

          accumulator_obj.logPerConfigurationGreensFunction(single_meas_G_k_w);

          accumulator_obj.clearSingleMeasurement();

        }

        if (accumulator_log && stamping_period)

          logSingleMeasurement(accumulator_obj, stamping_period, log_MFunction);

        accumulator_obj.finishMeasuring();

      }

      if (print)

test/integration/cluster_solver/stdthread_qmci/stdthread_ctint_cluster_solver_test.cpp

+3 −2
Original line number Diff line number Diff line 
// Copyright (C) 2018 ETH Zurich

// Copyright (C) 2018 UT-Battelle, LLC

// Copyright (C) 2022 ETH Zurich

// Copyright (C) 2022 UT-Battelle, LLC

// All rights reserved.

//

// See LICENSE.txt for terms of usage.

//  See CITATION.md for citation guidelines, if DCA++ is used for scientific publications.

//

// Author: Giovanni Balduzzi (gbalduzz@itp.phys.ethz.ch)

//         Peter Doak (doakpw@ornl.gov)

//

// No-change test for CT-INT posix wrapper.