-
Notifications
You must be signed in to change notification settings - Fork 1
/
mpi_tc.c
2039 lines (1587 loc) · 52.3 KB
/
mpi_tc.c
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
/*!
\file
\brief The various triangle counting routines
*/
#include "tc.h"
#include <mpi.h>
#include <stdlib.h>
#include <inttypes.h>
#include <math.h>
#ifndef DYNAMIC_CHUNK
#define DYNAMIC_CHUNK 16
#endif
#ifndef PPT_DYNAMIC_CHUNK
#define PPT_DYNAMIC_CHUNK 500
#endif
#ifndef factor
#define factor 0.5
#endif
#ifndef BLK_SZ
#define BLK_SZ 1
#endif
#ifndef NPROCS
#define NPROCS 1024
#endif
#define adj_t int64_t
#define MPI_ADJ_INT MPI_LONG_LONG_INT
#define blk_t int
#define MPI_BLK_CNT_INT MPI_INT
//#define ADJNCY_ADM_MAX 1431655766.0
//#define XADJ_ADM_MAX 134217728
#define ADJNCY_ADM_MAX 1431655766.0
#define XADJ_ADM_MAX 1431655766.0
/* OSX timer includes */
#ifdef __MACH__
#include <mach/mach.h>
#include <mach/mach_time.h>
#endif
//double tct, pkt, unpkt, commt;
//double dcntt, ineedt1, ineedt2, ineedt3, ineedt4;
//double ult1, ult2, ult3, ult4, sortt;
//double end1, end2, end3;
/**
* @brief Return the number of seconds since an unspecified tim
e (e.g., Unix
* epoch). This is accomplished with a high-resolution m
onotonic timer,
* suitable for performance timing.
*
* @return The number of seconds.
*/
static inline double monotonic_seconds()
{
#ifdef __MACH__
/* OSX */
static mach_timebase_info_data_t info;
static double seconds_per_unit;
if(seconds_per_unit == 0) {
mach_timebase_info(&info);
seconds_per_unit = (info.numer / info.denom) / 1e9;
}
return seconds_per_unit * mach_absolute_time();
#else
/* Linux systems */
struct timespec ts;
clock_gettime(CLOCK_MONOTONIC, &ts);
return ts.tv_sec + ts.tv_nsec * 1e-9;
#endif
}
static void print_time(double const seconds) {
printf("partial time: %0.04fs\n", seconds);
}
/*************************************************************************/
/*! Determine the iperm for the key order using counting sort.
*/
/*************************************************************************/
ssize_t *gk_sskvipermi(vtxs_t n, ssize_t *xadj)
{
ssize_t i;
ssize_t *iperm, *perm;
int32_t *counts;
gk_i32kv_t *cand;
int range;
/* TODO: check if we need to change this? */
cand = gk_i32kvmalloc(n, "cand");
for (i=0; i<n; i++) {
//printf(" %zd ", (xadj[i+1]-xadj[i]));
cand[i].key = (int32_t)(xadj[i+1]-xadj[i]);
cand[i].val = i;
}
//printf("\n");
for (range=0, i=0; i<n; i++) {
if (cand[i].key > range)
range = cand[i].key;
}
range++;
counts = gk_i32smalloc(range+1, 0, "counts");
for (i=0; i<n; i++)
counts[cand[i].key]++;
MAKECSR(i, range, counts);
//iperm = gk_i32smalloc(n, 0, "iperm");
iperm = (ssize_t *) malloc(n*sizeof(ssize_t));
perm = (ssize_t *) malloc(n*sizeof(ssize_t));
memset(iperm, 0, n*sizeof(ssize_t));
for (i=0; i<n; i++)
iperm[counts[cand[i].key]++] = i;
for (i=0; i<n; i++) {
perm[iperm[i]] = i;
}
gk_free((void **)&counts, &cand, &iperm, LTERM);
return perm;
}
int mod(int a, int b) {
int r = a%b;
return r < 0 ? r + b : r;
}
int ceil1(double x) {
if((int)x * (double)(1.0) == x)
return x;
else
return (int)x + 1;
}
int64_t mpitc_kernelJIKBlockCyclic7(
int shift,
matrix_info_tn *la_matrix,
blob_info_t *rblob,
blob_info_t *cblob)
{
/* TODO: Include checks that will ensure that it doesn't
count triangles for sub blocks of the form i<j<k. */
matrix_info_tn *ra_matrix, *rb_matrix;
int mype, npes, rootp;
MPI_Comm_size(MPI_COMM_WORLD, &npes);
MPI_Comm_rank(MPI_COMM_WORLD, &mype);
int64_t ntriangles=0;
int32_t ltriangles=0; /* Do we need to change this to int64_t too? */
ssize_t la_vi, la_vj;
ssize_t vi, vj, vk;
double start;
double ltime, gtime, atime, mtime;
adj_t l, nc, hmsize, *hmap;
adj_t rb_vtx;
ssize_t *la_xadj, *ra_xadj, *rb_xadj;
adj_t *la_adjncy, *ra_adjncy, *rb_adjncy;
rootp = (int) sqrt(npes);
/* unpkt */
/* Unpack 1. */
ra_matrix = mpitc_unpackBlobBlockCyclic(rblob);
/* Unpack 2. */
rb_matrix = mpitc_unpackBlobBlockCyclic(cblob);
la_xadj = la_matrix->xadj;
ra_xadj = ra_matrix->xadj;
rb_xadj = rb_matrix->xadj;
la_adjncy = la_matrix->adjncy;
ra_adjncy = ra_matrix->adjncy;
rb_adjncy = rb_matrix->adjncy;
//MPI_Barrier(MPI_COMM_WORLD);
//start = MPI_Wtime();
/* Figuring out the size of hmap. */
for (hmsize=0, vi=0; vi<rb_matrix->g_nvtxs; vi++) {
hmsize = gk_max(hmsize, (adj_t)(rb_xadj[vi+1]-rb_xadj[vi]));
}
for (l=1; hmsize>(1<<l); l++);
hmsize = (1<<(l+4))-1;
hmap = gk_i64smalloc(hmsize+1, -1, "hmap");
/* For each communicated ra-chunk, find the number of
residing triangles. */
int32_t g_off = BLK_SZ*rootp;
adj_t ch_vtx = -1;
/* tct */
adj_t nnz_idx;
/* nvtxs corresponds to either unvtxs/lnvtxs. */
/* This helps you navigate the nnz idxs (verify). */
vtxs_t vtxs = rb_matrix->nvtxs;
ssize_t lcnt;
for (la_vi=0; la_vi<rb_matrix->nvtxs; la_vi++) {
nnz_idx = rb_matrix->nnz_idx[la_vi];
if (la_xadj[nnz_idx+1]==la_xadj[nnz_idx])
continue;
for (lcnt=0, la_vj=la_xadj[nnz_idx]; la_vj<la_xadj[nnz_idx+1]; la_vj++) {
lcnt += ra_xadj[la_adjncy[la_vj]+1]-ra_xadj[la_adjncy[la_vj]];
}
if (!lcnt)
continue;
if (rb_adjncy[rb_xadj[nnz_idx+1]-1]-rb_adjncy[rb_xadj[nnz_idx]]+1 > hmsize) { /*Possible collisions.*/
/* Hash rb adjncy to perform lookups, i-th chunk in focus. */
for (nc=0, vi=rb_xadj[nnz_idx+1]-1; vi>=rb_xadj[nnz_idx]; vi--) {
vj = rb_adjncy[vi];
for (l=vj&hmsize; hmap[l]!=-1; l=((l+1)&hmsize), nc++);
hmap[l] = vj;
}
ch_vtx = rb_adjncy[rb_xadj[nnz_idx]];
if (nc > 0) {
for (ltriangles=0, la_vj=la_xadj[nnz_idx]; la_vj<la_xadj[nnz_idx+1]; la_vj++) {
rb_vtx = la_adjncy[la_vj];
for (vj=ra_xadj[rb_vtx+1]-1; vj>=ra_xadj[rb_vtx]; vj--) {
vk = ra_adjncy[vj];
if (vk<ch_vtx)
break;
for (l=(vk&hmsize); hmap[l]!=vk && hmap[l]!=-1; l=((l+1)&hmsize)) {
}
if (hmap[l]==vk) {
ltriangles++;
}
}
}
ntriangles += ltriangles;
for (vi=rb_xadj[nnz_idx+1]-1; vi>=rb_xadj[nnz_idx]; vi--) {
vj = rb_adjncy[vi];
for (l=vj&hmsize; hmap[l]!=vj; l=((l+1)&hmsize)) {
}
hmap[l] = -1;
}
} else {
for (ltriangles=0, la_vj=la_xadj[nnz_idx]; la_vj<la_xadj[nnz_idx+1]; la_vj++) {
rb_vtx = la_adjncy[la_vj];
/* Intersection --
only intersect in case of elements being persisted in hmap. */
for (vj=ra_xadj[rb_vtx+1]-1; vj>=ra_xadj[rb_vtx]; vj--) {
vk = ra_adjncy[vj];
if (vk<ch_vtx)
break;
if (hmap[vk&hmsize]==vk) {
ltriangles++;
}
}
}
ntriangles += ltriangles;
for (vi=rb_xadj[nnz_idx+1]-1; vi>=rb_xadj[nnz_idx]; vi--) {
vj = rb_adjncy[vi];
hmap[vj&hmsize] = -1;
}
}
} else { /*no collisions.*/
/* Hash rb adjncy to perform lookups, i-th chunk in focus. */
for (vi=rb_xadj[nnz_idx+1]-1; vi>=rb_xadj[nnz_idx]; vi--) {
vj = rb_adjncy[vi];
hmap[vj&hmsize] = vj;
}
ch_vtx = rb_adjncy[rb_xadj[nnz_idx]];
for (ltriangles=0, la_vj=la_xadj[nnz_idx]; la_vj<la_xadj[nnz_idx+1]; la_vj++) {
rb_vtx = la_adjncy[la_vj];
/* Intersection --
only intersect in case of elements being persisted in hmap. */
for (vj=ra_xadj[rb_vtx+1]-1; vj>=ra_xadj[rb_vtx]; vj--) {
vk = ra_adjncy[vj];
if (vk<ch_vtx)
break;
if (hmap[vk&hmsize]==vk) {
ltriangles++;
}
}
}
ntriangles += ltriangles;
for (vi=rb_xadj[nnz_idx+1]-1; vi>=rb_xadj[nnz_idx]; vi--) {
vj = rb_adjncy[vi];
hmap[vj&hmsize] = -1;
}
}
}
/*
MPI_Reduce(<ime, >ime, 1,
MPI_DOUBLE, MPI_SUM, 0, MPI_COMM_WORLD);
MPI_Reduce(<ime, &mtime, 1,
MPI_DOUBLE, MPI_MAX, 0, MPI_COMM_WORLD);
//printf("rank: %d ltime: %f. \n", mype, ltime);
if (mype==0) {
atime = gtime/ (double) npes;
printf("rank: %d tct max: %f, avg: %f, imb: %0.4f. \n", mype, mtime, atime, (mtime/atime));
}
MPI_Barrier(MPI_COMM_WORLD);
*/
free(hmap);
free(ra_xadj);
free(rb_xadj);
free(ra_adjncy);
free(rb_adjncy);
free(ra_matrix->nnz_idx);
free(rb_matrix->nnz_idx);
free(ra_matrix);
free(rb_matrix);
return ntriangles;
}
int intcmp(const void *x, const void *y)
{
return(*((int32_t *)x) - *((int32_t *)y));
}
void SetUpRecvInfo(SendInfoTypeGK *sinfo, RecvInfoTypeGK *rinfo) {
int npes, mype;
MPI_Comm_size(MPI_COMM_WORLD, &npes);
MPI_Comm_rank(MPI_COMM_WORLD, &mype);
ssize_t i, j;
MPI_Request send_req[NPROCS], recv_req[NPROCS];
MPI_Status statuses[NPROCS];
int32_t *recv_row = (int32_t *) malloc(npes*sizeof(int32_t));
int32_t *send_row = (int32_t *) malloc(npes*sizeof(int32_t));
for(i=0; i<npes; i++)
send_row[i] = 0;
for(i=0; i<sinfo->n_nbrs; i++) {
send_row[sinfo->nbr_ids[i]] = (int32_t) sinfo->ptr[i+1] - sinfo->ptr[i];
}
/* I decide what I want to request - that's in the SEND row, but it's actually the recv row.
* I will get need to populate the RECV row, but, that's the send row.. */
MPI_Alltoall(send_row, 1, MPI_INT, recv_row, 1, MPI_INT, MPI_COMM_WORLD);
/* What will I receive? */
/* Setup. */
/* ineedt */
ssize_t nrecv;
rinfo->n_nbrs = 0;
for(nrecv=i=0; i<npes; i++) {
if(recv_row[i]) {
nrecv+=recv_row[i];
rinfo->n_nbrs++;
}
}
rinfo->ptr = (ssize_t *) malloc((rinfo->n_nbrs+1) * sizeof(ssize_t));
rinfo->nbr_ids = (int *) malloc(rinfo->n_nbrs * sizeof(int));
rinfo->ind = (adj_t *) malloc(nrecv * sizeof(adj_t));
ssize_t n_inds = 0;
rinfo->n_nbrs = 0;
rinfo->ptr[0] = 0;
for(i=0; i<npes; i++) {
if(recv_row[i]) {
rinfo->nbr_ids[rinfo->n_nbrs++] = (int32_t) i;
n_inds += recv_row[i];
rinfo->ptr[rinfo->n_nbrs] = n_inds;
}
}
int tag1;
tag1 = 2;
/* All non-blocking calls - helps not deadlock the processing. */
#ifdef all2all
int *rcount = (int *) malloc(npes * sizeof(int));
int *rdispls = (int *) malloc(npes * sizeof(int));
int *scount = (int *) malloc(npes * sizeof(int));
int *sdispls = (int *) malloc(npes * sizeof(int));
//if (mype==0)
//gk_startwctimer(vault->timer_3);
j=0;
for(i=0; i<npes; i++) {
if(j<rinfo->n_nbrs && rinfo->nbr_ids[j] == i) {
rcount[i] = rinfo->ptr[j+1] - rinfo->ptr[j];
j++;
} else {
rcount[i] = 0;
}
if(i==0)
rdispls[i] = 0;
else {
rdispls[i] = rcount[i-1] + rdispls[i-1];
}
}
j=0;
for(i=0; i<npes; i++) {
if(j<sinfo->n_nbrs && sinfo->nbr_ids[j] == i) {
scount[i] = sinfo->ptr[j+1] - sinfo->ptr[j];
j++;
} else {
scount[i] = 0;
}
if(i==0)
sdispls[i] = 0;
else {
sdispls[i] = scount[i-1] + sdispls[i-1];
}
}
MPI_Alltoallv(sinfo->ind, scount, sdispls, MPI_INT, rinfo->ind, rcount, rdispls, MPI_INT, MPI_COMM_WORLD);
free(rcount);
free(rdispls);
free(scount);
free(sdispls);
#endif
//if (mype==0)
//gk_stopwctimer(vault->timer_3);
/* Recv. */
for(i=0; i<rinfo->n_nbrs; i++) {
MPI_Irecv(
rinfo->ind+rinfo->ptr[i],
rinfo->ptr[i+1]-rinfo->ptr[i],
MPI_ADJ_INT, rinfo->nbr_ids[i], tag1,
MPI_COMM_WORLD, recv_req+i);
}
/* Sends. */
for(i=0; i<sinfo->n_nbrs; i++) {
MPI_Isend(
sinfo->ind+sinfo->ptr[i],
sinfo->ptr[i+1]-sinfo->ptr[i],
MPI_ADJ_INT, sinfo->nbr_ids[i], tag1,
MPI_COMM_WORLD, send_req+i);
}
MPI_Waitall(rinfo->n_nbrs, recv_req, statuses);
MPI_Waitall(sinfo->n_nbrs, send_req, statuses);
#ifdef isr
#endif
free(recv_row);
free(send_row);
}
/*
n: Total unique nbrs in processor p_i
m: size of unique vtxids - (computed before hand) - xadj_list for local list.
*/
void SetUpSendInfo(
SendInfoTypeGK *sinfo,
ssize_t m, adj_t *nbr_list,
ssize_t *vtxdist,
AccumPRDetails *req_data,
vtxs_t nvtxs)
{
int npes, mype;
MPI_Comm_size(MPI_COMM_WORLD, &npes);
MPI_Comm_rank(MPI_COMM_WORLD, &mype);
ssize_t offset, until;
ssize_t i, j, k, l, nunique;
offset = vtxdist[mype];
until = vtxdist[mype+1];
/* if (mype==npes-1)
until = nvtxs;
else
until = partition*(mype+1); */
sinfo->ind = (adj_t *) malloc(m * sizeof(adj_t));
sinfo->nbr_ids = (int *) malloc(npes * sizeof(int));
sinfo->ptr = (ssize_t *) malloc((npes+1) * sizeof(ssize_t));
/*
Step 1: Determine which vertex ranks need to be received from other processors.
It's the unfortunately populated in the SEND row, but, it's the RECV row.
*/
/* This is what I will recv - unique number of nvtxs,
* m (nunique) is traversed to find the vtx ids outside. */
req_data->nvtxs = 0;
//printf("mype: %d nbr list: ", mype);
for(i=0; i<m; i++) {
//printf(" %"PRId32" ", nbr_list[i]);
if(nbr_list[i]<offset || nbr_list[i]>=until) {
req_data->vtx_ids[req_data->nvtxs] = nbr_list[i];
sinfo->ind[req_data->nvtxs++] = nbr_list[i];
}
}
//printf("\n");
/* j has the count of unique entries. */
nunique = req_data->nvtxs;
/* printf("mype: %d in setup sinfo, ind ", mype);
for (i=0; i<nunique; i++) {
printf(" %"PRId32" ", sinfo->ind[i]);
}
printf("\n"); */
sinfo->n_nbrs = 0;
sinfo->ptr[0] = 0;
ssize_t supposed_until;
supposed_until = 0;
for(j=i=0; i<npes; i++) {
l=j; //some random index to be happy.
/* if(i==npes-1)
supposed_until = nvtxs;
else
supposed_until = supposed_offset+supposed_chunk; */
supposed_until = vtxdist[i+1];
for(; j<nunique; j++) {
if(sinfo->ind[j] >= supposed_until)
break;
}
if(j>l) {
sinfo->nbr_ids[sinfo->n_nbrs] = i; /* id of the ith nbr. */
sinfo->n_nbrs++;
sinfo->ptr[sinfo->n_nbrs] = j; /* i+1th entry gives the end index for nbr i. */
}
//supposed_offset = supposed_offset + supposed_chunk;
//supposed_offset = vtxdist[i+1];
}
}
/**
* Requesting rank values for adjancent vertices that are not local. Also send to guys who are away from you.
*/
void GetValues(
ssize_t *local_vals,
ssize_t *remote_vals,
RecvInfoTypeGK *rinfo,
SendInfoTypeGK *sinfo) {
int npes, mype;
ssize_t i, j;
MPI_Comm_size(MPI_COMM_WORLD, &npes);
MPI_Comm_rank(MPI_COMM_WORLD, &mype);
MPI_Request recv_reqs[NPROCS], send_reqs[NPROCS];
MPI_Status statuses[NPROCS];
/*
Rinfo values are what I want to get, what happens to sinfo? Needs to be sent as well.
Do the same isend/irecv - seems to be faster = but why?
*/
#ifdef all2allv
int *rcount = (int *) malloc(npes * sizeof(int));
int *rdispls = (int *) malloc((npes+1) * sizeof(int));
int *scount = (int *) malloc(npes * sizeof(int));
int *sdispls = (int *) malloc((npes+1) * sizeof(int));
j=0;
for(i=0; i<npes; i++) {
if(j<rinfo->n_nbrs && rinfo->nbr_ids[j] == i) {
rcount[i] = rinfo->ptr[j+1] - rinfo->ptr[j];
j++;
} else {
rcount[i] = 0;
}
if(i==0)
rdispls[i] = 0;
else {
rdispls[i] = rcount[i-1] + rdispls[i-1];
}
}
j=0;
for(i=0; i<npes; i++) {
if(j<sinfo->n_nbrs && sinfo->nbr_ids[j] == i) {
scount[i] = sinfo->ptr[j+1] - sinfo->ptr[j];
j++;
} else {
scount[i] = 0;
}
if(i==0)
sdispls[i] = 0;
else {
sdispls[i] = scount[i-1] + sdispls[i-1];
}
}
MPI_Alltoallv(
local_vals, rcount,
rdispls, MPI_LONG_LONG_INT,
remote_vals, scount,
sdispls, MPI_LONG_LONG_INT,
MPI_COMM_WORLD);
free(rcount);
free(rdispls);
free(scount);
free(sdispls);
#endif
/* Post the recvs. */
for(i=0; i<sinfo->n_nbrs; i++) {
MPI_Irecv(
remote_vals+sinfo->ptr[i],
sinfo->ptr[i+1]-sinfo->ptr[i],
MPI_LONG_LONG_INT,
sinfo->nbr_ids[i], 0,
MPI_COMM_WORLD, send_reqs+i);
}
/* Send the required values as well. */
for(i=0; i<rinfo->n_nbrs; i++) {
MPI_Isend(
local_vals+rinfo->ptr[i],
rinfo->ptr[i+1]-rinfo->ptr[i],
MPI_LONG_LONG_INT,
rinfo->nbr_ids[i], 0,
MPI_COMM_WORLD, recv_reqs+i);
}
MPI_Waitall(sinfo->n_nbrs, send_reqs, statuses);
MPI_Waitall(rinfo->n_nbrs, recv_reqs, statuses);
#ifdef isr
#endif
}
matrix_info_ta *mpitc_ExtractUpperLowerAdjncy(
gk_graph_t *graph,
ssize_t *perm)
{
int mype, npes, rootp;
MPI_Comm_rank(MPI_COMM_WORLD, &mype);
MPI_Comm_size(MPI_COMM_WORLD, &npes);
ssize_t ej, ejend, nedges;
ssize_t hrange, lrange, drange, range;
ssize_t x_offset, total, vi_val, vj_val, proc;
ssize_t vi_off, vi_t, vj_t, nunique;
ssize_t *xadj, *uxadj, *lxadj, *vtxdist;
ssize_t *all_uxadj, *all_lxadj, *all_ruxadj, *all_rlxadj;
ssize_t *urank_array, *rrank_array, *vi_offset, *remote_vals, *all_perm;
vtxs_t nvtxs;
int32_t vi, vj, vk, vl, rank;
adj_t *all_uadjncy, *all_ladjncy, *all_ruadjncy, *all_rladjncy;
adj_t *adjncy, *uadjncy, *ladjncy, *dadjncy, *unadjncy;
int32_t *counts;
chunk_t *iperm;
ssize_t *nvtxs_count, *rnvtxs_count, *nvtxs_displs, *rnvtxs_displs;
//gk_i32kv_t *cand;
gk_i64kv_t *cand;
matrix_info_ta *mat = (matrix_info_ta *) malloc(sizeof(*mat));
/* Access to my partition ranks - have to send these to requesting processors. */
nvtxs = graph->nvtxs;
xadj = graph->xadj;
adjncy = graph->adjncy;
vtxdist = graph->vtxdist;
ssize_t offset = (ssize_t) vtxdist[mype];
ssize_t until = (ssize_t) vtxdist[mype+1];
total = xadj[nvtxs];
dadjncy = (adj_t *) malloc(total*sizeof(adj_t));
/* sorting to find duplicates */
#ifdef ancy_chngs
cand = gk_i32kvmalloc(total, "cand");
#endif
cand = gk_i64kvmalloc(total, "cand");
for (vi_t=0; vi_t<total; vi_t++) {
cand[vi_t].key = adjncy[vi_t];
cand[vi_t].val = (int64_t) vi_t;
}
#ifdef ancy_chngs
gk_i32kvsorti(total, cand);
#endif
gk_i64kvsorti(total, cand);
/* Ineed array here - get unique. */
/* Has issues when #vtxs < #procs. */
dadjncy[0] = (adj_t) cand[0].key;
for (nunique=vi_t=1; vi_t<total; vi_t++)
if (cand[vi_t].key!=cand[vi_t-1].key)
dadjncy[nunique++]=(adj_t)cand[vi_t].key;
SendInfoTypeGK sinfo;
RecvInfoTypeGK rinfo;
AccumPRDetails sdata;
sdata.vtx_ids = (adj_t *) malloc(nunique*sizeof(adj_t));
/* Step 2: Setup send info - RECV info in reality. */
SetUpSendInfo(&sinfo, nunique, dadjncy, vtxdist, &sdata, vtxdist[npes]);
free(dadjncy);
/* Step 3: Setup recv info - SEND info in reality. */
SetUpRecvInfo(&sinfo, &rinfo);
/* ineedt */
remote_vals = (ssize_t *) malloc((sinfo.ptr[sinfo.n_nbrs]) * sizeof(ssize_t));
sdata.vtx_vals = (ssize_t *) malloc(rinfo.ptr[rinfo.n_nbrs]*sizeof(ssize_t));
for (vj=vi=0; vi<rinfo.ptr[rinfo.n_nbrs]; vi++)
sdata.vtx_vals[vj++] = perm[rinfo.ind[vi]-offset];
GetValues((&sdata)->vtx_vals, remote_vals, &rinfo, &sinfo);
free(sdata.vtx_ids);
free(sdata.vtx_vals);
/* ineedt */
for (vi_t=0, vj_t=0; vi_t<total; vi_t++) {
vk = cand[vi_t].key;
if (vk>=offset && vk<until)
adjncy[cand[vi_t].val] = (adj_t) perm[(ssize_t) vk-offset];
else {
if (vk!=sinfo.ind[vj_t])
vj_t++;
adjncy[cand[vi_t].val] = (adj_t) remote_vals[vj_t];
}
}
free(cand);
free(remote_vals);
ssize_t *lcounts;
nvtxs_count = (chunk_t *) malloc(npes*sizeof(chunk_t));
rnvtxs_count = (chunk_t *) malloc(npes*sizeof(chunk_t));
nvtxs_displs = (chunk_t *) malloc((npes+1)*sizeof(chunk_t));
rnvtxs_displs = (chunk_t *) malloc((npes+1)*sizeof(chunk_t));
lcounts = (ssize_t *) malloc(npes*sizeof(ssize_t));
memset(nvtxs_count, 0, npes*sizeof(ssize_t));
memset(lcounts, 0, npes*sizeof(ssize_t));
//mat->uvtxs_displs = rnvtxs_displs;
/* SUB MATRIX INFO - 1 */
rootp = sqrt((double) npes);
/* ult */
urank_array = (ssize_t *) malloc(nvtxs*rootp*sizeof(ssize_t));
vi_offset = (ssize_t *) malloc(nvtxs*sizeof(ssize_t));
/* Figuring out which set of rootp
* procs a vi will be assigned to -
* so I can send adj vertices to these procs. */
for (vi=0; vi<nvtxs; vi++){
vi_val = (perm[vi]/BLK_SZ)%rootp;
vi_offset[vi] = (ssize_t) nvtxs_count[vi_val*rootp];
nvtxs_count[vi_val*rootp]++;
lcounts[vi_val*rootp]++;
}
for (vi=0; vi<npes; vi++) {
nvtxs_count[vi] = nvtxs_count[(vi/rootp)*rootp];
lcounts[vi] = lcounts[(vi/rootp)*rootp];
}
free(sinfo.ind);
free(sinfo.nbr_ids);
free(sinfo.ptr);
free(rinfo.ind);
free(rinfo.nbr_ids);
free(rinfo.ptr);
MPI_Alltoall(
nvtxs_count, 1,
MPI_CHUNK_INT,
rnvtxs_count, 1,
MPI_CHUNK_INT,
MPI_COMM_WORLD);
nvtxs_displs[0] = 0;
for (vi=1; vi<=npes; vi++)
nvtxs_displs[vi] = nvtxs_displs[vi-1]+nvtxs_count[vi-1];
rnvtxs_displs[0] = 0;
for (vi=1; vi<=npes; vi++)
rnvtxs_displs[vi] = rnvtxs_displs[vi-1]+rnvtxs_count[vi-1];
free(rnvtxs_count);
/* ult */
ssize_t p;
for (vi=0; vi<nvtxs; vi++) {
p = perm[vi];
vi_val = (p/BLK_SZ)%rootp;
for (vj=0; vj<rootp; vj++) {
urank_array[nvtxs_displs[vi_val*rootp+vj]++] = p;
}
}
nvtxs_displs[0] = 0;
for (vi=1; vi<=npes; vi++)
nvtxs_displs[vi] = nvtxs_displs[vi-1]+nvtxs_count[vi-1];
free(nvtxs_count);
ssize_t *lcounts_cp;
lcounts_cp = (ssize_t *) malloc(npes*sizeof(ssize_t));
memcpy(lcounts_cp, lcounts, npes*sizeof(ssize_t));
rrank_array = ancy_all2allv_xadj(lcounts, urank_array, MPI_COMM_WORLD);
free(urank_array);
/* ult */
all_uxadj = (ssize_t *) malloc((nvtxs_displs[npes]+1)*sizeof(ssize_t));
all_lxadj = (ssize_t *) malloc((nvtxs_displs[npes]+1)*sizeof(ssize_t));
memset(all_uxadj, 0, ((nvtxs_displs[npes]+1)*sizeof(ssize_t)));
memset(all_lxadj, 0, ((nvtxs_displs[npes]+1)*sizeof(ssize_t)));
for (vi_t=0; vi_t<nvtxs; vi_t++) {
p = perm[vi_t];
vi_val = (p/BLK_SZ)%rootp;
vi_off = vi_offset[vi_t];
for (vj_t=xadj[vi_t]; vj_t<xadj[vi_t+1]; vj_t++) {
rank = adjncy[vj_t];
vj_val = (rank/BLK_SZ)%rootp;
proc = vi_val*rootp + vj_val;
if (p<rank) {
all_uxadj[(ssize_t) nvtxs_displs[proc]+vi_off]++;
} else if (p>rank) {
all_lxadj[(ssize_t) nvtxs_displs[proc]+vi_off]++;
}
}
}
//all_ruxadj = (ssize_t *) malloc((rnvtxs_displs[npes]+1)*sizeof(ssize_t));
//all_rlxadj = (ssize_t *) malloc((rnvtxs_displs[npes]+1)*sizeof(ssize_t));
memcpy(lcounts, lcounts_cp, npes*sizeof(ssize_t));
all_ruxadj = ancy_all2allv_xadj(lcounts, all_uxadj, MPI_COMM_WORLD);
all_rlxadj = ancy_all2allv_xadj(lcounts_cp, all_lxadj, MPI_COMM_WORLD);
/* all_uxadj, all_lxadj makecsr */
MAKECSR(vi, nvtxs_displs[npes], all_uxadj);
MAKECSR(vi, nvtxs_displs[npes], all_lxadj);
MAKECSR(vi, rnvtxs_displs[npes], all_ruxadj);
MAKECSR(vi, rnvtxs_displs[npes], all_rlxadj);
/* ult */
all_uadjncy = (adj_t *) malloc(all_uxadj[nvtxs_displs[npes]]*sizeof(adj_t));
all_ladjncy = (adj_t *) malloc(all_lxadj[nvtxs_displs[npes]]*sizeof(adj_t));
ssize_t test_l, t_disp;
t_disp = all_lxadj[nvtxs_displs[npes]];
for (vi_t=0; vi_t<nvtxs; vi_t++) {
p = perm[vi_t];
vi_val = (p/BLK_SZ)%rootp;
vi_off = vi_offset[vi_t];
for (vj_t=xadj[vi_t]; vj_t<xadj[vi_t+1]; vj_t++) {
rank = adjncy[vj_t];
vj_val = (rank/BLK_SZ)%rootp;
proc = vi_val*rootp + vj_val;
if (p<rank) {
test_l = all_uxadj[(ssize_t) nvtxs_displs[proc]+vi_off];
all_uadjncy[test_l] = rank;
all_uxadj[(ssize_t) nvtxs_displs[proc]+vi_off]++;
} else if (p>rank) {
test_l = all_lxadj[(ssize_t) nvtxs_displs[proc]+vi_off];
all_ladjncy[test_l] = rank;
all_lxadj[(ssize_t) nvtxs_displs[proc]+vi_off]++;
}
}
}
SHIFTCSR(vi, nvtxs_displs[npes], all_uxadj);
SHIFTCSR(vi, nvtxs_displs[npes], all_lxadj);
ssize_t ru_sz = all_ruxadj[rnvtxs_displs[npes]];
ssize_t rl_sz = all_rlxadj[rnvtxs_displs[npes]];
//all_ruadjncy = (adj_t *) malloc(ru_sz*sizeof(adj_t));
//all_rladjncy = (adj_t *) malloc(rl_sz*sizeof(adj_t));
/* lcounts -- counts for each processor id. */
for (vi=0; vi<npes; vi++)
lcounts[vi] = all_uxadj[nvtxs_displs[vi+1]]-all_uxadj[nvtxs_displs[vi]];
free(all_uxadj);
all_ruadjncy = ancy_all2allv(lcounts, all_uadjncy, MPI_COMM_WORLD);
free(all_uadjncy);
for (vi=0; vi<npes; vi++)
lcounts[vi] = all_lxadj[nvtxs_displs[vi+1]]-all_lxadj[nvtxs_displs[vi]];
free(all_lxadj);
all_rladjncy = ancy_all2allv(lcounts, all_ladjncy, MPI_COMM_WORLD);
free(all_ladjncy);
free(lcounts);
free(lcounts_cp);
free(nvtxs_displs);
/* arrays of interest - rrank_array, all_ruxadj,
* all_rlxadj, allruadjncy, allrladjncy, rnvtxs_displs. */
/* sortt */
/* sort by rank. */
for (hrange=0, lrange=0, vi_t=0; vi_t<rnvtxs_displs[npes]; vi_t++) {
if ((p=rrank_array[vi_t]) > hrange)
hrange = p;
else if (p < lrange)
lrange = p;
}
drange = hrange-lrange;
drange++;
counts = gk_i32smalloc(drange+1, 0, "counts");
for (vi_t=0; vi_t<rnvtxs_displs[npes]; vi_t++)
counts[rrank_array[vi_t]-lrange]++;
MAKECSR(vi, drange, counts);
//iperm = gk_i32smalloc(rnvtxs_displs[npes], 0, "iperm");
iperm = (chunk_t *) malloc(rnvtxs_displs[npes]*sizeof(chunk_t));
for (vi_t=0; vi_t<rnvtxs_displs[npes]; vi_t++)
iperm[counts[rrank_array[vi_t]-lrange]++] = (chunk_t) vi_t;
gk_free((void **)&counts, LTERM);