1
1
openmpi/ompi/mca/coll/libnbc/nbc_ireduce_scatter.c
Torsten Hoefler 6b22641669 added LibNBC (http://www.unixer.de/NBC) as collv1 (blocking) component.
I know it does not make much sense but one can play around with the
performance. Numbers are available at http://www.unixer.de/research/nbcoll/perf/.
This is the first step towards collv2. Next step includes the addition
of non-blocking functions to the MPI-Layer and the collv1 interface.

It implements all MPI-1 collective algorithms in a non-blocking manner.
However, the collv1 interface does not allow non-blocking collectives so
that all collectives are used blocking by the ompi-glue layer.

I wanted to add LibNBC as a separate subdirectory, but I could not
convince the buildsystem (and had not the time). So the component looks
pretty messy. It would be great if somebody could explain me how to move
all nbc*{c,h}, and {hb,dict}*{c,h} to a seperate subdirectory.

It's .ompi_ignored because I did not test it exhaustively yet.

This commit was SVN r11401.
2006-08-24 16:47:18 +00:00

133 строки
5.5 KiB
C

#include "nbc.h"
/* an reduce_csttare schedule can not be cached easily because the contents
* ot the recvcounts array may change, so a comparison of the address
* would not be sufficient ... we simply do not cache it */
/* binomial reduce to rank 0 followed by a linear scatter ...
*
* Algorithm:
* pairwise exchange
* round r:
* grp = rank % 2^r
* if grp == 0: receive from rank + 2^(r-1) if it exists and reduce value
* if grp == 1: send to rank - 2^(r-1) and exit function
*
* do this for R=log_2(p) rounds
*
*/
int NBC_Ireduce_scatter(void* sendbuf, void* recvbuf, int *recvcounts, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, NBC_Handle* handle) {
int peer, rank, maxr, p, r, res, count, offset, firstred;
MPI_Aint ext;
char *redbuf, *sbuf, inplace;
NBC_Schedule *schedule;
NBC_IN_PLACE(sendbuf, recvbuf, inplace);
res = NBC_Init_handle(handle, comm);
if(res != NBC_OK) { printf("Error in NBC_Init_handle(%i)\n", res); return res; }
res = MPI_Comm_rank(comm, &rank);
if (MPI_SUCCESS != res) { printf("MPI Error in MPI_Comm_rank() (%i)\n", res); return res; }
res = MPI_Comm_size(comm, &p);
if (MPI_SUCCESS != res) { printf("MPI Error in MPI_Comm_size() (%i)\n", res); return res; }
MPI_Type_extent(datatype, &ext);
if (MPI_SUCCESS != res) { printf("MPI Error in MPI_Type_extent() (%i)\n", res); return res; }
schedule = malloc(sizeof(NBC_Schedule));
if (NULL == schedule) { printf("Error in malloc()\n"); return NBC_OOR; }
res = NBC_Sched_create(schedule);
if(res != NBC_OK) { printf("Error in NBC_Sched_create (%i)\n", res); return res; }
maxr = (int)ceil((log(p)/LOG2));
count = 0;
for(r=0;r<p;r++) count += recvcounts[r];
handle->tmpbuf = malloc(ext*count*2);
if(handle->tmpbuf == NULL) { printf("Error in malloc()\n"); return NBC_OOR; }
redbuf = ((char*)handle->tmpbuf)+(ext*count);
/* copy data to redbuf if we only have a single node */
if((p==1) && !inplace) {
res = NBC_Copy(sendbuf, count, datatype, redbuf, count, datatype, comm);
if (NBC_OK != res) { printf("Error in NBC_Copy() (%i)\n", res); return res; }
}
firstred = 1;
for(r=1; r<=maxr; r++) {
if((rank % (1<<r)) == 0) {
/* we have to receive this round */
peer = rank + (1<<(r-1));
if(peer<p) {
res = NBC_Sched_recv(0, true, count, datatype, peer, schedule);
if (NBC_OK != res) { free(handle->tmpbuf); printf("Error in NBC_Sched_recv() (%i)\n", res); return res; }
/* we have to wait until we have the data */
res = NBC_Sched_barrier(schedule);
if (NBC_OK != res) { free(handle->tmpbuf); printf("Error in NBC_Sched_barrier() (%i)\n", res); return res; }
if(firstred) {
/* take reduce data from the sendbuf in the first round -> save copy */
res = NBC_Sched_op(redbuf-(unsigned long)handle->tmpbuf, true, sendbuf, false, 0, true, count, datatype, op, schedule);
firstred = 0;
} else {
/* perform the reduce in my local buffer */
res = NBC_Sched_op(redbuf-(unsigned long)handle->tmpbuf, true, redbuf-(unsigned long)handle->tmpbuf, true, 0, true, count, datatype, op, schedule);
}
if (NBC_OK != res) { free(handle->tmpbuf); printf("Error in NBC_Sched_op() (%i)\n", res); return res; }
/* this cannot be done until handle->tmpbuf is unused :-( */
res = NBC_Sched_barrier(schedule);
if (NBC_OK != res) { free(handle->tmpbuf); printf("Error in NBC_Sched_barrier() (%i)\n", res); return res; }
}
} else {
/* we have to send this round */
peer = rank - (1<<(r-1));
if(firstred) {
/* we have to send the senbuf */
res = NBC_Sched_send(sendbuf, false, count, datatype, peer, schedule);
} else {
/* we send an already reduced value from redbuf */
res = NBC_Sched_send(redbuf-(unsigned long)handle->tmpbuf, true, count, datatype, peer, schedule);
}
if (NBC_OK != res) { free(handle->tmpbuf); printf("Error in NBC_Sched_send() (%i)\n", res); return res; }
/* leave the game */
break;
}
}
res = NBC_Sched_barrier(schedule);
if (NBC_OK != res) { free(handle->tmpbuf); printf("Error in NBC_Sched_barrier() (%i)\n", res); return res; }
/* rank 0 is root and sends - all others receive */
if(rank != 0) {
res = NBC_Sched_recv(recvbuf, false, recvcounts[rank], datatype, 0, schedule);
if (NBC_OK != res) { free(handle->tmpbuf); printf("Error in NBC_Sched_recv() (%i)\n", res); return res; }
}
if(rank == 0) {
offset = 0;
for(r=1;r<p;r++) {
offset += recvcounts[r-1];
sbuf = ((char *)redbuf) + (offset*ext);
/* root sends the right buffer to the right receiver */
res = NBC_Sched_send(sbuf-(unsigned long)handle->tmpbuf, true, recvcounts[r], datatype, r, schedule);
if (NBC_OK != res) { free(handle->tmpbuf); printf("Error in NBC_Sched_send() (%i)\n", res); return res; }
}
res = NBC_Sched_copy(redbuf-(unsigned long)handle->tmpbuf, true, recvcounts[0], datatype, recvbuf, false, recvcounts[0], datatype, schedule);
if (NBC_OK != res) { free(handle->tmpbuf); printf("Error in NBC_Sched_copy() (%i)\n", res); return res; }
}
/*NBC_PRINT_SCHED(*schedule);*/
res = NBC_Sched_commit(schedule);
if (NBC_OK != res) { free(handle->tmpbuf); printf("Error in NBC_Sched_commit() (%i)\n", res); return res; }
res = NBC_Start(handle, schedule);
if (NBC_OK != res) { free(handle->tmpbuf); printf("Error in NBC_Start() (%i)\n", res); return res; }
/* tmpbuf is freed with the handle */
return NBC_OK;
}