Commit 9edf7896 authored by gbalduzz's avatar gbalduzz
Browse files

fix 217: DCA_HAVE_CUDA_AWARE_MPI flag.

parent 8c956625

include/dca/phys/dca_step/cluster_solver/shared_tools/accumulation/tp/tp_accumulator_mpi_gpu.hpp

+6 −6
Original line number Diff line number Diff line
@@ -123,10 +123,10 @@ private: 
  std::array<MPI_Request, 2> recv_requests_{MPI_REQUEST_NULL, MPI_REQUEST_NULL};

  std::array<MPI_Request, 2> send_requests_{MPI_REQUEST_NULL, MPI_REQUEST_NULL};



#ifndef DCA_WITH_CUDA_AWARE_MPI

#ifndef DCA_HAVE_CUDA_AWARE_MPI

  std::array<std::vector<Complex>, 2> sendbuffer_;

  std::array<std::vector<Complex>, 2> recvbuffer_;

#endif  // DCA_WITH_CUDA_AWARE_MPI

#endif  // DCA_HAVE_CUDA_AWARE_MPI

};



template <class Parameters>
@@ -366,7 +366,7 @@ void TpAccumulator<Parameters, linalg::GPU, DistType::MPI>::send(const std::arra 
  using dca::parallel::MPITypeMap;

  const auto g_size = data[0].size().first * data[0].size().second;



#ifdef DCA_WITH_CUDA_AWARE_MPI

#ifdef DCA_HAVE_CUDA_AWARE_MPI

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

    MPI_Isend(data[s].ptr(), g_size, MPITypeMap<Complex>::value(), target, thread_id_ + 1,

              MPI_COMM_WORLD, &request[s]);
@@ -381,7 +381,7 @@ void TpAccumulator<Parameters, linalg::GPU, DistType::MPI>::send(const std::arra 
    MPI_Isend(sendbuffer_[s].data(), g_size, MPITypeMap<Complex>::value(), target, thread_id_ + 1,

              MPI_COMM_WORLD, &request[s]);

  }

#endif  // DCA_WITH_CUDA_AWARE_MPI

#endif  // DCA_HAVE_CUDA_AWARE_MPI

}



template <class Parameters>
@@ -390,7 +390,7 @@ void TpAccumulator<Parameters, linalg::GPU, DistType::MPI>::receive( 
  using dca::parallel::MPITypeMap;

  const auto g_size = data[0].size().first * data[0].size().second;



#ifdef DCA_WITH_CUDA_AWARE_MPI

#ifdef DCA_HAVE_CUDA_AWARE_MPI

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

    MPI_Irecv(data[s].ptr(), g_size, MPITypeMap<Complex>::value(), source, thread_id_ + 1,

              MPI_COMM_WORLD, &request[s]);
@@ -408,7 +408,7 @@ void TpAccumulator<Parameters, linalg::GPU, DistType::MPI>::receive( 
    cudaMemcpy(data[s].ptr(), recvbuffer_[s].data(), g_size * sizeof(Complex),

               cudaMemcpyHostToDevice);

  }

#endif  // DCA_WITH_CUDA_AWARE_MPI

#endif  // DCA_HAVE_CUDA_AWARE_MPI

}



}  // namespace accumulator

test/unit/phys/dca_step/cluster_solver/shared_tools/accumulation/tp/tp_accumulator_mpi_gpu_test.cpp

+1 −1
Original line number Diff line number Diff line
@@ -12,7 +12,7 @@ 






#include "dca/phys/dca_step/cluster_solver/shared_tools/accumulation/tp/tp_accumulator_gpu.hpp"

#include "dca/phys/dca_step/cluster_solver/shared_tools/accumulation/tp/tp_accumulator_mpi_gpu.hpp"



#include <array>

#include <functional>