forked from cellularmitosis/ADC-reference-library-2009-july
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathlisting18.html
executable file
·830 lines (698 loc) · 27.6 KB
/
listing18.html
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
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN"
"http://www.w3.org/TR/1999/REC-html401-19991224/loose.dtd">
<html>
<head>
<!-- BEGIN META TAG INFO -->
<meta http-equiv="content-type" content="text/html; charset=UTF-8">
<link rel="home" href="http://developer.apple.com/">
<link rel="find" href="http://developer.apple.com/search/">
<link rel="stylesheet" type="text/css" href="../../documentation/css/adcstyle.css" title="fonts">
<script language="JavaScript" src="../../documentation/js/adc.js" type="text/javascript"></script>
<!-- END META TAG INFO -->
<!-- BEGIN TITLE -->
<title>dist_fft - /transpose_mpi.c</title>
<!-- END TITLE -->
<script language="JavaScript">
function JumpToNewPage() {
window.location=document.scpopupmenu.gotop.value;
return true;
}
</script>
</head>
<!-- BEGIN BODY OPEN -->
<body>
<!--END BODY OPEN -->
<!-- START CENTER OPEN -->
<center>
<!-- END CENTER OPEN -->
<!-- BEGIN LOGO AND SEARCH -->
<!--#include virtual="/includes/adcnavbar"-->
<!-- END LOGO AND SEARCH -->
<!-- START BREADCRUMB -->
<div id="breadcrumb">
<table width="680" border="0" cellpadding="0" cellspacing="0">
<tr>
<td scope="row"><img width="340" height="10" src="images/1dot.gif" alt=""></td>
<td><img width="340" height="10" src="images/1dot.gif" alt=""></td>
</tr>
<tr valign="middle">
<td align="left" colspan="2">
<a href="http://developer.apple.com/">ADC Home</a> > <a href="../../referencelibrary/index.html">Reference Library</a> > <a href="../../samplecode/index.html">Sample Code</a> > <a href="../../samplecode/Performance/index.html">Performance</a> > <a href="../../samplecode/Performance/idxVelocityEngine-date.html">Velocity Engine</a> > <A HREF="javascript:location.replace('index.html');">dist_fft</A> >
</td>
</tr>
<tr>
<td colspan="2" scope="row"><img width="680" height="35" src="images/1dot.gif" alt=""></td>
</tr>
</table>
</div>
<!-- END BREADCRUMB -->
<!-- START MAIN CONTENT -->
<!-- START TITLE GRAPHIC AND INTRO-->
<table width="680" border="0" cellpadding="0" cellspacing="0">
<tr align="left" valign="top">
<td><h1><div id="pagehead">dist_fft</div></h1></td>
</tr>
</table>
<!-- END TITLE GRAPHIC AND INTRO -->
<!-- START WIDE COLUMN -->
<table width="680" border="0" cellpadding="0" cellspacing="0">
<tr align="left" valign="top">
<td id="scdetails">
<h2>/transpose_mpi.c</h2>
<form name="scpopupmenu" onSubmit="return false;" method=post>
<p><strong>View Source Code:</strong>
<select name="gotop" onChange="JumpToNewPage();" style="width:340px"><option selected value="ingnore">Select File</option>
<option value="listing1.html">/dist_fft.c</option>
<option value="listing2.html">/dist_fft.h</option>
<option value="listing3.html">/dist_fft_prefetch.c</option>
<option value="listing4.html">/dist_fft_prefetch.h</option>
<option value="listing5.html">/dist_fft_transpose.c</option>
<option value="listing6.html">/dist_fft_transpose.h</option>
<option value="listing7.html">/dist_fft_twiddle.c</option>
<option value="listing8.html">/dist_fft_twiddle.h</option>
<option value="listing9.html">/dist_fft_types.h</option>
<option value="listing10.html">/LnxMPI/LnxMPI_S.c</option>
<option value="listing11.html">/LnxMPI/mpi.h</option>
<option value="listing12.html">/sched.c</option>
<option value="listing13.html">/sched.h</option>
<option value="listing14.html">/test_main.c</option>
<option value="listing15.html">/test_main.h</option>
<option value="listing16.html">/TOMS_transpose.c</option>
<option value="listing17.html">/TOMS_transpose.h</option>
<option value="listing18.html">/transpose_mpi.c</option>
<option value="listing19.html">/transpose_mpi.h</option></select>
</p>
</form>
<p><strong><a href="dist_fft.zip">Download Sample</a></strong> (“dist_fft.zip”, 716.6K)<BR>
<strong><a href="dist_fft.dmg">Download Sample</a></strong> (“dist_fft.dmg”, 800.0K)</p>
<!--
<p><strong><a href="#">Download Sample</a></strong> (“filename.sit”, 500K)</p>
-->
</td>
</tr>
<tr>
<td scope="row"><img width="680" height="10" src="images/1dot.gif" alt=""><br>
<img height="1" width="680" src="images/1dot_919699.gif" alt=""><br>
<img width="680" height="20" src="images/1dot.gif" alt=""></td>
</tr>
<tr>
<td scope="row">
<!--googleon: index -->
<pre class="sourcecodebox">/*
* Copyright (c) 1997-1999, 2003 Massachusetts Institute of Technology
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*
*/
/*
* Changes made by Advanced Computation Group July 2004
* Copyright (c) 2004 Apple Computer, Inc
*/
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include "transpose_mpi.h"
#include "sched.h"
#include "TOMS_transpose.h"
#include "dist_fft_transpose.h"
/**************************************************************************/
void transpose_mpi_die(const char *error_string)
{
int my_pe;
MPI_Comm_rank(MPI_COMM_WORLD, &my_pe);
fprintf(stderr, "transpose: process %d: %s", my_pe, error_string);
MPI_Abort(MPI_COMM_WORLD, EXIT_FAILURE);
}
static int transpose_mpi_get_block_size(int n, int n_pes)
{
return((n + n_pes - 1) / n_pes);
}
void transpose_mpi_get_local_size(int n, int my_pe, int n_pes,
int *local_n, int *local_start)
{
int block_size;
block_size = transpose_mpi_get_block_size(n,n_pes);
n_pes = (n + block_size - 1) / block_size;
if (my_pe >= n_pes) {
*local_n = 0;
*local_start = 0;
}
else {
*local_start = my_pe * block_size;
if (my_pe == n_pes - 1)
*local_n = n - *local_start;
else
*local_n = block_size;
}
}
#define MAX2(a,b) ((a) > (b) ? (a) : (b))
int transpose_mpi_get_local_storage_size(int nx, int ny,
int my_pe, int n_pes)
{
int local_nx, local_ny, local_x_start, local_y_start;
transpose_mpi_get_local_size(nx,my_pe,n_pes,&local_nx,&local_x_start);
transpose_mpi_get_local_size(ny,my_pe,n_pes,&local_ny,&local_y_start);
return MAX2(1, MAX2(local_nx*ny, nx*local_ny));
}
static int gcd(int a, int b)
{
int r;
do {
r = a % b;
a = b;
b = r;
} while (r != 0);
return a;
}
/**************************************************************************/
transpose_mpi_plan transpose_mpi_create_plan(int nx, int ny,
MPI_Comm transpose_comm)
{
transpose_mpi_plan p;
int my_pe, n_pes, pe;
int x_block_size, local_nx, local_x_start;
int y_block_size, local_ny, local_y_start;
transpose_mpi_exchange *exchange = 0;
int step, send_block_size = 0, recv_block_size = 0, num_steps = 0;
int **sched, sched_npes, sched_sortpe, sched_sort_ascending = 0;
int *perm_block_dest = NULL;
int num_perm_blocks = 0, perm_block_size = 0, perm_block;
char *move = NULL;
int move_size = 0;
int *send_block_sizes = 0, *send_block_offsets = 0;
int *recv_block_sizes = 0, *recv_block_offsets = 0;
MPI_Comm comm;
/* create a new "clone" communicator so that transpose
communications do not interfere with caller communications. */
MPI_Comm_dup(transpose_comm, &comm);
MPI_Comm_rank(comm,&my_pe);
MPI_Comm_size(comm,&n_pes);
/* work space for in-place transpose routine: */
move_size = (nx + ny) / 2;
move = (char *) malloc(sizeof(char) * move_size);
x_block_size = transpose_mpi_get_block_size(nx,n_pes);
transpose_mpi_get_local_size(nx,my_pe,n_pes,&local_nx,&local_x_start);
y_block_size = transpose_mpi_get_block_size(ny,n_pes);
transpose_mpi_get_local_size(ny,my_pe,n_pes,&local_ny,&local_y_start);
/* allocate pre-computed post-transpose permutation: */
perm_block_size = gcd(nx,x_block_size);
num_perm_blocks = (nx / perm_block_size) * local_ny;
perm_block_dest = (int *) malloc(sizeof(int) * num_perm_blocks);
for (perm_block = 0; perm_block < num_perm_blocks; ++perm_block)
perm_block_dest[perm_block] = num_perm_blocks;
/* allocate block sizes/offsets arrays for out-of-place transpose: */
send_block_sizes = (int *) malloc(n_pes * sizeof(int));
send_block_offsets = (int *) malloc(n_pes * sizeof(int));
recv_block_sizes = (int *) malloc(n_pes * sizeof(int));
recv_block_offsets = (int *) malloc(n_pes * sizeof(int));
for (step = 0; step < n_pes; ++step)
send_block_sizes[step] = send_block_offsets[step] =
recv_block_sizes[step] = recv_block_offsets[step] = 0;
if (local_nx > 0 || local_ny > 0) {
sched_npes = n_pes;
sched_sortpe = -1;
for (pe = 0; pe < n_pes; ++pe) {
int pe_nx, pe_x_start, pe_ny, pe_y_start;
transpose_mpi_get_local_size(nx,pe,n_pes,
&pe_nx,&pe_x_start);
transpose_mpi_get_local_size(ny,pe,n_pes,
&pe_ny,&pe_y_start);
if (pe_nx == 0 && pe_ny == 0) {
sched_npes = pe;
break;
}
else if (pe_nx * y_block_size != pe_ny * x_block_size
&& pe_nx != 0 && pe_ny != 0) {
if (sched_sortpe != -1)
transpose_mpi_die("BUG: More than one PE needs sorting!\n");
sched_sortpe = pe;
sched_sort_ascending =
pe_nx * y_block_size > pe_ny * x_block_size;
}
}
sched = make_comm_schedule(sched_npes);
if (!sched) {
MPI_Comm_free(&comm);
return 0;
}
if (sched_sortpe != -1) {
sort_comm_schedule(sched,sched_npes,sched_sortpe);
if (!sched_sort_ascending)
invert_comm_schedule(sched,sched_npes);
}
send_block_size = local_nx * y_block_size;
recv_block_size = local_ny * x_block_size;
num_steps = sched_npes;
exchange = (transpose_mpi_exchange *)
malloc(num_steps * sizeof(transpose_mpi_exchange));
if (!exchange) {
free_comm_schedule(sched,sched_npes);
MPI_Comm_free(&comm);
return 0;
}
for (step = 0; step < sched_npes; ++step) {
int dest_pe;
int dest_nx, dest_x_start;
int dest_ny, dest_y_start;
int num_perm_blocks_received, num_perm_rows_received;
exchange[step].dest_pe = dest_pe =
exchange[step].block_num = sched[my_pe][step];
if (exchange[step].block_num == -1)
transpose_mpi_die("BUG: schedule ended too early.\n");
transpose_mpi_get_local_size(nx,dest_pe,n_pes,
&dest_nx,&dest_x_start);
transpose_mpi_get_local_size(ny,dest_pe,n_pes,
&dest_ny,&dest_y_start);
exchange[step].send_size = local_nx * dest_ny;
exchange[step].recv_size = dest_nx * local_ny;
send_block_sizes[dest_pe] = exchange[step].send_size;
send_block_offsets[dest_pe] = dest_pe * send_block_size;
recv_block_sizes[dest_pe] = exchange[step].recv_size;
recv_block_offsets[dest_pe] = dest_pe * recv_block_size;
/* Precompute the post-transpose permutation (ugh): */
if (exchange[step].recv_size > 0) {
num_perm_blocks_received =
exchange[step].recv_size / perm_block_size;
num_perm_rows_received = num_perm_blocks_received / local_ny;
for (perm_block = 0; perm_block < num_perm_blocks_received;
++perm_block) {
int old_block, new_block;
old_block = perm_block + exchange[step].block_num *
(recv_block_size / perm_block_size);
new_block = perm_block % num_perm_rows_received +
dest_x_start / perm_block_size +
(perm_block / num_perm_rows_received)
* (nx / perm_block_size);
if (old_block >= num_perm_blocks ||
new_block >= num_perm_blocks)
transpose_mpi_die("bad block index in permutation!");
perm_block_dest[old_block] = new_block;
}
}
}
free_comm_schedule(sched,sched_npes);
} /* if (local_nx > 0 || local_ny > 0) */
p = (transpose_mpi_plan)
malloc(sizeof(transpose_mpi_plan_struct));
if (!p) {
free(exchange);
MPI_Comm_free(&comm);
return 0;
}
p->comm = comm;
p->nx = nx; p->ny = ny;
p->local_nx = local_nx; p->local_ny = local_ny;
p->my_pe = my_pe; p->n_pes = n_pes;
p->exchange = exchange;
p->send_block_size = send_block_size;
p->recv_block_size = recv_block_size;
p->num_steps = num_steps;
p->perm_block_dest = perm_block_dest;
p->num_perm_blocks = num_perm_blocks;
p->perm_block_size = perm_block_size;
p->move = move;
p->move_size = move_size;
p->send_block_sizes = send_block_sizes;
p->send_block_offsets = send_block_offsets;
p->recv_block_sizes = recv_block_sizes;
p->recv_block_offsets = recv_block_offsets;
p->all_blocks_equal = send_block_size * n_pes * n_pes == nx * ny &&
recv_block_size * n_pes * n_pes == nx * ny;
if (p->all_blocks_equal)
for (step = 0; step < n_pes; ++step)
if (send_block_sizes[step] != send_block_size ||
recv_block_sizes[step] != recv_block_size) {
p->all_blocks_equal = 0;
break;
}
if (nx % n_pes == 0 && ny % n_pes == 0 && !p->all_blocks_equal)
transpose_mpi_die("n_pes divided dimensions but blocks are unequal!");
/* Set the type constant for passing to the MPI routines;
here, we assume that TRANSPOSE_EL_TYPE is one of the
floating-point types. */
if (sizeof(TRANSPOSE_EL_TYPE) == sizeof(double))
p->el_type = MPI_DOUBLE;
else if (sizeof(TRANSPOSE_EL_TYPE) == sizeof(float))
p->el_type = MPI_FLOAT;
else
transpose_mpi_die("Unknown TRANSPOSE_EL_TYPE!\n");
return p;
}
/**************************************************************************/
void transpose_mpi_destroy_plan(transpose_mpi_plan p)
{
if (p) {
if (p->exchange)
free(p->exchange);
if (p->perm_block_dest)
free(p->perm_block_dest);
if (p->move)
free(p->move);
if (p->send_block_sizes)
free(p->send_block_sizes);
if (p->send_block_offsets)
free(p->send_block_offsets);
if (p->recv_block_sizes)
free(p->recv_block_sizes);
if (p->recv_block_offsets)
free(p->recv_block_offsets);
MPI_Comm_free(&p->comm);
free(p);
}
}
/**************************************************************************/
static void exchange_elements(TRANSPOSE_EL_TYPE *buf1,
TRANSPOSE_EL_TYPE *buf2, int n)
{
int i;
TRANSPOSE_EL_TYPE t0,t1,t2,t3;
for (i = 0; i < (n & 3); ++i) {
t0 = buf1[i];
buf1[i] = buf2[i];
buf2[i] = t0;
}
for (; i < n; i += 4) {
t0 = buf1[i];
t1 = buf1[i+1];
t2 = buf1[i+2];
t3 = buf1[i+3];
buf1[i] = buf2[i];
buf1[i+1] = buf2[i+1];
buf1[i+2] = buf2[i+2];
buf1[i+3] = buf2[i+3];
buf2[i] = t0;
buf2[i+1] = t1;
buf2[i+2] = t2;
buf2[i+3] = t3;
}
}
static void do_permutation(TRANSPOSE_EL_TYPE *data,
int *perm_block_dest,
int num_perm_blocks,
int perm_block_size)
{
int start_block;
/* Perform the permutation in the perm_block_dest array, following
the cycles around and *changing* the perm_block_dest array
to reflect the permutations that have already been performed.
At the end of this routine, we change the perm_block_dest
array back to its original state. (ugh) */
for (start_block = 0; start_block < num_perm_blocks; ++start_block) {
int cur_block = start_block;
int new_block = perm_block_dest[start_block];
while (new_block > 0 && new_block < num_perm_blocks &&
new_block != start_block) {
exchange_elements(data + perm_block_size*start_block,
data + perm_block_size*new_block,
perm_block_size);
perm_block_dest[cur_block] = -1 - new_block;
cur_block = new_block;
new_block = perm_block_dest[cur_block];
}
if (new_block == start_block)
perm_block_dest[cur_block] = -1 - new_block;
}
/* reset the permutation array (ugh): */
for (start_block = 0; start_block < num_perm_blocks; ++start_block)
perm_block_dest[start_block] = -1 - perm_block_dest[start_block];
}
TRANSPOSE_EL_TYPE *transpose_allocate_send_buf(transpose_mpi_plan p,
int el_size)
{
TRANSPOSE_EL_TYPE *send_buf = 0;
/* allocate the send buffer: */
if (p->send_block_size > 0) {
send_buf = (TRANSPOSE_EL_TYPE *)
malloc(p->send_block_size * el_size
* sizeof(TRANSPOSE_EL_TYPE));
if (!send_buf)
transpose_mpi_die("Out of memory!\n");
}
return send_buf;
}
void transpose_in_place_local(transpose_mpi_plan p,
int el_size, TRANSPOSE_EL_TYPE *local_data,
transpose_in_place_which which)
{
switch (which) {
case BEFORE_TRANSPOSE:
if (el_size == 1)
TOMS_transpose_2d(local_data, p->local_nx, p->ny, p->move, p->move_size);
else
TOMS_transpose_2d_arbitrary(local_data,
p->local_nx, p->ny,
el_size,
p->move, p->move_size);
break;
case AFTER_TRANSPOSE:
do_permutation(local_data, p->perm_block_dest,
p->num_perm_blocks, p->perm_block_size * el_size);
break;
}
}
/**************************************************************************/
static void local_transpose_copy(TRANSPOSE_EL_TYPE *src,
TRANSPOSE_EL_TYPE *dest,
int el_size, int nx, int ny)
{
int x, y;
if (el_size == 1) {
if (0) {
dist_fft_transpose_2d(src, 1, dest, 1, nx, ny);
}
else {
for (x = 0; x < nx; ++x)
for (y = 0; y < ny; ++y)
dest[y * nx + x] = src[x * ny + y];
}
}
else if (el_size == 2) {
if (1) {
dist_fft_transpose_2d(src, 2, dest, 2, nx, ny);
dist_fft_transpose_2d(src+1, 2, dest+1, 2, nx, ny);
}
else {
for (x = 0; x < nx; ++x) {
for (y = 0; y < ny; ++y) {
dest[y * (2 * nx) + 2*x] = src[x * (2 * ny) + 2*y];
dest[y * (2 * nx) + 2*x + 1] = src[x * (2 * ny) + 2*y + 1];
}
}
}
}
else
for (x = 0; x < nx; ++x)
for (y = 0; y < ny; ++y)
memcpy(&dest[y * (el_size*nx) + (el_size*x)],
&src[x * (el_size*ny) + (el_size*y)],
el_size * sizeof(TRANSPOSE_EL_TYPE));
}
/* Out-of-place version of transpose_mpi (or rather, in place using
a scratch array): */
static void transpose_mpi_out_of_place(transpose_mpi_plan p, int el_size,
TRANSPOSE_EL_TYPE *local_data,
TRANSPOSE_EL_TYPE *work)
{
local_transpose_copy(local_data, work, el_size, p->local_nx, p->ny);
if (p->all_blocks_equal)
MPI_Alltoall(work, p->send_block_size * el_size, p->el_type,
local_data, p->recv_block_size * el_size, p->el_type,
p->comm);
else {
int i, n_pes = p->n_pes;
for (i = 0; i < n_pes; ++i) {
p->send_block_sizes[i] *= el_size;
p->recv_block_sizes[i] *= el_size;
p->send_block_offsets[i] *= el_size;
p->recv_block_offsets[i] *= el_size;
}
MPI_Alltoallv(work, p->send_block_sizes, p->send_block_offsets,
p->el_type,
local_data, p->recv_block_sizes, p->recv_block_offsets,
p->el_type,
p->comm);
for (i = 0; i < n_pes; ++i) {
p->send_block_sizes[i] /= el_size;
p->recv_block_sizes[i] /= el_size;
p->send_block_offsets[i] /= el_size;
p->recv_block_offsets[i] /= el_size;
}
}
do_permutation(local_data, p->perm_block_dest, p->num_perm_blocks,
p->perm_block_size * el_size);
}
/**************************************************************************/
void transpose_mpi(transpose_mpi_plan p, int el_size,
TRANSPOSE_EL_TYPE *local_data,
TRANSPOSE_EL_TYPE *work)
{
/* if local_data and work are both NULL, we have no way of knowing
whether we should use in-place or out-of-place transpose routine;
if we guess wrong, MPI_Alltoall will block. We prevent this
by making sure that transpose_mpi_get_local_storage_size returns
at least 1. */
if (!local_data && !work)
transpose_mpi_die("local_data and work are both NULL!");
if (work)
transpose_mpi_out_of_place(p, el_size, local_data, work);
else if (p->local_nx > 0 || p->local_ny > 0) {
int step;
TRANSPOSE_EL_TYPE *send_buf = transpose_allocate_send_buf(p,el_size);
transpose_in_place_local(p, el_size, local_data, BEFORE_TRANSPOSE);
for (step = 0; step < p->num_steps; ++step) {
transpose_finish_exchange_step(p, step - 1);
transpose_start_exchange_step(p, el_size, local_data,
send_buf, step, TRANSPOSE_SYNC);
}
transpose_finish_exchange_step(p, step - 1);
transpose_in_place_local(p, el_size, local_data, AFTER_TRANSPOSE);
if (send_buf)
free(send_buf);
} /* if (local_nx > 0 || local_ny > 0) */
}
/**************************************************************************/
/* non-blocking routines for overlapping communication and computation: */
#define USE_SYNCHRONOUS_ISEND 1
#if USE_SYNCHRONOUS_ISEND
#define ISEND MPI_Issend
#else
#define ISEND MPI_Isend
#endif
void transpose_get_send_block(transpose_mpi_plan p, int step,
int *block_y_start, int *block_ny)
{
if (p->local_nx > 0) {
*block_y_start =
p->send_block_size / p->local_nx * p->exchange[step].block_num;
*block_ny = p->exchange[step].send_size / p->local_nx;
}
else {
*block_y_start = 0;
*block_ny = 0;
}
}
void transpose_start_exchange_step(transpose_mpi_plan p,
int el_size,
TRANSPOSE_EL_TYPE *local_data,
TRANSPOSE_EL_TYPE *send_buf,
int step,
transpose_sync_type sync_type)
{
if (p->local_nx > 0 || p->local_ny > 0) {
transpose_mpi_exchange *exchange = p->exchange;
int block = exchange[step].block_num;
int send_block_size = p->send_block_size;
int recv_block_size = p->recv_block_size;
if (exchange[step].dest_pe != p->my_pe) {
/* first, copy to send buffer: */
if (exchange[step].send_size > 0)
memcpy(send_buf,
local_data + el_size*send_block_size*block,
el_size * exchange[step].send_size *
sizeof(TRANSPOSE_EL_TYPE));
#define DO_ISEND \
if (exchange[step].send_size > 0) { \
ISEND(send_buf, \
exchange[step].send_size * el_size, \
p->el_type, \
exchange[step].dest_pe, 0, \
p->comm, \
&p->request[0]); \
}
p->request[0] = MPI_REQUEST_NULL;
p->request[1] = MPI_REQUEST_NULL;
if (sync_type == TRANSPOSE_ASYNC) {
/* Note that we impose an ordering on the sends and
receives (lower pe sends first) so that we won't
have deadlock if Isend & Irecv are blocking in some
MPI implementation: */
if (p->my_pe < exchange[step].dest_pe)
DO_ISEND;
if (exchange[step].recv_size > 0) {
MPI_Irecv(local_data + el_size*recv_block_size*block,
exchange[step].recv_size * el_size,
p->el_type,
exchange[step].dest_pe, MPI_ANY_TAG,
p->comm,
&p->request[1]);
}
if (p->my_pe > exchange[step].dest_pe)
DO_ISEND;
}
else /* (sync_type == TRANSPOSE_SYNC) */ {
MPI_Status status;
MPI_Sendrecv(send_buf,
exchange[step].send_size * el_size,
p->el_type,
exchange[step].dest_pe, 0,
local_data + el_size*recv_block_size*block,
exchange[step].recv_size * el_size,
p->el_type,
exchange[step].dest_pe, MPI_ANY_TAG,
p->comm, &status);
}
}
else if (exchange[step].recv_size > 0 &&
recv_block_size != send_block_size)
memmove(local_data + el_size*recv_block_size*block,
local_data + el_size*send_block_size*block,
exchange[step].recv_size * el_size *
sizeof(TRANSPOSE_EL_TYPE));
}
}
void transpose_finish_exchange_step(transpose_mpi_plan p, int step)
{
if ((p->local_nx > 0 || p->local_ny > 0) && step >= 0
&& p->exchange[step].dest_pe != p->my_pe) {
MPI_Status status[2];
MPI_Waitall(2,p->request,status);
}
}
</pre>
<!--googleoff: index -->
</td>
</tr>
</table>
<!-- END WIDE COLUMN -->
<!-- END MAIN CONTENT -->
<table width="680" border="0" cellpadding="0" cellspacing="0">
<tr>
<td><div style="width: 100%; height: 1px; background-color: #919699; margin-top: 5px; margin-bottom: 15px"></div></td>
</tr>
<tr>
<td align="center"><br/>
<table border="0" cellpadding="0" cellspacing="0" class="graybox">
<tr>
<th>Did this document help you?</th>
</tr>
<tr>
<td>
<div style="margin-bottom: 8px"><a href="http://developer.apple.com/feedback/?v=1&url=/samplecode/dist_fft/listing18.html%3Fid%3DDTS10003377-1.0&media=dvd" target=_new>Yes</a>: Tell us what works for you.</div>
<div style="margin-bottom: 8px"><a href="http://developer.apple.com/feedback/?v=2&url=/samplecode/dist_fft/listing18.html%3Fid%3DDTS10003377-1.0&media=dvd" target=_new>It’s good, but:</a> Report typos, inaccuracies, and so forth.</div>
<div><a href="http://developer.apple.com/feedback/?v=3&url=/samplecode/dist_fft/listing18.html%3Fid%3DDTS10003377-1.0&media=dvd" target=_new>It wasn’t helpful</a>: Tell us what would have helped.</div>
</td>
</tr>
</table>
</td>
</tr>
</table>
<!-- START BOTTOM APPLE NAVIGATION -->
<!--#include virtual="/includes/footer"-->
<!-- END BOTTOM APPLE NAVIGATION -->
<!-- START CENTER CLOSE -->
</center>
<!-- END CENTER CLOSE -->
</body>
</html>