Commit ba208e1b authored by gbalduzz's avatar gbalduzz
Browse files

Compute single particle greens function always in double precision.

parent 9d4e80e0

cmake/dca_config.cmake

+7 −7
Original line number Diff line number Diff line
@@ -237,12 +237,12 @@ endif() 


################################################################################

# Single precision measurements

# TODO: change to ON by default after merging and testing the two particle accumulator.

option(DCA_WITH_SINGLE_PRECISION_MEASUREMENTS "Measure in single precision." OFF)

mark_as_advanced(DCA_WITH_SINGLE_PRECISION_MEASUREMENTS)

# TODO: maybe change to ON by default.

option(DCA_WITH_SINGLE_PRECISION_TP_MEASUREMENTS "Measure in single precision." OFF)

mark_as_advanced(DCA_WITH_SINGLE_PRECISION_TP_MEASUREMENTS)



if (DCA_WITH_SINGLE_PRECISION_MEASUREMENTS)

  dca_add_config_define(DCA_WITH_SINGLE_PRECISION_MEASUREMENTS)

if (DCA_WITH_SINGLE_PRECISION_TP_MEASUREMENTS)

  dca_add_config_define(DCA_WITH_SINGLE_PRECISION_TP_MEASUREMENTS)

endif()



################################################################################
@@ -278,9 +278,9 @@ else() 
endif()



if (DCA_WITH_SINGLE_PRECISION_MEASUREMENTS)

  set(MC_ACCUMULATION_SCALAR float)

  set(TP_ACCUMULATION_SCALAR float)

else()

  set(MC_ACCUMULATION_SCALAR double)

  set(TP_ACCUMULATION_SCALAR double)

endif()



option(DCA_WITH_MANAGED_MEMORY "Use managed memory allocator." OFF)

include/dca/config/accumulation_options.hpp.in

+1 −1
Original line number Diff line number Diff line
@@ -23,7 +23,7 @@ namespace config { 
// dca::config::



struct AccumulationOptions {

  using MCAccumulationScalar = @MC_ACCUMULATION_SCALAR@;

  using TPAccumulationScalar = @TP_ACCUMULATION_SCALAR@;



  static constexpr bool memory_savings = @MEMORY_SAVINGS@;

include/dca/config/cmake_options.hpp

+1 −2
Original line number Diff line number Diff line
@@ -40,8 +40,7 @@ struct CMakeOptions { 
  static const std::string dca_profiler;

  static const std::string dca_with_autotuning;

  static const std::string dca_with_gnuplot;

  static const std::string dca_with_reduced_vertex_function;

  static const std::string dca_with_single_precision_measurements;

  static const std::string dca_with_single_precision_tp_measurements;

  static const std::string dca_with_memory_savings;

  static const std::string dca_with_managed_memory;

include/dca/phys/dca_data/dca_data.hpp

+1 −1
Original line number Diff line number Diff line
@@ -66,7 +66,7 @@ public: 
  using Concurrency = typename Parameters::concurrency_type;

  using Lattice = typename Parameters::lattice_type;

  constexpr static int DIMENSION = Lattice::DIMENSION;

  using TpAccumulatorScalar = typename Parameters::MC_measurement_scalar_type;

  using TpAccumulatorScalar = typename Parameters::TP_measurement_scalar_type;



  using TDmn = func::dmn_0<domains::time_domain>;

  using WDmn = func::dmn_0<domains::frequency_domain>;

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

+6 −7
Original line number Diff line number Diff line
@@ -167,9 +167,8 @@ private: 
  void accumulate_single_particle_quantities();



  void accumulate_equal_time_quantities();

  using AccumType = typename Parameters::MC_measurement_scalar_type;

  void accumulate_equal_time_quantities(const std::array<linalg::Matrix<AccumType, linalg::GPU>, 2>& M);

  void accumulate_equal_time_quantities(const std::array<linalg::Matrix<AccumType, linalg::CPU>, 2>& M);

  void accumulate_equal_time_quantities(const std::array<linalg::Matrix<double, linalg::GPU>, 2>& M);

  void accumulate_equal_time_quantities(const std::array<linalg::Matrix<double, linalg::CPU>, 2>& M);



  void accumulate_two_particle_quantities();
@@ -192,8 +191,8 @@ protected: 


  std::array<std::vector<vertex_singleton_type>, 2> hs_configuration_;



  std::array<dca::linalg::Matrix<AccumType, device_t>, 2> M_;

  std::array<dca::linalg::Matrix<AccumType, linalg::CPU>, 2> M_host_;

  std::array<dca::linalg::Matrix<double, device_t>, 2> M_;

  std::array<dca::linalg::Matrix<double, linalg::CPU>, 2> M_host_;



  func::function<double, func::dmn_0<domains::numerical_error_domain>> error;

  func::function<double, func::dmn_0<Feynman_expansion_order_domain>> visited_expansion_order_k;
@@ -437,7 +436,7 @@ void CtauxAccumulator<device_t, Parameters, Data>::accumulate_equal_time_quantit 


template <dca::linalg::DeviceType device_t, class Parameters, class Data>

void CtauxAccumulator<device_t, Parameters, Data>::accumulate_equal_time_quantities(

    const std::array<linalg::Matrix<AccumType, linalg::GPU>, 2>& M) {

    const std::array<linalg::Matrix<double, linalg::GPU>, 2>& M) {

  for (int s = 0; s < 2; ++s)

    M_host_[s].setAsync(M[s], thread_id, s);

  for (int s = 0; s < 2; ++s)
@@ -448,7 +447,7 @@ void CtauxAccumulator<device_t, Parameters, Data>::accumulate_equal_time_quantit 


template <dca::linalg::DeviceType device_t, class Parameters, class Data>

void CtauxAccumulator<device_t, Parameters, Data>::accumulate_equal_time_quantities(

    const std::array<linalg::Matrix<AccumType, linalg::CPU>, 2>& M) {

    const std::array<linalg::Matrix<double, linalg::CPU>, 2>& M) {

  equal_time_accumulator_.accumulateAll(hs_configuration_[0], M[0], hs_configuration_[1], M[1],

                                        current_sign);