int ompi_coll_base_reduce_intra_binomial( const void *sendbuf, void *recvbuf,
int count, ompi_datatype_t* datatype,
ompi_op_t* op, int root,
ompi_communicator_t* comm,
mca_coll_base_module_t *module,
uint32_t segsize,
int max_outstanding_reqs )
{
int segcount = count;
size_t typelng;
mca_coll_base_module_t *base_module = (mca_coll_base_module_t*) module;
mca_coll_base_comm_t *data = base_module->base_data;
OPAL_OUTPUT((ompi_coll_base_framework.framework_output,"coll:base:reduce_intra_binomial rank %d ss %5d",
ompi_comm_rank(comm), segsize));
COLL_BASE_UPDATE_IN_ORDER_BMTREE( comm, base_module, root );
/**
* Determine number of segments and number of elements
* sent per operation
*/
ompi_datatype_type_size( datatype, &typelng );
COLL_BASE_COMPUTED_SEGCOUNT( segsize, typelng, segcount );
return ompi_coll_base_reduce_generic( sendbuf, recvbuf, count, datatype,
op, root, comm, module,
data->cached_in_order_bmtree,
segcount, max_outstanding_reqs );
}
/* Todo: gather_intra_generic, gather_intra_binary, gather_intra_chain,
* gather_intra_pipeline, segmentation? */
int
ompi_coll_base_gather_intra_binomial(const void *sbuf, int scount,
struct ompi_datatype_t *sdtype,
void *rbuf, int rcount,
struct ompi_datatype_t *rdtype,
int root,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module)
{
int line = -1, i, rank, vrank, size, total_recv = 0, err;
char *ptmp = NULL, *tempbuf = NULL;
ompi_coll_tree_t* bmtree;
MPI_Status status;
MPI_Aint sextent, slb, strue_lb, strue_extent;
MPI_Aint rextent, rlb, rtrue_lb, rtrue_extent;
mca_coll_base_module_t *base_module = (mca_coll_base_module_t*) module;
mca_coll_base_comm_t *data = base_module->base_data;
size = ompi_comm_size(comm);
rank = ompi_comm_rank(comm);
OPAL_OUTPUT((ompi_coll_base_framework.framework_output,
"ompi_coll_base_gather_intra_binomial rank %d", rank));
/* create the binomial tree */
COLL_BASE_UPDATE_IN_ORDER_BMTREE( comm, base_module, root );
bmtree = data->cached_in_order_bmtree;
ompi_datatype_get_extent(sdtype, &slb, &sextent);
ompi_datatype_get_true_extent(sdtype, &strue_lb, &strue_extent);
vrank = (rank - root + size) % size;
if (rank == root) {
ompi_datatype_get_extent(rdtype, &rlb, &rextent);
ompi_datatype_get_true_extent(rdtype, &rtrue_lb, &rtrue_extent);
if (0 == root){
/* root on 0, just use the recv buffer */
ptmp = (char *) rbuf;
if (sbuf != MPI_IN_PLACE) {
err = ompi_datatype_sndrcv((void *)sbuf, scount, sdtype,
ptmp, rcount, rdtype);
if (MPI_SUCCESS != err) { line = __LINE__; goto err_hndl; }
}
} else {
/* root is not on 0, allocate temp buffer for recv,
* rotate data at the end */
tempbuf = (char *) malloc(rtrue_extent + ((ptrdiff_t)rcount * (ptrdiff_t)size - 1) * rextent);
if (NULL == tempbuf) {
err= OMPI_ERR_OUT_OF_RESOURCE; line = __LINE__; goto err_hndl;
}
ptmp = tempbuf - rtrue_lb;
if (sbuf != MPI_IN_PLACE) {
/* copy from sbuf to temp buffer */
err = ompi_datatype_sndrcv((void *)sbuf, scount, sdtype,
ptmp, rcount, rdtype);
if (MPI_SUCCESS != err) { line = __LINE__; goto err_hndl; }
} else {
/* copy from rbuf to temp buffer */
err = ompi_datatype_copy_content_same_ddt(rdtype, rcount, ptmp,
(char *)rbuf + (ptrdiff_t)rank * rextent * (ptrdiff_t)rcount);
if (MPI_SUCCESS != err) { line = __LINE__; goto err_hndl; }
}
}
total_recv = rcount;
} else if (!(vrank % 2)) {
/* other non-leaf nodes, allocate temp buffer for data received from
* children, the most we need is half of the total data elements due
* to the property of binimoal tree */
tempbuf = (char *) malloc(strue_extent + ((ptrdiff_t)scount * (ptrdiff_t)size - 1) * sextent);
if (NULL == tempbuf) {
err= OMPI_ERR_OUT_OF_RESOURCE; line = __LINE__; goto err_hndl;
}
ptmp = tempbuf - strue_lb;
/* local copy to tempbuf */
err = ompi_datatype_sndrcv((void *)sbuf, scount, sdtype,
ptmp, scount, sdtype);
if (MPI_SUCCESS != err) { line = __LINE__; goto err_hndl; }
/* use sdtype,scount as rdtype,rdcount since they are ignored on
* non-root procs */
rdtype = sdtype;
rcount = scount;
rextent = sextent;
total_recv = rcount;
} else {
/* leaf nodes, no temp buffer needed, use sdtype,scount as
* rdtype,rdcount since they are ignored on non-root procs */
ptmp = (char *) sbuf;
total_recv = scount;
}
if (!(vrank % 2)) {
/* all non-leaf nodes recv from children */
for (i = 0; i < bmtree->tree_nextsize; i++) {
int mycount = 0, vkid;
/* figure out how much data I have to send to this child */
vkid = (bmtree->tree_next[i] - root + size) % size;
mycount = vkid - vrank;
if (mycount > (size - vkid))
mycount = size - vkid;
mycount *= rcount;
OPAL_OUTPUT((ompi_coll_base_framework.framework_output,
"ompi_coll_base_gather_intra_binomial rank %d recv %d mycount = %d",
rank, bmtree->tree_next[i], mycount));
err = MCA_PML_CALL(recv(ptmp + total_recv*rextent, (ptrdiff_t)rcount * size - total_recv, rdtype,
bmtree->tree_next[i], MCA_COLL_BASE_TAG_GATHER,
comm, &status));
if (MPI_SUCCESS != err) { line = __LINE__; goto err_hndl; }
total_recv += mycount;
}
}
if (rank != root) {
/* all nodes except root send to parents */
OPAL_OUTPUT((ompi_coll_base_framework.framework_output,
"ompi_coll_base_gather_intra_binomial rank %d send %d count %d\n",
rank, bmtree->tree_prev, total_recv));
err = MCA_PML_CALL(send(ptmp, total_recv, sdtype,
bmtree->tree_prev,
MCA_COLL_BASE_TAG_GATHER,
MCA_PML_BASE_SEND_STANDARD, comm));
if (MPI_SUCCESS != err) { line = __LINE__; goto err_hndl; }
}
if (rank == root) {
if (root != 0) {
/* rotate received data on root if root != 0 */
err = ompi_datatype_copy_content_same_ddt(rdtype, (ptrdiff_t)rcount * (ptrdiff_t)(size - root),
(char *)rbuf + rextent * (ptrdiff_t)root * (ptrdiff_t)rcount, ptmp);
if (MPI_SUCCESS != err) { line = __LINE__; goto err_hndl; }
err = ompi_datatype_copy_content_same_ddt(rdtype, (ptrdiff_t)rcount * (ptrdiff_t)root,
(char *) rbuf, ptmp + rextent * (ptrdiff_t)rcount * (ptrdiff_t)(size-root));
if (MPI_SUCCESS != err) { line = __LINE__; goto err_hndl; }
free(tempbuf);
}
} else if (!(vrank % 2)) {
/* other non-leaf nodes */
free(tempbuf);
}
return MPI_SUCCESS;
err_hndl:
if (NULL != tempbuf)
free(tempbuf);
OPAL_OUTPUT((ompi_coll_base_framework.framework_output, "%s:%4d\tError occurred %d, rank %2d",
__FILE__, line, err, rank));
return err;
}
