Commit b96860b1 authored by gbalduzz's avatar gbalduzz
Browse files

fixup! Merge branch 'fix_develop_137' into ct_int

parent 45175835

test/performance/phys/accumulation/sp_accumulation_performance_test.cpp

+4 −4
Original line number Diff line number Diff line
@@ -143,19 +143,19 @@ int main(int argc, char** argv) { 
  // Allows memory to be assigned.

  gpu_accumulator.resetAccumulation();

  gpu_accumulator.accumulate(M_dev, config, sign);

  cudaStreamSynchronize(gpu_accumulator.get_streams()[0]);

  cudaStreamSynchronize(gpu_accumulator.get_streams()[1]);

  cudaStreamSynchronize(*gpu_accumulator.get_streams()[0]);

  cudaStreamSynchronize(*gpu_accumulator.get_streams()[1]);

  gpu_accumulator.resetAccumulation();



  Profiler::start();

  cudaProfilerStart();



  // Profile Single invocation.

  start_event.record(gpu_accumulator.get_streams()[0]);

  start_event.record(*gpu_accumulator.get_streams()[0]);

  dca::profiling::WallTime host_start_time;

  gpu_accumulator.accumulate(M_dev, config, sign);

  dca::profiling::WallTime host_end_time;

  stop_event.record(gpu_accumulator.get_streams()[1]);

  stop_event.record(*gpu_accumulator.get_streams()[1]);



  const double host_time = duration(host_end_time, host_start_time);

  const double dev_time = dca::linalg::util::elapsedTime(stop_event, start_event);

test/performance/phys/accumulation/tp_accumulation_performance_test.cpp

+3 −3
Original line number Diff line number Diff line
@@ -130,7 +130,7 @@ int main(int argc, char** argv) { 
    std::cout << "\nN positive frequencies:\t" << parameters.get_four_point_fermionic_frequencies();

    std::cout << "\nN bands:\t" << BDmn::dmn_size();

    std::cout << "\nN cluster sites:\t" << RDmn::dmn_size();

    std::cout << "\nType:\t" << parameters.get_four_point_type();

    std::cout << "\nType:\t" << dca::phys::toString(parameters.get_channels().at(0));

    std::cout << "\n\nTpAccumulation CPU time [sec]:\t " << time << "\n";

  }
@@ -153,11 +153,11 @@ int main(int argc, char** argv) { 
  cudaProfilerStart();



  // Time a single execution.

  start_event.record(gpu_accumulator.get_streams()[0]);

  start_event.record(*gpu_accumulator.get_stream());

  dca::profiling::WallTime host_start_time;

  gpu_accumulator.accumulate(M_dev, config, sign);

  dca::profiling::WallTime host_end_time;

  stop_event.record(gpu_accumulator.get_streams()[0]);

  stop_event.record(*gpu_accumulator.get_stream());



  const double host_time = duration(host_end_time, host_start_time);

  const double dev_time = dca::linalg::util::elapsedTime(stop_event, start_event);