/*@
MPI_Exscan - Computes the exclusive scan (partial reductions) of data on a
collection of processes
Input Parameters:
+ sendbuf - starting address of send buffer (choice)
. count - number of elements in input buffer (integer)
. datatype - data type of elements of input buffer (handle)
. op - operation (handle)
- comm - communicator (handle)
Output Parameters:
. recvbuf - starting address of receive buffer (choice)
Notes:
'MPI_Exscan' is like 'MPI_Scan', except that the contribution from the
calling process is not included in the result at the calling process
(it is contributed to the subsequent processes, of course).
.N ThreadSafe
.N Fortran
.N collops
.N Errors
.N MPI_SUCCESS
.N MPI_ERR_COMM
.N MPI_ERR_COUNT
.N MPI_ERR_TYPE
.N MPI_ERR_BUFFER
.N MPI_ERR_BUFFER_ALIAS
@*/
int MPI_Exscan(const void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype,
MPI_Op op, MPI_Comm comm)
{
int mpi_errno = MPI_SUCCESS;
MPID_Comm *comm_ptr = NULL;
int errflag = FALSE;
MPID_MPI_STATE_DECL(MPID_STATE_MPI_EXSCAN);
MPIR_ERRTEST_INITIALIZED_ORDIE();
MPIU_THREAD_CS_ENTER(ALLFUNC,);
MPID_MPI_COLL_FUNC_ENTER(MPID_STATE_MPI_EXSCAN);
/* Validate parameters, especially handles needing to be converted */
# ifdef HAVE_ERROR_CHECKING
{
MPID_BEGIN_ERROR_CHECKS;
{
MPIR_ERRTEST_COMM(comm, mpi_errno);
}
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;
{
MPID_Datatype *datatype_ptr = NULL;
MPID_Op *op_ptr = NULL;
int rank;
MPID_Comm_valid_ptr( comm_ptr, mpi_errno, FALSE );
if (mpi_errno != MPI_SUCCESS) goto fn_fail;
MPIR_ERRTEST_COMM_INTRA(comm_ptr, mpi_errno);
MPIR_ERRTEST_COUNT(count, mpi_errno);
MPIR_ERRTEST_DATATYPE(datatype, "datatype", mpi_errno);
MPIR_ERRTEST_OP(op, mpi_errno);
if (HANDLE_GET_KIND(datatype) != HANDLE_KIND_BUILTIN) {
MPID_Datatype_get_ptr(datatype, datatype_ptr);
MPID_Datatype_valid_ptr( datatype_ptr, mpi_errno );
if (mpi_errno != MPI_SUCCESS) goto fn_fail;
MPID_Datatype_committed_ptr( datatype_ptr, mpi_errno );
if (mpi_errno != MPI_SUCCESS) goto fn_fail;
}
rank = comm_ptr->rank;
MPIR_ERRTEST_USERBUFFER(sendbuf,count,datatype,mpi_errno);
if (rank != 0) {
MPIR_ERRTEST_RECVBUF_INPLACE(recvbuf, count, mpi_errno);
MPIR_ERRTEST_USERBUFFER(recvbuf,count,datatype,mpi_errno);
}
if (mpi_errno != MPI_SUCCESS) goto fn_fail;
if (HANDLE_GET_KIND(op) != HANDLE_KIND_BUILTIN) {
MPID_Op_get_ptr(op, op_ptr);
MPID_Op_valid_ptr( op_ptr, mpi_errno );
if (mpi_errno != MPI_SUCCESS) goto fn_fail;
}
if (HANDLE_GET_KIND(op) == HANDLE_KIND_BUILTIN) {
mpi_errno =
( * MPIR_OP_HDL_TO_DTYPE_FN(op) )(datatype);
}
if (mpi_errno != MPI_SUCCESS) goto fn_fail;
if (sendbuf != MPI_IN_PLACE && count != 0)
MPIR_ERRTEST_ALIAS_COLL(sendbuf, recvbuf, mpi_errno);
}
MPID_END_ERROR_CHECKS;
}
# endif /* HAVE_ERROR_CHECKING */
/* ... body of routine ... */
mpi_errno = MPIR_Exscan_impl(sendbuf, recvbuf, count, datatype, op, comm_ptr, &errflag);
if (mpi_errno) goto fn_fail;
/* ... end of body of routine ... */
fn_exit:
MPID_MPI_COLL_FUNC_EXIT(MPID_STATE_MPI_EXSCAN);
MPIU_THREAD_CS_EXIT(ALLFUNC,);
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_exscan",
"**mpi_exscan %p %p %d %D %O %C", sendbuf, recvbuf, count, datatype, op, comm);
}
# endif
mpi_errno = MPIR_Err_return_comm( comm_ptr, FCNAME, mpi_errno );
goto fn_exit;
/* --END ERROR HANDLING-- */
}
static int MPIR_Scan_generic (
const void *sendbuf,
void *recvbuf,
int count,
MPI_Datatype datatype,
MPI_Op op,
MPID_Comm *comm_ptr,
int *errflag )
{
MPI_Status status;
int rank, comm_size;
int mpi_errno = MPI_SUCCESS;
int mpi_errno_ret = MPI_SUCCESS;
int mask, dst, is_commutative;
MPI_Aint true_extent, true_lb, extent;
void *partial_scan, *tmp_buf;
MPID_Op *op_ptr;
MPI_Comm comm;
MPIU_THREADPRIV_DECL;
MPIU_CHKLMEM_DECL(2);
if (count == 0) return MPI_SUCCESS;
/* check if multiple threads are calling this collective function */
MPIDU_ERR_CHECK_MULTIPLE_THREADS_ENTER( comm_ptr );
comm = comm_ptr->handle;
comm_size = comm_ptr->local_size;
rank = comm_ptr->rank;
MPIU_THREADPRIV_GET;
/* set op_errno to 0. stored in perthread structure */
MPIU_THREADPRIV_FIELD(op_errno) = 0;
if (HANDLE_GET_KIND(op) == HANDLE_KIND_BUILTIN) {
is_commutative = 1;
}
else {
MPID_Op_get_ptr(op, op_ptr);
if (op_ptr->kind == MPID_OP_USER_NONCOMMUTE)
is_commutative = 0;
else
is_commutative = 1;
}
/* need to allocate temporary buffer to store partial scan*/
MPIR_Type_get_true_extent_impl(datatype, &true_lb, &true_extent);
MPID_Datatype_get_extent_macro(datatype, extent);
MPIU_CHKLMEM_MALLOC(partial_scan, void *, count*(MPIR_MAX(extent,true_extent)), mpi_errno, "partial_scan");
/* This eventually gets malloc()ed as a temp buffer, not added to
* any user buffers */
MPID_Ensure_Aint_fits_in_pointer(count * MPIR_MAX(extent, true_extent));
/* adjust for potential negative lower bound in datatype */
partial_scan = (void *)((char*)partial_scan - true_lb);
/* need to allocate temporary buffer to store incoming data*/
MPIU_CHKLMEM_MALLOC(tmp_buf, void *, count*(MPIR_MAX(extent,true_extent)), mpi_errno, "tmp_buf");
/* adjust for potential negative lower bound in datatype */
tmp_buf = (void *)((char*)tmp_buf - true_lb);
/* Since this is an inclusive scan, copy local contribution into
recvbuf. */
if (sendbuf != MPI_IN_PLACE) {
mpi_errno = MPIR_Localcopy(sendbuf, count, datatype,
recvbuf, count, datatype);
if (mpi_errno) MPIU_ERR_POP(mpi_errno);
}
if (sendbuf != MPI_IN_PLACE)
mpi_errno = MPIR_Localcopy(sendbuf, count, datatype,
partial_scan, count, datatype);
else
mpi_errno = MPIR_Localcopy(recvbuf, count, datatype,
partial_scan, count, datatype);
if (mpi_errno) MPIU_ERR_POP(mpi_errno);
mask = 0x1;
while (mask < comm_size) {
dst = rank ^ mask;
if (dst < comm_size) {
/* Send partial_scan to dst. Recv into tmp_buf */
mpi_errno = MPIC_Sendrecv_ft(partial_scan, count, datatype,
dst, MPIR_SCAN_TAG, tmp_buf,
count, datatype, dst,
MPIR_SCAN_TAG, comm,
&status, errflag);
if (mpi_errno) {
/* for communication errors, just record the error but continue */
*errflag = TRUE;
MPIU_ERR_SET(mpi_errno, MPI_ERR_OTHER, "**fail");
MPIU_ERR_ADD(mpi_errno_ret, mpi_errno);
}
if (rank > dst) {
mpi_errno = MPIR_Reduce_local_impl(
tmp_buf, partial_scan, count, datatype, op);
if (mpi_errno) MPIU_ERR_POP(mpi_errno);
mpi_errno = MPIR_Reduce_local_impl(
tmp_buf, recvbuf, count, datatype, op);
if (mpi_errno) MPIU_ERR_POP(mpi_errno);
}
else {
if (is_commutative) {
mpi_errno = MPIR_Reduce_local_impl(
tmp_buf, partial_scan, count, datatype, op);
if (mpi_errno) MPIU_ERR_POP(mpi_errno);
}
else {
mpi_errno = MPIR_Reduce_local_impl(
partial_scan, tmp_buf, count, datatype, op);
if (mpi_errno) MPIU_ERR_POP(mpi_errno);
mpi_errno = MPIR_Localcopy(tmp_buf, count, datatype,
partial_scan,
count, datatype);
if (mpi_errno) MPIU_ERR_POP(mpi_errno);
}
}
}
mask <<= 1;
}
if (MPIU_THREADPRIV_FIELD(op_errno)) {
mpi_errno = MPIU_THREADPRIV_FIELD(op_errno);
if (mpi_errno) MPIU_ERR_POP(mpi_errno);
}
fn_exit:
MPIU_CHKLMEM_FREEALL();
/* check if multiple threads are calling this collective function */
MPIDU_ERR_CHECK_MULTIPLE_THREADS_EXIT( comm_ptr );
if (mpi_errno_ret)
mpi_errno = mpi_errno_ret;
else if (*errflag)
MPIU_ERR_SET(mpi_errno, MPI_ERR_OTHER, "**coll_fail");
return mpi_errno;
fn_fail:
goto fn_exit;
}
