1 #include "colls_private.h"
3 static inline int MPIU_Mirror_permutation(unsigned int x, int bits)
5 /* a mask for the high order bits that should be copied as-is */
6 int high_mask = ~((0x1 << bits) - 1);
7 int retval = x & high_mask;
10 for (i = 0; i < bits; ++i) {
11 unsigned int bitval = (x & (0x1 << i)) >> i; /* 0x1 or 0x0 */
12 retval |= bitval << ((bits - i) - 1);
19 int smpi_coll_tuned_reduce_scatter_mpich_pair(void *sendbuf, void *recvbuf, int recvcounts[],
20 MPI_Datatype datatype, MPI_Op op, MPI_Comm comm)
22 int rank, comm_size, i;
23 MPI_Aint extent, true_extent, true_lb;
26 int mpi_errno = MPI_SUCCESS;
27 int total_count, dst, src;
29 comm_size = smpi_comm_size(comm);
30 rank = smpi_comm_rank(comm);
32 extent =smpi_datatype_get_extent(datatype);
33 smpi_datatype_extent(datatype, &true_lb, &true_extent);
35 if (smpi_op_is_commute(op)) {
39 disps = (int*)xbt_malloc( comm_size * sizeof(int));
42 for (i=0; i<comm_size; i++) {
43 disps[i] = total_count;
44 total_count += recvcounts[i];
47 if (total_count == 0) {
52 if (sendbuf != MPI_IN_PLACE) {
53 /* copy local data into recvbuf */
54 smpi_datatype_copy(((char *)sendbuf+disps[rank]*extent),
55 recvcounts[rank], datatype, recvbuf,
56 recvcounts[rank], datatype);
59 /* allocate temporary buffer to store incoming data */
60 tmp_recvbuf = (void*)xbt_malloc(recvcounts[rank]*(max(true_extent,extent))+1);
61 /* adjust for potential negative lower bound in datatype */
62 tmp_recvbuf = (void *)((char*)tmp_recvbuf - true_lb);
64 for (i=1; i<comm_size; i++) {
65 src = (rank - i + comm_size) % comm_size;
66 dst = (rank + i) % comm_size;
68 /* send the data that dst needs. recv data that this process
69 needs from src into tmp_recvbuf */
70 if (sendbuf != MPI_IN_PLACE)
71 smpi_mpi_sendrecv(((char *)sendbuf+disps[dst]*extent),
72 recvcounts[dst], datatype, dst,
73 COLL_TAG_SCATTER, tmp_recvbuf,
74 recvcounts[rank], datatype, src,
75 COLL_TAG_SCATTER, comm,
78 smpi_mpi_sendrecv(((char *)recvbuf+disps[dst]*extent),
79 recvcounts[dst], datatype, dst,
80 COLL_TAG_SCATTER, tmp_recvbuf,
81 recvcounts[rank], datatype, src,
82 COLL_TAG_SCATTER, comm,
85 if (is_commutative || (src < rank)) {
86 if (sendbuf != MPI_IN_PLACE) {
88 tmp_recvbuf, recvbuf, &recvcounts[rank],
93 tmp_recvbuf, ((char *)recvbuf+disps[rank]*extent),
94 &recvcounts[rank], &datatype);
95 /* we can't store the result at the beginning of
96 recvbuf right here because there is useful data
97 there that other process/processes need. at the
98 end, we will copy back the result to the
99 beginning of recvbuf. */
103 if (sendbuf != MPI_IN_PLACE) {
105 recvbuf, tmp_recvbuf, &recvcounts[rank], &datatype);
106 /* copy result back into recvbuf */
107 mpi_errno = smpi_datatype_copy(tmp_recvbuf, recvcounts[rank],
109 recvcounts[rank], datatype);
110 if (mpi_errno) return(mpi_errno);
114 ((char *)recvbuf+disps[rank]*extent),
115 tmp_recvbuf, &recvcounts[rank], &datatype);
116 /* copy result back into recvbuf */
117 mpi_errno = smpi_datatype_copy(tmp_recvbuf, recvcounts[rank],
121 recvcounts[rank], datatype);
122 if (mpi_errno) return(mpi_errno);
127 /* if MPI_IN_PLACE, move output data to the beginning of
128 recvbuf. already done for rank 0. */
129 if ((sendbuf == MPI_IN_PLACE) && (rank != 0)) {
130 mpi_errno = smpi_datatype_copy(((char *)recvbuf +
132 recvcounts[rank], datatype,
134 recvcounts[rank], datatype );
135 if (mpi_errno) return(mpi_errno);
139 xbt_free(tmp_recvbuf);
145 int smpi_coll_tuned_reduce_scatter_mpich_noncomm(void *sendbuf, void *recvbuf, int recvcounts[],
146 MPI_Datatype datatype, MPI_Op op, MPI_Comm comm)
148 int mpi_errno = MPI_SUCCESS;
149 int comm_size = smpi_comm_size(comm) ;
150 int rank = smpi_comm_rank(comm);
154 int recv_offset, send_offset;
155 int block_size, total_count, size;
156 MPI_Aint true_extent, true_lb;
162 smpi_datatype_extent(datatype, &true_lb, &true_extent);
166 while (pof2 < comm_size) {
171 /* begin error checking */
172 xbt_assert(pof2 == comm_size); /* FIXME this version only works for power of 2 procs */
174 for (i = 0; i < (comm_size - 1); ++i) {
175 xbt_assert(recvcounts[i] == recvcounts[i+1]);
177 /* end error checking */
179 /* size of a block (count of datatype per block, NOT bytes per block) */
180 block_size = recvcounts[0];
181 total_count = block_size * comm_size;
183 tmp_buf0=( void *)xbt_malloc( true_extent * total_count);
184 tmp_buf1=( void *)xbt_malloc( true_extent * total_count);
185 void *tmp_buf0_save=tmp_buf0;
186 void *tmp_buf1_save=tmp_buf1;
188 /* adjust for potential negative lower bound in datatype */
189 tmp_buf0 = (void *)((char*)tmp_buf0 - true_lb);
190 tmp_buf1 = (void *)((char*)tmp_buf1 - true_lb);
192 /* Copy our send data to tmp_buf0. We do this one block at a time and
193 permute the blocks as we go according to the mirror permutation. */
194 for (i = 0; i < comm_size; ++i) {
195 mpi_errno = smpi_datatype_copy((char *)(sendbuf == MPI_IN_PLACE ? recvbuf : sendbuf) + (i * true_extent * block_size), block_size, datatype,
196 (char *)tmp_buf0 + (MPIU_Mirror_permutation(i, log2_comm_size) * true_extent * block_size), block_size, datatype);
197 if (mpi_errno) return(mpi_errno);
204 for (k = 0; k < log2_comm_size; ++k) {
205 /* use a double-buffering scheme to avoid local copies */
206 char *incoming_data = (buf0_was_inout ? tmp_buf1 : tmp_buf0);
207 char *outgoing_data = (buf0_was_inout ? tmp_buf0 : tmp_buf1);
208 int peer = rank ^ (0x1 << k);
212 /* we have the higher rank: send top half, recv bottom half */
216 /* we have the lower rank: recv top half, send bottom half */
220 smpi_mpi_sendrecv(outgoing_data + send_offset*true_extent,
221 size, datatype, peer, COLL_TAG_SCATTER,
222 incoming_data + recv_offset*true_extent,
223 size, datatype, peer, COLL_TAG_SCATTER,
224 comm, MPI_STATUS_IGNORE);
225 /* always perform the reduction at recv_offset, the data at send_offset
226 is now our peer's responsibility */
228 /* higher ranked value so need to call op(received_data, my_data) */
230 incoming_data + recv_offset*true_extent,
231 outgoing_data + recv_offset*true_extent,
233 /* buf0_was_inout = buf0_was_inout; */
236 /* lower ranked value so need to call op(my_data, received_data) */
238 outgoing_data + recv_offset*true_extent,
239 incoming_data + recv_offset*true_extent,
241 buf0_was_inout = !buf0_was_inout;
244 /* the next round of send/recv needs to happen within the block (of size
245 "size") that we just received and reduced */
246 send_offset = recv_offset;
249 xbt_assert(size == recvcounts[rank]);
251 /* copy the reduced data to the recvbuf */
252 result_ptr = (char *)(buf0_was_inout ? tmp_buf0 : tmp_buf1) + recv_offset * true_extent;
253 mpi_errno = smpi_datatype_copy(result_ptr, size, datatype,
254 recvbuf, size, datatype);
255 xbt_free(tmp_buf0_save);
256 xbt_free(tmp_buf1_save);
257 if (mpi_errno) return(mpi_errno);
263 int smpi_coll_tuned_reduce_scatter_mpich_rdb(void *sendbuf, void *recvbuf, int recvcounts[],
264 MPI_Datatype datatype, MPI_Op op, MPI_Comm comm)
266 int rank, comm_size, i;
267 MPI_Aint extent, true_extent, true_lb;
269 void *tmp_recvbuf, *tmp_results;
270 int mpi_errno = MPI_SUCCESS;
271 int dis[2], blklens[2], total_count, dst;
272 int mask, dst_tree_root, my_tree_root, j, k;
274 MPI_Datatype sendtype, recvtype;
275 int nprocs_completed, tmp_mask, tree_root, is_commutative=0;
276 comm_size = smpi_comm_size(comm);
277 rank = smpi_comm_rank(comm);
279 extent =smpi_datatype_get_extent(datatype);
280 smpi_datatype_extent(datatype, &true_lb, &true_extent);
282 if (smpi_op_is_commute(op)) {
286 disps = (int*)xbt_malloc( comm_size * sizeof(int));
289 for (i=0; i<comm_size; i++) {
290 disps[i] = total_count;
291 total_count += recvcounts[i];
294 /* noncommutative and (non-pof2 or block irregular), use recursive doubling. */
296 /* need to allocate temporary buffer to receive incoming data*/
297 tmp_recvbuf= (void *) xbt_malloc( total_count*(max(true_extent,extent)));
298 /* adjust for potential negative lower bound in datatype */
299 tmp_recvbuf = (void *)((char*)tmp_recvbuf - true_lb);
301 /* need to allocate another temporary buffer to accumulate
303 tmp_results = (void *)xbt_malloc( total_count*(max(true_extent,extent)));
304 /* adjust for potential negative lower bound in datatype */
305 tmp_results = (void *)((char*)tmp_results - true_lb);
307 /* copy sendbuf into tmp_results */
308 if (sendbuf != MPI_IN_PLACE)
309 mpi_errno = smpi_datatype_copy(sendbuf, total_count, datatype,
310 tmp_results, total_count, datatype);
312 mpi_errno = smpi_datatype_copy(recvbuf, total_count, datatype,
313 tmp_results, total_count, datatype);
315 if (mpi_errno) return(mpi_errno);
319 while (mask < comm_size) {
322 dst_tree_root = dst >> i;
325 my_tree_root = rank >> i;
328 /* At step 1, processes exchange (n-n/p) amount of
329 data; at step 2, (n-2n/p) amount of data; at step 3, (n-4n/p)
330 amount of data, and so forth. We use derived datatypes for this.
332 At each step, a process does not need to send data
333 indexed from my_tree_root to
334 my_tree_root+mask-1. Similarly, a process won't receive
335 data indexed from dst_tree_root to dst_tree_root+mask-1. */
337 /* calculate sendtype */
338 blklens[0] = blklens[1] = 0;
339 for (j=0; j<my_tree_root; j++)
340 blklens[0] += recvcounts[j];
341 for (j=my_tree_root+mask; j<comm_size; j++)
342 blklens[1] += recvcounts[j];
346 for (j=my_tree_root; (j<my_tree_root+mask) && (j<comm_size); j++)
347 dis[1] += recvcounts[j];
349 mpi_errno = smpi_datatype_indexed(2, blklens, dis, datatype, &sendtype);
350 if (mpi_errno) return(mpi_errno);
352 smpi_datatype_commit(&sendtype);
354 /* calculate recvtype */
355 blklens[0] = blklens[1] = 0;
356 for (j=0; j<dst_tree_root && j<comm_size; j++)
357 blklens[0] += recvcounts[j];
358 for (j=dst_tree_root+mask; j<comm_size; j++)
359 blklens[1] += recvcounts[j];
363 for (j=dst_tree_root; (j<dst_tree_root+mask) && (j<comm_size); j++)
364 dis[1] += recvcounts[j];
366 mpi_errno = smpi_datatype_indexed(2, blklens, dis, datatype, &recvtype);
367 if (mpi_errno) return(mpi_errno);
369 smpi_datatype_commit(&recvtype);
372 if (dst < comm_size) {
373 /* tmp_results contains data to be sent in each step. Data is
374 received in tmp_recvbuf and then accumulated into
375 tmp_results. accumulation is done later below. */
377 smpi_mpi_sendrecv(tmp_results, 1, sendtype, dst,
379 tmp_recvbuf, 1, recvtype, dst,
380 COLL_TAG_SCATTER, comm,
385 /* if some processes in this process's subtree in this step
386 did not have any destination process to communicate with
387 because of non-power-of-two, we need to send them the
388 result. We use a logarithmic recursive-halfing algorithm
391 if (dst_tree_root + mask > comm_size) {
392 nprocs_completed = comm_size - my_tree_root - mask;
393 /* nprocs_completed is the number of processes in this
394 subtree that have all the data. Send data to others
395 in a tree fashion. First find root of current tree
396 that is being divided into two. k is the number of
397 least-significant bits in this process's rank that
398 must be zeroed out to find the rank of the root */
407 tmp_mask = mask >> 1;
409 dst = rank ^ tmp_mask;
411 tree_root = rank >> k;
414 /* send only if this proc has data and destination
415 doesn't have data. at any step, multiple processes
416 can send if they have the data */
418 (rank < tree_root + nprocs_completed)
419 && (dst >= tree_root + nprocs_completed)) {
420 /* send the current result */
421 smpi_mpi_send(tmp_recvbuf, 1, recvtype,
422 dst, COLL_TAG_SCATTER,
425 /* recv only if this proc. doesn't have data and sender
427 else if ((dst < rank) &&
428 (dst < tree_root + nprocs_completed) &&
429 (rank >= tree_root + nprocs_completed)) {
430 smpi_mpi_recv(tmp_recvbuf, 1, recvtype, dst,
432 comm, MPI_STATUS_IGNORE);
440 /* The following reduction is done here instead of after
441 the MPIC_Sendrecv_ft or MPIC_Recv_ft above. This is
442 because to do it above, in the noncommutative
443 case, we would need an extra temp buffer so as not to
444 overwrite temp_recvbuf, because temp_recvbuf may have
445 to be communicated to other processes in the
446 non-power-of-two case. To avoid that extra allocation,
447 we do the reduce here. */
449 if (is_commutative || (dst_tree_root < my_tree_root)) {
452 tmp_recvbuf, tmp_results, &blklens[0],
455 ((char *)tmp_recvbuf + dis[1]*extent),
456 ((char *)tmp_results + dis[1]*extent),
457 &blklens[1], &datatype);
463 tmp_results, tmp_recvbuf, &blklens[0],
466 ((char *)tmp_results + dis[1]*extent),
467 ((char *)tmp_recvbuf + dis[1]*extent),
468 &blklens[1], &datatype);
470 /* copy result back into tmp_results */
471 mpi_errno = smpi_datatype_copy(tmp_recvbuf, 1, recvtype,
472 tmp_results, 1, recvtype);
473 if (mpi_errno) return(mpi_errno);
477 smpi_datatype_free(&sendtype);
478 smpi_datatype_free(&recvtype);
484 /* now copy final results from tmp_results to recvbuf */
485 mpi_errno = smpi_datatype_copy(((char *)tmp_results+disps[rank]*extent),
486 recvcounts[rank], datatype, recvbuf,
487 recvcounts[rank], datatype);
488 if (mpi_errno) return(mpi_errno);
491 xbt_free(tmp_recvbuf);
492 xbt_free(tmp_results);