This class holds data buffers for either immediate MPI send or recv calls. More...

#include <AMReX_NonLocalBC.H>

Public Attributes
TheFaArenaPointer	the_data = nullptr
	Holds 'em all in one data pointer.
Vector< int >	rank {}
	Stores MPI ranks. For recvs it is the 'from' rank and for sends it is the 'to' rank.
Vector< char * >	data {}
	Pointers to the_data that can be used for each single data transaction.
Vector< std::size_t >	offset {}
	All offsets of data in the_data.
Vector< std::size_t >	size {}
	The size in bytes for each data transaction.
Vector< MPI_Request >	request {}
	The associated MPI_Request for each data transaction.
Vector< MPI_Status >	stats {}
	For each request the corresponding MPI_status, used for debugging.
Vector< const FabArrayBase::CopyComTagsContainer * >	cctc {}
	For each request the copy comm tags for the corresponding data FABs.
Vector< int >	comm_rank {}
	rank translated into the communicator used by Alltoall and Alltoallv.
Vector< int >	counts {}
	size and offset per rank of that communicator, as MPI_Alltoallv wants them.
Vector< int >	displs {}
std::size_t	total_bytes = 0
	Size of the_data in bytes, so that it can be released and reacquired.
std::uint64_t	id = std::numeric_limits<std::uint64_t>::max()

Detailed Description

This class holds data buffers for either immediate MPI send or recv calls.

The data buffers need to be kept alive until all MPI_Request have been completed. All member variables are resized a corresponding call to PrepareCommBuffers and the data transactions are initiated either by calls to PostRecvs or PostSends.

Member Data Documentation

◆ cctc

Vector<const FabArrayBase::CopyComTagsContainer*> amrex::NonLocalBC::CommData::cctc {}

For each request the copy comm tags for the corresponding data FABs.

◆ comm_rank

Vector<int> amrex::NonLocalBC::CommData::comm_rank {}

rank translated into the communicator used by Alltoall and Alltoallv.

◆ counts

Vector<int> amrex::NonLocalBC::CommData::counts {}

size and offset per rank of that communicator, as MPI_Alltoallv wants them.

◆ data

Vector<char*> amrex::NonLocalBC::CommData::data {}

Pointers to the_data that can be used for each single data transaction.

◆ displs

Vector<int> amrex::NonLocalBC::CommData::displs {}

◆ id

std::uint64_t amrex::NonLocalBC::CommData::id = std::numeric_limits<std::uint64_t>::max()

◆ offset

Vector<std::size_t> amrex::NonLocalBC::CommData::offset {}

All offsets of data in the_data.

◆ rank

Vector<int> amrex::NonLocalBC::CommData::rank {}

Stores MPI ranks. For recvs it is the 'from' rank and for sends it is the 'to' rank.

◆ request

Vector<MPI_Request> amrex::NonLocalBC::CommData::request {}

The associated MPI_Request for each data transaction.

◆ size

Vector<std::size_t> amrex::NonLocalBC::CommData::size {}

The size in bytes for each data transaction.

◆ stats

Vector<MPI_Status> amrex::NonLocalBC::CommData::stats {}

For each request the corresponding MPI_status, used for debugging.

◆ the_data

TheFaArenaPointer amrex::NonLocalBC::CommData::the_data = nullptr

Holds 'em all in one data pointer.

◆ total_bytes

std::size_t amrex::NonLocalBC::CommData::total_bytes = 0

Size of the_data in bytes, so that it can be released and reacquired.

The documentation for this struct was generated from the following file:

/home/runner/work/amrex/amrex/Src/Base/AMReX_NonLocalBC.H

Public Attributes