📄 barrier.c
字号:
/* -*- Mode: C; c-basic-offset:4 ; -*- *//* * * (C) 2001 by Argonne National Laboratory. * See COPYRIGHT in top-level directory. */#include "mpiimpl.h"/* -- Begin Profiling Symbol Block for routine MPI_Barrier */#if defined(HAVE_PRAGMA_WEAK)#pragma weak MPI_Barrier = PMPI_Barrier#elif defined(HAVE_PRAGMA_HP_SEC_DEF)#pragma _HP_SECONDARY_DEF PMPI_Barrier MPI_Barrier#elif defined(HAVE_PRAGMA_CRI_DUP)#pragma _CRI duplicate MPI_Barrier as PMPI_Barrier#endif/* -- End Profiling Symbol Block *//* Define MPICH_MPI_FROM_PMPI if weak symbols are not supported to build the MPI routines */#ifndef MPICH_MPI_FROM_PMPI#define MPI_Barrier PMPI_Barrier/* This is the default implementation of the barrier operation. The algorithm is: Algorithm: MPI_Barrier We use the dissemination algorithm described in: Debra Hensgen, Raphael Finkel, and Udi Manbet, "Two Algorithms for Barrier Synchronization," International Journal of Parallel Programming, 17(1):1-17, 1988. It uses ceiling(lgp) steps. In step k, 0 <= k <= (ceiling(lgp)-1), process i sends to process (i + 2^k) % p and receives from process (i - 2^k + p) % p. Possible improvements: End Algorithm: MPI_Barrier This is an intracommunicator barrier only!*//* not declared static because it is called in ch3_comm_connect/accept */int MPIR_Barrier( MPID_Comm *comm_ptr ){ static const char FCNAME[] = "MPIR_Barrier"; int size, rank, src, dst, mask, mpi_errno=MPI_SUCCESS; MPI_Comm comm; size = comm_ptr->local_size; /* Trivial barriers return immediately */ if (size == 1) return MPI_SUCCESS; rank = comm_ptr->rank; comm = comm_ptr->handle; /* Only one collective operation per communicator can be active at any time */ MPIDU_ERR_CHECK_MULTIPLE_THREADS_ENTER( comm_ptr ); mask = 0x1; while (mask < size) { dst = (rank + mask) % size; src = (rank - mask + size) % size; mpi_errno = MPIC_Sendrecv(NULL, 0, MPI_BYTE, dst, MPIR_BARRIER_TAG, NULL, 0, MPI_BYTE, src, MPIR_BARRIER_TAG, comm, MPI_STATUS_IGNORE); /* --BEGIN ERROR HANDLING-- */ if (mpi_errno) { mpi_errno = MPIR_Err_create_code(mpi_errno, MPIR_ERR_RECOVERABLE, FCNAME, __LINE__, MPI_ERR_OTHER, "**fail", 0); return mpi_errno; } /* --END ERROR HANDLING-- */ mask <<= 1; } MPIDU_ERR_CHECK_MULTIPLE_THREADS_EXIT( comm_ptr ); return mpi_errno;}#if 0/* This is the default implementation of the barrier operation. The algorithm is: Algorithm: MPI_Barrier Find the largest power of two that is less than or equal to the size of the communicator. Call tbis twon_within. Divide the communicator by rank into two groups: those with rank < twon_within and those with greater rank. The barrier executes in three steps. First, the group with rank >= twon_within sends to the first (size-twon_within) ranks of the first group. That group then executes a recursive doubling algorithm for the barrier. For the third step, the first (size-twon_within) ranks send to the top group. This is the same algorithm used in MPICH-1. Possible improvements: The upper group could apply recursively this approach to reduce the total number of messages sent (in the case of of a size of 2^n-1, there are 2^(n-1) messages sent in the first and third steps). End Algorithm: MPI_Barrier This is an intracommunicator barrier only!*/int MPIR_Barrier( MPID_Comm *comm_ptr ){ int size, rank; int twon_within, n2, remaining, gap, partner; MPID_Request *request_ptr; int mpi_errno = MPI_SUCCESS; size = comm_ptr->remote_size; rank = comm_ptr->rank; /* Trivial barriers return immediately */ if (size == 1) return MPI_SUCCESS; /* Only one collective operation per communicator can be active at any time */ MPIDU_ERR_CHECK_MULTIPLE_THREADS_ENTER( comm_ptr ); /* Find the twon_within (this could be cached if more routines need it) */ twon_within = 1; n2 = 2; while (n2 <= size) { twon_within = n2; n2 <<= 1; } remaining = size - twon_within; if (rank < twon_within) { /* First step: receive from the upper group */ if (rank < remaining) { MPID_Recv( 0, 0, MPI_BYTE, twon_within + rank, MPIR_BARRIER_TAG, comm_ptr, MPID_CONTEXT_INTRA_COLL, MPI_STATUS_IGNORE, &request_ptr ); if (request_ptr) { mpi_errno = MPIC_Wait(request_ptr); MPID_Request_release(request_ptr); /* --BEGIN ERROR HANDLING-- */ if (mpi_errno != MPI_SUCCESS) { goto fn_exit; } /* --END ERROR HANDLING-- */ } } /* Second step: recursive doubling exchange */ for (gap=1; gap<twon_within; gap <<= 1) { partner = (rank ^ gap); MPIC_Sendrecv( 0, 0, MPI_BYTE, partner, MPIR_BARRIER_TAG, 0, 0, MPI_BYTE, partner, MPIR_BARRIER_TAG, comm_ptr->handle, MPI_STATUS_IGNORE ); } /* Third step: send to the upper group */ if (rank < remaining) { MPID_Send( 0, 0, MPI_BYTE, rank + twon_within, MPIR_BARRIER_TAG, comm_ptr, MPID_CONTEXT_INTRA_COLL, &request_ptr ); if (request_ptr) { mpi_errno = MPIC_Wait(request_ptr); MPID_Request_release(request_ptr); /* --BEGIN ERROR HANDLING-- */ if (mpi_errno != MPI_SUCCESS) { goto fn_exit; } /* --END ERROR HANDLING-- */ } } } else { /* For the upper group, step one is a send */ MPID_Send( 0, 0, MPI_BYTE, rank - twon_within, MPIR_BARRIER_TAG, comm_ptr, MPID_CONTEXT_INTRA_COLL, &request_ptr ); if (request_ptr) { mpi_errno = MPIC_Wait(request_ptr); MPID_Request_release(request_ptr); /* --BEGIN ERROR HANDLING-- */ if (mpi_errno != MPI_SUCCESS) { goto fn_exit; } /* --END ERROR HANDLING-- */ } /* There is no second step; for the third step, recv */ MPID_Recv( 0, 0, MPI_BYTE, rank - twon_within, MPIR_BARRIER_TAG, comm_ptr, MPID_CONTEXT_INTRA_COLL, MPI_STATUS_IGNORE, &request_ptr ); if (request_ptr) { mpi_errno = MPIC_Wait(request_ptr); MPID_Request_release(request_ptr); /* --BEGIN ERROR HANDLING-- */ if (mpi_errno != MPI_SUCCESS) { goto fn_exit; } /* --END ERROR HANDLING-- */ } } fn_exit: MPIDU_ERR_CHECK_MULTIPLE_THREADS_EXIT( comm_ptr ); return mpi_errno;}#endif/* not declared static because a machine-specific function may call this one in some cases */int MPIR_Barrier_inter( MPID_Comm *comm_ptr ){ static const char FCNAME[] = "MPIR_Barrier_inter"; int rank, mpi_errno, i, root; MPID_Comm *newcomm_ptr = NULL; rank = comm_ptr->rank; /* Get the local intracommunicator */ if (!comm_ptr->local_comm) MPIR_Setup_intercomm_localcomm( comm_ptr ); newcomm_ptr = comm_ptr->local_comm; /* do a barrier on the local intracommunicator */ mpi_errno = MPIR_Barrier(newcomm_ptr); /* --BEGIN ERROR HANDLING-- */ if (mpi_errno) { mpi_errno = MPIR_Err_create_code(mpi_errno, MPIR_ERR_RECOVERABLE, FCNAME, __LINE__, MPI_ERR_OTHER, "**fail", 0); return mpi_errno; } /* --END ERROR HANDLING-- */ /* rank 0 on each group does an intercommunicator broadcast to the remote group to indicate that all processes in the local group have reached the barrier. We do a 1-byte bcast because a 0-byte bcast will just return without doing anything. */ /* first broadcast from left to right group, then from right to left group */ if (comm_ptr->is_low_group) { /* bcast to right*/ root = (rank == 0) ? MPI_ROOT : MPI_PROC_NULL; mpi_errno = MPIR_Bcast_inter(&i, 1, MPI_BYTE, root, comm_ptr); /* --BEGIN ERROR HANDLING-- */ if (mpi_errno) { mpi_errno = MPIR_Err_create_code(mpi_errno, MPIR_ERR_RECOVERABLE, FCNAME, __LINE__, MPI_ERR_OTHER, "**fail", 0); return mpi_errno; } /* --END ERROR HANDLING-- */ /* receive bcast from right */ root = 0; mpi_errno = MPIR_Bcast_inter(&i, 1, MPI_BYTE, root, comm_ptr); /* --BEGIN ERROR HANDLING-- */ if (mpi_errno) { mpi_errno = MPIR_Err_create_code(mpi_errno, MPIR_ERR_RECOVERABLE, FCNAME, __LINE__, MPI_ERR_OTHER, "**fail", 0); return mpi_errno; } /* --END ERROR HANDLING-- */ } else { /* receive bcast from left */ root = 0; mpi_errno = MPIR_Bcast_inter(&i, 1, MPI_BYTE, root, comm_ptr); /* --BEGIN ERROR HANDLING-- */ if (mpi_errno) { mpi_errno = MPIR_Err_create_code(mpi_errno, MPIR_ERR_RECOVERABLE, FCNAME, __LINE__, MPI_ERR_OTHER, "**fail", 0); return mpi_errno; } /* --END ERROR HANDLING-- */ /* bcast to left */ root = (rank == 0) ? MPI_ROOT : MPI_PROC_NULL; mpi_errno = MPIR_Bcast_inter(&i, 1, MPI_BYTE, root, comm_ptr); /* --BEGIN ERROR HANDLING-- */ if (mpi_errno) { mpi_errno = MPIR_Err_create_code(mpi_errno, MPIR_ERR_RECOVERABLE, FCNAME, __LINE__, MPI_ERR_OTHER, "**fail", 0); return mpi_errno; } /* --END ERROR HANDLING-- */ } return mpi_errno;}#endif#undef FUNCNAME#define FUNCNAME MPI_Barrier/*@MPI_Barrier - Blocks until all processes in the communicator havereached this routine. Input Parameter:. comm - communicator (handle) Notes:Blocks the caller until all processes in the communicator have called it; that is, the call returns at any process only after all members of thecommunicator have entered the call..N ThreadSafe.N Fortran.N Errors.N MPI_SUCCESS.N MPI_ERR_COMM@*/int MPI_Barrier( MPI_Comm comm ){ static const char FCNAME[] = "MPI_Barrier"; int mpi_errno = MPI_SUCCESS; MPID_Comm *comm_ptr = NULL; MPID_MPI_STATE_DECL(MPID_STATE_MPI_BARRIER); MPIR_ERRTEST_INITIALIZED_ORDIE(); MPID_CS_ENTER(); MPID_MPI_COLL_FUNC_ENTER(MPID_STATE_MPI_BARRIER); /* Validate parameters, especially handles needing to be converted */# ifdef HAVE_ERROR_CHECKING { MPID_BEGIN_ERROR_CHECKS; { MPIR_ERRTEST_COMM(comm, mpi_errno); if (mpi_errno != MPI_SUCCESS) goto fn_fail; } MPID_END_ERROR_CHECKS; }# endif /* HAVE_ERROR_CHECKING */ /* Convert MPI object handles to object pointers */ MPID_Comm_get_ptr( comm, comm_ptr ); /* Validate parameters and objects (post conversion) */# ifdef HAVE_ERROR_CHECKING { MPID_BEGIN_ERROR_CHECKS; { /* Validate communicator */ MPID_Comm_valid_ptr( comm_ptr, mpi_errno ); if (mpi_errno) goto fn_fail; } MPID_END_ERROR_CHECKS; }# endif /* HAVE_ERROR_CHECKING */ /* ... body of routine ... */ if (comm_ptr->coll_fns != NULL && comm_ptr->coll_fns->Barrier != NULL) { mpi_errno = comm_ptr->coll_fns->Barrier(comm_ptr); } else { MPIR_Nest_incr(); if (comm_ptr->comm_kind == MPID_INTRACOMM) { mpi_errno = MPIR_Barrier( comm_ptr ); } else { /* intercommunicator */ /* mpi_errno = MPIR_Err_create_code( MPI_SUCCESS, MPIR_ERR_RECOVERABLE, FCNAME, __LINE__, MPI_ERR_COMM, "**intercommcoll", "**intercommcoll %s", FCNAME ); */ mpi_errno = MPIR_Barrier_inter( comm_ptr ); } MPIR_Nest_decr(); } /* --BEGIN ERROR HANDLING-- */ if (mpi_errno != MPI_SUCCESS) goto fn_fail; /* --END ERROR HANDLING-- */ /* ... end of body of routine ... */ fn_exit: MPID_MPI_COLL_FUNC_EXIT(MPID_STATE_MPI_BARRIER); MPID_CS_EXIT(); return mpi_errno; fn_fail: /* --BEGIN ERROR HANDLING-- */# ifdef HAVE_ERROR_CHECKING { mpi_errno = MPIR_Err_create_code( mpi_errno, MPIR_ERR_RECOVERABLE, FCNAME, __LINE__, MPI_ERR_OTHER, "**mpi_barrier", "**mpi_barrier %C", comm); }# endif mpi_errno = MPIR_Err_return_comm( comm_ptr, FCNAME, mpi_errno ); goto fn_exit; /* --END ERROR HANDLING-- */}
⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -